Reply to Chen Et Al, Uchikoba Et Al, Siberry Et Al, and Vuorio Et Al

To the Editor—We read with great interest the letters by Uchikoba et al [1], Chen et al [2], Siberry et al [3], and Vuorio et al [4] related to our study “Effectiveness of Paxlovid in Reducing Severe Coronavirus Disease 2019 and Mortality in High-Risk Patients” recently published in Clinical Infectious Diseases [5]. In their letter, Uchikoba et al [1] queried the transition to inpatient treatment and indicators for coronavirus disease 2019 (COVID-19) severity. In Israel, Paxlovid is prescribed to COVID-19 patients almost exclusively on an outpatient basis, and all patients are followed by “COVID-19 operational rooms” through telephone calls (and in-person visits, as needed). Patients are transitioned to inpatient care if their condition deteriorates. While hospitalized, their clinical status is reported to the Israeli Ministry of Health on a daily basis. Concerns regarding the use of stepwise backward selection in the multivariable Cox regression models were raised by Uchikoba et al. They claimed that this method is problematic due to overfitting and multiple testing and that “rather than simply entering variables into a stepwise regression, substantive expert knowledge should guide the logical selection of variables.” As clearly stated in our Methods section, all candidate confounders included in the multivariable Cox regression model were selected based on established prior epidemiologic evidence of their association with high risk for severe COVID-19 [6]. Moreover, the biased estimate with backward selection is a concern only in small datasets with few events per variable [7]. This was not the case in our study that included a large dataset and an estimated 60 events per variable, considered to be very high. Running the Cox proportional hazard regression model with all candidate variables forced into the model (without backward selection) yielded exactly the same hazard ratios (HRs) and 95% confidence intervals (CIs) for Paxlovid and adequate COVID-19 vaccination status: HR, 0.54 (95% CI, .39–.75) and HR, 0.20 (95% CI, .17–.22), respectively.