Characterization of temporal relationships of comorbidities developed following cancer diagnoses in veterans.

Conference Paper

e18049 Background: Understanding patient trajectories and common sequences of comorbidity accrual among those newly diagnosed with cancer is critical for precision approaches to care and prevention. The Veterans Affairs (VA) Cooperative Studies Program (CSP) #380 cohort includes 3,121 healthy asymptomatic veterans who underwent screening colonoscopy and were followed for at least ten years. The current analysis leverages computational approaches to characterize the temporal relationships of diagnoses in CSP #380 participants following diagnosis of colorectal or other cancers. Methods: Patients enrolled in CSP #380 with at least 5 years of linked electronic health record data from the VA Corporate Data Warehouse (October 1999-December 2015) were included. Cancer diagnoses and their most common subsequent new diagnoses were identified per patient by the first instance of each three-digit ICD-9 diagnosis affecting at least 50 patients. Pairwise chronological relative risks (RR) between subsequent diagnoses were represented as a directed network graph, which maps the probability of developing a diagnosis following a prior diagnosis. Results: A total of 2,210 patients were included. The most common cancer diagnoses were prostate (436), thoracic (169), bladder (120), colon (72), and kidney cancers (65). Most first diagnoses following a cancer diagnosis were related to progressive cancer or acute/subacute treatment toxicity. For prostate cancer, comorbidities with greatest RR were carcinoma in situ (RR 6.85), unspecified (NOS) metastases (2.75), and urethral stricture (2.53). For lung cancer, they were metastases of respiratory and digestive sites (12.24), lymph nodes (6.47), and NOS (5.68), pneumothorax and air leak (4.16), and convalescence and palliative care (3.07). In bladder cancer, they were carcinoma in situ (9.00), cystitis (6.78), kidney or other urinary cancer (6.19), attention to artificial openings (3.40), and urethral stricture (2.78). These and other results were visualized with network graphs. Conclusions: Computational techniques can identify and visualize future health concerns following cancer diagnoses. In this cohort of initially healthy and asymptomatic veterans on a prospective screening colonoscopy study, most subsequent diagnoses were related to cancer or toxicities of therapy, as might be expected in an aging cohort. Future work may focus on streamlining in-clinic identification of potential high likelihood comorbidities.

Full Text

Duke Authors

Cited Authors

  • Hong, JC; Hauser, ER; Redding, TS; Sims, KJ; Gellad, ZF; O'Leary, M; Madison, A; Qin, X; Weiss, DG; Bullard, AJ; Williams, CD; Sullivan, B; Lieberman, DA; Provenzale, DT

Published Date

  • May 20, 2019

Published In

Volume / Issue

  • 37 / 15_suppl

Start / End Page

  • e18049 - e18049

Published By

Electronic International Standard Serial Number (EISSN)

  • 1527-7755

International Standard Serial Number (ISSN)

  • 0732-183X

Digital Object Identifier (DOI)

  • 10.1200/jco.2019.37.15_suppl.e18049