Disparities in Intensive Care Unit Admission and Mortality Among Patients With Schizophrenia and COVID-19

A National Cohort Study

Guillaume Fond; Vanessa Pauly; Marc Leone; Pierre-Michel Llorca; Veronica Orleans; Anderson Loundou; Christophe Lancon; Pascal Auquier; Karine Baumstarck; Laurent Boyer


Schizophr Bull. 2021;47(3):624-634. 

In This Article


Study Design and Data Sources

In this population-based cohort study, we used data from Programme de Médicalisation des Systèmes d'Information (PMSI database), the French national hospital database in which administrative and medical data are systematically collected for acute and psychiatric care. The PMSI database is based on diagnosis-related groups, with all diagnoses coded according to the 10th revision of the International Classification of Diseases (ICD-10) and procedural codes from the Classification Commune des Actes Médicaux (CCAM). In our study, we included all hospitalized patients between February 1, 2020, and June 9, 2020, aged 15 years or older with identified COVID-19 (ICD-10 codes: U07.10 or U07.12 or U07.14) and respiratory symptoms (ICD-10 = U07.10 or U07.11) and a length of hospital stay > 24 h (in order not to take into account pauci- or asymptomatic COVID-19 forms that did not actually require hospitalization) except if the patients died within 24 h. We excluded patients with a severe mental illness diagnosis other than SCZ: bipolar disorder or recurrent major depression (ICD-10 codes = F30* or F31* or F33*).

The PMSI database is used to determine financial resources and is frequently and thoroughly verified by both its producer and the paying party, with possible financial and legal consequences.[4] Data from the PMSI database are anonymized and can be reused for research purposes.[2,5] Due to its suitable accuracy and exhaustive data collection, no patients were lost to follow-up during the study period.


We defined 2 populations. Cases were patients who had a diagnosis of SCZ according to specific ICD-10 codes (ie, F20*, F22*, or F25*) in either the acute care or psychiatric PMSI database. Controls were patients who did not have a diagnosis of severe mental illness according to specific ICD-10 codes in the acute care PMSI database and who were not listed in the PMSI psychiatry database.

The primary outcome was in-hospital mortality. The secondary outcome was ICU admission. We gathered patients' sociodemographic data (age classes: <55, 55–65, 65–80, and >80 years; sex; social deprivation: favored/deprived[6]), clinical data at baseline (smoking status: yes/no; overweight and obesity: yes/no; Charlson Comorbidity Index score[7] and main comorbidities: yes/no), stay data (origin of patients: from home or from hospital-institution; length of ICU and hospital stay), management data (Simplified Acute Physiology Score II (SAPS II) for ICU stay; recourse to mechanical ventilation: yes/no; recourse to renal replacement therapy: yes/no), hospital data (hospital category: public, university, or private; number of hospital stays for COVID-19), and geographical areas of hospitalization (4 areas grouped according to pandemic exposure from the highest to the lowest: Ile-de-France, northeast, southeast, and west, data from Sante Publique France: supplementary figure S1).

Figure S1.

Geographical areas of hospitalization grouped according to pandemic exposure from the highest to the lowest: Ile-de-France, northeast, southeast and west; data from SantePublique France. Peak of the epidemic in France: 04/04/2020. https://www.gouvernement.fr/info-coronavirus/carte-et-donnees.

Statistical Analysis

Continuous variables are expressed as medians and interquartile ranges. Categorical variables are summarized as counts and percentages. No imputation was made for missing data.

The 2 outcomes were assessed with unadjusted (model 1) and multivariable (models 2 and 3) models. Univariable and multivariable generalized linear models with random effects and correlation matrices (to take into account the clustered effect of the hospitals) were used to estimate the association between SCZ and the 2 outcomes.

Model 2 incorporated sociodemographic data (ie, age, sex, social deprivation), clinical data at baseline (ie, smoking status, overweight and obesity, Charlson Comorbidity Index), stay data (ie, origin of the patient), hospital data (ie, hospital category, number of hospital stays for COVID-19), and geographical areas of hospitalization (ie, Ile-de-France, northeast, southeast, and west).

Model 3 incorporated model 2 plus 2 interaction terms, SCZ × age and SCZ × geographical areas of hospitalization, to check whether the association between SCZ and the 2 outcomes was homogenous across ages and geographical areas of hospitalization according to pandemic exposure. The 2 interactions were determined based on a previous work reporting the influence of age and overcrowding on the COVID-19 prognosis.[8] In addition to aggregate analysis, we conducted stratified analyses when an interaction was statistically significant.

A significance threshold of P < .05 was used. All analyses were performed in SAS (version 9.4).