More Like this

1. Data heterogeneity in federated learning with Electronic Health Records: Case studies of risk prediction for acute kidney injury and sepsis diseases in critical care

   https://doi.org/10.1371/journal.pdig.0000117  

   Rajendran, Suraj ; Xu, Zhenxing ; Pan, Weishen ; Ghosh, Arnab ; Wang, Fei ( March 2023 , PLOS Digital Health)

   Frasch, Martin G. (Ed.)

   With the wider availability of healthcare data such as Electronic Health Records (EHR), more and more data-driven based approaches have been proposed to improve the quality-of-care delivery. Predictive modeling, which aims at building computational models for predicting clinical risk, is a popular research topic in healthcare analytics. However, concerns about privacy of healthcare data may hinder the development of effective predictive models that are generalizable because this often requires rich diverse data from multiple clinical institutions. Recently, federated learning (FL) has demonstrated promise in addressing this concern. However, data heterogeneity from different local participating sites may affect prediction performance of federated models. Due to acute kidney injury (AKI) and sepsis’ high prevalence among patients admitted to intensive care units (ICU), the early prediction of these conditions based on AI is an important topic in critical care medicine. In this study, we take AKI and sepsis onset risk prediction in ICU as two examples to explore the impact of data heterogeneity in the FL framework as well as compare performances across frameworks. We built predictive models based on local, pooled, and FL frameworks using EHR data across multiple hospitals. The local framework only used data from each site itself. The pooled framework combined data from all sites. In the FL framework, each local site did not have access to other sites’ data. A model was updated locally, and its parameters were shared to a central aggregator, which was used to update the federated model’s parameters and then subsequently, shared with each site. We found models built within a FL framework outperformed local counterparts. Then, we analyzed variable importance discrepancies across sites and frameworks. Finally, we explored potential sources of the heterogeneity within the EHR data. The different distributions of demographic profiles, medication use, and site information contributed to data heterogeneity. 

   more » « less
2. Detection of COVID-19 features in lung ultrasound images using deep neural networks

   https://doi.org/10.1038/s43856-024-00463-5  

   Zhao, Lingyi ; Fong, Tiffany Clair ; Bell, Muyinatu A. Lediju ( March 2024 , Communications Medicine)

   Deep neural networks (DNNs) to detect COVID-19 features in lung ultrasound B-mode images have primarily relied on either in vivo or simulated images as training data. However, in vivo images suffer from limited access to required manual labeling of thousands of training image examples, and simulated images can suffer from poor generalizability to in vivo images due to domain differences. We address these limitations and identify the best training strategy.

   We investigated in vivo COVID-19 feature detection with DNNs trained on our carefully simulated datasets (40,000 images), publicly available in vivo datasets (174 images), in vivo datasets curated by our team (958 images), and a combination of simulated and internal or external in vivo datasets. Seven DNN training strategies were tested on in vivo B-mode images from COVID-19 patients.

   Here, we show that Dice similarity coefficients (DSCs) between ground truth and DNN predictions are maximized when simulated data are mixed with external in vivo data and tested on internal in vivo data (i.e., 0.482 ± 0.211), compared with using only simulated B-mode image training data (i.e., 0.464 ± 0.230) or only external in vivo B-mode training data (i.e., 0.407 ± 0.177). Additional maximization is achieved when a separate subset of the internal in vivo B-mode images are included in the training dataset, with the greatest maximization of DSC (and minimization of required training time, or epochs) obtained after mixing simulated data with internal and external in vivo data during training, then testing on the held-out subset of the internal in vivo dataset (i.e., 0.735 ± 0.187).

   DNNs trained with simulated and in vivo data are promising alternatives to training with only real or only simulated data when segmenting in vivo COVID-19 lung ultrasound features.

    

   more » « less
3. Understanding Clipping for Federated Learning: Convergence and Client-Level Differential Privacy

   Xinwei Zhang ; Xiangyi Chen ; Mingyi Hong ; Zhiwei Steven Wu ; Jinfeng Yi ( July 2022 , International Conference on Machine Learning)

   Providing privacy protection has been one of the primary motivations of Federated Learning (FL). Recently, there has been a line of work on incorporating the formal privacy notion of differential privacy with FL. To guarantee the client-level differential privacy in FL algorithms, the clients’ transmitted model updates have to be clipped before adding privacy noise. Such clipping operation is substantially different from its counterpart of gradient clipping in the centralized differentially private SGD and has not been well-understood. In this paper, we first empirically demonstrate that the clipped FedAvg can perform surprisingly well even with substantial data heterogeneity when training neural networks, which is partly because the clients’ updates become similar for several popular deep architectures. Based on this key observation, we provide the convergence analysis of a differential private (DP) FedAvg algorithm and highlight the relationship between clipping bias and the distribution of the clients’ updates. To the best of our knowledge, this is the first work that rigorously investigates theoretical and empirical issues regarding the clipping operation in FL algorithms. 

   more » « less
4. Guiding Federated Learning with Inferenced Formal Logic Properties

   https://doi.org/10.1145/3576841.3589633  

   An, Ziyan ; Ma, Meiyi ( May 2023 , Proceedings of the ACM/IEEE 14th International Conference on Cyber-Physical Systems)

   Recent progressions in federated learning (FL) have facilitated the development of decentralized collaborative Internet-of-Things (IoT) applications. However, data-driven FL algorithms face the challenge of heterogeneity in participating IoT devices, including their deployment environment and calibration settings. Fail to follow these device-specific properties can degenerate the model performance. To address this issue, we present FedSTL in this poster abstract, which is a two-staged personalized FL framework with clustering for sequential prediction tasks in IoT. FedSTL first identifies client properties as Signal Temporal Logic (STL) specifications. Then, a partitioning component of FedSTL associates each client to an aggregation center, while the framework continues to infer properties for the cluster. At the training stage, both cluster and client models are encouraged to follow customized properties to achieve a hierarchical property enhancing strategy. Further, we show preliminary results of FedSTL in this poster abstract under a synthetic multitask IoT environment and a real-world traffic prediction scenario. 

   more » « less
5. A prospective cohort study of preconception COVID-19 vaccination and miscarriage

   https://doi.org/10.1093/humrep/dead211  

   Yland, Jennifer J. ; Wesselink, Amelia K. ; Regan, Annette K. ; Hatch, Elizabeth E. ; Rothman, Kenneth J. ; Savitz, David A. ; Wang, Tanran R. ; Huybrechts, Krista F. ; Hernández-Díaz, Sonia ; Eisenberg, Michael L. ; et al ( October 2023 , Human Reproduction)

   To what extent is preconception maternal or paternal coronavirus disease 2019 (COVID-19) vaccination associated with miscarriage incidence?

   COVID-19 vaccination in either partner at any time before conception is not associated with an increased rate of miscarriage.

   Several observational studies have evaluated the safety of COVID-19 vaccination during pregnancy and found no association with miscarriage, though no study prospectively evaluated the risk of early miscarriage (gestational weeks [GW] <8) in relation to COVID-19 vaccination. Moreover, no study has evaluated the role of preconception vaccination in both male and female partners.

   An Internet-based, prospective preconception cohort study of couples residing in the USA and Canada. We analyzed data from 1815 female participants who conceived during December 2020–November 2022, including 1570 couples with data on male partner vaccination.

   Eligible female participants were aged 21–45 years and were trying to conceive without use of fertility treatment at enrollment. Female participants completed questionnaires at baseline, every 8 weeks until pregnancy, and during early and late pregnancy; they could also invite their male partners to complete a baseline questionnaire. We collected data on COVID-19 vaccination (brand and date of doses), history of SARS-CoV-2 infection (yes/no and date of positive test), potential confounders (demographic, reproductive, and lifestyle characteristics), and pregnancy status on all questionnaires. Vaccination status was categorized as never (0 doses before conception), ever (≥1 dose before conception), having a full primary sequence before conception, and completing the full primary sequence ≤3 months before conception. These categories were not mutually exclusive. Participants were followed up from their first positive pregnancy test until miscarriage or a censoring event (induced abortion, ectopic pregnancy, loss to follow-up, 20 weeks’ gestation), whichever occurred first. We estimated incidence rate ratios (IRRs) for miscarriage and corresponding 95% CIs using Cox proportional hazards models with GW as the time scale. We used propensity score fine stratification weights to adjust for confounding.

   Among 1815 eligible female participants, 75% had received at least one dose of a COVID-19 vaccine by the time of conception. Almost one-quarter of pregnancies resulted in miscarriage, and 75% of miscarriages occurred <8 weeks’ gestation. The propensity score-weighted IRR comparing female participants who received at least one dose any time before conception versus those who had not been vaccinated was 0.85 (95% CI: 0.63, 1.14). COVID-19 vaccination was not associated with increased risk of either early miscarriage (GW: <8) or late miscarriage (GW: 8–19). There was no indication of an increased risk of miscarriage associated with male partner vaccination (IRR = 0.90; 95% CI: 0.56, 1.44).

   The present study relied on self-reported vaccination status and infection history. Thus, there may be some non-differential misclassification of exposure status. While misclassification of miscarriage is also possible, the preconception cohort design and high prevalence of home pregnancy testing in this cohort reduced the potential for under-ascertainment of miscarriage. As in all observational studies, residual or unmeasured confounding is possible.

   This is the first study to evaluate prospectively the relation between preconception COVID-19 vaccination in both partners and miscarriage, with more complete ascertainment of early miscarriages than earlier studies of vaccination. The findings are informative for individuals planning a pregnancy and their healthcare providers.

   This work was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development, the National Institute of Health [R01-HD086742 (PI: L.A.W.); R01-HD105863S1 (PI: L.A.W. and M.L.E.)], the National Institute of Allergy and Infectious Diseases (R03-AI154544; PI: A.K.R.), and the National Science Foundation (NSF-1914792; PI: L.A.W.). The funders had no role in the study design, data collection, analysis and interpretation of data, writing of the report, or the decision to submit the paper for publication. L.A.W. is a fibroid consultant for AbbVie, Inc. She also receives in-kind donations from Swiss Precision Diagnostics (Clearblue home pregnancy tests) and Kindara.com (fertility apps). M.L.E. received consulting fees from Ro, Hannah, Dadi, VSeat, and Underdog, holds stock in Ro, Hannah, Dadi, and Underdog, is a past president of SSMR, and is a board member of SMRU. K.F.H. reports being an investigator on grants to her institution from UCB and Takeda, unrelated to this study. S.H.-D. reports being an investigator on grants to her institution from Takeda, unrelated to this study, and a methods consultant for UCB and Roche for unrelated drugs. The authors report no other relationships or activities that could appear to have influenced the submitted work.

   N/A.

    

   more » « less