[Development of a risk prediction model and a sample risk chart for long-term care certification based on the...

Objectives　The first aim of this study was to develop risk prediction models based on age, sex, and functional health to estimate the absolute risk of the 3-year incidence of long-term care certification and to evaluate its performance. The second aim was to produce risk charts showing the probability of the incident long-term care certification as a tool for prompting older adults to engage in healthy behaviors.Methods　This study's data was obtained from older adults, aged ≥65 years, without any disability (i.e., they did not certify≥care level 1) and residing in Yabu, Hyogo Prefecture, Japan (n=5,964). A risk prediction model was developed using a logistic regression model that incorporated age and the Kihon Checklist (KCL) score or the Kaigo-Yobo Checklist (KYCL) score for each sex. The 3-year absolute risk of incidence of the long-term care certification (here defined as≥care level 1) was then calculated. We evaluated the model's discrimination and calibration abilities using the area under the receiver operating characteristic curves (AUC) and the Hosmer-Lemeshow goodness-of-fit test, respectively. For internal validity, the mean AUC was calculated using a 5-fold cross-validation method.Results　After excluding participants with missing KCL (n=4) or KYCL (n=1,516) data, we included 5,960 for the KCL analysis and 4,448 for the KYCL analysis. We identified incident long-term care certification for men and women during the follow-up period: 207 (8.2%) and 390 (11.3%) for KCL analysis and 128 (6.6%) and 256 (10.2%) for KYCL analysis, respectively. For calibration, the χ² statistic for the risk prediction model using KCL and KYCL was: P=0.26 and P=0.44 in men and P=0.75 and P=0.20 in women, respectively. The AUC (mean AUC) in the KCL model was 0.86 (0.86) in men and 0.83 (0.83) in women. In the KYCL model, the AUC was 0.86 (0.85) in men and 0.85 (0.85) in women. The risk charts had six different colors, suggesting the predicted probability of incident long-term care certification.Conclusions　The risk prediction model demonstrated good discrimination, calibration, and internal validity. The risk charts proposed in our study are easy to use and may help older adults in recognizing their disability risk. These charts may also support health promotion activities by facilitating the assessment and modification of the daily behaviors of older adults in community settings. Further studies with larger sample size and external validity verification are needed to promote the widespread use of risk charts.