Introduction

The major challenge in diabetes management is preventing or delaying the progression of diabetes-related complications. The landmark trials in type 1 [1,2] and type 2 [3,4] diabetic patients have demonstrated a strong inverse relationship between blood glucose levels and the occurrence of long-term microvascular complications.

Recently an increasing body of evidence, not yet fully supported by randomized clinical trials, has suggested that postprandial hyperglycemia is directly implicated in cardiovascular complications [5-7]. The hypothesis is that a better control of postprandial glucose (PPG) levels may be an important therapeutic target in the prevention of cardiovascular complications in diabetic patients.

The use of rapid-acting insulin analogues (Lispro, Aspart, Glulisine) provides lower postprandial glucose excursions in people with diabetes [8-11]. Nevertheless their effects on long-term outcomes are unknown.

A recent observational study showed that the rapid-acting insulin analog gluslisine was associated with a reduced incidence of macro- and microvascular events in patients with type 2 diabetes in comparison with human regular insulin [12]. However, the primary care setting used in that study didn’t allow to capture all events but only those registered by the general practitioners (GPs).

The aim of this study is therefore to compare the use of rapid-acting insulin analogues and human regular insulin in relation to the first occurrence of any diabetes-related complication in a cohort of diabetic patients without macrovascular disease at baseline through the analysis of administrative databases.

Methods

We conducted a retrospective, population-based cohort study using administrative data from four local health authorities in the Abruzzo Region of Central Italy, that comprise about 900,000 inhabitants (68% of the overall regional population).

Details regarding data sources and study design have been published previously [13]. Briefly, a record-linkage analysis of outpatient drug prescriptions, hospital discharges, prescriptions for laboratory tests, services use and specialistic consultations and the civil registry was performed including data from January 1, 2005 to December 31, 2008. All information was linked through a unique and anonymous personal identification code. In order to protect the patient’s privacy the inverse process was prevented by the deletion of the conversion table. Because this automated system is anonymous, neither ethical committee approval nor informed consent was required. Furthermore the anonymous data file is routinely used by the regional health authorities for epidemiological and administrative purposes. Permission to use it for the present study was granted by the responsible authority.

Study population

A flow chart of the study population selection is depicted in Figure 1. From the source population we identified individuals with at least four prescriptions of insulin agents between January 1, 2005 and December 31, 2005 (baseline year).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Patient selection flowchart.

https://doi.org/10.1371/journal.pone.0079762.g001

At baseline, we excluded patients if they had (i) at least one prescription of antiplatelets (excluding low-dose aspirin), pentoxifylline or nitrates, which are commonly used for the treatment of ischemic cardiovascular diseases, or a diagnosis of macrovascular disease in hospital discharge or procedure codes reported on prescriptions for services use [13]; (ii) no prescription of either human regular insulin or rapid-acting insulin analogues;(iii) prescriptions of both human regular insulin or rapid-acting insulin analogues during the whole study period (January 1, 2005 to December 31, 2008). Individuals who met the criteria were divided into two mutually exclusive groups (study cohorts): those treated with human regular insulin and those treated with rapid-acting insulin analogues throughout the entire follow-up. In both groups the bolus insulin was used alone or in combination with other types of insulin (premixed insulin and basal insulin) or oral hypoglycemic drugs, or both.

Each individual accumulated person-years of follow-up from January 1, 2006 until the date of hospital admission for any diabetic complications, censoring (death or emigration), or December 31, 2008 (i.e. the end of the follow-up), whichever occurred first.

Results

Follow-up

During the 3-year follow-up, there were 286 incident events (31.3%) in the human regular insulin group and 235 (17.1%) in the rapid-acting insulin analogue group, with a difference between the two treatment groups in the cumulative HR for any diabetic complications that increased over time (p<0.0001 by the log-rank test) (Figure 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Cumulative hazard according to the two treatment group of (A) Any diabetes-related complications (B) Macrovascular complications, (C) Microvascular complications and (D) Metabolic complications.

https://doi.org/10.1371/journal.pone.0079762.g002

After multivariable adjustments, using human regular insulin group as the reference group, the HR was 0.56 (95% CI: 0.45-0.69) for any diabetic complications in the rapid-acting insulin analogue group. Regardless of secondary outcomes, we found 203 macrovascular events (22.2%) in the human regular insulin group and 147 (10.7%) in the rapid-acting insulin analogue group, with a significant reduction in macrovascular complications associated with the use of rapid-acting insulin analogues (HR=0.56; 95% CI:0.43-0.72) (Table 2). Among macrovascular events, a highly significant reduction was observed for each specific clinical condition in the rapid-acting insulin analogue users. A statistically significant difference was also found in the risk of metabolic complications. Among these, only the difference in hyperglycemic events remained significant after multivariate adjustments. Conversely, no difference between the two groups was found for microvascular disease rates (Table 2).

	Unmatched cohorts				Propensity-score matched cohorts
Complications	Human regular insulin n (%)	Rapid-acting Insulin Analogues n (%)	Adjusted HRa (95% CI)	Human regular insulin n (%)		Rapid-acting Insulin Analogues n (%)	HR (95% CI)
Any complication	286 (31.3)	235 (17.1)	0.56 (0.45-0.69)	165 (29.1)		132 (23.3)	0.73 (0.58-0.92)
Macrovascular	203 (22.2)	147 (10.7)	0.56 (0.43-0.72)	110 (19.4)		87 (15.4)	0.73 (0.55-0.95)
Cardiovascular disease	107 (11.7)	95 (6.9)	0.62 (0.44-0.86)	63 (11.1)		53 (9.4)	0.77 (0.54-1.10)
Peripheral vascular disease	21 (2.3)	14 (1.0)	0.32 (0.14-0.75)	14 (2.5)		8 (1.4)	0.53 (0.22-1.27)
Cerebrovascular disease	75 (8.2)	38 (2.8)	0.55 (0.35-0.88)	33 (5.8)		26 (4.6)	0.73 (0.44-1.20)
Microvascular	43 (4.7)	53 (3.9)	0.73 (0.45-1.20)	27 (4.8)		26 (4.6)	0.89 (0.54-1.48)
Metabolic	52 (5.7)	37 (2.7)	0.36 (0.21-0.59)	35 (6.2)		21 (3.7)	0.55 (0.32-0.94)
Hyperglycemia	34 (3.7)	21 (1.5)	0.36 (0.21-0.59)	21 (3.7)		11 (1.9)	0.48 (0.23-0.99)
Hypoglycemia	19 (2.1)	16 (1.2)	0.51 (0.23-1.12)	14 (2.5)		12 (2.1)	0.62 (0.29-1.34)

Table 2. Cumulative incidence and Cox Proportional Hazard Ratios (HRs) of diabetes-related complications in patients treated with human regular insulin or a rapid-acting insulin analogue.

IA= Insulin Analogues. Human regular insulin reference group. ^a Adjusted for variables reported in Table 1 plus mean daily dosage of both insulin and oral hypoglycemic agents in quartiles.

CSV

Download CSV

The PS matched Cox models supported these results (Figure S3). Specifically, the differences between the two groups in relation to the risk of any diabetic (HR=0.73; 95% CI:0.58-0.92), macrovascular (HR=0.73; 95% CI:0.55-0.95) and metabolic complications (HR=0.55; 95% CI:0.32-0.94) remained statistically significant. However among macrovascular events, no difference between the two groups was found for any specific clinical condition. The analysis confirmed the lack of statistical significance for the difference in microvascular complication rates (HR=0.89; 95% CI:0.54-1.48) (Table 2).

Discussion

The major finding of our study is that patients given rapid-acting insulin analogues developed fewer diabetic complications (especially, fewer macrovascular events) than those given human regular insulin.

Due to the lack of randomization, significant differences in baseline characteristics exist between the two treatment study group. Patients on human regular insulin are expected to be at higher risk of developing cardiovascular complications than those on rapid-acting insulin analogue, because they were older and more likely to use low-dose aspirin and antiarrhythmic and antihypertensive agents. On the other hand, patients treated with bolus insulin analogues were more often men, reported a higher number of HbA1c tests and a higher mean daily dose of all anti-diabetic agents than patients receiving human regular insulin. Despite these differences between the two study cohorts, the reduction of both primary and secondary outcomes remained significant even after adjusting for these and other potential confounders evaluated in this study. These results were also supported by PS matched-pair analyses performed on the two balanced subsamples.

Moreover, based on our previous finding that a lower risk of any diabetic complication was associated with the use of insulin glargine [13] we carried out the analyses on the subcohorts of patients who did not receive any prescription of insulin glargine during the whole study period in order to evaluate any possible interference of glargine use on the study results. In this subgroup analysis, a statistically significant reduction for any diabetes-related complication was still observed in patients treated with rapid-acting insulin analogues. Similarly, macrovascular and metabolic events were markedly decreased in patients given rapid-acting insulin analogues as compared with those receiving human regular insulin.

Our findings were in line with a previous observation suggesting that the use of rapid-acting insulin analogue glulisine could be associated with a risk reduction of 24% for macrovascular events in patients with type 2 diabetes as compared with the use of human regular insulin [12]. In that study, the assessment of the outcomes was based on ICD codes registered by GPs only. The relevance of our study is the use of a large-scale, population-based cohort and the ability to draw information from a real-world setting through the analysis of administrative data. As a matter of fact, the majority of diabetes-related trials do not focus on significant cardiovascular outcomes and are performed in study groups that are not representative of the general diabetic population [15]. We believe that the strength of administrative data for our study question is in the ability to capture all acute diabetes complication events that is extremely important for the assessment of the quality of care in the real-world.

Nevertheless, we acknowledge that an observational study using administrative data is prone to potential biases. In particular, our findings may be subject to confounding by indication due to the lack of randomization. In fact the allocation of treatment is based on the clinical judgment of physicians and may be related to a different cardiovascular risk profile between the two groups. However, all PS matching resulted in full balance of all patients characteristics and is likely to have minimized confounding by indication enhancing the comparability of patients. In addition, Cox regression analyses were conducted with macrovascular complications as the endpoint after stratification for gender or age (<65 years versus ≥65 years); the results remained essentially unchanged when the association between study treatment and macrovascular events was examined in the different subgroups (Table S3).

The presence of unrecognized confounders could lead us to over-estimate the magnitude of the association between exposure and outcome as compared with the results of randomized clinical trials. The analysis of confounders that were included in the Cox-models was limited to variables collected by administrative data. A good deal of information on the degree of metabolic control, severity and duration of diabetes and on lifestyle attitudes, which are known to influence the likelihood of developing complications [16], is not available in our data sources. Nevertherless, we attempted to limit the role of these confounders by adjusting for age, sex, concomitant drugs and the presence of microvascular or metabolic events at baseline (the latter two factors may be a proxy of duration and severity of diabetes); moreover we reported all analyses in the two cohorts after PS matching.

Another limitation of our study is that our database does not allow us to ascertain with confidence the type of diabetes as a potential confounder. The higher proportion of younger patients in the rapid-acting insulin analogue group suggests the possibility of a higher prevalence of patients with type 1 diabetes. Therefore, we attempted to explore the effect of this variable on our results by using the following criteria for identifying type 1 diabetes cases: no records of oral hypoglycemic agents, specific ICD9 code for type 1 diabetes (250.x1 or 250.x3), diabetic ketoacidosis episode (250.x1), age<40 year [17]. According to this analysis, the percentage of type 1 diabetic patients was 5.6% in the human regular insulin group and 15.7% in the rapid-acting insulin analogue group. However, our results remain consistent after the inclusion of this variable in all models or after the exclusion of type 1 patients from the analyses (data not shown).

Our findings could be explained by a better quality of care in the group treated with bolus insulin analogues that could be associated with a more favorable prognosis in that group [18]. However, we also adjusted for the number of HbA1c tests as a proxy for the quality of diabetes care in all models.

Additionally, our study population included diabetic patients free of macrovascular diseases at baseline, who may have been misclassified because of medical errors in the medication and diagnosis codes or the short baseline period (1 year). However, this type of error does not seem to be very high because the cohort of diabetic patients with macrovascular disease at baseline represented about 30% of the overall sample, which is consistent with results from previous landmark studies [2,3]. Finally, because outcome data were drawn from hospital discharges, these data can be misclassified. However, outcome misclassification can be expected to occur independently of the insulin exposure under study and therefore a non-differential misclassification would have biased the results of the study towards the null hypothesis [19].

Despite these intrinsic limitations of observational data, several speculative explanations for our findings should be considered. In particular there are three demonstrated clinical effects of the rapid-acting insulin analogues that could explain why they may be more effective than the human regular insulin in reducing the risk of diabetic complications.

1. Overall glycemic control may be better with rapid-acting insulin analogues than with human regular insulin. These agents can be injected immediately before the meals, whereas human regular insulin should be injected 30-45 min before the meals [20]. In reality, most patients do not follow the strict time injection required for the human regular insulin administration and therefore in many cases the PPG levels are not optimally controlled with this agent [19,20]. In a randomized crossover study Annuzzi et al [21] showed that insulin lispro, in combination with neutral protamine Hagedorn (NPH) insulin, significantly reduced HbA1c (0.2%) in type 1 diabetic patients. Furthermore, in a large-scale multicentre trial Hermansen et al [22] demonstrated that the combination of the basal-bolus regimen (insulin detemir and insulin aspart) provided a significant reduction in HbA1c by 0.22% when compared with traditional insulins (NPH and human regular insulin) in patients with type 1 diabetes after 18 weeks of treatment. More recently, in a meta-analysis Mannucci et al [23] showed that the use of rapid-acting insulin analogues in type 2 diabetic patients provides a small but statistically significant reduction of HbA1c in comparison with human regular insulin. In our opinion the magnitude of the HbA1c reduction found in these studies in patients treated with rapid-acting insulin analogues is not sufficient to explain our findings.
2. Another plausible account might be that the greater beneficial effect of bolus insulin analogues on the reduction of diabetic complications (particularly macrovascular events), might be also attributable to their ability to better control PPG excursions. In line with those results, the Diabetes Intervention Study (DIS) [24] and The San Luigi Gonzaga Diabetes Study [25] demonstrated the importance of postprandial blood glucose control in preventing cardiovascular events in patients with type 2 diabetes. Conversely, the HEART2D trial [26] did not show any beneficial effect of specifically treating postprandial hyperglycemia in relation to the risk of cardiovascular events in diabetic patients. However, a post-hoc subgroup analysis of the HEART2D on a specific population (older patients) has found a lower risk of a subsequent cardiovascular events associated with the use of prandial insulin lispro [27]. In the light of these findings, the role of postprandial hyperglycemia in the development or progression of diabetic complication remains a sensitive and controversial issue. Our results refer to a cohort of diabetic patients without a recent cardiovascular event; in this respect they don’t resemble the profile of the HEART2D cohort. Unfortunately, the potential role of PPG levels in reducing the risk of developing long-term diabetes outcomes cannot be evaluated in our study given the lack of information on metabolic parameters in the database.
3. Finally, several studies have also shown a lower occurrence of hypoglycemia with the use of rapid-acting insulin analogues [28,29]. In our study we found a lower risk of metabolic events in the rapid-acting insulin analogue group. However, the incidence of hypoglycemic events, when analyzed separately, was not statistically different after adjusting for the covariates. Only hypoglycemic episodes requiring hospitalization were taken into account in our analyses and therefore it is possible that we underestimated the overall incidence of hypoglycemic episodes and the potential difference between the two treatment groups in that these conditions are more likely to be treated in an outpatient setting [30,31]. Hypoglycemic events have been recently recognized as important markers of cardiovascular risk in diabetic patients [32].

Conclusions

This analysis from a real-world setting suggests that the use of rapid-acting insulin analogues, in comparison with human regular insulin, is associated with a lower risk of diabetic complications and particularly macrovascular events. Future trials may be required to determine the potential long-term clinical benefits of bolus insulin analogues in preventing cardiovascular diseases in diabetic patients.

Supporting Information

Acknowledgments

The author also express their gratitude to the following persons for their collaboration in data collection: Vito Di Candia and Angelo Muraglia, Abruzzo Regional Health Care Agency (ASR-Abruzzo), Italy; Francesco Di Stanislao, Polytechnic University of the Marche Region, Italy; Angelo Mucciconi, Camillo Odio and Lanfranco Venturoni, Department of Health Care Abruzzo Region.