Abstract: TH-PO013
Machine Learning Algorithm in Predicting Non-Diabetic Kidney Disease in Type 2 Diabetes Mellitus: Development and Validation of a Noninvasive Predictor Scoring Model
Session Information
- AI, Digital Health, Data Science - I
November 02, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Veeranki, Vamsidhar, Sanjay Gandhi Post Graduate Institute of Medical Sciences, Lucknow, Uttar Pradesh, India
- Prasad, Narayan, Sanjay Gandhi Post Graduate Institute of Medical Sciences, Lucknow, Uttar Pradesh, India
- Meyyappan, Jeyakumar, Sanjay Gandhi Post Graduate Institute of Medical Sciences, Lucknow, Uttar Pradesh, India
Group or Team Name
- Nephro-SGPGI.
Background
Identifying non-diabetic kidney disease (NDKD) is essential in retarding the progression of chronic kidney disease among patients with type-2 diabetes mellitus(T2DM). Renal biopsy, despite being the gold standard in detecting the presence of NDKD, has an inherent risk of life-threatening complications. The current study aims to develop a non-invasive scoring model to predict the presence of NDKD using clinical and laboratory parameters.
Methods
Patients with T2DM who underwent biopsy for various indications were included and were divided into derivational and validation cohorts. Using the variables significantly associated with presence of NDKD on biopsy on univariate analysis, a model was developed using multivariate logistic-regression based machine learning algorithm. The model was then run on the derivational(internal validation) and validation cohort(temporal-validation) and the performance was assessed by receiver operating characteristic(ROC) curve.
Results
A total of 538 patients with T2DM were included in the study analysis; 376 in derivational cohort and 162 in validation cohort. The final model consists- diabetes mellitus duration< 5 years, absence of coronary artery disease, absence of diabetic retinopathy, presence of oliguria, acute rise in creatinine, & low serum complement-C3 level significantly predicted presence of NDKD on renal-biopsy. The model performed robustly with AUC-ROC of 0.869(95%CI:0.805-0.933) in validation cohort.
Conclusion
The clinical and laboratory parameter-based prediction model robustly predicted the NDKD among T2DM patients, and a cut-off total score of ≥ 6 has a high sensitivity & specificity of 86% &80% in predicting NDKD.
Derivation and validation of the prediction model