Positive predictive value of an algorithm used for cancer surveillance in the U.S. Armed Forces

Recent large-scale epidemiologic studies of cancer incidence in the U.S. Armed Forces have used International Classification of Disease, 9th and 10th Revision (ICD-9 and ICD-10, respectively) diagnostic codes from administrative medical encounter data archived in the Defense Medical Surveillance System. Cancer cases are identified and captured according to an algorithm published by the Armed Forces Health Surveillance Branch. Standardized chart reviews were performed to provide a gold standard by which to validate the case definition algorithm. In a cohort of active component U.S. Air Force, Navy, and Marine Corps officers followed from 1 October 1995 through 31 December 2017, a total of 2,422 individuals contributed 3,104 algorithm-derived cancer cases. Of these cases, 2,108 (67.9%) were classified as confirmed cancers, 568 (18.3%) as confirmed not cancers, and 428 (13.8%) as unclear. The overall positive predictive value (PPV) of the algorithm was 78.8% (95% confidence interval [CI]: 77.2-80.3). For the 12 cancer sites with at least 50 cases identified by the algorithm, the PPV ranged from a high of 99.6% for breast and testicular cancers (95% CI: 97.8-100.0 and 97.7-100.0, respectively) to a low of 78.1% (95% CI: 71.3-83.9) for non-Hodgkin lymphoma. Of the 568 cases confirmed as not cancer, 527 (92.7%) occurred in individuals with at least 1 other confirmed cancer, suggesting algorithmic capture of metastases as additional primary cancers.