Development of a frailty screening tool using components from clinical parameters and frailty criteria for Thai community-dwelling older adults
Issued Date
2026-12-01
Resource Type
eISSN
14712318
Scopus ID
2-s2.0-105038126240
Journal Title
BMC Geriatrics
Volume
26
Issue
1
Rights Holder(s)
SCOPUS
Bibliographic Citation
BMC Geriatrics Vol.26 No.1 (2026)
Suggested Citation
Sutiwisesak R., Korsirikoon C., Khamrangsee S., Techaniyom P., Chitta P., Pratumvinit B., Dawangpa A., Wongloet W., Kaewboonruang W., Assantachai P., Muangpaisan W., Maneesing T.U., Supapueng O., Soongsathitanon J., Dharakul T., Boonnak K., Borkowski J.J., Sae-Lee C. Development of a frailty screening tool using components from clinical parameters and frailty criteria for Thai community-dwelling older adults. BMC Geriatrics Vol.26 No.1 (2026). doi:10.1186/s12877-026-07411-z Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/116752
Title
Development of a frailty screening tool using components from clinical parameters and frailty criteria for Thai community-dwelling older adults
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Background: Comprehensive frailty assessment is crucial for longitudinal care in older adults but is often impractical in community, rural, or understaffed settings. This study aimed to determine key identifiers of frailty from different assessment tools and use them to develop a screening tool for frailty status using data from community-dwelling older adults in Bangkok, Thailand. Methods: This cross-sectional study recruited 848 community-dwelling participants aged above 60 years during October 2022 and November 2023. Comprehensive data, including demographics, socioeconomic status, physical health, lifestyle, and laboratory results were collected. Frailty status was determined using Fried Frailty Phenotype (FFP), the FRAIL scale, Kihon Checklist (KCL), and Thai Frailty Index (TFI). A decision tree classification model was developed using an exhaustive CHAID algorithm, with variables selected via initial screening for multicollinearity and multinomial logistic regression. The dataset was split into training (80%) and testing (20%) sets for validation. Results: Frailty status varied across assessment tools and showed some discrepancies. The final three-layer decision tree using five key predictors: 4-meter gait speed lower than 1 m/s, unintentional weight loss in one year more than 5%, experiencing difficulties in chewing hard/solid food, KCL1 (Do you go out by bus or train by yourself?), and KCL25 (In the last 2 weeks, have you felt tired without a reason?). Gait speed was identified as the most important predictor. The model achieved an overall accuracy of 0.751 on the training data and 0.776 on the testing data. It demonstrated strong performance in identifying robust individuals, with a correct classification rate (recall) of 1.000 in the testing dataset, and 0.603 and 0.731, for pre-frail and frail individuals respectively, in the testing dataset. The area under the receiver operating characteristic curve (AUC) values were 0.896 for robust, 0.740 for pre-frail, and 0.813 for frail classifications. For practicality, the decision tree can be collapsed down to only 3 questions to identify “non-robust” individuals for comprehensive frailty assessment referral. Conclusions: Our study developed a practical and scalable data-driven decision tree for rapid community-based frailty identification. By combining a simple physical test (gait speed), percentage of weight loss, and a specific KCL item, this tool offers a cost-effective method to classify frailty status, particularly robust individuals. Its simplicity makes it suitable for use even by less experienced staff, supporting early identification and referral of at-risk older adults who would benefit from timely intervention.