Factors associated with frailty status compared to pre-frailty in community-dwelling older adults: a cross-sectional...

To identify factors distinguishing frail from pre-frail status in community-dwelling older adults and construct a risk prediction model with logistic regression as the primary method to identify independent risk factors, and a neural network as a supplementary approach to explore complex relationships.

The FRAIL Frailty Screening Scale was used to screen adults aged 65 and above meeting inclusion criteria for pre-frailty and frailty. A cross-sectional survey collected basic information and disease status, while scales assessed nutritional risk, sarcopenia, and activities of daily living (ADL), alongside physical measurements. Binary logistic regression was used as the primary method to identify factors associated with frailty status. A multi-layer perceptron neural network was employed secondarily to explore complex, non-linear associative patterns.

Of 1,451 participants, 46.0% were pre-frailty, and 54.0% were frail. Age, education level, smoking status, number of chronic diseases, SNAQ nutritional risk, SARC-F sarcopenia, and ADL disability were independent associated with frailty status, with SARC-F sarcopenia showing the strongest association with frailty status. The neural network model identified SARC-F sarcopenia (100.0%), waist circumference (67.0%), and age (62.8%) as the most influential factors.

Our findings identify several factors strongly associated with frailty status in a cross-sectional sample. While these factors represent potential targets for intervention, future longitudinal studies are needed to confirm their predictive value for frailty progression. The logistic regression model provides clinically interpretable factors associated with frailty (e.g., SARC-F sarcopenia, ADL disability, nutritional risk). The neural network analysis corroborated the paramount importance of sarcopenia and highlighted additional non-linear associations. Community decisions should be primarily based on the interpretable outputs of the logistic regression model. Given the cross-sectional design, these findings represent associations at a single time point.