First Use of Multiple Imputation with the National Tuberculosis Surveillance System

Area of Science:

• Epidemiology
• Biostatistics
• Public Health Surveillance

Background:

• The National Tuberculosis Surveillance System (NTSS) dataset from the Centers for Disease Control and Prevention (CDC) has a high rate of missing human immunodeficiency virus (HIV) infection status.
• Missing data can introduce bias in epidemiological analyses, necessitating robust handling methods.

Purpose of the Study:

• To compare the efficacy of multiple imputation methods against case exclusion for handling missing covariate data in the NTSS.
• To assess the impact of increasing the number of imputed datasets on the estimated association between isoniazid resistance and death.

Main Methods:

• Multiple imputation techniques were employed to address missing HIV infection status.
• Regression analysis was performed using both multiple imputation and case exclusion (deleting subjects with missing data).
• The association between initial isoniazid resistance and death was evaluated under both methods.

Main Results:

• Multiple imputation yielded an odds ratio of 2.07 (95% CI 1.30, 3.29) for isoniazid resistance and death.
• Case exclusion resulted in a lower odds ratio of 1.53 (95% CI 0.83, 2.83).
• Using more than five imputed datasets did not significantly alter the study's findings.

Conclusions:

• Multiple imputation is a valuable method for epidemiological analysis, particularly with datasets like the NTSS.
• Careful consideration of missing covariate data and its potential impact is crucial throughout the analysis process.