A Novel Imputation Approach for Sharing Protected Public Health Data

Area of Science:

• Public Health
• Data Science
• Biostatistics

Background:

• Government administrative data, crucial for public health surveillance, often contains suppressed values due to low counts.
• Accurate data sharing and statistical analyses are hindered by these suppressed values, limiting research and policy development.

Purpose of the Study:

• To develop and evaluate an imputation method for estimating suppressed values in government administrative data.
• The goal is to facilitate accurate data sharing and improve statistical and spatial analyses for public health surveillance.

Main Methods:

• A novel imputation approach was developed using Massachusetts opioid prescription data (2011-2017).
• The method combines modified previous/next substitution, mean imputation, and count adjustment to estimate suppressed values.
• Four imputation methods were modeled and compared against baseline mean imputation.

Main Results:

• The developed imputation method significantly outperformed simple mean imputation.
• It retained 46% of the suppressed value's proportional variance, with 22% lower RMSE and 26% lower MAE.
• The technique effectively addresses adverse effects of low count suppression.

Conclusions:

• The proposed imputation technique is easy to implement and superior to mean imputation for handling suppressed public health data.
• This novel method is generalizable for researchers sharing protected public health surveillance data.
• It enhances the utility of administrative datasets for critical public health research.