Use Machine Learning Methods to Explore the Associated Risk Factors of Osteoporosis or Bone Loss in Chinese People With Type 2 Diabetes：A Clinical Prediction Model

Abstract

Background: Osteoporosis (OP) and bone loss are high risk factors for fractures. More and more studies have shown that compared with the general population, patients with type 2 diabetes mellitus (T2DM) are more likely to develop fractures. However, the relevant mechanism remains controversial.

Purpose: This study aimed to use machine learning (ML) methods to explore the associated risk factors with OP and bone loss in the Chinese T2DM population, so as to construct useful clinical prediction models.

Methods: This was a two-center, retrospective study. The data came from a epidemiological investigation database conducted in Ningde City and Wuyishan City, Fujian Province, China from March 2011 to December 2014. Finally, 798 T2DM patients who met the enrollment criteria were included in the final analysis. In order to control gender as a confounding factor that affects the results, we constructed two clinical prediction models based on different genders. We used the least absolute shrinkage and selection operator (LASSO) algorithm to filter relevant feature variables. The selected characteristic variables were modeled by logistic regression (LR), and clinical nomograms were used for more intuitive expression. The stability, clinical applicability and recognition of the model were evaluated by C-index, receiver operating characteristic (ROC) curve, calibration chart and decision curve analysis (DCA). Internal verification was achieved through bootstrapping validation.

Results: In the clinical prediction model of female patients. There were a total of 9 associated predictors, namely age, marital status, glutamyl transpeptidase, fracture, coronary heart disease, fruit-flavored drinks, moderate-intensity exercise, menopause and nap time were determined by LASSO analysis from a total of 69 variables. The model we constructed using these 9 associated predictors showed medium prediction ability (C-index value: 0.738, 95%CI[0.692, 0.784]), the C-index in bootstrapping validation was 0.714, and the area under the ROC curve (AUC) was 0.738. The DCA showed that if the risk threshold was between 4% and 100%, the nomogram could be used clinically. In the clinical prediction model of male patients. A total of 12 associated predictors were identified from 65 variables through LASSO analysis, including age, marital status, fasting serum insulin, alanine aminotransferase, coronary heart disease, respiratory diseases, diabetic retinopathy, seafood, desserts, fruit-flavored beverages, coffee, high-intensity exercise. The model we constructed using these 12 associated predictors showed medium prediction ability (C-index value: 0.751, 95%CI[0.694-0.808]), the C-index in bootstrapping validation was 0.704, and the AUC value was 0.751. The DCA showed that if the risk threshold was between 3% and 68%, the nomogram could be used clinically.

Conclusion: We explored the associated risk factors of OP or bone Loss in Chinese people with type 2 diabetes, and developed a risk nomogram with moderate predictive power.

Note: Funding Statement: This work was supported by the Chinese Medical Association Foundation and Chinese Endocrine Society (Grant 12020240314), the National Natural Science Foundation of China (Grant 81270874) and the Natural Science Foundation of Fujian Province (Grant 2011J06012). Declaration of Interests: The authors declare that they have no conflicts of interest for this work. Ethics Approval Statement: The research procedures were explained in detail to all subjects and the informed consent form was signed. The research protocol has been approved by the Ethics Committee of the Fujian Provincial Hospital (Ethics number: K2020-10-002).

Keywords: Type 2 diabetes mellitus, osteoporosis, bone loss, risk factors, nomogram, machine learning