Premium
Concordance between administrative data and clinical review for mortality in the randomized on/off bypass follow‐up study (ROOBY‐FS)
Author(s) -
Quin Jacquelyn A.,
Hattler Brack,
Shroyer Annie Laurie W.,
Kemp Darlene,
Almassi G Hossein,
Bakaeen Faisal G.,
Carr Brendan M.,
Bishawi Muath,
Collins Joseph F.,
Grover Frederick L.,
Wagner Todd H.
Publication year - 2017
Publication title -
journal of cardiac surgery
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.428
H-Index - 58
eISSN - 1540-8191
pISSN - 0886-0440
DOI - 10.1111/jocs.13379
Subject(s) - medicine , concordance , cause of death , randomized controlled trial , national death index , sudden cardiac death , veterans affairs , medical record , clinical trial , emergency medicine , cardiac surgery , cohen's kappa , confidence interval , hazard ratio , disease , machine learning , computer science
Background The optimal methodology to identify cardiac versus non‐cardiac cause of death following cardiac surgery has not been determined. Methods The Randomized On/Off Bypass Trial was a multicenter, randomized, controlled clinical trial of 2203 patients (February 2002‐May 2008) comparing 1‐year cardiac outcomes between off‐pump and on‐pump bypass surgery. In 2013, the Veterans Affairs (VA) Cooperative Studies Program funded a follow‐up study to assess 5‐year outcomes including mortality. Deaths were identified and confirmed using the National Death Index (NDI), VA Vital Status file, and medical records. An Endpoints Committee (EC) reviewed patient medical records and classified each cause of death as cardiac, non‐cardiac, or unknown. Using pre‐determined ICD‐10 codes, NDI death certificates were independently used to classify deaths as cardiac or non‐cardiac. Cause of death was compared between the NDI and EC classifications and concordance measured, using Kappa statistics. Results Of the 297 5‐year deaths identified by the NDI and/or VA vital status file and confirmed by the EC, 219 had adequate patient records for EC cause of death determination. The EC adjudicated 141 of these deaths as non‐cardiac and 78 as cardiac, while the NDI classified 150 as non‐cardiac and 69 as cardiac; agreement was 77.6% (kappa 0.500; P < 0.001). Conclusions Since concordance between EC and NDI cause of death classifications was only moderate, caution should be exercised in relying exclusively on NDI data to determine cause of death. A hybrid approach, integrating multiple information sources, may provide the most accurate approach to classifying cause of death.