Figure 1 . Data sources and curation (see text for details).A. Raw data was prepared by the Croatian Institute for Public Health from several databases that it maintains. COVID-19 patients were identified based on positive polymerase chain reaction (PCR) or rapid antigen testing (RAT) performed at dedicated public testing points or hospitals (ICD-10 code U07.1), or based on epidemiological/clinical criteria (ICD-10 code U07.2) and individual data were linked to databases on vaccination, deceased persons, hospitalizations and Central Heath Information System. B . Anonymized data were “tidied-up” by exclusion of patients first diagnosed by PCR testing when seeking hospital assistance regardless of the reason, so as to retain only outpatients; subjects younger than 16 years and those with missing/erroneous entries on key variables. Also, repeated COVID-19 episodes were excluded and cut-off date for index COVID-19 diagnosis was set at August 15, 2021. Based on International Classification of Disease version 10 (ICD-10) code entries and Anatomical Therapeutic Chemical (ATC) code entries patients were classified into subsets (Group A, Group B, Group C) in respect to issuance of prescriptions for fluvoxamine and underlying morbidity. Definitions of Group A (patients burdened with conditions requiring antidepressant/anxyolytic treatment and prescribed fluvoxamine around the time of COVID-19 diagnosis), Group B (patients patients burdened with conditions requiring antidepressant/anxyolytic treatment not prescribed with fluvoxamine) and Group C patients (patients free of psychiatric difficulties and of respective treatments) are detailed in Table 1.
Figure 2 . Raw incidence (percentage) of COVID-related hospitalization, all-cause 30-day hospitalization and of COVID-related mortality (composite) (see Patients and Methods for definitions) across the patient subsets: Group A (patients burdened with conditions requiring antidepressant/anxyolytic treatment and prescribed fluvoxamine around the time of COVID-19 diagnosis), Group B (patients burdened with similar psychiatric difficulties, but not prescribed with fluvoxamine) and Group C (patients free of psychiatric difficulties and of respective treatments).
Figure 3. Analysis of outcomes (see Patients and Methods for definitions) in matched sets of patients burdened with conditions requiring antidepressant/anxyolytic treatment and prescribed (Group A) or not prescribed (Group B) fluvoxamine, and those free of such difficulties and related treatments (Group C) around the time of COVID-19 diganosis. Depicted are proportions (percentages) of patients with outcomes in each matched set and respective relative risks (RR). Priors for Bayes estimates: skeptical is moderately informative normal prior centered at 0.0 for Ln(RR) with standard deviation 0.355 – gives equal (50%) probability to an RR above and an RR below unity with 95% probability for an RR between 0.5 and 2.0; optimistic is a moderately informative normal prior centered at -0.199 for Ln(RR) (i.e., 18% relative risk reduction) with standard deviation 0.40 – suggests a benefit but leaves 30% probability of an RR >1.0;pesimistic is a weakly informative normal prior centered at 0.199 for Ln(RR) (i.e., 22% relative risk increase) with standard deviation 0.77 – suggests harm (of the same extent as benefit suggested by the optimistic prior), but leaves 40% probability of an RR <1.0.
Figure 4 . Differences between Group A patients (burdened with conditions requiring antidepressant/anxyolytic treatment and prescribed fluvoxamine around the time of COVID-19 diagnosis) and Group B patients (suffer similar psychiatric difficulties, but are not prescribed with fluvoxamine), i.e., control patients, regarding the outcomes of interest (see Patients and Methods for definitions) generated in network meta-analysis (frequentist and Bayes) that included Group A vs. Group B, Group A vs. Group C and Group B vs. Group C comparisons in matched sets.A . Meta-analysis based on weighted proportions. Direct, indirect and total (combined, network) differences. B.Meta-analysis based on Ln (RR) generated in primary Bayesian analysis with moderately informative skeptical prior (shown in Figure 3). Only total (combined) effects are shown (as in A, direct and indirect effects were consistent).