Comments on Paper 1
The answer of comment 1 : We think the high initial HRs are in part driven by increased healthcare contact. As noted in the
discussion: "First, electronic health records are routinely collected data for health care provision and so
only capture conditions diagnosed and recorded by the health care professional rather than true incidence
in the population. Unvaccinated people may have been less likely to contact health services and to test for
SARS-CoV-2 infection, leading to underestimated effects. People with recorded COVID-19, particularly
COVID-19 with hospitalization, may be more likely to have mental illnesses recorded due to greater
contact with health services. This may underpin the particularly high HRs observed initially, especially in
those hospitalized, and the rapid fall as service contact is likely highest early after diagnosis. However, this
is unlikely to fully explain adverse effects, given the persistent elevation of incidence of mental illnesses
following COVID-19 with hospitalization and the variation across mental illnesses."
The answer of comment 2 : We cannot fully mitigate unmeasured confounding as electronic
health record data are not intended for research, so do not
capture everything we would want to adjust for (let alone things
we might not have thought of!). I think the best approach to this
and all research questions is triangulation, where we use different
methods with different biases to gather evidence. I would
recommend this paper on triangulation: https://doi.org/10.1093/ije/
dyw314
discussion: "First, electronic health records are routinely collected data for health care provision and so
only capture conditions diagnosed and recorded by the health care professional rather than true incidence
in the population. Unvaccinated people may have been less likely to contact health services and to test for
SARS-CoV-2 infection, leading to underestimated effects. People with recorded COVID-19, particularly
COVID-19 with hospitalization, may be more likely to have mental illnesses recorded due to greater
contact with health services. This may underpin the particularly high HRs observed initially, especially in
those hospitalized, and the rapid fall as service contact is likely highest early after diagnosis. However, this
is unlikely to fully explain adverse effects, given the persistent elevation of incidence of mental illnesses
following COVID-19 with hospitalization and the variation across mental illnesses."
The answer of comment 2 : We cannot fully mitigate unmeasured confounding as electronic
health record data are not intended for research, so do not
capture everything we would want to adjust for (let alone things
we might not have thought of!). I think the best approach to this
and all research questions is triangulation, where we use different
methods with different biases to gather evidence. I would
recommend this paper on triangulation: https://doi.org/10.1093/ije/
dyw314