Estimating the proportion of relapse following treatment of Visceral Leishmaniasis: meta-analysis using Infectious Diseases Data Observatory (IDDO) systematic review

Summary

Background

Occurrences of relapse after 6-months post-treatment has been reported in recent Visceral Leishmaniasis (VL) efficacy studies. A meta-analysis was carried out to quantify the proportion of relapses observed at and beyond 6-months using the Infectious Diseases Data Observatory (IDDO) systematic review (SR) database.

Methods

Studies in the IDDO SR database (1983–2021; 160 studies) were eligible for inclusion if follow-up was at least 6-months, relapse was clearly reported, and patients with HIV coinfections were excluded. Meta-analysis of single proportion was undertaken and the estimates were reported with 95% confidence intervals (CI).

Findings

Overall, 131 studies enrolling 27,687 patients were included; 1193 patients relapsed. In the Indian sub-continent (ISC), relapse estimates at 6-months was 4.5% [95% CI: 2.6%–7.5%; I² = 66.2%] following single dose liposomal amphotericin B (L-AmB) and 1.5% [95% CI: 0.7%–3.3%; I² = 0%] for L-AmB in a combination therapy. In East Africa (EA), corresponding estimates were 3.8% [95% CI: 1.3%–10.9%; I² = 75.8%] following pentavalent antimony (PA), and 13.0% [95% CI: 4.3%–33.6%; I² = 0%] for PA + paromomycin. From 21 studies with follow-up longer than 6-months, 0.6% [95% CI: 0.2%–1.8%; I² = 0%] of patients relapsed after 6-months and estimated 27.6% [95% CI: 11.2%–53.4%; I² = 12%] of relapses would have been missed by a 6-month follow-up.

Interpretation

The estimated relapse proportion ranged from 0.5% to 4.5% in ISC and 3.8%–13.0% in EA with the currently recommended drugs. Over one-quarter of relapses would be missed with 6-months follow-up suggesting a longer follow-up may be warranted.

Funding

Wellcome Trust (ref: 208378/Z/17/Z).

Research in context.

Evidence before this study

Therapeutic efficacy in visceral leishmaniasis is assessed by estimating the proportion of patients achieving initial cure and definitive cure. Initial cure is assessed at the end of treatment completion and definitive cure is assessed typically after 6-months post-treatment follow-up. An important outcome assessed during the follow-up is relapse, which is the recurrence of the disease among patients who had achieved initial cure. Studies have shown that relapses can help sustain the transmission of the infection and hence quick identification and treatment of relapse is crucial from a public health standpoint. Most of the studies with immunocompetent patients have reported some proportion of patients relapsing after attaining initial cure. Despite the occurrence and reporting of relapses in VL studies, data on expected distribution of relapse in different endemic regions or for different treatment regimens is lacking.

Research studies conducted in the Indian sub-continent have revealed that the standard 6-month post-treatment follow-up fails to capture a considerable proportion of relapse. The cumulative evidence from clinical trials and observational studies conducted over the past decade underscores the need to reconsider the length of post-treatment follow-up in these regions to capture late relapses.

Added value of this study

This study estimates the baseline proportion of relapses observed in VL clinical trials conducted over the past 40 years at the conventionally adopted follow-up period of 6-months for different drug regimens across different geographical regions. Using studies with longer follow-up, this study reports the proportion of late relapses (occurring after 6 month) and evaluates the adequacy of post-treatment follow-up duration of 6-months. In particular, an estimated 27.6% of the relapses would have been missed with the conventionally adopted 6-months follow-up period.

Implications of all the available evidence

Our results reveal that over a quarter of all relapses are missed by truncating the follow up at 6-months and suggest that longer duration of follow-up might be needed. The estimates presented for treatment regimens across different geographic regions can serve as a benchmark for forthcoming studies.

Introduction

Visceral Leishmaniasis (VL) is a protozoan parasitic disease with 50,000–90,000 new cases occurring annually.¹ Untreated cases of VL are highly fatal in over 95% of cases with a significant reduction in mortality following treatment.²

The efficacy for current drug regimens adopted as first line therapies is characterised by high regional variability.³ The assessment of therapeutic efficacy in VL trials relies on two key endpoints: initial cure and definitive cure (or final cure).³ Initial cure is assessed upon completion of the standard course of a drug regimen and is based on clinical improvement and/or parasitological clearance of the infection. Definitive cure is measured at the end of the post-treatment follow-up (usually at 6-months) and is defined based on the composite of achievement of an initial cure at end of treatment, no death and absence of clinical signs and symptoms of VL during follow-up (i.e., absence of relapse).³ Cumulative evidence from clinical trials and observational studies with extended follow-up conducted over the past decade in the Indian sub-continent (ISC) suggests that the conventionally adopted 6-months post-treatment follow-up fails to capture a substantial proportion of relapse.4, 5, 6 From a public health perspective, identification of relapse is crucial as they are linked with increased infectiousness.⁷ Therefore, monitoring the underlying temporal trend in observed relapse could potentially be used as early signs of drug-resistance,⁸ and prompt treatment of these cases would contribute to the overall reduction of infective pool of parasite in the host and would reduce further transmission of disease.⁹ However, there is a lack of reliable baseline information on expected proportion of relapse at different follow-up time-points for different drug regimens across the endemic regions.

The overall aim of this study is to quantify the proportion of relapse among patients who were initially cured for VL, excluding HIV co-infected patients. The specific objectives were i) to establish a baseline proportion of relapse observed in VL clinical trials at the conventionally adopted follow-up period of 6-months, ii) to quantify the proportion of relapse by drug regimen, region, and nature of infection (primary cases vs previously treated cases), and iii) to assess the adequacy of post-treatment follow-up duration of 6-months to capture the late relapses using data from studies with a follow-up longer than 6-months.

Methods

Information sources and search strategy

All the articles indexed in Infectious Diseases Data Observatory (IDDO) VL library of clinical studies were eligible for inclusion.¹⁰ The IDDO VL library is a living systematic review (LSR) updated bi-annually and searches the following databases: PubMed, Embase, Scopus, Web of Science, Cochrane, clinicaltrials.gov, WHO ICTRP, Global Index Medicus, IMEMR, IMSEAR, and LILACS. Details of the search strategy adopted for each of the databases has been described elsewhere previously (PROSPERO registration: CRD42021284622). The iteration of the systematic review upon which this review is based was updated on Nov-2021 and had indexed 160 published studies from 1983 to 2021.¹⁰

Study selection

Any study indexed in the IDDO VL LSR were eligible for inclusion if the study had a minimum post-treatment follow up duration of 6-months. The exclusion criteria were: i) included patients with HIV co-infections, and ii) did not clearly report the number of patients who achieved initial cure or the number of patients who developed relapse during follow-up period. iii) not reporting relapse numbers at 6-month time-point.

Data extraction

Data were extracted on study location, year, drug regimen and dosage, the number of patients who were enrolled into a study arm, the number of patients who achieved initial cure and the number of relapses occurring among the patients who were declared as initially cured. The time-period of the occurrence of relapse when reported as well as further information on the type of infection of the patients or type of VL infection (primary vs previously treated infections) and the methodology used for defining relapse or relapse diagnosis method (clinical or parasitological) were also extracted. A single reviewer extracted the data initially (SSP) and two reviewers (PD, RC) verified the extracted data on all the studies. REDCap software was used to extract data from the publications.

Statistical analyses

Meta-analysis of single proportion was carried out at study arm level. The proportion of relapse was estimated for the patients who were declared as initially cured at the end of the treatment. A random-effects meta-analysis was undertaken, and the estimates were reported together with the 95% confidence interval (CI). We used a generalised linear mixed model (GLMM), specifically a random intercept logistic regression model, using the logit transformation for the meta-analysis of proportions. Additionally, a fixed-effect meta-analysis was carried out when data were available from less than 5 arms as reliable estimation of between-study heterogeneity remains challenging with few studies.¹¹ Between-study heterogeneity was summarised using Tau squared (Tau²) and I² statistics. Tau squared estimates (Tau²) represents the absolute measure (heterogeneity) and I² represents the proportion of total variability in observed effect due to between-study heterogeneity.¹² Prediction Intervals (95% PI) were also reported to convey the extent of heterogeneity in treatment effects across studies in a clinically relevant context. Hartung-Knapp adjustment was undertaken for estimating the 95% CI. The primary objective was to estimate the proportion of relapse at 6-months. The relapse proportion was estimated by the type of infection (primary vs previously treated infections), geographical regions, drug regimens, and time-period (1980–1989; 1990–1999; 2000–2009; 2010–2021).

In order to gauge the adequacy of the 6-months follow-up to capture relapses, meta-analysis was carried out using subset of studies with follow-up duration longer than 6-months; the proportion of relapses captured at 6-months and between 6 and 12 months were quantified. For single-dose liposomal amphotericin B (L-AmB) and miltefosine-based regimens, odds ratios for relapse comparing combination regimens to monotherapy were calculated using a generalized linear (mixed-effects) model. For L-AmB regimen, which has been introduced in different dosages (single and multiple dosage) in the Indian sub-continent,¹³ meta-regression was carried out to quantify the impact of total weighted adjusted (mg/kg) dosage of L-AmB on relapse, adjusting for type of VL infection and publication year of the study. All relapses reported are for 6-months’ time-point unless specified otherwise.

All statistical analyses were carried out using “R” statistical software: meta-analysis of proportion was carried out using the metaprop function in the meta package using logit transformation.¹⁴

Sensitivity analysis and risk of bias assessment

Small study effects (selection bias) were explored using Egger's test and adjusted using Copas selection approach (or trim-and-fill approach was used in cases of non-convergence of Copas model).¹⁵ Risk of bias in randomised studies was assessed using Cochrane risk of bias tool and in non-randomised studies based on Newcastle-Ottawa scale. PRISMA guidelines were used for reporting the results of this meta-analysis.¹⁶

Role of funding source

The review was funded by a biomedical resource grant from Wellcome to the Infectious Diseases Data Observatory (Recipient: PJG; ref: 208378/Z/17/Z) and the University of Oxford. The funders had no role in the design and analysis of the research or the decision to publish the work.

Results

Of the 160 studies indexed in the IDDO VL LSR, 29 studies were excluded. The reasons for exclusion were lack of information on either the number of cured patients or the number of relapses (n = 3 studies), inclusion of patients with VL-HIV coinfection (n = 10 studies), follow-up < 6-months/follow-up data not reported/no active treatment follow-up (n = 12 studies) or unclear reporting of data/efficacy outcomes (n = 4 studies). Data from a total of 276 arms from the eligible 131 studies (25,718 patients who achieved initial cure) were included in this meta-analysis (Fig. 1).

Fig. 1.

PRISMA flowchart.

Of the 131 studies, 101 (77.1%) were from the Indian sub-continent (ISC) (India, Nepal and Bangladesh), 14 (10.7%) from East Africa (EA), 5 (3.8%) from the Mediterranean region, 6 (4.6%) from Latin America, 3 (2.3%) from Central Asia, and 2 (1.5%) were multi-regional. In 21 (16%) studies, patient follow-up was longer than 6-months with relapse numbers reported at 6-months and 12-months. Confirmation of relapse differed across the studies; 105 (80.2%) studies required parasitological confirmation of relapse, 5 (3.8%) used only clinical suspicion, 1 (0.8%) used molecular (PCR) method, 1 (0.8%) used parasitological or serological method and 19 (14.5%) did not clearly define the methodology adopted for confirming relapse (Table 1). Overall, 17 different drug regimens were used across the included studies with amphotericin B, liposomal amphotericin B (L-AmB), miltefosine, paromomycin and sodium stibogluconate being the major ones (Supplemental File 1: Screening and extracted data).

Table 1.

Meta-analysis of relapse following treatment.

	Number of study arms	Total treated/Initially Cured/Relapsed	Proportion of relapse Random effects estimate [95% confidence interval]	Tau Squared	I2	Bias adjusted estimate [95% confidence interval]a	95% prediction interval
Region
Indian sub-continent	209	24,667/23,196/1002	3% [95% CI: 2.3%–3.7%]	2	65.7%	8.4% [95% CI: 7.2%-9.9%]	0.2%–32.6%
Eastern Africa	29	1864/1446/78	5.9% [95% CI: 3.5%–9.8%]	1.2	57.8%	9.1% [95% CI: 6.2%–13%]	0.6%–39.6%
South America	12	518/457/20	4% [95% CI: 1.4%–11.1%]	1.5	56.1%	6.8% [95% CI: 3.3%–13.5%]	0.2%–44.9%
Central Asia	7	170/170/2	1.2% [95% CI: 0.2%–6.3%]	0	0%	2.7% [95% CI: 1.1%-6.7%]	0.2%–6.9%
Mediterranean region	11	293/287/10	3.5% [95% CI: 1.7%–6.9%]	0	0%	5.4% [95% CI: 3.1%-9.3%]	1.7%–7%
Multi-regional	8	175/162/11	6.8% [95% CI: 3.3%–13.5%]	0	0%	9% [95% CI: 5.2%–15.3%]	3.2%–13.8%
Drug regimen
Amphotericin B (fat/lipid/colloid/cholesterol)	28	1302/1264/92	6.5% [95% CI: 4.1%–10%]	0.7	32.1%	9.9% [95% CI: 7.3%–13.4%]	1.1%–29.1%
Amphotericin B deoxycholate	30	4631/4530/31	0.5% [95% CI: 0.3%–1.1%]	1	11.9%	1.2% [95% CI: 0.8%–1.8%]	0.1%–4.4%
Amphotericin B Unknown	14	1189/1150/1	0% [95% CI: 0%–8.7%]	1.3	0%	1% [95% CI: 0.5%–2.1%]	0%–14.8%
Liposomal Amphotericin B (Multiple dose) in combination regimen	1	72/63/0	0% [95% CI: 0%–5.7%]e	–	–	–	–
Liposomal Amphotericin B (Multiple dose)	34	1294/1258/45	3.2% [95% CI: 1.8%–5.7%]	1.3	0%	10.2% [95% CI: 7%–14.7%]	0.3%–26.4%
Liposomal Amphotericin B (Single dose) in combination regimen	13	1505/1458/35	1.7% [95% CI: 0.8%–3.7%]	0.9	47.5%	3.9% [95% CI: 2.2%–6.7%]	0.2%–13.7%
Liposomal Amphotericin B (Single dose)	16	3261/3145/123	4.5% [95% CI: 2.6%–7.5%]	0.6	66.2%	5.4% [95% CI: 3.3%–8.7%]	0.8%–21.2%
Miltefosine & Paromomycin	3	813/802/4	0.5% [95% CI: 0.2%–1.3%]b 0.5% [95% CI: 0.1%–4.2%]c	0	0%	0.7% [95% CI: 0.3%–1.6%]	0%–74.5%
Miltefosine	30	5932/5443/368	5.8% [95% CI: 4.3%–7.8%]	0.3	65%	7.2% [95% CI: 5.6%–9.3%]	1.8%–17.2%
Paromomycin	14	1685/1625/94	6.6% [95% CI: 4.6%–9.5%]	0.2	54.7%	7.2% [95% CI: 5.2%–10%]	2.6%–15.9%
Pentamidine	4	189/157/17	10.8% [95% CI: 6.8%–16.7%]b 10.8% [95% CI: 5.1%–21.6%]c	0	21.8%	12.1% [95% CI: 7.6%–18.6%]	3.9%–26.8%
Pentavalent antimonial	53	4535/3787/206	4.1% [95% CI: 2.5%–6.5%]	2.2	58.7%	14.3% [95% CI: 11.2%–18.1%]	0.2%–46%
Pentavalent antimonial & Paromomycin	10	283/230/14	6.1% [95% CI: 3.2%–11.2%]	0.1	0%	8.4% [95% CI: 5.1%–13.7%]	2.5%–14.2%
Pentavalent antimonial combination	6	314/291/32	8.9% [95% CI: 3.3%–21.7%]	0.3	2.9%	13% [95% CI: 9.4%–17.8%]	1.6%–36.9%
Sitamaquine	12	296/239/16	6.7% [95% CI: 3.9%–11.3%]	0	0%	8.7% [95% CI: 5.5%–13.5%]	3.9%–11.3%
Type of infection
Primary infection	134	15,794/14,332/648	3.4% [95% CI: 2.6%–4.4%]	1.6	66.8%	7.7% [95% CI: 6.1%–9.6%]	0.3%–30.6%
Previously treated (Unresponsive/relapses)	43	2628/2548/114	2.1% [95% CI: 1%–4.3%]	3	28.7%	12.1% [95% CI: 9.3%–15.5%]	0.1%–43.2%
Mixture of primary and previously treated	37	2464/2388/97	4% [95% CI: 2.7%–6.1%]	0.8	54.6%	5.6% [95% CI: 4%–7.9%]	0.7%–21.3%
Unclear	62	6801/6450/264	2.9% [95% CI: 1.8%–4.7%]	2.3	59.4%	8.7% [95% CI: 6.1%–12.2%]d	0.1%–39.7%
Relapse method
Solely based on clinical suspicion	10	460/417/10	2.3% [95% CI: 0.9%–6%]	0	0%	3.5% [95% CI: 2.1%–5.9%]	0.8%–6.9%
Parasitological demonstration	221	23,794/22,167/911	3.1% [95% CI: 2.4%–3.9%]	2	63.9%	8.4% [95% CI: 7.2%–9.8%]	0.2%–34.2%
Parasitological/Serological	3	75/75/1	1.3% [95% CI: 0.2%–8.9%]b 1.3% [95% CI: 0%–50.7%]c	0	0%	2.8% [95% CI: 0.7%–10.4%]	0%–100%
Molecular	2	46/45/3	6.7% [95% CI: 2.2%–18.7%]b 6.7% [95% CI: 2.2%–18.7%]c	0	0%	–	–
Diagnosis Unclear	40	3312/3014/198	4.9% [95% CI: 3.3%–7.2%]	0.9	59%	8.3% [95% CI: 6.2%–10.9%]	0.7%–27%
Studies with follow-up longer than 6-months (n = 21 studies)
Overall at 12-months	34	4082/3893/204	1.8% [95% CI: 0.7%–4.5%]	3	57%	6.9% [95% CI: 4.5%–10.3%]	0%–41.5%
Estimates at 6 months	34	4082/3893/139	0.9% [95% CI: 0.3%–3.2%]	4.4	44.5%	6.3% [95% CI: 4.2%–9.3%]	0%–44.8%
Estimates at 12 months	34	4082/3893/65	0.6% [95% CI: 0.2%–1.8%]	1.8	0%	2.8% [95% CI: 1.9%–4.1%]	0%–10.2%
Time-period
1980 to ≤ 1989	15	607/547/59	9.1% [95% CI: 5.1%–15.8%]	0.7	0%	14.9% [95% CI: 11.5%–19.2%]	1.5%–39.6%
1990 to ≤ 1999	108	5661/5310/251	3.7% [95% CI: 2.6%–5.3%]	2	37.1%	14.4% [95% CI: 11.3%–18.1%]	0.2%–39.9%
2000 to ≤ 2009	91	9180/8197/297	3.1% [95% CI: 2.3%–4.1%]	1.1	60%	6.8% [95% CI: 5.6%–8.2%]	0.4%–21.5%
2010 to ≤ 2021	62	12,239/11,664/516	2.2% [95% CI: 1.4%–3.4%]	2.1	70%	6.6% [95% CI: 4.5%–9.4%]d	0.1%–30%

CI = confidence interval.

^aBias adjusted estimate derived using Copas selection model.

^bEstimate reported from fixed effect meta-analysis when there are <5 studies.

^cEstimate from random-effects meta-analysis.

^dBias adjusted estimates could not be derived through Copas selection model and adjusted estimates using trim and fill are reported instead.

^e95% confidence interval derived using Wilson's method.

Overall estimates from meta-analysis of single proportion

Overall, 27,687 treated patients had follow-up information available, out of which 25,718 were declared initially cured. A total of 1193 (4.3%) relapse were reported during follow-up period up to the last study visit. From meta-analysis, the overall proportion of relapse at 6-months was 3.2% [95% CI: 2.6%–3.9%; Tau² = 1.77; I² = 61.8%; 95% PI: 0.2%–31.4%]. After stratifying by geographical region, the estimate of relapse at 6-months was 3.0% [95% CI: 2.3%–3.7%; Tau² = 1.95; I² = 65.7%; 95% PI: 0.2%–32.6%] in the ISC, 5.9% [95% CI: 3.5%–9.8%; Tau² = 1.24; I² = 57.8%; 95% PI: 0.6%–39.6%] in EA, 3.5% [95% CI: 1.7%–6.9%; Tau² = 0; I² = 0%; 95% PI: 1.7%–7.0%] in the Mediterranean region, 4.0% [95% CI: 1.4%–11.1%; Tau² = 1.53; I² = 56.1%; 95% PI: 0.2%–44.9%] in South America, and 1.2% [95% CI: 0.2%–6.3%; Tau² = 0; I² = 0%; 95% PI: 0.2%–6.9%] in Central Asia (Table 1).

Estimates for each drug regimens across geographical regions

Indian sub-continent (n = 101 studies, 209 study arms)

In the ISC, when stratified by drug regimens, the estimates of relapse at 6-months were: 5.8% [95% CI: 3.3%–10.1%; Tau² = 1.8; I² = 51.4%; 95% PI: 0.4%–51.7%] for pentavalent antimony (PA), 4.5% [95% CI: 2.6%–7.5%; Tau² = 0.6; I² = 66.2%; 95% PI: 0.8%–21.2%] for single dose L-AmB, 1.5% [95% CI: 0.7%–3.3%; Tau² = 0.4; I² = 0%; 95% PI: 0.3%–7.1%] for L-AmB as a combination therapy (L-AmB + miltefosine or L-AmB + paromomycin), and 0.5% [95% CI: 0.2%–1.3%; Tau² = 0; I² = 0%; 95% PI: 0%–74.5%] for miltefosine and paromomycin combination (Table 2). For single dose L-AmB, the odds ratio of relapse was 0.32 [95% CI: 0.14–0.84] when it was adopted as a combination regimen compared to when it was used as a monotherapy. For miltefosine based regimens, the odds ratio of relapse was 0.08 [95% CI: 0.02–0.28] for miltefosine and paromomycin combination regimen as compared to monotherapy.

Table 2.

Estimates for sub-group meta-analysis in studies from Indian sub-continent.

	Number of study arms	Total treated/Initially Cured/Relapsed	Proportion of relapse Random effects estimate [95% confidence interval]	Tau Squared	I2	Bias adjusted estimate [95% confidence interval]a	95% prediction interval
Indian sub-continent	209	24,667/23,196/1002	3% [95% CI: 2.3%–3.7%]	2	65.7%	8.4% [95% CI: 7.2%–9.9%]	0.2%–32.6%
Drug regimen
Liposomal Amphotericin B (Single dose)	16	3261/3145/123	4.5% [95% CI: 2.6%–7.5%]	0.6	66.2%	5.4% [95% CI: 3.3%–8.7%]	0.8%–21.2%
Amphotericin B deoxycholate	29	4581/4482/30	0.5% [95% CI: 0.2%–1.1%]	1	14.7%	1.2% [95% CI: 0.8%–1.8%]	0.1%–4.5%
Amphotericin B Unknown	14	1189/1150/1	0% [95% CI: 0%–8.7%]	1.3	0%	1% [95% CI: 0.5%–2.1%]	0%–14.8%
Liposomal Amphotericin B (Single dose) in combination regimen	10	1293/1265/26	1.5% [95% CI: 0.7%–3.3%]	0.4	0%	3.7% [95% CI: 2.4%–5.6%]	0.3%–7.1%
Pentavalent antimonial	29	2954/2440/162	5.8% [95% CI: 3.3%–10.1%]	1.8	51.4%	–	0.4%–51.7%
Miltefosine	27	5561/5246/352	5.5% [95% CI: 4%–7.7%]	0.3	65.5%	6.9% [95% CI: 5.2%–9%]	1.7%–16.8%
Miltefosine & Paromomycin	3	813/802/4	0.5% [95% CI: 0.2%–1.3%]b 0.5% [95% CI: 0.1%–4.2%]c	0	0%	0.7% [95% CI: 0.3%–1.6%]	0%–74.5%
Sitamaquine	5	161/138/6	4.3% [95% CI: 1.4%–12.7%]	0	0%	5.4% [95% CI: 2.5%–11.3%]	1.2%–14.6%
Amphotericin B (fat/lipid/colloid/cholesterol)	24	1260/1223/91	7% [95% CI: 4.4%–10.8%]	0.7	42.2%	11.9% [95% CI: 9.3%–15.1%]	1.2%–31.3%
Pentamidine	4	189/157/17	10.8% [95% CI: 6.8%–16.7%]b 10.8% [95% CI: 5.1%–21.6%]c	0	21.8%	12.1% [95% CI: 7.6%–18.6%]	3.9%–26.8%
Pentavalent antimonial & Paromomycin	9	260/207/11	5.3% [95% CI: 2.7%–10.3%]	0	0%	7.5% [95% CI: 4.2%–13%]	2.6%–10.5%
Liposomal Amphotericin B (Multiple dose)	13	749/739/14	1.5% [95% CI: 0.4%–5%]	1.8	0%	6.6% [95% CI: 3.9%–11%]	0.1%–27.1%
Pentavalent antimonial combination	6	314/291/32	8.9% [95% CI: 3.3%–21.7%]	0.3	2.9%	13% [95% CI: 9.4%–17.8%]	1.6%–36.9%
Paromomycin	11	1624/1572/88	6.2% [95% CI: 4.2%–9.2%]	0.2	60.7%	6.7% [95% CI: 4.7%–9.5%]	2.4%–15.5%
Liposomal Amphotericin B (Multiple dose) in combination regimen	1	72/63/0	0% [95% CI: 0%–5.7%]d	–	–	–	–
Type of infection
Primary infection	96	13,408/12,408/575	3–3% [95% CI: 2.4%–4.6%]	1.8	71.1%	7.6% [95% CI: 5.8%–10%]	0.2%–33.4%
Previously treated (Unresponsive/relapses)	43	2628/2548/114	2.1% [95% CI: 1%–4.3%]	3	28.7%	12.1% [95% CI: 9.3%–15.5%]	0.1%–43.2%
Mixture of primary and previously treated	32	2412/2336/90	3.8% [95% CI: 2.6%–5.6%]	0.6	48.4%	5% [95% CI: 3.6%–7%]	0.8%–16.1%
Unclear	38	6219/5904/223	2% [95% CI: 1%–4.3%]	3.3	69.9%	8.4% [95% CI: 5.8%–12.2%]	0%–47.9%
Relapse method
Parasitological demonstration	172	21,488/20,303/808	2.6% [95% CI: 2%–3.5%]	2.2	66.2%	11.7% [95% CI: 9.1%–14.9%]	0.1%–34%
Molecular	2	46/45/3	6.7% [95% CI: 2.2%–18.7%]b 6.7% [95% CI: 2.2%–18.7%]c	0	0%	–	–
Diagnosis Unclear	38	6219/5904/223	2% [95% CI: 1%–4.3%]	3.3	69.9%	8.4% [95% CI: 5.8%–12.2%]	0%–47.9%
Dose response for Liposomal Amphotericin B in Indian sub-continent
Total dose: 5 mg/kg	3	96/95/8	8.4% [95% CI: 4.3%–15.9%]b 8.4% [95% CI: 1.8%–31.1%]c	0	59.4%	7.7% [95% CI: 3.8%–15%]	0.1%–91%
Total dose: 7.5 mg/kg	2	213/205/17	8.3% [95% CI: 5.2%–12.9%]b 8.3% [95% CI: 5.2%–12.9%]c	1.6	93.4%	–	–
Total dose: 10 mg/kg	7	2783/2678/95	3.5% [95% CI: 2.8%–4.5%]	0	0%	–	2.7%–4.6%
Total dose: 15 mg/kg	4	169/167/3	1.8% [95% CI: 0.6%–5.4%]b 1.8% [95% CI: 0.3%–10.5%]c	0	0%	2.7% [95% CI: 1%–7%]	0.1%–18.3%
Time-period
1980 to ≤ 1989	9	507/458/50	8.8% [95% CI: 3.6%–19.9%]	1	0%	16.8% [95% CI: 12.1%–22.8%]	0.7%–56.4%
1990 to ≤ 1999	79	5089/4768/205	3.4% [95% CI: 2.2%–5.2%]	2.2	43.8%	12% [95% CI: 9.3%–15.4%]	0.2%–41.6%
2000 to ≤ 2009	73	7564/6972/265	3.1% [95% CI: 2.3%–4.3%]	1.2	63.1%	9.2% [95% CI: 7.5%–11.3%]	0.4%–22.8%
2010 to ≤ 2021	48	11,507/10,998/482	1.9% [95% CI: 1.1%–3.2%]	2.5	73.6%	93.2% [95% CI: 19%–99.9%]	0.1%–32.3%
Time trend for PA regimen
PA: 1980 to ≤ 1989	8	497/452/49	8.3% [95% CI: 3%–21.1%]	1.2	0%	5.9% [95% CI: 4.1%–8.4%]	0.5%–62.5%
PA: 1990 to ≤ 1999	12	1398/1303/58	5% [95% CI: 1.9%–12.6%]	1.9	41.2%	12% [95% CI: 8.7%–16.4%]	0.2%–57.5%
PA: 2000 to ≤ 2009	8	861/518/55	8.4% [95% CI: 3.4%–19.1%]	0.9	77.1%	14% [95% CI: 8.5%–22.1%]	0.8%–52.2%

CI = confidence interval.

^aBias adjusted estimate derived using Copas selection model.

^bEstimate reported from fixed effect meta-analysis when there are <5 studies.

^cEstimate from random-effects meta-analysis.

^d95% confidence interval derived using Wilson's method.

For the L-AmB regimen, the overall relapse at 6-months was 8.4% [95% CI: 4.3%–15.9%; Tau² = 0; I² = 59.4%; 95% PI: 0.1%–91%] for patients who received 5 mg/kg, 8.3% [95% CI: 5.2%–12.9%; Tau² = 1.6; I² = 93.4%] for patients who received 7.5 mg/kg, 3.5% [95% CI: 2.8%–4.5%; Tau² = 0; I² = 0%; 95% PI: 2.7%–4.6%] among those who received 10 mg/kg, and 1.8% [95% CI: 0.6%–5.4%; Tau² = 0; I² = 0%; 95% PI: 0.1%–18.3%] among those who received 15 mg/kg (Table 2, Fig. 2). In a meta-regression analysis, the odds ratio of relapse was 0.81 [95% CI: 0.72–0.91; P-value = 0.036] for every 1 unit increase in total L-AmB dosage (in mg/kg).

Fig. 2.

Estimate of relapse and 95% confidence interval at 6-months following different total dosage of single dose L-AmB regimen in Indian sub-continent. Legend: Further details are presented in Table 2.

For PA, the relapse estimate was 8.3% [95% CI: 3%–21.1%; Tau² = 1.2; I² = 0%; 95% PI: 0.5%–62.5%] during the period 1980–1989, 5.0% [95% CI: 1.9%–12.6%; Tau² = 1.9; I² = 41.2%; 95% PI: 0.2%–57.5%] for 1990–1999 and 8.4% [95% CI: 3.4%–19.1%; Tau² = 0.9; I² = 77.1%; 95% PI: 0.8%–52.2%] for 2000–2009 (Table 2). The odds ratio (OR) of relapse for PA regimens in ISC compared to L-AmB was 2.84 [95% CI: 1.42–5.68; P-value = 0.003].

Eastern Africa (n = 14 studies, 29 study arms)

In EA, the overall estimate of relapse at 6-months was 5.9% [95% CI: 3.5%–9.8%; Tau² = 1.2; I² = 57.8%; 95% PI: 0.6%–39.6%]. The estimates by drug regimen were: 8.6% [95% CI: 4.4%–16.3%; Tau² = 0.6; I² = 73.3%] for single dose L-AmB (10 mg/kg) in a combination regimen (2 arms, 100 patients, overall 8 relapses), 3.8% [95% CI: 1.3%–10.9%; Tau² = 2.7; I² = 75.8%; 95% PI: 0.1%–62.8%] for PA, 11.3% [95% CI: 5.2%–23%; Tau² = 0; I² = 0%; 95% PI: 0%–96.9%] for paromomycin, 8.1% [95% CI: 5%–12.8%; Tau² = 0.3; I² = 69.7%; 95% PI: 0%–99.7%] for miltefosine, and 13.0% [95% CI: 4.3%–33.6%] for PA + paromomycin regimen (Table 3).

Table 3.

Estimates for sub-group meta-analysis in studies from Eastern Africa.

	Number of study arms	Total treated/Initially Cured/Relapsed	Proportion of relapse Random effects estimate [95% confidence interval]	Tau Squared	I2	Bias adjusted estimatea [95% confidence interval]	95% prediction interval
Eastern Africa	29	1864/1446/78	5.9% [95% CI: 3.5%–9.8%]	1.2	57.8%	9.1% [95% CI: 6.2%–13%]	0.6%–39.6%
Drug regimen
Liposomal Amphotericin B (Single dose) in combination regimen	2	100/93/8	8.6% [95% CI: 4.4%–16.3%]b 8.6% [95% CI: 4.4%–16.3%]c	0.6	73.3%	–	–
Pentavalent antimonial	14	1196/988/37	3.8% [95% CI: 1.3%–10.9%]	2.7	75.8%	8.2% [95% CI: 4.2%–15.5%]	0.1%–62.8%
Miltefosine	3	371/197/16	8.1% [95% CI: 5%–12.8%]b 8.5% [95% CI: 1.7%–33.2%]c	0.3	69.7%	9% [95% CI: 4.2%–18.2%]	0%–99.7%
Sitamaquine	6	113/92/8	8.7% [95% CI: 3.5%–19.8%]	0	0%	10.1% [95% CI: 5.3%–18.3%]	3.3%–21%
Pentavalent antimonial & Paromomycin	1	23/23/3	13% [95% CI: 4.3%–33.6%]d	–	–	–	–
Paromomycin	3	61/53/6	11.3% [95% CI: 5.2%–23%]b 11.3% [95% CI: 1.9%–45.2%]c	0	0%	12% [95% CI: 5.5%–24.2%]	0%–96.9%
Type of infection
Primary infection	24	1702/1294/56	5% [95% CI: 2.7%–9.2%]	1.5	59%	8.3% [95% CI: 5.3%–12.6%]	0.4%–41.2%
Unclear	5	162/152/22	13.4% [95% CI: 6%–27.5%]	0.1	18.1%	14.4% [95% CI: 8%–24.6%]	3.4%–40.7%
Relapse method
Parasitological demonstration	27	1818/1408/74	5.6% [95% CI: 3.2%–9.6%]	1.3	59.8%	13% [95% CI: 8.9%–18.7%]	0.5%–40.6%
Diagnosis Unclear	5	162/152/22	13.4% [95% CI: 6%–27.5%]	0.1	18.1%	14.4% [95% CI: 8%–24.6%]	3.4%–40.7%
Time-Period
1980 to ≤ 1989	6	100/89/9	10.1% [95% CI: 4.4%–21.7%]	0	0%	12.4% [95% CI: 6.8%–21.7%]	4.1%–23%
1990 to ≤ 1999	6	134/119/22	18.5% [95% CI: 11%–29.4%]	0	0%	20.5% [95% CI: 13.9%–29.1%]	10.5%–30.4%
2000 to ≤ 2009	11	1407/1034/24	2.6% [95% CI: 1%–6.2%]	1.2	54.4%	3.7% [95% CI: 1.8%–7.1%]	0.2%–26.2%
2010 to ≤ 2021	6	223/204/23	10.8% [95% CI: 5.4%–20.2%]	0.1	14.1%	14.2% [95% CI: 9.1%–21.6%]	3.1%–31.1%

CI = confidence interval; PA = Pentavalent Antimony.

^aBias adjusted estimate derived using Copas selection model.

^bEstimate reported from fixed effect meta-analysis when there are <5 studies.

^cEstimate from random-effects meta-analysis.

^d95% confidence interval derived using Wilson's method.

The majority of the studies from EA included in this analysis (except for PA) had small sample sizes (median sample size per arm was 21, interquartile range (IQR) = 12–50; n = 29 arms). There was data from only a single study on PA and paromomycin combination regimen at the time of this analysis—the current first line regimen in use in EA. Further detailed breakdown by drug regimen and other characteristics are presented in Table 3. The overall relapse estimate was 10.1% [95% CI: 4.4%–21.7%; Tau² = 0; I² = 0%; 95% PI: 4.1%–23%] during the period 1980–1989, 18.5% [95% CI: 11%–29.4%; Tau² = 0; I² = 0%; 95% PI: 10.5%–30.4%] for 1990–1999, 2.6% [95% CI: 1%–6.2%; Tau² = 1.2; I² = 54.4%; 95% PI: 0.2%–26.2%] for 2000–2009 period and 10.8% [95% CI: 5.4%–20.2%; Tau² = 0.1; I² = 14.1%; 95% PI: 3.1%–31.1%] for 2010–2021 period (Table 3).

South America (n = 6 studies, 12 study arms)

The overall estimate of relapse at 6-months was 4.0% [95% CI: 1.4%–11.1%; Tau² = 1.5; I² = 56.1%; 95% PI: 0.2%–44.9%] in South America; all the studies were from Brazil. The drug-specific estimates were 11.4% [95% CI: 1.8%–46.9%; Tau² = 0.82; I² = 74%; 95% PI: 0.1%–93.3%] for multiple dose L-AmB regimen and 3.5% [95% CI: 0.5%–20.8%] for amphotericin B in cholesterol dispersion (ABCD). Further details are presented in Supplemental Table S1.

The Mediterranean region (n = 5 studies, 11 study arms)

The overall estimate of relapse at 6-months from 5 studies in the Mediterranean region was 3.5% [95% CI: 1.7%–6.9%; Tau² = 0; I² = 0%; 95% PI: 1.7%–6.9%]. The drug-specific estimates were 3.2% [95% CI: 1.0%–9.5%; Tau² = 0.36; I² = 0%; 95% PI: 0.5%–17.5%] for multidose L-AmB regimen (9 arms; 4 studies) and 4.1% [95% CI: 1.1%–14.9%] for PA from a single study. Further details are presented in Supplemental Table S1.

Other regions

In central Asia, the overall estimate of relapse from 3 studies (all from Iran and all were treated with PA; 7 study arms) was 1.2% [95% CI: 0.2%–6.4%; Tau² = 0; I² = 0%; 95% PI: 0.2%–6.9%]. From 2 multi-regional studies (8 study arms), the overall estimate of relapse was 6.8% [95% CI: 3.3%–13.5%; Tau² = 0; I² = 0.0%; 95% PI: 3.2%–13.8%]; both were dose finding studies with multiple dose L-AmB regimen.^17,18 The first study was conducted across India, Brazil and Kenya tested L-AmB total dosage varying from 12 mg/kg to 25 mg/kg with a total of 8 relapses observed; the relapse proportion was 10.7% [95% CI: 3.5%–28.2%].¹⁷ The second was a multi-centre study conducted across Italy, Brazil and the United Kingdom and treated patients with total dose of L-AmB ranging from 6 mg/kg to 20 mg/kg with a total of 3 relapses; the relapse proportion was 3.5% [95% CI: 0.6%–18.8%].¹⁸

Relapses occurring after 6-months of post-treatment follow-up (n = 21 studies)

Overall, 24 studies followed up patients for more than 6-months and only 21 of them reported aggregated number of relapses at 6-months and at 12-months (Table 4). From the 21 studies, a total of 3893 patients were identified as initially cured of whom 204 relapsed subsequently. Of these 204 relapses, 139 were captured by 6-months, and the remaining 65 relapses occurred between 6 and 12 months (Table 4).

Table 4.

List of studies with follow-up duration longer than 6 months and proportion of relapse among initially cured.

IDDO tag	Study	Region	Drug arm	Total treated	Total initially cured	Total relapse	Relapses at 6 month	Relapses at 12 month	Relapse proportion at 6 month	Relapse proportion at 12 month
65	Dietze-2001	South America	Sitamaquine	22	9	3	2	1	22.2% [95% CI: 2.8%–60%]	11.1% [95% CI: 0.3%–48.2%]
56	Rijal 2013a	Indian sub-continent	Miltefosine	120	115	24	13	11	11.3% [95% CI: 6.2%–18.6%]	9.6% [95% CI: 4.9%–16.5%]
64	Sherwood-1994	Eastern Africa	Sitamaquine	16	5	0	0	0	8.3% [95% CI: 0%–52.2%]	8.3% [95% CI: 0%–52.2%]
82	Dietze 1993	South America	Amphotericin b (fat/lipid/colloid/cholesterol)	10	10	0	0	0	4.5% [95% CI: 0%–30.8%]	4.5% [95% CI: 0%–30.8%]
84	Dietze-1995	South America	Amphotericin b (fat/lipid/colloid/cholesterol)	10	10	1	1	0	10% [95% CI: 0.3%–44.5%]	4.5% [95% CI: 0%–30.8%]
97	Maji-2012	Indian sub-continent	Miltefosine	71	70	4	2	2	2.9% [95% CI: 0.3%–9.9%]	2.9% [95% CI: 0.3%–9.9%]
147	Sundar-2019	Indian sub-continent	Liposomal Amphotericin B (Single dose)	1143	1088	66	34	32	3.1% [95% CI: 2.2%–4.3%]	2.9% [95% CI: 2%–4.1%]
59	Davidson 1994	Mediterranean	LamB (Multiple dose) Mono regimen	20	20	0	0	0	2.4% [95% CI: 0%–16.8%]	2.4% [95% CI: 0%–16.8%]
126	Giri-1994a	Indian sub-continent	Amphotericin B Unknown	25	25	0	0	0	1.9% [95% CI: 0%–13.7%]	1.9% [95% CI: 0%–13.7%]
4	Giri-1993	Indian sub-continent	Amphotericin B Unknown	27	27	0	0	0	1.8% [95% CI: 0%–12.8%]	1.8% [95% CI: 0%–12.8%]
22	Chulay-1983	Eastern Africa	Pentavalent antimonial	30	28	2	2	0	7.1% [95% CI: 0.9%–23.5%]	1.7% [95% CI: 0%–12.3%]
95	Thakur-1996b	Indian sub-continent	Liposomal Amphotericin B (Multiple dose)	30	30	0	0	0	1.6% [95% CI: 0%–11.6%]	1.6% [95% CI: 0%–11.6%]
129	Ostyn-2014	Indian sub-continent	Miltefosine	1016	947	78	64	14	6.8% [95% CI: 5.2%–8.5%]	1.5% [95% CI: 0.8%–2.5%]
55	Thakur-2010	Indian sub-continent	Amphotericin B deoxycholate	230	230	3	0	3	0.2% [95% CI: 0%–1.6%]	1.3% [95% CI: 0.3%–3.8%]
93	Mishra 1994	Indian sub-continent	Amphotericin B Unknown & Pentavalent antimonial	40	40	0	0	0	1.2% [95% CI: 0%–8.8%]	1.2% [95% CI: 0%–8.8%]
94	Rijal 2010	Indian sub-continent	Pentavalent antimonial	198	167	2	0	2	0.3% [95% CI: 0%–2.2%]	1.2% [95% CI: 0.1%–4.3%]
58	Nyakundi 1994	Eastern Africa	Pentavalent antimonial	65	65	14	14	0	21.5% [95% CI: 12.3%–33.5%]	0.8% [95% CI: 0%–5.5%]
73	Davidson-1996	Multi-Regional	Liposomal Amphotericin B (Multiple dose)	88	87	3	3	0	3.4% [95% CI: 0.7%–9.7%]	0.6% [95% CI: 0%–4.2%]
109	Di Martino-1997	Mediterranean	Liposomal Amphotericin B (Multiple dose)	106	105	4	4	0	3.8% [95% CI: 1%–9.5%]	0.5% [95% CI: 0%–3.5%]
124	Giri-1994b	Indian sub-continent	Amphotericin B deoxycholate	100	100	0	0	0	0.5% [95% CI: 0%–3.6%]	0.5% [95% CI: 0%–3.6%]
116	Chowdhury-1991	Indian sub-continent	Pentavalent antimony	715	715	0	0	0	0.1% [95% CI: 0%–0.5%]	0.1% [95% CI: 0%–0.5%]

From meta-analysis of these 21 studies, the total relapse proportion was 1.8% [95% CI: 0.7%–4.5%; Tau² = 3; I² = 57%; 95% PI: 0%–41.5%] at 12-months. The estimated proportion of relapse at 6-months was 0.9% [95% CI: 0.3%–3.2%; Tau² = 4.4; I² = 44.5%; 95% PI: 0%–44.8%]; and a further 0.6% [95% CI: 0.2%–1.8%; Tau² = 1.8; I² = 0%; PI: 0%–10.2%] of the patients experienced relapse between 6 and 12 months (See Tables 1 and 4). Among the patients who had a relapse (n = 204 relapses) by 12-months, a 6-month follow-up was estimated to miss 27.6% [95% CI: 11.2.1–53.4%; Tau² = 1.7; I² = 12%; 95% PI: 1.8%–88.7%] of the total relapses compared to a 12-month of follow-up.

Further sub-group meta-analyses

Relapses by type of infection

Overall, 32 (24.4%) studies did not mention the type of VL patients enrolled as either primary or secondary (relapse or previously treated) cases, 62 (47.3%) studies enrolled only primary VL cases, 24 (18.3%) enrolled previously treated patients (secondary cases) and 17 (12.9%) studies had a mixture of primary and previously treated cases. The estimate of relapse at 6-months was 2.1% [95% CI: 1%–4.3%; Tau² = 3.0; I² = 28.7%; 95% PI: 0.1%–43.2%] among previously treated cases, 3.4% [95% CI: 2.6%–4.4%; Tau² = 1.6; I² = 66.8%; 95% PI: 0.3%–30.6%] among primary cases (odds ratio of relapse: 0.77 [95% CI: 0.43–1.36; P-value = 0.087] for the previously treated group compared to the primary infection).

Definition of relapse

The estimated proportion of relapse at 6-months was 3.1% [95% CI: 2.4%–3.9%; Tau² = 2.0; I² = 63.9%; 95% PI: 0.2%–34.2%; n = 105 studies] when the analysis was restricted to studies that required parasitological demonstration for confirming relapse, 2.3% [95% CI: 0.9%–6%; Tau² = 0; I² = 0%; 95% PI: 0.8%–6.9%; n = 5 studies] in studies that defined relapse solely based on clinical suspicion, 6.7% [95% CI: 2.2%–18.7%; n = 1 study] in a study that used molecular method and 4.9% [95% CI: 3.3%–7.2%; Tau² = 0.9; I² = 59%; 95% PI: 0.7%–27%; n = 19 studies] in studies with unclear definition of relapse (further breakdown is presented in Table 1).

Trend over time

Overall, there was a significant subgroup difference in relapse proportions across different decades (P-value < 0.001). The overall relapse proportion at 6-months was 9.1% [95% CI: 5.1%–15.8%; Tau² = 0.7; I² = 0%; 95% PI: 1.5%–39.6%] during 1980–1989, 3.7% [95% CI: 2.6%–5.3%; Tau² = 2.0; I² = 37.1%; 95% PI: 0.2%–39.9%] during 1990–99, 3.1% [95% CI: 2.3%–4.1%; Tau² = 1.1; I² = 60%; 95% PI: 0.4%–21.5%] during 2000–2009 and 2.2% [95% CI: 1.4%–3.4%; Tau² = 2.1; I² = 70%; 95% PI: 0.1%–30.0%] during 2010–2021 (Table 1). Further trend over time for EA and ISC are presented in Tables 2 and 3.

Sensitivity analyses and risk of bias assessment

The results for assessment of potential publication bias are presented in Supplemental Table S2. After adjusting for potential study small-effects, the bias-adjusted estimate for relapse proportion at 6-months was 12.2% [95% CI: 10%–14.9%; Tau² = 1.8; I² = 62%]. In general, the estimates obtained from bias-adjusted analysis were higher (See Table 1, Table 2, Table 3 and Supplemental Table S2). Risk of bias in studies included in the meta-analysis is presented in Supplemental Table S3 and Supplemental File S2.

Of the 58 randomised studies, a high risk of bias was observed in 40 (68.9%) studies for the blinding of participants and personnel domain and in 2 (3.4%) studies for outcome definitions (defined relapse solely based on clinical suspicion). Of the 73 non-randomised studies, 2 (2.7%) studies were considered to be at high risk of bias for the outcome definition (defined relapse solely based on clinical suspicion), 7 (9.6%) studies were at high risk of bias in participant selection, and 2 (2.7%) studies that had an imbalance in distributions of baseline characteristics were considered to be at high risk of bias (See Supplemental Table S3). Overall, the estimates of relapse were 3.1% [95% CI: 2.4%–3.9%; Tau² = 1.7; I² = 55%; 95% PI: 0.2%–30.3%] for randomised studies, and 3.4% [95% CI: 2.5%–4.7%; Tau² = 1.8; I² = 67%; 95% PI: 0.2%–34.1%] for non-randomised studies (See Supplemental Table S1).

Discussion

We estimated the proportion of relapses among initially cured patients post VL treatment using data from 131 published clinical trials. Majority of the studies included were from ISC where the point estimate for overall proportion of relapse remained below 5% for the currently recommended therapies; substantial differences in the estimates between drug regimens was observed. In the ISC, PA had been the mainstay treatment since the 1940s. In the time-period covered in this review, the relapse proportion was initially low (∼5%) for the PA regimen during the 1990s (Table 2). In the subsequent years, response to treatment decreased, with the emergence of PA resistance in the region and the relapse proportion reached 8.4% during the 2000s when it was replaced by alternative regimens such as miltefosine and L-AmB. The overall decreased rates of relapse in ISC over time likely demonstrates the effect of prompt changes in treatment recommendation and the importance of regularly conducting drug efficacy surveillance. Amphotericin B were tested in different formulation and dosages throughout the 2000s in the ISC prior to the adoption of single dose L-AmB as the recommended first line therapy. Currently, 10 mg/kg dosage of the single dose L-AmB is the first line regimen and had the estimated mean relapse of 3.5% (Table 2). In ISC, the overall relapse rate at 6-months for single dose L-AmB was 8.4% for 5 mg/kg dosage and 1.8% for 15 mg/kg dosage, indicating that higher doses of L-AmB may be more effective in preventing relapse. The lower relapse rates observed at higher doses of L-AmB may be due to the increased drug exposure and higher tissue concentrations achieved with higher dosages.¹⁹ It was also found that with L-AmB, for every 1 mg/kg increase in dosage, the odds of relapse decreases by 19% [95% CI: 9%–28%]. However, it is worth noting that there was significant heterogeneity (wide 95% prediction interval) observed suggesting that there may be other factors influencing the efficacy of the L-AmB regimen like patient characteristics, parasite susceptibility, and drug pharmacokinetics, beyond the dosage alone.

Contrary to ISC, there was limited data included in our analysis on L-AmB in East Africa, with a major study²⁰ testing different doses of L-AmB excluded from this meta-analysis due to unclear data on relapses at 6-months. Data was included only from 2 arms that tested single dose L-AmB in a combination regimen at a dose of 10 mg/kg with relapse proportion estimated to be less than 10%. Currently, the combination of PA and paromomycin for 17 days is the recommended first line therapy in the region. Data on the large comparative trial with this regimen and paromomycin monotherapy weren't eligible for inclusion in this meta-analysis as we excluded studies that enrolled HIV co-infected patients and/or in which number of relapses at the end of follow-up was not clearly defined.^21,22 Nevertheless, a recent publication (not included in this review) has indicated efficacy greater than 90% for the PA and paromomycin regimen.²³

This review also found lower relapse proportion following a combination regimen compared to a monotherapy (Table 2). Combination therapy in most regimens reduces the treatment duration (apart from single dose L-AmB) leading to an overall lower burden to the health systems and thus has been increasingly advocated for the treatment of VL.²⁴ This can also potentially delay the emergence of resistance either through synergistic drug action and/or due to the differences in mode of action between the two drugs.²⁵ In East Africa (EA), data regarding the use of combination therapies were limited (i.e., only 2 studies with combination regimen), hence no further exploration was possible regarding the usage of combination therapies in EA. There is also further need to review the safety of the combination therapies; this was beyond the scope of the current work.

Another important question of public health importance is if the currently adopted post-treatment follow-up duration of 6-months is long enough to capture all or a high proportion of relapses.^6,26 In this review, data were available from 21 studies that clearly reported information on occurrence (or non-occurrence) of relapses at 6- and 12-months. From these 21 studies, about two-thirds of all observed relapses (139/204) were detected by 6-months; this would lead to an absolute underestimation in proportion of relapse by 0.6 percentage points or in relative terms 27.6% of all observed relapses would be missed with a 6-months follow-up period. Furthermore, a recent retrospective analysis of data from routine care by Médecins Sans Frontières in South Sudan characterised that over 15% of all relapses occurred after one year.²⁷ This is important for disease control and elimination since identification and prompt treatment of cases is critical for reducing the infective pool of parasites that can sustain the disease transmission. Therefore, the findings of this meta-analysis taken together with existing literature4, 5, 6^,26 suggests that a longer follow-up duration is warranted in VL studies. It is also important to the weigh-up the benefit of extended follow-up against additional challenges of increased cost and increased risk of bias (due to lost to follow-up).²⁶ Therefore, there is a need to invest in parasite and/or host biomarkers predictive of relapse for future clinical trials. It also important to note that two-thirds of the studies used were from ISC and the data may not be extrapolated between regions and requires taking sources of heterogeneity into consideration. Individual participant data meta-analysis (IPD MA) allows adjustment for patient level covariates such as HIV status and further understand the distribution of failure times and would permit a nuanced characterisation of time of occurrence of each relapse, adjusted for individual covariates.²⁶

This meta-analysis also indicated heterogeneity in the estimates of relapse for the same drug regimen across the continents (Table 1, Table 2, Table 3). For example, the relapse proportion following single dose L-AmB combination regimen was 1.5% in the ISC whereas this was 8.6% in EA. Yet, it is important to stress that the number of studies and the median samples size (2 study arms, 100 patients) included in our analysis on single dose L-AmB in EA were small and may require further research. Similarly, for paromomycin, overall proportion of relapse was 6.2% in ISC (from 11 study arms) and 11.3% in EA (3 study arms). Although it was not possible to identify the reasons for these observed differences, such unexplained geographical variation in parasitological response has been widely observed and remains a key research question for VL.³

This review has several limitations. Information regarding the exact time of occurrence of relapse was found to be poorly described in the literature. The definition of relapse adopted also varied across the studies, thus contributing towards methodological heterogeneity. The usage of insufficiently sensitive methods for assessing final cure can lead to underestimation of relapse at the end of follow-up. In particular, the differences in the estimates of relapse proportion in studies that adopted parasitological measures from the studies that only adopted clinical method suggests the need for standardised definitions to be adopted across the studies (Table 1). The currently available rK39 RDTs cannot differentiate between the past VL infection and relapse because antibodies remain present over long period of time. Lack of confirmatory parasitological diagnosis may be due to limitations in performing invasive tissue aspiration in VL patients. New non-invasive tools are needed (either based on antigen detection or molecular biology) to allow for confirmatory diagnosis of relapse without the need of tissue aspiration.

Finally, determination if a relapse is indeed a “true relapse” i.e., identification if the same parasite that caused the initial infection was responsible for the subsequent relapse is critical for the interpretation of the results from therapeutic studies. Determination of true relapse necessitates a comparison of parasite's DNA collected at baseline and at the time of detection of relapse. The use of molecular genotyping is limited in the context of VL and a study in Nepal indicated that the probability of the new infection during the follow-up is rare.⁵ It is also thought that immunity following Leishmania infection is long-term, and this suggests low possibility of re-infection.²⁸ However, the probability of re-infection in highly endemic areas cannot be ruled out as immunity might be impaired due to existing immunosuppression.^28,29 Addressing this remains beyond the scope of this review. Despite these limitations, this review amalgamated data from 131 studies and presents several important findings which can help in designing future therapeutic trials. Some of these limitations can be addressed through individual patient data meta-analyses. Currently research is ongoing to identify determinants of relapse using this approach, facilitated by the IDDO VL data platform.³⁰ This resource can be further explored to quantify the temporal trend of relapses to gauge in the optimal follow-up duration.

This review estimated the expected proportion of relapse for different treatment regimens across the geographical regions which can serve as a benchmark for future comparisons and future trial design. Conventionally adopted 6-months follow-up period was estimated to miss 27.6% of all observed relapses compared to a 12-month follow-up. A longer duration of follow-up may be warranted to capture late relapses.

Contributors

Rutuja Chhajed (RC), Prabin Dahal (PD), Sauman Singh-Phulgenda (SSP), Matthew Brack (MB), Caitlin Naylor (CN), Shyam Sundar (SS), Fabiana Alves (FA), Kasia Stepniewska (KS), Philippe J Guerin (PJG).

Study conception: PD, SSP, FA, KS, PJG.

Data curation: RC.

Formal Analysis: RC, PD, KS.

Funding acquisition: PJG.

Investigation: RC, PD, SSP, FA, KS, PJG.

Methodology: RC, PD, KS.

Project administration: MB, CN.

Resources: MB, CN, PJG.

Software: RC.

Supervision: PD, KS, PJG.

Validation: PD, SSP.

Visualization: RC, PD.

Writing-original draft: RC, PD.

Writing-review and editing: PJG, KS, FA, SS, CN, MB, SSP, PD, RC.

All authors critically edited, read and approved the initial and final version.

Data sharing statement

All the data used for the purpose of this review are available as supplemental material (Supplementary File S3). The code used for generating all the estimates, figures, tables presented in the manuscript is available from https://github.com/ruutuu/VL-Relapses.

Declaration of interests

The authors declare no conflict of interest.