Linear mixed modelling of federated data when only the mean, covariance, and sample size are available

TL;DR

This paper introduces a novel method for fitting linear mixed models using only summary statistics from multiple data sources, preserving privacy and reducing communication compared to federated learning.

Contribution

The proposed framework achieves equivalent estimates to individual-level data analysis using only mean, covariance, and sample size, simplifying federated data analysis.

Findings

Achieves identical estimates to individual data analysis.

Requires only one round of summary statistic sharing.

Demonstrated on real patient data from 70 clinics.

Abstract

In medical research, individual-level patient data provide invaluable information, but the patients' right to confidentiality remains of utmost priority. This poses a huge challenge when estimating statistical models such as linear mixed models, which is an extension of linear regression models that can account for potential heterogeneity whenever data come from different data providers. Federated learning algorithms tackle this hurdle by estimating parameters without retrieving individual-level data. Instead, iterative communication of parameter estimate updates between the data providers and analyst is required. In this paper, we propose an alternative framework to federated learning algorithms for fitting linear mixed models. Specifically, our approach only requires the mean, covariance, and sample size of multiple covariates from different data providers once. Using the principle of statistical sufficiency within the framework of likelihood as theoretical support, this proposed framework achieves estimates identical to those derived from actual individual-level data. We demonstrate this approach through real data on 15 068 patient records from 70 clinics at the Children's Hospital of Pennsylvania (CHOP). Assuming that each clinic only shares summary statistics once, we model the COVID-19 PCR test cycle threshold as a function of patient information. Simplicity, communication efficiency, and wider scope of implementation in any statistical software distinguish our approach from existing strategies in the literature.