Scholarlink

<p>Chronic kidney disease (CKD) and osteoarthritis (OA) are highly prevalent chronic conditions that substantially impair quality of life in middle-aged and older adults. Large-sample longitudinal evidence regarding the association between CKD and incident OA risk in the Chinese population remains limited. In addition, interpretable machine learning approaches for OA identification in high-risk populations remain insufficiently explored. This study aimed to investigate the longitudinal association between baseline CKD and OA risk using a prospective cohort design and to develop an interpretable machine learning model for OA identification based on the China Health and Retirement Longitudinal Study (CHARLS) database. Data were derived from four independent prospective cohorts in the CHARLS database (2011–2013, 2011–2015, 2011–2018, and 2011–2020), encompassing 16,493 participants followed for 2–9 years. Multivariable logistic regression and Cox proportional hazards models with hierarchical adjustment were used to evaluate the independent association between baseline CKD and OA risk. The robustness of the findings was assessed through multiple pre-specified sensitivity analyses, including a dual-validation CKD definition combining estimated glomerular filtration rate (eGFR) and self-reported physician diagnosis. An automated machine learning pipeline based on the Tree-based Pipeline Optimization Tool (TPOT) was developed to construct OA identification models in both the overall population and the CKD-specific subgroup. Model interpretability was achieved using SHapley Additive exPlanations (SHAP) analysis. Baseline CKD was consistently associated with an increased likelihood of OA across all four follow-up cohorts, with fully adjusted odds ratios (ORs) ranging from 1.990 to 2.219 (all <i>P</i> &lt; 0.001). Cox proportional hazards analysis further demonstrated a significant association between baseline CKD and subsequent incident OA, with a fully adjusted hazard ratio (HR) of 1.502 (95% confidence interval [CI]: 1.431–1.576, <i>P</i> &lt; 0.001). Sensitivity analyses and cross-cohort validation consistently supported the robustness of the observed association. The optimized TPOT models demonstrated moderate discrimination for OA identification, achieving test-set AUC values of 0.768–0.777 in the overall population and 0.743 in the CKD subgroup. SHAP analysis identified knee pain, stomach illness, recent outpatient visits, and CKD status among the features that contributed most strongly to model output. In this large prospective cohort study, baseline CKD was associated with a substantially increased risk of subsequent OA among Chinese middle-aged and older adults. The machine learning models demonstrated moderate discrimination and provided interpretable information regarding the relative contribution of routinely collected clinical features to OA identification. These findings support the potential value of integrating CKD-related information into OA screening and risk stratification approaches, while further external validation, calibration, and assessment of clinical utility are needed before broader clinical implementation. Given the potential bidirectional relationship between CKD and OA, future longitudinal and causal studies are warranted to further clarify the directionality and mechanisms underlying this association.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Association between chronic kidney disease and osteoarthritis risk and machine learning-based osteoarthritis identification in CHARLS

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月30日

查看全文
  • Bing He,
  • Weibing Shuang,
  • Zhao Hou,
  • Qiwei Wang

摘要

Chronic kidney disease (CKD) and osteoarthritis (OA) are highly prevalent chronic conditions that substantially impair quality of life in middle-aged and older adults. Large-sample longitudinal evidence regarding the association between CKD and incident OA risk in the Chinese population remains limited. In addition, interpretable machine learning approaches for OA identification in high-risk populations remain insufficiently explored. This study aimed to investigate the longitudinal association between baseline CKD and OA risk using a prospective cohort design and to develop an interpretable machine learning model for OA identification based on the China Health and Retirement Longitudinal Study (CHARLS) database. Data were derived from four independent prospective cohorts in the CHARLS database (2011–2013, 2011–2015, 2011–2018, and 2011–2020), encompassing 16,493 participants followed for 2–9 years. Multivariable logistic regression and Cox proportional hazards models with hierarchical adjustment were used to evaluate the independent association between baseline CKD and OA risk. The robustness of the findings was assessed through multiple pre-specified sensitivity analyses, including a dual-validation CKD definition combining estimated glomerular filtration rate (eGFR) and self-reported physician diagnosis. An automated machine learning pipeline based on the Tree-based Pipeline Optimization Tool (TPOT) was developed to construct OA identification models in both the overall population and the CKD-specific subgroup. Model interpretability was achieved using SHapley Additive exPlanations (SHAP) analysis. Baseline CKD was consistently associated with an increased likelihood of OA across all four follow-up cohorts, with fully adjusted odds ratios (ORs) ranging from 1.990 to 2.219 (all P < 0.001). Cox proportional hazards analysis further demonstrated a significant association between baseline CKD and subsequent incident OA, with a fully adjusted hazard ratio (HR) of 1.502 (95% confidence interval [CI]: 1.431–1.576, P < 0.001). Sensitivity analyses and cross-cohort validation consistently supported the robustness of the observed association. The optimized TPOT models demonstrated moderate discrimination for OA identification, achieving test-set AUC values of 0.768–0.777 in the overall population and 0.743 in the CKD subgroup. SHAP analysis identified knee pain, stomach illness, recent outpatient visits, and CKD status among the features that contributed most strongly to model output. In this large prospective cohort study, baseline CKD was associated with a substantially increased risk of subsequent OA among Chinese middle-aged and older adults. The machine learning models demonstrated moderate discrimination and provided interpretable information regarding the relative contribution of routinely collected clinical features to OA identification. These findings support the potential value of integrating CKD-related information into OA screening and risk stratification approaches, while further external validation, calibration, and assessment of clinical utility are needed before broader clinical implementation. Given the potential bidirectional relationship between CKD and OA, future longitudinal and causal studies are warranted to further clarify the directionality and mechanisms underlying this association.