COVID-19-Associated Hospitalizations Among Vaccinated and Unvaccinated Adults 18 Years or Older in 13 US States, January 2021 to April 2022

Key Points Question How do COVID-19–associated hospitalization rates compare among adults who are unvaccinated and vaccinated, and what are the risk factors for hospitalization for COVID-19 among vaccinated persons? Findings In this cross-sectional study of US adults hospitalized with COVID-19 during January 2022 to April 2022 (during Omicron variant predominance), COVID-19-associated hospitalization rates were 10.5 times higher in unvaccinated persons and 2.5 times higher in vaccinated persons with no booster dose, respectively, compared with those who had received a booster dose. Compared with unvaccinated hospitalized persons, vaccinated hospitalized persons were more likely to be older and have more underlying medical conditions. Meaning The study results suggest that COVID-19 vaccines are strongly associated with prevention of serious COVID-19 illness.

6. Second raking: A second round of raking was applied to adjust the trimmed weights to sum to the population totals, using the same dimensions above. The final sample weight was obtained after iterating these dimensions to convergence.

Vaccination definitions, weighting of cases with known vaccination status
Vaccinated cases were defined as having received a second dose of a 2-dose series or one dose of a single-dose series ≥14 days before a positive SARS-CoV-2 test, regardless of booster dose status. If the SARS-CoV-2 test date was not available, hospital admission date was used. Patients who had received at least one dose of vaccine but had not completed a primary vaccination series were excluded from this analysis.
Vaccination status for hospitalized cases and vaccine coverage for the underlying catchment area were determined by state immunization information system (IIS) data, as previously described. 3   vaccination status (doses, dates administered, and product) was determined from state IISs for all sampled COVID-NET cases. In addition to the minimum data elements required for each case, some sites opted to collect vaccine information on all cases. With this additional information, non-sampled cases are able to be included in analyses regarding vaccination data. If a site did not opt to collect vaccine © 2022 Havers FP et al. JAMA Internal Medicine.
information on non-sampled cases, their original sample weight is applied and only sampled cases are included in analyses. The inclusion of non-sampled cases allows COVID-NET to retain a representative sample while allowing for much more precise estimates regarding vaccine data.

Propensity Score Analysis
In the analysis comparing outcomes between vaccinated and unvaccinated hospitalized patients, the main analysis was a multivariate logistic regression model that included the entire eligible sample, and adjusted for the following covariates: race/ethnicity; age category; sex; site; long-term care facility residence; and underlying medical conditions. However, to check the robustness of our results, we also accounted for potential confounding using a second method using a propensity score matched cohort. 3,4,5 The propensity score calculated for each patient estimated the probability of vaccination use based on baseline covariates, regardless of actual vaccination status. The propensity score for each individual was calculated using a multivariable logistic regression model; the dependent variable was vaccination status. The propensity score model used included the following covariates: race/ethnicity; age category; sex; site; long-term care facility residence; and underlying medical conditions, including obesity, diabetes, chronic lung disease, cardiovascular disease, neurologic disease, renal disease, immunosuppressive conditions, liver or gastrointestinal tract disease, blood disorder, and rheumatologic or autoimmune diseases.
Using the propensity score, we matched vaccinated cases with unvaccinated cases who had similar propensity scores using a 1:1 greedy matching algorithm. 3,4,5 The propensity-matched cohort selected vaccinated and unvaccinated cases who had similar baseline characteristics except for vaccination status (Supplementary Table 4). Using these propensity-matched cohorts, models for ICU status and in-hospital death used logistic regression. To account for residual confounding, the logistic regression model adjusted for the same variables used in the model that generated the propensity score. Data on race and ethnicity were categorized as follows: non-Hispanic White (White), non-Hispanic Black (Black), non-Hispanic Asian or Pacific Islander (Asian/Pacific Islander), non-Hispanic American Indian or Alaska Native (American Indian/Alaska Native) and Other/Unknown. If ethnicity was unknown (8% of cases), non-Hispanic ethnicity was assumed. d Includes multiple race (71, 0.7%) and unknown race (462, 4%).  for persons with a weakened immune system) cannot be distinguished from a booster dose. This is a relevant consideration because vaccines can be less effective in persons with a weakened immune system. c Statistical significance for univariate analyses was determined as p<0.10 d Overall condition categories as defined in Supplementary Table 1. e Obesity is defined as calculated body mass index (BMI) ≥30 kg/m2, and if BMI is missing, by International Classification of Diseases discharge diagnosis codes. f Cardiovascular disease excludes hypertension. g Time from vaccination to admission is the number of days between the most last vaccine dose received plus 14 days and date of hospital admission.