Introduction
The three main components of dysglycaemia in patients with diabetes are chronic hyperglycaemia, hypoglycaemia and glycemic variability (GV) [
1]. GV is a measure of changes in blood glucose (BG) levels, both throughout the day and over time. Long-term GV is most frequently evaluated by fluctuations over weeks or months in glycated haemoglobin (HbA1c), fasting plasma glucose (FPG) and/or postprandial glucose (PPG), while short-term GV involves the measurement of acute (within-day) glucose fluctuations [
1,
2]. The most obvious example of these is the change in BG levels after a meal (PPG). PPG levels are an important factor in terms of overall metabolic control in diabetes, with some studies finding a closer association between PPG levels and HbA1c than between FPG levels and HbA1c [
3,
4]. Reducing postprandial excursions is a valuable strategy for reducing GV in patients with diabetes [
5].
HbA1c is a well-established risk factor for adverse diabetes-related outcomes [
6], and both fasting and postprandial hyperglycaemia are also major risk factors for diabetic complications [
5]; however, data suggest that PPG excursions may represent a particularly relevant therapeutic target in patients with diabetes. Compared with long-term, sustained hyperglycaemia, BG fluctuations postprandially or during glucose ‘swings’ have a more specific triggering effect on oxidative stress, a factor that plays a pivotal role in the pathogenesis of various diabetic complications [
7]. There is also evidence that postprandial hyperglycaemia is a greater predictor of cardiovascular disease than elevated FPG levels [
8].
Rapid-acting insulin analogues (RAIAs; aspart, glulisine and lispro) and regular human insulin (RHI) have substantially different pharmacokinetic and pharmacodynamic profiles. Compared with RHI, RAIAs more closely mimic normal mealtime insulin excursions and their pharmacodynamic profiles are more similar to that of endogenous insulin (reviewed by Home [
9]). The faster onset and offset of action [
10], a lower incidence of hypoglycaemia [
10], and less intrapatient variability on absorption [
11] seen with RAIAs mean that these agents are preferred in patients with symptomatic hyperglycaemia, or those with glycaemia uncontrolled by basal insulin alone [
10]. However, despite the advantages of RAIAs over RHI, both RAIAs and RHI are still widely used in the management of PPG excursions and GV in patients with diabetes [
10,
12]. The principal objective of the current meta-analysis is to update and consolidate the literature on this topic by using data from randomised controlled trials (RCTs) to address the impact of RAIAs and RHI on glycemic control, including PPG and long- and short-term GV [as measured by end-of-treatment (EOT) HbA1c and pre- to postprandial change in BG, respectively], in patients with type 1 diabetes (T1D) or type 2 diabetes (T2D). Our hypothesis is that RAIAs are more effective than RHI in reducing PPG excursions and improving glycemic control in patients with T1D or T2D.
Methods
Data Sources and Searches
We searched PubMed for studies published between 1999 and 29 June 2016, using the following search string: ‘insulin, short-acting’ OR ‘insulin lispro’ OR ‘insulin aspart’ OR ‘insulin glulisine’ OR ‘Novorapid’ OR ‘Apidra’ OR ‘Humalog’ OR ‘protamine suspension’ OR ‘insulin aspart protamine drug combination 30:70’ OR ‘Novolog’ NOT (‘pregnancy’ OR ‘pregnant women’ OR ‘hospitalization’ OR ‘institutionalization’ OR ‘paediatric’ OR ‘child’ OR ‘childhood’ OR ‘infant’ OR ‘newborn’). The search was then filtered to identify RCTs only.
The bibliographies of included studies and relevant reviews were searched to identify additional studies for inclusion.
Study Selection
We included all RCTs of patients with diabetes published from 1999 to 29 June 2016 that assessed the effects of RAIAs or RHI on glycemic control, with a focus on pre- and postprandial BG and HbA1c. All identified abstracts and study titles were initially reviewed independently by Antonio Nicolucci and Marco Orsini Federici. Subsequently, full articles of potentially appropriate trials were downloaded and screened for inclusion. The following article types and studies were excluded: reviews, editorials, case reports, clamp studies, studies of less than 4 weeks’ duration, studies involving healthy volunteers, animal studies, and studies published in languages other than English.
Two reviewers (Antonio Nicolucci and Marco Orsini Federici) discussed and decided upon the final studies for inclusion in the meta-analysis. Only studies that reported both means and standard deviations (SDs) for the outcomes of interest were included in the final analysis.
Data extraction was performed independently by Antonio Nicolucci and Marco Orsini Federici, who prepared a data extraction spreadsheet containing data collated from the final studies selected. Extracted data included study reference details, patient numbers, patient characteristics, measures of GV, incidence of hypoglycaemia, and other relevant study information. Any discrepancies in the data gathered were discussed by Antonio Nicolucci and Marco Orsini Federici until a consensus was reached.
Outcomes
The primary outcomes were the mean differences between RAIAs and RHI at the end of the study in PPG (mg/dL), preprandial BG (mg/dL) and EOT HbA1c (%). Outcomes for the sensitivity analysis were the mean differences between RAIAs and RHI at the end of the study regarding the difference between PPG and preprandial BG (mg/dL) and the change from baseline in HbA1c (%).
Data Synthesis and Analysis
The meta-analysis of the effects of RAIAs versus RHI on mean pre- and postprandial BG levels and EOT HbA1c was conducted as follows: weighted mean differences and 95% confidence intervals (CIs) between RAIAs and RHI were generated to yield the overall point estimate from the meta-analysis. Risk estimates from individual studies were pooled by using random- or fixed-effects models [
13]. The meta-analysis did not adjust for covariates that may have affected the results of the included studies. When the
P value for heterogeneity was less than 0.10, a random-effect meta-analysis was performed. For
P values of 0.10 or higher, a fixed-effects model was used. Heterogeneity across studies was assessed using the heterogeneity
χ2 (Cochran
Q) statistic and the
I2 test [
14]. These analyses were also performed using data showing the effects of RAIAs and RHI on the difference between postprandial and preprandial BG levels, and the difference between EOT and baseline HbA1c (sensitivity analyses). Risk of bias was not assessed.
All analyses were stratified according to the type of diabetes and were performed using a macro routine written in SAS language (SAS Release 9.4; Cary, NC, USA).
Compliance with Ethics Guidelines
This article is based on an analysis of previously conducted studies already in the public domain and does not include any studies with human participants or animals performed by any of the authors. Data were obtained via a published literature search.
Discussion
The results of this meta-analysis, which incorporated 27 studies and more than 7000 patients, have shown that the use of RAIAs significantly reduced PPG compared with the use of RHI in patients with T1D. EOT HbA1c was also significantly lower with RAIAs than RHI in patients with T1D. Sensitivity analyses showed that the favourable results with RAIAs versus RHI were maintained when the difference between pre- and postprandial BG and the change from baseline in HbA1c were analysed.
These results are in agreement with other studies comparing the effect of RAIAs versus RHI on PPG/glucose excursions in patients with T1D. These include a crossover study of 18 patients who received various rapid-/short-acting insulins, including insulin lispro, and RHI following stabilisation of preprandial glycaemia and before the ingestion of a standardised meal [
42]. In this study the maximum postprandial glycaemia in the first 3 h post-injection was higher with RHI than with insulin lispro, and the area under the curve was also higher with RHI. These data demonstrate a reduction in postprandial glycemic excursions with insulin lispro versus RHI [
42]. In a separate study involving 21 patients with T1D, a single-dose administration of premeal insulin glulisine resulted in a lower maximum glucose excursion, a lower total BG exposure, and a lower maximum BG concentration compared with that seen after the administration of RHI [
43]. These results confirm those of several older studies in patients with T1D that showed a reduction in PPG/glucose excursions with RAIAs versus RHI [
27,
44‐
46].
In the current meta-analysis, the number of studies involving patients with T2D was insufficient to allow us to draw valid conclusions; however, we would expect the results to be similar to those seen in patients with T1D given the available literature on the topic. For example, a randomised, multicentre study of 29 patients with T2D who received insulin aspart or RHI at a dose aimed at achieving a PPG below 140 mg/dL for 24 months showed that for the first 9 months of the study, patients receiving insulin aspart who had PPG levels of less than 140 mg/dL were on a significantly lower dose than those receiving RHI who achieved the same endpoint [
47]. A randomised crossover trial involving 13 patients with T2D who received insulin aspart or RHI demonstrated a significantly lower BG increase after a standardised meal with insulin aspart versus RHI [
48]. Subanalyses of the A
1chieve and IMPROVE™ studies showed that switching from RHI to insulin aspart led to significant decreases in both HbA1c and PPG [
49,
50].
In terms of estimation of GV, there is continued debate about the most suitable metrics for assessment of this outcome [
1,
51]. HbA1c is a measure of long-term, average glycemic control [
51] and serial measurements of this metric can be used—as in our sensitivity analysis—as a marker of long-term GV [
1,
12]. Short-term, within-day GV primarily reflects fluctuations between the fasted (preprandial) and fed (postprandial) states and can be assessed using data generated via self-monitoring of BG or continuous glucose monitoring [
51]. In this analysis, we used the change in BG following a normal meal at home (pre- vs PPG) as an estimator of short-term GV in the sensitivity analysis. Both HbA1c and change in BG showed significant between-group differences in favour of RAIAs (change in HbA1c from baseline to EOT, T1D and T2D; change in BG following a meal, T1D). Since baseline HbA1c is a strong predictor of HbA1c response to insulin therapy [
52‐
54], it was important to investigate whether potential differences in baseline HbA1c could have impacted the EOT HbA1c results obtained in our analysis. That the change from baseline in HbA1c still showed between-group differences in favour of RAIAs strengthens the main analysis results. Of the nine studies included in the analysis of preprandial BG, eight used neutral protamine Hagedorn insulin as the basal insulin, and one used insulin glargine; this lack of variation in the basal insulin used in the included studies prevented analysis of whether the choice of basal insulin affected preprandial glucose levels (and hence the change in pre- to postprandial BG level).
The outcomes assessed in the current analysis have additional value in the assessment of patients with diabetes: both FPG and PPG have been identified as major risk factors for complications of diabetes [
5,
6], and PPG has been shown to independently correlate with HbA1c, making it a suitable marker for evaluating glycemic control. Although analysing the impact of decreased GV on adverse outcomes in patients with diabetes is beyond the scope of this analysis, it has been known for many years that chronic hyperglycaemia is associated with both macrovascular and microvascular complications [
5‐
7]; therefore, agents that adequately control GV are also likely to control diabetes complications.
In some of the analyses reported here, the
χ2 values, their associated
P values and the
I2 values indicated evidence of heterogeneity. This was particularly the case for the change in HbA1c from baseline to EOT in the patients with T1D (
χ2 = 540.52;
P < 0.0001;
I2 = 100%; 95% CI 63–100%) (Fig.
4a). However, the heterogeneity estimates reported herein, particularly those relating to patients with T2D, should be interpreted with caution; Cochrane’s
Q (the
χ2 statistic) has low power to reject the null hypothesis of homogeneity in meta-analyses involving small numbers of studies, and the
I2 statistic should also be interpreted cautiously because of its potential for bias in this situation [
55]. The small number of studies included in many of the current analyses resulted from the absence of required data in many of the studies selected for inclusion. This and the heterogeneity of the studies are limitations of the present analysis. As a result, valid conclusions regarding the effect of RAIAs versus RHI on glycemic control in patients with T2D could not be drawn. Additional limitations of the study are that the meta-analysis was not adjusted for covariates, and a risk-of-bias analysis was not performed.