Data Privacy and Specimen Pooling: Using an old tool for New Challenges

TL;DR

This paper demonstrates that specimen pooling, an epidemiologic tool, can be adapted as a privacy-preserving method to analyze confidential data in case-control studies, maintaining data utility while protecting individual privacy.

Contribution

The study introduces a novel application of specimen pooling for privacy-preserving data analysis in epidemiology, enabling estimation of odds ratios from aggregate data.

Findings

Parameter estimates from pooled data closely match individual data estimates.

Pooling maintains similar standard errors and confidence intervals.

Method is effective for both matched and unmatched case-control designs.

Abstract

Background: In the context of ongoing debate over data confidentiality versus shared use of research data, as raised following the new EU General Data Protection Regulation, we seek to find alternate techniques that can balance these two issues. In particular, we demonstrate that an existing epidemiologic tool, specimen pooling, can be adapted as a privacy-preserving method to enable data analysis while maintaining data confidentiality. Specimen pooling is a cost-effective tool in studying the effect of an expensive-to-measure exposure on a disease outcome, for both unmatched and matched case-control designs. We propose the technique in a new context to analyze confidential data and demonstrate that it can be successfully used to estimate OR of covariates based on aggregate data when individual patient data cannot be shared. Methods: We demonstrate the application of specimen pooling based on aggregate covariate level and show that aggregated covariate levels can be used in a conditional logistic regression model to estimate individual-level odds ratio parameters of interest. We then show how to adapt the technique as a privacy-preserving method to analyze data from a matched case-control design. A similar approach can be applied for an unmatched design and unconditional logistic regression. Results: The parameter estimates from the standard conditional logistic regression are compared to those based on aggregated data. The parameter estimates are similar and have similar standard errors and confidence interval coverage. Conclusions: Pooling can be used effectively to analyze confidential data arising from distributed data networks and will be extremely useful in pharmacoepidemiology.