We thank Drs. Aalders and Pouwer for sharing their concerns (1). The authors raise some important issues, such as the assessment of health-related quality of life (HRQoL) in observational studies. However, a number of definitions of HRQoL exists and there is no universal agreement on an appropriate definition. In our study (2), we have used the physical and mental unhealthy days measure developed by the Centers for Disease Control and Prevention (CDC) as a proxy measure for HRQoL. Having the advantages of simplicity and brevity, this measure has been used in other telephone surveys, including the Behavioral Risk Factor Surveillance System (BRFSS), for a long time. Furthermore, several studies show that physical and mental unhealthy days are strongly associated with personal health behaviors, socioeconomic and health-related data, measures of disability, and other HRQoL scales (3,4). We pointed out in our limitations section that unhealthy days are a general indicator of HRQoL, and we agree that a more comprehensive assessment would provide a more detailed picture of HRQoL. We opted to use the CDC measure as it is an easily administered and validated proxy for HRQoL that provides comparable data to other North American telephone surveys, such as the BRFSS.
Aalders and Pouwer also raised concerns that “transformation of a continuous score (based on the two questions) into a dichotomized score has resulted in a considerable further loss of information” (1). Regular regression models are more appropriate for normally distributed outcome variables. However, our unhealthy days outcome is not normally distributed nor a strictly continuous score. It is a composite score that is based on two variables with skewed, truncated, unknown distributions. Dichotomizing the variable might not be the best solution from a purely statistical point of view; higher-order (nonlinear) models that take the complex distributions into account might be an alternative strategy. Although dichotomization may have resulted in the loss of some information, the ease of interpretation of this strategy may be more meaningful for the clinician (5). Our cutoff point for the dichotomization was not data-driven but based on clinical evidence from the literature. We have also conducted several sensitivity analyses where we have used different cutoff points and where we have trichotomized our outcome variable. Our findings remained largely unchanged in these sensitivity analyses.
Finally, Aalders and Pouwer (1) suggested a more careful interpretation of the results. We agree with the authors that more research and monitoring/intervention studies are needed before we can implement intervention and prevention programs. Nevertheless, to our knowledge the Montreal Diabetes Health and Well-Being Study (DHS) is the only population-based study of people with diabetes where a wide spectrum of outcome and risk factors was assessed annually over 5 years, providing unique evidence regarding recurrent subthreshold depressive symptoms. As pointed out in our conclusion, recurrent subthreshold depressive symptoms might be a risk factor for poor health outcomes in diabetes. We hope that our results will stimulate further research in this area to get a better understanding of the role of recurrent subthreshold depressive symptoms in the course of diabetes.
Article Information
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
