Patients
We identified 504912 COVID-19 diagnoses in outpatient settings, eventually resulting in 416030 eligible first COVID-19 episodes in adult outpatients (Fig 1B). Of those: (i) in Study 1, 1016 patients met the criteria for Cohort A, 95984 met the criteria for Cohort B and 275804 met the criteria for Cohort C (Fig 1B); (ii) in Study 2, 944 patients met the criteria for Cohort Fluvoxamine and 1796 met the criteria for Cohort Paroxetine (Fig 1B). Raw data (berfore matching) across the studies/cohorts indicated (Table 3): (i) < 1% of the patients in each cohort were fully vaccinated and 95% received no vaccination whatsoever; (ii) in Study 1, Cohort C patients (free of psychiatric difficulties/treatments) appeared younger and had clearly lower prevalence of all comorbidities than Cohort A (psychiatric difficulties, prescribed fluvoxamine) and Cohort B (psychiatric difficulties, not prescribed fluvoxamine) patients, who were closely similar in all aspects (Table 3); (ii) in Study 2, patients in the two cohorts were closely similar in respect to all pre-COVID-19 characteristics, and were in this respect also similar to Cohort A and Cohort B in Study 1 (Table 3).
In Study 1, raw incidence of all outcomes was closely similar in Cohort A and Cohort B - 3.35% vs. 3.25%, 12.5% vs. 11.5% and 3.74% vs. 4.44% for COVID-related hospitalizations, 30-day all-cause hospitalizations and COVID-related mortality, respectively (Table 3) - and was clearly higher than in Cohort C patients (0.94%, 5.18%, 1.05%, respectively) (Table 3). In Study 2, raw incidence of all outcomes was closely similar in the two cohorts and also similar to Cohorts A and B in Study 1 (Table 3).
Analysis in matched sets: Study 1
Eventually, 749 Cohort A patients were matched to 31336 Cohort B patients; and 866 of the former and 82323 of the latter were matched to 222792 and to 268778 Cohort C patients, respectively (Online Resource 2 – Supplemental results – provides all pairwise data before and after matching).
Incidence of all outcomes was lower in all matched sets than before matching, and there was no indication that outpatients prescribed fluvoxmine around the time of COVID-19 diagnosis (Cohort A) were at a reduced risk of any outcome as compared to their peers burdened with similar psychiatric difficulties but not prescribed fluvoxamine (Cohort B), or as compared to COVID-19 outpatients free of psychiatric difficulties and related treatments (Cohort C) (Fig 2): all relative risk estimates were around 1.0 or somewhat higher than 1.0. Comparisons between matched cohorts B and C yielded more precise estimates (higher nubmer of subjects) (Fig 2), but in terms of the point-estimates, A vs. C and B vs. C differences were closely similar (Fig 2).
In the network analysis, direct and indirect Cohort A to Cohort B comparisons were consistent (Fig 3): there was no indication that Cohort A patients were at a reduced risk of any of the outcomes compared to their Cohort B peers (Fig 3).
Analysis in matched sets: Study 2
Eventually, 344 Cohort Fluvoxamine patients were exactly matched to 535 Cohort Paroxetine patients (Online resource 2 – Supplemental Results - lists all covariates before and after matching). Incidence of all outcomes was lower than before matching (Fig 4). Incidence of “all-cause 30-day hospitalization” (Fig 4) was resonable (weighted event counts 21.9 vs. 28.8). There was no signal that outpatients prescribed fluvoxmine (but not paroxetine) around the time of COVID-19 diagnosis were at a reduced risk of this outcome as compared to their peers prescribed paroxetine (and not fluvoxamine) (Fig 4): all RR estimates were around 1.0. Incidence of COVID-19-related hospitalization and of COVID-19-related mortality was vary low (<1.0%) (Fig 4), hence estimates were imprecise – however, point-estimates were closely similar to the estimates for Cohort A vs. Cohort B in Study 1 not indicating any benefit of fluvoxamine (Fig 2).
Sensitivity to unmeasured confounding
Estimates corrected for (hypothetical) bias arising from a 1:40 imbalance in co-treatment with an effective (30% risk reduction) non-fluvoxamine therapy between fluvoxamine-exposed and control subjects (Cohort B in Study 1, Cohort Paroxetine in Study 2) did not relevantly differ from the acutally observed estimates (Table 4).