Detect The Extremes That Usually Remain Undetected In Conventional Observational Studies

In medicine, clinicians and researchers tend to score the quality of available evidence according to the famous pyramid of evidence–based medicine in which controlled trials and their meta-analyses are seen as superior to observational studies (i.e., yielding evidence of a better quality). Indeed, we frequently say that what is suggested in observational studies must be confirmed in 1 or more controlled trials. This approach is not always practical and is sometimes even unrealistic, as not everything can or will be captured in the setting of a controlled trial. Under certain circumstances, observational studies could perform better, in particular when they are large enough to capture reasonable numbers of certain events. Examples of this are observational studies using data mining of blood glucose concentrations in critically ill patients.Glucose control in the critically ill has been extensively studied by clinicians and researchers over the last 2 decades. Numerous observational studies have shown a clear association between hyperglycemia and mortality in intensive care unit (ICU)4 patients (1–3), suggesting that a strategy aiming at normoglycemia with intravenous insulin could reduce mortality. Indeed, a series of controlled trials in different cohorts of critically ill patients showed that this strategy, frequently called strict glycemic control (SGC), improved the outcome of critically ill patients (4–6). This provides an excellent example of how evidence–based medicine works: what was suggested in observational studies was confirmed in controlled trials. Unfortunately, subsequent controlled trials could not confirm the benefit of intensive insulin therapy, for several reasons explained elsewhere (7). One reason could be the higher incidence of severe hypoglycemia. Indeed, observational studies showed that in critically ill patients, severe hypoglycemia is associated with death as well, and the …