When managing type 2 diabetes the general focus is on achieving and maintaining good glycaemic control while minimising the potential for adverse events such as hypoglycaemia. The value of this approach has been shown from evidence gained in landmark clinical and epidemiological studies where the reduced incidence of micro- and macrovascular complications was apparent with intensive glycaemic control [7-14] and has further been confirmed in a published meta-analysis of observational studies [15]. To achieve these reductions in the incidence of complications, current UK guidelines from the National Institute for Health and Clinical Excellence (NICE) recommend target glycosylated haemoglobin (HbA1c) levels of between 6.5 and 7.5% (targets that are in line with guidance from other national and international bodies) [16]. Although this dogma has come under scrutiny and been challenged by some studies [17] it remains the central principal for the majority of treatment regimens [18] even if the initiation of insulin is often resisted by both patients and physicians [19].

Despite published guidance on the management of diabetes it is apparent that many patients fail to receive the appropriate standard of care (although the reasons are unclear and blame cannot be easily apportioned). The magnitude of this issue has been demonstrated by two studies examining type 2 diabetes management in the UK primary care setting [20,21]. In these retrospective cohort studies the authors found that not only were HbA1c measurements poorly recorded for the majority of patients, regardless of numbers of oral antidiabetic agents (OADs) prescribed, but even where data were available at least 40% of individuals failed to achieve a modest HbA1c target of 7.5% [21]. Despite this poor control the average time spent on monotherapy was 3.8 years and even after failing to achieve glycaemic control with two or more OADs the median time before commencing insulin therapy was 7.7 years from initiation of the final OAD [20,21]. This implies an average time to initiation of insulin of at least 11.5 years from initial diagnosis of type 2 diabetes. From the published data it is clear then that patients would not be maintaining their glycaemic control targets for a substantial proportion of this time. This has the obvious implication that patients are unnecessarily at an elevated risk of diabetes-related complications.

Even though there is a notable risk of hypoglycaemia which would at least partly offset improvements in quality-adjusted life expectancy associated with insulin administration there were still substantial benefits for the immediate versus delayed initiation of insulin. By performing a sensitivity analysis where no quality of life disutility was accounted for hypoglycaemia events the effect on outcomes of this consequence of improved glycaemic control was evaluated. Overall the incidence of hypoglycaemia associated with insulin therapy led directly to a decrease in quality-adjusted life expectancy of 0.21 QALYs compared to the base case analysis (0.55 versus 0.34 QALYs).

Model outcomes have been presented here as undiscounted values to more clearly demonstrate the potential outcomes for immediate versus delayed initiation of insulin. Although NICE recommends a discount rate of 3.5% per annum in the UK setting we have deliberately chosen not to conduct this study from an economic standpoint, but rather, relate only the potential clinical benefits that are foregone by delaying insulin initiation. However, as a matter of record, where outcomes were discounted the results for the base case showed a benefit for immediate versus delayed initiation of insulin in terms of life expectancy of 0.36 years (8.57 versus 8.20 years) and for quality-adjusted life expectancy of 0.18 QALYs (5.73 versus 5.55 QALYs). Although these discounted figures may not appear as dramatic as the undiscounted results they still represent substantial improvements that are in line or better than many health interventions seeking market approval.

As noted, this study makes no attempt to quantify the economic consequences of delaying insulin initiation. However there are likely to be direct medical cost savings associated with immediate initiation of insulin due to diabetes-related complications avoided. There are also other potential cost savings from a societal perspective, e.g. less need for carers, sustained productivity due to premature mortality or retirement avoided, etc. that are not accounted here.

Although the results from this modelling analysis unequivocally project improved outcomes for those initiating insulin and attaining HbA1c targets earlier than currently observed in the UK there are limitations to the study and these must be acknowledged. Most obviously this is a hypothetical modelling study and the outcomes are not based on randomised controlled clinical trial data (which would be ethically impossible to acquire). However, the inputs are based on recent real world data from the UK setting and where possible all data required to generate the analyses have been taken from the primary study publications [20,21] and then supplemented with other UK-specific data [30,31].

Additionally we have attempted to avoid bias by performing the analysis without a product-specific focus and it represents no particular commercial standpoint. In terms of the tool used to perform the analyses, the computer simulation model was developed as, and remains, a non-product-specific decision analysis tool that has been validated against a number of published clinical trials and shown to project robust and plausible outcomes [23]. In order to avoid unnecessary complexity and aid transparency the analysis was designed to allow minimal opportunities for bias or error (most inputs are unchanged between comparators) and was simplified where necessary. For example, treatment effects associated with insulin in terms of changes to systolic blood pressure and serum lipids (total cholesterol, high- and low-density lipoprotein cholesterol and triglycerides) were not modelled. This was intended to both simplify and maintain a conservative approach as improvements in these parameters were likely to reduce cardiovascular outcomes and one would predict that this would favour the immediate versus delayed initiation of insulin scenario.

Another potential criticism of this analysis is that it does not account any insulin specific effects outside of change in HbA1c and risk of hypoglycaemia. Although there is a distinct absence of reliable data regarding safety issues with insulin, one area that merits specific attention is in individuals administering insulin subsequent to an MI event. The published epidemiological data from the DIGAMI-2 trial reported a significant increase in the cumulative endpoint of death, reinfarction and stroke (hazard ratio of 1.42) for those administering insulin subsequent to an MI event [32]. However, these results are in contrast to outcomes from the original DIGAMI study [33] and the two conclusions are challenging to resolve with each other. As the authors of the post-hoc analysis of the DIGAMI-2 trial themselves point out, there is the strong possibility that results may be confounded by unknown covariates, and in a complex disease such as diabetes this is a common concern. As there is a limited amount of data regarding this in the context of the analysis presented here no attempt has been made to account for a potential increase in further cardiovascular events following MI when administering insulin. It should be noted though that in the base case analysis the time to onset for a first MI event was greater than the delay to initiation of insulin and so for the majority of cases there would have been no difference in outcomes (i.e. both arms would have been administering insulin).

Additionally, a similar analysis could be carried out whereby glycaemic control is postulated to be better maintained by any manner (more appropriate prescription of OADs or other pharmaceutical interventions, compliance issues, patient education, physician support, etc.), however it was decided that with the published evidence available the use of insulin represented a more realistic scenario. It should be noted that the publications by Calvert and colleagues which prompted this analysis did not directly assess the make up of the OAD based regimens and whether there were any issue with patient compliance. However, glycaemic control measurements for individuals were often not recorded perhaps indicating a systematic failure to manage the patients condition regardless of the intervention used. Regardless of the specific management practice addressed in this context the results clearly highlight the potential benefits for type 2 diabetes patients in the UK who attain glycaemic control targets compared with the current "real world" observations.

By considering only insulin effects we excluded the possibility that patients may have been prescribed one of the new classes of diabetes interventions (DPP-IV inhibitor or GLP-1 agonist) prior to initiating insulin. As these new compounds reach the market and become available for prescription they may well be used prior to insulin initiation, however, the data available to us largely predated their use and so they were not considered in this analysis. Their use and effect on the glycaemic control of type 2 diabetes subjects in the UK remains to be demonstrated.

We have performed this study using a computer simulation model that incorporates risk formula generated from studies that demonstrated benefits associated with intensive glycaemic control in terms of hard clinical endpoints. Following on from these findings the attainment and maintenance of good glycaemic control as judged by HbA1c measurements is still recommended in most patient groups by national healthcare bodies. It has to be acknowledged though that some patients, such as the very elderly and seriously ill, are unlikely to realise the full long-term benefit of tight glycaemic control especially when associated with elevated risk of hypoglycaemia. There have also been publications questioning the benefits of intensive glycaemic control and the consequences for cardiovascular outcomes, such as the findings from the ACCORD study [17]. The cause of the increased mortality with intensive versus standard therapy in this study have still to be elucidated but they may indicate that aggressive titration to very low HbA1c levels is not appropriate for all (or indeed any) patients. Until there is an adequate explanation and association between the mortality risk and level of glycaemic control we are unable to effectively model this outcome with adequate understanding.

An issue not accounted in this study is the findings of an observational study by Bowker et al. [34] who reported a significantly higher rate of cancer-related mortality in patients exposed to sulfonylureas and insulin compared with patients exposed to metformin. This issue has received considerable attention recently with the publication of a study by Hemkens et al. [35] who reported a positive association between cancer incidence (with diagnosis of malignant neoplasm defined as the primary endpoint) and insulin glargine. However, the results were derived from studies conducted over a relatively short period of time (5 years), potentially suggesting that insulin (glargine) may accelerate the rate of development of existing tumours rather than stimulating malignant transformation and the formation of new ones [36]. Despite this concern the FDA has stated that variations in patient characteristics across treatment groups could have driven any difference in risk and has not acknowledged any confirmed link or advised any discontinuation of treatment [37]. With no firm evidence to link the use of insulin (or any other diabetes therapy) to increased cancer risk this has not been addressed in this analysis.

There are other implications that have not been addressed by this modelling analysis in relation to the management of diabetes patients that deserve further investigation. Although not reported in the two studies by Calvert and colleagues [20,21] there is accumulating evidence that type 2 diabetes patients are also poorly managed in terms of blood pressure and cholesterol levels [38-40]. It is highly likely that further clinical benefits not captured in this analysis could be realised if these additional patient management strategies were adhered to. Also, the similarities and differences between the levels of glycaemic control and time to initiation of insulin in the UK and other settings are unknown. It is not possible perhaps to translate these results directly to other countries but it would be surprising indeed if similar trends were not observed elsewhere and hence comparable outcomes.