Abstract
Most screening tests for T2DM in use today were developed using multivariate regression methods that are often further simplified to allow transformation into a scoring formula. The increasing volume of electronically collected data opened the opportunity to develop more complex, accurate prediction models that can be continuously updated using machine learning approaches. This study compares machine learning-based prediction models (i.e. Glmnet, RF, XGBoost, LightGBM) to commonly used regression models for prediction of undiagnosed T2DM. The performance in prediction of fasting plasma glucose level was measured using 100 bootstrap iterations in different subsets of data simulating new incoming data in 6-month batches. With 6 months of data available, simple regression model performed with the lowest average RMSE of 0.838, followed by RF (0.842), LightGBM (0.846), Glmnet (0.859) and XGBoost (0.881). When more data were added, Glmnet improved with the highest rate (+ 3.4%). The highest level of variable selection stability over time was observed with LightGBM models. Our results show no clinically relevant improvement when more sophisticated prediction models were used. Since higher stability of selected variables over time contributes to simpler interpretation of the models, interpretability and model calibration should also be considered in development of clinical prediction models.
Highlights
Most screening tests for Type 2 diabetes mellitus (T2DM) in use today were developed using multivariate regression methods that are often further simplified to allow transformation into a scoring formula
After manually removing variables like ‘Region’, ‘Finnish Diabetes Risk Score (FINDRISC) groups’ and various precomputed scores [for age, body mass index (BMI), waist circumference, FINDRISC], we reduced the number of variables from 111 to 103
We studied differences in performance, calibration and interpretability of machine learning-based prediction models and multivariable regression models
Summary
Most screening tests for T2DM in use today were developed using multivariate regression methods that are often further simplified to allow transformation into a scoring formula. This study compares machine learning-based prediction models (i.e. Glmnet, RF, XGBoost, LightGBM) to commonly used regression models for prediction of undiagnosed T2DM. Due to the increased prevalence of T2DM, various techniques have been used to build predictive models and models for early disease diagnosis, such as logistic and Cox proportional hazard regression m odels[5,6,7], Random F orest[8,9], boosted ensembles[10,11], etc. There have been a number of important developments in relation to machine learning methods in recent years that can be tested for predictive modelling through interrogation of electronic healthcare record (EHR) data; these techniques include AdaBoost, random forest, support vector regression, decision tree[13] and neural network-based on Stacked Denoising Autoencoders (SDA)[14]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
