Tao Yang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Tao Yang has a total of 79 co-authors. Here are the top co-authors by number of shared publications:
- Kongfa Hu
School of Artificial Intelligence and Information Technology, Nanjing University of Chinese Medicine
6 shared publications - Guo-Zheng Li
Tongji University
2 shared publications - Haiyang He
Information Office, Henan University of Chinese Medicine, Zhengzhou 450046, China
2 shared publications - Jinwan Wang
School of Information Management Nanjing University Nanjing China
2 shared publications - Mark Xuefang Zhu
School of Information Management, Nanjing University
2 shared publications - Qing‐Feng Yin
Department of Sports Medicine The Second Hospital of Shandong University Jinan China
2 shared publications
Latest Publications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 72 (4) , 12922-12933
A Frequency-Conditioned Transformer for Gastrointestinal Image Segmentation in Consumer Healthcare Electronics
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 72 (4) , 12922-12933
A Frequency-Conditioned Transformer for Gastrointestinal Image Segmentation in Consumer Healthcare Electronics
- Open Access
- 3 September 2026
- Published by Springer Nature in BMC Nephrology
- Vol. 27 (1)
Characteristics associated with efficacy of hypoxia-inducible factor-prolyl hydroxylase inhibitors in renal anemia patients: a retrospective cohort study
- 3 September 2026
- Published by Springer Nature in BMC Nephrology
- Vol. 27 (1)
Open AccessCharacteristics associated with efficacy of hypoxia-inducible factor-prolyl hydroxylase inhibitors in renal anemia patients: a retrospective cohort study
- Open Access
- 5 February 2026
- Published by Springer Nature in Research Square
Dual-Model Collaboration with Consistency Calibration for Intelligent Acupuncture Prescription Recommendation
Background Acupuncture is a core therapeutic modality in traditional Chinese medicine (TCM), widely used for symptom relief and functional regulation, yet its clinical effectiveness relies heavily on individualized acupoint selection. To address the personalized requirements for selecting acupuncture points in TCM, this study investigates a dual-model collaboration strategy for intelligent acupuncture prescription recommendation. The aim is to provide auxiliary decision support for clinical acupuncture treatment. Methods A dual-model collaborative framework for intelligent acupuncture prescription recommendation is proposed. Specifically, symptom-to-acupoint mapping is formulated as an end-to-end text generation task and an intelligent recommendation method based on dual-model collaboration with consistency calibration(DM3C) is proposed. Firstly, a Chinese Bidirectional and Auto-Regressive Transformers(BART) model is fine-tuned on clinical symptom-acupoint pairs to generate baseline acupoint sequences. Then, a retrieval-augmented prompting strategy is used to query large language models (LLMs), producing additional candidate prescriptions from similar historical cases. Finally, a random forest-based consistency calibration module integrates multi-source signals to score and filter candidate acupoints, yielding a reliability-weighted final recommendation. Results Experimental results demonstrate that the DM3C significantly improves acupuncture prescription recommendation performance. When more prompts are retrieved (top-k = 21), the DeepSeek-V3-based collaborative achieves an F1 score of 0.3821, a Recall of 0.5216, and a Coverage of 0.3372, representing a 0.1062 increase in F1 over the original DeepSeek-V3 without model collaboration. The GPT-4o-based model attains an F1 score of 0.3780 and a Recall of 0.4833, with a 0.0169 improvement in Recall compared with its retrieval-enhanced counterpart. Conclusion By combining a domain-fine-tuned generator, retrieval-augmented LLM outputs, and a consistency-calibrated random forest filter, the proposed the DM3C enhances the generalizability and Recall of intelligent acupuncture prescription recommendations, providing effective support for clinical decision-making in TCM acupuncture treatment.
- 5 February 2026
- Published by Springer Nature in Research Square
Open AccessDual-Model Collaboration with Consistency Calibration for Intelligent Acupuncture Prescription Recommendation
Background Acupuncture is a core therapeutic modality in traditional Chinese medicine (TCM), widely used for symptom relief and functional regulation, yet its clinical effectiveness relies heavily on individualized acupoint selection. To address the personalized requirements for selecting acupuncture points in TCM, this study investigates a dual-model collaboration strategy for intelligent acupuncture prescription recommendation. The aim is to provide auxiliary decision support for clinical acupuncture treatment. Methods A dual-model collaborative framework for intelligent acupuncture prescription recommendation is proposed. Specifically, symptom-to-acupoint mapping is formulated as an end-to-end text generation task and an intelligent recommendation method based on dual-model collaboration with consistency calibration(DM3C) is proposed. Firstly, a Chinese Bidirectional and Auto-Regressive Transformers(BART) model is fine-tuned on clinical symptom-acupoint pairs to generate baseline acupoint sequences. Then, a retrieval-augmented prompting strategy is used to query large language models (LLMs), producing additional candidate prescriptions from similar historical cases. Finally, a random forest-based consistency calibration module integrates multi-source signals to score and filter candidate acupoints, yielding a reliability-weighted final recommendation. Results Experimental results demonstrate that the DM3C significantly improves acupuncture prescription recommendation performance. When more prompts are retrieved (top-k = 21), the DeepSeek-V3-based collaborative achieves an F1 score of 0.3821, a Recall of 0.5216, and a Coverage of 0.3372, representing a 0.1062 increase in F1 over the original DeepSeek-V3 without model collaboration. The GPT-4o-based model attains an F1 score of 0.3780 and a Recall of 0.4833, with a 0.0169 improvement in Recall compared with its retrieval-enhanced counterpart. Conclusion By combining a domain-fine-tuned generator, retrieval-augmented LLM outputs, and a consistency-calibrated random forest filter, the proposed the DM3C enhances the generalizability and Recall of intelligent acupuncture prescription recommendations, providing effective support for clinical decision-making in TCM acupuncture treatment.
- Open Access
- 19 November 2025
- Published by Springer Nature in Chinese Medicine
- Vol. 20 (1) , 193
Reinforcement learning for LLM-based explainable TCM prescription recommendation with implicit preferences from small language models
Objective: In an effort to reinforce both the interpretability and accuracy of prescription recommendations in Traditional Chinese Medicine (TCM), this study puts forward a two-stage training framework that integrates knowledge distillation from a teacher model with implicit preference-driven reinforcement learning grounded in a compact model. Methods: Above all, GPT-4o is employed to parse structured TCM clinical records, creating high-quality distillation samples. These are employed to guide Low-Rank Adaptation (LoRA)-based fine-tuning of the Qwen2.5-7B model, enabling it to generate explainable outputs in the format of "symptom analysis—prescription recommendation—prescription explanation". Then, a lightweight BART (Bidirectional and Auto-Regressive Transformers) model is trained to learn the mapping relation between symptoms and prescriptions. Its outputs are compared with those of the large model to construct preference pairs, which are subsequently utilized in Direct Preference Optimization (DPO)-based reinforcement tuning to further align the model with potentially better recommendations. Results: The suggested model achieves a P @30 of 35.62% and F1 @30 of 37.36%, outperforming existing baselines. Knowledge distillation contributes to the improvement of the model's generalization and explainability, while implicit preference-based reinforcement further enhances F1 @30 by 2.01%. Overall, the model obtains more desirable performance in both accuracy and explainability. Conclusion: The recommended approach not only improves the quality and transparency of TCM prescription recommendations, but also offers a fruitful strategy for building trustworthy and clinically applicable intelligent TCM decision-support systems.
- 19 November 2025
- Published by Springer Nature in Chinese Medicine
- Vol. 20 (1) , 193
Open AccessReinforcement learning for LLM-based explainable TCM prescription recommendation with implicit preferences from small language models
Objective: In an effort to reinforce both the interpretability and accuracy of prescription recommendations in Traditional Chinese Medicine (TCM), this study puts forward a two-stage training framework that integrates knowledge distillation from a teacher model with implicit preference-driven reinforcement learning grounded in a compact model. Methods: Above all, GPT-4o is employed to parse structured TCM clinical records, creating high-quality distillation samples. These are employed to guide Low-Rank Adaptation (LoRA)-based fine-tuning of the Qwen2.5-7B model, enabling it to generate explainable outputs in the format of "symptom analysis—prescription recommendation—prescription explanation". Then, a lightweight BART (Bidirectional and Auto-Regressive Transformers) model is trained to learn the mapping relation between symptoms and prescriptions. Its outputs are compared with those of the large model to construct preference pairs, which are subsequently utilized in Direct Preference Optimization (DPO)-based reinforcement tuning to further align the model with potentially better recommendations. Results: The suggested model achieves a P @30 of 35.62% and F1 @30 of 37.36%, outperforming existing baselines. Knowledge distillation contributes to the improvement of the model's generalization and explainability, while implicit preference-based reinforcement further enhances F1 @30 by 2.01%. Overall, the model obtains more desirable performance in both accuracy and explainability. Conclusion: The recommended approach not only improves the quality and transparency of TCM prescription recommendations, but also offers a fruitful strategy for building trustworthy and clinically applicable intelligent TCM decision-support systems.
- 11 November 2025
- Published by Elsevier in Journal of Affective Disorders
- Vol. 394 (Pt B) , 120656
Development of explainable machine learning-based prediction models of depression with mild and severe insomnia in middle-aged and elderly Chinese people
- 11 November 2025
- Published by Elsevier in Journal of Affective Disorders
- Vol. 394 (Pt B) , 120656
Development of explainable machine learning-based prediction models of depression with mild and severe insomnia in middle-aged and elderly Chinese people
- Open Access
- 7 March 2025
- Published by Springer Nature in Aging Clinical and Experimental Research
- Vol. 37 (1) , 1-15
An explainable machine learning-based prediction model for sarcopenia in elderly Chinese people with knee osteoarthritis
Background: Sarcopenia is an age-related progressive skeletal muscle disease that leads to loss of muscle mass and function, resulting in adverse health outcomes such as falls, functional decline, and death. Knee osteoarthritis (KOA) is a common chronic degenerative joint disease among elderly individuals who causes joint pain and functional impairment. These two conditions often coexist in elderly individuals and are closely related. Early identification of the risk of sarcopenia in KOA patients is crucial for developing intervention strategies and improving patient health. Methods: This study utilized data from the China Health and Retirement Longitudinal Study (CHARLS), selecting symptomatic KOA patients aged 65 years and above and analyzing a total of 95 variables. Predictive factors were screened via least absolute shrinkage and selection operator (LASSO) regression and logistic regression. Eight machine learning algorithms were employed to construct predictive models, with internal cross-validation and independent test validation performed. The final selected model was analyzed via the SHapley Additive exPlanations (SHAP) method to enhance interpretability and clinical applicability. To facilitate clinical use, we developed a web application based on this model ( http://106.54.231.169/ ). Results: The results indicate that six predictive factors—body mass index, upper arm length, marital status, total cholesterol, cystatin C, and shoulder pain—are closely associated with the risk of sarcopenia in KOA patients. CatBoost demonstrated excellent overall performance in both calibration analyses and probability estimates, reflecting accurate and dependable predictions. The final results on the independent test set (accuracy = 0.8902; F1 = 0.8627; AUC = 0.9697; Brier score = 0.0691) indicate that the model possesses strong predictive performance and excellent generalization ability, with predicted probabilities closely aligning with actual occurrence rates and thereby underscoring its reliability. Conclusion: From the perspective of public health and aging, this study constructed an interpretable sarcopenia risk prediction model on the basis of routine clinical data. This model can be used for early screening and risk assessment of symptomatic KOA patients, assisting health departments and clinicians in the early detection and follow-up of relevant populations, thereby improving the quality of life and health outcomes of elderly individuals.
- 7 March 2025
- Published by Springer Nature in Aging Clinical and Experimental Research
- Vol. 37 (1) , 1-15
Open AccessAn explainable machine learning-based prediction model for sarcopenia in elderly Chinese people with knee osteoarthritis
Background: Sarcopenia is an age-related progressive skeletal muscle disease that leads to loss of muscle mass and function, resulting in adverse health outcomes such as falls, functional decline, and death. Knee osteoarthritis (KOA) is a common chronic degenerative joint disease among elderly individuals who causes joint pain and functional impairment. These two conditions often coexist in elderly individuals and are closely related. Early identification of the risk of sarcopenia in KOA patients is crucial for developing intervention strategies and improving patient health. Methods: This study utilized data from the China Health and Retirement Longitudinal Study (CHARLS), selecting symptomatic KOA patients aged 65 years and above and analyzing a total of 95 variables. Predictive factors were screened via least absolute shrinkage and selection operator (LASSO) regression and logistic regression. Eight machine learning algorithms were employed to construct predictive models, with internal cross-validation and independent test validation performed. The final selected model was analyzed via the SHapley Additive exPlanations (SHAP) method to enhance interpretability and clinical applicability. To facilitate clinical use, we developed a web application based on this model ( http://106.54.231.169/ ). Results: The results indicate that six predictive factors—body mass index, upper arm length, marital status, total cholesterol, cystatin C, and shoulder pain—are closely associated with the risk of sarcopenia in KOA patients. CatBoost demonstrated excellent overall performance in both calibration analyses and probability estimates, reflecting accurate and dependable predictions. The final results on the independent test set (accuracy = 0.8902; F1 = 0.8627; AUC = 0.9697; Brier score = 0.0691) indicate that the model possesses strong predictive performance and excellent generalization ability, with predicted probabilities closely aligning with actual occurrence rates and thereby underscoring its reliability. Conclusion: From the perspective of public health and aging, this study constructed an interpretable sarcopenia risk prediction model on the basis of routine clinical data. This model can be used for early screening and risk assessment of symptomatic KOA patients, assisting health departments and clinicians in the early detection and follow-up of relevant populations, thereby improving the quality of life and health outcomes of elderly individuals.
- Open Access
- 10 January 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (1) , e0316130
Sex differences in the association between composite dietary antioxidant index and hyperlipidemia: Insights from NHANES
Background: Previous studies have shown that both the composite dietary antioxidant index (CDAI) and sex are strongly associated with a variety of cardiovascular diseases, but sex differences between CDAI and hyperlipidemia are unknown. Objective: This study utilized data from the National Health and Nutrition Examination Survey (NHANES) to investigate the sex differences between CDAI and hyperlipidemia. Method: We calculated the CDAI of the six dietary antioxidants using data from NHANES, explored the relationship between CDAI and the prevalence of hyperlipidemia using multivariate logistic regression analysis, and analyzed for potential nonlinear associations using restricted cubic spline. Finally, the association between CDAI and hyperlipidemia was further explored using multivariate logistic regression in different genders. Results: The study included a total of 34,754 participants with a mean age of 47.04 years, of whom 49.37% were man. In a fully adjusted multivariable binary logistic regression model, CDAI was negatively associated with the prevalence of hyperlipidemia (OR = 0.99, 95% CI:0.98–0.99). In addition, participants in the highest quartile had a lower risk of hyperlipidaemia compared with the lowest quartile of CDAI (OR = 0.83, 95%CI: 0.76–0.92). We also found a non-linear relationship (non-linear P = 0.003, Inflection point = -0.179). Finally, we found that the association between CDAI and the prevalence of hyperlipidemia was significantly stronger in the female population than in the male population (P for interaction <0.05). Conclusion: Our study highlights the L-shaped association between CDAI and the prevalence of hyperlipidemia in the general adult population. In addition, this association was more significant in the female population than in the male population.
- 10 January 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (1) , e0316130
Open AccessSex differences in the association between composite dietary antioxidant index and hyperlipidemia: Insights from NHANES
Background: Previous studies have shown that both the composite dietary antioxidant index (CDAI) and sex are strongly associated with a variety of cardiovascular diseases, but sex differences between CDAI and hyperlipidemia are unknown. Objective: This study utilized data from the National Health and Nutrition Examination Survey (NHANES) to investigate the sex differences between CDAI and hyperlipidemia. Method: We calculated the CDAI of the six dietary antioxidants using data from NHANES, explored the relationship between CDAI and the prevalence of hyperlipidemia using multivariate logistic regression analysis, and analyzed for potential nonlinear associations using restricted cubic spline. Finally, the association between CDAI and hyperlipidemia was further explored using multivariate logistic regression in different genders. Results: The study included a total of 34,754 participants with a mean age of 47.04 years, of whom 49.37% were man. In a fully adjusted multivariable binary logistic regression model, CDAI was negatively associated with the prevalence of hyperlipidemia (OR = 0.99, 95% CI:0.98–0.99). In addition, participants in the highest quartile had a lower risk of hyperlipidaemia compared with the lowest quartile of CDAI (OR = 0.83, 95%CI: 0.76–0.92). We also found a non-linear relationship (non-linear P = 0.003, Inflection point = -0.179). Finally, we found that the association between CDAI and the prevalence of hyperlipidemia was significantly stronger in the female population than in the male population (P for interaction <0.05). Conclusion: Our study highlights the L-shaped association between CDAI and the prevalence of hyperlipidemia in the general adult population. In addition, this association was more significant in the female population than in the male population.
- Open Access
- 1 January 2025
- Published by American Association for the Advancement of Science (AAAS) in Computational and Structural Biotechnology Journal
- Vol. 27, 4915-4935
HerbComb: An integrated database for the discovery of novel combinational therapies from herbal medicines
Drug combinations have gained increasing interest due to their potential for optimal efficacy and for overcoming drug resistance. Herbal medicine is a complex system with multiple ingredients and has illustrated apparent therapeutic effects through thousands of years of clinical experience, offering a rich resource for combinational therapies. However, the underlying mechanisms of their synergistic therapeutic effects in disease treatment remain obscure for most herbal medicines. To fill this gap, we developed an integrated database, HerbComb (https://herbcomb.com/), to discover combinatorial therapies in herbal medicine, featuring three unique functions: drug prepositioning, customized combinatorial analysis, and biological and pharmacological features. Additionally, transcriptional gene signatures and ADMET properties were also interrogated as interactive functionalities to facilitate combinational analysis. Using Tongxinluo for Stroke as a case study, we identified and validated Oleanolic acid and Ferulic acid as synergistic ingredients. In summary, HerbComb is a versatile data exploration platform designed to characterize synergistic interactions among herbal medicines, enhance understanding of synergistic mechanisms, and facilitate the discovery of effective drug combinations.
- 1 January 2025
- Published by American Association for the Advancement of Science (AAAS) in Computational and Structural Biotechnology Journal
- Vol. 27, 4915-4935
Open AccessHerbComb: An integrated database for the discovery of novel combinational therapies from herbal medicines
Drug combinations have gained increasing interest due to their potential for optimal efficacy and for overcoming drug resistance. Herbal medicine is a complex system with multiple ingredients and has illustrated apparent therapeutic effects through thousands of years of clinical experience, offering a rich resource for combinational therapies. However, the underlying mechanisms of their synergistic therapeutic effects in disease treatment remain obscure for most herbal medicines. To fill this gap, we developed an integrated database, HerbComb (https://herbcomb.com/), to discover combinatorial therapies in herbal medicine, featuring three unique functions: drug prepositioning, customized combinatorial analysis, and biological and pharmacological features. Additionally, transcriptional gene signatures and ADMET properties were also interrogated as interactive functionalities to facilitate combinational analysis. Using Tongxinluo for Stroke as a case study, we identified and validated Oleanolic acid and Ferulic acid as synergistic ingredients. In summary, HerbComb is a versatile data exploration platform designed to characterize synergistic interactions among herbal medicines, enhance understanding of synergistic mechanisms, and facilitate the discovery of effective drug combinations.
- 5 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 30 (4) , 2889-2902
Advanced Camera-Based Scoliosis Screening via Deep Learning Detection and Fusion of Trunk, Limb, and Skeleton Features
Scoliosis significantly impacts quality of life, highlighting the need for effective early scoliosis screening (SS) and intervention. However, current SS methods often involve physical contact, undressing, or radiation exposure. This study introduces an innovative, non-invasive SS approach utilizing a monocular RGB camera that eliminates the need for undressing, sensor attachment, and radiation exposure. We introduce a novel approach that employs Parameterized Human 3D Reconstruction (PH3DR) to reconstruct 3D human models, thereby effectively eliminating clothing obstructions, seamlessly integrated with an ISANet segmentation network, which has been enhanced by Multi-Scale Fusion Attention (MSFA) module we proposed for facilitating the segmentation of distinct human trunk and limb features (HTLF), capturing body surface asymmetries related to scoliosis. Additionally, we propose a Swin Transformer-enhanced CMU-Pose to extract human skeleton features (HSF), identifying skeletal asymmetries crucial for SS. Finally, we develop a fusion model that integrates the HTLF and HSF, combining surface morphology and skeletal features to improve the precision of SS. The experiments demonstrated that PH3DR and MSFA significantly improved the segmentation and extraction of HTLF, whereas ST-based CMU-Pose substantially enhanced the extraction of HSF. Our final model achieved a comparable F1 (0.895 $\pm$ 0.014) to the best-performing baseline model, with only 0.79% of the parameters and 1.64% of the FLOPs, achieving 36 FPS–significantly higher than the best-performing baseline model (10 FPS). Moreover, our model outperformed two spine surgeons, one less experienced and the other moderately experienced. With its patient-friendly, privacy-preserving, and easily deployable solution, this approach is particularly well-suited for early SS and routine monitoring.
- 5 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 30 (4) , 2889-2902
Advanced Camera-Based Scoliosis Screening via Deep Learning Detection and Fusion of Trunk, Limb, and Skeleton Features
Scoliosis significantly impacts quality of life, highlighting the need for effective early scoliosis screening (SS) and intervention. However, current SS methods often involve physical contact, undressing, or radiation exposure. This study introduces an innovative, non-invasive SS approach utilizing a monocular RGB camera that eliminates the need for undressing, sensor attachment, and radiation exposure. We introduce a novel approach that employs Parameterized Human 3D Reconstruction (PH3DR) to reconstruct 3D human models, thereby effectively eliminating clothing obstructions, seamlessly integrated with an ISANet segmentation network, which has been enhanced by Multi-Scale Fusion Attention (MSFA) module we proposed for facilitating the segmentation of distinct human trunk and limb features (HTLF), capturing body surface asymmetries related to scoliosis. Additionally, we propose a Swin Transformer-enhanced CMU-Pose to extract human skeleton features (HSF), identifying skeletal asymmetries crucial for SS. Finally, we develop a fusion model that integrates the HTLF and HSF, combining surface morphology and skeletal features to improve the precision of SS. The experiments demonstrated that PH3DR and MSFA significantly improved the segmentation and extraction of HTLF, whereas ST-based CMU-Pose substantially enhanced the extraction of HSF. Our final model achieved a comparable F1 (0.895 $\pm$ 0.014) to the best-performing baseline model, with only 0.79% of the parameters and 1.64% of the FLOPs, achieving 36 FPS–significantly higher than the best-performing baseline model (10 FPS). Moreover, our model outperformed two spine surgeons, one less experienced and the other moderately experienced. With its patient-friendly, privacy-preserving, and easily deployable solution, this approach is particularly well-suited for early SS and routine monitoring.
- Open Access
- 2 August 2023
- Published by Springer Nature in BMC Medical Informatics and Decision Making
- Vol. 23 (1) , 1-12
Study on risk factors of diabetic peripheral neuropathy and establishment of a prediction model by machine learning
Background: Diabetic peripheral neuropathy (DPN) is a common complication of diabetes. Predicting the risk of developing DPN is important for clinical decision-making and designing clinical trials.Methods: We retrospectively reviewed the data of 1278 patients with diabetes treated in two central hospitals from 2020 to 2022. The data included medical history, physical examination, and biochemical index test results. After feature selection and data balancing, the cohort was divided into training and internal validation datasets at a 7:3 ratio. Training was made in logistic regression, k-nearest neighbor, decision tree, naive bayes, random forest, and extreme gradient boosting (XGBoost) based on machine learning. The k-fold cross-validation was used for model assessment, and the accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (AUC) were adopted to validate the models’ discrimination and clinical practicality. The SHapley Additive exPlanation (SHAP) was used to interpret the best-performing model.Results: The XGBoost model outperformed other models, which had an accuracy of 0·746, precision of 0·765, recall of 0·711, F1-score of 0·736, and AUC of 0·813. The SHAP results indicated that age, disease duration, glycated hemoglobin, insulin resistance index, 24-h urine protein quantification, and urine protein concentration were risk factors for DPN, while the ratio between 2-h postprandial C-peptide and fasting C-peptide(C2/C0), total cholesterol, activated partial thromboplastin time, and creatinine were protective factors.Conclusions: The machine learning approach helped established a DPN risk prediction model with good performance. The model identified the factors most closely related to DPN.
- 2 August 2023
- Published by Springer Nature in BMC Medical Informatics and Decision Making
- Vol. 23 (1) , 1-12
Open AccessStudy on risk factors of diabetic peripheral neuropathy and establishment of a prediction model by machine learning
Background: Diabetic peripheral neuropathy (DPN) is a common complication of diabetes. Predicting the risk of developing DPN is important for clinical decision-making and designing clinical trials.Methods: We retrospectively reviewed the data of 1278 patients with diabetes treated in two central hospitals from 2020 to 2022. The data included medical history, physical examination, and biochemical index test results. After feature selection and data balancing, the cohort was divided into training and internal validation datasets at a 7:3 ratio. Training was made in logistic regression, k-nearest neighbor, decision tree, naive bayes, random forest, and extreme gradient boosting (XGBoost) based on machine learning. The k-fold cross-validation was used for model assessment, and the accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (AUC) were adopted to validate the models’ discrimination and clinical practicality. The SHapley Additive exPlanation (SHAP) was used to interpret the best-performing model.Results: The XGBoost model outperformed other models, which had an accuracy of 0·746, precision of 0·765, recall of 0·711, F1-score of 0·736, and AUC of 0·813. The SHAP results indicated that age, disease duration, glycated hemoglobin, insulin resistance index, 24-h urine protein quantification, and urine protein concentration were risk factors for DPN, while the ratio between 2-h postprandial C-peptide and fasting C-peptide(C2/C0), total cholesterol, activated partial thromboplastin time, and creatinine were protective factors.Conclusions: The machine learning approach helped established a DPN risk prediction model with good performance. The model identified the factors most closely related to DPN.