Abstract
OBJECTIVE: Our aim was to apply state-of-the-art machine learning algorithms to predict the risk of future progression to diabetes complications, including diabetic kidney disease (≥30% decline in eGFR) and diabetic retinopathy (mild, moderate or severe).
RESEARCH DESIGN AND METHODS: Using data in a cohort of 537 adults with type 1 diabetes we predicted diabetes complications emerging during a median follow-up of 5.4 years. Prediction models were computed first with clinical risk factors at baseline (17 measures) and then with clinical risk factors and blood-derived molecular data (965 molecules) at baseline. Participants were classified into different groups: type 1 diabetes stable (n=195), type 1 diabetes with progression to diabetic kidney disease (≥30% decline in eGFR; n=79) or diabetic retinopathy (mild, moderate or severe; n=111). Progression of diabetic kidney disease and diabetic retinopathy were predicted in two complication-specific models. Area under the receiver operating characteristic curves (AUROCs) were validated by both k-fold and Monte-Carlo simulation for each model. Clinical risk factor and molecule importance were explained with the SHAP algorithm. Accuracy, precision, recall, and F-score were used to evaluate clinical utility.
RESULTS: During a median follow-up of 5.4 years, 79 (21 %) of the participants (mean±SD: age 54.8 ± 13.7 years) progressed in diabetic kidney disease and 111 (29 %) of the participants progressed to diabetic retinopathy. The two predictive models for diabetic kidney disease progression were highly accurate with 15 clinical risk factors: AUROCs of 0.92±0.01 and 0.96±0.25 (k-fold and Monte Carlo simulation values respectively) and when six omics predictors were included: AUROC of 0.99±0.001 and 0.96±0.06. The top three important features were albuminuria, estimated glomerular filtration rate and retinopathy status at baseline for both models.
Models for diabetic retinopathy progression with 12 clinical risk predictors achieved an AUROC of 0.81±0.10 and 0.75±0.14 and with additional seven omics included an AUROC of 0.87±0.001 and 0.79±0.16. The top features were hemoglobin A1c, albuminuria and retinopathy status at baseline.
CONCLUSIONS: application of machine learning effectively predicted five-year progression of diabetes complications, in particular diabetic kidney disease. Further replication of machine learning tools in a real-world context will facilitate implementation in the clinic.
RESEARCH DESIGN AND METHODS: Using data in a cohort of 537 adults with type 1 diabetes we predicted diabetes complications emerging during a median follow-up of 5.4 years. Prediction models were computed first with clinical risk factors at baseline (17 measures) and then with clinical risk factors and blood-derived molecular data (965 molecules) at baseline. Participants were classified into different groups: type 1 diabetes stable (n=195), type 1 diabetes with progression to diabetic kidney disease (≥30% decline in eGFR; n=79) or diabetic retinopathy (mild, moderate or severe; n=111). Progression of diabetic kidney disease and diabetic retinopathy were predicted in two complication-specific models. Area under the receiver operating characteristic curves (AUROCs) were validated by both k-fold and Monte-Carlo simulation for each model. Clinical risk factor and molecule importance were explained with the SHAP algorithm. Accuracy, precision, recall, and F-score were used to evaluate clinical utility.
RESULTS: During a median follow-up of 5.4 years, 79 (21 %) of the participants (mean±SD: age 54.8 ± 13.7 years) progressed in diabetic kidney disease and 111 (29 %) of the participants progressed to diabetic retinopathy. The two predictive models for diabetic kidney disease progression were highly accurate with 15 clinical risk factors: AUROCs of 0.92±0.01 and 0.96±0.25 (k-fold and Monte Carlo simulation values respectively) and when six omics predictors were included: AUROC of 0.99±0.001 and 0.96±0.06. The top three important features were albuminuria, estimated glomerular filtration rate and retinopathy status at baseline for both models.
Models for diabetic retinopathy progression with 12 clinical risk predictors achieved an AUROC of 0.81±0.10 and 0.75±0.14 and with additional seven omics included an AUROC of 0.87±0.001 and 0.79±0.16. The top features were hemoglobin A1c, albuminuria and retinopathy status at baseline.
CONCLUSIONS: application of machine learning effectively predicted five-year progression of diabetes complications, in particular diabetic kidney disease. Further replication of machine learning tools in a real-world context will facilitate implementation in the clinic.
| Originalsprog | Engelsk |
|---|---|
| Udgiver | medRxiv |
| DOI | |
| Status | Udgivet - 8 okt. 2021 |