5772
Validation analysis of a composite real-world mortality endpoint for US cancer patients
Qianyi Zhang1, Anala Gossai1, Shirley Monroe1, Nathan C Nussbaum1, Christina M Parrinello1
1Flatiron Health, New York, NY, United States
Objective: In oncology research, mortality, as a variable, and overall survival (OS), as an efficacy endpoint, are critical. As real-world data gains use in supporting regulatory decisions, assessing the validity of mortality data is important. Mortality data derived solely from electronic health records (EHRs) has known gaps. We developed a mortality variable based on multiple sources of death data and benchmarked it against the National Death Index (NDI) data as a gold standard within 4 cancer types (Curtis et al, 2018). We refreshed and expanded validity assessment to 18 cancer types.

Methods: Patients diagnosed with at least 1 of 18 cancer types between 1/1/2011 and 12/31/2017 were included from the Flatiron Health EHR-derived de-identified US database. To develop a composite mortality variable, we amalgamated multiple sources of real-world death data, linking patient-level structured (EHR, commercial, Social Security Death Index) and unstructured data curated via technology-enabled abstraction. We calculated validity metrics (sensitivity, specificity, ±15-day accuracy) by benchmarking against the NDI data in each cancer type, as a composite variable, as well as by individual and combinations of death data sources. OS was estimated using the Kaplan-Meier method.

Results: There were 160,436 patients included in the study cohort. In the 18 cancer types, sensitivity ranged from 83.9 - 91.5% (17 out of 18 had sensitivity ≥85%), specificity ranged from 93.5 - 99.7%, and ±15-day accuracy ranged from 95.6 - 97.6%, compared to the NDI (Table). Median OS estimates for the composite mortality variable when compared to that from the NDI ranged from 3 - 13% greater across all cancer types.

Conclusions: We observed high sensitivity, specificity, and date accuracy of our composite mortality variable across all cancer types. Future research will investigate the validity of our mortality variable stratified by demographic and clinical characteristics to understand performance in subpopulations.

Cancer Type Note: some patients could have >1 cancer typeNSensitivity (%)Specificity (%)+/- 15-day Accuracy (%)
Solid Tumor
Breast Cancer (Early)166983.9 (77.4, 90.3)99.7 (99.5, 100)96.3 (92.7, 99.9)
Breast Cancer (Metastatic)1647388.3 (87.6, 88.9)97.7 (97.4, 98.0)96.9 (96.6, 97.3)
Colorectal Cancer (Metastatic)1723286.4 (85.7, 87.1)97.1 (96.8, 97.5)96.3 (95.9, 96.8)
Gastro-Esophageal Cancer (Advanced)716988.5 (87.6, 89.4)94.3 (93.3, 95.3)97.1 (96.6, 97.6)
Head and Neck Cancer (Advanced)527189.3 (88.2, 90.3)95.9 (95.0, 96.8)96.9 (96.3, 97.5)
Hepatocellular Carcinoma278485.0 (83.3, 86.7)95.7 (94.5, 96.9)95.6 (94.5, 96.6)
Malignant Pleural Mesothelioma170091.0 (89.4, 92.6)95.6 (93.7, 97.4)97.6 (96.7, 98.5)
Melanoma (Advanced)703189.5 (88.4, 90.6)98.7 (98.4, 99.1)97.4 (96.8, 98)
Non-small Cell Lung Cancer (Advanced)4507090.4 (90.1, 90.7)95.1 (94.8, 95.5)97.2 (97.0, 97.4)
Ovarian Cancer496486.8 (85.2, 88.3)98.7 (98.3, 99.1)96.8 (95.9, 97.6)
Pancreatic Cancer (Metastatic)545890.2 (89.3, 91.1)93.5 (92.2, 94.8)97.2 (96.7, 97.7)
Prostate Cancer (Metastatic)849589.0 (88.0, 90)98.7 (98.4, 99.0)97.1 (96.5, 97.7)
Renal Cell Carcinoma (Metastatic)577088.5 (87.4, 89.6)97.7 (97.1, 98.3)97.2 (96.6, 97.8)
Small Cell Lung Cancer472490.4 (89.4, 91.4)95.8 (94.7, 96.8)97.5 (97.0, 98.1)
Urothelial Cancer (Advanced)629390.2 (89.3, 91.1)96.6 (95.8, 97.4)97.5 (97.0, 98.0)
Liquid Tumor
Chronic Lymphocytic Leukemia903591.5 (90.4, 92.7)99.3 (99.1, 99.5)96.8 (96.1, 97.6)
Diffuse Large B-cell Lymphoma434489.1 (87.4, 90.8)99 (98.7, 99.4)96.1 (95.0, 97.3)
Multiple Myeloma780388.5 (87.3, 89.7)99.0 (98.8, 99.3)97.0 (96.3, 97.7)




Session: Descriptive Epidemiology, Record Linkage, and Big Data (Virtual Poster Session)
Category: EPIDEMIOLOGY