In a paper published in Diabetologia last year, Martin et al. [1] reported that self-monitoring of blood glucose (SMBG) in the observational retrospective cohort Retrolective Study ‘Self-monitoring of Blood Glucose and Outcome in Patients with Type 2 Diabetes’ (ROSSO) was associated with significantly decreased diabetes-related morbidity and all-cause mortality even after adjusting for age, sex, comorbidities at diagnosis of diabetes (hypertension, CHD and history of stroke), fasting plasma glucose concentrations, triacylglycerol levels and treatment. These findings remained true for patients not taking insulin. In contrast, in the present issue of Diabetologia, Davis et al. [2] report that, in the observational, prospective Fremantle study, SMBG was not associated with improved outcomes of diabetes-related morbidity, cardiac death or all-cause mortality after adjustment for confounders. The confounders adjusted for in the Fremantle study were age, sex, duration of diabetes, prior CHD, prior peripheral arterial disease, neuropathy, retinopathy, microalbuminuria, systolic blood pressure, total serum cholesterol and current smoking. Surprisingly, there was a significant 79% increased risk of cardiovascular mortality in those patients who were not taking insulin but were performing SMBG. How can the disparate results of these two large studies be explained and what lessons can we learn from them? First, there are many factors that might impinge on the outcomes, especially the macrovascular ones, and taking all the appropriate confounders into account may be problematic. Second, there may be a self-selection process for physicians in terms of who is recommended for, and patients who perform, SMBG. Evidence for the former is the finding that those who perform SMBG have higher glycaemia than those who do not [1, 3]. Evidence for the latter was supplied by a large observational retrospective health plan study in which patients who performed SMBG more frequently had lower HbA1c levels [4]. However, a self-administered questionnaire or a computer-assisted telephone interview administered to plan members (83% of whom responded) revealed that self-care practices and healthy lifestyle behaviours were significantly more common in patients who performed SMBG more frequently. By definition, an observational study is not an experimental one, i.e. interventions are chosen rather than randomly assigned. This leads to the possibility of selection bias which, in turn, means that outcomes may not be caused by the intervention under study [5]. Although identifiable differences can be taken into account when the data are analysed, one cannot be certain that these adjustments are adequate because all of the relevant differences may not be Diabetologia (2007) 50:497–499 DOI 10.1007/s00125-006-0582-z
Read full abstract