Efficacy and Safety of Basal Analog Regimens in Type 2 Diabetes Mellitus: Systematic Review and Meta-Analysis of Randomized Controlled Trials

A comprehensive literature search of MEDLINE, Embase, and the Cochrane Central Register of Control Trials (CENTRAL) was conducted, covering the period from January 1, 1997 to October 31, 2017. A standardized review protocol was used to define the eligibility criteria applied when searching for and screening references, guided by the population, intervention, comparator, outcome, timing, setting, and study design [PICO(TSS)] framework (Table S1 in the Electronic supplementary material, ESM) [10]. Inclusion criteria included primary RCTs with adult T2DM patients, an intervention group who received a BA (insulin glargine or detemir), a comparator group who received a PM insulin, a minimum follow-up period of 12 weeks, and a minimum of 30 patients per treatment arm. The studies also needed to report data on at least one efficacy outcome (HbA1c, fasting glucose, 2-h postprandial glucose, and total insulin dose) or safety outcome (body weight and hypoglycemia). Keywords used in the search strategy included type 2 diabetes, basal insulin, glargine, detemir, premix, biphasic insulin AND 50/50, 75/25, 70/30. The search strategy employed for each of these databases is described in more detail in Table S2a–c of the ESM. The present article is based on previously conducted studies and does not contain any studies with human participants or animals performed by any of the authors.

Study selection and screening were conducted using the web-based platform Digital Outcome Conversion (DOC) Library (version 2.0; Doctor Evidence, LLC, Santa Monica, CA, USA), according to a screening protocol based on the PICO(TSS) criteria. An assessment of the quality of the included studies was conducted by two reviewers using the Cochrane Collaborations tool for assessing the risk of bias in randomized trials [11]. This instrument is used to evaluate seven domains of bias: random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias), incomplete outcome data (attrition bias), selective reporting (reporting bias), and other sources of bias (other bias). Data were stored and managed in Microsoft Excel. Discrepancies were resolved by an independent reviewer.

Data extraction was conducted using the DOC Data version 2.0 software platform (Doctor Evidence) and its universal electronic extraction form, based on a standardized data configuration protocol. Continuous outcomes included change from baseline (reported or calculated) in HbA1c, fasting glucose, 2-h postprandial glucose, and body weight. Data on total insulin dose at the end of follow-up were also collected. Change from baseline was calculated for selected outcomes when the study did not report this data. This was possible if the outcome was reported at the endpoint and baseline. The change was calculated by subtracting the baseline value from the endpoint value. The standard deviation (SD) in the outcome change, assuming a correlation of 0.8, was calculated based on Eq. (1) in the Cochrane Handbook [10]:

Fasting glucose and postprandial glucose data extracted included both plasma and capillary glucose values. Postprandial glucose measurements obtained 2 h after breakfast (or, for one study, after lunch) were included. In the total insulin dose analysis, a unit conversion of U or IU to U/kg to adjust for patient body weight was performed for analytical purposes. Categorical outcomes included the proportion of patients achieving HbA1c levels < 7% or ≤ 7% and total hypoglycemia by the end of follow-up. The latter was evaluated based on the author’s definition and included symptomatic and asymptomatic events, measured glucose events, and self-diagnosed events. Primary efficacy outcomes were change from baseline in HbA1c and HbA1c levels < 7% or ≤ 7%. Secondary efficacy outcomes were fasting glucose, total insulin dose at the end of the study, and 2-h postprandial glucose.

A pairwise meta-analysis (MA) was performed for the outcomes of interest (i.e., HbA1c, fasting glucose, 2-h postprandial glucose, total insulin dose, body weight, and hypoglycemia). The conventional DerSimonian-Laird random-effects model was utilized. We calculated heterogeneity across studies using the Cochran Q test and the I² statistic. I² > 50% indicated significant heterogeneity [10]. All analyses were done in R 3.5.2 using the “metafor” package [12]. The comparative efficacy for each outcome was represented by the odds ratio (OR) and the associated 95% confidence interval (CI) for categorical data, or the mean difference (MD) and the associated 95% CI for continuous data. The original comparison was stratified by BA and PM insulin frequencies, including BA once daily (QD) with or without (±) OAD vs. PM insulin BID ± OAD (9 studies: [13‐21]), basal-bolus insulin given once a day (BB insulin 1×) vs. PM insulin BID (3 studies: [2, 22, 23]), basal-bolus insulin given thrice a day (BB insulin 3×) vs. PM insulin BID (2 studies: [24, 25]), BB insulin 3× vs. PM insulin TID (2 studies: [26, 27]), and basal-bolus insulin given zero to thrice daily [BB insulin stepwise (0–3×)] vs. PM insulin given zero to twice a day [PM insulin stepwise (0–2×)] (6 studies: [28‐33]). These original treatment comparisons are reported in Table 1. Exploratory analyses were conducted for comparisons based on PM insulin type, PM insulin ratio, BA insulin type, race, study follow-up period, and baseline HbA1c. The studies included in each of these secondary analysis comparisons are summarized in Table S3 of the ESM.

The search of relevant databases identified a total of 251 studies. Following title and abstract screening, a total of 213 studies were excluded for not aligning with the prespecified PICO(TSS) criteria. Thirty-eight references were full-text screened, of which 16 did not fit the prespecified criteria for intervention (BA insulin), outcomes (change in HbA1c, fasting glucose, 2-h postprandial glucose, and body weight, insulin dose, hypoglycemia), outcome stratification (data not reported for intervention or comparator), study design (RCTs, phase 2–4), or number of participants (≥ 30 patients per arm), and were therefore excluded. The remaining 22 studies met the PICO(TSS) criteria and were considered for data configuration and included in the meta-analysis [2, 13‐33]. A PRISMA flowchart of the study selection process is presented in Fig. 1.

A summary of study characteristics is provided in Table 1. A total of 9691 patients were included in the analysis. The most common BA insulin treatments were insulin glargine 100 U/mL (19 studies: [2, 13‐15, 17‐24, 26‐32]) and insulin aspart (11 studies: [15, 16, 18‐20, 22, 23, 25, 31‐33]). The patient characteristics in each trial are also reported in Table 1. Fasting plasma glucose ranged from a mean of 5.8 mmol/L (105.1 mg/dL) in insulin glargine 100U/mL + glulisine 0–2×-treated patients [31] to 14 mmol/L (252 mg/dL) in insulin aspart BID + OAD-treated patients [20]. HbA1c levels ranged from a mean of 8.07% in patients who received insulin glargine 100 U/mL + sitagliptin [15] to 11.4% in patients who received insulin glargine 100 U/mL + metformin + sulfonylurea [13].

Table S2d in the ESM presents a summary of the quality of the 22 studies. There was generally a low risk of selection bias across the studies, with some studies judged as having an unclear risk of such bias. In eight studies [13, 16, 18, 20, 25, 28, 32, 33], inadequate methods of generating the randomization sequence were reported; similarly, seven studies were judged to have an unclear risk of allocation concealment due to insufficient details [2, 13, 15, 16, 28, 32, 33]. Risk of performance bias (pertaining to the blinding of participants and personnel) and risk of detection bias (pertaining to blinding of outcome assessment) were primarily judged to be high, as most of the studies had an open-label study design. All of the studies had a low risk of attrition and reporting bias. Finally, nearly all the studies had a low risk of other sources of bias.

The primary efficacy outcome of mean change from baseline in HbA1c was reported or calculated in all 22 trials and included in the analysis [2, 13‐33]. The MD values for the overall comparisons and for each of the five primary analysis comparisons are reported in Fig. 2a. There was no statistically significant difference in HbA1c change from baseline between patients who used BA regimens and patients who used PM insulin [− 0.04% (95% CI − 0.14, 0.07)], with significant heterogeneity of effect size across the included trails (I² = 77.4%, p value < 0.05) (Fig. 2a). A comparison of BB insulin 3× to PM insulin BID highlighted a statistically significant difference in HbA1c reduction [− 0.37% (95% CI − 0.64, − 0.10)] (Fig. 2a). In contrast, BA therapy ± OAD led to a similar HbA1c change to that achieved with PM insulin in combination or without OAD [0.00% (95% CI − 0.21, 0.20)], suggesting that the change observed in the BB insulin 3× group resulted from the bolus injections (Fig. 2a). Exploratory analyses were conducted for comparisons based on BA or PM insulin types and patient/trial characteristics, as further summarized in Table S3 of the ESM. When examining PM insulin stratified by type, there was a statistically significant difference in HbA1c reduction between patients on BA insulin regimens and those receiving regular human PM insulin [− 0.39% (95% CI − 0.60, − 0.18)] (Table S4 in the ESM). A secondary analysis of PM insulin ratio also revealed a statistically significant difference in the change in HbA1c between those receiving BA insulin regimens and those receiving PM 50/50 insulin [− 0.22% (95% CI − 0.40, − 0.04)] (Table S4 in the ESM). None of the other secondary analysis comparisons showed statistically significant differences in change from baseline HbA1c (Table S4 in the ESM).

HbA1c ≤ 7% was reported and analyzed in all 22 trials [2, 13‐33]. ORs for the overall comparisons and for each of the five primary analysis comparisons are reported in Fig. 2b. Overall, although there were no statistically significant differences between patients who used BA insulin regimens and those who used PM insulin [1.14 (95% CI 0.94, 1.40)], there was significant heterogeneity of effect size across the included trials (I² = 78.7%, p value < 0.05) (Fig. 2b). The odds of achieving HbA1c ≤ 7% were highest with BB insulin 3× compared to PM insulin BID [1.76 (95% CI 1.17, 2.64)] (Fig. 2b).

Change from baseline in fasting glucose was an average of 0.61 mmol/L smaller in patients who used BA than in patients who used PM insulin regimens [− 0.61 mmol/L (95% CI − 0.90, − 0.32)] (Table 2). Three primary analyses showed statistically significant differences, including BB insulin stepwise versus PM insulin stepwise [− 0.99 mmol/L (95% CI − 1.42, − 0.55)], BB insulin 3× versus PM insulin TID [− 0.74 mmol/L (95% CI − 1.46, − 0.02)], and BA insulin QD ± OAD versus PM insulin BID ± OAD [− 0.59 mmol/L, 95% CI − 1.06, − 0.12)] (Table 2).

Although not statistically significant, the evidence suggests that total insulin dose was on average 0.06 U/kg lower in patients who used BA compared to those who used PM insulin [− 0.06 U/kg (95% CI − 0.13, 0.01)] (Table 2). A significant decrease in total insulin dose was reported for the BA insulin QD ± OAD patients compared to the PM insulin BID ± OAD patients [− 0.107 U/kg (95% CI − 0.32, − 0.01)] (Table 2).

An increase from baseline in 2-h postprandial glucose was observed in patients who used BA insulin regimens compared to patients who used PM insulin, but this was not statistically significant [0.09 mmol/L (95% CI − 0.31, 0.49)] (Table 2).

Total hypoglycemia was suitable for analysis in 19 trials [2, 14‐27, 29‐31, 33]. Results for overall comparisons and each of the five primary analysis comparisons are reported in Fig. 3a. Patients who used BA insulin regimens had a statistically significantly reduced likelihood of total hypoglycemia compared to those who used PM insulin regimens [0.77 (95% CI 0.64, 0.92)], with significant heterogeneity of effect size across the included trials (I² = 65.3%, p value < 0.05) (Fig. 3a). The only statistically significant difference in hypoglycemia incidence in a primary comparison was between the BA insulin QD ± OAD and PM insulin BID ± OAD groups [0.53 (95% CI 0.40, 0.70)] (Fig. 3a). Other primary comparisons showed no statistical significant difference in hypoglycemia incidence (Fig. 3a). The results of exploratory analyses of total hypoglycemia for comparisons based on BA or PM insulin types and patient/trial characteristics are summarized in Table S5 in the ESM.

Mean change from baseline in body weight was suitable for analysis in 21 trials [2, 13‐31, 33], with MDs for overall comparisons and each of the five primary analysis comparisons reported in Fig. 3b. There was a statistically significant difference in body weight between patients who used BA insulin regimens and patients who used PM insulin [− 0.48 kg (95% CI − 0.86, − 0.11)], with significant heterogeneity of effect size across the included trials (I² = 75.7%, p value < 0.05) (Fig. 3b). The comparison of the BA QD ± OAD and PM insulin BID ± OAD groups showed a significant difference in body weight change [− 1.23 kg (95% CI − 1.99, − 0.47)] (Fig. 3b). Body weight change was also significantly different for BB insulin 1×-treated patients and PM insulin BID-treated patients [− 0.64 kg (95% CI − 1.06, − 0.22)] (Fig. 3b).