Maintenance Notice

Due to necessary scheduled maintenance, the JMIR Publications website will be unavailable from Wednesday, July 01, 2020 at 8:00 PM to 10:00 PM EST. We apologize in advance for any inconvenience this may cause you.

Who will be affected?

Accepted for/Published in: Journal of Medical Internet Research

Date Submitted: Mar 2, 2023
Date Accepted: Sep 29, 2023

The final, peer-reviewed published version of this preprint can be found here:

The Price of Explainability in Machine Learning Models for 100-Day Readmission Prediction in Heart Failure: Retrospective, Comparative, Machine Learning Study

Soliman A, Agvall B, Etminani K, Hamed O, Lingman M

The Price of Explainability in Machine Learning Models for 100-Day Readmission Prediction in Heart Failure: Retrospective, Comparative, Machine Learning Study

J Med Internet Res 2023;25:e46934

DOI: 10.2196/46934

PMID: 37889530

PMCID: 10638630

The Price of Explainability in Machine Learning Models for 100-days Readmission Prediction in Heart Failure: Retrospective, Comparative, Machine Learning Study

  • Amira Soliman; 
  • Björn Agvall; 
  • Kobra Etminani; 
  • Omar Hamed; 
  • Markus Lingman

ABSTRACT

Background:

Sensitive and explainable machine learning models can assist clinicians in managing heart failure (HF) patients at discharge by highlighting individual aspects associated with high-risk of readmission. In this cohort study, we investigate the driving forces for potential applicability of classification models as decision support tools for readmission prediction of HF patients.

Objective:

Evaluating the trade-off between adopting deep and traditional machine learning models for identifying 100-days readmission risk of HF patients and explaining the predictions given by presenting important features globally among patient cohort and locally for an individual patient.

Methods:

The retrospective data was collected from the regional healthcare information platform (RHIP) in Region Halland, Sweden. The cohort includes HF patients >40 years old who were hospitalized at least once between 2017 and 2019. Data analysis was performed from January 1, 2017, to December 31, 2019. Two machine learning models were developed and validated for 100-days readmission prediction associated with explainability of taken decisions. Models were developed based on decision trees and recurrent neural architecture. Model explainability was extracted using machine learning explainer. The predictive performance of the models was compared against two risk assessment tools using multiple performance metrics.

Results:

A traditional and explainable model informed by clinical knowledge can perform on par with deep model and outperform conventional scores in predicting readmission among HF patients. Explainable models provide actionable insights that might improve care planning.

Conclusions:

A widely used deep prediction model did not outperform an explainable machine learning model when predicting readmissions among HF patients. Model transparency does not have to come at a price of lower performance, which could support clinical adoption.


 Citation

Please cite as:

Soliman A, Agvall B, Etminani K, Hamed O, Lingman M

The Price of Explainability in Machine Learning Models for 100-Day Readmission Prediction in Heart Failure: Retrospective, Comparative, Machine Learning Study

J Med Internet Res 2023;25:e46934

DOI: 10.2196/46934

PMID: 37889530

PMCID: 10638630

Download PDF


Request queued. Please wait while the file is being generated. It may take some time.

© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.