Overestimation of clinical diagnostic performance caused by low necropsy rates
- 1Department of Medicine, University of California San Francisco, CA, USA
- 2Department of Pathology and Laboratory Medicine, Baylor Health Care System, USA
- 3Center for Primary Care and Outcomes Research, Stanford University, CA, USA
- Correspondence to: Dr K G Shojania The Ottawa Hospital - Civic Campus, 1053 Carling Avenue, Room C403, Box 693, Ottawa, ON, Canada K1Y 4E9;
- Accepted 31 July 2005
Background: Diagnostic sensitivity is calculated as the number of correct diagnoses divided by the sum of correct diagnoses plus the number of missed or false negative diagnoses. Because missed diagnoses are generally detected during clinical follow up or at necropsy, the low necropsy rates seen in current practice may result in overestimates of diagnostic performance. Using three target conditions (aortic dissection, pulmonary embolism, and active tuberculosis), the prevalence of clinically missed cases among necropsied and non-necropsied deaths was estimated and the impact of low necropsy rates on the apparent sensitivity of antemortem diagnosis determined.
Methods: After reviewing case series for each target condition, the most recent study that included cases first detected at necropsy was selected and the reported sensitivity of clinical diagnosis adjusted by estimating the total number of cases that would have been detected had all decedents undergone necropsy. These estimates were based on available data for necropsy rates, time period, country (US v non-US), and case mix.
Results: For all three target diagnoses, adjusting for the estimated prevalence of clinically missed cases among non-necropsied deaths produced sensitivity values outside the 95% confidence interval for the originally reported values, and well below sensitivities reported for the diagnostic tests that are usually used to detect these conditions. For active tuberculosis the sensitivity of antemortem diagnosis decreased from an apparent value of 96% to a corrected value of 83%, with a plausible range of 42–91%; for aortic dissection the sensitivity decreased from 86% to 74%; and for pulmonary embolism the reduction fell only modestly from 97% to 91% but was still lower than generally reported values of 98% or more.
Conclusions: Failure to adjust for the prevalence of missed cases among non-necropsied deaths may substantially overstate the performance of diagnostic tests and antemortem diagnosis in general, especially for conditions with high early case fatality.
This article is based in part on work performed by the UCSF-Stanford Evidence-based Practice Center under contract to the Agency for Healthcare Research and Quality (Contract No. 290-970013), Rockville, MD. The authors are responsible for the contents of this article. No statement in this article should be construed as an official position of the Agency for Healthcare Research and Quality or of the US Department of Health and Human Services.
Dr Shojania holds a Canada Research Chair in Patient Safety and Quality Improvement.