Comparison of an ensemble machine learning model to a Cox regression model to predict colorectal cancer risk among people with HIV using retrospective nationwide cohort data in Sweden: a study protocol
Open Access
Abstract
Introduction: There are currently no colorectal cancer (CRC) screening recommendations specifically outlined for people with HIV (PWH). Screening measures used for people without HIV (PWoH) have been previously discussed as sufficient for use among PWH, despite observations of higher CRC prevalence and CRC reportedly appearing at earlier ages among PWH in comparison to PWoH. Machine learning (ML) methods are regarded as robust approaches that may enhance predictive performance, particularly in the context of complex or high-dimensional data. This study aims to develop an ensemble ML model to predict CRC risk in PWH using comprehensive nationwide datasets. The model’s predictive performance will be evaluated and compared with a baseline Cox proportional regression model. The better-performing method will be implemented to develop a CRC risk prediction model with the aim of personalising screening recommendations for PWH. Methods and analysis: The study population will include all PWH and PWoH born between 1940 and 2008, aged 18 or older and living in Sweden sometime between 1983 and 2024. The study population will be linked to six nationwide demographic and healthcare registers. Follow-up will continue until the first incident of CRC, emigration or death. The outcome of interest is CRC. PWH will be matched to negative controls 1:10. A Cox regression analysis will be completed first, and the results will be used as a baseline comparison to the ensemble ML results. A range of ML methods will be used to develop the ensemble model using stacking. Ethics and dissemination: This study has ethical approval from the Regional Ethical Committee in Sweden (Dnr: 2024-04185-02, 2024-06783-02, 2023-00191-01, 2022-02897-02, 2022-05624-01, 2018/11-31/2). Given that the study is retrospective and register-based, using only pseudonymised data, there are minimal physical, psychological or privacy risks to included individuals. All results will be presented at the population level with no possibility of identification. The results of this study will be submitted for publication in a peer-reviewed journal.Keywords
Funding Information
- Vetenskapsrådet (2024-01769)
This publication has 43 references indexed in Scilit:
- Employee Attrition Prediction Using Bayesian Optimized Stacked Ensemble Learning and Explainable AISN Computer Science, 2025
- Nordic colorectal cancer screening programmes: A comparison of organization, operation, and quality indicatorsEuropean Journal of Cancer, 2025
- Comparing AI/ML approaches and classical regression for predictive modeling using large population health databases: Applications to COVID-19 case predictionGlobal Epidemiology, 2024
- Consistency between the National Patient Register and the Swedish Cancer RegisterPharmacoepidemiology and Drug Safety, 2024
- Young-onset colon cancer among people living with HIV in metropolitan AtlantaInternational Journal of Colorectal Disease, 2022
- Occurrence of comorbidity with colorectal cancer and variations by age and stage at diagnosisCancer Epidemiology, 2022
- Incidence and mortality of non-AIDS-defining cancers among people living with HIV: A systematic review and meta-analysiseClinicalMedicine, 2022
- The longitudinal integrated database for health insurance and labour market studies (LISA) and its use in medical researchEuropean Journal of Epidemiology, 2019
- The Swedish cause of death registerEuropean Journal of Epidemiology, 2017
- Registers of the Swedish total population and their use in medical researchEuropean Journal of Epidemiology, 2016