[Validation of using disqualification from long-term care insurance registry as a proxy of death]

Objectives　The national database for long-term care insurance (LTCI) of Japan (Kaigo DB) enables researchers to access comprehensive data from its LTCI registry, eligibility assessment records, claims for service usage, and information about service providers. However, studies regarding the death or mortality of beneficiaries cannot be conducted because Kaigo DB does not contain death records, and researchers are not allowed to link Kaigo DB to other databases, such as national death records. Therefore, we aimed to assess the validity of using an insurer's disqualification from an LTCI beneficiary as a proxy of death.Methods　We used 510,751,798 monthly beneficiary records between April 2007 and March 2017 from the LTCI registry, while excluding data for ineligible persons for LTCI benefit or those younger than 65 years. We identified insurer cases disqualified from LTCI beneficiaries and linked them to national death records using deterministic linkage methods by dates of birth and death, sex, and residence. We considered the cases as positive if they were disqualified and their record was linked to a death. We used sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) as validity indices.Results　We identified 5,986,991 (1.17%) disqualified and 5,295,961 death cases. Sensitivity, specificity, PPV, and NPV of disqualification for death were 100%, 99.86%, 88.46%, and 100%, respectively. After stratification, PPV of disqualification was between 85% and 88% before 2012, 91% after 2012, 91.9% in men, and 85.9% in women. PPV increased with age (65-69 years: 80.6%, 70-74 years: 86.7%, 75-79 years: 86.4%, 80-84 years: 86.7%, 85-89 years: 88.0%, 90-94 years: 90.6%, and 95+ years: 93.4%) and level of care needed (support level: 72.2%, care level (CL) 1: 79.7%, CL2: 85.9%, CL3: 89.3%, CL4: 92.3%, and CL5: 94.0%).Conclusions　Disqualification from the LTCI registry is an inappropriate measure to estimate mortality accurately because it has a 10% false-positive rate. However, it appears sufficiently valid to use disqualification as a proxy outcome of death, although the main effect or confounding of a possible predictor of death could be slightly underestimated.