Related Experiment Videos
Validated methods for identifying tuberculosis patients in health administrative databases: systematic review
L A Ronald1, D I Ling2, J M FitzGerald3
1Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, Quebec, Division of Respiratory Medicine, Faculty of Medicine, University of British Columbia, Vancouver, Centre for Clinical Epidemiology and Evaluation, Vancouver Coastal Health Research Institute, Vancouver.
Summary
Health administrative databases are increasingly used for tuberculosis (TB) research, but methods for identifying TB patients vary widely. This review highlights significant limitations in diagnostic accuracy, especially when using International Classification of Diseases (ICD) codes alone.
Area of Science:
- Public Health
- Epidemiology
- Health Informatics
Background:
- Health administrative databases are increasingly utilized for tuberculosis (TB) research.
- However, significant limitations exist in accurately identifying TB patients within these databases.
Purpose of the Study:
- To systematically review and summarize validated methods for identifying TB cases in health administrative databases.
- To assess the diagnostic accuracy of algorithms used for TB identification.
Main Methods:
- A systematic literature search was conducted in Ovid Medline and Embase databases (January 1980-January 2016).
- Diagnostic accuracy studies evaluating algorithms based on drug prescription, International Classification of Diseases (ICD) codes, and/or laboratory data were included.
Main Results:
- 14 studies were included from 2413 unique citations.
- Algorithms for TB identification showed wide variability in diagnostic accuracy, with positive predictive values from 1.3% to 100% and sensitivity from 20% to 100%.
Conclusions:
- Diagnostic accuracy for identifying TB in health administrative databases varies considerably across studies.
- Sole reliance on ICD diagnostic codes, particularly for out-patient records, may lead to inaccurate TB case estimations due to current coding limitations.