Scholarlink

Objective <p>Chronic pain is common among middle-aged and older Chinese adults, yet latent pain patterns, temporal transitions, and individual susceptibility factors remain unclear. Using five waves of the China Health and Retirement Longitudinal Study, we assessed latent pain classes, their dynamic changes, and key predictors of low back pain (LBP) in adults aged ≥ 45 years.</p> Methods <p>Latent class analysis (LCA) was used to identify pain-site patterns at each wave, and Kendall correlations assessed associations among sites. Latent transition analysis (LTA) evaluated temporal shifts and the modifying effects of age and sex. For LBP, we developed machine-learning models (logistic regression, random forest, decision tree, extreme gradient boosting, light gradient boosting machine, support vector machine, artificial neural network), optimized through multiple imputation, feature engineering, and grid search. A nomogram was constructed from features consistently important across artificial neural network, logistic regression, and extreme gradient boosting, followed by subgroup and mediation analyses.</p> Results <p>LCA identified 2–4 classes per wave, summarized into three pain states: widespread pain, localized pain, and no pain. Lumbar pain was most common. LTA confirmed the three-state model and nine transition pathways; the no-pain state showed highest stability (82.3%). Men were more likely to enter or remain in pain states (<i>p</i> &lt; 0.05), and older adults without pain were more likely to shift to localized pain (<i>p</i> &lt; 0.05). Artificial neural network achieved the best performance in the test set. SHapley additive explanations analyses based on multiple machine-learning models consistently identified the center for epidemiologic studies depression scale-10 items, cumulative chronic disease burden, and activities of daily living function as the strongest predictors of LBP. Seven stable predictors were used to build a nomogram with favorable discrimination (C-index = 0.783), calibration, and clinical utility.</p> Conclusion <p>Pain patterns in middle-aged and older Chinese adults show substantial heterogeneity and dynamic transitions. Psychological status, chronic disease burden, and functional capacity are the key predictors of LBP.</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

Latent pain class identification, longitudinal transitions, and machine learning prediction of incident low back pain in middle-aged and older Chinese adults

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月6日

查看全文
  • Junpeng Liu,
  • Zhiheng Zhao,
  • Shuhuan Li,
  • Xinglin Liu,
  • Sheyang Xu,
  • Bowen Lu,
  • Xianglong Meng

摘要

Objective

Chronic pain is common among middle-aged and older Chinese adults, yet latent pain patterns, temporal transitions, and individual susceptibility factors remain unclear. Using five waves of the China Health and Retirement Longitudinal Study, we assessed latent pain classes, their dynamic changes, and key predictors of low back pain (LBP) in adults aged ≥ 45 years.

Methods

Latent class analysis (LCA) was used to identify pain-site patterns at each wave, and Kendall correlations assessed associations among sites. Latent transition analysis (LTA) evaluated temporal shifts and the modifying effects of age and sex. For LBP, we developed machine-learning models (logistic regression, random forest, decision tree, extreme gradient boosting, light gradient boosting machine, support vector machine, artificial neural network), optimized through multiple imputation, feature engineering, and grid search. A nomogram was constructed from features consistently important across artificial neural network, logistic regression, and extreme gradient boosting, followed by subgroup and mediation analyses.

Results

LCA identified 2–4 classes per wave, summarized into three pain states: widespread pain, localized pain, and no pain. Lumbar pain was most common. LTA confirmed the three-state model and nine transition pathways; the no-pain state showed highest stability (82.3%). Men were more likely to enter or remain in pain states (p < 0.05), and older adults without pain were more likely to shift to localized pain (p < 0.05). Artificial neural network achieved the best performance in the test set. SHapley additive explanations analyses based on multiple machine-learning models consistently identified the center for epidemiologic studies depression scale-10 items, cumulative chronic disease burden, and activities of daily living function as the strongest predictors of LBP. Seven stable predictors were used to build a nomogram with favorable discrimination (C-index = 0.783), calibration, and clinical utility.

Conclusion

Pain patterns in middle-aged and older Chinese adults show substantial heterogeneity and dynamic transitions. Psychological status, chronic disease burden, and functional capacity are the key predictors of LBP.