A total of 412 potentially eligible studies were identified: 267 from Embase and 145 from Scopus (Fig. 1). After excluding 69 duplicates, 343 documents underwent an initial title and abstract screen, of which 287 were excluded. The most common reasons for exclusion were preclinical study (n = 83), review article (52), RCT (30), drug other than MOV (22), and safety study (21). The remaining 56 studies, plus an additional three records identified via supplementary literature surveillance, underwent full-text review. Fifty records were excluded, with the most common reason being non-population-based study (17), active comparator only (14), and no eligible outcomes (8).

Flow chart of systematic literature review to identify, screen, and select eligible real-world studies evaluating the effectiveness of molnupiravir. PICOTS population, intervention, comparator, outcome, time, and study design. A Additional documents were identified via ongoing literature surveillance

Nine studies (six peer-reviewed studies and three pre-prints) met all PICOTS criteria and were included in the SLR (Table 2) [29,30,31,32,33,34,35,36,37]. Two of the three pre-prints have subsequently been published as peer-reviewed articles [38, 39]. The peer-reviewed version of Paraskevis et al. has a different title and abstract compared to the pre-print version, but the data appear to be the same [33, 39]. As the study period and results in the peer-reviewed version of Bajema et al. have been updated compared to the pre-print version, we revised the analysis to incorporate the published version of Bajema et al. rather than the pre-print [30, 38].

One of the nine included studies was conducted in Greece [33], three in Hong Kong [34, 35, 37], two in Israel [29, 32], one in the UK [31], and two in the US [30, 36, 38]. Some studies from the same country used the same data source (electronic health records from the Hong Kong Hospital Authority in three studies [34, 35, 37], the Clalit Health System in two Israeli studies [combined with another data source in one of these studies] [29, 32], and the US Veterans Health Administration COVID-19 Shared Data Resource in two studies [combined with other data sources in one of these studies]) [30, 36, 38]. However, studies using the same data source had different study periods and methodologies. The study periods were of different lengths, but all began between December 16, 2021 and February 26, 2022 and ended between February 28 and October 20, 2022; Omicron variants of SARS-CoV-2 were thus dominant during all study periods and in all study locations [29,30,31,32,33,34,35,36,37,38, 40].

The size of the MOV-treated study population ranged from 359 to 7818, and all study populations had age-related and/or other risk factors for progression to severe COVID-19 [29,30,31,32,33,34,35,36,37,38]. Prior immunity to SARS-CoV-2 (defined in various ways based on vaccination and/or previous infection) was assessed directly in seven studies [29,30,31,32,33, 35, 36, 38] and inferred from age- and sex-stratified population-level vaccination data in one study [37]. The proportion of the MOV-treated group with prior SARS-CoV-2 immunity ranged from 16.1% in a Hong Kong-based study [35] to 98.2% in the UK study [31]. The most common MOV treatment initiation window was ≤ 5 days after a positive test (five studies) [29, 32, 35,36,37]; a 3-day window was used in one study [33], a 7-day window in two studies [31, 34], and a 10-day window in one study [30, 38]. All studies used cohort designs with longitudinal data, patient-level follow-up, and robust statistical methods (e.g., multivariate regression or propensity score-based approaches) to address potential confounding [29,30,31,32,33,34,35,36,37,38].

Two studies had a serious overall risk of bias, assigned in both cases due to an assessment of serious risk of bias in the confounding domain (Table 3) [31, 33]. In the case of Paraskevis et al., the serious risk of confounding was assigned because the authors were unable to obtain information on comorbidities among the untreated control group [33]. The authors attempted to address this limitation by excluding individuals < 65 years of age and by matching treated individuals to controls by age, since the number of comorbidities generally increases with age [33]. Nevertheless, comorbidities are an important risk factor for severe COVID-19 outcomes [7], and thus the study was assigned a serious risk of bias in the confounding domain. In the other study with a serious risk of confounding bias, Evans et al. did not use matching or propensity score adjustment for comparisons between the MOV-treated and control groups [31]. Although no statistical tests of differences between the treated and control groups were reported, the authors noted differences between the two groups that could potentially bias the study results in favor of MOV: the treated group were on average younger than the control group (mean age 53 versus 57 years), had fewer comorbidities (e.g., 74.6 vs. 62.8% had a Charlson comorbidity index of 0–10), and had a higher degree of prior immunity (36.3 vs. 17.6% had received ≥ 4 doses of a SARS-CoV-2 vaccine) [31]. The study reported univariate association analyses showing that younger individuals (i.e., those < 60 years of age) and those who had received ≥ 4 vaccine doses were more likely to avoid hospital admission or death within 28 days than were older or less vaccinated individuals, respectively [31]. Both studies had a low or moderate risk of bias in all other domains.

The remaining seven studies were determined to have a moderate overall risk of bias. The domains in which these seven studies were found to have a moderate, rather than a low, risk of bias were confounding (all seven studies; moderate is the lowest possible risk of confounding bias for a non-randomized study [28]), deviations from intended intervention (five studies [30, 34,35,36,37,38]), outcomes measurement (four studies [29, 32, 35, 36]), and selection bias (three studies [34, 35, 37]).

Seven studies were appropriately designed to accurately classify all time at risk of the outcome for both the MOV-treated and the control groups, and thereby mitigated the risk of immortal time bias. However, two of the Hong Kong-based studies did not account for immortal time bias due to differences in index dates and follow-up periods between the MOV-treated and control groups [