Paper Breakdown: Imperial Omicron Hospitalization Analysis
Read a paper with me
Hello from a biostatistician. Trying out some new formats, but here I take notes from my reading of Imperial’s Report 50, primarily focused on the structure of the statistical analysis.
Summary of “Report 50 - Hospitalisation risk for Omicron cases in England”
Scientific goal: Assess differences in the risk of hospitalization for infections by: (1) variant type, (2) whether it is a re-infection, and (3) vaccination type.
Linked data sources:
Infection data = PCR-confirmed SARS-CoV-2 cases captured by routine community testing (“Pillar 2”)
Variant data = S-gene target failure (SGTF), yes/no
Vaccination data = National Immunisation Management System (NIMS)
Hospitalization data = NHS Emergency Care and Secondary Use Services
Reinfection data = list of known reinfections
Study location: England
Study time period: PCR-confirmed cases tested between December 1-14, 2022
Note: A crude analysis of severity would calculate the hospitalization rate among Delta infections and compare this to the hospitalization rate among Omicron infections. This would be misleading for several reasons:
The age distribution of Delta and Omicron cases are different. Omicron cases are on average younger.
Omicron cases have had less follow-up time overall, including less time for progression to severe disease.
Analysis 1 - Hospitalization risk by variant type and previous infection
Outcome: Hospitalization (any record of attendance at a hospital in the 14 days after the last positive PCR test)
Predictors: Variant (Omicron vs. Delta), previously infected vs. not previously infected
Stratifying Variables: vaccination status, 10-year age band, sex, ethnicity, NHS region, specimen date*
* By stratifying on specimen date, it ensures that comparisons are between people with the same amount of follow-up time.
Model: Conditional Poisson regression model (looking within strata, how do numbers of hospitalizations compare across the predictors?)
Results: See Table 2. Rates of re-infection are higher for Omicron versus Delta cases. Without adjustment for misclassification of re-infections, the hazard of hospitalization among Omicron infections is 0.75x (or 25% lower) than the hazard of hospitalization among Delta infections. The hazard for hospitalizations among reinfections is 0.50x (or 50% lower) than the hazard of hospitalization among first infections. These effects are averaged across all strata (and thus reflects the fact that vaccination levels are high within England, and so may not be directly generalizable).
After adjusting for missed first infections (assuming only 33% are captured by surveillance), these hazard ratios are 0.89x and 0.43x for Omicron versus delta and for reinfections versus first infections, respectively. Additional analyses reported in the table consider different definitions of case and different definitions of hospitalization.
Analysis 2 - Hospitalization risk by vaccination status
Outcome: Hospitalization
Predictors: Groups defined by the combination of variant (Omicron vs. Delta) and vaccination type, all compared against a reference group of an unvaccinated Delta infection in a previously uninfected person (“primary infection”).
Stratifying Variables: 10-year age band, sex, ethnicity, NHS region, specimen date
Model: Conditional Poisson regression model
Results: See Table 3. Hazard ratios estimate the combined impact of variant and vaccination type, all compared to the same reference group of unvaccinated Delta primary infection. The number of groups and the narrow reference group make the estimates more difficult to interpret. The numbers are also small in some groups.
Picking out one interesting point of discussion:
“Hazard ratios for hospital attendance with Omicron for those who received [Pfizer/Moderna] as their primary vaccination schedule are similar to those seen for Delta in those vaccination categories, while Omicron hazard ratios are generally lower than for Delta for those who received AZ as their primary vaccination.” …. “We would also emphasise that these are estimates which condition upon infection; overall vaccine effectiveness against hospital attendance may not vary between the vaccines, given that [Pfizer/Moderna] maintain higher effectiveness against symptomatic infection with Omicron than AZ.”
Key messages:
Importance of adjusting for prior immunity
“Overall, we find evidence of a reduction in the risk of hospitalisation for Omicron relative to Delta infections, averaging over all cases in the study period.”
“These reductions must be balanced against the larger risk of infection with Omicron, due to the reduction in protection provided by both vaccination and natural infection.”
I was disappointed that the authors failed to report results stratified by age. These findings are not generalizable. To apply these results to another geography, results need to be stratified by both vaccination status and age. The study does not report the latter.