To the Editor:
We have read with great interest the review1 of our recent article2. Lim and Feldman provide valuable insights into the multiple pitfalls and limitations of retrospective studies, aiming to guide readers in better understanding the methodological underpinnings of different study designs and to help in selecting optimal studies to inform their practice. We thank Lim and Feldman for their comments; we take this opportunity to clarify several points.
First, we point out that our study does not meet the definition of a “prognostic study,” because “prognostic studies” imply a directional causal relationship between a predictor and an outcome. Indeed, Lim and Feldman quoted an article by Altman that defines prognostic studies as “clinical studies of variables predictive of future events as well as epidemiological studies of etiological risk factors”3. We explicitly stated throughout our report that our study was not designed to evaluate the causal/temporal relationship between hydroxychloroquine (HCQ) use and the decrease in antiphospholipid antibody (aPL) levels. Further, the information presented was not intended to aid clinicians in clinical decision making. Rather, because there was no previously published data looking at the association between HCQ and aPL levels, we believed it was important to communicate our findings to other researchers. The information learned from our study can be used to design future large-scale prospective studies to evaluate whether HCQ use may lead to a decrease in aPL levels and, through this effect, whether HCQ can be effective in primary and secondary prevention of the antiphospholipid syndrome (APS).
In our discussion section, we acknowledged most of the limitations and potential biases that were mentioned by Lim and Feldman, including differential selection, differential and nondifferential misclassification, and lack of information about disease activity. We also recognized that because of the retrospective design of our study, complete information on medication dosage, duration, and compliance was not available. Therefore, we chose HCQ exposure “ever” as our main variable of interest, similar to previous retrospective studies that evaluated associations between use of medications and aPL4,5. Although we were unfortunately unable to describe our methods in detail because of the word limit prescribed by the editors for this brief report, our models were indeed tested for confounding, interactions, and goodness of fit using the Hosmer-Lemeshow method6. While our sample represented a diverse group of patients with systemic lupus erythematosus (SLE) with a wide range of disease severity, we included only patients who met the American College of Rheumatology criteria for SLE7, and who had aPL measured twice at least 12 weeks apart, in accordance with the classification criteria for APS8. Finally, we performed several sensitivity analyses using different definitions of aPL positivity, and stratified by use of immunosuppressives to evaluate for a possible “by indication” bias and to test the robustness of our findings.
Despite the limitations described in our article and further analyzed by Lim and Feldman, our manuscript has several important strengths, including a relatively large sample from an ethnically diverse urban tertiary care center with defined SLE and with clinically significant aPL levels. We hope that our paper, as well as the comments by Lim and Feldman, will help design better studies to answer whether HCQ may be helpful in primary and secondary prevention of APS.