Clinical Data for COVID-19 Mortality Prediction in Patients with Mental Disorders

A retrospective cohort of 439 hospitalized patients with mental disorders and laboratory-confirmed COVID-19 was collected from 10 hospitals in Henan Province, China. The data was used to develop and validate machine-learning models for predicting 31-day mortality, with results published by Yi Zhang in 2026. The dataset includes clinical and laboratory variables, with patients split into training (n=308) and test (n=131) cohorts.

Use Cases

• Training mortality prediction models based on routine clinical and laboratory variables.
• Analyzing feature importance for COVID-19 outcomes based on SHAP-identified predictors like concomitant hormone therapy.
• Benchmarking machine learning algorithms for imbalanced clinical data based on the described oversampling approach.
• Studying risk stratification in vulnerable populations based on the model-defined high and low-risk groups.

Strengths

• Includes data from 439 patients across 10 hospitals, suggesting a multicenter source.
• Model validation was performed on an independent test cohort of 131 patients.
• The best-performing model achieved an AUC of 0.911 in the test cohort.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data reflects geographic bias inherent to the regional cohort from Henan Province, China.

Provenance

Source

figshare, author Yi Zhang.

Collection Method

Retrospectively collected from hospital records.

Freshness

Last updated 2026-05-26 06:03:49; freshness should be verified.

Geography

Henan Province, China.