Foetal haemoglobin inducers for reducing blood transfusion in non‐transfusion‐dependent beta‐thalassaemias

Abstract

Background

Non‐transfusion‐dependent β‐thalassaemia (NTDβT) is a subset of inherited haemoglobin disorders characterised by reduced production of the β‐globin chain of haemoglobin leading to anaemia of varying severity. Although blood transfusion is not a necessity for survival, it may be required to prevent complications of chronic anaemia, such as impaired growth and hypercoagulability. People with NTDβT also experience iron overload due to increased iron absorption from food sources which becomes more pronounced in those requiring blood transfusion.

People with a higher foetal haemoglobin (HbF) level have been found to require fewer blood transfusions, thus leading to the emergence of treatments that could increase its level. HbF inducers stimulate HbF production without altering any gene structures. Evidence for the possible benefits and harms of these inducers is important for making an informed decision on their use.

Objectives

To compare the effectiveness and safety of the following for reducing blood transfusion for people with NTDβT:

1. HbF inducers versus usual care or placebo;

2. single HbF inducer with another HbF inducer, and single dose with another dose; and

3. combination of HbF inducers versus usual care or placebo, or single HbF inducer.

Search methods

We used standard, extensive Cochrane search methods. The latest search date was 21 August 2022.

Selection criteria

We included randomised controlled trials (RCTs) or quasi‐RCTs comparing single HbF inducer with placebo or usual care, with another single HbF inducer or with a combination of HbF inducers; or comparing different doses of the same HbF inducer.

Data collection and analysis

We used standard Cochrane methods. Our primary outcomes were blood transfusion and haemoglobin levels. Our secondary outcomes were HbF levels, the long‐term sequelae of NTDβT, quality of life and adverse events.

Main results

We included seven RCTs involving 291 people with NTDβT, aged two to 49 years, from five countries. We reported 10 comparisons using eight different HbF inducers (four pharmacological and four natural): three RCTs compared a single HbF inducer to placebo and seven to another HbF inducer. The duration of the intervention lasted from 56 days to six months. Most studies did not adequately report the randomisation procedures or whether and how blinding was achieved.

HbF inducer against placebo or usual care

Three HbF inducers, HQK‐1001, Radix Astragali or a 3‐in‐1 combined natural preparation (CNP), were compared with a placebo. None of the comparisons reported the frequency of blood transfusion. We are uncertain whether Radix Astragali and CNP increase haemoglobin at three months (mean difference (MD) 1.33 g/dL, 95% confidence interval (CI) 0.54 to 2.11; 1 study, 2 interventions, 35 participants; very low‐certainty evidence). We are uncertain whether Radix Astragali and CNP have any effect on HbF (MD 12%, 95% CI −0.74% to 24.75%; 1 study, 2 interventions, 35 participants; very low‐certainty evidence). Only medians on haemoglobin and HbF levels were reported for HQK‐1001.

Adverse effects reported for HQK‐1001 were nausea, vomiting, dizziness and suprapubic pain. There were no prespecified adverse effects for Radix Astragali and CNP.

HbF inducer versus another HbF inducer

Four studies compared a single inducer with another over three to six months. Comparisons included hydroxyurea versus resveratrol, hydroxyurea versus thalidomide, hydroxyurea versus decitabine and Radix Astragali versus CNP. No study reported our prespecified outcomes on blood transfusion. Haemoglobin and HbF were reported for the comparison Radix Astragali versus CNP, but we are uncertain whether there were any differences (1 study, 24 participants; low‐certainty evidence).

Different doses of the same HbF inducer

Two studies compared two different types of HbF inducers at different doses over two to six months. Comparisons included hydroxyurea 20 mg/kg/day versus 10 mg/kg/day and HQK‐1001 10 mg/kg/day, 20 mg/kg/day, 30 mg/kg/day and 40 mg/kg/day. Blood transfusion, as prespecified, was not reported. In one study (61 participants) we are uncertain whether the lower levels of both haemoglobin and HbF at 24 weeks were due to the higher dose of hydroxyurea (haemoglobin: MD −2.39 g/dL, 95% CI −2.80 to −1.98; very low‐certainty evidence; HbF: MD −10.20%, 95% CI −16.28% to −4.12%; very low‐certainty evidence). The study of the four different doses of HQK‐1001 did not report results for either haemoglobin or HbF. We are not certain if major adverse effects may be more common with higher hydroxyurea doses (neutropenia: risk ratio (RR) 9.93, 95% CI 1.34 to 73.97; thrombocytopenia: RR 3.68, 95% CI 1.12 to 12.07; very low‐certainty evidence). Taking HQK‐1001 20 mg/kg/day may result in the fewest adverse effects.

A combination of HbF inducers versus a single HbF inducer

Two studies compared three combinations of two inducers with a single inducer over six months: hydroxyurea plus resveratrol versus resveratrol or hydroxyurea alone, and hydroxyurea plus l‐carnitine versus hydroxyurea alone. Blood transfusion was not reported.

Hydroxyurea plus resveratrol may reduce haemoglobin compared with either resveratrol or hydroxyurea alone (MD −0.74 g/dL, 95% CI −1.45 to −0.03; 1 study, 54 participants; low‐certainty evidence). We are not certain whether the gastrointestinal disturbances, headache and malaise more commonly reported with hydroxyurea plus resveratrol than resveratrol alone were due to the interventions.

We are uncertain whether hydroxyurea plus l‐carnitine compared with hydroxyurea alone may increase mean haemoglobin, and reduce pulmonary hypertension (1 study, 60 participants; very low‐certainty evidence). Adverse events were reported but not in the intervention group.

None of the comparisons reported the outcome of HbF.

Authors' conclusions

We are uncertain whether any of the eight HbF inducers in this review have a beneficial effect on people with NTDβT. For each of these HbF inducers, we found only one or at the most two small studies. There is no information on whether any of these HbF inducers have an effect on our primary outcome, blood transfusion. For the second primary outcome, haemoglobin, there may be small differences between intervention groups, but these may not be clinically meaningful and are of low‐ to very low‐certainty evidence. Data on adverse effects and optimal doses are limited. Five studies are awaiting classification, but none are ongoing.

Plain language summary

Medicines to increase foetal haemoglobin levels and reduce the need for blood transfusion in people with non‐transfusion‐dependent thalassaemia

Review question

Do medicines that increase foetal haemoglobin (HbF) levels in people with non‐transfusion‐dependent β‐thalassaemia (NTDβT) reduce their need for blood transfusion?

What is thalassaemia?

Thalassaemia is a genetic (inherited) blood disorder that causes defects in adult haemoglobin (the oxygen carrying component of red blood cells) leading to destruction of the red blood cells and anaemia with different degrees of severity. Persistent anaemia can affect general health and reduce quality of life. People with NTDβT may require periodic blood transfusion to replace the red blood cells and this could lead to excess iron being deposited in various organs in the body, affecting their function.

People with NTDβT have higher levels of HbF (the main form of haemoglobin found during the development of a baby before birth) which persists after birth. The amount of HbF that persists varies and people with a higher HbF level require less frequent blood transfusions.

What are HbF inducers?

HbF inducers are substances which increase HbF levels without alteration to the gene. They may reduce the need for blood transfusion in people with NTDβT. However, it is not known which HbF inducers are effective and safe, and if so, what the optimal dose is and at what age treatment should be started.

What did we do?

We searched medical databases for studies comparing single HbF inducer with placebo (dummy treatment) or usual care, with another inducer or with a combination of inducers; or comparing different doses for a same inducer.

What did we find?

We found seven very small randomised controlled trials (where people taking part in the trial had equal chances of being in the treatment or the control group), involving 291 people with NTDβT, aged between two and 49 years, from five countries. These studies varied widely in the type of HbF inducers investigated and their comparison, the doses and how outcomes were reported. The duration of the trials ranged from two to six months. The inducers used include hydroxyurea, decitabine, HQK‐1001, thalidomide, Radix Astragali, resveratrol, l‐carnitine and combined natural preparation (CNP).

Main results

None of the studies reported our main outcome of changes to the frequency of blood transfusion.

All inducers may have caused a small increase in haemoglobin and HbF when compared to placebo, but we are very uncertain about this.

Four studies compared a single HbF inducer against another, for three to six months. There were changes to haemoglobin and HbF, but we cannot be certain if a single HbF inducer or a combination of HbF inducers would work better than another.

Two studies, each compared a different dose of the same inducer. Lower doses of hydroxyurea appeared to increase haemoglobin and HbF levels more than the higher doses, but we are very uncertain. The other study used four different doses of HQK‐1001 but did not actually look for a difference in the effect of these four different doses on haemoglobin or HbF.

Two studies compared a combination of two HbF inducers with a single HbF inducer for six months. We are very uncertain whether the combination or a single inducer improves haemoglobin.

None of the studies reported whether HbF inducers have any effect on quality of life.

Adverse (unwanted) drug effects were reported for each HbF inducer, but there was little information available to guide us on the safety of these substances.

What are the limitations of the evidence?

The main reason we are very uncertain about the effects of HbF inducers are because the studies were all very small and had weaknesses in their design. None of the studies lasted long enough to provide meaningful information for many of the outcomes we measured.

How up to date is this evidence?

The evidence is current to 21 August 2022.

Summary of findings

Summary of findings 1. Single foetal haemoglobin inducer at any dose or duration compared with usual care or placebo.

Single HbF inducer at any dose or duration compared with usual care or placebo for reducing blood transfusion in people with non‐transfusion‐dependent β‐thalassaemia
Patient or population: people with non‐transfusion‐dependent β‐thalassaemia Setting: specialised outpatient clinic in Thailand, Lebanon and China Intervention: single HbF inducer Comparison: placebo
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with placebo or usual care	Risk with single HbF inducer
Frequency of transfusion (transfused volume/year)	None of the studies reported this outcome.
Transfusion‐free interval (days)	None of the studies reported this outcome.
Haemoglobin (g/dL) Follow‐up: 2–3 months	The mean haemoglobin level (g/dL) at 3 months was 7.3 (SD 1.1) g/dL	MD 1.33 g/dL higher (0.54 higher to 2.11 higher)	—	35 (1 RCT, 2 interventions)	⊕⊝⊝⊝ Very lowa, b	The meta‐analysis include Radix Astragali and CNP, but not HQK‐1001. For HQK‐1001, there was no difference in median haemoglobin for any of the 4 doses compared with placebo.
HbF (%) Follow‐up: 2–3 months	The mean HbF level (%) at 3 months was 54% (SD 17%)	MD 12% higher (−0.74% lower to 24.75% higher)	—	35 (1 RCT, 2 interventions)	⊕⊝⊝⊝ Very lowa, c	The meta‐analysis include Radix Astragali and CNP, but not HQK‐1001. Of the 4 different doses of HQK‐1001 (21 participants), there was an increase in median HbF only with the 20 mg/kg dose when compared with placebo.
QoL	None of the studies reported QoL.
Adverse events Follow‐up: 2–3 months	HQK‐1001 The reported adverse events for each dose of HQK‐1001 were URTI, headache, fever, fatigue, nausea, dizziness, palpitations, suprapubic pain, gastritis, gastroenteritis and severe URTI. The 20 mg/kg/day dose was reported as the dose with the least adverse events – on this dose, 2/9 participants reported URTI, 3/9 reported headache, 1/9 reported suprapubic pain, 2/9 reported nausea and 2/9 reported dizziness, and were free from other adverse effects. However, these events were too few to determine whether higher doses were associated with more or fewer adverse effects or whether there were more adverse effects in the HbF inducer or placebo groups. Radix Astragali No adverse events were reported; no deranged blood counts (white blood cells, neutrophils, platelets), liver (GGT, ALT, AST) and renal (BUN, creatinine) blood profiles. CNP No adverse events were reported; no deranged blood counts (white blood cells, neutrophils, platelets), liver (GGT, ALT, AST) and renal (BUN, creatinine) blood profiles			56 (2 RCTs, 3 interventions)	⊕⊝⊝⊝ Very lowa, b	An overall summary of adverse effects reported in the included studies is provided in Table 2.
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). ALT: alanine transaminase; AST: aspartate aminotransferase; BUN: blood urea nitrogen; CI: confidence interval; CNP: combined natural preparation; GGT: gamma‐glutamyl transferase; HbF: foetal haemoglobin; MD: mean difference; QoL: quality of life; SD: standard deviation; URTI: upper respiratory tract infection.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

^aDowngraded one level for indirectness because the follow‐up duration was short and insufficient to see changes.
^bDowngraded two levels for imprecision because of very small information size well below the optimal information size.
^cDowngraded two levels for imprecision because optimal information size was not achieved and there were wide CIs that included both appreciable benefit and appreciable harm.

1. Overview of adverse events reported in included studies.

HbF inducer	Study	Adverse effects prespecified in study method	Adverse effects reporteda
Hydroxyurea	Bohara 2014	Evaluation on blood investigation (complete blood count, kidney and liver function tests).	20 mg/kg/day Severe neutropenia (9/29 participants) Thrombocytopenia (10/29 participants) Raised ALT and AST (5/29 participants) Gastrointestinal adverse effects (9/29 participants) 10 mg/kg/day Severe neutropenia (1/32 participants) Thrombocytopenia (3/32 participants) Raised ALT and AST (2/32 participants) Gastrointestinal adverse effects (3/32 participants)
	Haghpanah 2018	Evaluation on blood investigation (complete blood count, serum ferritin levels, kidney and liver function tests, and fasting blood sugar), and history and physical examination of any occurrence of adverse events.	Gastrointestinal events included severe nausea, vomiting, abdominal pain, diarrhoea, and gastrointestinal bleeding (1/18 participants). Headache and malaise were also reported for some participants who had gastrointestinal symptoms (no specification to which group the participants were randomised into).
	Karimi 2010a	Evaluation of clinical complaints and physical findings. Biochemical parameters and blood counts.	Nausea and vomiting (6 participants), headache (1 participant) and abdominal discomfort (1 participant) without specification to which group the participants were randomised into. No changes to the biochemical parameters and blood counts.
	Jain 2019	Tolerability and safety.	Mild gastrointestinal upset (number of participants not reported).
	Jha 2019	Biochemical parameters, cytopenia, infection.	Respiratory tract infection (10/15 participants) Neurological (1/15 participants) Fever (2/15 participants)
HQK‐1001	Fucharoen 2013	Biochemical parameters, coagulation profile, blood counts, urinalysis and ECG.	URTI 10 mg/kg/day: 5/8 participants 20 mg/kg/day: 2/9 participants 30 mg/kg/day: 1/6 participants 40 mg/kg/day: 1/9 participants placebo: 3/8 participants Headache 10 mg/kg/day: 5/8 participants 20 mg/kg/day: 3/9 participants 30 mg/kg/day: 2/6 participants 40 mg/kg/day: 1/9 participants placebo: 2/8 participants Fever 10 mg/kg/day: 3/8 participants 20 mg/kg/day: 0/9 participants 30 mg/kg/day: 2/6 participants 40 mg/kg/day: 1/9 participants placebo: 2/8 participants Fatigue 10 mg/kg/day: 1/8 participants 20 mg/kg/day: 0/9 participants 30 mg/kg/day: 0/6 participants 40 mg/kg/day: 4/9 participants placebo: 1/8 participants Nausea 10 mg/kg/day: 0/8 participants 20 mg/kg/day: 2/9 participants 30 mg/kg/day: 0/6 participants 40 mg/kg/day: 1/9 participants placebo: 0/8 participants Dizziness 10 mg/kg/day: 2/8 participants 20 mg/kg/day: 2/9 participants 30 mg/kg/day: 0/6 participants 40 mg/kg/day: 1/9 participants placebo: 0/8 participants Palpitations 10 mg/kg/day: 1/8 participants 20 mg/kg/day: 0/9 participants 30 mg/kg/day: 0/6 participants 40 mg/kg/day: 0/9 participants placebo: 0/8 participants Suprapubic pain 10 mg/kg/day: 0/8 participants 20 mg/kg/day: 1/9 participants 30 mg/kg/day: 0/6 participants 40 mg/kg/day: 0/9 participants placebo: 0/8 participants Gastritis 10 mg/kg/day: 0/8 participants 20 mg/kg/day: 0/9 participants 30 mg/kg/day: 1/6 participants 40 mg/kg/day: 0/9 participants placebo: 0/8 participants Gastroenteritis 10 mg/kg/day: 0/8 participants 20 mg/kg/day: 0/9 participants 30 mg/kg/day: 0/6 participants 40 mg/kg/day: 0/9 participants placebo: 1/8 participants Severe URTI 10 mg/kg/day: 0/8 participants 20 mg/kg/day: 0/9 participants 30 mg/kg/day: 0/6 participants 40 mg/kg/day: 0/9 participants placebo: 1/8 participants
Thalidomide	Jain 2019	Tolerability and safety.	Somnolence and headache Number of participants with events not reported.
Decitabine	Jha 2019	Biochemical parameters, cytopenia, infection.	Respiratory tract infection (2/15 participants) Neurological (0/15 participants) Fever (0/15 participants)
Radix Astragali (黄芪) in a tea bag	Lu 2012	Blood counts (WBC, neutrophils, platelets), liver function (GGT, AST, ALT), renal profile (BUN, creatinine).	None occurred.
CNP in a tea bag. CNP: combination of Radix Astragali (黄芪), Codonopsis pilosula (党参) and tortoise plastron (龟板)
Resveratrol	Haghpanah 2018	Evaluation on blood investigation (complete blood count, serum ferritin levels, kidney and liver function tests, and fasting blood sugar), and history and physical examination of any occurrence of adverse events.	Gastrointestinal events included severe nausea, vomiting, abdominal pain, diarrhoea and gastrointestinal bleeding (5/16 participants). Headache and malaise were also reported for some of the participants who had gastrointestinal symptoms (no specification to which group the participants were randomised into).
Hydroxyurea plus resveratrol	Haghpanah 2018		Gastrointestinal events included severe nausea, vomiting, abdominal pain, diarrhoea, and gastrointestinal bleeding (5/20 participants) Headache and malaise were also reported for some of the participants who had gastrointestinal symptoms (no specification to which group the participants were randomised into).
Hydroxyurea plus l‐carnitine	Karimi 2010a	Evaluation of clinical complaints and physical findings. Biochemical parameters and blood counts.	Nausea and vomiting (6 participants), headache (1 participant) and abdominal discomfort (1 participant) without specification to which group the participants were randomised into. No changes to the biochemical parameters and blood counts.

'None occurred' means that adverse effects were specifically looked for, but none were identified.
ALT: alanine transaminase; AST: aspartate aminotransferase; BUN: blood urea nitrogen; CNP: combination natural preparation; ECG: electrocardiogram; GGT: gamma‐glutamyl transferase; URTI: upper respiratory tract infection; WBC: white blood cell count.

Summary of findings 2. Single foetal haemoglobin inducer at any dose or duration compared with another foetal haemoglobin inducer for reducing blood transfusion in people with non‐transfusion‐dependent β‐thalassaemia.

Single HbF inducer at any dose or duration compared with another HbF inducer for reducing blood transfusion in people with non‐transfusion‐dependent β‐thalassaemia
Patient or population: people with non‐transfusion‐dependent β‐thalassaemia Setting: specialised outpatient clinic in Iran, India and China Intervention: single HbF inducer Comparison: another single HbF inducer
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with single HbF inducer	Risk with another single HbF inducer
Frequency of transfusion (transfused volume/year) Follow‐up: 6 months	None of the studies reported this outcome using the scale transfused volume/year.		—	—	—	1 study reported frequency of transfusion (30 participants, 15 per arm) as mean number of units of blood/participant/year. SDs were not reported. The hydroxyurea group needed a mean of 3 units/participant/year of blood volume (range 1 to 6), while 0 participants in the thalidomide group needed any units of blood.
Transfusion‐free interval (days)	None of the studies reported transfusion‐free interval.
Haemoglobin (g/dL) Follow‐up: 3–6 months	Hydroxyurea lowered mean haemoglobin compared to resveratrol by MD 0.30 g/dL (95% CI −1.14 to 0.54). Radix Astragali lowered mean haemoglobin compared to CNP by MD 0.50 g/dL (95% CI −1.38 to 0.38).		—	118 (4 RCTs, 4 pair‐wise intervention comparisons)	⊕⊕⊝⊝ Lowa	Comparisons include: hydroxyurea vs resveratrol; hydroxyurea vs thalidomide; hydroxyurea vs decitabine; and Radix Astragali vs CNP. The study comparing hydroxyurea vs thalidomide did not report this outcome and the study comparing hydroxyurea vs decitabine reported this outcome as peak median haemoglobin.
HbF (%) Follow‐up: 3–6 months	There was no difference between thalidomide and hydroxyurea, MD 3.50% (95% CI −1.41% to 8.41%). Radix Astragali increased the mean HbF levels compared to natural combination preparation by 5.00% (95% CI −11.44% to 21.44%).		—	84 (3 RCTs, 3 pair‐wise intervention comparisons)	⊕⊕⊝⊝ Lowb	Comparisons include hydroxyurea vs thalidomide; hydroxyurea vs decitabine; and Radix Astragali vs CNP. The study comparing hydroxyurea vs decitabine reported this outcome but did not provide sufficient information to compare the 2 groups.
QoL	None of the studies reported QoL.
Adverse events Follow‐up: 3–6 months	Hydroxyurea mild gastrointestinal upset (Jain 2019) respiratory tract infection (Jha 2019) neurological problems (Jha 2019) fever (Jha 2019) Thalidomide somnolence and headache (Jain 2019) Decitabine respiratory tract infection (Jha 2019) Resveratrol gastrointestinal events included severe nausea, vomiting, abdominal pain, diarrhoea and gastrointestinal bleeding, headache and malaise (Haghpanah 2018) Radix Astragali None of the prespecified adverse drug effects (abnormalities to blood counts, liver enzymes levels and renal profile) occurred (Lu 2012). CNP None of the prespecified adverse drug effects (abnormalities to blood counts, liver enzymes levels and renal profile) occurred (Lu 2012).			118 (4 RCTs, 4 pair‐wise intervention comparisons)	⊕⊝⊝⊝ Very lowc, d	An overall summary of adverse effects reported in the included studies is provided in Table 2.
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; CNP: combined natural preparation; HbF: foetal haemoglobin; MD: mean difference; QoL: quality of life; SD: standard deviation.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

^aDowngraded two levels for imprecision because data from only one small trial and the CIs included both appreciable benefit and appreciable harm.
^bDowngraded two levels for imprecision because data were based on one small trial with very wide CIs.
^cDowngraded one level for lack of blinding.
^dDowngraded two levels for imprecision, data from only one small study with very small information size.

Summary of findings 3. One dose or dose regimen of a foetal haemoglobin inducer compared with another dose or dose regimen of the same foetal haemoglobin inducer for reducing blood transfusion in people with non‐transfusion‐dependent β‐thalassaemia.

One dose or dose regimen of an HbF inducer compared with another dose or dose regimen of the same HbF inducer for reducing blood transfusion in people with non‐transfusion‐dependent β‐thalassaemia
Patient or population: people with non‐transfusion‐dependent β‐thalassaemia Setting: specialised outpatient clinic in India Intervention: 1 dose of an HbF inducer Comparison: a different dose of the same HbF inducer
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with 1 dose (lower) of HbF inducer	Risk with another dose (higher) of HbF inducer
Frequency of transfusion (transfused volume/ year)	None of the studies reported frequency of transfusion.
Transfusion‐free interval (days)	None of the studies reported transfusion‐free interval.
Haemoglobin (g/dL) Follow‐up: 6 months	The mean haemoglobin with 10 mg/kg/day was 9.12 (SD 0.70)g/dL	The mean haemoglobin with 20 mg/kg/day was2.39 g/dL lower (95% CI 2.80 lower to 1.98 lower)	—	61 (1 RCT)	⊕⊝⊝⊝ Very lowa,b	Study compared hydroxyurea 10 mg/kg/day vs 20 mg/kg/day. Not downgraded for lack of blinding because of outcome considered to be objective.
HbF (%) Follow‐up: 6 months	The mean HbF for 10 mg/kg/day was 35.4% (SD 12.3%)	The mean HbF with 20 mg/kg/day was 10.20% lower (95% CI 16.28% lower to 4.12% lower)	—	61 (1 RCT)	⊕⊝⊝⊝ Very lowa,b	Study compared hydroxyurea 10 mg/kg/day vs 20 mg/kg/day. Not downgraded for lack of blinding because outcome considered to be objective.
QoL	None of the studies reported QoL.
Adverse events Follow‐up: 6 months	—			82 (2 RCTs, 2 interventions, 6 different doses)	⊕⊝⊝⊝ Very lowb,d	1 study compared hydroxyurea 10 mg/kg/day vs 20 mg/kg/day; 1 study compared 4 doses of HQK‐1001: 10 mg/kg/day; 20 mg/kg/day; 30 mg/kg/day; 40 mg/kg/day (for a summary of these findings at 2 months, see Table 1). An overall summary of adverse effects reported in the included studies is provided in Table 2.
Neutropenia (WBC < 1 × 103/μL, ANC < 0.5 × 103/μL)	31 per 1000 with hydroxyurea 10 mg/kg/day	310 per 1000 (42 to 1000) with hydroxyurea 20 mg/kg/day	RR 9.93 (1.34 to 73.67)	61 (1 RCT)	⊕⊝⊝⊝ Very lowc,d	Neutropenia would influence the decision to continue the intervention but not the decision to transfuse blood. Not downgraded for lack of blinding because outcome considered to be objective.
Thrombocytopenia (platelet count < 50 × 103/μL)	94 per 1000 with hydroxyurea 10 mg/kg/day	345 per 1000 (105 to 1000) with hydroxyurea 20 mg/kg/day	RR 3.68 (1.12 to 12.07)	61 (1 RCT)	⊕⊝⊝⊝ Very lowc,d	Thrombocytopenia would influence the decision to continue the intervention but not the decision to transfuse blood. Not downgraded for lack of blinding because outcome considered to be objective.
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). ANC: absolute neutrophil count; CI: confidence interval; HbF: foetal haemoglobin; MD: mean difference; QoL: quality of life; RR: risk ratio; SD: standard deviation; WBC: white blood cell count.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

^aDowngraded one level for risk of bias because of unclear allocation concealment.
^bDowngraded two levels for imprecision because data were based on one small trial with very small information size.
^cDowngraded two levels for imprecision because data were based on one small trial with very wide CIs.
^dDowngraded one level for indirectness because the follow‐up duration was short and insufficient to see changes.

Summary of findings 4. Two or more foetal haemoglobin inducers at any dose or duration compared with a usual management protocol, placebo or a single foetal haemoglobin inducer for reducing blood transfusion in people with non‐transfusion‐dependent β‐thalassaemia.

Two or more HbF inducers at any dose or duration compared with a usual management protocol, placebo or a single HbF inducer for reducing blood transfusion in people with non‐transfusion‐dependent β‐thalassaemia
Patient or population: people with non‐transfusion‐dependent β‐thalassaemia Setting: outpatient clinics in Iran Intervention: combined ≥ 2 HbF inducers Comparison: single HbF inducer
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with single HbF inducer	Risk with ≥ 2 HbF inducers
Frequency of transfusion (transfused volume/year)	None of the studies reported frequency of transfusion.
Transfusion‐free interval (days)	None of the studies reported transfusion‐free interval.
Haemoglobin level (g/dL) Follow‐up: 6 months	The mean haemoglobin level (g/dL) with a single HbF inducer ranged from 8.5 (SD 1.2) g/dL to 8.8 (SD 1.3) g/dL	The mean haemoglobin level with ≥ 2 HbF inducers was 0.74 g/dL lower (1.45 lower to 0.03 lower)	—	54 (1 RCT, 3 intervention combinations)	⊕⊕⊝⊝ Lowa	The single HbF inducer here refers to either hydroxyurea or resveratrol alone. Not downgraded for lack of blinding because outcome considered to be objective.
HbF levels (%)	None of the studies reported HbF levels.
QoL	None of the studies reported QoL.
Adverse events Follow‐up: 6 months	Hydroxyurea vs resveratrol vs combined hydroxyurea plus resveratrol (Haghpanah 2018) gastrointestinal events including severe nausea, vomiting, abdominal pain, diarrhoea, and gastrointestinal bleeding headache malaise Hydroxyurea vs combined hydroxyurea plus l‐carnitine (Karimi 2010a) nausea and vomiting (6 participants) headache (1 participant) abdominal discomfort (1 participant) no changes to the biochemical parameters and blood counts			104 (2 RCTs, 3 pair‐wise comparisons)	⊕⊝⊝⊝ Very lowa,b	Karimi 2010a reported adverse events, but data could not be separated from study arms that we had excluded from this review. An overall summary of adverse effects reported in the included studies is provided in Table 2.
*The risk in the intervention group (and its 95% CI) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; HbF: foetal haemoglobin; MD: mean difference; QoL: quality of life; RCT: randomised controlled trial; SD: standard deviation.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

^aDowngraded two levels for imprecision because based on one small trial with very small information size.
^bDowngraded two levels for risk of bias because unclear randomisation, allocation concealment and blinding.

Background

Description of the condition

Thalassaemia is an inherited disease resulting in abnormal haemoglobin. This affects both the production and life span of red cells resulting in anaemia of varying degrees. It is found mainly in populations originating from malaria‐endemic areas (Weatherall 2001), such as Africa (Weatherall 2012), Asia (Kohne 2010; Weatherall 2012), and the Mediterranean (De Sanctis 2017). Globally, thalassaemia affects approximately 56,000 newborns annually (Modell 2008), and a third of the population with thalassaemia is considered non‐transfusion‐dependent (NTD) (Weatherall 2012). This population lies between those who require regular transfusion for survival (transfusion‐dependent thalassaemia (TDT)) and those who do not require transfusion beyond that of a person without the condition (carrier state or thalassaemia trait). The types of NTD thalassaemia include β‐thalassaemia intermedia, haemoglobin E combined with β‐thalassaemia (HbE/β‐thalassaemia) and haemoglobin H (HbH) disease. The β‐globin chain is affected in HbE/β‐thalassaemia and β‐thalassaemia intermedia, whereas α‐globin chain is affected in HbH disease (Modell 2008; Weatherall 2012). The inheritance can be confirmed by Hb electrophoresis and DNA analysis for α‐ and β‐globin chain.

In this review, β‐thalassaemia intermedia and HbE/β‐thalassaemia will be collectively referred to as non‐transfusion‐dependent β‐thalassaemia (NTDβT), the major contributors to NTD thalassaemia. They present with an extremely diverse clinical spectrum of chronic anaemia and face a risk of developing complications such as bone deformities, extramedullary haemopoiesis, pulmonary hypertension and hypercoagulable episodes (Taher 2010). These changes occur over a number of years. The diversity arises from multiple genetic differences that determine the lifespan of a haemoglobin molecule (Camaschella 1995; Kohne 2011), and the presence of persistent higher levels of foetal haemoglobin (HbF) (Musallam 2012; Weatherall 2001). The proportion of HbF could differ within one genetic variation (Nuinoon 2010; Sankaran 2013), but since it has some functional properties, the persistence of HbF might be important in preventing red cell transfusion (Carrocini 2011; Sankaran 2013). As a result, depending on the proportion and type of HbF, and its affinity for oxygen, a clinical spectrum is observed which ranges from mild asymptomatic anaemia to the occasional need for transfusion.

Red blood cell (RBC) transfusion is the mainstay of treatment (Asadov 2018; Borgna‐Pignatti 2007), and is determined by factors such as stunted growth and manifestation of extramedullary erythropoiesis in children; and during demanding conditions such as pregnancy and heart failure in adults (Taher 2010; TIF 2018a).

However, each transfusion will exacerbate the existing problem of high iron deposition from ongoing increased gut iron absorption secondary to the low hepcidin concentrations found in NTDβT (Nemeth 2009; Origa 2007; Pippard 1979) and haem breakdown (Taher 2009); this causes iron overload and its complications, although it is less severe than in people with TDT (Taher 2010; Weatherall 2001). Transfusion also increases risks of blood‐transmitted infections, RBC antibody formation, hypersensitivity reactions and cholecystitis. Chelating agents to remove excessive iron deposits are initiated at lower iron levels than that for TDT (Taher 2009), but they too have their own disadvantages (Botzenhardt 2018; Cappellini 2006; Fisher 2013).

Description of the intervention

HbF inducers are natural or synthetic agents which increase HbF levels without altering the globin gene sequence (Bianchi 2009), such as preventing the replacement of HbF with adult haemoglobin and inhibiting the actions of transcription factor genes silencers (Uda 2008), or enzymes that regulate γ‐globin expression (Fard 2013; Sankaran 2013; Shearstone 2016; Witt 2003).

In a number of studies in people with β‐thalassaemia, HbF inducers have been found to increase haemoglobin and reduce blood transfusion (Bradai 2003; Singer 2005a). Therapies to raise hepcidin levels and erythroid cell maturating agents are alternative approaches to the use of HbF inducers to reduce the need for transfusion in people with thalassaemia, but such interventions are still under investigation (Casu 2018; Taher 2021). One such therapy is luspatercept, an erythrocyte‐maturing agent (Casu 2018).

Hydroxyurea, a ribonucleotide reductase inhibitor, is the most widely studied inducer for people with NTDβT, sickle cell disease (SCD) (Rankine‐Mullings 2022) and TDT (Ansari 2019), and is the only one that is currently in use for NTDβT. It has been included in a guideline published by the Thalassaemia International Federation (TIF 2018b). Other promising inducers include histone deacetylase (HDAC) inhibitors (Migliaccio 2008), such as butyrates (Perrine 2011; Reid 2014), short fatty acids (Pace 2002), decitabine (Kalantri 2018; Olivieri 2011), and valproate acid (Selby 1997); immunomodulators such as thalidomide (Aerbajinai 2007); and natural agents (Rodrigue 2001), such as resveratrol (Bianchi 2009), l‐carnitine (Perrine 1989), Yisui Shengxue granules (YSSXG) (Cheng 2016), bergamot (Guerrini 2009), and Fructus Trichosanthis (Li 2011), but their usages are mainly in the experimental phase.

All are administered orally, except decitabine which is given subcutaneously. Information is sparse about the duration needed for each inducer to exert its peak response and maintain its effectiveness beyond the last therapeutic dose (Musallam 2013a; Steinberg 2010). Hydroxyurea is effective within six months of treatment (Kinney 1999; Strouse 2012; Zimmerman 2004), and reaches its peak levels within four hours of ingestion (Charache 1992), but the duration of effect is uncertain. Two observational studies suggested hydroxyurea loses its efficacy after 12 months (Mancuso 2006; Rigano 2010). However, one study has suggested improved outcomes up to 24 months (Italia 2009). There are no data beyond 24 months. Decitabine may be effective within two weeks (Olivieri 2011), and its efficacy is sustained for eight weeks (Kalantri 2018). It has been reported that YSSXG may sustain its efficacy for four weeks. Details about other inducers remain uncertain and these are still at an experimental stage. Most inducers do not have follow‐up data beyond six months. Their effectiveness could also be influenced by genetic variations (Musallam 2013a; Pourfarzad 2013), and pretreatment HbF levels (Pourfarzad 2013).

Each inducer has its own reported adverse effects, which could be dose‐dependent, transient or reversible after discontinuing treatment (Strouse 2012; Zimmerman 2004); these include neutropenia, oligospermia, skin rashes with hydroxyurea (Zimmerman 2004); gastrointestinal disturbances, headache, elevated liver enzymes with HQK‐1001 (Patthamalai 2014); carcinogenic effects, the thrombotic risk with decitabine (Olivieri 2011); rash (Ren 2018), and teratogenic effects with thalidomide.

Inducers have been prescribed to children with NTDβT but the optimal age for initiating therapy and the appropriate dose is unknown (Bradai 2007; Wang 2009). Some inducers may have additional antioxidant properties possibly reducing erythrocyte destruction (Ng 2014). Combinations of inducers may have synergistic effects (Singer 2005b).

How the intervention might work

HbF is highest during foetal life and after birth there is a natural decline. In people with β‐thalassaemia, defective haemoglobins are produced once the HbF level has declined. High levels of HbF have been found to reduce the complications of chronic anaemia (Gamberini 2004; Meo 2008), and the need for transfusion (Bradai 2007; Kosaryan 2014). The mechanisms that enhance the activity and expression of RBC gamma‐globin (γ‐globin), a major component of HbF (α₂γ₂) could raise HbF levels. This rise in turn could help to stabilise the imbalance of the α/β chain, thus improving ineffective erythropoiesis, reducing RBC haemolysis and causing a lesser degree of anaemia; and ultimately, resulting in a reduction in the need of blood transfusion. This is why higher levels HbF are found in people with β‐thalassaemia and why HbF inducers might be an effective treatment to reduce blood transfusions.

The effect of HbF inducers on haemoglobin varies and may be affected by different doses or types of inducers. There is little information if any about the onset or duration of effect (Hoppe 1999; Singer 2005a).

Why it is important to do this review

Considering the problems faced with blood transfusion, the use of therapies aimed at increasing HbF has become of interest and if found to be safe and effective could impact the quality of life (QoL) by reducing blood transfusion. HbF inducers, mainly hydroxyurea, have already been recommended in guidelines (TIF 2018b). It remains uncertain what their impact might have on healthcare resources, such as cost of treatment, human resources and blood bank reserves. It is possible that this intervention could completely change how this unique population is managed. A systematic assessment of the benefits and harms is essential to provide appropriate recommendations for people with NTDβT.

Objectives

To compare the effectiveness and safety of the following for reducing blood transfusion for people with NTDβT:

1. HbF inducers versus usual care or placebo;

2. a single HbF inducer with another HbF inducer, and single dose with another dose; and

3. combination HbF inducers versus usual care or placebo, or a single HbF inducer.

Methods

Criteria for considering studies for this review

Types of studies

Randomised controlled trials (RCTs) and quasi‐RCTs.

Types of participants

People of any age who have been defined as having NTDβT, either β‐thalassaemia intermedia or Hb E/β‐thalassaemia.

Since there is a wide range of transfusion‐dependency ranging from occasional to regular transfusion during certain circumstances, NTDβT is difficult to define. For this review, we considered NTD to be present when an individual had one of the above‐mentioned conditions and transfusion was based on circumstances such as poor growth, excessive extramedullary haemopoiesis, severe infection and pregnancy.

We described the ages of the included participants.

Types of interventions

1. A single HbF inducer at any dose or duration compared with usual care^a or placebo.

2. A single HbF inducer at any dose or duration compared with another HbF inducer.

3. A single dose or dose regimen of an HbF inducer compared with another dose or dose regimen of the same HbF inducer.

4. Two or more HbF inducers at any dose or duration compared with usual care, placebo or a single HbF inducer.

^aWe defined usual care in this review as a standard management protocol that might include regular surveillance to identify chronic anaemia or when blood transfusion might be needed.

We may need to revise the types of interventions included in the future updates of this Cochrane Review as more information about the precise mechanism of action of these substances becomes better known. Specifically, we did not include interventions that would normally be classified as gene therapy. We also excluded treatment modalities such as splenectomy, stem cell therapy, erythropoiesis enhancers and the use of other blood products.

We excluded studies where the intervention period was too short to have a measurable effect, which may have varied between different HbF inducers depending on their mode of action (Description of the intervention).

Types of outcome measures

We planned to assess the following outcome measures.

Primary outcomes

1. Blood transfusion

   1. frequency of transfusion measured by transfused volume per year (if there are insufficient data to calculate this, we planned to report this as units transfused per kg per year or other similar measures)

   2. transfusion‐free interval (in months)

2. Haemoglobin level (in grams per decilitre or equivalent at intervals after commencing the intervention, e.g. every three to six months)

Secondary outcomes

1. HbF level (%) at intervals after commencing the intervention (e.g. every three to six months)

2. Long‐term sequelae of NTDβT (at the longest time point reported by study authors)

   1. growth

   2. puberty

   3. maxillary hyperplasia

   4. severity of extramedullary haematopoiesis (liver and spleen size)

   5. pulmonary hypertension

3. Quality of life (QoL) (physical, psychological, social relationship and environmental (e.g. the World Health Organization Quality of Life (WHOQOL) (WHO 2012)) (at the longest time point)

4. Adverse outcomes (reported at any time during the study)

   1. iron overload (such as by serum ferritin measurement)

   2. need for iron chelation (at any time during the study)

   3. adverse drug effects

Search methods for identification of studies

We searched for all relevant published and unpublished studies without restrictions on language, year or publication status.

Electronic searches

The Cochrane Cystic Fibrosis and Genetic Disorders Group's Information Specialist conducted a systematic search of the Group's Haemoglobinopathies Trials Register for relevant studies using the following terms: ((thalassaemia:kw) OR ((haemoglobinopathies AND general):kw)) AND ((foetal haemoglobin:kw) OR (hydroxurea OR resveratrol* OR YiSui* OR bergamot OR fructus* OR Valproic acid:kw) OR (((fetal OR foetal) AND (haemoglobin OR hemoglobin)) OR hbf:ti,ab,em,mh)).

The Haemoglobinopathies Trials Register is compiled from electronic searches of the Cochrane Central Register of Controlled Trials (CENTRAL) (updated each new issue of the Cochrane Library) and weekly searches of MEDLINE. Unpublished work is identified by searching the abstract books of five major conferences: the European Haematology Association conference; the American Society of Hematology conference; the British Society for Haematology Annual Scientific Meeting; the Caribbean Public Health Agency Annual Scientific Meeting (formerly known as the Caribbean Health Research Council Meetings) and the National Sickle Cell Disease Program Annual Meeting. For full details of all searching activities for the register, please see the relevant section of the Cochrane Cystic Fibrosis and Genetic Disorders Group's website (cfgd.cochrane.org/our-specialised-trials-registers).

Date of most recent search: 16 September 2022.

We also searched the following databases and study registries:

1. CENTRAL in the Cochrane Library (searched 21 August 2022);

2. PubMed (www.ncbi.nlm.nih.gov/pubmed/; 1946 to 21 August 2022);

3. Embase Ovid (1974 to 21 August 2022);

4. US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (www.clinicaltrials.gov; searched 21 August 2022);

5. World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) (trialsearch.who.int/; searched 21 August 2022).

For details of our search strategies, see Appendix 1.

Searching other resources

We checked the bibliographies of included studies and any relevant systematic reviews identified for further references to relevant trials. We contacted the authors of included trials in an attempt to identify further trials they may be involved in.

We were made aware of four Chinese language databases (China Network Knowledge Infrastructure (CNKI) (www.cnki.net/); Chinese Scientific Journals Database (VIP) (www.cqvip.com/); Wan Fang data (www.wanfangdata.com.cn/index.html); and SinoMed (www.sinomed.ac.cn/)). We could access three of the databases (CNKI, VIP, Wan Fang), but not SinoMed. We searched these databases using simple keywords 地中海贫血症 (thalassaemia), HbF 诱导剂 (HbF inducer), 随机对照试验 (RCT).

Date of most recent search: 21 August 2022.

Data collection and analysis

Selection of studies

Three review authors (WCF, CKL, DL) independently screened the results (titles and abstracts) of the literature search for potentially relevant studies. We retrieved full reports of the potentially relevant studies and independently determined if they meet the review's predefined inclusion criteria using a pretested eligibility form (Lefebvre 2022). We resolved discrepancies through discussion, and on several occasions consulted a fourth review author (JJH). We also consulted an editor within the Cochrane Cystic Fibrosis and Genetic Disorders Group where necessary. We contacted study authors and co‐authors for further information when a study eligibility was unclear. We listed all excluded studies and documented the reason for excluding them in the Characteristics of excluded studies table (Li 2022; Review Manager 2014).

Data extraction and management

We extracted data from the included studies onto a specially designed data collection form. Three review authors (WCF, CKL and DL) independently collected study details and outcome data. We then entered the information into the agreed form, compared results and resolved any disagreements by discussion, and if required by consultation with a fourth review author (JJH). We extracted the following details:

1. general information (title, authors, source, country, year of publication, setting);

2. study characteristics (design, risk of bias);

3. participant characteristics (inclusion and exclusion criteria, type of NTDβT, severity of anaemia, indication for HbF inducer, ongoing usual care including the indication for blood transfusion, sample size, losses to follow‐up, baseline characteristics such as age, ethnic group, pre‐existing comorbidities, duration of transfusion, prior splenectomy and iron chelation);

4. intervention (data regarding the usage of HbF inducers, e.g. amount, frequency of delivery, timing from diagnosis, length of treatment, participant responsiveness);

5. comparison (data regarding usual care with or without a placebo; in studies using placebo, we collected data on amount and frequency of delivery);

6. other outcomes as specified under Types of outcome measures; and

7. additional data (adverse events, QoL measurements and outcomes not described under Types of outcome measures); for each study, we extracted data on adverse drug effects prespecified in the study protocol or methods and adverse drug effects reported in the results or discussion.

Where necessary we contacted the original investigators in an effort to obtain any data that needed further clarification or was not reported.

We planned to group outcome data into specific time periods depending on what was reported. We also grouped studies using different doses of the same HbF inducer or different combinations of HbF inducers accordingly.

For multiple‐arm studies, we only included the relevant arms that involved an HbF inducer as part of the intervention in one or more groups (Karimi 2010a). For the three‐arm study comparing two different HbF inducers to placebo, we included both arms in a single analysis by dividing the placebo group between the two intervention groups (Lu 2012). For the three‐arm study comparing two different HbF inducers to a combined HbF inducer, we included both arms in a single analysis by dividing the combined HbF inducer group between the HbF intervention groups (Haghpanah 2018).

Assessment of risk of bias in included studies

Three review authors (WCF, CKL, DL) independently assessed the risk of bias of each included study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2017). We resolved any disagreements by discussion with a fourth review author (JJH). Where there was inadequate information to make a judgement, we attempted to correspond with the study authors.

For each study included, we assessed the following criteria.

Domain	Assessment method	Judgement
Sequence generation	We described the methods used to generate the allocation sequence in terms of sufficient details and appropriateness.	Low risk of bias (any truly random process, e.g. random number table; computer random number generator) High risk of bias (any non‐random process, e.g. odd or even date of birth; hospital or clinic record number) Unclear risk of bias
Allocation concealment	We described the methods used to conceal allocation to interventions prior to assignment to judge whether the intervention allocation could have been foreseen in advance of, during recruitment or changed after assignment.	Low risk of bias (e.g. telephone or central randomisation; consecutively numbered sealed opaque envelopes) High risk of bias (open random allocation; unsealed or non‐opaque envelopes, alternation; date of birth) Unclear risk of bias
Blinding to treatment (of clinician (person delivering the treatment), participant and outcome assessors)	We described the methods used, if any, to blind participants and personnel from knowledge of which a participant received. We made a judgement on whether the blinding was adequate. We judged each outcome to be either subjective or objective based on whether we considered lack of blinding could affect the measurement of the outcome. We considered that studies were at low risk of bias if they were blinded, or if we judged that the lack of blinding would be unlikely to affect the results. We assessed blinding separately for objective and subjective outcomes.	Low, high or unclear risk of bias for participants Low, high or unclear risk of bias for personnel Low, high or unclear risk of bias for outcome assessors
Completeness of the outcome data (possible attrition bias through withdrawals, loss to follow‐up and protocol violations)	We described whether there were significant missing outcome data based on the number of participants recruited or randomised and the number analysed for each outcome at the end of the study period. If the percentage loss to follow‐up exceeded 10%, we made a decision to use an appropriate sensitivity analysis.	Low risk of bias (e.g. no missing outcome data; missing outcome data balanced across groups) High risk of bias (e.g. numbers or reasons for missing data imbalanced across groups; 'as treated' analysis done with substantial departure of intervention received from that assigned at randomisation) Unclear risk of bias
Selective reporting bias (whether all prespecified outcomes and all expected outcomes of interest to the review were reported)	We attempted to describe how we investigated the possibility of selective outcome reporting bias and what we found including non‐significant results that were not reported adequately or missed important outcome variables.	Low risk of bias (where it was clear that all the study's prespecified outcomes and all expected outcomes of interest to the review had been reported) High risk of bias (where not all the study's prespecified outcomes had been reported; ≥ 1 reported primary outcomes were not prespecified; outcomes of interest were reported incompletely and so could not be used; study failed to include results of a key outcome that would have been expected to have been reported) Unclear risk of bias
Other sources of bias	We described any important concerns we had about other possible sources of bias such as baseline discrepancies and the possibility of skewed data. For sequential dose‐escalation or cross‐over studies, we assessed the risk of bias related to the washout period by taking into consideration the intervention's half‐life, any possibility of carry‐over effects and the potential life span of thalassaemia erythrocytes.	Low risk of other bias High risk of other bias Unclear whether there is risk of other bias

We made a judgement on the overall risk of bias assessment (based on the items above). We made explicit judgements whether studies were at high risk of bias, according to the criteria given in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2017). With reference to the domains we assessed, we judged the likely magnitude and direction of the bias and whether we considered it as likely to impact the findings. We explored the impact of the level of bias through undertaking sensitivity analysis. We used the overall risk of bias as part of grading the certainty of evidence for the summary of findings tables.

Measures of treatment effect

We would have reported frequency of transfusion in millilitres per year, as a mean difference (MD) or if reported as a rate, as a rate ratio. We would have analysed a further primary outcome 'transfusion‐free interval' as time‐to‐event data and calculated the hazard ratio (HR).

For dichotomous data, we presented the results as a summary risk ratio (RR).

For continuous outcomes (all other outcomes listed above), where outcomes were measured in the same way between studies, we reported the mean relative change from baseline or the mean postintervention value. If we had encountered outcomes reported on different scales, such as QoL scales, we would have made a judgement whether they were reporting the same measure and, therefore, could be reasonably combined. Such data analyses would have been done by using standardised mean difference (SMD).

We reported 95% confidence intervals (CI) for all measures of treatment effect.

Unit of analysis issues

We treated the participant as the unit of analysis (Deeks 2022). We included dose‐escalation studies and would have included cross‐over studies.

For dose‐escalation or cross‐over studies, we examined the washout period and assessed the adequacy and whether there would be any carryover effect.

If we had encountered cluster‐RCTs, we had planned to follow the methods described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022). We would have used an estimate of the intracluster correlation coefficient (ICC) derived from the study (if possible), or from another source.

For multiple‐arm studies (studies using one or more treatment groups), we selected only those arms that were relevant to the review. Where relevant we would have combined groups to create a single pair‐wise comparison using the methods described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2017). When a multiple‐arm study contributed more than one comparison to a particular meta‐analysis, we divided the control group between the two intervention groups.

Dealing with missing data

For missing or unclear data, we attempted to contact the authors. We contacted 12 investigators and two institutions for additional information on all the included studies (Bohara 2014; Fucharoen 2013; Haghpanah 2018; Jain 2019; Jha 2019; Karimi 2010a; Lu 2012), one excluded study (Li 2016), and all five studies awaiting assessment (Cheng 2019; Huang 2016; Jain 2021; NCT04411082; Wu 2007). We tried twice or thrice; only one author from an included study responded, but without fully providing the requested (Fucharoen 2013). If data were not available, we would have described the level of attrition, attempted to explore whether there was any pattern associated with missing data and analyse the available data. Missing summary statistics, such as the standard deviation (SD), would be estimated by conversion of available statistics, such as the standard error (SE). We also attempted to establish whether the primary study used an intention‐to‐treat analysis (i.e. we analysed the participants in the groups to which they were randomised regardless of whether they received the allocated intervention). We tried contacting the study investigators for more information or any full reports if the included study was published in an abstract format or presented at meetings only (Deeks 2022).

We did not attempt to impute missing SDs.

Assessment of heterogeneity

We tested for heterogeneity between studies using the I² statistic and the Chi² test (Deeks 2022; Higgins 2003). We regarded heterogeneity as substantial if an I² value was greater than 80% and either Tau² was greater than zero, or there was a low P value (less than 0.10) in the Chi² test. When we found that heterogeneity was too severe to meaningfully combine the studies, we did not perform a meta‐analysis. We noted a low P value (less than 0.10) in the Chi² test and interpret it in light of the number of included studies. We explored any heterogeneity by subgroup analysis.

Assessment of reporting biases

We attempted to minimise publication and reporting bias by conducting a comprehensive search. If we had obtained a sufficient numbers of studies (at least 10), we planned to assess publication bias by constructing and reviewing the symmetry of a funnel plot. If we had found any funnel plot asymmetry, we would have considered whether this could have been due to publication bias (Page 2022). Where reports were available, we compared the study protocols or methods sections with results presented in final study reports for any selective outcome reporting.

Data synthesis

Where studies were clinically and methodologically comparable, we performed meta‐analysis using a fixed‐effect model. We would have analysed rate ratio or time‐to‐event data using a generic inverse variance meta‐analysis. If no meta‐analysis was possible, or if we encountered substantial heterogeneity (as defined in Assessment of heterogeneity), we reported a descriptive qualitative critical appraisal of the outcome information from these included studies (Deeks 2022).

If we had encountered any cross‐over studies, we would have included them and analysed them separately (Elbourne 2002), and followed the methods in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022).

Subgroup analysis and investigation of heterogeneity

We identified substantial clinical heterogeneity in our included studies and investigated it using the subgroup analyses as follows (Deeks 2022). Where meaningful, we carried out subgroup analysis for different types and combinations of HbF inducers. However, we were unable to conduct our other two preplanned subgroup analyses (type of NTDβT and baseline HbF proportion) in this review due to a lack of data. We plan to include these analyses in future updates of the review if the necessary data become available.

Sensitivity analysis

We planned to conduct sensitivity analyses to assess the robustness of our review results by repeating the analysis with the following adjustments, exclusion of studies with:

1. unclear or high risk of bias regarding allocation concealment;

2. unclear or high risk of bias regarding blinding of outcomes assessment;

3. unclear or high risk of bias regarding completeness of follow‐up;

4. high risk of selective outcome reporting.

We also planned to use a sensitivity analysis to test the robustness of our choice of model used for meta‐analysis (Deeks 2022).

Summary of findings and assessment of the certainty of the evidence

We used GRADE to reflect the certainty of evidence (Schünemann 2022a) and used GRADEpro GDT software (GRADEpro GDT) to create a summary of findings table for each comparison we presented, which had the following components:

1. summarised key findings (participants, comparison and baseline information, outcome) (Schünemann 2022b);

2. summarised statistical results; and

3. summarised certainty of evidence, magnitude of the effect including the source of any external information used in the 'Assumed risk' column or any departures from the standard methods, and reasons behind the decisions.

We included the following outcomes in our summary of finding stables (Schünemann 2022b).

1. Frequency of transfusion

2. Transfusion‐free interval (post hoc change)

3. Haemoglobin level (post hoc change)

4. HbF level

5. QoL

6. Adverse events (iron overload, need for iron chelation, severe immunosuppression (e.g. requiring hospitalisation or cessation of the intervention or death)

Results

Description of studies

For further details, see Characteristics of included studies; Characteristics of excluded studies; and Characteristics of studies awaiting classification tables.

Results of the search

Our search retrieved 1434 records of which 609 records were duplicates (Figure 1). We excluded 795 records by title alone because none of them fit into our inclusion criteria, and assessed 29 records (21 studies) for eligibility. We proceeded to look for confirmatory evidence from the literature if the intervention had any HbF inducing properties: hydroxyurea (Baliga 2000; Rodriguez 1998), HQK‐1001 (Perrine 2011), thalidomide (Aerbajinai 2007), decitabine (5‐aza‐2'‐deoxycytidine) (Perrine 1989; Kalantri 2018), l‐carnitine (Pekala 2011; Perrine 1989), Radix Astragali (the root of Astragalus propinquus) (Lu 2016), resveratrol (DeSimone 2002), and YSSXG (Fang 2007). We finally included seven studies (13 records) in the review and excluded nine studies (10 records). We found five studies (six records) that we have listed as awaiting classification since we could not reach the authors for further clarification (Jain 2021; NCT04411082) or had other concerns (Cheng 2019; Wu 2007). One other study, Huang 2016 replied that they would provide further information in the future.

1.

Included studies

We included seven studies (13 records) in this review (see Characteristics of included studies table). An overview of the main clinical characteristics of the included studies by type of HbF inducer is found in the table (Table 6).

2. Overview of included studies by foetal haemoglobin inducers.

Intervention	Study ID	No. of participants	Age at enrolment (years)	No. of splenectomised participants	Dose and duration
Single HbF inducer at any dose or duration compared with usual care or placebo
HQK‐1001 (oral tablet)	Fucharoen 2013	21	Range 17–45	16	Sequential dose‐escalation from 10 mg/kg/day then to 30 mg/kg/day from 20 mg/kg/day then to 40 mg/kg/day 56 intervention days with in between 8 weeks intervention‐free period
Radix Astragali (tea bag – oral drink)	Lu 2012	24	Mean 6.5, SD 3.6	Not reported	Dose: according to age Duration: 3 months
CNPa (tea bag – oral drink)	Lu 2012	22	Mean 6.5, SD 3.6	Not reported	Dose: according to age Duration: 3 months
Single HbF inducer at any dose or duration compared with another HbF inducer
Hydroxyurea vs decitabine (oral HUb vs subcutaneous)	Jha 2019	30	> 12	Not reported	Daily HUb Twice a week decitabine (no other details) Duration: 3 months
Hydroxyurea vs thalidomide (appearance not stated. Both inducers were taken orally)	Jain 2019	30	Range 6–45	Not reported	Hydroxyurea: 10 mg/kg/day Thalidomide: fixed‐dose 50 mg/day Duration: 6 months
Hydroxyurea vs resveratrolc (HUb vs capsule)	Haghpanah 2018	34	Range 18–42	Not reported	Hydroxyurea: 8–12 mg/kg/day Resveratrol: 4 capsules daily Duration: 6 months
Radix Astragali vs CNPa (both in similar looking tea bag – oral drinks)	Lu 2012	24	Mean 6.5, SD 3.6	Not reported	Dose: according to age Duration: 3 months
1 dose or dose regimen of an HbF inducer compared with another dose or dose regimen of the same HbF inducer
Hydroxyurea vs hydroxyurea (tablets)	Bohara 2014	61	Mean 16.68, SD 6.05	Not reported	20 mg/kg/day vs 10 mg/kg/day Duration: 6 months
≥ 2 HbF inducers at any dose or duration compared with usual care, placebo or a single HbF inducer
Combined resveratrolc and hydroxyurea vs hydroxyurea alone	Haghpanah 2018	36	Range 18–42	Not reported	Hydroxyurea: 8–12 mg/kg/day Resveratrol: 4 capsules daily Duration: 6 months
Combined resveratrol and hydroxyurea vs hydroxyurea alone (Resveratrol capsules, HUb)	Haghpanah 2018	38	Range 18–42	Not reported	Hydroxyurea: 8–12 mg/kg/day Resveratrol: 4 capsules daily Duration: 6 months
Combined l‐carnitine and hydroxyurea vs hydroxyurea alone (L‐carnitine tablets, hydroxyurea capsules)	Karimi 2010a	50	Range 4–35	60/120 participants (inclusive of groups on magnesium chloride)	Hydroxyurea: 8–12 mg/kg/day L‐carnitine: 50 mg/kg/day Duration: 6 months

^aCNP (combined natural preparation): combination of Radix Astragali (黄芪), Codonopsis pilosula (党参) and tortoise plastron (龟板).
^bHU: unsure if the hydroxyurea was in the form of a capsule or tablet, but it was taken orally.
^cResveratrol: came in a form of a capsule that contained micronised trans‐resveratrol 500 mg and piperine 10 mg. Piperine was added to the capsule to increase the oral bioavailability of resveratrol.

Design

Of the seven included RCTs, one was a four‐arm study (Karimi 2010a), two were three‐arm studies (Haghpanah 2018; Lu 2012), and three were two‐arm studies (Bohara 2014; Jain 2019; Jha 2019). These studies were all of parallel design. The final study was a dose‐escalation RCT, where two cycles compared four different doses of an HbF inducer with placebo (Fucharoen 2013). Each cycle lasted 12 weeks (eight weeks of the intervention and four weeks of follow‐up), with a washout period of at least four weeks; participants were rerandomised for the second cycle.

We did not identify any cross‐over studies. Two studies were phase 2 trials (Fucharoen 2013; Karimi 2010a), one was a phase 3 trial (Haghpanah 2018), but we are unsure about the remaining four studies.

Sample sizes

Overall there were 373 participants with NTDβT included in the studies, of whom only 291 received interventions relevant to this review. The sample size of each study ranged from 21 participants (Fucharoen 2013) to 61 participants (Bohara 2014).

Setting

The included studies were mainly conducted in Asia, the Middle East and North Africa. One was from China (Lu 2012), three were from India (Bohara 2014; Jain 2019; Jha 2019), and two were from Iran (Haghpanah 2018; Karimi 2010a). One study had participants from South East Asia (Thailand) and the Middle East (Lebanon) (Fucharoen 2013).

All studies were conducted in upper‐middle income countries except those from India, a lower‐middle income country (World Bank 2021).

Study authors' declarations of interest

Four studies reported no conflict of interest (Bohara 2014; Haghpanah 2018; Jain 2019; Jha 2019). The remaining studies did not state this information (Fucharoen 2013; Karimi 2010a; Lu 2012).

Sources of funding

Three studies received university or government funding (Haghpanah 2018; Karimi 2010a; Lu 2012), and one study was funded by a pharmaceutical company (Fucharoen 2013). The remaining studies did not report the source of funding (Bohara 2014; Jain 2019; Jha 2019).

Study dates

Five studies reported their study dates, which ranged from 2007 to 2017 (Bohara 2014; Fucharoen 2013; Haghpanah 2018; Karimi 2010a; Lu 2012). Two studies did not mention the study dates (Jain 2019; Jha 2019).

Participants

The participants had HbE/β‐thalassaemia with the phenotype of β‐thalassaemia intermedia (Bohara 2014; Jain 2019; Jha 2019), a mixture of HbE/β‐thalassaemia and β⁺/β⁰ thalassaemia (Fucharoen 2013), β‐thalassaemia intermedia only (Haghpanah 2018; Karimi 2010a), and β‐thalassaemia of mild to moderate severity (Lu 2012).

Some participants from two included studies are not considered in this review (Karimi 2010a; Lu 2012). In one study, there were 60 participants in two of the four treatment arms who did not use an HbF inducer (magnesium chloride) and were not relevant to this review (Karimi 2010a). In the second study, 22 participants were transfusion‐dependent (Lu 2012). The studies reported these results separately from the NTDβT participants.

The definitions of NTDβT were not consistent across all studies. Two studies considered participants to have NTDβT if they had a transfusion‐free interval of six months or more (Haghpanah 2018; Karimi 2010a). One study defined NTDβT as at least three months transfusion‐free with the added criteria of no administration of haematinics for the past six months (Lu 2012). One study defined NTDβT as "transfusion independent" despite participants persistently being in an anaemic state and potentially requiring intermittent transfusion during recurrent illnesses such as fever (Bohara 2014). The remaining three studies did not describe the frequency of participants' transfusions and described participants as "non‐transfusion‐dependent thalassaemia" or "not on regular transfusion" (Fucharoen 2013; Jain 2019; Jha 2019).

The age of the participants eligible for recruitment ranged from two years to 49 years. Four studies analysed a mixed population of children and adults with the mean age ranging from 13.76 (SD 9.05) years to 20.15 (SD 4.4) years (Bohara 2014; Jain 2019; Jha 2019; Karimi 2010a); none of these studies reported their results by age. Two studies included only adults whose mean age ranged from 28.2 (SD 5.6) to 33.9 (SD 8.2) years (Fucharoen 2013; Haghpanah 2018); and one study only analysed children aged between two years and 18 years (Lu 2012).

Three studies enrolled some participants who had undergone splenectomy (Fucharoen 2013; Haghpanah 2018; Karimi 2010a), while the other studies did not report this. One study also reviewed the spleen sizes as one of the pre‐ and postintervention outcomes (Karimi 2010a). None of the studies performed a subgroup analysis between splenectomised and non‐splenectomised participants to look for any effect on the postintervention total haemoglobin.

Interventions

The seven included studies tested eight different HbF inducers, four pharmacological and four natural in origin. The four pharmacological HbF inducers were hydroxyurea, HQK‐1001, thalidomide and decitabine. The four natural HbF inducers were resveratrol, l‐carnitine, Radix Astragali (黄芪) and a 3‐in‐1 herbal‐animal mixture that comprised Radix Astragali (黄芪), Codonopsis pilosula (党参) and tortoise plastron (龟板) in a tea bag. We referred to the 3‐in‐1 mixture as 'combined natural preparation' (CNP) in this review. Piperine capsules were to be taken together with resveratrol to increase the oral bioavailability of resveratrol. It is a natural substance with no HbF inducer properties or any effects on red cell stabilisation and production.

Decitabine was given subcutaneously. Studies did not describe the appearances of the HQK‐1001 and thalidomide, but they were taken orally. The remaining inducers were taken orally, as tablets, capsules or a drink after immersing a tea bag that contained a mixture of natural ingredients in warm water.

Comparisons

Single foetal haemoglobin inducer versus placebo or usual care

Two studies (one of which had three arms) compared an HbF inducer against placebo or usual care (Fucharoen 2013; Lu 2012). The three‐arm study compared two different HbF inducers to placebo; we included both active arms in a single analysis by dividing the placebo group between the two intervention groups (Lu 2012). The comparisons were:

1. HQK‐1001 at four different doses (10 mg/kg, 20 mg/kg, 30 mg/kg, 40 mg/kg) versus placebo (Fucharoen 2013);

2. Radix Astragali versus placebo (Lu 2012); and

3. CNP versus placebo (Lu 2012).

One study stated that the placebo was made to resemble the intervention in a tea bag (Lu 2012), but the other study did not describe the placebo (Fucharoen 2013).

Single foetal haemoglobin inducer versus another foetal haemoglobin inducer

Four studies compared a single HbF against a different HbF (Haghpanah 2018; Jain 2019; Jha 2019; Lu 2012). The comparisons were:

1. hydroxyurea versus resveratrol (Haghpanah 2018);

2. hydroxyurea versus thalidomide (Jain 2019);

3. hydroxyurea versus decitabine (Jha 2019); and

4. Radix Astragali versus CNP (Lu 2012).

The study comparing hydroxyurea to resveratrol had three arms and compared hydroxyurea plus placebo versus resveratrol plus placebo versus hydroxyurea plus resveratrol. This study was also eligible for the 'two or more HbF inducers versus a single HbF inducer, placebo or usual care' comparison (Haghpanah 2018).

Comparing two different doses of the same foetal haemoglobin inducer

Two studies compared two or more different doses of the same HbF inducer (Bohara 2014; Fucharoen 2013).

The first study compared hydroxyurea 20 mg/kg/day versus hydroxyurea 10 mg/kg/day (Bohara 2014). For this review, we considered the 20 mg/kg/day dose of hydroxyurea to be the intervention and the 10 mg/kg/day dose as the comparison (Bohara 2014).

The second study compared four different doses of HQK‐1001 against each other: 10 mg/kg, 20 mg/kg, 30 mg/kg, 40 mg/kg (Fucharoen 2013).

Two or more foetal haemoglobin inducers versus a single foetal haemoglobin inducer, placebo or usual care

Two studies compared a combination of HbF inducers against a single HbF inducer (Haghpanah 2018; Karimi 2010a). The comparisons were:

1. hydroxyurea plus resveratrol versus hydroxyurea plus placebo (Haghpanah 2018);

2. hydroxyurea plus resveratrol versus resveratrol plus placebo (Haghpanah 2018); and

3. hydroxyurea plus l‐carnitine versus hydroxyurea alone (Karimi 2010a).

Two studies compared a combination of two HbF inducers with a single HbF inducer (Haghpanah 2018; Karimi 2010a). We included the single HbF inducer intervention arms of Haghpanah 2018 in a single analysis by dividing the participants in the third arm, a combination treatment, between the two groups. In this study, the participants in resveratrol group were also given piperine capsules to increase the oral bioavailability of resveratrol (Haghpanah 2018). The second study compared hydroxyurea plus l‐carnitine to hydroxyurea (Karimi 2010a). This study did not use an HbF inducer in two out of its four arms (instead it used magnesium chloride), but given the potential effects of magnesium chloride on haemoglobin levels, we have only included the two arms using HbF inducer(s) in this review (Karimi 2010a). Magnesium chloride may potentially sustain higher haemoglobin levels by stabilising the RBC membranes (Wagner 2002); and since haemoglobin levels are used as the main determinant for a blood transfusion, we decided to only include the two study arms that used either hydroxyurea alone or a combination of hydroxyurea and l‐carnitine.

Duration of intervention

The duration of the intervention ranged from two months (Fucharoen 2013) to six months (Bohara 2014; Haghpanah 2018; Jain 2019; Jha 2019; Karimi 2010a).

Outcomes

Three included studies had published protocols (Fucharoen 2013; Haghpanah 2018; Karimi 2010a).

No study reported on our primary outcome of frequency of blood transfusion or the transfusion‐free interval; however one study reported the transfusion units per participant per year, but we were unable to calculate a rate ratio (Jain 2019). Three studies reported the number of participants who required a transfusion during the study (Bohara 2014; Haghpanah 2018; Karimi 2010a).

Four studies reported the mean haemoglobin levels before and after the intervention period over varying time points from six weeks up to six months (Bohara 2014; Haghpanah 2018; Karimi 2010a; Lu 2012). One study reported the median changes in total haemoglobin for at least eight weeks (Fucharoen 2013). One study reported the peak mean increment in total haemoglobin, increment range and % of participants with a rise in haemoglobin greater than 1.0 g/dL from baseline (Jha 2019). The final study only reported the baseline median haemoglobin level and the number of participants who experienced a change from baseline (responders) (Jain 2019).

Two studies measured and reported absolute % HbF at specific time points (Bohara 2014; Lu 2012); one of these reported both HbF in grams per decilitre and % (Bohara 2014). One study reported % HbF as the change from baseline to the end of the intervention (Fucharoen 2013), and two studies reported it as mean increment from baseline to end of therapy (Jain 2019; Jha 2019). One study reported mean % HbF at baseline, but not for any other time points (Karimi 2010a). One study did not report % HbF (Haghpanah 2018).

No studies reported any long‐term sequelae of NTDβT such as growth, puberty and maxillary hyperplasia; although one study did measure pulmonary hypertension and spleen size before and after the intervention (Karimi 2010a).

No studies used a validated QoL measuring tool; however, one study measured improvement in energy level and sense of well‐being (Bohara 2014).

No studies reported the adverse outcomes of iron overload or the need for iron chelation. All seven studies reported adverse drug effects (Bohara 2014; Fucharoen 2013; Haghpanah 2018; Jain 2019; Jha 2019; Karimi 2010a; Lu 2012). Two studies reported that they would look for adverse effects, but did not specify any in particular (Haghpanah 2018; Jain 2019). Five studies prespecified the adverse effects they would measure, but none of these studies included any of our prespecified adverse effects. One study measured neutropenia, thrombocytopenia, gastrointestinal disturbances and raised liver enzymes (Bohara 2014). One study used the National Cancer Institute Common Terminology Criteria for Adverse Events (NCI CTCAE) scoring to assess the severity of adverse events (Fucharoen 2013). One study reported infections, cytopenia and deranged biochemical parameters (Jha 2019). One study reported adverse effects of the drug on the liver and kidneys (Karimi 2010a), and one study measured white cell counts, neutrophil counts and platelet counts, liver function tests (alanine transaminase (ALT), aspartate aminotransferase (AST), gamma‐glutamyl transferase (GGT)) and renal profiles (urea, sodium, potassium, creatinine) (Lu 2012).

More details are presented in Table 2.

Excluded studies

We excluded nine studies (10 records) (Biswas 2014; Chai 2005; Cheng 2016; Elalfy 2013; Fang 2007; Guo 2014; Li 2016; Loukopoulos 1998; Patthamalai 2014).

Four studies (four records) were not RCTs (Chai 2005; Cheng 2016; Loukopoulos 1998; Patthamalai 2014), and one study (one record) was withdrawn after publication (Guo 2014).

The excluded studies used either intervention that are considered RBC stabilisers (magnesium chloride) or RBC enhancers without any effect on HbF production (erythropoietin, colla corii asini). We excluded one RCT (one record) that included a co‐intervention (human erythropoietin), which we did not consider an HbF inducer (Elalfy 2013). We excluded one RCT (two records) that used colla corii asini, a bone marrow erythrocyte progenitor cells activator that improves anaemia but reduces HbF and HbA₂ levels (Li 2016).

The remaining two excluded studies enrolled participants with transfusion‐dependent HbE/β‐thalassaemia or β‐thalassaemia major, but used HbF inducers (hydroxyurea, Radix Astragali, YSSXG) as their interventions (Biswas 2014; Fang 2007).

The intervention used in the attempt to increase the haemoglobin levels or reduce the severity of anaemia differed between the studies. Details of these excluded trials can be found in the Characteristics of excluded studies table and Appendix 2.

Ongoing studies

We did not identify any ongoing studies.

Studies awaiting classification

Five studies are awaiting classification (Cheng 2019; Huang 2016; Jain 2021; NCT04411082; Wu 2007). We are waiting further information from the authors of two of these (Jain 2021; NCT04411082). We are currently waiting for further information from the authors of one study (Huang 2016). We are currently attempting to verify the data from the remaining two studies (Cheng 2019; Wu 2007). Further details of these studies can be found in the Characteristics of studies awaiting classification table.

Risk of bias in included studies

Details of the risk of bias assessment for each included study are presented in the Characteristics of included studies table and summary descriptions of the assessments are presented in Figure 2 and Figure 3.

2.

3.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation

Sequence generation

We judged four studies at low risk of bias for sequence generation because they described a random component in the sequence generation process such as using random number table, using computer‐generated random numbers and random numbers generated via a numbering software (Bohara 2014; Fucharoen 2013; Haghpanah 2018; Lu 2012). We judged three studies at unclear risk of bias for sequence generation because they did not adequately describe how sequence generation was accomplished (Jain 2019; Jha 2019; Karimi 2010a).

Allocation concealment

We judged all studies at unclear risk of bias for allocation concealment. None reported whether the allocation was concealed or how it was done (Bohara 2014; Fucharoen 2013; Haghpanah 2018; Jain 2019; Jha 2019; Karimi 2010a; Lu 2012).

Blinding

Performance bias (blinding of participants and personnel)

We judged three studies at high risk of performance bias (Jain 2019; Jha 2019; Karimi 2010a). Jain 2019 provided no information on how hydroxyurea and thalidomide were designed to be as similar as possible in terms of appearance, smell, taste and frequency of administration. In Jha 2019, the intervention and comparator were distinguishable by mode of administration and possibly appearance; hydroxyurea was given orally, but decitabine was given subcutaneously. In Karimi 2010a, there were differences in the frequency of administration and appearance; hydroxyurea was given once a day while l‐carnitine was given more than once a day.

We judged three studies at unclear risk of performance bias (Bohara 2014; Haghpanah 2018; Lu 2012). Bohara 2014 did not report the blinding of the participants and personnel for all outcomes. Haghpanah 2018 did not describe the placebo. In Lu 2012, it may have been possible to distinguish the intervention from placebo through taste and smell, although both were made to look similar and were administered similarly.

We judged the remaining study at low risk of performance bias as the study used a matching placebo to ensure proper blinding of participants and personnel (Fucharoen 2013).

Detection bias (blinding of outcome assessments)

We assessed detection bias separately for objective and subjective outcomes. We judged haemoglobin and HbF levels to be objective and unlikely to be affected by lack of blinding. We judged blood transfusion, long‐term sequelae, QoL and adverse events to be subjective and hence at high risk of detection bias where there was a lack of blinding.

Objective outcomes

We judged all studies to have low risk of detection bias for haemoglobin levels and HbF because these outcomes were laboratory assessments, judged not to influence response to interventions (Bohara 2014; Fucharoen 2013; Haghpanah 2018; Jain 2019; Jha 2019; Karimi 2010a; Lu 2012).

Subjective outcomes

Four studies reported on blood transfusion. We judged two studies to have high risk of detection bias because the blinding was not possible (Jain 2019; Karimi 2010a), and two studies to have an unclear risk because the blinding of the outcome assessors was either not reported or could not be determined (Bohara 2014; Haghpanah 2018).

Only one study reported long‐term sequelae and was at high risk of bias because blinding was not possible and the measurements (splenic size and pulmonary echocardiogram findings) were operator‐dependent (Karimi 2010a).

Only one study assessed QoL and was at unclear risk of bias because it was unclear whether the participants were blinded, and they were the ones who reported any changes in energy level and sense of well‐being (Bohara 2014).

For adverse events, we judged one study to have low risk of detection bias because the adverse events reported were based on laboratory measurements (Lu 2012). We judged three studies at high risk of detection bias because blinding was not possible and the participants were the ones who reported the adverse events (Jain 2019; Jha 2019; Karimi 2010a). We judged three studies at unclear risk of detection bias because the blinding of outcome assessors was unclear and the adverse events reported depended on the clinical judgement or participants' perceptions, or both (Bohara 2014; Fucharoen 2013; Haghpanah 2018).

Incomplete outcome data

We judged four studies to have low risk of attrition bias because all enrolled participants including dropouts (if any) were analysed at the end of the study and they were analysed in the group to which they were assigned (Bohara 2014; Fucharoen 2013; Haghpanah 2018; Lu 2012). We judged one study at high risk of attrition bias because it did not analyse all the recruited participants (Karimi 2010a). We judged two studies at unclear risk of attrition bias because they were abstracts and there were no details on the numbers of participants enrolled or who withdrew (Jain 2019; Jha 2019).

Selective reporting

We judged three studies at low risk of reporting bias because they reported all expected outcomes (Fucharoen 2013; Haghpanah 2018; Lu 2012). We judged one study at high risk of reporting bias because the methods section reported participants to have undergone an interview regarding their sense of well‐being but did not report the result of this interview (Bohara 2014). We judged three studies at unclear risk of reporting bias (Jain 2019; Jha 2019; Karimi 2010a). Two studies were abstracts with limited information (Jain 2019; Jha 2019), and one study did not clearly state the time point of measurement for HbF level and reported only baseline HbF (Karimi 2010a).

Other potential sources of bias

We judged six studies at low risk of other bias (Bohara 2014; Haghpanah 2018; Jain 2019; Jha 2019; Karimi 2010a; Lu 2012). We judged one study to have unclear risk of bias because we were unclear about the washout period (Fucharoen 2013).

Effects of interventions

See: Table 1; Table 3; Table 4; Table 5

A single foetal haemoglobin inducer versus usual care or placebo

See Table 1.

Two studies compared an HbF inducer with placebo (Fucharoen 2013; Lu 2012). One assessed HQK‐1001 given by sequential dose‐escalation with a washout period between doses (Fucharoen 2013); one compared Radix Astragali versus CNP versus placebo (Lu 2012).

Primary outcomes

1. Blood transfusion

Neither study reported any results for frequency of transfusion measured by transfused volume per year or transfusion‐free interval (Fucharoen 2013; Lu 2012).

2. Haemoglobin

One study reported haemoglobin at three months postintervention (Lu 2012). Since this was a three‐arm study where two different types of HbF inducers were compared against a placebo control group, we made a judgement to split the controls into almost equal numbers when we meta‐analysed the three arms (Lu 2012). The point estimates for the two different HbF inducers compared to placebo favoured HbF inducers (MD 1.33 g/dL, 95% CI 0.54 to 2.11; 1 study, 35 participants; very low‐certainty evidence; Analysis 1.1). We downgraded the certainty of the evidence by one level for indirectness because the follow‐up duration was short and insufficient to identify potential changes and two levels for imprecision because the conclusion was based on a very small sample size (Table 1).

1.1. Analysis.

Comparison 1: Single HbF inducer at any dose or duration compared with usual care or placebo, Outcome 1: Mean haemoglobin (g/dL) at 3 months postintervention, subgroup by types of HbF inducer

One study (21 participants) comparing four different doses of HQK‐1001 with placebo reported that there was no difference in median haemoglobin for any dose compared with placebo at 12 weeks (Fucharoen 2013).

Secondary outcomes

1. Foetal haemoglobin (%)

One study reported HbF at three months postintervention (Lu 2012). For the meta‐analysis we split the number of participants in the control group into almost equal portions, one for each intervention. There may be an increase in HbF for both HbF inducers compared with placebo (MD 12.00%, 95% CI −0.74% to 24.75%; 1 study, 35 participants; very low‐certainty evidence; Analysis 1.2). We downgraded the certainty of the evidence one level for indirectness because the follow‐up duration was short and insufficient to identify potential changes and two levels for imprecision because optimal sample size was not achieved and there were wide CIs that included both appreciable benefit and appreciable harm.

1.2. Analysis.

Comparison 1: Single HbF inducer at any dose or duration compared with usual care or placebo, Outcome 2: Mean HbF (%) at 3 months, subgroup by types of HbF inducer

In the second study, investigators reported an increase in median HbF at 12 weeks only with one of the four doses, the HQK‐1001 20 mg/kg/dose, when compared with placebo; data were presented without a measure of dispersion (Fucharoen 2013).

2. Long‐term sequelae of non‐transfusion‐dependent β‐thalassaemia

Neither study reported long‐term sequelae of NTDβT in terms of changes to growth, puberty, extramedullary haematopoiesis and maxillary hyperplasia.

3. Quality of life

Neither study reported QoL.

4. Adverse outcomes

a. Iron overload

Neither study reported QoL.

b. Need for iron chelation

Neither study reported the need for iron chelation.

c. Adverse drug effects

Table 2 provides an overall summary of the prespecified adverse effects, reported adverse effects and adverse effects that were not reported in the included studies.

In one study, the five most common adverse drug effects reported in the HQK‐1001 10 mg/kg group were upper respiratory tract infection, headache, fatigue, nausea and dizziness; other effects reported were fever, palpitations, suprapubic pain, gastritis, gastroenteritis, upper abdominal pain, back pain and severe upper respiratory tract infection (Fucharoen 2013). Nausea, vomiting and abdominal pain occurred more frequently among the HQK‐1001 20 mg/kg and 40 mg/kg groups, and fatigue occurred more frequently among the HQK‐1001 40 mg/kg group. There were severe adverse effects, namely upper abdominal pain, gastritis, suprapubic pain and back pain, in each dose of HQK‐1001. Headache and gastroenteritis occurred in the placebo group. There were too few events to be able to assess whether higher doses were associated with more adverse effects or whether there were more adverse effects in the HQK‐1001 or placebo groups (Fucharoen 2013). More details about the adverse events for each dose of HQK‐1001 are in Table 2.

None of the adverse drug effects prespecified by the second study occurred (abnormalities in blood counts, liver enzyme levels and renal profile) (Lu 2012).

The certainty of evidence was very low for all studies for adverse drug effects. We downgraded one level for indirectness because the follow‐up duration was short and insufficient to see changes and two levels for imprecision because of the very small sample size well below the optimal sample size (Table 1).

A single foetal haemoglobin inducer at any dose compared with another foetal haemoglobin inducer

See Table 3.

Four studies compared a single HbF inducer with another HbF inducer (Haghpanah 2018; Jain 2019; Jha 2019; Lu 2012).

The comparisons were:

1. hydroxyurea versus thalidomide (Jain 2019);

2. hydroxyurea versus decitabine (Jha 2019);

3. Radix Astragali versus CNP (Lu 2012); and

4. hydroxyurea versus resveratrol – participants from both groups also received a placebo (Haghpanah 2018).

Primary outcomes

1. Blood transfusion

None of the studies reported frequency of transfusion in terms of transfused volume per year or transfusion‐free interval.

One study reported that there was a significant reduction in transfusion for thalidomide compared with hydroxyurea (Jain 2019). There was a mean of 3 units/participant/year (range 1 unit/participant/year to 6 units/participant/year) transfused at six months postintervention for the hydroxyurea group (15 participants) but none for the thalidomide group (15 participants). The study did not report SDs and we could not contact the authors for further information. The certainty of evidence was very low; downgraded one level for lack of blinding and two levels for imprecision because conclusions were based on one small study with very little information.

A second study reported three participants required blood transfusion within a six‐month intervention period, one from the resveratrol group and two from the hydroxyurea group (Haghpanah 2018).

2. Haemoglobin

All four studies reported haemoglobin at different time points using different comparisons (Haghpanah 2018; Jain 2019; Jha 2019; Lu 2012).

In one study, at three months mean haemoglobin was lower in the Radix Astragali group compared to the CNP group (MD −0.50 g/dL, 95% CI −1.38 to 0.38; 24 participants; low‐certainty evidence; Analysis 2.1; Lu 2012). In one study, at six months mean haemoglobin was lower in the hydroxyurea group when compared to resveratrol (MD −0.30 g/dL, 95% CI −1.14 to 0.54; 34 participants; low‐certainty evidence; Analysis 2.1; Haghpanah 2018). We downgraded both analyses twice for imprecision because findings were based on one small study and the CIs included both appreciable benefit and appreciable harm (Table 3).

2.1. Analysis.

Comparison 2: Single HbF inducer at any dose or duration compared with another HbF inducer, Outcome 1: Mean haemoglobin (g/dL), subgroup by types of HbF inducer

Two studies reported results we were unable to analyse. One study reported that the peak increase in haemoglobin in the hydroxyurea group was 0.98 g/dL (range 0.1 g/dL to 1.4 g/dL) and in the decitabine group was 1.28 g/dL (range 0.5 g/dL to 1.6 g/dL), but the time point for this observation was not stated (Jha 2019). One study stated narratively that there was a significant increase in haemoglobin at one, three and six months postintervention in the thalidomide group compared to the hydroxyurea group (Jain 2019).

Secondary outcomes

1. Foetal haemoglobin (%)

Three studies reported HbF at three months (Lu 2012) and six months (Jain 2019; Jha 2019) postintervention.

At three months, there was no difference in mean HbF between the CNP and Radix Astragali groups (MD 5.00%, 95% CI −11.44% to 21.44%; 1 study, 24 participants; low‐certainty evidence; Analysis 2.2; Lu 2012). We downgraded the certainty of the evidence twice for imprecision because findings were based on one small trial with very wide CIs (Table 3).

2.2. Analysis.

Comparison 2: Single HbF inducer at any dose or duration compared with another HbF inducer, Outcome 2: Mean HbF (%), subgroup by type of HbF inducer

At six months, there was no difference in mean HbF between thalidomide and hydroxyurea (MD 3.50%, 95% CI −1.41% to 8.41%; 1 study; 30 participants; low‐certainty evidence; Analysis 2.2; Jain 2019). We downgraded twice for imprecision because findings were based on one small trial with very wide CIs (Table 3).

At six months, one study reported a mean increase in HbF of 5.13% in the hydroxyurea group and 5.38% in the decitabine group (Jha 2019). There was no measure of dispersion and attempts to contact the authors were unsuccessful.

2. Long‐term sequelae of non‐transfusion‐dependent β‐thalassaemia

None of the studies reported growth, puberty, maxillary hyperplasia and severity of extramedullary haematopoiesis (liver and spleen size).

3. Quality of life

None of the studies reported QoL.

4. Adverse outcomes

a. Iron overload

One study reported the baseline ferritin level, but did not report the effect of the intervention on this outcome (Haghpanah 2018).

b. Need for iron chelation

None of the studies reported the need for iron chelation.

c. Adverse drug effects

We present an overall summary of the prespecified adverse effects, reported adverse effects and adverse effects that were not reported in the included studies in Table 2.

Two studies prespecified adverse drug effects for each HbF inducer and reported the frequency of the observed events (Jha 2019; Lu 2012), and two studies prespecified and reported the types of adverse drug effects without the frequency (Haghpanah 2018; Jain 2019). The certainty of evidence was very low for all studies in this comparison. We downgraded one level for lack of blinding and two levels for imprecision because findings were based on one small study with a very small sample size (Table 3).

Hydroxyurea

One study reported mild gastrointestinal upset (Jain 2019), while one study reported a respiratory tract infection, neurological problems and fever (Jha 2019).

Thalidomide

One study reported somnolence and headache (Jain 2019).

Decitabine

One study reported a respiratory tract infection (Jha 2019).

Resveratrol

One study reported gastrointestinal events included severe nausea, vomiting, abdominal pain, diarrhoea, gastrointestinal bleeding, headache and malaise (Haghpanah 2018).

Radix Astragali

One study reported that none of the prespecified adverse drug effects (abnormalities in blood counts, liver enzyme levels and renal profile) occurred (Lu 2012).

Combined natural preparation

One study reported that none of the prespecified adverse drug effects (abnormalities in blood counts, liver enzyme levels and renal profile) occurred (Lu 2012).

A single dose or dose regimen of a foetal haemoglobin inducer compared with another dose or dose regimen of the same foetal haemoglobin inducer

See Table 4.

Two studies compared different doses of the same HbF inducer (Bohara 2014; Fucharoen 2013).

One trial involving 61 participants compared hydroxyurea 10 mg/kg/day against hydroxyurea 20 mg/kg/day (Bohara 2014) and one trial involving 21 participants included four different doses of HQK‐1001 (10 mg/kg/day, 20 mg/kg/day, 30 mg/kg/day and 40 mg/kg/day) (Fucharoen 2013).

Primary outcomes

1. Blood transfusion

Neither study presented their results as frequency of transfusion measured by transfused volume per year or transfusion‐free interval.

However, one study reported that 7/29 participants in the hydroxyurea 20 mg/kg/day group and 0/32 participants in the hydroxyurea 10 mg/kg/day group required a blood transfusion during the six‐month intervention (Bohara 2014).

2. Haemoglobin

One study reported that mean haemoglobin was lower in the hydroxyurea 20 mg/kg/day group at 6, 12 and 24 weeks (24 weeks: MD −2.39 g/dL, 95% CI −2.80 to −1.98; 61 participants; very low‐certainty evidence; Analysis 3.1; Bohara 2014). We downgraded the certainty of the evidence by one level for risk of bias because allocation concealment was unclear and two levels for imprecision because findings were based on one small study with a very small sample size (Table 4).

3.1. Analysis.

Comparison 3: One dose or dose regimen of a HbF inducer compared with another dose or dose regimen of the same HbF inducer: hydroxyurea 20 mg/kg/day versus hydroxyurea 10 mg/kg/day, Outcome 1: Mean haemoglobin (g/dL)

One study presented the haemoglobin measurements graphically as medians without a measure of dispersion and did not compare the four different doses of HQK‐1001 with each (Fucharoen 2013).

Secondary outcomes

1. Foetal haemoglobin

One study reported that at 24 weeks, the mean HbF was lower in the hydroxyurea 20 mg/kg/day group compared with hydroxyurea 10 mg/kg/day (MD −10.20%, 95% CI −16.28% to −4.12%; 61 participants; very low‐certainty evidence; Analysis 3.2; Bohara 2014). We downgraded one level for risk of bias because the concealment of allocation was unclear and two levels for imprecision because findings were based on one small study with a very small information size (Table 4).

3.2. Analysis.

Comparison 3: One dose or dose regimen of a HbF inducer compared with another dose or dose regimen of the same HbF inducer: hydroxyurea 20 mg/kg/day versus hydroxyurea 10 mg/kg/day, Outcome 2: Mean HbF (%)

2. Long‐term sequelae of non‐transfusion‐dependent β‐thalassaemia

Neither study reported long‐term sequelae (growth, puberty, maxillary hyperplasia and severity of extra‐medullary haematopoiesis).

3. Quality of life

Neither study reported QoL. The study of hydroxyurea prespecified that 'sense of well‐being' and 'state of energy' would be measured, but these outcomes were not reported (Bohara 2014).

4. Adverse outcomes

a. Iron overload

Neither study reported iron overload.

b. Need for iron chelation

Neither study reported the need for iron chelation.

c. Adverse drug effects

We present an overall summary of the prespecified adverse effects, reported adverse effects and adverse effects that were not reported in the included studies in Table 2. Both studies in this comparison prespecified the adverse effects of the inducer being compared.

One study reported found a higher risk of neutropenia (defined by the authors as absolute neutrophil count below 0.5 × 10³/μL) in the hydroxyurea 20 mg/kg/day group (RR 9.93, 95% CI 1.34 to 73.67; 61 participants; very low‐certainty evidence; Analysis 3.3; Bohara 2014). This study also found a higher risk of thrombocytopenia (defined as a platelet count below 50 × 10³/μL) in the hydroxyurea 20 mg/kg/day group (RR 3.68, 95% CI 1.12 to 12.07; 61 participants; very low‐certainty evidence; Analysis 3.3). We downgraded both analyses one level for indirectness because the duration of the intervention was short and insufficient to identify potential changes and two levels for imprecision because findings were based on one small study with very wide CIs (Table 4).

3.3. Analysis.

Comparison 3: One dose or dose regimen of a HbF inducer compared with another dose or dose regimen of the same HbF inducer: hydroxyurea 20 mg/kg/day versus hydroxyurea 10 mg/kg/day, Outcome 3: Major adverse effects (at 24 weeks)

The adverse drug effects reported by the HQK‐1001 study are presented in the first comparison 'Single HbF inducer versus usual care or placebo' (Fucharoen 2013).

Two or more foetal haemoglobin inducers at any dose or duration compared with a usual care, placebo or a single foetal haemoglobin inducer

See Table 5.

Two studies compared a combination of two HbF inducers with a single HbF inducer (Haghpanah 2018; Karimi 2010a).

The comparisons were:

1. hydroxyurea plus placebo versus hydroxyurea plus resveratrol (Haghpanah 2018);

2. hydroxyurea versus hydroxyurea plus l‐carnitine (Karimi 2010a); and

3. resveratrol plus placebo versus hydroxyurea plus resveratrol (Haghpanah 2018).

Haghpanah 2018 gave no details about the placebo used alongside the single HbF inducers. Furthermore, the participants in resveratrol group were also given piperine capsules to increase the oral bioavailability of resveratrol (Haghpanah 2018).

No studies compared a combination of two or more HbF inducers against a placebo or usual care.

Primary outcomes

1. Blood transfusion

Neither study presented their results as frequency of transfusion measured by transfused volume per year or transfusion‐free interval.

The study comparing hydroxyurea to hydroxyurea plus l‐carnitine (120 participants) reported that 10 participants needed a blood transfusion within the six‐month intervention period, but did not specify which intervention they had received (Karimi 2010a).

2. Haemoglobin

One study reported haemoglobin at six months postintervention (Haghpanah 2018). Haemoglobin levels were lower in the group with resveratrol plus hydroxyurea compared to one of the study's HbF inducers (either resveratrol alone or hydroxyurea alone) (MD −0.74 g/dL, 95% CI −1.45 to −0.03; 54 participants; low‐certainty evidence; Analysis 4.1). We downgraded the certainty of the evidence by two levels for imprecision because findings were based on one small trial with a very small information size (Table 5).

4.1. Analysis.

Comparison 4: Two or more HbF inducers at any dose or duration compared with usual care, placebo or a single HbF inducer, Outcome 1: Mean haemoglobin (g/dL) at 6 months postintervention, subgroup by types of HbF inducer

One study reported an increase in mean haemoglobin from baseline in the group receiving hydroxyurea plus l‐carnitine, but no changes in the hydroxyurea alone group. However, the study did not directly compare the two groups (Karimi 2010a).

Secondary outcomes

1. Foetal haemoglobin

Neither study reported HbF.

2. Long‐term sequelae of non‐transfusion‐dependent β‐thalassaemia

One study comparing hydroxyurea to hydroxyurea plus l‐carnitine (120 participants) reported pulmonary acceleration time (PAT), which provides a measure of pulmonary hypertension (Karimi 2010a). Shorter PAT indicates higher pulmonary pressure (Mallery 1991). There was an increase in the PAT mean duration from 110.64 (SD 28.73) ms to 116.66 (SD 24.83) ms for the hydroxyurea plus l‐carnitine group. There were no data for the hydroxyurea alone group. We were unable to contact the authors for further details.

3. Quality of life

Neither study reported QoL.

4. Adverse outcomes

Two studies reported iron overload as baseline ferritin levels, but not the effect of the intervention on this outcome nor the need for any iron chelation (Haghpanah 2018; Karimi 2010a).

a. Iron overload

Two studies reported iron overload as baseline ferritin levels, but not the effect of the intervention on iron overload (Haghpanah 2018; Karimi 2010a).

b. Need for iron chelation

Neither study reported need for any iron chelation.

c. Adverse drug effects

Both studies prespecified and reported adverse drug effects (Table 2).

One study reported gastrointestinal events including severe nausea, vomiting, abdominal pain, diarrhoea, and gastrointestinal bleeding in the resveratrol alone group and hydroxyurea plus resveratrol group (Haghpanah 2018). The investigators also reported that some participants experienced headaches and malaise with gastrointestinal symptoms, but did not specify which group of participants experienced these events. None of the participants receiving hydroxyurea alone had adverse events.

One study comparing hydroxyurea to hydroxyurea plus l‐carnitine combined reported nausea and vomiting (six participants), headache (one participant) and abdominal discomfort (one participant) without specifying the intervention group (Karimi 2010a). The study reported no changes to the biochemical parameters and blood counts.

The certainty of evidence for adverse effects was very low (Table 5). We downgraded two levels for risk of bias because of unclear randomisation, allocation concealment and blinding, and two levels for imprecision because findings were based on one small study with little information.

We present an overall summary of the prespecified adverse effects, reported adverse effects and adverse effects that were not reported in the included studies in Table 2.

Discussion

Summary of main results

Our review included seven studies with 291 participants with NTDβT receiving eight different interventions that were relevant to this review (Bohara 2014; Fucharoen 2013; Haghpanah 2018; Jain 2019; Jha 2019; Karimi 2010a; Lu 2012). The age of the participants eligible for recruitment ranged from two years to 50 years. The actual enrolled participants were aged between two and 49 years. There were four pharmacological and four natural HbF inducers either used alone or in combination with another inducer. The duration of intervention ranged from two months to six months.

We are uncertain whether any of the eight HbF inducers in this review has a beneficial effect due to each comparison having only one or two studies, and all the studies were very small. Where we found differences between the intervention groups, these differences were small and may not be clinically meaningful.

Single foetal haemoglobin inducer versus usual care or placebo

Neither of the two studies comparing HbF versus placebo reported our first primary outcome, the capability of HbF inducers for reducing the need for blood transfusion by increasing haemoglobin or HbF levels, or both (Fucharoen 2013; Lu 2012). Although the evidence was very low certainty, all the HbF inducers in this comparison (Radix Astragali, CNP and HQK‐1001) may increase both haemoglobin and HbF levels, but more importantly, the data to inform us whether this could result in a reduction in transfusion volume or an increase in transfusion‐free interval was not available.

We have no information on long‐term sequelae of NTDβT or QoL.

Adverse effects, iron overload and the need for chelation therapy were not reported. Adverse drug effects for HQK‐1001 included upper respiratory tract infection, gastrointestinal symptoms such as abdominal pain and vomiting, dizziness, palpitation and headache. None of the prespecified adverse effects of Radix Astragali and CNP was reported to have occurred.

Single foetal haemoglobin inducer at any dose compared with another foetal haemoglobin inducer

We found four studies for this comparison (Haghpanah 2018; Jain 2019; Jha 2019; Lu 2012). Two studies assessed the need for transfusion, one comparing thalidomide versus hydroxyurea (Jain 2019), and one comparing resveratrol versus hydroxyurea (Haghpanah 2018). There are very little data on the effects of any interventions on transfusion and, therefore, we are very uncertain whether any intervention is able to reduce transfusion. We do not know the effects of a single HbF inducer compared to another for the outcomes of haemoglobin levels, HbF levels and adverse effects because there was only one study per pair‐wise comparison.

We have no information on long‐term sequelae of NTDβT or QoL.

A single dose or dose regimen of a foetal haemoglobin inducer compared with another dose or dose regimen of the same foetal haemoglobin inducer

Two studies compared HbF inducers given at different doses (Bohara 2014; Fucharoen 2013).

We are uncertain whether there is any difference in the need for transfusion for the two doses of hydroxyurea that were compared (Bohara 2014), and we have no information on the need for transfusion for HQK‐1001 (Fucharoen 2013).

The studies assessed different doses, but there is insufficient information to determine whether there are any dose‐dependent effects. Haemoglobin and HbF may be higher at 24 weeks in the group receiving hydroxyurea 10 mg/kg/day compared with the hydroxyurea 20 mg/kg/day group. Investigators in the study of HQK‐1001 reported median haemoglobin and HbF levels graphically, but the groups could not be compared. These comparisons were of low‐ to very low‐certainty evidence.

The study that prespecified QoL assessment did not report the outcome (Bohara 2014).

The adverse effects for each inducer were reported, but we do not know if there is a difference in adverse effects of a single inducer compared with another because there was only one study per pair‐wise comparison with very low‐certainty evidence.

Two or more foetal haemoglobin inducers at any dose or duration compared with usual care, placebo or a single foetal haemoglobin inducer

There were no studies that compared two or more HbF inducers with placebo or usual care, but we identified two studies comparing two HbF inducers against a single HbF inducer (Haghpanah 2018; Karimi 2010a).

For the comparison of hydroxyurea plus l‐carnitine versus hydroxyurea alone, participants needing transfusion during the study period were reported but could not be separated from study arms that we had excluded from this review (Karimi 2010a).

Haemoglobin may be lower in the group receiving hydroxyurea plus resveratrol when compared with hydroxyurea alone or resveratrol alone, but we are uncertain (Haghpanah 2018). There was a rise in haemoglobin reported with hydroxyurea plus l‐carnitine but not with hydroxyurea alone (Karimi 2010a).

We have no information on HbF or QoL.

The adverse effects reported in both studies were mainly related to the gastrointestinal system (Haghpanah 2018; Karimi 2010a). We do not know if there were differences in adverse effects of each single HbF inducer over another because there was only one study per pair‐wise comparison with very low‐certainty evidence.

Overall completeness and applicability of evidence

All included studies reported the effect of HbF inducers on haemoglobin and HbF, but we do not have information on their effect on blood transfusion. This is important because of the two primary outcomes in this review, blood transfusion would be a more clinically meaningful outcome and the most important outcome for both patients and policymakers. Both haemoglobin and HbF might be considered surrogate outcomes, since haemoglobin levels are used to determine whether a transfusion is needed. We also have insufficient information on QoL and the long‐term sequelae of chronic anaemia to NTDβT. The durations of the interventions were too short for any of the included comparisons to measure long‐term sequelae.

Besides information on long‐term outcomes, we do not know whether the effect of HbF inducers is sustained when they are given for longer times. Observational studies on hydroxyurea suggested outcomes were improved up to 24 months (Italia 2009), but not beyond (Mancuso 2006; Rigano 2010).

We have included one sequential‐dose analysis study, but due to the lack of analysable data, we were unable to draw any conclusions from this. There was also a lack of justification for the choice of washout period in the study (Fucharoen 2013). For any study that is designed to continue at another dose such as cross‐over study or sequential dose‐escalation study, we need to avoid a carryover effect; therefore, not only do we need to know the half‐life of the substance but, more importantly, we need to determine the duration of the clinical effects of the HbF inducer on the target sites such as the bone marrow or erythrocytes. The half‐life of HQK‐1001 was between nine and 15 hours (Perrine 2011), but the duration of its clinical effects on target sites, presumably the erythrocytes, is unknown and likely to be much longer. The duration of clinical effects might depend on the duration of RBC survival and this might vary between different types of NTDβT as well as the predominance of HbF, and therefore could vary considerably between individuals. In people with HbE/β‐thalassaemia, RBC lifespan was reduced when compared with a healthy population (Singer 2005b). Similarly, RBCs with predominant HbF had a shorter lifespan than healthy RBCs with HbA₁ predominance (Pearson 1967). The washout period might also be dependent on whether the person has had their spleen removed. Thus, predetermining the most appropriate washout period could be difficult.

RBC lifespan could also be affected by splenectomy (Musallam 2013b; Sharma 2019). Only three of the included studies considered the possibility of splenectomised participants responding differently (Fucharoen 2013; Haghpanah 2018; Karimi 2010a).

The five studies listed in the Studies awaiting classification table have 275 participants (NCT04411082; Cheng 2019; Huang 2016; Jain 2021; Wu 2007). Since the included studies had only 291 participants, the inability at this present time to include these 275 potential participants, who represent a comparatively large cohort compared with the included participants, may affect the conclusions of the review. It might also potentially result in the inclusion of information from RCTs on another HbF inducer not currently described in this review. One of our included studies was not published in English and was translated by more than one person to avoid any misinterpretation of data or missing any important information (Lu 2012).

Quality of the evidence

Using the GRADE assessment tool, the overall certainty of evidence for our primary outcomes ranged from low to very low for all comparisons (Table 1; Table 3; Table 4; Table 5). The main reasons for downgrading the evidence were for risk of bias, indirectness and imprecision. We had concerns about the risk of bias, mainly due to lack or poor descriptions of randomisation and blinding, and selective reporting.

We downgraded all outcomes for imprecision because the number of participants was small, the optimal information size was not met and the CIs were very wide incorporating both a substantial benefit and substantial harm. In addition, we downgraded some outcomes for indirectness because the study period was too short to observe any substantial changes or long‐term sequelae of the condition or long‐term adverse effects such as iron overload or need for chelation.

Potential biases in the review process

A potential bias in the review process was our decision to include one sequential dose‐escalation study, in spite of uncertainties about the duration of washout period and the possibility of carry‐over effects (Fucharoen 2013). We included this study because it met our inclusion criteria for participants and interventions and also reported our primary outcome. We could not perform a sensitivity analysis to test this decision because there was only one study for comparison. However, we noted from the data given in the study that haemoglobin levels did not appear to return to baseline during the washout period.

We reported outcomes that were expressed differently to our prespecified outcomes, for example, rather than reporting our primary outcome (volume of blood transfusion per participant per year), three studies reported the number of participants who required a transfusion during the study period. One study reported pulmonary hypertension measurements as long‐term sequelae of NTDβT. Although we had included this in the background section of the protocol (Foong 2015), we had not listed it in our outcomes; we have added it post hoc.

We had to make judgements about whether an intervention could be an HbF inducer rather than a form of gene therapy based on the current knowledge of their mechanism of action. Nevertheless, this may change as more information becomes available on the mechanism of action of the included interventions.

It is always possible that relevant unpublished data may have been missed. We were made aware of some Chinese language databases (CNKI, VIP, Wan Fang, SinoMed) that we had not searched and which may potentially have contained relevant studies. However, surveying these databases, except SinoMed, revealed that few if any studies in the database were RCTs, and most studies reported outcomes reflecting the Traditional Chinese Medicine concepts of Yin and Yang rather than the type of measurable clinical outcomes we prespecified in this review. We made a judgement that the potential low yield from these databases did not justify searching them systematically. Those studies that we did identify in the Chinese language were not obtained from any of these databases, and they reported both clinical outcomes and outcomes related to Traditional Chinese Medicine practices.

Agreements and disagreements with other studies or reviews

Hydroxyurea remains the most commonly studied HbF inducer. The previous version of this review, which only included hydroxyurea as the HbF inducer, found no RCTs comparing hydroxyurea with placebo or usual care (Foong 2016). The only included RCT in that review was the comparison between two different doses of hydroxyurea that has also been included in the current version of this review (Bohara 2014). However, we are aware of several observational studies assessing responses with various doses of hydroxyurea, from 8 mg/kg/day up to 30 mg/kg/day (Amoozgar 2011; El‐Beshlawy 2014; Fucharoen 1996; Karimi 2010b). These studies have reported similar bone marrow suppression, hepatotoxicity and gastrointestinal symptoms as found in the included study (Bohara 2014), and also at the higher doses of hydroxyurea (El‐Beshlawy 2014; Karimi 2010b; Taher 2010).

There are two other Cochrane Reviews that have reported the effectiveness of hydroxyurea in reducing blood transfusion, in SCD (Rankine‐Mullings 2022) and in TDT (Ansari 2019). There were no studies included in the TDT review (Ansari 2019). Besides finding improvements related to SCD, the authors of the SCD review found moderate‐certainty evidence for an improvement in HbF and provide further data on adverse effects (Rankine‐Mullings 2022). As for NTDβT, there is one systematic review on hydroxyurea for NTDβT which meta‐analysed 11 studies with 859 participants (Algiraigri 2017). The review included studies which were of a pre‐ and postdesign, but there were no RCTs. The authors of that review described serious limitations to the quality of the studies, which limited their conclusions about the efficacy and safety of hydroxyurea.

We found no systematic reviews that specifically reviewed trials on HbF inducers for NTDβT. However, there are several narrative reviews that suggested other natural or pharmacological products that might ameliorate anaemia. It is possible that some of these might work by increasing HbF levels. There are several other compounds which are currently being investigated as HbF inducers (rapamycin, angelicin, linear psoralens, ethanol extract, mithramycine, cucurbitacin‐D, fructus Trichosanthis, citropten, bergapten, 5‐azacytidine, trichostatin‐A and short‐chain fatty acids derivatives) but to date, there are no RCTs of these interventions (Bianchi 2009; Fard 2013; Kumari 2011; Musallam 2013a; Ng 2014).

Authors' conclusions

Implications for practice.

We are uncertain whether any of the eight foetal haemoglobin (HbF) inducers in this review have a beneficial effect on people with non‐transfusion‐dependent β‐thalassaemia (NTDβT). For each of these HbF inducers, we found only one or two small studies. There is no information on whether any of these HbF inducers have an effect on our primary outcome, blood transfusion. For the second primary outcome, haemoglobin, there may be small differences between intervention groups, but these may not be clinically meaningful and are of very low‐certainty evidence. Although we found HbF inducers may increase HbF, the evidence is very uncertain. There is no information about which single inducer might be more beneficial than another, or whether a combination of inducers might be better than a single inducer. We are very uncertain about the potential adverse effects of each inducer. Although evidence was very low‐certainty, at higher doses increased adverse effects were seen. Five studies are awaiting classification which may provide more information if we are able to include these in an update of the review. More evidence is needed for recommendation of the use of HbF inducers, be them alone or in combination, to reduce the need for blood transfusion through a rise in haemoglobin or HbF levels. There are insufficient data on which to make a treatment of choice.

Implications for research.

This review suggests several areas where further research is needed.

This Cochrane Review has shown that more studies are needed for all the currently studied HbF inducers. In particular, studies comparing an HbF inducer with a placebo are needed. Hydroxyurea appears to have been introduced into clinical practice for the treatment of NTDβT without adequate evaluation. No studies have compared hydroxyurea with a placebo. Therefore, there is an urgent need for large well‐designed randomised controlled trials comparing HbF inducers with placebo or usual care in people with NTDβT and, wherever possible, placebo‐controlled studies are preferable.

Since placebos such as teas or soups may be used in future studies, and these potentially could have some activity of their own, future studies need to adequately describe the nature of any placebo used, including whether participants are able to distinguish it from the intervention.

The primary outcome in future studies should be blood transfusion. Studies should be of sufficient duration to determine this outcome and criteria for blood transfusion need to be well‐defined. Since there is some evidence that the effects of HbF inducers might not be sustained, studies of sufficient duration to detect these are needed, perhaps up to two years, and outcomes might need to be measured at several prespecified time points prior to this. Longer study durations are also needed to detect changes in quality of life, changes to the long‐term sequelae of chronic anaemia, as well as enabling further monitoring for any adverse drug events and the complications of both the disease and blood transfusion.

Studies comparing a single HbF inducer or a combination of inducers with another HbF inducer should only proceed once the effectiveness of the inducer has been determined in placebo‐controlled studies.

History

Protocol first published: Issue 10, 2020

Notes

This review is set to replace the currently published review 'Hydroxyurea for reducing blood transfusion in non‐transfusion‐dependent thalassaemias' (Foong 2016). This new review includes other HbF inducers besides hydroxyurea.

Appendices

Appendix 1. Electronic search strategies

Database	Search strategy	Date searched
CENTRAL	#1 MeSH descriptor: [Thalassemia] explode all trees #2 (thalassaemia or thalassemia):ti,ab,kw (Word variations have been searched) #3 #1 or #2 #4 MeSH descriptor: [Fetal Hemoglobin] explode all trees #5 (fetal NEAR/2 (haemoglobin or hemoglobin)):ti,ab,kw (Word variations have been searched) #6 (foetal NEAR/2 (haemoglobin or hemoglobin)):ti,ab,kw (Word variations have been searched) #7 (Haemoglobin F):ti,ab,kw #8 (gamma globin):ti,ab,kw #9 (y‐globin):ti,ab,kw #10 #4 or #5 or #6 or #7 or #8 or #9 #11 #3 and #10	21 August 2022
PubMed	#1 "Thalassemia"[Mesh] #2 thalassaemia or thalassemia #3 #1 OR #2 #4 "Fetal Hemoglobin"[Mesh] #5 fetal haemoglobin #6 (fetal AND (haemoglobin OR hemoglobin)) #7 (foetal AND (haemoglobin OR hemoglobin)) #8 HbF #9 Hemoglobin F #10 Haemoglobin F #11 gamma globin #12 γ ‐globin #13 #4 OR #5 OR #6 OR #7 OR #8 OR #9 OR #10 OR #11 OR #12 #14 #3 AND #13 #15 randomized controlled trial [pt] #16 controlled clinical trial [pt] #17 randomized [tiab] #18 placebo [tiab] #19 drug therapy [sh] #20 randomly [tiab] #21 trial [tiab] #22 groups [tiab] #23 #15 OR #16 OR #17 OR #18 OR #19 OR #20 OR #21 OR #22 #24 animals [mh] NOT humans [mh] #25 #23 NOT #24 #26 #14 AND #25	21 August 2022
Embase	#1 'thalassemia'/exp #2 thalassaemia OR thalassemia #3 #1OR #2 #4 'hemoglobin f'/exp #5 ‘fetal hemoglobin’ #6 ‘fetal AND (haemoglobin OR hemoglobin)’ #7 ‘foetal AND (haemoglobin OR hemoglobin)’ #8 ‘hbf’ #9 ‘hemoglobin f’ #10 ‘haemoglobin f’ #11 ‘gamma globin’ #12 'γ globin' #13 #4 OR #5 OR #6 OR #7 OR #8 OR #9 OR #10 OR #11 OR #12 #14 #3 AND #13 #15 'randomized controlled trial':ti AND [humans]/lim #16 'controlled clinical trial':ti AND [humans]/lim #17 'randomized':ti,ab AND [humans]/lim #18 'placebo':ti,ab AND [humans]/lim #19 'drug therapy':lnk AND [humans]/lim #20 'randomly':ti,ab AND [humans]/lim #21 'trial':ti,ab AND [humans]/lim #22 'groups':ti,ab AND [humans]/lim #23 #15 OR #16 OR #17 OR #18 OR #19 OR #20 OR #21 OR #22 #24 #14 AND #23	21 August 2022
US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (www.clinicaltrials.gov)	[Advanced Search] Condition or disease: thalassemia or thalassaemia Other terms: HbF or fetal haemoglobin or fetal hemoglobin or foetal haemoglobin or foetal haemoglobin Study type: Interventional Studies (Clinical Trials)	21 August 2022
WHO ICTRP (apps.who.int/trialsearch).	[Advanced Search] Condition: thalassaemia or thalassemia AND Intervention: HbF or fetal haemoglobin or fetal hemoglobin or foetal haemoglobin or foetal hemoglobin Recruitment status is: All	21 August 2022
Chinese language databases China Network Knowledge Infrastructure (CNKI) (www.cnki.net/) Chinese Scientific Journals Database (VIP) (www.cqvip.com/) Wan Fang data www.wanfangdata.com.cn/index.html) SinoMed (www.sinomed.ac.cn/)	地中海贫血症 HbF 诱导剂 随机对照试验	21 August 2022

Appendix 2. Overview of HbF inducers in excluded and awaiting classification studies

HbF inducer	Study	Pharmacological or natural	Mode of administration	Considered as an HbF inducer
Colla corii asini	Li 2016	Natural	Oral drink	No
HQK‐1001 (2,2‐dimethylbutyrate)	Patthamalai 2014	Pharmacological	Oral capsule or tablet	Yes
Human erythropoietin	Elalfy 2013	Pharmacological	Subcutaneous	No
Hydroxyurea	Biswas 2014; Elalfy 2013; Huang 2016; Jain 2021; Loukopoulos 1998	Pharmacological	Oral capsule or tablet	Yes
IMR‐687 (phosphodiesterase‐9‐inhibitor)	NCT04411082	Pharmacological	Not reported in protocol	Yes
Radix Astragali	Guo 2014	Natural	Not reported	Yes
Valproate acid	Biswas 2014	Pharmacological	Not reported in abstract	Yes
Yisui Shengxue granules (益髓生血颗 粒), which contained 11 different types of herbs (Fructus Corni, Prepared Radix Polygoni multifl ori, Radix Rehmanniae preparata, Radix Astragali, Radix Codonopsis, Radix Angelicae sinensis, Fructus Psoraleae, Colla corii asini, Caulis Spatholobi, Carapax Trionycis and Fructus Amomi)	Cheng 2016; Cheng 2019; Fang 2007; Wu 2007	Natural	Oral drink	Yes
Thalidomide	Jain 2021	Pharmacological	Not reported	Yes

HbF: foetal haemoglobin.

Data and analyses

Comparison 1. Single HbF inducer at any dose or duration compared with usual care or placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Mean haemoglobin (g/dL) at 3 months postintervention, subgroup by types of HbF inducer	1	35	Mean Difference (IV, Fixed, 95% CI)	1.33 [0.54, 2.11]
1.1.1 Combined natural preparation (CNP)	1	16	Mean Difference (IV, Fixed, 95% CI)	1.60 [0.44, 2.76]
1.1.2 Radix Astragali	1	19	Mean Difference (IV, Fixed, 95% CI)	1.10 [0.04, 2.16]
1.2 Mean HbF (%) at 3 months, subgroup by types of HbF inducer	1	35	Mean Difference (IV, Fixed, 95% CI)	12.00 [‐0.74, 24.75]
1.2.1 Combined natural preparation (CNP)	1	16	Mean Difference (IV, Fixed, 95% CI)	9.00 [‐11.17, 29.17]
1.2.2 Radix Astragali	1	19	Mean Difference (IV, Fixed, 95% CI)	14.00 [‐2.44, 30.44]

Comparison 2. Single HbF inducer at any dose or duration compared with another HbF inducer.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 Mean haemoglobin (g/dL), subgroup by types of HbF inducer	2		Mean Difference (IV, Fixed, 95% CI)	Subtotals only
2.1.1 Hydroxyurea versus resveratrol	1	34	Mean Difference (IV, Fixed, 95% CI)	‐0.30 [‐1.14, 0.54]
2.1.2 Radix Astragali versus combined natural preparation	1	24	Mean Difference (IV, Fixed, 95% CI)	‐0.50 [‐1.38, 0.38]
2.2 Mean HbF (%), subgroup by type of HbF inducer	2		Mean Difference (IV, Fixed, 95% CI)	Subtotals only
2.2.1 Thalidomide versus hydroxyurea	1	30	Mean Difference (IV, Fixed, 95% CI)	3.50 [‐1.41, 8.41]
2.2.2 Radix Astragali versus combined natural preparation	1	24	Mean Difference (IV, Fixed, 95% CI)	5.00 [‐11.44, 21.44]

Comparison 3. One dose or dose regimen of a HbF inducer compared with another dose or dose regimen of the same HbF inducer: hydroxyurea 20 mg/kg/day versus hydroxyurea 10 mg/kg/day.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
3.1 Mean haemoglobin (g/dL)	1	61	Mean Difference (IV, Fixed, 95% CI)	‐2.39 [‐2.80, ‐1.98]
3.2 Mean HbF (%)	1	61	Mean Difference (IV, Fixed, 95% CI)	‐10.20 [‐16.28, ‐4.12]
3.3 Major adverse effects (at 24 weeks)	1		Risk Ratio (M‐H, Fixed, 95% CI)	Subtotals only
3.3.1 Neutropenia	1	61	Risk Ratio (M‐H, Fixed, 95% CI)	9.93 [1.34, 73.67]
3.3.2 Thrombocytopenia	1	61	Risk Ratio (M‐H, Fixed, 95% CI)	3.68 [1.12, 12.07]

Comparison 4. Two or more HbF inducers at any dose or duration compared with usual care, placebo or a single HbF inducer.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
4.1 Mean haemoglobin (g/dL) at 6 months postintervention, subgroup by types of HbF inducer	1	54	Mean Difference (IV, Fixed, 95% CI)	‐0.74 [‐1.45, ‐0.03]
4.1.1 Hydroxyurea	1	28	Mean Difference (IV, Fixed, 95% CI)	‐0.60 [‐1.58, 0.38]
4.1.2 Resveratrol	1	26	Mean Difference (IV, Fixed, 95% CI)	‐0.90 [‐1.93, 0.13]

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Bohara 2014.

Study characteristics
Methods	RCT Single centre in India Trial date: April 2012 to September 2012 Trial duration: 6 months
Participants	61 participants aged ≥ 8 years with HbE/β‐thalassaemia with the phenotype of β‐thalassaemia intermedia Sex: 31 males, 30 females Inclusion criteria: diagnosed with HbE/β‐thalassaemia and had a phenotype of β‐thalassaemia intermedia Blood transfusion status: transfusion‐independent, but persistently anaemic state or intermittent transfusion that was needed during recurring illnesses such as fever, or both Exclusion criteria: people with renal and hepatic dysfunction (serum creatinine 41.5 times normal or ALT and AST 44 times normal or significant conjugated hyperbilirubinemia, or a combination of these), people with WBC < 4 × 103/μL and platelet count < 100 × 103/μL, recent transfusion history (past 12 weeks) and people unable to follow‐up regularly Number of splenectomised participants prior to study not reported Hydroxyurea 10 mg/kg/day group: mean age 16.68 (SD 6.05) years Hydroxyurea 20 mg/kg/day hydroxyurea group: mean age 16.17 (SD 5.75) years
Interventions	Arm 1: hydroxyurea 10 mg/kg/day (designated control in this review), given for 6 months (n = 32) Arm 2: hydroxyurea 20 mg/kg/day, given for 6 months (n = 29) Hydroxyurea was planned to be taken over 24 weeks, provided the participants continued to show a rise in haemoglobin levels of ≥ 0.6 – 1.0 g/dL at any time during the study period and did not experience any adverse events. Adjustments were made to the dose and duration of intervention when: adverse events occurred (e.g. a reduction in haemoglobin > 0.5 g/dL that led to a need for transfusion, ANC < 1.5 × 103/μL, platelet count < 50 × 103/μL, significant hepatic or renal dysfunction); a participant was deemed non‐responsive at the 12‐week cut‐off point (when hydroxyurea caused a rise in haemoglobin of < 0.5 g/dL or a reduction in haemoglobin from prehydroxyurea value). Hydroxyurea therapy was withheld for 1 week when adverse events occurred. These participants "resumed the therapy as per study protocol" once the toxicity had resolved but would continue hydroxyurea at a dose of 5 mg/kg/day lower than the dose at which the toxicity had occurred. Hydroxyurea was permanently withheld for non‐responsiveness. There were 5 non‐responsive participants from the 10 mg/kg/day group and 12 from the 20 mg/kg/day group. The number of participants remaining on hydroxyurea on each arm declined markedly as the study progressed. At the end of the study, 25 participants remained receiving the lower dose of hydroxyurea while 13 participants received the higher dose. There were no participants randomised to receive either placebo or usual care.
Outcomes	Haemoglobin (g/dL) measured at baseline, 6, 12 and 24 weeks HbF (%) measured at baseline, 6, 12 and 24 weeks Adverse effects (neutropenia, thrombocytopenia, gastrointestinal disturbances and raised liver enzymes) Response rates (good response, intermediate response and no response) based on arbitrary criteria Sense of well‐being and state of energy
Notes	No funding statement. Authors reported no conflict of interest. No pre‐existing published protocol. Available as published abstract and full‐text publication.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote: "Patients were randomised to one of two study arms, group A and B using a random number table."
Allocation concealment (selection bias)	Unclear risk	Allocation concealment was not reported.
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Blinding of participants or trial personnel was not reported for all outcomes.
Blinding of outcome assessment (detection bias) Subjective outcomes: transfusion rates, long‐term sequelae, QoL and adverse events	Unclear risk	Unclear whether the participants were blinded and all these outcomes depended on assessors' clinical judgement or participants' perception, or both.
Blinding of outcome assessment (detection bias) Objective outcomes: haemoglobin and HbF levels	Low risk	Outcomes were laboratory assessments, judged to not influence response to hydroxyurea.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Although not all participants received the intervention that they were assigned for the full duration of the trial, all enrolled participants were analysed in the group they were assigned to "irrespective of how long they received hydroxyurea therapy."
Selective reporting (reporting bias)	High risk	In the methods section, participants were reported to have undergone an interview regarding their sense of well‐being and state of energy, and this was not stated in the results.
Other bias	Low risk	The baseline characteristics between the 2 groups had no significant difference in age, gender, β genotype, reticulocytes, and the mean for haemoglobin, HbF and HbE. However, there was a significant difference in the mean of the MCV levels, judged to not influence response to hydroxyurea.

Fucharoen 2013.

Study characteristics
Methods	Phase 1/2 sequential dose‐escalation RCT comparing 4 different doses of HQK‐1001 with a placebo and between each different doses. Multicentre in Thailand and Lebanon Trial period: 2009 – 2010 Trial duration: 8 weeks (56 days) of intervention, followed by 4 weeks of follow‐up and ≥ 8 weeks of wash‐out period
Participants	21 participants aged 19 – 49 years with β‐thalassaemia intermedia, either of subtype HbE/β0 thalassaemia (n = 14) or β0/β+ thalassaemia mutations (n = 7) Sex: 4 males, 17 females Inclusion criteria: diagnosed with β intermedia and had total haemoglobin levels of 5 – 9 gm/dL, aged 18 – 50 years; had a spleen dimension of ≤ 2 cm from the left costal margin for non‐splenectomised participants, based on lower responses in extreme splenomegaly noted in prior trials of other HbF inducers Blood transfusion status: not on regular transfusion Exclusion criteria: need for regular RBC transfusions or had RBCs transfusion within 2 months prior to start of the study, taking other experimental medications, pregnant females 16 were splenectomised prior to study. Mean age for each cohort not reported.
Interventions	HQK‐1001 capsules given for ≥ 8 weeksa (56 days) Cohort 1: HQK‐1001 10 mg/kg/day (n = 8) Cohort 2: HQK‐1001 20 mg/kg/day (n = 9) Cohort 3: HQK‐1001 30 mg/kg/day (n = 6) Cohort 4: HQK‐1001 40 mg/kg/day (n = 9) aThis was a sequential dose‐escalation study where cohorts moved to a higher dose: from 10 mg/kg to 30 mg/kg (Cohort 1 became Cohort 3) and 20 mg/kg to 40 mg/kg (Cohort 2 became Cohort 4). Before moving on as the next cohort, the participants in that cohort were rerandomised after an 8‐week washout period. For Cohort 1, the randomisation was at a ratio of 7:2 (7 received intervention to 2 received placebo). For Cohorts 2, 3 and 4, the randomisation was at a ratio of 6:1 (6 received intervention to 1 received placebo). Placebo: matching capsules administered once a day, orally, for at least 56 days. Cohort 1: n = 2 Cohort 2: n = 2 Cohort 3: n = 2 Cohort 4: n = 2 The cohort from Thailand received intervention for 8 weeks and the cohort from Lebanon received intervention for 10 weeks.
Outcomes	Safety and tolerability of HQK‐1001 measured at baseline and every 2 weeks to completion of study Pharmacokinetic profiles of HQK‐1001 within 24 hours, measured on day 13 and day 55 of intervention Adverse effects of HQK‐1001 on haemoglobin monitored at baseline and every 2 weeks to completion of study Measurement of changes in HbF from baseline
Notes	Supported by HemaQuest Pharmaceuticals. No declaration on conflict of interest. Published protocol is available (NCT00790127) in ClinicalTrials.gov. Written to Susan P Perrine (sperrine@bu.edu) and Suthat Fuchareon (suthat.fuc@mahidol.ac.th) for more details. Susan P Perrine replied.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote "… subjects were randomised to receive active treatment or placebo according to a schedule by third party and employing a table of random numbers."
Allocation concealment (selection bias)	Unclear risk	Allocation concealment not reported.
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Matching placebo given in a similar regimen.
Blinding of outcome assessment (detection bias) Subjective outcomes: transfusion rates, long‐term sequelae, QoL and adverse events	Unclear risk	Blinding of outcome assessors was unclear and adverse events outcomes depended on the clinical judgement or participants' perception, or both. The other outcomes were not assessed.
Blinding of outcome assessment (detection bias) Objective outcomes: haemoglobin and HbF levels	Low risk	Matching placebo received a similar regimen. The authors stated, "Participants, caregivers, investigators and laboratory personnel were blinded to interventions."
Incomplete outcome data (attrition bias) All outcomes	Low risk	Quote: "All subjects were included in the subgroup analyses."
Selective reporting (reporting bias)	Low risk	All outcomes were reported.
Other bias	Unclear risk	Quote: "The half‐life of HQK‐1001 is 11 hours in normal human volunteers." Comment: we are unsure if the wash‐out period of 8 weeks was sufficient to ensure no carry‐over effects before the next escalation dose. The haemoglobin level did not reach baseline before the next escalation dose.

Haghpanah 2018.

Study characteristics
Methods	Phase 3 RCT Single centre in Iran Trial date: October 2016 to March 2017 Trial duration: 6 months
Participants	54 people with non‐transfusion‐dependent β‐thalassaemia intermedia Sex: 29 males, 25 females Inclusion criteria: aged > 18 years with haemoglobin levels 7 – 10 g/dL Blood transfusion status: no requirement of blood transfusion for the past 6 months Exclusion criteria: had underlying diabetes mellitus, overt heart failure, and viral hepatitis; considerable decline in haemoglobin levels and need for blood transfusion; severe gastrointestinal complications or unwillingness to continue consuming the drug 25 were splenectomised prior to study. Arm 1: mean age 29.7 (SD 4.7) years Arm 2: mean age 26.5 (SD 5.8) years Arm 3: mean age 28.1 (SD 6.2) years
Interventions	Arm 1: hydroxyurea 8 – 15 mg/kg bodyweight/day plus placebo, given for 6 months (n = 18) Arm 2: resveratrola 4 capsules a day in 4 divided doses plus placebo, given for 6 months (n = 16) Arm 3: hydroxyurea 8 – 15 mg/kg/day and 4 resveratrola capsules a day (taken in 4 divided doses), given for 6 months (n = 20) aResveratrol: combined resveratrol and piperine capsules contained micronised transresveratrol 500 mg (with 98% purity) plus piperine 10 mg (with 95% purity). Resveratrol was an HbF inducer while piperine was added to increase the therapeutic effect of resveratrol by increasing resveratrol's oral bioavailability. Since resveratrol was the active inducer, for this review, the combined resveratrol‐piperine capsules were labelled as resveratrol only. No description of the placebo was reported. Placebo was added to the 2 intervention groups that had 1 of the HbF inducers to avoid performance or detection bias.
Outcomes	Blood transfusion within the study period Haemoglobin level (g/dL) measured monthly CBC, serum ferritin, kidney and liver function tests, and fasting blood sugar measured monthly Adverse events monitored monthly Clinical examination for any adverse events monitored monthly Response rate (good response and poor response) based on changes in haemoglobin levels and blood transfusion requirement
Notes	Funded by Shiraz University of Medical Sciences. Authors declared no conflict of interest. Published protocol available (IRCT2014112320051N2 2015). Attempts to contact the authors for further clarification or details were unsuccessful.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote: "… randomly allocated into three groups by simple randomization method, using computer‐generated random numbers."
Allocation concealment (selection bias)	Unclear risk	Allocation concealment not reported.
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	The nature of hydroxyurea and placebo were not described. Resveratrol were capsules. The intake frequency of the hydroxyurea plus placebo arm was not reported. However, resveratrol plus placebo and resveratrol plus hydroxyurea were given as 4 capsules a day in 4 divided doses.
Blinding of outcome assessment (detection bias) Subjective outcomes: transfusion rates, long‐term sequelae, QoL and adverse events	Unclear risk	It was unclear whether the participants were blinded and transfusion rates and adverse events depended on assessors' clinical judgement or participants' perception, or both.QoL and long‐term sequelae outcomes were not assessed.
Blinding of outcome assessment (detection bias) Objective outcomes: haemoglobin and HbF levels	Low risk	These outcomes were laboratory assessments, judged to not influence response to the given interventions.
Incomplete outcome data (attrition bias) All outcomes	Low risk	There were 3 dropouts in hydroxyurea arm, 10 in resveratrol arm and 8 in hydroxyurea plus resveratrol arm. All participants including dropouts were analysed at end of study.
Selective reporting (reporting bias)	Low risk	All outcomes were reported. However, the criteria for transfusion were not reported.
Other bias	Low risk	Quote: "The baseline characteristics between the three groups had no significant difference in age, gender, number of splenectomised participants and mean haemoglobin, BMI, and median serum ferritin."

Jain 2019.

Study characteristics
Methods	RCT Single centre in India Trial date: not reported Trial duration: 6 months
Participants	30 people with HbE‐β‐thalassaemia Sex: male:female ratio 1.1:1 Inclusion criteria: non‐transfusion‐dependent HbE/β‐thalassaemia Blood transfusion status: non‐transfusion‐dependent Exclusion criteria: pregnant women and those planning for pregnancy Number of splenectomised participants prior to study not reported Arm 1: median 17 (range 6 – 32) years Arm 2: median 27 (range 7 – 45) years
Interventions	Arm 1: hydroxyurea 10 mg/kg/day (n = 15) Arm 2: thalidomide 50 mg/day (fixed dose regardless of weight) (n = 15) Duration of the intervention not specified but can be assumed being given for 6 months based on the furthest time of the assessments.
Outcomes	Increment in haemoglobin (g/dL) measured at baseline, 1 month, 3 months and 6 months Increment in HbF levels measured at baseline, and 1 month and 3 months Tolerability and safety of both hydroxyurea and thalidomide Mean blood volume requirement (units/participant/year) Quote: "Responder was defined as increment of haemoglobin ≥ 1 g/dL and non‐responder as rise by < 1.0 g/dL or drop in haemoglobin level from the baseline."
Notes	This was only available in the form of an abstract. Efforts to contact investigators or contact authors for more information were unsuccessful. The authors declared no conflict of interest. Funding and presence of published protocol not reported. No contact details of the authors were available for clarification or details.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Quote: "A randomized controlled trial of …" Details of how this sequence generation was performed not reported.
Allocation concealment (selection bias)	Unclear risk	Allocation concealment not reported.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Hydroxyurea and thalidomide were most likely to be different in appearance, smell and taste. No further information on how these interventions were designed to be as similar as possible.
Blinding of outcome assessment (detection bias) Subjective outcomes: transfusion rates, long‐term sequelae, QoL and adverse events	High risk	Blinding was not possible and transfusion rates and adverse events depended on assessors' clinical judgement or participants' perception, or both. QoL and long‐term sequelae outcomes not assessed.
Blinding of outcome assessment (detection bias) Objective outcomes: haemoglobin and HbF levels	Low risk	Haemoglobin and HbF were laboratory assessments, judged to not influence response to the given treatments.
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	This was an abstract. The study flow diagram was not shown.
Selective reporting (reporting bias)	Unclear risk	This was an abstract.
Other bias	Low risk	The baseline for age between the 2 groups were significantly different but not for haemoglobin and HbF levels. A median value, instead of a mean, was given for both age and haemoglobin level. However, these were judged to not influence response to inducers.

Jha 2019.

Study characteristics
Methods	RCT Single centre in India Trial date: not reported Trial duration: 12 weeks
Participants	30 people with HbE/β‐thalassaemia Sex: not reported Inclusion criteria: aged > 12 years, with non‐transfusion‐dependent HbE/β‐thalassaemia Blood transfusion status: non‐transfusion‐dependent Exclusion criteria: not reported The number of splenectomised participants prior to the study was not reported. Arm 1: mean age not reported Arm 2: mean age not reported
Interventions	Arm 1: hydroxyurea, orally once a day for 12 weeks (n = 15) Arm 2: decitabine, subcutaneously twice a week for 12 weeks (n = 15) Doses for each intervention were not reported. The appearance of hydroxyurea was not reported.
Outcomes	CBC (haemoglobin (g/dL) levels) every 2 weeks HbF (%) at baseline and then every month until completion of 12 weeks of intervention Adverse events (infections, cytopenia, deranged biochemical parameters) Molecular evaluation for β mutation, α deletion and XMN1 polymorphism
Notes	This was only available in the form of an abstract. No contact details were available for the first author and attempts to contact co‐authors or the university were unsuccessful. Authors declared no conflict of interest. Funding and presence of published protocol not reported.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No description of random sequence generation, although the author stated that "… were randomly assigned into two treatment arms …"
Allocation concealment (selection bias)	Unclear risk	Allocation concealment not reported.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Hydroxyurea was given orally and decitabine was given subcutaneously. Blinding was not possible because the 2 interventions could be distinguished from appearance and mode of administration.
Blinding of outcome assessment (detection bias) Subjective outcomes: transfusion rates, long‐term sequelae, QoL and adverse events	High risk	Blinding was not possible and adverse event outcomes depended on clinical judgement or participants' perception, or both. Transfusion rates, QoL, and long‐term sequelae were not assessed.
Blinding of outcome assessment (detection bias) Objective outcomes: haemoglobin and HbF levels	Low risk	Haemoglobin and HbF levels were laboratory assessments judged to not influence response to the given treatments.
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	This was an abstract. The study flow diagram was not shown.
Selective reporting (reporting bias)	Unclear risk	This was an abstract. Any derangements in biochemical parameters and cytopenia were not reported in the results.
Other bias	Low risk	This was an abstract. The baseline characteristics were not reported, but this was judged to not influence response to inducers.

Karimi 2010a.

Study characteristics
Methods	Phase 2 RCT Single centre in Iran Trial date: June 2007 to November 2007 Trial duration: 6 months
Participants	120 people with β‐thalassaemia intermedia receiving hydroxyurea for the past 6 months Sex: 60 males, 60 females Inclusion criteria: aged 4 – 35 years, responding well (judged by haematological parameters) to hydroxyurea for the past 6 months and presented with a mean haemoglobin ≥ 7 g ⁄ dL Blood transfusion status: need for blood transfusion with an interval > 6 months or without a need for blood transfusion Exclusion criteria: have underlying diabetes mellitus, hepatitis B and hepatitis C, positive tests for human immunodeficiency virus, clinical cardiac symptoms or receiving drug for cardiac disease, other haemoglobinopathies except thalassaemia intermedia, and pregnant or lactating women 60 were splenectomised prior to study, but there is no distribution report to which these participants belonged. Arm 1: mean age 17.63 (SD 6.2) years Arm 2: mean age 20.15 (SD 4.4) years Arm 3: mean age 18.22 (SD 6.9) years Arm 4: mean age 20.75 (SD 7.2) years
Interventions	Arm 1: hydroxyurea once a day at a mean dose of 10 mg/kg/day (range 8 – 12 mg/kg/day) (n = 30) Arm 2: hydroxyurea once a day at a mean dose of 10 mg/kg/day (range 8 – 12 mg/kg/day) and l‐carnitine 50 mg/kg/day in divided doses per day (n = 30) Arm 3: hydroxyurea once a day at a mean dose of 10 mg/kg/day (range 8 – 12 mg/kg/day) and magnesium chloride powder suspension, administered in divided doses at 0.6 meq/kg/day (n = 30) Arm 4: hydroxyurea once a day at a mean dose of 10 mg/kg/day (range 8 – 12 mg/kg/day), l‐carnitine 50 mg/kg/day in divided doses per day and magnesium chloride powder suspension, administered in divided doses at 0.6 meq/kg/day (n = 30) Hydroxyurea came in the form of 500 mg capsules (from Medac, Hamburg, Germany). L‐carnitine came in the form of 250 mg tablets (from Minoo Company, Tehran, Iran). Magnesium chloride came in the form of powder and would be formulated into suspension, making each millilitre to contain 1 meg of magnesium chloride (from Tabib Thjhiz Nik Company, Tehran, Iran). Since hydroxyurea was given to all groups, we decided Arm 1 (hydroxyurea alone) would be the control and Arm 2 (hydroxyurea plus l‐carnitine) the intervention. Since magnesium chloride is known to stabilise the erythrocyte membrane and morphology and not an HbF inducer, we did not include Arms 3 and 4 in the review.
Outcomes	CBC measured monthly Liver function, renal profile creatinine levels measured 2 monthly (to monitor adverse effects on liver and kidneys) Echocardiogram at baseline and 6‐month postintervention Clinical examination including splenic measurements was performed monthly Physical findings for any adverse events (gastrointestinal symptoms and headache) measured monthly
Notes	Funded by The Shiraz Hematology Research Center of Shiraz University of Medical Sciences. No declaration on conflict of interest. Published protocol is available (NCT00809042) in ClinicalTrials.gov. Attempts to contact the authors for further clarifications or details were unsuccessful. Data from participants who required a transfusion during the course of intervention were removed from the analysis.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Quote: "Patients were randomly assigned into four groups …" There was no further elaboration on how this was done.
Allocation concealment (selection bias)	Unclear risk	Allocation concealment was not reported.
Blinding of participants and personnel (performance bias) All outcomes	High risk	All 3 interventions were distinguishable based on appearance and frequency of administration. Blinding was not possible. Knowledge of the intervention or placebo could influence participants' intake or clinicians' management ways.
Blinding of outcome assessment (detection bias) Subjective outcomes: transfusion rates, long‐term sequelae, QoL and adverse events	High risk	Blinding was not possible and transfusion rates, long‐term sequelae and adverse events depended on clinical judgement or participants' perception, or both. QoL was not assessed.
Blinding of outcome assessment (detection bias) Objective outcomes: haemoglobin and HbF levels	Low risk	Haemoglobin and HbF levels were laboratory assessments judged to not influence response to the given treatment.
Incomplete outcome data (attrition bias) All outcomes	High risk	16 participants were excluded from the final analyses due to adverse events (n = 8), poor adherence to study medications (n = 6) and required blood transfusion (n = 2).
Selective reporting (reporting bias)	Unclear risk	The baseline levels for HbF were stated, but the level at each measurement time point was not reported clearly.
Other bias	Low risk	The baseline characteristics between the 4 groups showed no significant difference in age, gender and the level of haemoglobin and HbF.

Lu 2012.

Study characteristics
Methods	RCT Single centre in China Trial date: January 2011 to November 2011 Trial duration: 12 weeks
Participants	57 people with β‐thalassaemia, mean age 6.5 (SD 3.6) years with haemoglobin levels 4.76 – 9.50 g/dL (mean haemoglobin level 7.4, SD 0.9 g/dL) Sex: 29 males, 28 females Inclusion criteria: aged 2 – 18 years with haemoglobin levels 4.5 – 10 g/dL Blood transfusion status: depend on severity; people with severe β‐thalassaemia had their last transfusion ≥ 4 weeks ago, while people with intermediate β‐thalassaemia had their last transfusion ≥ 12 weeks ago. Exclusion criteria: immunocompromised; with underlying heart, liver, kidney, hormonal or blood disorders; hypersensitivity problems and mental illness Number of splenectomised participants prior to study was not reported. Mean age for each arm was not reported.
Interventions	Participants with intermediate β‐thalassaemia (n = 35) Arm 1: combined natural preparation (n = 11) Arm 2: Radix Astragali package (黄芪) (n = 13) Arm 3: placebo (dextrin and starch) packed and taken in a similar regimen (n = 11) Participants with severe β‐thalassaemia (n = 22) Arm 1: combined natural preparation (n = 7) Arm 2: Radix Astragali package (黄芪) (n = 8) Arm 3: placebo (dextrin and starch) packed and taken in a similar regimen (n = 7) Each intervention was prepared in tea bags. Combined natural preparation has a mixture of Radix Astragali (黄芪 1 g), Codonopsis pilosula (党参 3 g) and tortoise plastron (龟板 0.7 g) in a tea bag. Radix Astragali (黄芪) package had Radix Astragali (黄芪 10 g) in a tea bag. The number of tea bags to be consumed depended on age. Arm 1: For children aged 2 years, 1 tea bag per day For children aged 6 years, 2 tea bags per day For children aged 12 – 18 years, 3 tea bags per day Arm 2: For children aged 2 years, 3 tea bags per day For children aged 6 years, 6 tea bags per day For children aged 12 – 18 years, 9 tea bags per day Arm 3: similar to Arm 2 The tea bags were immersed in warm water and drunk daily according to the schedule above for 12 weeks. For Arm 1 and Arm 2: the number of tea bags per day differed because they were made to be of equivalent strength to the traditional mixed preparation of Radix Astragali (黄芪 10 g), Codonopsis pilosula (党参 10 g) and tortoise plastron (龟板 10 g). The type of substance used as the placebo was not described but was said to be packed similarly in appearance.
Outcomes	Haemoglobin, RBC, MCV and MCH levels measured at baseline, 4 weeks, 8 weeks and 12 weeks HbF levels and reticulocyte counts measured at baseline and 12 weeks WBC, neutrophil counts and platelet counts measured at baseline, 4 weeks, 8 weeks and 12 weeks postintervention Liver function test (ALT, AST, GGT) and renal profile (urea, sodium, potassium, creatinine) measured at baseline and 12 weeks postintervention Response rate (good, fair or poor response) based on changes in haemoglobin and HbF levels before and after intervention
Notes	Funded by Scientific Research Project of GuangDong Provincial Bureau of Traditional Chinese Medicine (2010285) and Joint Scientific Research Project of GuangDong Provincial Department of Science and Technology – GuangDong Academy of Traditional Chinese Medicine (2011B032200002). No declaration of conflict of interest. The study was published in Chinese with an additional abstract in English. No pre‐existing published protocol. Attempts to contact the authors for further clarification or details were unsuccessful.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Participants were assigned to groups by random numbers generated via a numbering software.
Allocation concealment (selection bias)	Unclear risk	Allocation concealment not reported.
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	The 3 groups were given tea bags identical in appearance administered in a similar regimen. However, we were unable to judge whether the participants or personnel were able to guess what they were drinking because there could be differences in smell and taste after the tea bags were dipped in water.
Blinding of outcome assessment (detection bias) Subjective outcomes: transfusion rates, long‐term sequelae, QoL and adverse events	Low risk	The adverse events were based on laboratory measurements that were objective and unlikely to influence responses to the given intervention. Transfusion rates, long‐term sequelae and quality of life were not assessed.
Blinding of outcome assessment (detection bias) Objective outcomes: haemoglobin and HbF levels	Low risk	Haemoglobin and HbF levels are laboratory assessments judged to not influence response to the given intervention.
Incomplete outcome data (attrition bias) All outcomes	Low risk	All participants were accounted for in the results.
Selective reporting (reporting bias)	Low risk	All outcomes were reported.
Other bias	Low risk	Quote: "The baseline characteristics between the three arms for both intermediate and severe cohorts had no significant difference in age, gender and the level of haemoglobin."

ALT: alanine aminotransferase; ANC: absolute neutrophil count; AST: aspartame aminotransferase; BMI: body mass index; CBC: complete blood count; GGT: gamma‐glutamyl transferase; HbE: haemoglobin E; HbF: foetal haemoglobin; MCH: mean corpuscular haemoglobin; MCV: mean corpuscular volume; n: number of participants; QoL: quality of life; RBC: red blood cell; RCT: randomised controlled trial; SD: standard deviation; WBC: white blood cell count.

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Biswas 2014	2‐arm RCT comparing at hydroxyurea and valproate acid. These are both potential HbF inducers. However, the group of participants involved had severe transfusion‐dependent thalassaemia. Published as an abstract only.
Chai 2005	A 2‐arm before‐after study that used Yisui Shengxue granules on children with β‐thalassaemia.
Cheng 2016	Observational before‐after study on Yisui Shengxue granules to people with mild, moderate and severe β‐thalassaemia.
Elalfy 2013	2‐arm RCT where hydroxyurea was given to both groups and 1 group received additional human erythropoietin which is not an HbF inducer but another potential intervention to increase haemoglobin levels.
Fang 2007	Observational study assessing the efficacy of Yisui Shengxue granules on people with β‐thalassaemia major.
Guo 2014	Abstract of study of Radix Astragali that has been withdrawn. From the abstract, the study appeared to be laboratory‐based.
Li 2016	RCT using colla corii asini (阿 胶, donkey‐hide gelatine) on pregnant women with thalassaemia. Colla corii asini is unlikely to be an HbF inducer because it acts by activating the erythrocyte progenitor cells in bone marrow. It is reported to improve anaemia but reduces HbF and HbA2 levels.
Loukopoulos 1998	Before‐after study looking at hydroxyurea.
Patthamalai 2014	Letter to the editors about HQK‐1001.

HbA₂: haemoglobin A₂; HbF: foetal haemoglobin; RCT: randomised controlled trial.

Characteristics of studies awaiting classification [ordered by study ID]

Cheng 2019.

Methods	RCT Single centre in China Trial date: July 2011 to October 2011 (as written in the protocol) Trial duration: 3 months
Participants	40 participants diagnosed with β‐thalassaemia Sex: 25 males, 15 females Inclusion criteria: aged 3 – 40 years and diagnosed with β‐thalassaemia according to "Diagnosis and curative effect evaluation standard of haematopathy" (Zhang 2007) and "The principles of usage of the new Chinese herbs" (Zheng 2002) for Chinese medicine syndrome of Gan (Liver) and Shen (Kidney) yin deficiencya, have insufficiency of blood and essence, and have signed consent to the treatment plan (minors signed the consent under supervision of their guardians). Blood transfusion status: no blood transfusion or use of haematinics in the past 6 months Exclusion criteria: the presence of other primary diseases, upper respiratory tract infection, allergy to the experimental substance, blood transfusion within 45 days and psychiatric diagnosis or pregnancy. Number of splenectomised participants not reported Arm 1: mean age 13.76 (SD 9.05) years Arm 2: mean age 16.28 (SD 9.63) years
Interventions	Arm 1: Yisui Shengxue granules (益髓生血颗 粒) (12 g per tea bag, taken orally after immersing in warm water) drunk 3 times a day for 3 months (n = 20) Arm 2: placebo (dextrin and starch) packed and drunk in a similar regimen (n = 20)
Outcomes	Clinical efficacy (haemoglobin (g/L), RBC, reticulocyte and HbF levels (%) detected through peripheral blood film) measured at baseline, 1 month, 2 months and 3 months postintervention. Global DNA‐methylation assay measured at baseline and at end of study period (this was used for only 5 selected participants with a good response to YSSX)
Notes	aThe precise criteria were not available to us and neither were the quoted sources. Supported by the National Basic Research Program of China (No. 2010CB530406), National Natural Science Foundation of China (No. 81173167) and the Natural Science Foundation of Guangxi (No. 2012GXNSFAA053156). Authors declared no conflict of interest. Published protocol (NCT01549080) in ClinicalTrials.gov. Noted baseline discrepancies that were reported as no significant differences. There were also discrepancies between the protocol and full report. We are currently attempting to verify the data.

Huang 2016.

Methods	Prospective study with 2 groups of participants. Randomisation and allocation concealment were not reported.
Participants	29 participants with β‐thalassaemia intermedia
Interventions	Arm 1: hydroxyurea Arm 2: no hydroxyurea No further details available
Outcomes	Haemoglobin levels Reticulocytes percentage Serum ferritin Blood transfusion dependency Adverse effects
Notes	Unable to determine the study design. 1 authors, Dr Yao (Yao Hong Xia), replied that they will supply the further information that we are currently awaiting.

Jain 2021.

Methods	Prospective study with 3 groups of participants Randomisation and allocation concealment were not reported.
Participants	45 participants with β‐thalassaemia
Interventions	Arm 1: thalidomide plus folic acid Arm 2: hydroxyurea plus folic acid Arm 3: folic acid alone
Outcomes	Efficacy Safety profile
Notes	Unable to determine the study design. In the process of contacting the authors. Folic acid deficiency leads to impaired DNA synthesis and thus erythroblast apoptosis. We are unsure of the role of folic acid in this study.

NCT04411082.

Methods	RCT
Participants	120 participants with β‐thalassaemia or HbE/β‐thalassaemia including those with concomitant single α gene deletion, duplication or triplication. The participants had either TDT or NTDT. Age: 18 – 65 years Inclusion criteria (as stated in the protocol for participants with NTDT): documented diagnosis of β‐thalassaemia or HbE/β‐thalassaemia in their medical history, concomitant α gene deletion, duplication, or triplication is allowed; documentation of dates of transfusion events and the number of all pRBC units per event within the 12 weeks prior to baseline (day 1) visit; willing and able to complete all study assessments and procedures, and to communicate effectively with the investigator and site staff; transfusion independent, defined as 0 – 3 units of pRBCs received during the 12‐week period prior to baseline (day 1) visit, must not be on a regular transfusion programme, must be RBC transfusion‐free for ≥ 4 weeks prior to randomisation and must not be scheduled to start a regular; haematopoietic stem cell transplantation within 9 months; people with NTDT: must have haemoglobin ≤ 10.0 g/dL at screening, the screening Hb sample must be collected 7 – 28 days prior to randomisation, haemoglobin values within 21 days post‐transfusion will be excluded; ECOG performance score of 0 – 1; females must not be pregnant or breastfeeding and be highly unlikely to become pregnant and males must be unlikely to impregnate a partner. Exclusion criteria (as stated in the protocol for participants with NTDT): diagnosis of α‐thalassaemia (e.g. haemoglobin H) or haemoglobin S/β‐thalassaemia; body mass index < 17.0 kg/m2 or total bodyweight < 45 kg; or BMI > 35 kg/m2; known active hepatitis A, hepatitis B or hepatitis C, with active or acute event of malaria, or who are known to be positive for HIV; stroke requiring medical intervention ≤ 24 weeks prior to randomisation; platelet count > 1000 × 109/L; participated in another clinical study of an investigational agent (or device) within 30 days or 5‐half‐lives of the date of informed consent, whichever is longer, or is currently participating in another study; receiving iron chelation therapy at time of signing consent form, initiation of iron chelation therapy < 24 weeks before predicted randomisation date; prior exposure to sotatercept or luspatercept, IMR‐687 or gene therapy within 6 months prior to randomisation (day 1); people who have had major organ damage.
Interventions	Arm 1: phosphodiesterase‐9‐inhibitor (IMR‐687) dose 1 Arm 2: phosphodiesterase‐9‐inhibitor (IMR‐687) dose 2 Arm 3: placebo All taken once daily for 36 weeks
Outcomes	Primary outcome Events incidence and severity of adverse events Secondary outcomes (measurement levels up to 36 weeks) (as stated in the protocol for participants with NTDT) Increase from baseline of ≥ 1.0 g/dL in mean haemoglobin at weeks 12 – 24 in the absence of transfusions Proportion of participants with an increase from baseline of ≥ 3% in mean HbF values at weeks 12 – 24 in absence of transfusions Proportion of participants with an increase from baseline of ≥ 1.0 g/dL in haemoglobin at weeks 24 – 36 in the absence of transfusions Proportion of participants with an increase from baseline of ≥ 3% in mean HbF values at weeks 24 – 36 in absence of transfusions
Notes	From the information obtained from the entry on ClinicalTrials.gov, we found this study was terminated by the sponsors in March 2022 before scheduled completion in May 2022. Reason given was that IMR‐687 was generally well‐tolerated but failed to show any meaningful benefit in transfusion burden or improvement in most disease‐related biomarkers. Some usable outcome data were uploaded on 6 June 2022, and we are currently contacting the researchers for further information, with the aim of potentially including this study in an update of this review.

Wu 2007.

Methods	RCT Single centre in China Trial date: July 2004 to January 2005 Trial duration: 3 months
Participants	60 people with β‐thalassaemia. 41 have intermediate and 19 have severe type Sex: 41 males, 19 females Inclusion criteria: age 2 – 31 years, haemoglobin levels < 100 g/L, HbF > 20%, reticulocytes 3 – 10%, have not been on any medication or treatment that prevents anaemia for the past 6 months Blood transfusion status: last transfusion ≥ 2 months ago Exclusion criteria: have immune deficiency; liver, kidney, blood system or other primary diseases 6 participants were splenectomised prior to study: 2 from Arm 1 and 4 from Arm 2 Recruited participants aged 2 – 18 years
Interventions	Arm 1: Yisui Shengxue granules (YSSXG) 10 g per tea bag taken daily according to age (n = 30) Arm 2: placebo (dextrin and starch) taken daily in a similar manner (n = 30) Each tea bag of YSSXG contained 10 g granules which main ingredients included shan zhu yu (山茱萸), zhi he shou wu (制何首乌), shu de huang (熟 地 黄) and huang qi (黄芪). Each 10 g of YSSXG granules is equivalent to 23.68 g of raw mixed herbs. aAge 2 – 6 years: half tea bag twice a day aAge 6 – 10 years: 1 tea bag twice a day Age > 10 years: 1 tea bag 3 times a day aPrecise age‐grouping is unclear The tea bags were immersed in warm water and drunk daily according to the schedule above for 3 months. The placebo was packed in similar looking tea bags and was given in the similar manner as YSSXG.
Outcomes	Haemoglobin levels, RBC and haematological parameters (HbF, reticulocyte counts, etc.) measured at baseline, 1 month, 2 months and 3 months postintervention. Liver and spleen size and consistency, measured via ultrasound at baseline and 3 months. Any response to intervention with a cut‐off haemoglobin level of 5 g/L. The responses were then categorised according to Cao 1996. Physical changes categorised according to the severity of: sallow appearance, dizziness, palpitation, lethargy, breathlessness and poor appetite. Criteria for level of severity of each component based on MOH China 1995 and Shen 1986.
Notes	Grants from National Natural Science Foundation of China (No. 90409003 and No. 3017119). No declaration on conflict of interest. Study published in Chinese with an additional abstract in English. No pre‐existing published protocol. Noted unreported baseline discrepancies and found possibility of skewed data for certain outcomes. We are currently attempting to verify the data.

ECOG: Eastern Cooperative Oncology Group; HbE: haemoglobin E; HbF: foetal haemoglobin; NTDT: non‐transfusion‐dependent thalassaemia; pRBC: packed red blood cell; RBC: red blood cell; RCT: randomised controlled trial; SD: standard deviation; TDT: transfusion‐dependent thalassaemia.

Differences between protocol and review

We changed the title of the review from 'Hydroxyurea for reducing blood transfusion in non‐transfusion‐dependent beta thalassaemias' to 'Foetal haemoglobin inducers for reducing blood transfusion in non‐transfusion‐dependent beta thalassaemias' to include all HbF inducers.

We developed the background section on the description of the condition and newer HbF inducers.

We added to the methods section on how we would incorporate dose‐escalation studies, studies with HbF inducers and an active co‐intervention that we considered was not an HbF inducer.

We modified the criteria for assessing measurement (detection) bias. We assessed detection bias separately for objective and subjective outcomes.

We added pulmonary hypertension as one of the assessments for long‐term sequelae of NTDβT (a secondary outcome).

We added that we would use the fixed‐effect model when performing meta‐analysis.

We added haemoglobin to the list of outcomes for the summary of findings tables because the decision for a transfusion is still based on haemoglobin levels.

We added transfusion‐free interval to the list of outcomes for the summary of findings tables.

We identified some HbF inducers according to whether they are pharmacological or natural in origin.

We included two additional tables to provide a summary of some commonly used HbF inducers and their potential adverse effects described in the literature (Table 6; Table 2).

We included the Chinese Databases in the databases searched (Appendix 1).

Contributions of authors

Conceiving the review	All review authors
Designing the review	All review authors
Co‐ordinating the review	WCF, CKL
Data collection for the review	WCF, CKL, DL
Designing search strategies	WCF, CKL, DL, Search Co‐ordinator
Undertaking searches	WCF, CKL, DL, Search Co‐ordinator
Screening search results	WCF, CKL, DL
Organising retrieval of papers	WCF, CKL, DL
Screening retrieved papers against eligibility criteria	WCF, CKL, DL
Appraising quality of papers	All review authors
Extracting data from papers	WCF, CKL, DL
Writing to authors of papers for additional information	WCF, CKL, DL
Providing additional data about papers	WCF, CKL, DL
Obtaining and screening data on unpublished studies	WCF, CKL, DL
Data management for the review	All review authors
Entering data into Review Manager	WCF, CKL, DL
Analysis of data	All review authors
Interpretation of data	All review authors
Providing a methodological perspective	All review authors
Providing a clinical perspective	All review authors
Providing a policy perspective	All review authors
Providing a consumer perspective	None
Writing the review (or protocol)	All review authors
Providing general advice on the review	All review authors
Securing funding for the review	None
Performing previous work that was the foundation of the current review	WCF, CKL, JJH

The contributions of the current review authors are described according to an adapted scheme (Yank 1999). All review authors have discussed and agreed on their respective descriptions of contribution before the protocol or review is submitted for publication. WCF is the contact person for the review. When the review is updated, this section will be checked and revised as necessary to ensure that it is accurate and up to date.

Sources of support

Internal sources

• RCSI & UCD Malaysia Campus (formerly Penang Medical College), Malaysia

  This review is supported by RCSI & UCD Malaysia Campus where WCF and JJH work.

• Universiti Kebangsaan Malaysia, Malaysia

  This review is supported by Universiti Kebangsaan Malaysia where CKL and DL work.

External sources

• National Institute for Health Research, UK

  This systematic review was supported by the National Institute for Health Research, via Cochrane Infrastructure funding to the Cochrane Cystic Fibrosis and Genetic Disorders Group.

Declarations of interest

WCF: is involved in the management of people with non‐transfusion‐dependent thalassaemia and has received small pharmaceutical grants to attend conferences, all of which were unrelated to any HbF inducers.

CKL: is involved in the management of people with non‐transfusion‐dependent thalassaemia and has received small pharmaceutical grants to attend conferences, all of which were unrelated to any HbF inducers.

JJH: none.

DL: is involved in the management of people with non‐transfusion‐dependent thalassaemia and has received small pharmaceutical grants to attend conferences, all of which were unrelated to any HbF inducers.

New