Development and validation of PRE-FRA (PREdiction of FRAilty Risk in community older adults) frailty prediction model

Taiping Lin¹Xiaotao Huang²Xiang Wang³Miao Dai^3*Jirong Yue^1*

• ¹National Clinical Research Center for Geriatric Diseases, West China Hospital, Sichuan University, Chengdu, Sichuan Province, China
• ²Jiangyou 903 Hospital, Mianyang, China
• ³Jiujiang First People's Hospital, Jiujiang, China

As the global population ages, identifying older adults at risk of frailty becomes increasingly important for targeted interventions. This study aimed to develop and validate a 1-year frailty onset prediction model for initially non-frailty or pre-frailty, community-dwelling older adults.We enrolled 1,079 community-dwelling older adults aged >60 years without baseline frailty (i.e., non-frailty or pre-frailty) for the development cohort. Lasso regression was used to screen potential predictors. Subsequently, logistic regression analysis was conducted to create a nomogram, which was internally validated using 500 bootstrap resamples. Additionally, temporal validation was performed to ensure the model's generalizability. This validation involved an external cohort of 481 older adults, all aged over 60 years and without frailty at baseline. Discrimination was assessed using the area under the receiver operating characteristic curve (AUROC), and calibration was evaluated with calibration plots.: In the development cohort, we enrolled 1,079 older adults with a median age of 68.00 years (interquartile range: 64.00 -72.00), including 673 females. Over a 1-year follow-up, 73 cases of frailty were identified. Key predictors identified by the model included age, history of falls within the past month, coughing while drinking water, pre-frailtyty status, cognitive impairment, 5-time chair stand test, and calf circumference. The developed model exhibited favorable discriminative ability in the development cohort (AUROC = 0.81, 95% confidence interval 0.76 -0.87). Internal validation through bootstrapping yielded consistent results (AUROC = 0.80), while temporal validation confirmed its robustness (AUROC = 0.73). Calibration plots demonstrated favorable agreement in both the development and temporal validation cohorts. To enhance usability, an online web-based calculator was developed (accessible at: https://frailtyriskprediction.shinyapps.io/dynnomapp/). The model showed high sensitivity (0.92) for frailty exclusion at a 2.5% threshold and specificity (0.89) for frailty identification at a 15% threshold. Conclusions: This 1-year frailty onset prediction model for initially non-frailty or pre-frailty older adults integrates accessible variables and demonstrates robust validation. It aids clinical decision-making by identifying high-risk individuals for early intervention.