Optimal Indicator of Death for Using Real-World Cancer Patients' Data From the Healthcare System

Abstract

Background: Information on patient’s death is a major outcome of health-related research, but it is not always available in claim-based databases. Herein, we suggested the operational definition of death as an optimal indicator of real death and aim to examine its validity and application in patients with cancer.Materials and methods: Data of newly diagnosed patients with cancer between 2006 and 2015 from the Korean National Health Insurance Service—National Sample Cohort data were used. Death indicators were operationally defined as follows: 1) in-hospital death (the result of treatment or disease diagnosis code from claims data), or 2) case wherein there are no claims within 365 days of the last claim. We estimated true-positive rates (TPR) and false-positive rates (FPR) for real death and operational definition of death in patients with high-, middle-, and low-mortality cancers. Kaplan−Meier survival curves and log-rank tests were conducted to determine whether real death and operational definition of death rates were consistent.Results: A total of 40,970 patients with cancer were recruited for this study. Among them, 12,604 patients were officially reported as dead. These patients were stratified into high- (lung, liver, and pancreatic), middle- (stomach, skin, and kidney), and low- (thyroid) mortality groups consisting of 6,626 (death: 4,287), 7,282 (1,858), and 6,316 (93) patients, respectively. The TPR was 97.08% and the FPR was 0.98% in the high mortality group. In the case of the middle and low mortality groups, the TPR (FPR) was 95.86% (1.77%) and 97.85% (0.58%), respectively. The overall TPR and FPR were 96.68 and 1.27%. There was no significant difference between the real and operational definition of death in the log-rank test for all types of cancers except for thyroid cancer.Conclusion: Defining deaths operationally using in-hospital death data and periods after the last claim is a robust alternative to identifying mortality in patients with cancer. This optimal indicator of death will promote research using claim-based data lacking death information.

Impact of aging on acute myeloid leukemia epidemiology and survival outcomes: A real-world, population-based longitudinal cohort study

“…Another database-related limitation of our study is the accuracy of the mortality data in the claims database. Although claims data are known to have limitations in capturing mortality information because they only include hospital deaths, the death indicator of the NHIS has a true positive rate of 97% for high-risk cancers [ 44 ]. This study acknowledges a limitation regarding the assessment of the impact of the venetoclax plus HMA regimen on the elderly population.…”

Comparison of partitioned survival modeling with state transition modeling approaches with or without consideration of brain metastasis: a case study of Osimertinib versus pemetrexed-platinum

“…Second, the date of death was extracted using the operational definition described in the Methods section. However, we adapted an operational definition validated in a previous study on lung cancer patients in South Korea to minimize misclassification [26]. Third, long-term data were not available, as reimbursement for Osimertinib was effective from December 2017.…”

A nationwide analysis of the treatment patterns, survival, and medical costs in Korean patients with relapsed or refractory diffuse large B-cell lymphoma

“…Therefore, if mortality was only marked by the “in-hospital death” code, patient survival rates would be highly overestimated. To address this limitation, we defined “out-of-hospital” death as the date of the last claims filed for patients with no further records for 6 months, a method that has been adopted in previous research and validated in high-mortality cancer patients ( 20 , 21 ). We then assessed the survival outcomes from the first date of each line of treatment.…”