Data were extracted from the CHS database. CHS is the largest integrated payer-provider healthcare organization in Israel. The CHS database contains extensive medical histories of CHS’s 4.7 million members, including COVID-19 test results and outcomes. These data repositories have been previously described in detail.13,14
Study design and population
We conducted a cohort study using CHS data to estimate the real-world effectiveness of REGEN-COV in preventing severe COVID-19-related outcomes. Eligibility criteria included: A documented first positive SARS-CoV-2 polymerase-chain-reaction (PCR) test result; a determination of being high-risk for severe COVID-19 based on medical history and clinical characteristics; age 12 years or older; and at least one year of continuous CHS membership as of the infection date. We excluded patients who were known to be infected with the omicron variant (based on sequencing of viral samples or on the S-gene target failure (SGTF) technique; the prevalence of omicron during the study period was negligible). We also excluded patients with invalid outcome data and those who received a positive PCR result during a hospitalization for another condition.
To emulate a target trial, treated patients were individually matched with untreated patients. Treated patients were those with a first positive PCR test result between September 19, 2021 and December 8, 2021 who received REGEN-COV treatment; Untreated patients were those with a first positive PCR test result between July 1, 2021 and December 8, 2021 who did not receive REGEN-COV treatment. The recruitment period for the untreated patients was a few weeks longer than for the treated patients to increase the sample size of the untreated group and allow for 1:5 matching of treated to untreated individuals.
REGEN-COV was not administered to certain high-risk individuals for a range of possible reasons, including: The patient was diagnosed before REGEN-COV was being offered by the healthcare system, logistic complexity prevented distribution of the treatment to the patient’s home, or the patient refused to receive the treatment. The index date for the treated patients was defined as the date of REGEN-COV treatment. Untreated matched patients were given an index date based on the time from infection diagnosis to treatment of the matched treated patient: e.g., if the treated patient received REGEN-COV two days after their positive PCR test result, the index date for the matched untreated patient was set to 2 days after their positive PCR test result.
Outcomes and Follow-up
Three outcomes were examined: COVID-19 related hospitalization, severe COVID-19 illness, and death due to COVID-19. The treated and untreated patients were followed until the occurrence of the outcome or until 28-days from the index date.
Adjustment was performed in two phases. First, the treated and untreated patients were matched on an initial set of confounders. Then, further adjustment was performed with a regression model. Subjects were matched on: Age, population sector (Jewish, Arab, Ultra-Orthodox), sex, socioeconomic status (SES, based on place of residence and categorized into 20 levels), body mass index (BMI, as a continuous variable), immunosuppression status, pregnancy, and calendar week of first vaccination dose. Confounders that were adjusted for in the model included: Age, sex, population sector, SES (as above), number of flu vaccines received within the 5 years prior to COVID-19 diagnosis, BMI (as a categorical variable: underweight, normal and obese), smoking status, number of COVID-19 vaccination doses received, "recent full vaccination” status, and calendar week of first vaccination dose . We also adjusted for the presence of the chronic conditions described in Supplemental Table 1. We included again in the regression model some of the variables were matched for to better control possible residual confounding, as mentioned in the statistical analysis. All variables were extracted according to the most recently documented value before the positive testing date, as recorded in the patients’ medical records. Full variable definitions are presented in Supplemental Table 2.
Matched untreated individuals who experienced an outcome between their positive PCR test date and their assigned index date were excluded. Because the index date was only set after matching (based on the timing the matched treated counterpart was treated), this exclusion could only happen after matching.
Matching was performed at a ratio of 1:5 treated to untreated individuals, using an optimal matching scheme. The Mahalanobis distance metric was used for continuous variables, and exact matching was used for categorical variables.15,16 Optimal matching minimizes the overall pair-wise distances without dependency on the order of matching.
After the matching, Cox proportional hazards models were fit for each outcome, adjusting for the abovementioned potential confounders. We report one minus the hazard ratio (HR) with 95% confidence intervals as the measure of treatment effect. Adjustment was performed using both matching and Cox modeling for two reasons: First, some of the variables were continuous and the matching was not exact. Second, not all the treated subjects had the same number of controls, due to the exclusion of controls post-matching. Missing data are rare in CHS database for the variables used, thus we used complete case analysis.
A subgroup analysis by age group (<60 or ≥60 year old) was conducted as a secondary analysis.
This study was approved by the CHS Institutional Review Board.