Impact of concomitant oral glucose-lowering medications on the success of basal insulin titration in insulin-naïve patients with type 2 diabetes: a systematic analysis

Abstract

Basal insulin treatment for type 2 diabetes is usually initiated on a background of oral glucose-lowering medications (OGLM). We wanted to examine the influence of various OGLMs on fasting plasma glucose (FPG) and hemoglobin A_1c (HbA_1c) values achieved after titration. A PubMed literature search retrieved 42 publications (clinical trials introducing basal insulin in 17 433 insulin-naïve patients with type 2 diabetes on a defined background of OGLM) and reporting FPG, HbA_1c, target achievement, hypoglycemic events, and insulin doses. 60 individual study arms were grouped by OGLM (combinations) allowed during the titration process: (a) metformin only; (b) sulfonylureas only; (c) metformin and sulfonylureas; or (d) metformin and dipeptidyl peptidase-4 (DPP-4) inhibitors. For all OGLM categories, weighted means and SD were calculated for baseline and end-of-treatment FPG, HbA_1c, target achievement, incidence of hypoglycemic events, and insulin doses. Primary end point was a difference in FPG after titration between OGLM categories. Statistics: analysis of variance and post hoc comparisons. Sulfonylureas, alone or in combination with metformin, impair the titration of basal insulin (insulin doses 30%–40% lower, more hypoglycemic episodes), thus leading to poorer final glycemic control (p<0.05 for FPG and HbA_1c after titration). Conversely, adding a DPP-4 inhibitor to metformin is superior to metformin alone (p<0.05 for FPG and HbA_1c achieved) in patients with type 2 diabetes initiating basal insulin therapy. In conclusion, OGLM are a major determinant of the success of basal insulin therapy. Sulfonylureas impair, while DPP-4 inhibitors (added to metformin) may facilitate the achievement of ambitious fasting glucose targets. PROSPERO registration number CRD42019134821.

What is already known on this topic

• Occasional studies have pointed to the importance of concomitantly used oral glucose-lowering medications for achieving targets after initiating basal insulin therapy.

Waht this study adds

• Compared to metformin treatment, sulfonylurea treatment worsens the outcomes (fasting plasma glucose, HbA1c, hypoglycaemia), while adding DPP-4 inhibitors improves outcomes.

How this study might affect research, practice or policy

• Our study should prompt more research into the role of oral glucose-lowering medications used in addition to basal insulin. It suggests that sulfonylureas should be avoided in this context

Introduction

The preferred option for initiating insulin therapy in patients with type 2 diabetes, in whom tablets alone no longer achieve individualized glycemic targets is the introduction and titration of basal insulin in addition to pre-existing therapy with oral glucose-lowering medications (OGLM).^1–3 Such combination therapy of basal insulin and OGLM has been described in patients continuing any pre-existing therapy with oral agents.^4–8 Since OGLM differ regarding their mechanism of action and associated risks for adverse events (eg, hypoglycemic episodes in the case of sulfonylureas,⁹ which stimulate insulin secretion relatively independent from current plasma glucose concentrations¹⁰), the choice of oral agents may determine the success of basal insulin therapy in type 2 diabetes. In a landmark study, Yki-Järvinen et al¹¹ compared metformin, the sulfonylurea glyburide, or a combination of both together with titrated intermediate-acting insulin preparation neutral protamine Hagedorn subcutaneously injected at bedtime. In this publication, the combination of NPH (Neutral Protamine Hagedorn) insulin with metformin yielded the most favorable results in terms of lowering fasting plasma glucose (FPG), hemoglobin A_1c (HbA_1c), body weight gain, and the risk for hypoglycemic episodes, while glyburide resulted in less improvement in glycemic control, while at the same time provoking more weight gain and hypoglycemia.¹¹ The combination of metformin and glyburide was worse than metformin alone with respect to these end points. Since 1999, new OGLM have been approved for the treatment of type 2 diabetes and for the combination with basal insulin, for example, inhibitors of dipeptidyl peptidase-4 (DPP-4 I)¹² and sodium-glucose co-transporter-2 inhibitors (SGLT-2 I).¹³ Both new classes do not act by stimulating insulin secretion independent from ambient plasma glucose concentrations and, thus, do not enhance the risk for hypoglycemic episodes,^{14 15} when used without concomitant insulin treatment. In particular, DPP-4 Is have been described to be associated with better glycemic control and fewer hypoglycemic episodes, when employed on top of basal insulin treatment.^16–20

We wanted to systematically study the influence of OGLM used individually or in combination on the success of basal insulin therapy. To this end, we performed a literature search aiming to identify publications describing the use of basal insulin in insulin-naïve patients with type 2 diabetes on a background of well-defined oral glucose-lowering agents used individually or in combinations. Based on frequently used individual glucose-lowering medications and their combinations, study arms were pooled, and FPG and HbA_1c concentrations as well as target achievements for FPG and HbA_1c (≤7.0%; <6.5%), the insulin dose and weight gain after titration, and the proportion of patients with any or severe hypoglycemic episodes were compared. The results indicate significant advantages and disadvantages associated with certain oral glucose-lowering agents and their combinations.

Patients and methods

Search strategy and study selection

For the present analysis, articles reporting clinical trials of initiating basal insulin treatment in insulin-naïve patients with type 2 diabetes on a background of a well-defined therapy with single or combined oral glucose-lowering agents were identified through a systematic PubMed search. The search terms are displayed in online supplemental table S1. We searched for prospective, randomized, clinical trials published between 1999 and October 2020 providing details on the basal insulin titration process like (a) insulin preparations used; (b) background OGLM (by class and by agent); (c) person performing the titration (investigator or patient); (d) initial insulin doses, (e) titration intervals (categorized as daily, every 3 days or twice a week, weekly, or in association with study visits (including telephone contacts) only) or titration opportunities (multiplying the occasions per week with the total study duration); (f) ‘stringency’ of the titration algorithm (steepness of the relationship between fasting hyperglycemia and the recommended increment in insulin dose), and (g) FPG titration targets. Additional inclusion criteria were (h) study duration ≥24 weeks, (i) a minimum number of 50 patients per study arm, and (k) report of essential information regarding baseline characteristics (age, sex, duration of diabetes, body weight and body mass index, FPG and HbA_1c) and relevant outcomes at the completion of the study (FPG, HbA_1c, HbA_1c target achievement <7.0% (53.0 mmol/mol) and ≤6.5% (47.5 mmol/mol)), insulin dose after titration (per day and/or per kg body weight and day), body weight changes versus baseline and the proportion of patients reporting any symptomatic or severe hypoglycemia. Open-label and blinded randomized, prospective studies were eligible. Exclusion criteria were publications reporting cross-over studies, concerning other types of diabetes, reporting results relevant for specific ethnic groups other than Caucasian or internationally mixed, representative populations (ie, patient populations with a different pathophysiology regarding insulin sensitivity and insulin secretory defects like Asians²¹), studies allowing concomitant use of glucagon-like peptide 1 (GLP-1) receptor agonists (RAs) (exception: overall <5.0% of the study population as a consequence of protocol violations), studies with >10 % patients with pre-existing basal insulin therapy, studies reporting >5% of patients treated with rapid-acting insulin preparations as part of the rescue strategy. Of 1060 records identified initially, 42 publications representing 61 study arms could be used. Exclusion criteria are described in online supplemental figure S1 according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement.²² We registered our protocol with PROSPERO (https://www.crd.york.ac.uk/prospero/; identification no. CRD42019134821).

Design of the analysis

Individual study arms were analyzed, if they fulfilled the inclusion and exclusion criteria. For our systematic analysis, these study arms were grouped by frequently employed oral glucose-lowering agents and their combinations: (a) metformin only, (b) sulfonylureas only, (c) metformin+sulfonylureas (±thiazolidinediones in a minority of patients), and (d) metformin±DPP-4 Is. Since with basal insulin therapy, the achievement of FPG concentrations in the normal fasting range is a prerequisite for reaching HbA_1c concentrations in the target range, our analysis focusses on FPG (primary end point) and HbA_1c (secondary end point) after titration. Only a single study contributed a single arm describing the effects of SGLT-2 Is in combination with basal insulin degludec.²³ This study was not included in the systematic analysis but will be reported and discussed in relation to the therapies that were part of the systematic analysis. For comparison, we also pooled studies reporting the initiation of basal insulin therapy in insulin-naïve patients with type 2 diabetes on any pre-existing oral therapy with glucose-lowering agents.^4–8

Quality assessment

Study quality was assessed applying the Jadad score²⁴ and the Risk of Bias tool (https://www.riskofbias.info/).²⁵ All publications turned out to be suitable for our analysis.

Data extraction

Relevant data were extracted into prestructured paper forms listing variables of interest. Data were extracted by JW and DW. In case of questions or discrepancies, MAN was consulted. In case of differences that could not be resolved, MAN had the final decision.

Data synthesis and systematic analysis

Differences according to OGLM categories in FPG achieved after basal insulin titration was defined as the primary end point. Secondary end points were HbA_1c after basal insulin titration and HbA_1c target achievements (<7.0% (53.0 mmol/mol) and ≤6.5 (47.5 mmol/mol)), and the proportion of patients reporting any symptomatic or severe hypoglycemia. Exploratory end points were the insulin dose after titration (per day and/or per kg body weight and day), and body weight gain (study end vs baseline). All end points were compared between study arms belonging to the oral glucose-lowering medication categories defined above. Within-group weighted means and pooled SD were calculated using established equations assuming a normal distribution of data. Heterogeneity was reported as Q-value, the associated p value, and I².²⁶

Regression analyses

A linear regression analysis was performed relating FPG and HbA_1c achieved after basal insulin titration. The regression equation, r² and the respective p values are reported for this association.

Estimation of fasting plasma glucose target achievement

We analyzed the proportion of patients reaching their individual FPG titration targets from mean values±SD, assuming a normal distribution, using the function ‘normal distribution’ implemented in Microsoft Excel (V.16.49).

Statistical analysis

Baseline patient characteristics and results at study end are reported as means±SD or proportions (percentages); 95% CIs were derived from SD and the number of patients in the respective category. Weighted mean values and pooled SD for all studies belonging to one subgroup or all studies pooled were calculated using standard equations. For continuous variables, p values for significant differences were calculated by analysis of variance assuming that SD were different (Brown-Forsythe and Welch’s method) for comparing three or more groups (oral glucose-lowering agent categories) with post hoc comparisons between individual groups by Games-Howell tests.²⁷ For continuous variables, a χ² test for larger than 2×2 contingency tables and Fisher’s exact test for 2×2 contingency tables (eg, post hoc tests to identify significant differences between specific titration targets) were used. No adjustment was made for multiple comparisons. Exact p values are presented. P values <0.05 were taken to indicate significant differences.

Results

Selection of publications

The search terms for the retrieval of publications and the selection of study arms for the present analysis is illustrated in online supplemental table S1 and online supplemental figure S1. Overall, 60 arms from 41 publications could be used for the present analysis, representing 17 433 patients divided among 4 categories of different oral glucose-lowering regimens used concomitantly with basal insulin.

Quality assessment

The quality of the studies assessed by the Jadad score²⁴ (online supplemental table S2) and the Cochrane Collection Risk of Bias tool²⁵ (online supplemental figure S2) was found to be sufficient for the inclusion of all retrieved publications and relevant study arms.

Baseline characteristics

Baseline patient characteristics of all studies analyzed, summarized by oral glucose-lowering medication used concomitantly with basal insulin, are shown in table 1, more details in online supplemental tables S3 and S4. Patient age was somewhat higher in the sulfonylurea-only group (table 1). The proportion of females was around 45%, with subtle, but significant differences between groups defined by their oral glucose-lowering medication. Patients receiving sulfonylureas only had significantly lower body mass index and body weight. The known duration of diabetes mellitus was around 9 years, with subtle, but significant differences between groups. Baseline FPG and HbA_1c were higher in the group receiving sulfonylureas only (table 1). Study duration was significantly shorter in those on metformin or sulfonylureas only, and higher with metformin+DPP-4 Is than with metformin+sulfonylurea (±thiazolidinedione) combination therapy (table 1).

Table 1.

Baseline characteristics of insulin-naïve patients with type 2 diabetes initiating basal insulin therapy sort by oral glucose-lowering drug subgroups

Outcome parameter	Unit	OGLM 1: metformin only	OGLM 2: sulfonylureas only	OGLM 3: metformin+ sulfonylureas± thiazolidinediones	OGLM 4: metformin±DPP-4 inhibitors	Overall significance (p value)
Age	Years	57±10	61±9*	57±9†	58±10*, †, ‡	<0.0001
Sex	Female/Male (% female)	1866/2036 (47.8)	606/713 (45.9)	4067/4843 (45.6)	1411/1891 (42.7)	<0.0001
BMI	kg/m²	31.7±5.2	28.6±4.0*	31.2±4.9*, †	32.0±5.6†, ‡	<0.0001
Weight	kg	89.0±18.5	82.1±14.7*	88.2±18.2†	91.8±19.7*, †, ‡‡	<0.0001
Duration of diabetes	Years	8.7±5.9	9.3±8.9	8.8±8.4	9.4±6.0*, ‡	<0.0001
HbA1c	%	8.3±0.8	9.0±1.1*	8.6±0.9*, †,	8.4±0.9*, †, ‡	<0.0001
HbA1c	mmol/mol	67.3±9.1	74.5±11.5*	70.3±10.0*, †	68.2±9.9*, †	<0.0001
Fasting plasma glucose	mmol/L	9.8±2.5	11.6±3.2*	9.7±2.9†	9.9±2.7†, ‡	<0.0001
Study duration	Weeks	26.8±3.0	24.0±0.1*	34.1±11.3*, †	31.2±10.5*, †, ‡	<0.0001

Weighted group means±common SD (continuous variables) or number fulfilling/not fulfilling the criterion and the proportion (percentage) fulfilling the criterion in question (categorical variables). Statistical significance was assessed using one-way analysis of variance (Welch’s test) for continuous variables and χ² test for larger than 2×2 contingency tables and Fisher’s exact test for 2×2 contingency tables (eg, post hoc tests to identify significant differences between specific oral glucose-lowering drugs subgroups; exact p values are presented).

*Significantly different (p<0.05) vs OAD 1: metformin.

†Significantly different (p<0.05) vs OAD 2: sulfonylurea.

‡Significantly different (p<0.05) vs OAD 3: metformin, sulfonylureas and thiazolidinediones.

BMI, body mass index; DPP-4, dipeptidyl peptidase-4; HbA_1c, hemoglobin A_1c; OGLM, oral glucose-lowering medications.

In patients treated with metformin, 3.5% also received thiazolidinediones, and none received sulfonylureas, meglitinides, DPP-4 Is, SGLT-2 Is, or α-glucosidase inhibitors. In patients treated with sulfonylureas or meglitinides, all patients received sulfonylureas, and no other medication was used. Patients treated with metformin (94.9 %)+sulfonylureas (83.2 %) also received thiazolidinediones in 9.3% and acarbose in 0.2%. In those receiving metformin (97.7 %)±DPP-4 Is (30.5 %), no other oral glucose-lowering medication was used (online supplemental table S5).

Study characteristics including differences in the basal insulin titration strategy

Further study protocol details are shown in table 2. There were no differences in the proportions treated with different basal insulin preparations, in the person responsible for conducting titration, and aspects of titration stringency. However, there were significant differences in the titration interval (only at official study visits with sulfonylureas only) and in the starting insulin dose (higher with sulfonylurea only treatment; table 2).

Table 2.

Study characteristics by oral glucose-lowering agents/combinations for publications/study arms included in the present systematic analysis of insulin-naïve patients with type 2 diabetes initiating basal insulin therapy in combination with oral agents

Oral glucose-lowering drugs	OGLM 1: metformin only	OGLM 2: sulfonylurea only	OGLM 3: metformin+ sulfonylurea± thiazolidinediones	OGLM 4: metformin±DPP- 4 inhibitors	Overall significance (p value)
Basal insulin					0.42
Neutral protamine Hagedorn	0 (0.0)	1 (20.0)	2 (6.9)	0 (0.0)
Glargine U100	10 (62.5)	4 (80.0)	19 (65.5)	6 (60.0)
Glargine U300	0 (0.0)	0 (0.0)	0 (0.0)	1 (10.0)
Detemir	3 (18.8)	0 (0.0)	3 (10.3)	1 (10.0)
Degludec	3 (18.8)	0 (0.0)	1 (3.4)	2 (20.0)
Peglispro	0 (0.0)	0 (0.0)	1 (3.4)	0 (0.0)
Insulin lispro protamine	0 (0.0)	0 (0.0)	3 (10.3)	0 (0.0)
Titration interval					< 0.0001
Every 3 days or twice a week	2 (12.5)	0 (0.0)	1 (3.4)	0 (0.0)
Weekly	10 (62.5)	0 (0.0)	18 (62.1)	7 (70.0)
Only during official study visits	0 (0.0)	5 (100.0)	0 (0.0)	2 (20.0)
Other	3 (18.8)	0 (0.0)	10 (34.5)	1 (10.0)
Not reported	1 (6.3)	0 (0.0)	0 (0.0)	0 (0.0)
Person performing titration					0.27
Investigator	9 (56.3)	5 (100.0)	18 (62.1)	5 (50.0)
Participant	7 (43.8)	0 (0.0)	11 (37.9)	5 (50.0)
Formal patient education associated with recruitment into the study	9 (56.3)	5 (100.0)	9 (31.0)	3 (30.0)	0.021
Starting dose of basal insulin					0.0007
<10 IU/day	0 (0.0)	0 (0.0)	9 (31.0)	1 (10.0)
10 IU/day	12 (75.0)	0 (0.0)	13 (44.8)	7 (70.0)
>10 IU/day	1 (6.3)	3 (60.0)	6 (20.7)	2 (20.0)
Not reported	3 (18.8)	2 (40.0)	1 (3.4)	0 (0.0)
Titration category					0.082
‘At investigator’s discretion’	2 (12.5)	0 (0.0)	0 (0.0)	2 (20.0)
One-step algorithm	3 (18.8)	0 (0.0)	4 (13.8)	0 (0.0)
Stepped algorithm, weak (max. 2–4 IU)	6 (37.5)	0 (0.0)	8 (27.6)	1 (10.0)
Stepped algorithm, strong (max. 5–8 IU)	5 (31.3)	5 (100.0)	17 (58.6)	7 (70.0)
Occasions for titration					0.027
10–30	8 (50.0)	3 (60.0)	12 (41.4)	7 (70.0)
31–72	7 (43.8)	0 (0.0)	16 (55.2)	1 (10.0)
Not reported	1 (6.3)	2 (40.0)	1 (3.4)	2 (20.0)

Displayed are the numbers of study arms and the proportion (percentage) fulfilling the criterion in question.

Statistical significance was assessed using χ² test for larger than 2×2 contingency tables and Fisher’s exact test for 2×2 contingency tables (eg, post hoc tests to identify significant differences between specific oral glucose-lowering drugs subgroups; exact p values are presented.

Primary end point

FPG at study end, that is, as the result of the basal insulin titration process, was lowest with metformin+DPP-4 Is, followed by metformin only, the combination of metformin+sulfonylureas (±thiazolidinediones), and remained highest with sulfonylurea treatment only (table 3). All differences between any groups defined by OGLM used concomitantly with basal insulin were significant (p<0.01). The largest difference at study end was observed between patients receiving sulfonylureas and those treated by metformin±DPP-4 Is (Δ 0.8 mmol/L), the second largest difference was between those treated with sulfonylureas versus those on metformin only (Δ 0.7 mmol/L).

Table 3.

Results achieved in clinical trials of basal insulin titration in insulin-naïve patients with type 2 diabetes using various basal insulin preparations in addition to well-defined single or combined oral glucose-lowering agents, trying to aim at different glucose-lowering agent treatments

Oral glucose-lowering agents	Outcome parameter	Fasting plasma glucose	HbA1c	HbA1c <7% (53.0 mmol/mol)	HbA1c ≤6.5% (47.5 mmol/mol)	Hypoglycemia	Severe hypoglycemia	Daily insulin dose	Daily insulin dose	Body weight change from baseline
	Unit	(mmol/L)	(%) (mmol/mol)	Yes/No (% yes)	Yes/No (% yes)	Yes/No (% yes)	Yes/No (% yes)	(IU/day)	(IU/kg/day)	kg
a. Results in groups defined by oral glucose-lowering agents/combinations employed
OGLM 1	Metformin only	6.4 (6.3 to 6.5)	7.1 (7.1 to 7.2) 54.1 (54.1 to 55.2)	1914/1988 (49.1)	1013/2889 (26.0)	1567/2113 (42.6)	32/3870 (0.8)	60 (59 to 60)	0.65 (0.64 to 0.66)	0.8 (0.7 to 1.0)
OGLM 2	Sulfonylurea only	7.1 (6.9 to 7.2)*	7.7 (7.6 to 7.7)* 60.7 (59.6 to 60.7)	438/881 (33.2)*	237/3970 (5.6)*	740/579 (56.1)*	21/1298 (1.6)*	36 (35 to 37)*	0.43 (0.42 to 0.45)*	2.7 (2.5 to 3.0)*
OGLM 3	Metformin+ sulfonylureas± thiazolidinediones	6.7 (6.6 to 6.7)*, †, ‡	7.2 (7.2 to 7.2)*,† 55.2 (55.2 to 55.2)	4056/4854 (45.5)*, †	2196/6714 (24.6)*, †	4525/3362 (57.4)*	96/8628 (1.1)	39 (39 to 40)*, †	0.43 (0.43 to 0.44)*	2.1 (1.9 to 2.2)*,†
OGLM 4	Metformin±DPP-4 inhibitors	6.2 (6.2 to 6.3)*, †, ‡	7.0 (7.0 to 7.0)*, †, ‡ 53.0 (53.0 to 53.0)	1668/1634 (50.5)†, ‡	993/2309 (30.1)*, †, ‡	1338/1963 (40.5)†,‡	27/3168 (0.8)†	56 (55 to 57), *†, ‡	0.59 (0.58 to 0.60)*, †,‡	1.4 (1.2 to 1.5)*, †, ‡
	Overall significance (p value)	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	0.0632	<0.0001	<0.0001	<0.0001
b: Differences between groups defined by oral glucose-lowering agents/combinations employed
Comparison	OGLM 1 vs 2	0.65 (0.52 to 0.78)§	0.53 (0.46 to 0.60)§ 5.8 (5.0 to 6.6)	-15.8 (-18.8 to -12.8)§	-8.0 (-10.5 to -5.4)§	-13.5 (-16.7 to -10.4)§	-0.8 (-1.4 to 0.1)§	-§23 (-25 to -22)§	-0.22 (-0.24 to -0.20)§	1.9 (1.63 to 2.19)§
	OGLM 1 vs 3	0.29 (0.21 to 0.36)§	0.05 (0.01 to 0.08)§ 0.5 (0.1 to 0.9)	-4.0 (-5.9 to -2.1)§	-10.6 (-12.2 to -9.0)§	-14.8 (-16.8 to -12.9)§	-0.3 (-0.7 to 0.1)	-20 (-21 to -19)§	-0.22 (-0.23 to -0.21)§	1.20 (1.03 to 1.37)§
	OGLM 1 vs 4	-0.16 (-0.25 to -0.06)§	-0.13 (-0.17 to -0.08)§ 1.4 (-1.9 to -0.9)	-1.5 (-3.8 to 0.9)	-4.1 (-6.2 to -2.0)§	-2.0 (-4.4 to 0.3)	0.0 (-0.5 to 0.4)	-4 (-5 to -2)§	-0.06 (-0.08 to -0.05)§	0.52 (0.33 to 0.71)§
	OGLM 2 vs 3	-0.37 (-0.49 to -0.24)§	0.48 (-0.55 to -0.41)§ 5.2 (-6.0 to -4.5)	-11.9 (-14.7 to -9.1)§	-2.6 (-5.0 to -0.4)§	-1.3 (-4.2 to 1.7)	-0.5 (-1.4 to 0.1)	3 (2 to 4)§	0.00 (-0.01 to 0.00)	-0.71 (-0.98 to -0.44)§
	OGLM 2 vs 4	-0.81 (-0.95 to -0.67) §	0.66 (-0.73 to -0.59)§ 7.2 (-8.0 to -6.4)	-17.3 (-20.5 to -14.3)§	-12.1 (-14.8 to -9.5)§	-15.6 (-18.7 to -12.4)§	-0.7 (-1.7 to 0.0)§	20 (18 to 21)§	0.16 (0.14 to 0.18)§	-1.39 (-1.68 to -1.10)§
	OGLM 3 vs 4	-0.44 (-0.54 to -0.35)§	0.18 (-0.22 to -0.14)§ 2.0 (-2.4 to -1.5)	-5.4 (-7.4 to -3.4)§	-14.7 (-16.5 to -12.9)§	-16.8 (-18.8 to -14.8)§	-0.3 (-0.6 to 0.2)	17 (15 to 18) §	0.16 (0.14 to 0.17)§	-§0.68 (-0.87 to -0.49)§

Continuous variables are presented as means and their 95% CIs, categorical variables are presented as number fulfilling/not fulfilling the criterion and the proportion (percentage) fulfilling the criterion in question. Statistical significance was assessed using one-way analysis of variance (Welch’s test) for continuous variables and χ² test for larger than 2×2 contingency tables and Fisher’s exact test for 2×2 contingency tables (eg, post hoc tests to identify significant differences between specific titration targets), including the ‘attributable difference’ expressed as a percentage and its 95% CI. For overall comparisons, exact p values are presented.

*Significantly different (p<0.05) vs OGLM 1: metformin.

†Significantly different (p<0.05) vs OGLM 2: sulfonylurea.

‡Significantly different (p<0.05) vs OGLM 3: metformin, sulfonylureas and thiazolidinediones.

§: Significantly different (p< .05)

HbA_1c, hemoglobin A_1c.

Secondary end points

The distribution of HbA_1c concentrations at study end followed the pattern seen with FPG (table 3, figure 1). All differences were significant. The largest difference was observed between patients receiving sulfonylureas and those treated by metformin±DPP-4 Is (Δ 0.7%/7 mmol/mol). There was a highly significant correlation of FPG and HbA_1c (online supplemental figure S3), and values at baseline fell onto the same regression line as values at study end. Heterogeneity measures are presented in online supplemental table S6.

Figure 1.

Forrest plot of fasting plasma glucose (A) and hemoglobin A_1c (HbA_1c) (B) concentrations achieved after basal insulin titration according to categories of oral glucose-lowering agent categories 1 (metformin (Met), green symbols), 2 (sulfonylurea (SU), red symbols), 3 (Met±SU±thiazolidinediones (TZD), brown symbols) and 4 (Met±dipeptidyl peptidase-4 inhibitors ((DPP-4 I), blue symbols). Individual studies are shown with filled circles, pooled results reflecting the four categories of oral glucose-lowering agents are shown as diamonds. SEM±SD. For statistical analysis, see table 3.

HbA_1c target achievement (<7.0% and ≤6.5% (<53.0 and ≤47.5 mmol/mol)) was dramatically lower in those treated with sulfonylureas only (33.2% and 5.6%) compared with all other OGLM (45%–51% and 25%–30%, respectively), and were highest with metformin±DPP-4 Is treatment (table 3). The largest differences amounted to 17.3% and 5.4% in terms of target achievement (<7.0%/<53.0 mmol/mol) between metformin±DPP-4 I and sulfonylurea treatment and metformin and sulfonylureas only, respectively, and to 12.1% and 14.7% in terms of target achievement (≤6.5%/47.5 mmol/mol) between metformin±DPP-4 I and sulfonylurea treatment and metformin and metformin+sulfonylurea treatment, respectively.

The proportion of patients reporting any symptomatic hypoglycemia was substantially and significantly higher in both groups receiving sulfonylurea treatment compared with those in all other categories of OGLM (table 3). The difference amounted to a proportion of patients reporting symptomatic hypoglycemic episodes higher by 13.5%–16.8% with sulfonylureas only or metformin+sulfonylureas versus metformin alone or metformin±DPP-4 Is (table 3). Severe hypoglycemia showed a pattern similar to any symptomatic hypoglycemia, however, at a much lower prevalence (table 3). Severe hypoglycemia clearly was highest with sulfonylurea treatment only.

Exploratory end points

Insulin doses after titration were lower in all categories of OGLM containing sulfonylureas compared with metformin only and metformin±DPP-4 Is (table 3). Overall, there were significant differences in insulin doses achieved after completing titration between all groups defined by their use of OGLM‚ p<0.0001).

Body weight increased substantially and significantly more in patients treated with sulfonylureas only compared with all other categories of OGLM. This was somewhat attenuated by combining sulfonylureas with metformin. Patients treated with metformin±DPP-4 Is increased their body weight slightly more than those treated with metformin alone (table 3, online supplemental table S7).

Fasting plasma glucose titration target achievement

Looking at FPG target achievement (against the individually defined fasting plasma titration targets), it was reached in substantially lower proportions of patients treated with sulfonylureas only compared with all other groups (online supplemental figure S4).

SGLT-2 inhibitors plus basal insulin

A single study reporting the titration of basal insulin (inulin glargine) on a background of SGLT-2 I (plus metformin, ±thiazolidinediones),²³ resulted in a mean FPG concentration of 6.2 mmol/L, an HbA_1c of 6.7% (49.7 mmol/mol), achieving HbA_1c targets of <7.0% and ≤6.5% (<53.0 and ≤47.5 mmol/mol) in 71.3% and 49.5% of all patients on an insulin dose of 53.5 IU/day. Weight gain was 1.9 kg on average and a proportion of patients reporting any symptomatic hypoglycemic episode of 19.5% (severe hypoglycemia not reported).

Discussion

The main finding of the present analysis is that combining basal insulin with metformin or metformin±DPP-4 Is results in a more efficient achievement of both conventional HbA_1c targets (<7.0 and ≤6.5% (<53.0 and ≤47.5 mmol/mol)), as well as lower FPG and HbA_1c concentrations at the end of the titration periods, and less patients reporting any symptomatic or even severe hypoglycemia, especially when compared with patients concomitantly treated with sulfonylureas alone or in combination with metformin (figure 1, table 3). At the same time, insulin doses achieved by titration were substantially lower in patients receiving sulfonylurea treatment, alone or in combination with metformin (table 3). While clearly the results were worst in the case of using sulfonylureas alone, some of the differences to metformin only or metformin±DPP-4 I treatment were attenuated when combining sulfonylureas with metformin (table 3): this was the case regarding FPG and HbA_1c achieved at study end the proportion of patients achieving conventional HbA_1c targets, and body weight gain, but not regarding the number of patients reporting any symptomatic or even severe hypoglycemic episode (table 3). The fact that baseline HbA_1c and FPG concentrations in trials examining sulfonylureas during the titration of basal insulin were higher than with other categories of concomitant oral glucose-lowering agents used in the same setting may suggest that adaptation to higher degrees of hyperglycemia may provoke symptoms of hypoglycemia even in the presence of slightly elevated plasma glucose concentrations, thus hampering the progress with basal insulin titration.

Patients receiving metformin±DPP-4 Is had slightly, but significantly better treatment outcomes compared with those on metformin only with respect to FPG (online supplemental figure S4) and achievement of the lower HbA_1c target (≤6.5%/≤47.5 mmol/mol), but not regarding HbA_1c target <7.0% (<53.0 mmol/mol; table 3). Weight gain was slightly, but significantly worse with metformin±DPP-4 Is as compared with metformin alone (table 3). These differences are somewhat surprising, since only 30.5% of the patients in the metformin±DPP-4 I group received DPP-4 Is. We can only speculate about effects of using DPP-4 Is in addition to metformin in all patients. Clearly, improvements in outcomes of combinations of metformin and DPP-4 Is (sitagliptin or vildagliptin) have been shown in previous studies^16–20 which could, however not be included in the present analysis, because, they did not fulfill all our inclusion criteria. Namely, the oral medication was changed after insulin treatment was initiated.^16–20

Our results concerning the use of sulfonylureas in combination with basal insulin recapitulate those by,^{11 11} and extend their findings to the use of sulfonylureas different from glibenclamide/glyburide and to insulin preparations beyond NPH insulin alone (table 2). It seems that a combination of sulfonylureas and basal insulin provokes episodes of hypoglycemia at much lower insulin doses, which requires stopping the titration process or even reduction in the insulin dose found to elicit hypoglycemic episodes. This is supported by the uniquely low FPG target achievement with sulfonylurea treatment (online supplemental figure S4). It is unlikely that the lower HbA_1c target achievement on a background of sulfonlyureas is related to the shorter duration of the respective studies (table 1). Rather, as also indicated by a higher risk to report any or even severe hypoglycemic episodes (table 3), and the lower insulin doses reached by titration (table 3), that the lower HbA_1c target achievement is the consequence of intercurrent hypoglycemic episodes interfering with further escalations of the basal insulin doses. Overall, our results challenge the apparently widespread use of sulfonylureas (meglitinides) in combination with basal insulin. This conclusion is supported by looking at titration results in studies which have added basal insulin preparations to any mix of oral glucose-lowering agents (containing various proportions of patients treated with sulfonylureas; online supplemental tables S8 and S9): results indicated a higher end-of-titration FPG (6.6 vs 6.4 and 6.2 mmol/L compared with those treated with metformin or metformin±DPP-4 Is), similar HbA_1c concentrations (7.1 vs 7.1 or 7.0% to those treated with metformin or metformin±DPP-4 Is), slightly worse HbA_1c target achievements (<7.0%: 48.6 vs 49.0 or 50.5 in those treated with metformin or metformin±DPP-4 Is), a substantially higher proportion experiencing hypoglycemic episodes (53.2 vs 42.6 and 40.5% in those treated with metformin or metformin±DPP-4 Is), slightly lower insulin doses and minimally higher body weight gain (1.5 vs 0.8 and 1.4 kg in those treated with metformin or metformin±DPP-4 Is; online supplemental table S9). In the two studies reporting HbA_1c target achievement ≤6.5% when basal insulin was titrated on a background of any pre-existing oral glucose-lowering treatment,^{4 5} the value of 19.2% is less than what could be achieved with metformin only or metformin±DPP-4 Is (26.0% and 30.1%, respectively). Thus, initiating basal insulin treatment disregarding details of concomitant oral glucose-lowering treatment (online supplemental tables S9 and S10) will lead to results somewhat worse compared with metformin only and metformin±DPP-4 I treatment (table 3).

Imbalances in baseline characteristics (table 1) and details of the basal insulin titration procedure (table 2) need to be considered as potential confounders with an influence on the outcomes of our study. Obviously, the group with the worst outcome, those treated with sulfonylureas only, had the hardest baseline conditions with higher age and FPG and HbA_1c concentrations to begin with (table 1) and studies on sulfonylureas used the least aggressive titration algorithms (table 2). It had been noted previously that a high baseline HbA_1c is associated with worse chances for achieving ambitious FPG and HbA_1c targets.²⁸ The lower body weight and body mass index may indicate a lower degree of insulin resistance (perhaps reflected in lower insulin doses achieved after titration and higher proportions of patients reporting any symptomatic and severe hypoglycemic episodes (table 3). Furthermore, we believe that differences in baseline conditions should not necessarily have major impacts on the study outcomes, because the study duration was sufficient to finalize the insulin titration process. Those studies that reported time courses for insulin doses uniformly described a plateau reached after approximately 16 weeks (details not shown), indicating that other reasons, not the lack of time, must have determined the insulin dose and the associated quality of glycemic control.

GLP-1 RAs are used in combination with basal insulin, sometimes even as fixed combinations.^{6 29 30} Both GLP-1 RAs and basal insulin have the potential to reduce FPG and HbA_1c substantially, while in combination with oral glucose-lowering agents, the contribution of basal insulin will most likely be greater than that of OGLMs. Therefore, we excluded studies reporting the combination of GLP-1 RAs and basal insulin from our systematic analysis. The only available oral GLP-1 RA, oral semaglutide, when added to pre-existing insulin treatment, led to meaningful reduction in glycemic parameters and body weight without provoking hypoglycemic episodes.³¹ However, this study was not performed strictly using basal insulin, and premixed insulin and basal/bolus regimens. Therefore, it was outside the scope of our methodology, and could not be used to define a fith category of oral glucose-lowering agents used concomitantly with basal insulin titration.

As far as can be concluded based on a single publication, SGLT-2 Is are well suited for the combination with basal insulin, perhaps even when compared with metformin or metformin±DPP-4 Is. This conclusion, however, would require more studies with SGLT-2 Is.

Limitations of the present study are the mentioned imbalances in baseline characteristics (table 1) and in details of the titration protocols (table 2), the influence of which on the main study results remain uncertain (as discussed above), and the many aspects characterizing details of the basal insulin titration process which have not been reported in the publications used for the present systematic analysis, but still might be confounders with a hidden influence on the results. Studies selected for the present systematic analysis of clinical trials of basal insulin titration on various background OGLM often used the ‘last observation carried forward’ approach to deal with patients discontinuing their assigned treatment. This may be a reason for bias, which most likely has affected all categories of oral glucose-lowering agents in a similar way. It would be more straightforward to perform an analysis like our present one using individual patients’ data, but from publications, only summary data are available. Strengths of the present analysis are the systematic nature, the clear definition of inclusion and exclusion criteria, leading to treatment groups with a well-defined treatment with single or combined glucose-lowering agents, the large number of studies (as well as study arms and patient numbers) analyzed. Suitable clinical trials of sulfonylureas used concomitantly with titrating basal insulin have exclusively been reported for the long-acting sulfonylurea glimepiride.^{32 33} This may be viewed as a limitation of our study and certainly reduces the generalizability of conclusions extending to other compounds, especially more short-acting sulfonylureas. The fact that studies reporting basal insulin titration on a background of sulfonylurea agents uniformly were performed in patients with more prominent baseline hyperglycemia (table 1) likely impacted our findings and limits the comparability with studies using other concomitant oral agents, and thus, requires cautious conclusions regarding the less favorable results observed with sulfonylureas.

In conclusion, clinical trials reporting basal insulin titration in hitherto insulin-naïve patients with type 2 diabetes indicate better glycemic results (FPG and HbA_1c at study end) and a lower hypoglycemia risk with metformin or metformin±DPP-4 I treatment used concomitantly with basal insulin as compared with sulfonylureas used alone or in combination with metformin. The overall FPG target achievement was generally low, but lowest when basal insulin was used in combination with sulfonylureas. Our results shed serious doubts on the value of combining basal insulin with oral agents like sulfonylureas and/or meglitinides. In the absence of confirmation from other sources, we take our findings to recommend caution when initiating such combinations, but our findings do not justify the conclusion that such a combination should generally be regarded as contraindicated. Future studies should help decide, whether adding DPP-4 Is or SGLT-2 Is to metformin will further improve treatment outcomes.

Data availability statement

Data are available on reasonable request. All data analyzed for this manuscript are from selected publications.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

Not applicable.