A paper published in the journal Diabetologia has attempted to analyse the effect of taking statins on incidence of type 2 diabetes in men, reporting a small increase in absolute risk. This analysis was accompanied by roundup comments.
Title, Date of Publication & Journal
Increased risk of diabetes with statin treatment is associated with impaired insulin sensitivity and insulin secretion: a 6 year follow-up study of the METSIM cohort
March 4th 2015
Study’s main claims – and are they supported by the data
This study provides evidence for an association between the use of statins and the risk of diabetes but presents what is almost certainly an over-estimate of the association.
The paper does not establish a causal link.
From the results presented it cannot be concluded that treatment with statins reduces insulin sensitivity and insulin secretion since the analysis done on these outcomes is flawed.
Because individuals were already taking statins at baseline it is not possible to disentangle the effect of statins from the effect of confounding by indication (see Limitations for details).
This was a large population-based study, which collected information on many of the key confounding factors.
Unfortunately, the authors have not taken account of all the confounders in their analysis. In observational studies of drugs, a critical issue is confounding by indication. Statins are prescribed for reducing cholesterol (in order to reduce the risk of cardiovascular disease –
CVD) and those at high risk of developing CVD are also at high risk of developing diabetes. Therefore, anyone prescribed a statin is likely to already have a higher risk of developing diabetes. The results shown in this paper indicate that this is indeed the case since, at baseline, individuals on statins had higher levels of HbA1c and 2 hour glucose, and had lower levels of insulin secretion and sensitivity. In other words, at baseline a greater proportion of the statin group already had impaired glucose tolerance, which would of course mean that they were already at greater risk of developing diabetes.
The authors have taken confounding by indication into account to a certain extent by adjusting for some of the risk factors for diabetes and CVD (BMI, age, family history, smoking and so on). However, although the paper shows hazard ratios for diabetes additionally adjusted for baseline values of the glucose / insulin measures individually, the authors do not present any results adjusted for all of these key factors simultaneously. Since each one of these factors reduced the apparent association between statins and diabetes, the estimate obtained when adjusting for all of these factors simultaneously would definitely be lower than the 1.46 figure quoted in the abstract and the press release.
In a similar vein, when the authors analysed the difference in insulin secretion and sensitivity at follow up they did not adjust for baseline values of these measurements. Because there were differences between the statin and non-statin groups at baseline in terms of these measures, any analysis of these measures at follow-up will be biased (they will over-estimate the effect of statins).
The study has a serious design limitation which is not acknowledged in the paper. At baseline, when measurements of glucose and insulin (and other confounders) were made, individuals were already taking statins. In other words these “baseline” measurements were not truly at baseline. This means that it is not really possible to accurately estimate the effect of statins – since we do not know actual starting levels of glucose, insulin secretion, and so on. Further, because no information is given on how long these people had already been taking statins it is impossible to make a judgement on the likely impact of this.
Confounding factor (confounder)
A factor associated with the outcome of interest (diabetes in this case) as well as the exposure being studied (statins in this case) which, if not accounted for, may lead to over or under-estimation of the association between the exposure and outcome.
Confounding by indication
This is a particular type of confounding that can occur in observational studies of drugs/treatments (i.e. treatment/drug studies that are not randomised controlled trials). Individuals who are prescribed a particular drug are taking that drug for a reason (they have an “indication” for the drug). These individuals are inevitably different some way (i.e. in terms of prognostic factors, thus in terms of risk of future health outcomes) from those individuals who are not taking the drug.
Before The Headlines is a service provided to the SMC by volunteer statisticians: members of the Royal Statistical Society (RSS), Statisticians in the Pharmaceutical Industry(PSI) and experienced statisticians in academia and research. A list of contributors, including affiliations, is available here.