Characterization of long COVID temporal sub-phenotypes by distributed representation learning from electronic health record data: a cohort study

Arianna Dagliati, Zachary H. Strasser, Zahra Shakeri Hossein Abad, Jeffrey G. Klann, Kavishwar B. Wagholikar, Rebecca Mesa, Shyam Visweswaran, Michele Morris, Yuan Luo, Darren W. Henderson, Malarkodi Jebathilagam Samayamuthu, Bryce W. Q. Tan, Guillame Verdy, Gilbert S. Omenn, Zongqi Xia, Riccardo Bellazzi, James R. Aaron, Giuseppe Agapito, Adem Albayrak, Giuseppe Albi, M Alessiani, Anna Alloni, Danilo F. Amendola, François Angoulvant, Li L.l.j. Anthony, Bruce J. Aronow, Fatima Ashraf, Andrew M. Atz, Paul Avillach, Paula S. Azevedo, James Balshi, Brett K. Beaulieu‐Jones, Douglas S. Bell, Antonio Bellasi, Vincent Benoît, Michele Beraghi, José Luis Bernal-Sobrino, Mélodie Bernaux, Romain Bey, Surbhi Bhatnagar, Alvar Blanco-Martínez, Clara-Lea Bonzel, John Booth, Silvano Bosari, Florence T. Bourgeois, Robert L. Bradford, Gabriel A. Brat, Stéphane Breant, Nicholas W. Brown, Raffaele Bruno, William Bryant, Mauro Bucalo, Emily Bucholz, Anita Burgun, Tianxi Cai, Mario Cannataro, Aldo Carmona, Charlotte Caucheteux, Julien Champ, Jin Chen, Krista Y. Chen, Luca Chiovato, Lorenzo Chiudinelli, Kelly Cho, James J. Cimino, Tiago K. Colicchio, Sylvie Cormont, Sébastien Cossin, Jean B. Craig, Juan Luis Cruz-Bermúdez, Jaime Cruz‐Rojo, Mohamad Daniar, Christel Daniel, Priyam Das, Batsal Devkota, Audrey Dionne, Rui Duan, Julien Dubiel, Scott L. DuVall, Loïc Estève, Hossein Estiri, Shirley Fan, Robert W. Follett, Thomas Ganslandt, Noelia García Barrio, Lana X. Garmire, Nils Gehlenborg, Emily Getzen, Alon Geva, Tobias Gradinger, Alexandre Gramfort, Romain Griffier, Nicolas Griffon, Olivier Grisel, Alba Gutiérrez‐Sacristán, Larry Han, David A. Hanauer, Christian HaverkampEClinicalMedicine 2023 · DOI

TL;DR

This study introduces a novel distributed representation learning framework to identify and characterize distinct temporal sub-phenotypes of Long COVID using electronic health records (EHRs). By analyzing data from over 30,000 patients, they identified seven key sub-phenotypes, revealing that a significant portion of COVID-19 survivors experience persistent symptoms, with joint pain and dyspnea being the most prevalent.

What Problem Does This Paper Address?

Long COVID, or Post-Acute Sequelae of SARS-CoV-2 (PASC), is a complex and multifaceted condition characterized by a wide range of persistent symptoms that can significantly impact patients' lives. A major challenge in understanding and managing PASC is the lack of standardized definitions and the difficulty in characterizing its diverse sub-phenotypes and their temporal evolution. Existing research often relies on broad definitions or limited datasets, making it hard to identify specific patient groups, predict symptom trajectories, and develop targeted interventions. This paper aims to address this by developing a scalable, data-driven approach to precisely define and characterize PASC sub-phenotypes using longitudinal EHR data.

Methods

The researchers employed a two-phase approach. Phase 1 focused on defining PASC sub-phenotypes using a deductive framework that combined clinical knowledge with an iterative, data-driven approach. They utilized the Machine Learning for Health Outcomes (MLHO) pipeline, a distributed representation learning method, to analyze EHR data. This involved identifying "core features" (initial clinical indicators) and then using MLHO to discover "MLHO features" (additional EHR data elements) that were associated with PASC sub-phenotypes, particularly those appearing 90 days or more after initial COVID-19 hospitalization. This process was iterated to refine the definitions.

Phase 2 involved a meta-analysis of the derived sub-phenotype definitions across three healthcare systems to estimate population prevalence and evaluate temporal distributions. They used a Louvain community detection algorithm to cluster features based on their temporal patterns, providing insights into symptom evolution.

Datasets

The study utilized longitudinal electronic health record (EHR) data from 30,422 hospitalized COVID-19 patients across three healthcare systems participating in the Consortium for the Clinical Characterization of COVID-19 by EHR (4CE). Of these, 12,424 patients had at least one observation after 90 days from their initial positive test, forming the primary cohort for sub-phenotype analysis. No new benchmark datasets were introduced; rather, existing federated EHR data was leveraged.

Key Insights and Ideas

  • Temporal Dynamics Matter: The study highlights that PASC is not static; symptoms evolve over time. Their framework explicitly incorporates temporal information, allowing for a more nuanced understanding of sub-phenotype development.
  • Distributed Learning for Scalability: Using a distributed learning approach across multiple healthcare systems is crucial for developing generalizable PASC sub-phenotypes that are not specific to a single institution's data or practices.
  • Augmented Definitions are Key: Combining clinical expertise (core features) with data-driven discovery (MLHO features) creates richer, more accurate definitions of PASC sub-phenotypes, improving their clinical utility.
  • Sub-phenotypes Reveal Heterogeneity: The identification of seven distinct sub-phenotypes underscores the diverse nature of Long COVID, suggesting that a one-size-fits-all approach to diagnosis and treatment is insufficient.

Key Results

  • The framework successfully characterized seven PASC sub-phenotypes: Joint pain, Dyspnea, Chronic malaise and fatigue, Chest pain, Neurological deficits, Dyspepsia, and Cognitive changes.
  • An estimated 15.7% of hospitalized COVID-19 patients experienced at least one PASC symptom, with 5.98% experiencing multiple symptoms.
  • Joint pain (5.45% prevalence) and Dyspnea (4.53% prevalence) were the most common PASC sub-phenotypes.
  • The study found that Neurological deficits symptoms appeared earlier (average 289 days post-infection) compared to Chronic malaise and Fatigue and Dyspnea (average 330 days post-infection).
  • The MLHO-derived features demonstrated good discrimination performance for most sub-phenotypes, with AUROC values generally above 0.8.

Limitations and Future Work

  • Data Limitations: The study acknowledges limitations due to the pre-specified 4CE data extraction schema, meaning not all EHR data elements were available, potentially leading to overlooked features. The reliance on structured EHR data might also miss nuances of symptoms like chronic fatigue.
  • Unmeasured Confounders: As with any observational study using EHR data, unmeasured confounders (e.g., hospital characteristics, social policies, built environment) could influence associations.
  • Generalizability: While multi-site, the study population is from hospitalized COVID-19 patients, and future work could explore non-hospitalized cohorts or broader populations.
  • Future Directions: The authors suggest future work should focus on patient-level data for meta-analyses, incorporating methods to adjust for unmeasured confounding, and further investigating the temporal subtyping of PASC with rigorous causal analysis. They also propose that their machine learning approach can help identify cohorts for further study and aid clinicians in recognizing evolving PASC sub-phenotypes.

Summary updated

Abstract

Authors are supported by National Institute of Allergy and Infectious Diseases, National Institute on Aging, National Center for Advancing Translational Sciences, National Medical Research Council, National Institute of Neurological Disorders and Stroke, European Union, National Institutes of Health, National Center for Advancing Translational Sciences.
BibTeX
@inproceedings{dagliati2023characterization,
  author = {Arianna Dagliati and Zachary H. Strasser and Zahra Shakeri Hossein Abad and Jeffrey G. Klann and Kavishwar B. Wagholikar and Rebecca Mesa and Shyam Visweswaran and Michele Morris and Yuan Luo and Darren W. Henderson and Malarkodi Jebathilagam Samayamuthu and Bryce W. Q. Tan and Guillame Verdy and Gilbert S. Omenn and Zongqi Xia and Riccardo Bellazzi and James R. Aaron and Giuseppe Agapito and Adem Albayrak and Giuseppe Albi and M Alessiani and Anna Alloni and Danilo F. Amendola and François Angoulvant and Li L.L.J. Anthony and Bruce J. Aronow and Fatima Ashraf and Andrew M. Atz and Paul Avillach and Paula S. Azevedo and James Balshi and Brett K. Beaulieu‐Jones and Douglas S. Bell and Antonio Bellasi and Riccardo Bellazzi and Vincent Benoît and Michele Beraghi and José Luis Bernal-Sobrino and Mélodie Bernaux and Romain Bey and Surbhi Bhatnagar and Alvar Blanco-Martínez and Clara-Lea Bonzel and John Booth and Silvano Bosari and Florence T. Bourgeois and Robert L. Bradford and Gabriel A. Brat and Stéphane Breant and Nicholas W. Brown and Raffaele Bruno and William Bryant and Mauro Bucalo and Emily Bucholz and Anita Burgun and Tianxi Cai and Mario Cannataro and Aldo Carmona and Charlotte Caucheteux and Julien Champ and Jin Chen and Krista Y. Chen and Luca Chiovato and Lorenzo Chiudinelli and Kelly Cho and James J. Cimino and Tiago K. Colicchio and Sylvie Cormont and Sébastien Cossin and Jean B. Craig and Juan Luis Cruz-Bermúdez and Jaime Cruz‐Rojo and Arianna Dagliati and Mohamad Daniar and Christel Daniel and Priyam Das and Batsal Devkota and Audrey Dionne and Rui Duan and Julien Dubiel and Scott L. DuVall and Loïc Estève and Hossein Estiri and Shirley Fan and Robert W. Follett and Thomas Ganslandt and Noelia García Barrio and Lana X. Garmire and Nils Gehlenborg and Emily Getzen and Alon Geva and Tobias Gradinger and Alexandre Gramfort and Romain Griffier and Nicolas Griffon and Olivier Grisel and Alba Gutiérrez‐Sacristán and Larry Han and David A. Hanauer and Christian Haverkamp},
  title = {{Characterization of long COVID temporal sub-phenotypes by distributed representation learning from electronic health record data: a cohort study}},
  booktitle = {EClinicalMedicine},
  year = {2023},
  doi = {10.1016/j.eclinm.2023.102210},
}
Figure