Predicting Hospital Readmission for Schizophrenic Patients
Author Information
Author(s): Góngora Alonso Susel, Herrera Montano Isabel, De la Torre Díez Isabel, Franco-Martín Manuel, Amoon Mohammed, Román-Gallego Jesús-Angel, Pérez-Delgado María-Luisa
Primary Institution: University of Valladolid
Hypothesis
Can predictive models using particle swarm optimization and machine learning algorithms effectively assess the readmission risk of patients with schizophrenia?
Conclusion
The study developed predictive models that significantly improve the ability to assess the readmission risk of schizophrenia patients, leading to better patient care and reduced healthcare costs.
Supporting Evidence
- The Random Forest algorithm combined with PSO achieved an AUC of 0.860.
- The recall rate for the best model was 0.959.
- The accuracy of the optimized model was 0.844.
- The study included 6089 readmission records collected from 11 public hospitals.
- Significant gender differences were observed, with 67.21% of patients being male.
- Patients aged 31 to 50 years represented the largest age group.
- Readmission rates were significantly associated with various comorbidities and social factors.
Takeaway
This study created smart computer models to help doctors predict which patients with schizophrenia might need to go back to the hospital, so they can help them better and save money.
Methodology
The study used machine learning algorithms, specifically Random Forest, Support Vector Machine, and Multi-Layer Perceptron, optimized with particle swarm optimization to analyze hospital readmission data.
Potential Biases
The dataset was limited to a specific region, which may not generalize to other populations.
Limitations
The study relied solely on administrative data, lacking detailed clinical information that could enhance predictive accuracy.
Participant Demographics
The sample included 3065 patients with schizophrenia, predominantly male (67.21%) and aged between 31 to 50 years.
Statistical Information
P-Value
0.005
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website