In the present study, we investigated that whether the KCL can be used to predict incidence of LTCI certification (care level 1 or higher) in the short term. Eight KCL items that were strongly associated with LTCI certification included three items from the instrumental activities of daily living domain, one item from the physical strength domain, one item from the nutrition domain, one item from the memory domain, and two items from the mood domain. The AUC of these eight items with respect to incidence of LTCI certification was 0.93, which was as high as that of the main 20 items and all 25 items. This result suggests that the eight items sufficiently predicted short-term incidence of LTCI certification.
There was likely no significant difference in the age distribution of the participants in the present study and overall elderly population in Japan because we used large-scale data from complete enumeration obtained by the local government.16 Moreover, the proportion of participants with LTCI certification (support level 1–2) at baseline was comparable to that of the whole Japanese population;16, 17 thus, the data in the present study were representative of community-dwelling Japanese elderly. Tomata9 reported that the 2.2% of individuals required new LTCI certification (care level 1 or higher) during a one-year follow-up, versus 0.5% during a three-month follow-up in the present study; the latter value might have been affected by the shorter follow-up period.
Tomata reported that the items of the KCL that were associated with 1-year incidence of LTCI certification (support level 1 or higher), through forced-entry logistic regression analysis, were two items from the instrumental activities of daily living domain, two items from the physical strength domain, one item from the nutrition domain, one item from the isolation domain, and two items from the cognitive function domain.9 In addition to these items, items from the mood domain were selected in the present study. Fukutomi reported no association between the mood domain and two-year incidence of LTCI certification (support level 1 or higher).11 However, Hamazaki reported a significant association between the mood domain and two-year incidence of LTCI certification (care level 2 or higher).10 Therefore, items of the mood domain might be associated with short-term and more severe need for LTCI certification.
The utility of KCL items for predicting incidence of LTCI certification in the present study was notably greater than that reported in previous studies: The AUC in ROC analysis was 0.62–0.83 in Tomata et al.9 and 0.78 in Tsuji et al.13 The present study focused on the need for LTCI certification at care level 1 or higher over a shorter follow-up period than that in previous studies; the results suggested that the KCL was useful for predicting incidence of LTCI certification in the short term. Moreover, we found that the eight selected items were as efficacious for predicting incidence of LTCI certification as were the main 20 items and all 25 items. This result was also confirmed by a five-fold cross validation test.
The specificity and sensitivity for incidence of LTCI certification based on the selection criteria of the LTCI for people at high risk were 57.8% and 73.5%, respectively, according to national research in Japan, and 63.4% and 78.1%, respectively, in Tomata et al.9 Moreover, specificity and sensitivity were 73.1% and 70.5%, respectively in a previous study that used part of the KCL.13 Specificity and sensitivity in the present study were both more than 80%, values higher than those reported in previous studies.
In the present study, sensitivity using the eight selected items was highest at 92.6%, using a cut-off point of ≥ 3/8 items; however, specificity was highest at 91.3%, using a cut-off point of ≥ 4/8 items. These results suggest that it might be useful to change the cut-off point of the eight selected items according to various situations. The ≥ 3/8 cut-off point might be recommended if fewer false negatives are required during primary screening, while the ≥ 4/8 cut-off point might be recommended if fewer false positives are required during medical examinations.
From the results of the present study, if the local government did not determine LTCI certification at Municipal Certification Committees for service applicants with support level 1–2 based on the KCL, this would be expected to reduce social security costs by 6 billion Yen per year, given a cost of 10,000 Yen per person to determine LTCI certification at the Municipal Certification Committee.7,18
The present study had several limitations. We used data that were collected by the local government to screen for elderly persons with frailty; however, the response rate was 66.8%. Thus, the study population might not have included people at higher risk of incidence of LTCI certification. Because not all participants eligible for LTCI certification actually apply for LTCI certification, the present study might have included detection bias. Since we did not investigate disease and functional ability, how health status affected LTCI certification was unclear.
In conclusion, the present study showed that the main 20 items and all 25 items could be used to predict incidence of LTCI certification during a subsequent 3-month period with high accuracy. Moreover, eight selected items of the KCL also could be used to make this prediction. Therefore, the items of the KCL that are included could be changed according to various situations when predicting short-term incidence of LTCI certification.