Impact of COVID-19 Pandemic on Suicide Attempts in Paris

Study Methodology

To understand the impact of the COVID-19 pandemic on suicide attempts (SAs), a comprehensive study was conducted in Paris. This study adhered to the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) reporting guideline and received ethical approval from the Greater Paris University Hospitals’ institutional review board (IRB00011591).

Primary Measure: Suicide Attempts

The primary measure used in this study was the monthly number of hospitalizations due to SAs. This measure is a widely recognized indicator of mental health within a population.

Study Design and Participants

A retrospective cohort study was conducted involving all hospitalizations due to SAs in Paris’ Assistance Publique-Hôpitaux de Paris (AP-HP) university hospitals between August 2017 and June 2022. Only adult and pediatric hospitals with stable electronic health record (EHR) data collection were included (n = 15).

Data Sources

Data sources included administrative data, diagnoses from claim data, and clinical reports. The AP-HP clinical data warehouse combined data from EHR software and the claim database. Data was extracted on July 4th, 2022.

Natural Language Processing (NLP) for SA Detection

To accurately identify SA-caused hospitalizations, a Natural Language Processing (NLP) algorithm was employed. This algorithm utilized the RoBERTa neural network architecture and was trained on annotated data. Its performance was validated through a chart review by expert clinicians.

Variables Collected

For each SA-caused stay, the following variables were collected:

* Age at admission
* Sex
* Admission date
* Length of stay
* Death during stay
* Known SA risk factors
* Claim diagnostic codes
* Hospital location

Statistical Analysis

To assess the pandemic’s impact on SA-caused hospitalizations, a single-group interrupted time-series analysis was conducted, adjusting for mean, trend, and seasonality. Subgroup analyses were performed by sex and age. Sensitivity analyses evaluated alternative stay-classification algorithms, adjusted for missing data, and examined each hospital separately.

Study Methodology: Exploring Suicide Attempts Through Data Analysis

Reporting Guidelines and Ethical Considerations

Our study strictly adheres to the RECORD reporting guideline. Ethical approval was granted by Greater Paris University Hospitals (IRB00011591). Patient confidentiality is ensured through pseudonymization, and individuals who objected to data use were excluded.

Primary Measure: Suicide Attempts

We focus on the monthly count of hospitalizations due to suicide attempts (SA), a significant marker of population mental health.

Study Design, Setting, and Participants

We conducted a retrospective cohort study at 15 AP-HP university hospitals from 2017 to 2022. Our cohort includes all SA hospitalizations in adults and children.

Data Sources

Data is sourced from administrative data, claim diagnoses, and clinical reports. The AP-HP clinical data warehouse consolidates data from EHR software and the claim database.

Natural Language Processing (NLP) for SA Detection

NLP Algorithms for SA Detection

Screening Stage

We employ a dictionary of SA-related keywords in clinical reports.

Stay-Classification Stage

Keywords are extended and searched in discharge summaries. An NLP algorithm categorizes mentions as valid (SA-caused stay) or invalid.

Algorithm Development and Validation

Our NLP algorithm leverages the RoBERTa neural network architecture and is trained on annotated data. Expert clinicians assess its performance through chart reviews.

Variables Collected

For each SA-caused stay, we gather the following data:

• Age at admission
• Sex
• Admission date
• Length of stay
• Death during stay
• SA risk factors
• Claim diagnostic codes
• Hospital location

Statistical Analysis

Interrupted Time-Series Analysis

We assess the impact of the COVID-19 pandemic on SA-caused hospitalizations using a single-group interrupted time-series analysis. The model adjusts for mean, trend, and seasonality.

Subgroup and Sensitivity Analyses

Subgroup analyses are conducted by sex and age. We also perform sensitivity analyses with alternative stay-classification algorithms, adjust for missing data, and examine each hospital separately.

Conclusion

Our study methodology provides a comprehensive framework for analyzing suicide attempts using data from multiple sources. The application of NLP algorithms enhances the accuracy of SA detection. By examining data before, during, and after the COVID-19 pandemic, we aim to uncover the impact of this major event on population mental health. The results of our study will contribute to a deeper understanding of suicide prevention strategies and inform future interventions.