Risk prediction of QTc prolongation occurrence in cancer patients treated with commonly used oral tyrosine kinase inhibitors: machine learning modeling or conventional statistical analysis better?

Abstract Background Cancer patients receiving targeted therapies need to prevent QTc prolongation and life-threatening cardiovascular (CV) events to maintain a balanced benefit-risk ratio. This study aimed to develop an optimal prediction model for QTc prolongation risk and estimate its risk probabi...

Full description

Saved in:
Bibliographic Details
Main Authors: Hsiang-Wen Lin, Tien-Chao Lin, Chien-Ning Hsu, Tzu-Pei Yeh, Yu-Chieh Chen, Liang-Chih Liu, Chen-Yuan Lin
Format: Article
Language:English
Published: BMC 2025-08-01
Series:BMC Medical Informatics and Decision Making
Subjects:
Online Access:https://doi.org/10.1186/s12911-025-03091-8
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849234918750552064
author Hsiang-Wen Lin
Tien-Chao Lin
Chien-Ning Hsu
Tzu-Pei Yeh
Yu-Chieh Chen
Liang-Chih Liu
Chen-Yuan Lin
author_facet Hsiang-Wen Lin
Tien-Chao Lin
Chien-Ning Hsu
Tzu-Pei Yeh
Yu-Chieh Chen
Liang-Chih Liu
Chen-Yuan Lin
author_sort Hsiang-Wen Lin
collection DOAJ
description Abstract Background Cancer patients receiving targeted therapies need to prevent QTc prolongation and life-threatening cardiovascular (CV) events to maintain a balanced benefit-risk ratio. This study aimed to develop an optimal prediction model for QTc prolongation risk and estimate its risk probability in cancer patients treated with oral tyrosine kinase inhibitors (TKIs). Methods This retrospective cohort study analyzed electronic medical records (EMR) of cancer patients newly treated with commonly used oral TKIs at a medical center between January 2016 and December 2020. QTc prolongation was defined as ≥ 450 ms in males and ≥ 470 ms in females using Bazett’s formula. The study followed four key steps: (1) Managing missing data, (2) Identifying important variables, (3) Training and testing the best prediction models, (4). Estimating risk probability and determining cut-off points. Both univariate logistic regression (LR) and supervised machine learning (ML) approaches were used for variable selection. The backward LR method and seven ML algorithms were applied to train and test the prediction models. The best model was identified based on model performance, fitting criteria, area under the receiver operating characteristic curve (AUROC), risk probability cut-off points, and clinical relevance. Results The statistical 12-parameter model demonstrated excellent performance (AUROC = 0.89, sensitivity = 0.91, specificity = 0.75) and strong discrimination ability for risk probability prediction (AUROC = 0.78, cut-off = 0.46), outperforming other ML models. In the final best model: the baseline risk probability of QTc prolongation was 0.13, even in the absence of other contributing factors. Baseline QTc prolongation and a history of cardiovascular disease (excluding arrhythmia, cardiomyopathy, etc.) contributed the most to incremental risk probability (0.471 and 0.282, respectively), after controlling for other factors. The remaining 10 factors each contributed to an increased probability of QTc prolongation for more than 0.14 probability. Conclusions A logistic regression model utilizing 12 easily accessible variables from EMRs outperformed ML models in predicting the risk probability of QTc prolongation in cancer patients newly treated with five oral TKIs. These findings serve as a valuable clinical reference for integrating digital monitoring into cardiovascular care for cancer survivors undergoing targeted therapy with TKIs. They also underscore the importance of screening baseline ECG before initiating TKIs to assess the risk of QTc prolongation, facilitating early intervention and prevention in the future.
format Article
id doaj-art-96a92ef5e9dc48e1a16c6983e987a24c
institution Kabale University
issn 1472-6947
language English
publishDate 2025-08-01
publisher BMC
record_format Article
series BMC Medical Informatics and Decision Making
spelling doaj-art-96a92ef5e9dc48e1a16c6983e987a24c2025-08-20T04:02:56ZengBMCBMC Medical Informatics and Decision Making1472-69472025-08-0125111410.1186/s12911-025-03091-8Risk prediction of QTc prolongation occurrence in cancer patients treated with commonly used oral tyrosine kinase inhibitors: machine learning modeling or conventional statistical analysis better?Hsiang-Wen Lin0Tien-Chao Lin1Chien-Ning Hsu2Tzu-Pei Yeh3Yu-Chieh Chen4Liang-Chih Liu5Chen-Yuan Lin6School of Pharmacy and Graduate Institute, China Medical UniversityDepartment of Pharmacy, China Medical University HospitalDepartment of Pharmacy, Kaohsiung Chang Gung Memorial HospitalSchool of Nursing, China Medical UniversitySchool of Pharmacy and Graduate Institute, China Medical UniversitySurgical Department, China Medical University HospitalSchool of Pharmacy and Graduate Institute, China Medical UniversityAbstract Background Cancer patients receiving targeted therapies need to prevent QTc prolongation and life-threatening cardiovascular (CV) events to maintain a balanced benefit-risk ratio. This study aimed to develop an optimal prediction model for QTc prolongation risk and estimate its risk probability in cancer patients treated with oral tyrosine kinase inhibitors (TKIs). Methods This retrospective cohort study analyzed electronic medical records (EMR) of cancer patients newly treated with commonly used oral TKIs at a medical center between January 2016 and December 2020. QTc prolongation was defined as ≥ 450 ms in males and ≥ 470 ms in females using Bazett’s formula. The study followed four key steps: (1) Managing missing data, (2) Identifying important variables, (3) Training and testing the best prediction models, (4). Estimating risk probability and determining cut-off points. Both univariate logistic regression (LR) and supervised machine learning (ML) approaches were used for variable selection. The backward LR method and seven ML algorithms were applied to train and test the prediction models. The best model was identified based on model performance, fitting criteria, area under the receiver operating characteristic curve (AUROC), risk probability cut-off points, and clinical relevance. Results The statistical 12-parameter model demonstrated excellent performance (AUROC = 0.89, sensitivity = 0.91, specificity = 0.75) and strong discrimination ability for risk probability prediction (AUROC = 0.78, cut-off = 0.46), outperforming other ML models. In the final best model: the baseline risk probability of QTc prolongation was 0.13, even in the absence of other contributing factors. Baseline QTc prolongation and a history of cardiovascular disease (excluding arrhythmia, cardiomyopathy, etc.) contributed the most to incremental risk probability (0.471 and 0.282, respectively), after controlling for other factors. The remaining 10 factors each contributed to an increased probability of QTc prolongation for more than 0.14 probability. Conclusions A logistic regression model utilizing 12 easily accessible variables from EMRs outperformed ML models in predicting the risk probability of QTc prolongation in cancer patients newly treated with five oral TKIs. These findings serve as a valuable clinical reference for integrating digital monitoring into cardiovascular care for cancer survivors undergoing targeted therapy with TKIs. They also underscore the importance of screening baseline ECG before initiating TKIs to assess the risk of QTc prolongation, facilitating early intervention and prevention in the future.https://doi.org/10.1186/s12911-025-03091-8QTc prolongationOralTyrosine kinase inhibitorsLogistic regressionMachine learningAlgorithm
spellingShingle Hsiang-Wen Lin
Tien-Chao Lin
Chien-Ning Hsu
Tzu-Pei Yeh
Yu-Chieh Chen
Liang-Chih Liu
Chen-Yuan Lin
Risk prediction of QTc prolongation occurrence in cancer patients treated with commonly used oral tyrosine kinase inhibitors: machine learning modeling or conventional statistical analysis better?
BMC Medical Informatics and Decision Making
QTc prolongation
Oral
Tyrosine kinase inhibitors
Logistic regression
Machine learning
Algorithm
title Risk prediction of QTc prolongation occurrence in cancer patients treated with commonly used oral tyrosine kinase inhibitors: machine learning modeling or conventional statistical analysis better?
title_full Risk prediction of QTc prolongation occurrence in cancer patients treated with commonly used oral tyrosine kinase inhibitors: machine learning modeling or conventional statistical analysis better?
title_fullStr Risk prediction of QTc prolongation occurrence in cancer patients treated with commonly used oral tyrosine kinase inhibitors: machine learning modeling or conventional statistical analysis better?
title_full_unstemmed Risk prediction of QTc prolongation occurrence in cancer patients treated with commonly used oral tyrosine kinase inhibitors: machine learning modeling or conventional statistical analysis better?
title_short Risk prediction of QTc prolongation occurrence in cancer patients treated with commonly used oral tyrosine kinase inhibitors: machine learning modeling or conventional statistical analysis better?
title_sort risk prediction of qtc prolongation occurrence in cancer patients treated with commonly used oral tyrosine kinase inhibitors machine learning modeling or conventional statistical analysis better
topic QTc prolongation
Oral
Tyrosine kinase inhibitors
Logistic regression
Machine learning
Algorithm
url https://doi.org/10.1186/s12911-025-03091-8
work_keys_str_mv AT hsiangwenlin riskpredictionofqtcprolongationoccurrenceincancerpatientstreatedwithcommonlyusedoraltyrosinekinaseinhibitorsmachinelearningmodelingorconventionalstatisticalanalysisbetter
AT tienchaolin riskpredictionofqtcprolongationoccurrenceincancerpatientstreatedwithcommonlyusedoraltyrosinekinaseinhibitorsmachinelearningmodelingorconventionalstatisticalanalysisbetter
AT chienninghsu riskpredictionofqtcprolongationoccurrenceincancerpatientstreatedwithcommonlyusedoraltyrosinekinaseinhibitorsmachinelearningmodelingorconventionalstatisticalanalysisbetter
AT tzupeiyeh riskpredictionofqtcprolongationoccurrenceincancerpatientstreatedwithcommonlyusedoraltyrosinekinaseinhibitorsmachinelearningmodelingorconventionalstatisticalanalysisbetter
AT yuchiehchen riskpredictionofqtcprolongationoccurrenceincancerpatientstreatedwithcommonlyusedoraltyrosinekinaseinhibitorsmachinelearningmodelingorconventionalstatisticalanalysisbetter
AT liangchihliu riskpredictionofqtcprolongationoccurrenceincancerpatientstreatedwithcommonlyusedoraltyrosinekinaseinhibitorsmachinelearningmodelingorconventionalstatisticalanalysisbetter
AT chenyuanlin riskpredictionofqtcprolongationoccurrenceincancerpatientstreatedwithcommonlyusedoraltyrosinekinaseinhibitorsmachinelearningmodelingorconventionalstatisticalanalysisbetter