Authors' response.

We thank the authors for a close reading of our article1. Given the known biases in recruiting study participants for a case-control study, we decided to choose symptomatic HCWs who were tested for SARS-CoV-2 infection to maintain evenness in the way cases and controls were selected. We would like to posit that the reasons for which asymptomatic HCWs got tested were likely to be different from those of symptomatic HCWs. Hence, to maintain comparability between the cases and controls, we decided to include only symptomatic HCWs. We tried to adhere to the basic tenets of case-control investigations - the cases and controls should be comparable, except in that the case group experienced the outcome of interest. In addition, we would like to add that an analysis of one million tests conducted in India between January and April 2020 has shown that about 28 per cent of SARS-CoV-2-positive patients are asymptomatic2.

The standard practice in developing logistic regression models begins with the selection of independent variables using multiple strategies - known or established theories, existing evidence, exploratory analyses or a combination of these and other strategies3. The purpose of the univariate analysis was to identify the variables that were more likely to be statistically and biologically associated with the outcome of interest. To construct a parsimonious model, we chose to include biologically plausible variables which met a cut-off value (P<0.1). This is clarified in the subsection titled ‘multivariate analysis'. Further, we would like to emphasize that it is important to limit the number of independent variables to avoid a mathematically unstable model with limited generalizability beyond the current data4. In order for readers to appreciate the process, and to declare the associations observed through the univariate analyses, we chose to present both analyses.

While we acknowledge the lower response rate, this is a known shortcoming of telephone-based surveys. While in-person interviewing remains the method providing the highest yield in terms of response efficiency and representativeness, it was an untenable strategy given the realities of the ongoing pandemic and restrictions imposed on the movement of people by the nationwide lockdown. Also noteworthy is that, compared to online, mail, or self-reported data collection, telephone-based surveys provide better representativeness, more complete data and higher data yield56. To improve the response rate, we employed different strategies such as training of interviewers and multiple call attempts at different times of the day. Further, our study received higher response rates than similar methodologies employed to cover HCWs in India (paediatricians: 57%)7 and abroad (Germany: physicians, 56%8; France: physicians, 59%9 and USA: internists, 64%10).

The study participants were asked to declare the side effects experienced by them in our investigation. As noted in the ‘Results’ section, a very small proportion of the participants self-reported adverse effects linked to HCQ intake, and the frequency of occurrence of side effects was not significantly different across the case and control groups1.