Statistical Analysis

The primary study hypothesis was that cytotoxic chemotherapy within 35 days of COVID-19 diagnosis is associated with an increased hazard ratio (HR) of a severe or critical COVID-19 event. We repeated this investigation with longer (90 days) and shorter (14 days) cytotoxic chemotherapy treatment intervals from COVID-19 diagnosis. In prespecified secondary analyses (Data Supplement), HRs were similarly established using Cox proportional hazards regression models to define the association between severe or critical COVID-19 and each cancer-related variable as well as potential confounder covariates: age > 60 years, sex (male), obesity (BMI > 30 kg/m²), smoking history (former/current), pre-existing conditions, and Eastern Cooperative Oncology Group performance status (PS) ≥ 2 or Karnofsky PS < 80%. All HRs were formulated by dividing the hazard rate of patients positive for SARS-CoV-2 with the specific covariate characteristic by the hazard rate of patients positive for SARS-CoV-2 without the respective covariate characteristic. The Benjamini-Hochberg method was applied to adjust for multiplicity with a false discovery rate α of .10. For all analyses, censoring occurred at the time of documented symptom recovery or at the final observation end date of April 13, 2020, if patients continued in the study without documentation of a severe or critical COVID-19 event.

In a subsequent multivariable analysis, all covariates from the univariable analyses with P < .10 were analyzed together with potential confounders (age, sex, smoking, BMI, comorbid conditions, and performance status) in a Cox multivariable regression to test the association between each covariate and the primary end point in a single model. Variance inflation factors were defined for each covariate to assess for collinearity. Associations between variables were also assessed using paired χ² analysis.

Cox multivariable regression with the same covariates used in the primary analysis was also performed in defined patient subgroups, including patients with active disease, patients with inactive disease, patients treated with cytotoxic chemotherapy, and patients not treated with cytotoxic chemotherapy. The primary and secondary analyses were also repeated in sensitivity analyses using time from COVID-19 diagnosis as an alternate time onset point and three possible alternate end points: primary composite end point, time to ICU admission or death, or death.

Laboratory analysis that compared baseline (most recent value obtained between 14 and 90 days before SARS-CoV-2 RNA positive test) and peri–COVID-19 (closest value to date of SARS-CoV-2 RNA test within 3 days of positive test) values was also performed using paired Wilcoxon signed rank tests. We also compared peri–COVID-19 laboratory values between patients who did and did not meet the primary end point using Mann-Whitney U tests. All statistical analyses were performed using R version 3.6.1 software as well as packages survival_3.1-12 and rms_5.1-4 (R Foundation for Statistical Computing, Vienna, Austria).