Kidney Week

Abstract: TH-PO019

Development of Machine Learning-Based Models for Detection of Cognitive Impairment in Patients Receiving Maintenance Hemodialysis

Session Information

• Augmented Intelligence for Prediction and Image Analysis
  October 24, 2024 | Location: Exhibit Hall, 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

• Ling, Tsai-chieh, National Cheng Kung University Hospital, Tainan, Tainan, Taiwan

Tsai-chieh Ling, MD

• Chang, Chiung-Chih, Chang Gung Memorial Hospital Linkou, Kaohsiung, Taiwan

Chiung-Chih Chang

• Wu, Jia-Ling, National Cheng Kung University, Tainan, Taiwan

Jia-Ling Wu, MSc

• Lin, Wei-Ren, National Cheng Kung University Hospital, Tainan, Tainan, Taiwan

Wei-Ren Lin, MD

• Sun, Chien Yao, National Cheng Kung University Hospital, Tainan, Tainan, Taiwan

Chien Yao Sun, MD, MSc

• Huang, Chieh-Hsin, National Cheng Kung University Hospital, Tainan, Tainan, Taiwan

Chieh-Hsin Huang

• Chang, Yu Tzu, National Cheng Kung University Hospital, Tainan, Taiwan

Yu Tzu Chang

Background

Cognitive impairment is common but frequently undiagnosed in the dialysis population. There is no recommended screening tool specifically designed for them. We aimed to develop and validate a quick and accurate screening tool composed of only a few items in the Mini-Mental State Examination (MMSE) and Cognitive Abilities Screening Instrument (CASI) using machine-learning based approaches in hemodialysis patients.

Methods

We conducted a cross-sectional observational study in which the MMSE and CASI were administered in 508 hemodialysis patients and randomly divided into a derivation set (70%) and a validation set (30%). Cognitive impairment was defined as a CASI score below the 20th percentile of age- and education-matched norms. Using three to five key items from MMSE and CASI as predictors, we developed six machine learning models, including Lasso, classification and regression tree (CART), random forest (RF), extreme gradient boosting (XGboost), support vector machine (SVM), and artificial neural networks (ANN) in derivation set. We then evaluated the predictive performance of these models in the validation set.

Results

The derivation samples (n = 357) had a mean (SD) age of 64.13 (11.92) years and a mean education level of 8.76 (4.91) years. Around 40% participants were identified as cognitive impairment according to their total CASI score. Among all the developed machine-learning models, the RF model achieved the highest performance of prediction, with an accuracy of 0.94, an area under the curve (AUC) of 0.95, and an F1 score of 0.92 in the validation set. The other models, except for CART, performed equally well in terms of AUC. The RF and SVM models demonstrated the greatest net benefit over clinical thresholds according to decision curve analysis.

Conclusion

Our study demonstrates that using machine-learning models can efficiently identify patients with CASI score < 20 percentile of age- and education-matched norms with only several questions in CASI and MMSE within 5 minutes. Since cognitive impairment implicates decision making and is associated with poor outcomes, early detection might improves patient care and enables timely referral.

Funding

• Clinical Revenue Support