Fig. 2: Period-based comparison of prior vaccination, prior documented infection, and risk of progression to various clinical outcomes.

Panels illustrate A adjusted odds ratios, fitted via logistic regression models, for receipt of 5, 6, or ≥7 COVID-19 vaccine doses (relative to zero doses) among all outpatient cases diagnosed in the indicated periods relative to those diagnosed between 1 and 30 November, 2023; B adjusted odds ratios, fitted via logistic regression models, for documentation of 1, 2, or ≥3 prior SARS-CoV-2 infections (relative to zero documented prior SARS-CoV-2 infections) among all outpatient cases diagnosed in the indicated periods relative to those diagnosed between 1 and 30 November, 2023; and C adjusted hazard ratios, fitted via Cox proportional hazards models, for progression to emergency department (ED) presentation or hospital admission, due to any cause or in association with acute respiratory infection (ARI) diagnoses, comparing outpatient cases diagnosed in the indicated periods to those diagnosed between 1 and 30 November, 2023. All models adjust for age, sex, race/ethnicity, body mass index, history of cigarette smoking, prior-year healthcare utilization across all settings, Charlson comorbidity index, and median household income within cases’ census tract according to the categorization scheme indicated in Table 1. In addition, Cox proportional hazards models adjust for nirmatrelvir-ritonavir receipt as a time-varying exposure. Missing values were addressed via multiple imputation, with results pooled across five pseudo-dataset replicates. Tests for non-zero slopes in the Schoenfeld residuals of Cox proportional hazards models identified no violations of the proportional hazards assumption (two-sided p > 0.1 for all fitted models). Analyses include data from 46,067 eligible individuals throughout the study period. For all panels (a–c), points indicate maximum likelihood estimates, with surrounding lines delineating 95% confidence intervals; we generated estimates via Cox proportional hazards models (A, B) and conditional logistic regression models (c). Source data to replicate the figure are provided as a Source Data file (fig2_source.xlsx).