Leveraging the PEDSnet clinical research network and electronic health record data to enhance efficiency of trial enrollment for a rare pediatric rheumatic disease

Abstract

Background

We evaluated the PEDSnet clinical research network for study enrollment of juvenile spondyloarthritis, a rare rheumatic disease that includes enthesitis-related arthritis (ERA) and psoriatic arthritis (PsA).

Methods

An electronic health record (EHR)-based typology was developed by an interdisciplinary team to query EHR data for a spectrum of pediatric rheumatic diseases (2009–2023) from 8 PEDSnet centers. The prevalence, characteristics and drug exposures for juvenile spondyloarthritis was explored to gauge feasibility of leveraging the network for study enrollment. Next, the typology was adapted to identify subjects for a clinical trial; the efficiency of EHR typology query was compared to standard screening efforts.

Results

Code sets for 35 pediatric rheumatology conditions were developed to identify potentially eligible subjects in the PEDSnet network. 2510 unique patients with juvenile spondyloarthritis across the PEDSnet health care systems over the years of study were identified. Median age at 1st rheumatology visit was 12.9 years, 50.4% were female, and the median time from 1st to last rheumatology visit was 4.2 years. The spondyloarthritis typology was adapted to screen for eligible patients for the BACK-OFF JSpA trial. Over 3-months at one institution, the query saved 19.5 h and 1.9 h of effort compared to manual screening of all juvenile arthritis or enthesitis-related arthritis patient charts, respectively.

Conclusions

Results support the capacity of the PEDSnet clinical research network to facilitate identification of subjects for rare pediatric rheumatic disease studies. Typologies for these diseases were developed and can be leveraged for clinical trial recruitment to improve efficiency.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12969-025-01181-5.

Background

Many pediatric rheumatic diseases, such as juvenile idiopathic arthritis (JIA), spondyloarthritis (SpA), and systemic vasculitis, are rare. As such, the conduct of sophisticated and adequately powered clinical, translational, and epidemiologic studies is difficult without leveraging large electronic databases or networks. Clinical trials remain especially challenging to conduct. Barriers to conducting trials in these populations include low numbers of eligible patients at any one institution, limited resources to manually screen patients for trial eligibility, and high overhead costs [1]. Examples of high overhead costs include not only time for study start up including contracts, data use agreements, and applications to regulatory authorities, but also time required for study related activities including consent, assent for children, venipuncture, and costs of other procedures including imaging. Electronic health record (EHR)-based research networks and learning health systems provide an opportunity to facilitate identification of potential subjects using computable phenotypes and to streamline a population-based approach to trial enrollment for rare diseases. Learning health systems strive to incorporate knowledge generation into daily practice and clinical documentation to improve not only individual but also population health [2, 3].

PEDSnet (pedsnet.org) is a multicenter clinical research network comprised of 11 healthcare centers (as of 2024), all of which contain academic pediatric rheumatology divisions. PEDSnet was patterned after the learning health system concept and has been part of the Patient-Centered Outcomes Research network (PCORnet) since 2014 [4]. The vision of PCORnet was to create a network-of-networks composed of interconnected research infrastructures that would enable large-scale studies to be conducted faster, better (more responsive to patient and clinician priorities), and with more efficient use of resources [5]. Currently, PEDSnet is one of 8 clinical research networks (CRNs) in PCORnet, and the only one devoted exclusively to pediatric research [4, 6]. Currently, PEDSnet supports over 75 studies, funded by the Patient Centered Outcomes Research Institute (PCORI), the National Institutes of Health (NIH), the Food and Drug Association (FDA), Agency for Healthcare Research and Quality (AHRQ), Centers for Disease Control (CDC), and life sciences companies. Multiple PEDSnet studies have shown the ability of computable phenotypes to identify pediatric patients with high sensitivity and positive predictive value across a wide range of conditions, including glomerular disease, Crohn’s disease, leukemia, lupus, and lymphoma [7–10]. Computable phenotypes use combinations of data elements in the EHR for clinical features such as diagnoses, medications, procedures, both inclusionary and exclusionary, to accurately identify patients without need for chart review [11]. Phenotyping complex pediatric conditions facilitates the use of this data for conducting comparative effectiveness studies. For example, the Pediatric KIDney Stone (PKIDS) Care Improvement Network leveraged PEDSnet for a prospective cohort study to compare stone clearance, re-treatment, and unplanned healthcare encounters for 3 different standard of care interventions for children with kidney stones (ClinicalTrials.gov NCT04285658) [12, 13]. However, the potential to use PEDSnet to facilitate population-based clinical trial enrollment for rare diseases has largely been untapped.

Juvenile spondyloarthritis is an umbrella term which encompasses the juvenile idiopathic arthritis (JIA) categories of enthesitis-related arthritis (ERA) and psoriatic arthritis (PsA). Although juvenile spondyloarthritis is a rare disease it accounts for approximately 20–25% of JIA [14–17]. There has been significant development and study of targeted therapies for adults with spondyloarthritis; however, the evaluation of these therapies in children have not been done and are critically needed as over 1/3 of youth remain active despite treatment [18] and fewer than 20% achieve remission within five years of diagnosis [19]. As such, juvenile spondyloarthritis is a condition that would greatly benefit from innovative population-based methods to identify potentially eligible subjects to facilitate the feasibility of clinical trials.

The objective of this study was to assess the utility of PEDSnet for clinical trial recruitment for rare pediatric diseases by (1) developing code sets using Systematized Nomenclature of Medicine–Clinical Terminology (SNOMED CT) diagnostic codes for conditions treated by pediatric rheumatologists, (2) determining the frequency and demographics of patients within the PEDSnet network with juvenile spondyloarthritis, a rare rheumatic disease, to assess the utility of leveraging the network for research and real-world clinical data, and (3) demonstrating how an algorithm, comprised of a code set to identify a condition, drug code sets, and visits with a rheumatology provider, can be used to inform a population-based approach to trial recruitment in a rare pediatric rheumatic disease.

Methods

The study was reviewed and approved to be in compliance with the United States Department of Health and Human Services regulations as described in 45 CFR 46 by the institutional review board (IRB) at the Children’s Hospital of Philadelphia (CHOP) (IRB protocol number 17-014012) with a waiver of consent. The study also used the PEDSnet Master Reliance Agreement (protocol number 16-012878) process whereby participating institutions ceded IRB review to CHOP.

Data source

This retrospective cross-sectional study evaluated EHR data (2009–2023) within the PEDSnet Research Network (https://pedsnet.org), database version 5.0. This database includes patient-level demographic data and visit-level clinical data comprising clinician or facility specialty, prescription and inpatient administered drugs, diagnoses, resulted labs, ordered and billed procedures, anthropometrics, and vitals in the form of SNOMED CT terminology. Eight PEDSnet children’s healthcare organizations contributed data for this study: Children’s Hospital of Philadelphia; Cincinnati Children’s Hospital Medical Center; Children’s Hospital of Colorado; Nationwide Children’s Hospital; Nemours Children’s Health System (a Delaware and Florida health system); Seattle Children’s Hospital; Lurie Children’s Hospital; and Stanford Children’s Health. The data from each institution was extracted locally and transformed to the PEDSnet common data model (CDM), which is adapted from the Observational Medical Outcomes Partnership (OMOP) CDM [20]. The database is updated quarterly, and the transformed data undergoes a series of data quality assessments to maximize the quality of the data available for conducting research [8, 21]. Further details about the PEDSnet CDM and data quality program can be found on the PEDSnet website (https://pedsnet.org/data/common-data-model/).

Development of pediatric rheumatology diagnostic clusters

A team of 9 clinicians (7 pediatric rheumatologists, 2 general pediatricians with expertise in clinical informatics) participated in the development of the final typology that comprised 8 categories of conditions typically evaluated by pediatric rheumatologists. Categories had up to 9 clusters (35 in total) (Supplement Table 1 and supplementary file 1). For each cluster, a comprehensive list of diagnosis codes was generated by at least 2 investigators and reviewed by a third. The team of investigators aimed to capture the full spectrum of diagnostic codes for each rheumatologic condition, with the understanding that the specificity of each typology can be further augmented by incorporating subspecialty evaluation and/or drug exposures as needed. Leveraging expert opinion to develop code sets is commonly employed and recently used in a PEDSnet study of childhood mental health conditions [22].

Table 1.

Health care utilization by juvenile spondyloarthritis patients (Jan 2009-Jun 2023)

	All patients with ≥ 2 rheumatology visits	All Juvenile Spondyloarthritis	PsA	ERA
	N = 28,544	N = 2510	N = 808	N = 1776
	Median (IQR)
Age at first visit	8.6 (4.0-12.3)	9.9 (6.0-12.9)	9.3 (4.8–12.2)	10.1 (6.4–13.2)
Age at first rheumatology visit	12.0 (7.8–15.0)	12.9 (10.0-15.4)	12.0 (8.4–14.9)	13.2 (10.5–15.5)
Female, %	69/0%	50.4%	65.1%	43.6%
Duration of Follow-up, years	8.2 (5.2–11.4)	8.1 (5.2–11.1)	8.4 (5.6–11.5)	8.0 (5.1–10.9)
Duration followed in rheumatology, years	3.9 (1.8–6.8)	4.2 (2.2–6.6)	4.5 (2.4–7.5)	4.0 (2.2–6.4)
Rheumatology visits per person-year	3.7 (2.5–5.6)	3.6 (2.5–5.1)	3.6 (2.6–5.2)	3.6 (2.6–5.1)

Legend. Patients with 2 rheumatology visits (outpatient, inpatient, or ED) ≥ 28 days apart Jan 2009-Jun 2023. PsA = psoriatic arthritis, ERA = enthesitis-related arthritis

Study sample

The study period was from January 1, 2009 to June 13, 2023. The study sample included patients who had at least 2 encounters with a rheumatologist that fulfilled each of the following criteria: (1) both encounters before the age of 21 years; (2) each encounter with at least one applicable SNOMED CT code for juvenile spondyloarthritis (inclusive of the JIA clusters of ERA and PsA) (Supplement Table 1); and (3) at least 28 days apart in any inpatient, outpatient, or emergency department setting. Patients were excluded if they had less than 12 months in the PEDSnet database prior to 1st applicable SNOMED CT code. Requiring 2 encounters with a rheumatologist attempts to ensure continuity with the rheumatology department at the site and to avoid inclusion of patients with a single visit (e.g. patient seeking a second opinion).

To establish proof of concept of how a code set can inform a population-based approach to enhance study recruitment efforts, the juvenile spondyloarthritis phenotype was adapted to identify subjects for the Biologic Abatement and Capturing Kids’ Outcomes and Flare Frequency in Juvenile Spondyloarthritis (BACK-OFF JSpA) Study (ClinicalTrials.gov NCT04891640 [23]). The goal of the BACK-OFF JSpA trial is to evaluate tumor necrosis factor inhibitor (TNFi) de-escalation strategies in children with juvenile spondyloarthritis in sustained remission on standard TNFi dosing for a minimum of 6 months. This trial focused on the ERA subset of juvenile spondyloarthritis. Inclusion criteria for the trial include: (1) age 8–21 years, (2) diagnosis of ERA by treating rheumatologist, (3) standard dosing of a TNFi, and (4) inactive disease for at least 6 months. Exclusion criteria include uveitis, psoriasis, and inflammatory bowel disease as the presence of any of these associated comorbidities might drive systemic therapy decision making. Not all inclusion criteria are discernable based on an EHR query, but those that were discernable shortened the list of eligible patients; for example the EHR query could not identify which potentially eligible patients had criterion #4 of inactive disease. The presence/absence of inactive disease had to be determined through targeted chart review. To identify as many potentially eligible subjects as possible, the trial specific query included: 1) age 8–21 years, 2) ≥ 1 SNOMED CT code for SpA/ERA (Supplement Table 1) given by a rheumatologist, 3) TNFi prescribed within the past year; and 4) absence of a SNOMED CT code for any of the following excluded conditions: uveitis, psoriasis, and inflammatory bowel disease.

Data analysis

Figure 1 in the Supplement shows the iterative algorithm development of the ERA code set. To develop the ERA code set we computed attrition tables, utilization measures (e.g. specialty of physician diagnosing JSpA), diagnosis counts, and source system mappings of diagnostic codes to evaluate across all PEDSnet institutions. This allowed identification of potential flaws with the diagnostic code set and, combined with the validated chart review, informed the next iteration of the phenotype. Each iteration of the phenotype development was accompanied by focused chart reviews at one institution (CHOP), analysis, and refinement as needed. All charts were reviewed by at least 2 individuals blinded to cohort (JSpA/not JSpA) assignment. All discrepancies between the PEDSnet database and manual chart review were manually adjudicated. We continued to iterate until the algorithm demonstrated a high degree of accuracy. Our final algorithm was tested using 100 cases and 100 non-cases. Non-cases were defined as patients under the age of 25 years with any non-SpA juvenile arthritis diagnosis assigned by a rheumatologist. Many of the SNOMED CT codes used for spondyloarthritis are clinically interchangeable for ERA and PsA. As such, for the validation effort we wanted to be able to accurately identify the ERA patients and differentiate them from other categories of JIA (i.e. systemic JIA or polyarticular JIA) that would not be eligible for the study contemplated. Using the documented diagnosis in the chart as the reference standard, percent agreement, recall (sensitivity), precision, and F1 scores are reported since they are metrics independent of true negative cases.

Fig. 1.

Heatmap of ERA diagnosis code distribution across PEDSnet. Heatmap displaying the usage of the 10 most common ERA diagnoses across 8 sites. Among the 55 SNOMED CT codes for ERA, 32 codes were assigned at least once by rheumatologists for 4,156 patients. This heatmap shows the proportion of unique patients at each site diagnosed with any of the 10 most commonly used ERA codes. The JIA, ERA code is the darkest in color, reflecting the highest proportion

Clinical characteristics were examined using descriptive statistics and included demographics, participating institution, services (inpatient, outpatient, emergency department), SNOMED CT diagnostic codes, location and specialty of encounter physician, number of rheumatology subspecialty visits per person year (total number of rheumatology encounters ever/time followed in rheumatology), and duration of rheumatology follow-up (computed as the date of the most recent visit minus the date of the index visit).

The first-line drug was defined as the earliest relevant drug prescription or drug administration in the patient’s record during the study period after the index date (earliest relevant SNOMED CT code) and all subjects had at least 1 year in the database prior to the index date. Drug exposures were grouped into first, second, and third-line therapy which could each consist of more than 1 drug started concurrently. If 2 initial drug prescriptions or administrations occurred within the same 28-day window they were considered as combination/concurrent therapy. Subsequent lines of therapy were determined by looking for the initial prescription or administration of a different relevant drug that occurred at least 28 days after the index drug in the case of a second-line therapy, and at least 28 days after the second-line drug in the case of third-line therapy.

The efficiency of the EHR-based approach to identify potentially eligible subjects for the BACK-OFF JSpA trial was compared to (1) Manual screening of all JIA patients, (2) Manual screening of all ERA patients, (3) Manual screening of all ERA patients identified by local queries such as Epic System’s Slicer Dicer or Healthy Planet tools. Slicer Dicer and Healthy Planet are commonly used data extraction tools that enable the user to query EHR data about patients and populations across an Epic site (Epic Systems, Verona, WI). Efficiency was estimated using the number of charts that required screening and the cumulative time to review the charts.

Results

2,510 unique patients with juvenile spondyloarthritis across the 8 PEDSnet health care systems over the years of study were identified (808 and 1,776 with PsA and ERA, respectively, with 106 overlapping). Median age at 1st rheumatology visit was 12.9 (IQR 10.0-15.4), 50.4% were female, and the median time from 1st to last rheumatology visit was 4.2 years (IQR 2.2–6.6) (Table 1).

Median age of initial rheumatology visit for PsA patients was 9.3 years (4.8–12.2), 65.1% were female, and the median time from 1st to last rheumatology visit was 4.5 years (2.4–7.5) (Table 1). Table 2 lists 1st, 2nd and 3rd line medication therapies for PsA. 679 (84%) patients with PsA were treated with at least 1 disease-modifying antirheumatic drug (DMARD), biologic, or small molecule therapy (Table 2). The most common first-line (non-glucocorticoid) systemic therapies were DMARDs (66.0%) and TNFi (35.3%). Four hundred and twenty-nine (53.0%) and 189 (25.1%) patients required second- and/or third-line therapy. Some of the newer targeted agents, specifically inhibitors of the interleukin (IL)-12/23 and IL-17 axes for PsA were used more frequently as second- or third-line therapy. Specifically, ustekinumab, which blocks the IL-12/23 pathway, was the 3rd most common second-line therapy after TNFi and DMARDs and the most common third-line therapy.

Table 2.

First-, second- and third-line disease modifying, biologic, or small molecule therapies by rheumatology diagnosis

	First-line, N(%)	Second-line, N(%)	Third line, N(%)
PsA
Total N = 808	679 (84.0)	429 (53.0)	189 (25.1)
DMARDs	448 (66.0)	80 (18.6)	14 (6.9)
TNF inhibitors	240 (35.3)	256 (59.7)	16(7.9)
hydroxychloroquine	20 (2.9)	14 (3.7)	16 (7.9)
tocilizumab	< 11	18 (4.2)	25 (12.3)
ustekinumab	< 11	26 (6.1)	38 (18.7)
abatacept	< 11	18 (4.2)	30 (14.8)
tofacitinib	< 11	18 (4.2)	19 (9.4)
IL-17 inhibitors	< 11	< 11	29 (14.3)
apremilast	< 11	< 11	12 (5.9)
ERA
Total N = 1776	1367 (77.0)	703 (39.6)	219 (12.3)
DMARDs	781 (57.1)	203 (28.9)	16 (7.3)
TNF inhibitors	641 (46.9)	388 (55.2)	22 (10.0)
hydroxychloroquine	62 (4.5)	23 (3.3)	20 (9.1)
tocilizumab	< 11	30 (4.3)	49 (22.4)
abatacept	< 11	23 (3.3)	30 (13.7)
tofacitinib	< 11	21 (3.0)	30 (13.7)
IL-17 inhibitors	< 11	20 (2.8)	28 (12.8)
ustekinumab	< 11	< 11	12 (5.5)

Legend. Patients with 2 rheumatology encounters ≥ 28 days apart Jan 2009-Jun 2023. PsA = psoriatic arthritis, ERA = enthesitis-related arthritis

Median age of initial rheumatology visit for ERA patients was 10.1 years (6.4–13.2), 43.6% were female, and the median time from 1st to last rheumatology visit was 4.0 years (2.2–6.4) (Table 1). Table 2 lists 1st, 2nd and 3rd line medication therapies for ERA. One thousand three hundred sixty-seven (77.0%) patients with ERA were treated with at least 1 DMARD, biologic, or small molecule therapy. The most common first-line (non-glucocorticoid) therapies were DMARDs (57.1%) and TNFi (46.9%). Seven hundred and three (39.6%) and 219 (12.3%) patients required second- or third-line therapy. Similar to PsA, some of the newer targeted agents, specifically inhibitors of the IL-17 and Janus kinase-signal transducer and activator of transcription (JAK-STAT) pathways, were used more frequently as second- or third-line therapy.

Applying an EHR-based algorithm to identify subjects for a trial

Using the documented diagnosis in the chart as the reference standard, performance statistics of the final algorithm to identify patients with ERA demonstrated 94% agreement, with a recall of 90.0%, precision of 99% and F1 score of 0.94. Of the 200 patient records identified from the PEDSnet algorithm for validation by manual review – 100 cases and 100 controls – 110 patients had a documented diagnosis of JSpA and 90 had a non-JSpA diagnosis. There were 11 of 110 patients missed by the EHR algorithm that were positive during chart review and one patient of 90 during the chart review that was positive by computable phenotype but negative through chart review. Accuracy of the algorithm was reliant on provider coding accuracy and chart review applied classification criteria according to symptom presentation. The primary reason for discordance between the algorithm and chart review was due to providers using diagnosis codes as part of their clinical interpretation versus a classification definition (i.e., coding a patient as oligoarticular JIA when ILAR criteria for enthesitis-related arthritis was met). To explore the applicability of the code set across the 8 PEDSnet institutions a heatmap was created displaying the usage of the 10 most common ERA diagnoses across the 8 sites (Fig. 1). Figure 1 demonstrates that the 10 most common SNOMED CT codes for ERA are used somewhat differently across the 8 institutions, giving insight that the specific code set for each institution might need to be fine-tuned to match usage patterns and maximize efficiency and utility of query results.

EHR-discernable inclusion criteria (age 8–21 years; ≥1 SNOMED CT code for ERA given by a rheumatologist; TNFi prescribed within the past year) and exclusion criteria (absence of SNOMED CT codes for uveitis, psoriasis, and inflammatory bowel disease) were applied to the PEDSnet query for the first 3 months after the study was opened at a single center (CHOP; Nov 1, 2021 to March 4, 2022). The number of potentially eligible patients for the trial were identifiable in 4 ways: (1) Manual screening of all JIA patients, (2) Manual screening of all ERA patients, (3) Manual screening of all ERA patients identified by local queries such as Slicer Dicer or Healthy Planet, or (4) PEDSnet query with abbreviated chart review to verify eligibility. The cumulative number of charts that required review based on the 4 methods over 3 months at a single center were 417, 64, 40, and 26, respectively, to identify the same 15 trial-eligible patients (Fig. 2).

Fig. 2.

Number of charts requiring review based on screening methods. The number of potentially eligible patients across sites for the trial are identifiable in 4 ways: (1) PEDSnet query (PEDSnet), (2) Manual screen of all ERA patients identified by local queries such as Healthy Planet (HP ERA) or Slicer Dicer, (3) Manual screen of all ERA patients, (4) Manual screen of all JIA patients. Demonstrated above are the number of charts that required review for each method over the first 3 months of trial enrollment at a single PEDSnet site

Estimating approximately 3 min per chart review, Fig. 3 demonstrates the cumulative time required for each screening method to confirm trial eligibility over those same 3 months at that single center. By providing a smaller and more targeted list of eligible subjects, the PEDSnet ERA query saved approximately 19.5 h and 1.9 h of screening effort compared to manual screening of all juvenile arthritis or ERA patient charts, respectively, at a single center during the measured 3-month period.

Fig. 3.

Estimated cumulative time devoted to each method of screening for trial eligibility. Cumulative time devoted to each of 4 methods of screening for trial eligibility over the initial 3 months of trial enrollment at a single PEDSnet site, estimating approximately 3 min per chart

Discussion

A diagnostic code classification typology using EHR data was developed for a spectrum of rare pediatric rheumatic conditions and was used to estimate clinical characteristics and drug exposure(s) within the PEDSnet research network for juvenile spondyloarthritis, including ERA and PsA. A useful contribution of this study is the publicly available EHR-based typology (Supplement Table 1) that can augment methods and infrastructure for studying rare pediatric rheumatic diseases (https://pedsnetapps.chop.edu/metadata/handle/20.500.14642/921). Currently, most of the data about rare pediatric rheumatic diseases are cultivated from disease-specific registries that are costly to assemble and maintain, frequently require manual chart abstraction, may not capture the full spectrum of disease, and are prone to errors and missing data [3, 24, 25]. Additionally, data on the real-world 2nd - and 3rd -line drug use for rare pediatric rheumatology diseases including PsA and ERA are scarce. We demonstrated the usefulness of the PEDSnet network as a resource for real-world clinical and drug use data on rare diseases as well as a valuable source population for prospective observational and interventional studies.

The EHR holds tremendous promise to streamline recruitment efforts for clinical trials and other prospective research. The first EHR-based pragmatic trial to leverage the PCORnet was the Aspirin-Dosing: A Patient-Centric Trial assessing Benefits and Long-term Effectiveness (ADAPTABLE) trial [26]. This and other subsequent trials have identified several EHR-driven recruitment strategies including direct to patient messages through existing patient portals, lists of potentially eligible subjects based on queries for mailing/calls, point of care provider alerts, direct to research alerts, and patient/disease registries [27–31]. When applied to trials, queries using structured data from the EHR can include the condition of interest, comorbidities, medications, age and other demographics (including area-level variables linked via geocodes), and laboratory results to help narrow the potential pool of patients and reduce screen failures. When targeted query lists are cross-checked with clinic schedules on a weekly or biweekly basis, they can greatly reduce team screening efforts and augment ability to meet recruitment goals. Additionally, leveraging a centralized EHR-based recruitment model may result in improved participation from underrepresented populations [32–34]. However, perceived barriers to using EHR for trial recruitment include institutional limits on accessing records with consent, limits on ability to directly contact potential subjects without consent, and heterogeneous data structures across participating institutions, and lack of IT support [35]. PEDSnet as a large scale pediatric clinical research network has developed the technical, regulatory, ethical, and administrative infrastructure to overcome all these barriers. These advances are transforming how clinical research is being done in real-world settings, accelerating the pace and the breadth of knowledge generation that can have immediate patient impact.

The ERA typology developed herein was modified to facilitate population-based identification of potentially eligible subjects for the BACK-OFF JSpA trial [23] across the participating institutions. While the EHR query did not completely remove the screening burden, it did significantly reduce it. By using the targeted query list, research staff could easily cross-check upcoming clinic schedules for patients and perform a streamlined chart review for the eligibility criteria that could not be queried (inactive disease for at least 6 months, standard dosing of TNFi). In comparison to bi-weekly manual screening procedures of all juvenile arthritis or ERA patient charts, the PEDSnet query saved teams an estimated 19.5 and 1.9 h of effort, respectively, over just the first 3 months of enrollment at a single center. The implications on saved time and effort across all participating institutions over the duration of the trial will be significant. An additional benefit outside the direct timesaving is how the use of an EHR query allowed the data coordinating center to support participating institutions experiencing research staffing issues. Being able to periodically provide a short list of priority patients with potential eligibility to the site principal investigator allowed recruitment efforts to continue when a pause in research activity may have been necessary in other trials without a full complement of research staff. Similar approaches and frameworks can be leveraged by investigators who need to identify patients with rare disease for clinical trial participation.

There is a time commitment for each screening strategy outside the effort spent examining a patient chart for eligibility. In the comparison screening strategies, preparing the potential subject cohort ranged from zero effort strategies that were as simplistic as reviewing the entire upcoming clinic visit schedule for patients with a JIA or ERA diagnosis to the minimal effort strategy of accessing/creating a report with eligible diagnoses (Healthy Planet or Slicer Dicer). In contrast, design, setup, validation, and installation of a PEDSnet query utilizing a computable phenotype is a significant undertaking. However, once a diagnosis has a validated computable phenotype like those reported in this manuscript, the effort is reduced substantially and the commitment to install the strategy is reduced to verifying the correct population is being identified and to tweak the code set to account for local diagnostic code assignments, as necessary.

There are several limitations to our study. First, the code set for each rheumatic condition is exhaustive. However, this was intentional. The team of investigators aimed to capture the full spectrum of codes for each rheumatic condition, with the understanding that the specificity and validity of each typology can be further improved by incorporating subspecialty evaluation, disease manifestations, comorbid conditions, and/or drug exposures as needed. Second, diagnostic code sets were generated by clinical experts who care for children with these conditions and the use of various diagnostic codes for the same condition varies slightly across institutions (Fig. 1). Additionally, the ERA code set was validated at only one institution. However, the codes approach, which aggregates all codes for a specific condition, overcomes the idiosyncrasies of coding behavior. Therefore, some effort is required to refine the codes used at a particular site to increase accuracy of the query and to maximize efficiency for the team. Third, the time spent doing manual chart review was estimated and not a precise measurement. However, we estimated the amount of time conservatively at 3 min per chart while in many cases manual chart review could take much longer and may vary based upon the experience of the team member performing the screening. Fourth, by aggregating codes we may lose some of the nuances in diagnosis as not all diagnoses in a given cluster are treated the same given the heterogeneity in many rheumatic diseases. However, by making the individual codes for each code set available, researchers can easily narrow the spectrum of codes for a particular condition as needed. Lastly, differences in coding behavior, institutional infrastructure, or EHR system configuration might influence the transportability of the typology across other health systems. Broader external validation and implementation in non-PCORnet pediatric networks or international settings was beyond the scope of this project but certainly contemplatable as a future direction.

A diagnostic EHR-based typology was developed for a spectrum of rare pediatric rheumatic conditions and used to estimate prevalence and real-world drug exposure(s). Adaptation of the ERA code set was feasible and facilitated a population-based approach to subject enrollment within the PEDSnet research network that was also time saving for the research team.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

PFW- conception and design of the work; acquisition, analysis, and interpretation of data; Drafting and reviewing the manuscript; Final approval of the version to be published; Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolvedLU- data acquisition, analysis of data; Drafting and reviewing the manuscript; Final approval of the version to be published; Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.MM- data acquisition, analysis of data; Drafting and reviewing the manuscript; Final approval of the version to be published; HR- data acquisition, analysis of data; manuscript draft review and approval of final version. TGB- data acquisition, analysis of data; manuscript draft review and approval of final version. CM- data acquisition; manuscript draft review and approval of final version. CS- data acquisition; manuscript draft review and approval of final version. EJO- data acquisition; manuscript draft review and approval of final version. JC- data acquisition; manuscript draft review and approval of final version. KB- data acquisition; manuscript draft review and approval of final version. MB- data acquisition; manuscript draft review and approval of final version. EMM- data acquisition; manuscript draft review and approval of final version. CBF- conception and design of the work and data acquisition; manuscript draft review and approval of final version; Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Funding

This was supported by Patient Centered Outcomes Research Institute^® (PCORI^®) Award (CER-2020C1-19212). The study design, analysis, and interpretation of results presented in this publication are solely the responsibility of the authors and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute^® (PCORI^®), its Board of Governors or Methodology Committee.

Data availability

The data underlying this article were provided by PEDSnet under by permission. The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request when permitted by PEDSnet.

Declarations

Ethical approval

The study was approved by the institutional review board (IRB) at the Children’s Hospital of Philadelphia (CHOP) (IRB protocol number 17-014012) with a waiver of consent. The study also used the PEDSnet Master Reliance Agreement (protocol number 16-012878) process whereby participating institutions ceded IRB review to CHOP.

Consent for publication

Not applicable.

Competing interests

PFW – Royalties/licenses: Up-to-date (<$10K to author); Consulting fees: Site investigator for Pfizer and Abbvie Clinical Trials (Payment to institution), Advisory Board member: Lily, Novartis (all <$10K to author), and Consulting fees: Pfizer (payment to institution); Speaking payment or honoraria: 2022 Rheum Now Speaker (<$5K to author) and Spondyloarthritis Research and Treatment Network – honoraria for educational materials (<$5k to author). The other authors have no conflict of interest to declare.