Related Experiment Videos
Dealing with missing data in observational health care outcome analyses
C M Norris1, W A Ghali, M L Knudtson
1The Alberta Provincial Project for Outcome Assessment in Coronary Heart Disease "Approach," 8111 1st Floor ABC, 8440-112 Street, Edmonton, Alberta, Canada. cnorris@approach.org
Journal of Clinical Epidemiology
|April 29, 2000
Summary
Merging clinical and administrative databases improves observational outcome analyses when dealing with missing data. This data enhancement method offers a more valid approach than exclusion or assumption strategies for research validity.
Area of Science:
- Health Research
- Biostatistics
- Data Science
Background:
- Observational outcome analyses are common in health research.
- Clinical registries are preferred over administrative databases for these analyses.
- Missing data in clinical registries threaten the validity of observational studies.
Purpose of the Study:
- To compare methods for handling missing data in clinical registries.
- To evaluate the effectiveness of merging clinical and administrative databases.
Main Methods:
- Compared three approaches: data exclusion, assuming absence of risk, and merging databases.
- Developed a predictive model using merged data.
- Assessed model performance using C statistic and decile-of-risk analysis.
Main Results:
- The merged data model achieved a higher C statistic (0.770).
- The enhanced model demonstrated better goodness-of-fit and a larger risk gradient across deciles (46.3).
- A significant decrease in deviance (-2 log likelihood = 406.2) was observed with the merged data.
Conclusions:
- Merging clinical and administrative databases (data enhancement) is a superior method for handling nonrandom missing data.
- This methodology improves the validity and performance of observational outcome analyses.
- Researchers should consider data enhancement when faced with missing data challenges.