Development and validation of the BPDKC model for predicting sarcopenic knee osteoarthritis in community-dwelling
Guang-Ming Liu1,2, Bo Sun2, Jia-Yu Yang2
1Shuguang Hospital Affiliated to Shanghai University of Traditional Chinese Medicine, Shanghai, 201203, China.
Background:
Sarcopenia and knee osteoarthritis frequently coexist as sarcopenic knee osteoarthritis (SPKOA), a clinically significant and burdensome condition among older adults. Despite its high prevalence and functional impact, no validated, practical tool exists for early identification of SPKOA in community settings. This study aimed to develop and internally validate a parsimonious, clinically applicable prediction model for SPKOA risk stratification in community-dwelling older adults.
Methods:
We conducted a cross-sectional analysis of 640 community-dwelling adults aged 65-85 years, recruited between March and October 2024 from the Yuyuan Community Health Service Center in Shanghai, China (median age: 71 years; 57% female). SPKOA was diagnosed using established criteria for both sarcopenia and knee osteoarthritis. Participants were randomly split (7:3) into training (n = 448) and internal test (n = 192) sets. In the training set, we employed a hybrid variable selection strategy: age, gender, BMI, physical activity (PA), diabetes mellitus (DM), Knee Visual Analog Scale(KVAS), and calf circumference (CC) were prespecified based on clinical relevance; remaining candidate variables underwent univariate screening (P < 0.05), followed by forward stepwise logistic regression (entry: P < 0.05; removal: P > 0.10). Three candidate models were derived and compared: Model 1 (7 predictors: CC, KVAS, BMI, DM, PA, osteoporosis, age); Model 2 - the BPDKC model (5 predictors: BMI, PA, DM, KVAS, CC); and Model 3 (5 predictors: DM, PA, KVAS, CC, age). Model performance was evaluated using discrimination (AUC with 95% CI), calibration (Brier score, calibration plots), clinical utility (decision curve analysis). The optimal model was translated into a nomogram and an interactive web-based calculator.
Results:
Among 640 participants, 12.5% (n = 80) met SPKOA criteria, with balanced prevalence across training (12.5%, n = 56) and test sets (12.5%, n = 24). In the training set, all models showed strong discrimination (Model 1 AUC = 0.953; Model 2 AUC = 0.951; Model 3 AUC = 0.934) and excellent calibration (Brier scores: 0.053-0.062). However, in the internal test set, Model 2 (BPDKC) demonstrated the best generalizability and balance of performance: AUC = 0.861 (95% CI: 0.795-0.926), specificity = 94.6% (95% CI: 90.1-97.5%), precision = 50.0% (95% CI: 26.0-74.0%), F1-score = 0.429 (95% CI: 0.243-0.618), and Brier score = 0.1038 (95% CI: 0.068-0.139). Although sensitivity remained modest (37.5%, 95% CI: 18.8-59.4%), calibration curves and decision curve analysis confirmed robust risk estimation and superior net clinical benefit across clinically relevant risk thresholds (10-40%). Model 2 also achieved the best trade-off between performance and parsimony (AIC = 178.41 vs. Model 3 AIC = 194.79). A user-friendly web calculator implementing the BPDKC model is freely accessible at ( https://lgmyyyy.shinyapps.io/bpdkc_spkoa_app/ ) .
Conclusion:
The BPDKC model is a simple, internally validated, and clinically practical tool for SPKOA risk prediction in community-dwelling older adults. Its high specificity and calibration support its utility in ruling out low-risk individuals, while its web-based implementation facilitates integration into primary care workflows. Although sensitivity remains limited-reflecting the challenge of case-finding in low-prevalence settings-the model offers meaningful clinical value for targeted screening and early intervention. External validation in diverse populations is recommended prior to broader implementation.
