Differential follow-up patterns in COVID-19 and comparison cohorts – Authors' reply

We thank Rebecca Fuhrer and James Hanley for their comments. They are correct that the timing of COVID-19 and the control events differ (as shown in the appendix of the Article),¹ and that, in turn, the duration of available follow-up differs between cohorts. They are also right that, as a result, there is a possible risk that the matching between cohorts might be partly lost at follow-up. This loss of matching would occur, for instance, if patients who were diagnosed with COVID-19 in July, 2020, (ie, with less opportunity for follow-up) were systematically different in their baseline characteristics from patients who were diagnosed in March, 2020 (ie, with more opportunity for follow-up). Addressing this issue was the purpose of our sensitivity analysis, wherein cohorts were limited to people with a subsequent health visit within the 14 to 90 day period after diagnosis (as shown in the appendix of the Article).¹

With the benefits of the longer follow-up that is now available compared with when we did the study, we ran an additional sensitivity analysis in which the same cohorts of patients (ie, with COVID-19 and influenza) as in our primary analysis were followed up until Feb 12, 2021. This analysis allowed us to distinguish people who were potentially lost to follow up in the 3 months after their diagnosis from people who had not yet made contact with a health-care organisation from the network. In this additional analysis, there was substantially less difference in the follow-up rates between cohorts, particularly at 15 days, where the number of people who were at risk of psychiatric illness differed only by 0·2% (ie, 62·2% of the COVID-19 cohort vs 62·0% of the influenza cohort). In this analysis, at 90 days, the relative risk of having had a psychiatric illness in the COVID-19 versus influenza cohorts was 1·86 (95% CI 1·65–2·06), which is similar to that in the primary analysis. This similarity provides further evidence that differences in loss of patients to follow-up do not account for the observed differences in rates of psychiatric diagnoses.

Regarding the exclusion of patients from the analysis who had died after the index event, we included these patients as part of a sensitivity analysis in a follow-up study.² Similar findings were obtained when patients who had died were included.