Data science for pediatric infectious disease: utilizing COVID-19 as a model

Abstract

Purpose of review

During the COVID-19 pandemic, governments and public health agencies used data science tools and data sources in real time to evaluate pathogen transmissibility, disease burden, healthcare capacity, and evaluate treatment and preventive measures. The purpose of the review is to highlight the application of these data sources and methods during the COVID-19 response.

Recent findings

Advances in the development of common data models enabled multisite data networks to overcome healthcare data fragmentation, enabling national surveillance platforms, and offering unprecedented statistical power to conduct national surveillance and detect emerging clinical entities like MIS-C and long COVID in diverse pediatric populations. These integrated networks were also used in evaluating the effectiveness of vaccines and therapies. New surveillance approaches combining traditional clinical data with novel data sources including wastewater detection, web-based search engines, and mobility patterns yielded comprehensive ensemble approaches that informed public health policy.

Summary

The COVID-19 pandemic highlighted the importance of timely evidence for decision-making during outbreak responses and the benefits of using data science tools to help provide real time, actionable insights, which can help guide our public health response to infectious diseases threats in the future.

INTRODUCTION

The COVID-19 pandemic is considered one of the most severe in recent history, with long-standing social, health and economic impacts. Governments and health agencies needed to consider a number of factors simultaneously, including transmissibility, disease burden, healthcare capacity, economic impact, and behavioral factors, both on a local and global scale. This complexity required a response that utilized all available data from electronic health records (EHRs), public health departments, internet search engines, and virologic samples. Data science, which incorporates mathematics, statistics, specialized programming, advanced analytics, artificial intelligence and machine learning with subject matter expertise to guide decision making, aided in this response.

During the pandemic, several data science analytic approaches and sources were able to be activated rapidly to produce high-quality data accurately and consistently [1^▪. Built upon a decade of advances in clinical data infrastructure, large-scale collaborative networks supported insights in real time, as data were generated. These networks formed the epicenter of the outbreak countermeasures, helped inform decision making and public awareness regarding transmission risk and disease burden. These data helped inform many facets of the COVID response, including testing, variant and disease monitoring, healthcare resource utilization, as well as evaluating vaccination and treatment effectiveness (Fig. 1).

FIGURE 1.

Examples of data sources, infrastructure and outputs utilized during the COVID-19 pandemic. Footnote created in BioRender. Waxse, B. (2025) https://BioRender.com/dywtcix.

In this review, we outline ways in which data science enabled the COVID-19 response in multiple ways: [1^▪] the development of multiinstitutional networks [2], global surveillance and visualization methods [3], treatment evaluations [4], vaccine effectiveness monitoring, and [5] discovery of postacute sequelae.

THE ROLE OF DATA NETWORKS IN COVID-19

The COVID-19 pandemic catalyzed unprecedented collaboration in clinical data science, building upon a decade of advances in EHR research infrastructure. These collaborative data networks enabled researchers to study COVID-19 at an accelerated pace.

Modern EHR data are maintained by individual health systems, and even though only a few different EHR applications exist, standards and implementation differ dramatically. The challenge of harmonizing data from disparate sources necessitated common data models (CDMs) – standardized frameworks for collecting and querying multiinstitutional data.

Three primary CDMs were instrumental for pediatric COVID-19 research infrastructure: Observational Medical Outcomes Partnership (OMOP), National Patient-Centered Clinical Research Network (PCORnet), and Integrating Biology at the Bedside (i2b2) data repositories integrated by the SHRINE platform [2–6]. These prepandemic developments in data standardization created an environment where international collaborative networks could quickly analyze real-word patient data.

PEDSnet, established in 2013, emerged as a critical resource for pediatric COVID-19 research [7^▪]. PEDSnet, a network within the Patient Centered Outcomes Research Network (PCORnet), is a national clinical research network that has standardized EHR data for millions of children, and provides a novel resource for timely, efficient, and accurate surveillance of infectious diseases outbreaks in children. With data from 11 US pediatric health systems, PEDSnet’s infrastructure allowed for rapid, large-scale evaluation of COVID-19 impact in children [8]. By 2025, over a fifth of their 371 published studies were related to COVID-19 and the network has grown to over 14 million children [9].

The Consortium for Clinical Characterization of COVID-19 by EHR (4CE) became the first multicenter consortium to harmonize EHR data in the US and Europe, connecting 96 hospitals across 5 countries within just months of the pandemic’s onset [10]. Built upon i2b2 and OMOP frameworks, 4CE demonstrated that laboratory value trajectories and clinical presentations were remarkably consistent across international boundaries, although pediatric patients were underrepresented in this cohort [10].

The U.S.-based National COVID Cohort Collaborative (N3C), funded by the National Center for Advancing Translational Sciences, represented a data science collaboration unrivaled in scale and impact for both adult and pediatric research [11^▪]. Adopting the OMOP CDM, N3C harmonized data from diverse networks and CDMs into an open and secure enclave containing more than 12 billion clinical observations [11^▪]. By 2025, N3C had grown to include data from 84 sites and nearly 23 million patients, supporting 4300 researchers and generating 296 publications [12^▪]. Though only around 15% of N3C patients were children, its scale made it an invaluable contributor to pediatric COVID-19 research.

Just as preexisting data infrastructure supported rapid discovery during the pandemic, COVID-19 accelerated the development of clinical data infrastructure beyond scaling existing approaches. Today, lessons from these global data sharing initiatives are driving international collaboration, and the integration of EHR data with other modalities – genomic sequencing, participant surveys, social determinants of health, wearable technologies, and environmental exposures – is expanding our understanding of human health [13].

SURVEILLANCE, DATA VISUALIZATION, AND DATA DASHBOARDS

The COVID-19 pandemic revolutionized infectious diseases surveillance through the integration of traditional surveillance systems with innovative data science approaches. The Johns Hopkins University Center for Systems Science and Engineering (JHU CSSE) COVID-19 dashboard emerged as the world’s premier COVID-19 tracking resource. Launched within just three weeks of the World Health Organization’s announcement of the outbreak in Wuhan, China, it initially mirrored data from the Chinese CDC and WHO before evolving to incorporate data from over 3,500 locations across 195 countries and regions using automated web scraping complemented by manual data collection [14,15,16^▪▪]. This resource visualized cases, deaths, recoveries, testing, hospitalization, and vaccinations, serving as an objective source of information and setting a new standard for global pandemic surveillance [15,16^▪▪].

While dashboards tracked cases, other approaches provided critical insights into infection prevalence. The UK COVID-19 Infection Survey (CIS), coordinated by the Office for National Statistics, the University of Oxford, and other partners, exemplified a rigorous approach by randomly sampling households for PCR and antibody testing to capture both symptomatic and asymptomatic infections. This method estimated age-stratified transmission models that informed COVID-19 restrictions and related policies on a weekly basis [17]. As the pandemic evolved, data scientists also developed novel surveillance approaches that were particularly valuable for monitoring community transmission. Wastewater-based epidemiology (WBE) emerged as a powerful approach that could detect viral RNA 2–4 days before clinical testing by using quantified chemical and biological markers [18,19]. A comprehensive study of 99 counties from 40 states in the US demonstrated that wastewater surveillance not only predicted hospitalizations better than models based on cases and hospitalization records, but also revealed associations between vaccination rates, vulnerability indices, and COVID-19 outcomes [20]. As public health emergency orders ended and resources decreased, these approaches became increasingly valuable.

The pandemic also supported the development of ensemble forecasting approaches that combined multiple data types to improve prediction accuracy. In Austria, researchers compared wastewater data with Google mobility data to improve precision of the multivariate model [21]. The CDC developed forecasting systems that integrated case surveillance, hospitalization data, and wastewater monitoring to predict regional outbreaks with improved accuracy and found that median ensemble forecasts outperformed individual components, improving current modeling strategies [22,23].

RISK STRATIFICATION

The scale of massive integrated datasets during the COVID-19 pandemic enabled unmatched statistical power for identifying risk factors for COVID-19 severity and long-term complications [24]. Within months, large-scale EHR analysis from the United Kingdom’s National Health System identified risk factors associated with severe infection, providing clinicians with early guidance [25]. In other cases, standardized electronic phenotyping, machine learning, and novel data sources like wearables converged to create an enhanced understanding of a risk stratification and diseases mechanisms [11^▪,26^▪,27,28^▪,29,30].

EVALUATION OF TREATMENTS

During the early phase of the COVID-19 pandemic, there was a critical need to evaluate the effectiveness of treatments against SARS-CoV-2 in order to minimize disease burden and mortality. Observational data played a crucial role in evaluating treatment effectiveness in a timely manner, to aid clinicians as an adjunct to randomized controlled trials, or in settings in which RCT data were not unavailable, an issue particularly relevant to pediatrics [31,32]. However, such retrospective studies faced methodological challenges, including time-fixed and time-varying confounding and immortal time bias [33]. The target trial emulation framework has been proposed as a more rigorous method, overcoming several of these limitations, whereby a hypothetical RCT is designed, and then the target trial is emulated using observational data [34^▪,35^▪,36]. Other applications of EHR data used to aid treatment studies during the COVID pandemic were used to ascertain study feasibility, optimize study design, modernize recruitment through identifying potential participants for clinical trials, and to reduce trial data collection costs through automated clinical data extraction [37].

VACCINE EFFECTIVENESS

The development and deployment of SARS-CoV-2 vaccines signaled a turning point for the COVID-19 pandemic, and provided the general population with protection from severe SARS CoV-2 disease and mortality. Following the vaccine rollout through emergency use authorizations and conditional approvals, it became necessary to assess the real-world effectiveness of SARS-CoV-2 vaccines, and provide further evidence for its safety through surveillance. The CDC collaborated with a number of public health and academic partners throughout the US to collate EHR and observational data to provide timely assessments, and to monitor changes that may arise from emerging variants as well as waning immunity [38–40]. These programs helped inform the community, clinicians, and help shape vaccine policy, and included networks such as Investigating Respiratory Viruses in the Acutely Ill (IVY) [41], Virtual SARS-CoV-2, Influenza, and Other respiratory viruses Network (VISION) [42,43], Overcoming COVID [44,45] and Increasing Community Access to Testing, Treatment and Response (ICATT) [46].

EVALUATING THE POST-ACUTE SEQUELAE OF SARS-COV-2

Soon after the COVID-19 pandemic emerged, there was a recognition of a number of individuals suffering from postacute sequelae of SARS-CoV-2, also known as long COVID. Given its heterogeneous manifestations and novelty, there was an urgent need to better understand this condition both in adults and children [47]. In response to this need, the NIH funded the REsearching COVID to Enhance Recovery (RECOVER) Initiative, which was launched in 2021 to advance research into long COVID [48]. One of the cores of RECOVER includes EHR data efforts led by PEDSnet, PCORnet, and N3C. These networks have contributed vast amounts of clinical data that have been used to study aspects related to the detection, prevention, prediction and treatment of long COVID, and have helped to provide accelerated insights into an emerging health condition. Through the harmonization of data, development of computable phenotypes, use of machine learning and other advanced statistical approaches as well as collaboration with data scientists and clinicians, there has been a rapidly growing body of literature related to long COVID in children and adults [29,49–51].

CHALLENGES

The scale and speed of the COVID-19 data science research also revealed challenges and limitations that will need to inform pediatric data science in the future [52^▪]. Conduct of timely research during a global public health emergency must be balanced with ensuring adequate patient privacy, a central tenet of research. Networks such as N3C addressed this issue through a tiered access model with public researcher identification and comprehensive activity logging while maintaining transparency through publicly listed research abstracts [11^▪]. Alternatively, the OpenSAFELY model provided an innovative solution by providing researchers dummy data for the development of analyses that were executed with real patient records without granting direct access to underlying data [25].

The fragmented nature of healthcare delivery, particularly in the US, also posed significant challenges for data capture and validity. In N3C for example, many networks could only access mortality identified within contributing health systems, limiting the use and reliability of in-hospital mortality for studies [53^▪▪]. Data missingness, inappropriate or incomplete variable mapping among sites within large data networks for certain healthcare variables is another common challenge, requiring a rigorous data quality approach, and careful variable and cohort selection [12^▪]. In some circumstances, missing data required aggregation of data manually, a technique adopted for the Johns Hopkins University CSSE COVID-19 dashboard, for missing data from Arizona that was locked behind business intelligence tools [16^▪▪].

Two high-profile article retractions within 2 months of publication [54^▪▪] early in the pandemic emphasized the importance of data transparency and reproducibility when the data could not be independently verified [55,56]. These instances demonstrate the importance of using data that are “fit for purpose” and thoroughly explored before conducting analyses [57^▪]. In one study, a thorough consideration of data missingness, EHR discontinuity, and data quality resulted in only 15 out of 67 sites remaining in the final analysis, highlighting the trade-offs between data quality and ensuring a representative cohort [53^▪▪].

CONCLUSION

During a public health emergency response, leaders must make critical decisions regarding public health measures, provision of care and allocation of resources. These decisions are often reactive, occurring in a rapidly changing environment where there is little or incomplete information available [58]. Through a collaborative network of public health experts, epidemiologists, informaticians, data scientists and infectious diseases experts, the use of high quality, integrated data for analysis and surveillance met the call for an effective outbreak response, and provided guidance for the utility, scale and timing of prevention strategies. There is a growing use of AI tools for this purpose, including large-scale data abstraction, homogenization and modeling. Emerging infectious pathogens remain a significant threat to public health worldwide. The COVID-19 pandemic highlighted the importance of timely evidence for decision-making during outbreak responses and the benefits of using data science tools to help provide real time, actionable insights, which can help guide our public health response to infectious diseases threats in the future.

KEY POINTS.

• The COVID-19 pandemic accelerated data infrastructure development sharing in a period of discovery that would have been impossible through traditional approaches.

• Prepandemic advances in the development of common data models enabled researchers to overcome healthcare data fragmentation, offering unprecedented statistical power to detect emerging clinical entities like MIS-C and long COVID in diverse pediatric populations.

• New surveillance approaches combining traditional clinical data with wastewater epidemiology and mobility patterns yielded comprehensive ensemble approaches that informed public policy approaches.

• Target trial emulation and other advanced methods have helped overcome traditional limitations of observational data, enhancing the utility of large multicenter datasets and networks.

• A collaborative, open-science response to COVID-19 demonstrated that transparent, reproducible methodologies can accelerate discovery while maintaining scientific rigor and patient privacy.