The Healthy User Effect in Studies of Statins in the Critically Ill

    loading  Checking for direct PDF access through Ovid


We read with interest the article published in a recent issue of Critical Care Medicine by Mather et al (1) who examined the association between statin exposure and delirium in a medical ICU population. Their results are intriguing, suggesting that statins may reduce the risk of delirium by half. Given the implausibly large magnitude of this putative effect, the results seem more plausibly explained as a manifestation of the “healthy user effect” (2) than a true effect of statins. The healthy user effect is well described as a powerful source of unmeasured confounding and has been responsible for other implausibly large protective associations between statins and a variety of adverse outcomes (2). In the critically ill population, this phenomenon may account for the apparent protective effects of statins on acute kidney injury, acute respiratory distress syndrome, and chronic obstructive pulmonary disease exacerbation, all of which were refuted by subsequent randomized trials (3, 4). Although Mather et al (1) were unable to determine whether statin use was the result of continuation of home medications, the lack of routine indications for starting statin treatment in the acute medical ICU setting suggests that the majority of use was continuation of chronic treatment.
In addition, the healthy user effect may have been magnified by restricting the study to patients who received statins for the duration of the hospitalization, excluding those who only received statins intermittently. Chronic oral medications are often held in patients who cannot tolerate oral intake as a result of a high severity of illness. Consequently, including only patients able to receive statins for the entire duration may have selected out the healthiest statin users, who were compared with control patients who had no requirement for being able to tolerate oral medications. It therefore seems likely that statin-exposed patients had a lower severity of illness than control patients. Although the authors matched on a propensity score that included a comorbidity score, the score did not include variables that adequately reflect severity of illness in the ICU setting, such as vasopressor use or mechanical ventilation. The result of the analysis to gauge sensitivity to unmeasured confounding found a gamma of 1.69, which suggests that the results are fairly sensitive to bias according to the scale suggested by Rosenbaum (5).
Understanding the potential beneficial effects of statins in the critically ill continues to be an important area of inquiry. Any study that hopes to examine these effects in the absence of randomization must give careful attention to the healthy user effect, which can be particularly difficult to control using measured variables or scores based on them (2). In other settings, bias from the healthy user effect has been reduced by comparing statin users with “active controls” patients receiving a different chronic preventative medication that does not have any plausible effect on the outcome of interest (2). Such an approach might be a useful adjunct to the current study (1) to help understand the robustness of the findings. At a minimum, it might be useful to conduct a sensitivity analysis in patients receiving other oral medications throughout the duration of hospitalization.
    loading  Loading Related Articles