Natural language processing of multi-hospital electronic health records for public health surveillance of suicidality

There is an urgent need to monitor population mental health during crises such as COVID-19 to identify at-risk subgroups and target prevention. The authors developed and validated surveillance indicators related to suicidality: monthly hospitalisations due to suicide attempts (SA) and the prevalence of five risk factors among them. Indicators were computed automatically by applying NLP algorithms to EHRs from 15 university hospitals in the Paris area. In a retrospective cohort of 2,911,920 records, an interrupted time-series analysis compared pre- (Aug 1, 2017–Feb 29, 2020) and post-outbreak (Mar 1, 2020–Jun 31, 2022) periods. The slope of monthly SA increased after the outbreak by 3.7 (95% CI 2.1–5.3), mainly driven by girls aged 8–17 (trend variation 1.8, 95% CI 1.2–2.5). The study shows multi-hospital clinical text can be jointly analysed to provide timely mental health indicators, highlighting the need to better address violence against women, especially in early ages, in the aftermath of COVID-19.