Screening Carpal Tunnel Syndrome Risk in Occupational Populations — China, 2018–2023

Abstract

What is known about this topic?

Carpal tunnel syndrome is a common work-related disorder that is affected by personal, biomechanical, and psychosocial factors. Existing risk assessment methods often lack precision, hindering early intervention. As of August 1, 2025, occupational musculoskeletal disorders were recognized in China, emphasizing the need for effective preventive tools.

What has been added by this report?

This study developed machine learning models to identify carpal tunnel syndrome risk using a national dataset (n=88,609). The best model achieved 97.66% (3,130/3,131) accuracy, identifying key factors such as hand pain frequency and lifting load, and enhancing the early identification of high-risk workers.

What are the implications for public health practices?

These findings advocate for the use of data-driven screening tools in occupational health screening to enable early interventions. Employers should prioritize ergonomic improvements and symptom-reporting systems in alignment with China's proactive occupational disease policies.

ABSTRACT

Introduction: Carpal tunnel syndrome (CTS) poses a substantial public health burden, necessitating precise early detection strategies that align with the prevention-focused goals of the evolving occupational health policy in China.

Methods: Machine learning algorithms were employed to integrate multidimensional risk factors for CTS screening using a large-scale dataset of 88,609 valid responses, including 10,438 reported CTS cases. The SHapley Additive exPlanations analysis identified the key factors, and ablation tests confirmed the importance of these features.

Results: Among the developed models, extreme gradient boosting demonstrated superior performance [recall: 99.97% (3,130/3,131); precision: 84.43% (3,130/3,707)] in identifying high-risk individuals. The SHapley Additive exPlanations (SHAP) analysis revealed that subjective hand pain characteristics were the primary factors, followed by psychosocial stressors, including fatigue and biomechanical exposure.

Conclusion: These findings support the transition toward data-driven public health practices, enabling targeted early prevention and ergonomic interventions to mitigate the burden of CTS in at-risk workforces.

Carpal tunnel syndrome (CTS) has emerged as a critical occupational health challenge owing to the increasing prevalence of repetitive strain injuries in diverse work environments (1). Globally, approximately 14.4% of the population is affected by CTS (1). In China, its prevalence among office workers is 9.6%, which is notably higher than that observed in many Western populations (2). On August 1, 2025, occupational musculoskeletal disorders were officially included in the updated "Classification and Catalog of Occupational Diseases" in China. This major policy milestone places CTS within the scope of statutory occupational disease management, creating an urgent imperative for enterprises and public health authorities to implement effective and legally compliant preventive and control measures.

The etiology of CTS is multifactorial and involves complex interactions among biomechanical exposure (including repetitive hand movements), personal demographics (including age and sex), and psychosocial stressors (3–5). Epidemiological studies have identified specific predictors. For instance, Lund et al. (6) quantified the impact of wrist velocity on the prevalence of CTS, whereas Lee et al. (7) highlighted demographic disparities and noted a high prevalence among middle-aged women. However, traditional linear statistical methods often fail to capture the nonlinear and high-dimensional interactions among diverse risk factors (8). Consequently, the current screening approaches may lack the sensitivity required for early detection in large-scale occupational populations, representing a critical gap in meeting new prevention-oriented regulatory standards.

To address these limitations, this study applied advanced machine learning algorithms to synthesize multidimensional risk factors into a robust risk screening framework. By enhancing screening accuracy beyond traditional methods, this study aimed to provide a data-driven foundation for early screening and targeted intervention, thereby supporting the implementation of new occupational health policies and reducing the burden of CTS in the workforce.

Figure 1 presents a schematic overview of the methodology of the study. This study used data from a large-scale national occupational health survey conducted in China between 2018 and 2023. The dataset comprised 88,609 valid responses from workers across nine major industrial sectors and 29 distinct job types (9). A total of 10,438 participants reported CTS, corresponding to an overall prevalence of 11.78% in the study population. Data were collected using the electronic Chinese Musculoskeletal Disorder Questionnaire, integrated into the ergonomic assessment and WMSDs analysis system developed by the National Institute of Occupational Health and Poison Control, Chinese Center for Disease Control and Prevention (10). This study was approved by the Ethics Review Committee of the National Institute of Occupational Health and Poison Control, China. All participants provided informed consent prior to the study.

Figure 1.

Schematic overview of the study methodology.

Abbreviation: LR=logistic regression; DT=decision tree; RF=random forest; GBM=gradient boosting machine; XGBoost=extreme gradient boosting; AUC=area under the ROC-curve; SHAP=SHapley Additive exPlanations.

The primary outcome variable was self-reported CTS status (binary: CTS vs. non-CTS). The adoption of a standardized self-report-based CTS is fully aligned with the core objective of this study: to develop a scalable, high-efficiency risk-screening tool for large-scale occupational populations. To build a robust multidimensional screening model, potential risk factors were stratified into three domains: 1) personal factors (demographics, BMI, and exercise habits), 2) biomechanical factors (repetitive movements, force exertion, and posture), and 3) psychosocial factors (job demands and task load). Data preprocessing involved rigorous cleaning (removal of outliers and missing values) and encoding (standardization of continuous variables and ordinal or one-hot encoding of categorical variables). To ensure that the developed tool focused on practically meaningful risk factors rather than statistical noise, a dual-criteria feature-selection strategy was applied. Statistical tests (t-tests, chi-square tests, or nonparametric equivalents) were paired with effect size analyses. Only risk factors demonstrating both statistical significance (P<0.05) and moderate to strong effect sizes (φ>0.1, |r|>0.25, ε²>0.06) were retained as input features for the machine learning models. Statistical analyses were performed using SPSS (version 19.0, IBM Corp., Armonk, NY, USA).

Five machine learning algorithms — binary logistic regression, decision tree, random forest, gradient boosting, and extreme gradient boosting (XGBoost) — were developed to screen CTS risk. These models were selected for their capacity to handle nonlinear interactions inherent in occupational exposure data. The dataset was partitioned into training [70% (62,026/88,609)] and testing [30% (26,583/88,609)] sets. To address typical class imbalance in occupational disease data, the synthetic minority oversampling technique (SMOTE) was applied to the training set. The train-test split was performed at the beginning of the pipeline. All data-dependent procedures were performed exclusively within the training partitions. The held-out test set remained fully separate throughout all preprocessing and tuning steps to eliminate information leakage and ensure unbiased performance estimates.

SHapley Additive exPlanations (SHAP) values were used to identify the most relevant risk factors and to guide early intervention strategies. The model performance was evaluated using metrics prioritized for public health screening utility. While accuracy provides an overall baseline, emphasis is placed on sensitivity (recall) and the F1 score. High sensitivity is critical in occupational health surveillance to minimize false negatives (missed screenings) and ensure that high-risk workers receive timely ergonomic interventions. Precision was monitored to reduce false positives and unnecessary resource utilization. The area under the receiver operating characteristic curve was calculated to assess the discriminative power of the models across all classification thresholds.

SHAP analysis identified a clear hierarchy of risk factors and actionable targets for occupational CTS prevention (Table 1). Three risk tiers were proposed based on SHAP predictive weight, modifiability, and functional role in occupational CTS prevention. Tier 1: Core symptoms comprised the top-ranked hand pain and symptom variables with dominant explanatory power, serving as high-yield primary screening signals for identifying at-risk workers. Tier 2: Modifiable factors included psychosocial fatigue and biomechanical workload factors of moderate importance, representing the only intervention-actionable targets for workplace ergonomic optimization and fatigue management. Tier 3: Background covers stable personal and occupational factors, such as job tenure and general health, with small but consistent screening contributions used for baseline risk stratification rather than direct intervention. Subjective hand pain characteristics emerged as the strongest screening signals, with pain frequency and duration ranking as the top two contributors to model accuracy. Among the modifiable occupational risk factors, fatigue perception (psychosocial domain) and cumulative lifting load (biomechanical domain) demonstrated the highest importance, followed by repetitive hand movements and non-neutral wrist postures. Personal demographic factors, including age and sex, showed moderate but consistent contributions. This factor ranking directly informs workplace prevention strategies that should prioritize systematic symptom monitoring, fatigue management, and physical load reduction as core intervention components.

Table 1. Ranking of the top ten factors for carpal tunnel syndrome based on mean SHAP values and proposed risk tiers.

Rank	Factors	Category	Mean SHAP	Proposed risk tier
Abbreviation: SHAP=SHapley Additive exPlanations.
1	Frequency of hand pain/discomfort	Personal	2.029	Tier 1: Core Symptom
2	Hand pain/discomfort in past 12 months	Personal	1.356	Tier 1: Core Symptom
3	Tenure (years) in current job	Personal	0.053	Tier 3: Background
4	Physical health status	Personal	0.047	Tier 3: Background
5	Physical exercise	Personal	0.022	Tier 3: Background
6	Hand pain/discomfort in past 7 days	Personal	0.016	Tier 1: Core Symptom
7	Feel fatigue	Psychosocial	0.013	Tier 2: Modifiable
8	Hours until feeling fatigued	Psychosocial	0.013	Tier 2: Modifiable
9	Cumulative lifting load	Biomechanical	0.009	Tier 2: Modifiable
10	Duration of hand symptoms (Annual)	Personal	0.009	Tier 1: Core Symptom

Statistical screening initially identified 26 significant factors spanning the personal, biomechanical, and psychosocial domains (Supplementary Table S1, available at https://weekly.chinacdc.cn/). Ablation testing further confirmed the fundamentally multidimensional nature of CTS risk. Removing any single factor domain, including psychosocial factors, resulted in a notable degradation in the model screening performance (Figure 2). This finding validates the necessity of integrated multidomain occupational health surveillance rather than single-factor risk monitoring.

Figure 2.

Model accuracy decreased with the ablation of the screening factors.

Abbreviation: XGBoost=extreme Gradient Boosting.

Among the five machine learning algorithms evaluated, XGBoost emerged as the optimal model for large-scale occupational CTS screening (Supplementary Table S2, available at https://weekly.chinacdc.cn/). It achieved near-perfect sensitivity [99.97% (3,130/3,131)], which is critical for minimizing false negatives in public health surveillance, while maintaining high precision [84.43% (3,130/3,707)] and robust discriminative power, with an area under the receiver operating characteristic curve of 0.992 (Supplementary Figure S1, available at https://weekly.chinacdc.cn/). Although the decision tree also achieved a high recall, XGBoost delivered the most balanced performance between case detection capability and screening reliability for population-level screening applications.

DISCUSSION

This study aimed to develop and validate a generalizable high-performance machine learning screening model for CTS risk screening to support large-scale universal occupational health surveillance. The analytical framework of this study is strictly aligned with this objective. By synthesizing multidimensional risk factors using machine learning, this study established a high-precision risk-screening framework for CTS in occupational populations. The XGBoost model demonstrated superior risk-screening performance, particularly in terms of sensitivity (99.97%), ensuring minimal false-negative results during mass screening. SHAP analysis further identified subjective hand pain characteristics, fatigue perception, and biomechanical load as primary risk factors, validating the multifactorial etiology of CTS.

These findings have profound public health implications, particularly following the inclusion of musculoskeletal disorders in China's "Classification and Catalog of Occupational Diseases." First, the dominance of hand pain symptoms provided strong evidence for the establishment of symptom-based active sentinel monitoring. This approach offers a low-cost, direct pathway for early detection that is superior to passive reliance on clinical screening. Second, identifying fatigue as a critical factor bridges the gap between psychosocial stress and physical damage. This finding underscores the fact that effective CTS control requires the management of organizational factors, such as work rhythm, directly aligning with the Healthy China 2030 initiative's requirements for occupational stress management. Ultimately, this model serves as a quantifiable risk-grading tool, facilitating a paradigm shift in occupational health management from post-diagnosis compensation to pre-diagnosis prevention.

This study was limited by its cross-sectional design. The core screening variables of the established SHAP-based screening model were predominantly hand pain- and discomfort-related indicators extracted from the cross-sectional survey data, including the subjective pain frequency and duration. These symptomatic variables substantially overlapped with the symptomatic definition of self-reported CTS, which may partially account for the excellent screening performance of our model. This overlap indicates that the model primarily achieves symptom-based CTS identification rather than a definitive diagnosis. Furthermore, this study prioritized discrimination metrics for occupational health risk stratification and did not perform a formal calibration analysis. Future research should include calibration plots and Brier score evaluations to validate the reliability of the predicted risks and better support real-world decision-making.

From a practical application perspective, the proposed model was developed for the preliminary population-level screening of suspected carpal tunnel syndrome (CTS) in occupational health practice. It aims to support the early identification of at-risk workers and guide targeted follow-up arrangements and is not intended to serve as a standalone diagnostic tool. The definitive diagnosis of CTS still relies on formal specialist clinical assessment combined with objective electrophysiological examinations, such as nerve conduction studies, which remain the diagnostic gold standard.

The developed machine learning models can also support targeted employer interventions. For example, employers should establish nonpunitive, rapid symptom-reporting mechanisms (such as "one-click reporting" via mobile applications). Reports of frequent hand pain should serve as immediate triggers for on-site ergonomic investigations. Furthermore, to address the critical risks associated with fatigue and insufficient rest, enterprises should implement structured microbreak policies and optimize work pacing to prevent continuous high-intensity exposure. For the risks associated with cumulative lifting loads, priority should be given to engineering controls, such as introducing assistive handling devices and optimizing the workstation height to minimize non-neutral wrist postures.

These findings provide a scientific basis for refining screening and prevention standards for occupational CTS and guide the allocation of work injury prevention funds. Future initiatives should focus on the longitudinal validation of these models across diverse industries to establish causal relationships. Policymakers can use these data-driven insights to formulate ergonomic exposure limits that are suitable for the Chinese industrial context.

SUPPLEMENTARY DATA

Supplementary data to this article can be found online.

Funding Statement

Supported by the National Key R&D Program of China (Project Number: 2022YFC2503205), the Project of Occupational Health Risk Assessment and the National Occupational Health Standard Formulation of the National Institute of Occupational Health and Poison Control (Project Number: 102393220020090000020), Capital’s Funds for Health Improvement and Research (Capital Health Development 2024-1G-3101), and Key R&D Program of the Xinjiang Uygur Autonomous Region (2024B03040)

Conflicts of interest

No conflicts of interest.