Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2020 Mar 4:2019:313-322.
eCollection 2019.

Predicting Adverse Drug Reactions on Distributed Health Data using Federated Learning

Olivia Choudhury  1 , Yoonyoung Park  1 , Theodoros Salonidis  2 , Aris Gkoulalas-Divanis  3 , Issa Sylla  1 , Amar K Das  1
Affiliations

Predicting Adverse Drug Reactions on Distributed Health Data using Federated Learning

Olivia Choudhury et al. AMIA Annu Symp Proc. .

Abstract

Using electronic health data to predict adverse drug reaction (ADR) incurs practical challenges, such as lack of adequate data from any single site for rare ADR detection, resource constraints on integrating data from multiple sources, and privacy concerns with creating a centralized database from person-specific, sensitive data. We introduce a federated learning framework that can learn a global ADR prediction model from distributed health data held locally at different sites. We propose two novel methods of local model aggregation to improve the predictive capability of the global model. Through comprehensive experimental evaluation using real-world health data from 1 million patients, we demonstrate the effectiveness of our proposed approach in achieving comparable performance to centralized learning and outperforming localized learning models for two types of ADRs. We also demonstrate that, for varying data distributions, our aggregation methods outperform state-of-the-art techniques, in terms of precision, recall, and accuracy.

PubMed Disclaimer

Figures

Figure 1:
Figure 1:
System design of federated learning for ADR prediction. Each site maintains electronic health records for a number of patients. Once a global model is shared with each site, it is trained on the site’s local data. Updates to the local models’ parameters are aggregated to improve the global model. This process is repeated until a convergence criterion for the global model is satisfied.
Figure 2:
Figure 2:
Comparison of precision score for centralized learning (CL), federated learning (FL), and localized learning (LL) models using SVM, perceptron, and logistic regression with (a) opioid data and (b) antipsychotic data.
Figure 3:
Figure 3:
Comparison of recall score for centralized learning (CL), federated learning (FL), and localized learning (LL) models using SVM, perceptron, and logistic regression with (a) opioid data and (b) antipsychotic data.
Figure 4:
Figure 4:
Comparison of accuracy score for centralized learning (CL), federated learning (FL), and localized learning (LL) models using SVM, perceptron, and logistic regression with (a) opioid data and (b) antipsychotic data.
Figure 5:
Figure 5:
Effect of varying number of sites on precision and recall scores of federated learning models (SVM, per- ceptron, logistic regression) with (a) opioid data and (b) antipsychotic data.
See this image and copyright information in PMC

References

    1. Tatonetti Nicholas P, Patrick P Ye, Daneshjou Roxana, Altman Russ B. Data-driven prediction of drug effects and interactions. Science translational medicine. 2012;4(125):125ra31–125ra31. - PMC - PubMed
    1. Jensen Peter B, Jensen Lars J, Brunak SoÃÿren. Mining electronic health records: Towards better research applications and clinical care. Nature Reviews Genetics. 2012;13:395–405. - PubMed
    1. Madigan David, Ryan Patrick B, Schuemie Martijn, Stang Paul E, Overhage J Marc, Hartzema Abraham G, Suchard Marc A, DuMouchel William, Jesse A Berlin. Evaluating the impact of database heterogeneity on observational study results. American journal of epidemiology. 2013;178(4):645–651. - PMC - PubMed
    1. FDA Adverse Event Reporting System (FAERS). 2019 Feb; https://www.fda.gov/drugs/informationondrugs/ucm135151.htm Accessed:
    1. The Sentinel Initiative. 2019 Feb; https://www.sentinelinitiative.org/ Accessed: