Tempering optimism from repeated longitudinal mental health surveys – Authors' reply

We share the sentiment of Mark É Czeisler and colleagues that sources of bias must be carefully considered to determine whether they are the cause of changes when monitoring longitudinal trends in data. However, neither our methods nor other UK data from the same period suggest that our findings are due to non-response bias.

Our decision to restrict the analytical sample to participants with three repeated measurements had a technical basis, to meet the minimum timepoints needed for use of free time scores in latent growth modelling. The sample was weighted only after sample selection to remove the risk of attrition that would otherwise potentially lead to an analytical sample with improved mental health. The percentage of participants with diagnosed mental illness in the analytical sample (6679 [18·3%] of 36 520 participants) and excluded sample (724 [18·1%] of 4000 participants) was similar. Retention rates week-on-week in the study were high in the analytical sample (appendix). Although baseline Patient Health Questionnaire (PHQ) and Generalised Anxiety Disorder assessment (GAD) scores were both negatively correlated with the number of weeks observed (appendix of the Article), on further analysis the correlations were small (PHQ: Pearson's r=–0·16, GAD: r=–0·12). Additionally, our models applied full information maximum likelihood estimation, which uses the observed values to supplement loss of information due to missing data and has been shown to yield unbiased estimates of both parameters and their standard errors in simulation studies.¹

Comparing our results to studies published after ours, including those of both longitudinal and repeated cross-sectional samples, the same pattern of improvements as shown in our data are found.2, 3 However, our study only covers the first 20 weeks of the pandemic in the UK following the start of lockdown in March 23, 2020. So we agree that the improvements seen in this time should not be taken as conclusions that individuals showed recovery in mental health after the shock of the pandemic. Further data from the UCL COVID-19 Social Study and other studies suggest that mental health worsened again in England (and across Great Britian) in the autumn of 2020 as virus prevalence increased and restrictions were once again tightened.³ As the pandemic continues, monitoring of changes in mental health should be maintained to support planning and resources for mental health services.