Background

The global population of older adults was approximately 1 billion in 2019 and will increase to 1.4 billion by 2030 [1]. Ageing contributes to many chronic conditions, such as cognitive impairment and frailty, which have become increasingly significant public health problems [2, 3]. Frailty is influenced by multidomain factors, including age, sex, risk of malnutrition, and chronic diseases, as well as disability [4, 5]. Frailty and cognitive impairment interact in the ageing process, increasing the risk of adverse outcomes [6] such as dementia, disability, and mortality, but they have historically been studied separately. Consequently, the International Consensus Group from the International Academy of Nutrition and Aging (IANA) and the International Association of Gerontology and Geriatrics (IAGG) have proposed that cognitive frailty (CF) is a clinical condition characterized by the occurrence of both physical frailty and cognitive impairment and in the absence of dementia diagnosis [7]. CF may contribute to a higher risk of adverse outcomes than healthy older adults or those with physical frailty or cognitive impairment alone [8,9,10]. A meta-analysis indicated that the pooled prevalence of CF among community-dwelling older adults was 9% [11].

In order to identify individuals at high risk of CF and to facilitate the implementation of appropriate preventive measures and interventions [12,13,14], some prediction models have been developed [15,Candidate predictors

Candidate predictors were selected based on previous studies [24,25,26,27,28], medical knowledge and data available in the database. A total of 12 candidate predictors were chosen, and the detailed information is shown in Table 1. The physical disability was measured using the instrumental activity of daily living (IADL) and the basic activity of daily living (BADL), and it was defined if any item of IADL or BADL was judged as dependence [52]. Despite the lack of significant impact on CF, exercise remained a constituent of the final model. The predictors were computed following the AIC methodology and were informed by scientific insights, underscoring the pivotal role of exercise in CF [12, 52].

Strengths

Our study had some strengths. Firstly, the participants were representative because they were from a large population-based cohort. Secondly, the predictors included were non-invasive, low-cost, and easy to obtain, so the prediction model could be used in the primary care settings [53]. Thirdly, we performed external validation for portability and generalization, and the model displayed excellent discrimination and calibration in external validation. Lastly, we calculated the PAF to explore how modifiable risk factors contribute to CF.

Limitations

Our study also had some limitations. Firstly, most predictors were self-reported by older adults, potentially introducing information bias. Nevertheless, self-reported predictors were more easy-to-obtain and practical [54]. Secondly, some critical predictors were not included due to data limitations, such as depression [28], likely impacting the prediction performance. Thirdly, BMI, a part of the CF assessment, was selected as a predictor, which might introduce incorporation bias and optimist estimates of model performance [55]. However, the sensitivity analysis indicated that after removing the BMI, the prediction model maintained excellent performance. Therefore, the incorporation bias may have a negligible effect on the model performance.

Implications and clinical practice

This study presents some insights for future research. Firstly, the selection of predictive factors, encompassing easily accessible, non-invasive, and cost-effective variables, plays a pivotal role in prediction models applicable to clinical practice, especially within community healthcare and diverse clinical settings. Subsequent research endeavors should consider integrating addition predictive factors that share the accessibility, non-invasiveness, and cost-effectiveness criteria. Secondly, our study is grounded in the application of the logistic regression method for model development. Future research could explore alternative methodologies, such as machine learning techniques, to foster the evolution of predictive models. Lastly, the model's foundation is rooted in the Chinese population, prompting the necessity to examine its transferability to other demographics. This calls for comprehensive validation in diverse populations to establish its broader applicability in the future.

This study developed a prediction model for CF based on the characteristics of the Chinese population, utilizing practical, non-invasive, cost-effective, and easily obtainable variables. It can be applied in secondary prevention, enabling early identification, diagnosis, and treatment of CF. In tertiary disease prevention, the predictive model can be used to forecast recurrence, reduce mortality and disability [56]. Furthermore, it can provide community staffs with insights into the progression of CF in the older adults, allowing the identification of potential contributing factors for tailored preventive interventions [54].

Conclusion

The CF prediction model, following the TRIPOD statement, has been established and validated for older adults. It integrates six easily obtainable predictors and demonstrates excellent prediction performance. This model helps healthcare practitioners and nurses to identify older adults at a heightened risk of CF development over a six-year period and intervene proactively.