Introduction

Knee osteoarthritis (KOA) is the leading cause of disability worldwide, and its prevalence increases significantly with age1. The early detection of KOA patients is clinically pivotal for administering tiered therapeutic strategies that may decelerate, arrest, or potentially reverse pathological progression2. In many studies, incident knee OA is defined using either radiographic features (e.g. incident KL grade ≥ 2) or criteria based on symptoms (e.g. clinical/combined ACR criteria)3. Conventional imaging often fails to detect early structural changes in KOA4. This creates a diagnostic gap between anatomical damage and the onset of symptoms5. Measuring pain during daily activities helps detect functional decline and central sensitization earlier than imaging, overcoming the sensitivity limits of X-rays.. Movement-evoked pain (MEP), through its mechanistic link with mechanical stress, potentially offers superior pathophysiological relevance compared to resting pain for characterizing KOA-related nociceptive processes across disease stages6. Contemporary pain evaluation remains predominantly dependent on subjective rating tools, while the standardized MEP test lacks systematic incorporation into clinical assessment frameworks. The inadequate understanding of bidirectional relationships between pain phenotypes and therapeutic responses further hinders the formulation of precision treatment regimens.

In recent decades, researchers have formulated multiple risk prediction models incorporating multidimensional indicators to address challenges in early-stage diagnosis7,8. Current predictive models predominantly derived from tertiary medical institution data lack comprehensive validation regarding their feasibility in grassroots healthcare environments. Despite sporadic attempts to develop models using community cohorts, the scarcity of validation studies conducted in authentic community contexts significantly impedes their clinical implementation potential9. As a simple statistical visual tool, the nomogram is widely used to predict the occurrence, development, prognosis, and survival of diseases in recent years. Meanwhile, for the validation study conducted in a real community environment, we selected the community residents in Hangzhou as the subjects for early prediction of KOA. Therefore, this model mainly focuses on the pain-induced tests suitable for community conditions for the early prediction of KOA. This not only improves the accuracy of the prediction but also has significant clinical value.

Method

Study design and data sources

Strictly adhering to the TRIPOD statement10, we developed the prediction model using cross-sectional data from the Hangzhou KOA Community Cohort. This prospective study (May-December 2018) aimed to elucidate the KOA prevalence and risk factor profiles among permanent residents aged ≥ 40 in Hangzhou. The original protocol received ethical approval from the First Affiliated Hospital of Zhejiang Chinese Medical University (Approval No. 2018-ZX-026-01). The research was conducted in accordance with the Helsinki Declaration, and the informed consent of all subjects and/or their legal guardians was obtained. The current modeling analysis, as an extension of the prospective cohort, obtained new ethical clearance from the same institution (Approval No. 2025-KLS-335-01).

Sampling Strategy and Eligibility Criteria.

A three-stage stratified cluster random sampling strategy was implemented. First, 6 administrative districts were randomly selected from 13 Hangzhou districts using random number generation. Within each selected district, 2 neighborhood committees were chosen through probability proportional to size (PPS) sampling. Cluster surveys were conducted for all eligible permanent residents, resulting in 4676 baseline participants. Eligibility criteria were established as follows:

The inclusion criteria were: (1) Age ≥40 years with ≥5-year continuous community residency; (2) No history of knee surgery; (3) Capacity to complete standardized assessments independently; (4) Written informed consent

The exclusion criteria were: (1) Comorbid joint disorders: rheumatoid arthritis, psoriatic arthritis, ankylosing spondylitis, other autoimmune arthropathies; gouty, septic or traumatic arthritis. (2) Pain assessment confounders: chronic opioid use; peripheral neuropathy or CNS disorders affecting pain perception. (3) Functional or cognitive limitations: severe cognitive impairment or communication barriers; lower extremity musculoskeletal conditions potentially confounding knee pain assessment.

Standardized MEP test

Per OARSI recommendations for KOA clinical trials: “Pain can be assessed after specific activities (e.g., a walk test). If a pain assessment occurs after an activity then the study team should ensure consistency throughout the trial with the type and duration of activity as well as the timing of the pain assessment after the activity”11. Integrating these recommendations with WOMAC and KOOS core elements, we established a standardized MEP protocol comprising five dimensions: (1) Initiation pain: 6-m walk test after 15-min rest, recording pain during gait initiation; (2) Squatting pain: The peak pain that occurs during the process of slowly squatting down to the maximum extent and maintaining it for 15 s before standing up; (3) Stair climbing pain: The maximum degree of pain experienced during the process of climbing up and down 11 flights of stairs (each flight of stairs is 16 ± 1 cm high); (4) Housework pain: simulated pain that occurs within 3 min after mopping the floor; (5) Maximum pain: highest pain across all tests. All pain intensities were quantified using a Visual Analogue Scale (VAS, 0–10) administered by certified pain management nurses.

Data collection and quality control

Post-MEP assessments, standardized weight-bearing knee radiographs were obtained at once: 5° foot external rotation, 20° knee flexion, using posteroanterior projection. Two board-certified radiologists independently performed KL grading. Inter-rater reliability (weighted κ = 0.86, 95%CI:0.72–0.88) and intra-rater consistency (κ = 0.92, 95%CI:0.86–0.99) were established. KL grade ≥ 2 defined radiographic KOA. Treatment histories (physical therapy, oral/topical NSAIDs, intra-articular injections) were retrieved from electronic medical records. Exercise habits were additionally surveyed. Although dynamic pain assessment protocols were rigorously standardized, body mass index (BMI) remained unavailable due to heterogeneous documentation practices in community health records.

Statistical modeling and validation framework

The complete dataset was randomly split into training/validation sets (7:3 ratio). LASSO regression with tenfold cross-validation identified optimal penalty parameter (λ) for feature selection. Variables with non-zero coefficients underwent multivariable logistic regression to establish the nomogram prediction tool.

Analytical pipeline

The R software (version 4.2.2) was employed to develop the nomogram prediction model and the evaluation scale chart for this research. LASSO regression calculation was realized using the “glmnet” package. Multivariate logistic regression analysis was accomplished by using the “glm” function. Nomograms and calibration curve charts were generated by the “rms” package. The net benefit of the prediction model in this study was evaluated through the decision curve, which was plotted using the “rmda” package. The calculation of the Receiver Operating Characteristic curve (ROC) and the Area Under the Curve (AUC) was implemented by the “pROC” package.

Continuous data were presented as mean ± SD. Categorical variables were described using frequencies/percentages. T-tests analyzed continuous variables; chi-square/Fisher’s exact tests handled categorical variables. Statistical significance was defined as p < 0.05.

Results

Characteristics of the included patients

The final study population comprised 3374 participants (6748 knees) from 12 Hangzhou communities. Random 7:3 data partitioning yielded 4724 knees in the training set and 2024 in the validation set. The cohort had a mean age of 68.9 ± 8.4 years with 55.5% males (1874/3374). Radiographically confirmed KOA (KL grade ≥ 2) accounted for 78.4% (5296/6748) of cases. Baseline characteristic analysis revealed no statistically significant differences in demographic features, clinical indicators, or radiographic parameters between training and validation sets, confirming appropriate randomization (Table 1).

Table 1 Statistical description of the data set.
Full size table

Feature selection and model construction

Thirteen variables extracted from the cohort underwent LASSO regression. LASSO regression with tenfold cross-validation identified 13 predictors with non-zero coefficients (Fig. 1). Multivariable logistic regression ultimately identified 9 independent predictors (Table 2). MEP-related features included housework pain, squatting pain, stair climbing pain, and maximum pain. Treatment-related predictors encompassed oral NSAIDs, physical therapy, and intra-articular injections. Exercise habits and advancing age also emerged as independent factors.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Variable selection using the least absolute shrinkage and selection operator (LASSO) binary logistic regression model. (A)LASSO coefficient profiles of the 13 baseline features.(B) Tuning parameter (λ) selection in the LASSO model used tenfold cross-validation via minimum criteria variable selection.

Table 2 Logistics regression analysis of influencing factors of knee osteoarthritis.
Full size table

Model performance and clinical utility

Using logistic regression coefficients, we developed a nomogram for KOA prediction (Fig. 2). The model demonstrated AUC values of 0.889 (95% CI: 0.878–0.902) and 0.878 (95% CI: 0.859–0.898) in training and validation sets respectively (Fig. 3). Nonsignificant DeLong’s test (p = 0.327) suggested robust generalizability. CC demonstrated clinically acceptable agreement between predicted and observed probabilities (Fig. 4). DCA revealed that model application within 20–80% risk thresholds provided greater net clinical benefit compared to “treat-all” or “treat-none” strategies (Fig. 5). Hosmer–Lemeshow tests showed statistical significance in both training (χ2 = 919.97, p < 0.001) and validation sets (χ2 = 533.63, p < 0.001). Considering calibration curve patterns, we attribute these findings to enhanced test sensitivity from large sample sizes rather than clinically meaningful miscalibration.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Nomogram for predicting the incidence of KOA.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

The calculation of the receiver operating characteristic curve. AUC: Area Under the Curve.

Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

Calibration curve. (A)Training set. (B) Validation set.

Fig. 5
Fig. 5The alternative text for this image may have been generated using AI.
Full size image

decision curve analysis. (A)Training set. (B) Validation set.

Discussion

Our community-derived KOA prediction model represents the first diagnostic framework systematically integrating MEP as pivotal predictors. The model revealed that MEP-related features included housework pain, squatting pain, stair climbing pain, and maximum pain, treatment-related predictors encompassed oral NSAIDs, physical therapy, and intra-articular injections. Exercise habits and age also emerged as independent factors. For doctors and researchers, this might be a potentially useful tool for identifying the severity of KOA under simple conditions.

The imaging features and symptoms are both important but not the sole basis for diagnosing KOA12. Empirical evidence reveals a complex, non-linear relationship between patient-reported pain intensity and imaging-based disease severity13. While magnetic resonance imaging (MRI) outperforms conventional radiography in identifying early-stage structural pathologies of KOA14, the discordance between static imaging biomarkers and dynamic functional impairments could lead to diagnostic oversight of incipient pathological changes15.

The MEP test shows specificity and sensitivity in diagnosing KOA. Studies have revealed that in the natural course of KOA, pain related to the load is often one of the initial symptoms manifested16. The correlation between MEP and MRI indicates that they can capture unique nociceptive sensory pathways17,18. Heterogeneous contributions of joint pathologies to load-bearing vs resting pain 19,20, combined with load-reduction benefits21, reinforce mechanical stress-induced neuroinflammation as central to pain generation. This phenomenon may be related to the sensitization mechanisms of pain perception in the peripheral and central nervous systems, especially the imbalance of the pain regulation system mediated by abnormal serum β-endorphin levels22. Overall, MEP testing detects occult functional impairments and early sensitization undetectable at rest, offering novel biomarkers for early KOA identification 23. Previous studies have preliminarily verified the clinical application potential of MEP tests. Hensor et al.16 confirmed that knee flexion-related pain is the initial symptom manifestation of KOA. Naylor et al.24 conducted a six-minute walk test (6MWT) on patients awaiting knee or hip replacement surgeries. They found that the 6MWT exhibited the smallest measurement error among a series of tools typically used to assess disease severity, and thus could detect the smallest actual changes above the measurement error in daily clinical practice. Klokker’s dynamic assessment protocol (DAP)25, involving maximized knee flexion under time constraints, outperformed 6MWT in sensitivity metrics26.

Population analysis revealed exercise habits, age, and treatment modalities impact KOA. Our research suggests that certain exercise habits may act as risk factors for KOA. It should be noted that the impact of exercise on KOA remains controversial27. 2019 American College of Rheumatology/Arthritis Foundation Guideline (ACR) guidelines strongly recommend exercise as KOA prophylaxis, they emphasize needing research to identify knee-protective exercise modalities. The systematic review conducted by Migliorini et al.28 indicates that excessive exercise may lead to the early onset of KOA, while moderate exercise does not. However, there is no clear conclusion regarding whether appropriate exercise can prevent knee joint degeneration. Similar to Li et al.29, we identified aging as an independent risk factor, but we found no gender-risk association. This contradicts recent meta-analyses demonstrating female predisposition27. We speculate that this difference might be related to the confounding effect of the elderly cohort (68.9 ± 8.4 years old). Older age samples may weaken the significance of the gender effect, or reflect the interference effect of other confounding factors in the elderly population. For instance, the incidence of sarcopenia among older men may counterbalance the gender-specific risk30. The association between KOA risk and oral NSAIDs or physical therapy likely reflects “symptom-driven prescription bias”. Patients with severe pain are more likely to receive these treatments; therefore, the treatment itself does not necessarily cause disease progression.31. The observed protective effect of intra-articular injections aligns with their documented anti-inflammatory and chondroprotective properties32.

Over the past decade, KOA prediction modeling has evolved through technical advancements. The Nottingham study established community-based questionnaire models as methodological benchmarks8. Large longitudinal databases like OAI and CHARLS enabled multicenter external validation33,34. OAI-based breakthroughs first established MEP (particularly stair-climbing pain) as early OA markers, laying foundations for diagnostic research16. Integration of polygenic risk scores and molecular biomarkers has been pursued to enhance traditional clinical predictors35.The Rotterdam study advanced multidimensional integration but suggested limited biomarker-added value7. Losin et al.36 focused on established clinical factors to develop user-friendly primary care tools. Subsequent studies expanded to primary prevention models for high-risk groups, emphasizing clinical translation37. Contemporary innovations combine machine learning38 with biomarker-enhanced visual analytics39,40 to optimize clinical utility. Methodological progression has yielded models with enhanced scientific rigor, earlier detection capacity, and pragmatic clinical applicability. However, critical limitations persist. Although recognized as early markers, structured MEP tests remain absent from prognostic systems. Furthermore, there is insufficient evidence comparing biomarker-enhanced models against conventional clinical predictors35. Additionally, there is a scarcity of real-world validation in authentic community contexts. Addressing these requires biomechanically-informed dynamic assessment tools and implementable early intervention frameworks.

Strengths and limitations

Our study has several strengths. First, we included a large representative sample from Zhejiang communities, enhancing generalizability to Chinese populations. Good validation performance in community cohorts fills primary care application gaps, complementing OAI’s multicenter generalizability goals41. Second, we developed a practical KOA risk tool using clinically accessible variables, more suitable for resource-limited settings than complex biomarker-dependent models.

This study has the following limitations. Firstly, the key mechanical load indicator BMI was not included in this study. The potential collinearity between MEP and BMI might affect the discriminative efficacy of the model42. Nevertheless, by incorporating MEP, the model still partially captured the pathological features of mechanical stress overload. Second, KL ≥ 2 criteria may miss early , necessitating MRI and biomarker-enhanced diagnostic standards. Additionally, internal validation without external cohort testing limits generalizability conclusions.

Conclusion

In conclusion, in this study, we found that MEP test (housework pain, squatting pain, stair climbing pain and maximum pain), oral NSAIDs, physical therapy, intra-articular injections, exercise habits and age were predictors of KOA in community resident. Based on these predictors, we built a prediction nomogram for the early prediction of KOA, and our validation confirmed that this model was good. For each patient, higher total points reflected a greater risk of KOA. The visual and personalized model of predictors provides clinicians with a simple and intuitive tool for the early detection and identification of KOA, which may be of significance for decelerate, arrest, or potentially reverse pathological progression.