This analysis is based on the eVusheld Assessment Real World Effectiveness (VALOR-C19) observational, retrospective cohort study, which assessed the effectiveness of AZD7442 PrEP in reducing COVID-19 hospitalizations among AZD7442-eligible immunocompromised populations in Israel (ClinicalTrials.gov: NCT05712096) [24]. We used real-world data from the Clalit Health Services (CHS) electronic health record (EHR) database, the largest of Israel's four integrated payer/provider healthcare organizations, encompassing more than 4.7 million members (52%) of the Israeli population. CHS contains clinical and administrative inpatient and outpatient data, including COVID-19-specific data curated at the national level, polymerase chain reaction test results, and COVID-19–related admissions and deaths. COVID-19 data are collected by the IMoH from multiple sources and shared daily with healthcare providers, including Clalit. These data repositories have been previously described in detail [2, 25].

This study used de-identified patient data and was approved by the CHS institutional review board.

Individuals were eligible for inclusion if they were aged ≥ 12 years, weighed ≥ 40 kg during 2 years prior to the eligibility date, and met the IMoH criteria for AZD7442 PrEP during the study period (from 15 February to 11 December 2022). The study start date reflects the approval date of AZD7442 in Israel. The study end date was selected to restrict analysis to the year 2022 due to the study being divided into 4-week periods. Eligibility was defined as having at least one comorbidity or treatment causing moderate/severe immunosuppression [e.g., chimeric antigen receptor T-cell therapy, solid organ transplant, autologous or allogenic bone marrow transplant (within the previous year if allogenic), hypogammaglobulinemia, active lymphoma, active multiple myeloma, diagnosis of chronic lymphocytic leukemia (CLL) or receiving treatment for CLL, and B-cell depleting therapies]. The full list of disease definitions for immunocompromised individuals is included in Supplementary Table S1. A separate analysis of this study has previously characterized AZD7442 uptake among immunocompromised individuals eligible for AZD7442 administration as PrEP as per IMoH recommendations [24].

The primary outcome of interest was COVID-19 hospitalization, defined as any hospitalization reported by the IMoH due to SARS-CoV-2 infection, at 6 months of follow-up. An exploratory analysis also assessed COVID-19 hospitalization at 3 months of follow-up.

Individuals who received AZD7442 300 mg as PrEP against COVID-19 were considered AZD7442-exposed, and individuals who did not receive AZD7442 were considered unexposed.

We divided the study into 4-week calendar time periods. Within each of these 4-week intervals, newly AZD7442 exposed individuals were contemporaneously matched (1:1 ratio) to eligible AZD7442 unexposed individuals.

To minimize the impact of confounding by indication, matching was performed based on propensity scores (PS) estimated using the extreme gradient boosting model (XGBoost), and the matching was set up prospectively on retrospective data. The following covariates were included, based on relevant baseline characteristics: age, sex, district, peripheral rank, smoking status, chronic kidney failure, chronic obstructive pulmonary disease, asthma, other respiratory diseases, heart disease, cerebrovascular disease, body mass index, hypertension, type 1 diabetes, type 2 diabetes, neurological disease, liver disease, thalassemia, hematological cancer, organ transplant, bone marrow transplant, primary immunodeficiency, secondary immunodeficiency, number of physician visits in the last year, number of hospitalizations in the last year, and number of emergency department visits in the last year (the full definitions of these covariates are presented in Supplementary Table S2). Nearest-neighbor matching was performed using a caliper width between participants of 0.2 times the pooled standard deviation of the logit of the PS. After matching, a covariate was considered sufficiently balanced if the standardized mean difference (SMD) between the AZD7442-exposed group and the matched unexposed group was < 0.1. Data from 221 AZD7442-exposed individuals were utilized to construct the first PS model for the rolling cohort and consequently were excluded from the final analysis.

The index date for each PS-matched pair was the date of AZD7442 PrEP administration of the exposed unit. Each PS-matched pair was followed up from the index date until the earliest of COVID-19 hospitalization or censoring (death, 3 or 6 months from the index date, end-of-study period, or initiation of a subsequent dose of AZD7442 in the exposed or unexposed unit). For the primary analysis, we included a maximum follow-up time of up to 6 months from the index date. Furthermore, we conducted an additional analysis including follow-up of up to 3 months from the index date. Calendar time Cox proportional hazards regression models were used with adjustment for post-matched unbalanced covariates to estimate hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) for the association between AZD7442 exposure and COVID-19 hospitalization; effectiveness was estimated as 1 − HR. Two sets of HRs with 95% CI and the corresponding effectiveness measures were reported: (1) from the PS-matched population with no additional regression adjustment; and (2) from the PS-matched population with additional regression adjustment for unbalanced variables (SMD > 0.1).

In a pre-specified, exploratory analysis, we plotted non-parametric smoothed estimates of the instantaneous hazards for the AZD7442-exposed group and the PS-matched unexposed group to further determine the duration of protection conferred by AZD7442.

AZD7442 effectiveness against COVID-19 hospitalization was also estimated during pre-defined SARS-CoV-2 variant periods, by subdividing the study period into smaller calendar time units based on the dominant circulating SARS-CoV-2 variant at the time: BA.2 (from 13 March to 20 June 2022), BA.5 (from 21 June to 7 November 2022), BQ.1 (from 8 November to 31 December 2022), and BA.2 and BA.5 combined (from 13 March to 7 November 2022). Follow-up time within SARS-CoV-2 variant periods was calculated for each PS-matched pair to only include the person-time at risk that overlapped with that dominant-variant period. COVID-19 hospitalizations that occurred outside the dominant-variant period were censored.