JMIR Publications

ABSTRACT

Background:

The present study aimed to systematically review relevant literature on the predictive value of machine learning for platinum-based chemotherapy responses in ovarian cancer(OC) patients.

Objective:

This meta-analysis study aimed to systematically review relevant literature on the predictive value of machine learning for platinum response in ovarian cancer. Accurate prediction of platinum sensitivity is important to facilitate individualized treatment for ovarian cancer.The authors summarized previous studies that used machine learning methods, and concluded that machine learning can accurately predict the response of ovarian cancer. The review is extremely important and germane to the researchers and clinicians working in this field.

Methods:

Following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guideline, we systematically searched PubMed, Embase, Web of Science, and Cochrane for relevant studies on predictive models for platinum-based therapies for OC published before April 26, 2023. The Prediction Model Risk of Bias Assessment (PROBAST) was used to evaluate the risk of bias in the included articles. The C statistic, sensitivity, and specificity were used to evaluate the performance of prediction models to investigate the predictive value of machine learning for platinum chemotherapy responses in OC patients.

Results:

A total of 1749 articles were examined, and 19 of them involving 38 models were eligible. The most commonly used modeling methods are LR (n=18; 47%), XGBoost (n=4; 10%), and SVM (n=4; 10%). The training cohort reported C statistics in 38 predictive models, with a pooled value of 0.800; the validation cohort reported C statistics in 11 (28%) predictive models, with a pooled value of 0.827. SVM performed well in both the training and validation cohorts, with a C statistic of 0.942 and 0.877, respectively. The pooled sensitivity was 0.910, and the pooled specificity was 0.850.

Conclusions:

Machine learning can effectively predict how OC patients respond to platinum-based chemotherapy and may provide a reference for the development or updating of subsequent scoring systems.