Original research••

Identification and assessment of classification criteria attributes for systemic lupus erythematosus in a regional medical record data network

•,,,,,,,,.

...

Abstract

Objective To assess the application and utility of algorithms designed to detect features of SLE in electronic health record (EHR) data in a multisite, urban data network.

Methods Using the Chicago Area Patient-Centered Outcomes Research Network (CAPriCORN), a Clinical Data Research Network (CDRN) containing data from multiple healthcare sites, we identified patients with at least one positively identified criterion from three SLE classification criteria sets developed by the American College of Rheumatology (ACR) in 1997, the Systemic Lupus International Collaborating Clinics (SLICC) in 2012, and the European Alliance of Associations for Rheumatology and the ACR in 2019 using EHR-based algorithms. To measure the algorithms’ performance in this data setting, we first evaluated whether the number of clinical encounters for SLE was associated with a greater quantity of positively identified criteria domains using Poisson regression. We next quantified the amount of SLE criteria identified at a single healthcare institution versus all sites to assess the amount of SLE-related information gained from implementing the algorithms in a CDRN.

Results Patients with three or more SLE encounters were estimated to have documented 2.77 (2.73 to 2.80) times the number of positive SLE attributes from the 2012 SLICC criteria set than patients without an SLE encounter via Poisson regression. Patients with three or more SLE-related encounters and with documented care from multiple institutions were identified with more SLICC criteria domains when data were included from all CAPriCORN sites compared with a single site (p<0.05).

Conclusions The positive association observed between amount of SLE-related clinical encounters and the number of criteria domains detected suggests that the algorithms used in this study can be used to help describe SLE features in this data environment. This work also demonstrates the benefit of aggregating data across healthcare institutions for patients with fragmented care.

What is already known on this topic

Algorithms that identify attributes of SLE—formally defined in classification criteria sets—in electronic health record data have been developed using data from highly adjudicated gold standard cohorts.

What this study adds

This study assesses the ability of these algorithms to identify SLE attributes in a generalised and heterogeneous data context and highlights a specific use case in which these algorithms can be used to study SLE at the population level.

How this study might affect research, practice or policy

With further refinement, the algorithms highlighted in this study can be employed as the foundation for SLE surveillance systems, aid clinicians in assessing patients for the presence of definitive SLE and serve to investigate hypotheses related to healthcare patterns known to negatively impact patients with SLE.

Introduction

SLE is a chronic autoimmune disease with a wide range of clinical and laboratory manifestations.1 Given its complexity, SLE can be exceptionally difficult for clinicians to arrive at a timely diagnosis and initiate proper treatment, potentially leading to unnecessary accrual of disease-related damage.2 Furthermore, the degree of existing damage is associated with more severe disease progression and endpoints,3 making early diagnosis of SLE critical for optimising clinical care. Accurately identifying individuals with SLE is a labour-intensive process that often requires manual rheumatologist review of patient clinical data,4 presenting as a major barrier in identifying SLE cohorts for research and population surveillance strategies that identify SLE early. To help automate this process, our recent work has focused on developing algorithms that identify SLE attributes in electronic health record (EHR) data using three established classification criteria developed by the American College of Rheumatology (ACR) in 1997,5 6 Systemic Lupus International Collaborating Clinics (SLICC) in 20127 and the European Alliance of Associations for Rheumatology (EULAR)/ACR in 2019.8 9 While the best performing of these algorithms displayed a sensitivity of 76% and a specificity of 99% in identifying patients with SLE, this study used data from patients in a gold standard cohort derived from only one healthcare site, leaving their performance in broader and more representative data contexts as an open question.10

Clinical Data Research Networks (CDRNs) are partnerships between health systems that securely pool and link information among shared patients11 12 and have been successfully implemented in many research contexts.13–16 Recent applications of CDRNs in the setting of SLE have compared COVID-19 mortality rates with the general population using propensity score-matched analyses.17 To better understand the performance of our SLE classification criteria-based algorithms in a large, multisite urban EHR dataset, we adapted the algorithms to the Patient Centered Outcomes Research Network (PCORnet) common data model (CDM)11 and assessed them within the Chicago Area Patient-Centered Outcomes Research Network (CAPriCORN) CDRN.18

As a proof of concept of their utility in population-level surveillance, we also sought to use the algorithms employed in CAPriCORN to investigate care patterns and data availability among patients with documented encounters at multiple organisations, referred to as care fragmentation here. Care fragmentation is of particular importance in the context of SLE, as it has been associated with an increased risk of severe infections and a higher level of disease-related damage indicators.19 The underlying reasons for poorer outcomes in patients with fragmented care are likely multifactorial,19 and further work is needed in understanding the mechanisms contributing to this health disparity. One hypothesis that explains the poorer outcomes observed among patients with SLE with fragmented care is that the EHR data at one site are often incomplete, which may prevent clinicians from understanding the full extent of a patient’s disease state and alter their decision-making.19 Therefore, we aimed to use our algorithms across CAPriCORN to determine if more attributes from SLE criteria sets could be detected when data were aggregated from multiple sites when compared with information derived from only one healthcare organisation.

Methods

Patient and public involvement

CAPriCORN maintains a dedicated Patient and Community Advisory Community (PCAC) that works to include patient input on data governance and research conduct of projects approved by the network.18 This study was governed by the Chicago Area Institutional Review Board that serves as the Institutional Review Board for CAPriCORN, which is in part made up by the PCAC.

CAPriCORN is a partnership among 11 healthcare organisations, mediated by an honest broker, by which patient data derived from electronic health records are shared across the Chicago Metropolitan Area.18 As of 2019, this repository contains data representing approximately 11 million unique patients and includes EHR information derived from 32 hospitals, 20 federally qualified health centres, nearly 400 primary clinics, and over 700 outpatient specialty and surgery clinics.18 The query that produced the analytical dataset for this current work included data from all adult healthcare organisations that had a validated PCORnet CDM datamart at the time of the query: Northwestern Medicine, NorthShore University Health System, University of Illinois Hospital & Health Sciences System, University of Chicago, Cook County Health, AllianceChicago and Rush University Medical Center.

We adapted previously published algorithms for the identification of ACR, SLICC and EULAR/ACR classification criteria10 to V.6.0 of the PCORnet CDM used by CAPriCORN, which includes only structured data elements built on defined clinical terminologies.11 We identified patients who had one or more individual attributes of any of the three SLE classification criteria (ACR, SLICC or EULAR/ACR) within CAPriCORN from 1 January 2012 to 31 July 2019 to establish the initial cohort for this work. We also quantified the number of healthcare encounters with an International Classification of Disease Code (ICD) 9th (710.0) and 10th (M32.9) revisions pertaining to SLE, a simple method for cohort discovery previously shown to have reasonably high sensitivity for SLE among other health conditions.20–22 Finally, we classified patients into the following three groups based on the raw counts of SLE encounters identified by our data retrieval and on previous implementations of diagnosis code-based SLE phenotypes21 22: patients with zero SLE encounters, one to two SLE encounters, and three or more SLE encounters.

Study measures

Demographics

Age at time of the first SLE encounter as well as sex were collected for each patient. Race was based on definitions within the PCORnet CDM, V.3,11 23 which are based on the Office of Management and Budget Race Categories.24 In this study, we describe race as white, black or African American, and other, which includes all other race and ethnic categories identified within the dataset (‘Asian’, ‘Native Hawaiian or Other Pacific Islander’, ‘American Indian or Alaska Native’, ‘multiple race’ or ‘unknown’). This grouping of other racial groups was due to the relatively small number of individuals in the dataset from CAPriCORN and the requirement to preserve privacy of people with rare conditions. Ethnicity was not included in the analyses due to a small number of individuals self-designating ‘Hispanic or Latino’, which was likely due to missing data or differences in reporting between sites.

Healthcare utilisation

The total number of emergency room, inpatient and outpatient visits was extracted for each patient at participating sites over the study period.

Classification of SLE features

Each of the three classification criteria used in this study has specific conditions that need to be met to classify an individual with SLE. The 1997 ACR criteria definition requires that an individual displays four or more total positive criteria.5 6 The 2012 SLICC has two pathways by which someone may be classified as having SLE.7 The first requires that the patient is identified with four or more total criteria with at least one positive criterion derived from two general domains. The first domain consists of clinical manifestations and the second of laboratory procedures known to be associated with SLE. The second pathway includes individuals with a positive antinuclear antibody and SLE nephritis confirmed by renal biopsy. We excluded the second pathway for SLICC algorithm classification in this work given that we were unable to obtain pathology narratives from CAPriCORN. Finally, the 2019 EULAR/ACR criteria set defines SLE with an entry requirement of a positive ANA and a subsequent point-based system, where each criterion contributes points to a patient’s total score.9 Any patient with a positive ANA and a score of 10 or more is considered SLE. A summary of the individual criteria included in each set is included in online supplemental table 1.

EHR implementation of classification criteria sets

The identification of individual attributes and classification of SLE by each criteria set were previously translated into EHR data by expert identification of structured data elements representing diagnosis codes, laboratory measurements, procedures and medication orders related to criteria domains.10 Many of these domains also include exclusion conditions to maximise the possibility of identifying a domain that is attributable to SLE and likely contributed to the high specificity observed when implemented at a single site.10 For example, an inclusion condition for the EHR implementation of the leucopenia domain—a component of the 2012 SLICC criteria—is a blood lymphocyte count below 1.5 K/µL. However, one exclusion condition is a diagnosis of drug-induced neutropenia, which would indicate an instance of leucopenia attributed to a cause other than SLE. Our previous work reporting on the single-site implementation of these algorithms describes a comprehensive list of inclusion and exclusion conditions for each criteria domain.10

Care fragmentation

Care fragmentation was defined as having one or more encounters of any kind at two or more institutions within CAPriCORN during the study period. We were able to identify patients with care fragmentation in the dataset due to the privacy-preserving record linkage hashing algorithm that links patients across sites.11 18

Statistical analyses

All analyses described were conducted in the R software environment, V.4.1.2.

Overall cohort

Descriptive statistics, including race, age, sex and healthcare utilisation, were computed among three subgroups of patients: those with zero ICD-encoded SLE encounters, one to two SLE encounters, and three or more SLE encounters. Tests of statistical significance for these variables were conducted individually between the reference group—zero SLE encounters—and the two non-reference SLE encounter groups. Χ² tests were used for categorical variables and two-sample t-tests or their non-parametric equivalent on continuous variables.

Algorithm performance evaluation

Patients included in our cohort were de-identified and de-duplicated by CAPriCORN prior to transmission of data to our team. Therefore, we were unable to manually adjudicate patient medical records for a definitive diagnosis of SLE and directly evaluate our algorithms on performance metrics that we previously reported for our single-site implementation.10 As an indirect measure of the algorithms’ ability to identify criteria set features, we constructed Poisson regression models to quantify the association between the number of SLE encounter diagnoses and the amount of positive attributes identified from each criteria set. These models were used to compute prevalence ratios of criteria set attributes in the two non-referent SLE partition groups versus the referent group (zero SLE encounters). The positivity rates of individual criteria within the SLE encounter groups were also computed and intergroup comparison was conducted via Χ² tests. A summary of the criteria making up each set is provided in online supplemental table 1.

Care fragmentation

Descriptive statistics of demographic variables and healthcare utilisation were computed as above across the three SLE encounter groups, with additional partitions added to compare patients with and without fragmented care. Within-SLE partition tests of statistical significance were conducted to assess for differences between patients seen at one site and those with fragmented care.

Information gain

To assess the gain of non-redundant SLE information for patients with fragmented care, the mean number of positive criteria within each attribute set identified at the site with the most documented visits (‘top’) and at all sites was computed. A Wilcoxon signed-rank test was conducted for each SLE encounter group to assess if the change in the number of criteria identified was statistically significant. Patients were also assessed for the change in the rate of algorithm-classified SLE defined by each criteria set following multisite data inclusion, which were compared within each group via McNemar’s test.

Results

Of the approximately 11 million patients included in CAPriCORN in 2019, 1 213 130 patients were found with at least one positive domain in the ACR, SLICC or EULAR/ACR criteria sets from the seven participating sites. Further partitioning resulted in 1 201 999 individuals with zero SLE encounters, 4643 with one to two SLE encounters, and 6488 with three or more SLE encounters. The cohort characterisation and partitioning algorithms are summarised in figure 1.

Request permission

Figure 1

Cohort identification and partitioning. Algorithms designed to detect attributes from the 1997 ACR, 2012 SLICC, 2019 EULAR/ACR classification criteria sets were implemented at all CAPriCORN sites containing a validated PCORnet common data model datamart at the time of data capture. Any patient for which the algorithms identified one or more positive attributes from the three criteria sets was included for subsequent partitioning based on the number of SLE encounters identified during the study period. ACR, American College of Rheumatology; CAPriCORN, Chicago Area Patient-Centered Outcomes Research Network; CDRN, Clinical Data Research Network; EULAR, European Alliance of Associations for Rheumatology; PCORnet, Patient Centered Outcomes Research Network; SLICC, Systemic Lupus International Collaborating Clinics.

Patients with three or more SLE encounters were more often black or African American (42% vs 22%, p<0.001), younger (mean 51.0 vs 54.6 years, p<0.001), female (90% vs 55%, p<0.001) and used the healthcare system more across all visit types retrieved over the study period than did patients with zero SLE encounters. Descriptive statistics of the CAPriCORN cohort are shown in table 1 and a comparison with 2020 census data of the Chicago Metropolitan Area in online supplemental figure 1.

Table 1

Descriptive statistics of CAPriCORN cohort across SLE encounter partitions

Criteria identification among SLE partitions

A summary of criteria positivity rates across SLE encounter partitions and the number of mean positive SLICC criteria is shown in table 2. The positivity rates of all individual criteria in the three sets were significantly different across the SLE encounter partitions. Some clinical criteria were still present in only a small percentage of individuals across all SLE partitions. Most notably, oral ulcers, alopecia, arthritis and haemolytic anaemia were identified at rates of 3.5%, 5%, 6.8% and 4.8% among patients with three or more SLE encounters, respectively. Some immunological criteria were also detected at low rates in individuals with three or more SLE encounters. One of particular importance was ANA, which was only present in 26.4% of patients. Finally, leucopenia was identified at exceptionally high rates across all SLE encounter partitions, with rates of 66%, 62% and 76% among the zero, one to two, and three or more SLE encounter groups, respectively. The mean number of SLICC criteria identified were 1.2, 2.1 and 3.4 for patients with zero, one to two, and three or more SLE encounters, respectively. Compared with the reference group with no documented SLE encounters, prevalence ratios for the SLICC criteria were 1.69 (1.66 to 1.72, p<0.001) and 2.77 (2.73 to 2.8, p<0.001) among the non-referent partitions in order of increasing SLE encounters. This trend was significant even when controlling for the degree of healthcare utilisation, shown in online supplemental figure 2. As reported in online supplemental tables 2 and 3, similar patterns were observed for the ACR and EULAR/ACR attribute sets, respectively.

Table 2

SLICC criteria identification across SLE partitions

Care fragmentation

Across all SLE encounter groups, patients with fragmented care were more often black or African American and had more documented inpatient, outpatient and emergency room visits than their counterparts with care from only one health system. These trends were especially pronounced in individuals with three or more SLE encounters; individuals with fragmented care in this group were 58% black or African American with 102, 5.52, and 8.8 mean outpatient, inpatient, and emergency room visits, respectively. This contrasts with patients within the same SLE encounter partition without fragmented care, of which only 37% were black or African American with 91, 2.9, and 3.7 mean outpatient, inpatient, and emergency room visits, respectively. Summary statistics of the SLE encounter partitions segmented based on care fragmentation status are shown in table 3.

Table 3

Demographic summary of patients with and without fragmented care

A summary of criterion positivity increases in the total number of attributes as well as across each individual SLICC criterion is shown in table 4A. In patients with fragmented care, the average number of positive SLICC attributes increased from 1.43 to 2.74, 2.15 to 3.93 and 3.34 to 5.7 in the zero, one to two, and three or more SLE encounter partitions, respectively, when comparing data from the topmost visited site (top site) versus data from all sites. These increases were statistically significant across all SLE encounter groups (p<0.001). At the individual criterion level, particular immunological domains displayed some of the largest per cent increases among patients with three or more SLE encounters. Of note, the rate of detection for ANA rose from 25% to 38% and dsDNA rose from 34% to 47% in this subgroup. Similar patterns were observed for the 1997 ACR and 2019 EULAR/ACR criteria sets, and are presented in online supplemental tables 4 and 5, respectively.

Table 4

(A) 2012 SLICC classification criteria attribute identification among patients with fragmented care; (B) rates of algorithm-classified SLE (all criteria sets)

The change in the number of individuals with algorithm-classified SLE following multisite data inclusion is summarised in table 4B. The percentage of individuals classified with SLE by the SLICC algorithm increased from 1.1% to 2.4% (p<0.05), from 15% to 26% (p<0.05) and from 38% to 56% (p<0.05) among the SLE partitions in order of increasing SLE encounters. The largest difference occurred within the subgroup of patients with three or more SLE encounters (18%, p<0.001). Also shown in table 4B, similar increases in algorithm-classified SLE were observed for the ACR and EULAR/ACR algorithms following multisite data inclusion.

Discussion

This study aimed to assess the performance and utility of algorithms designed to detect SLE attributes by implementing them in a large, urban data network. We first examined if these algorithms identified SLE attributes more often in patients with a greater number of SLE encounters as a measure of their performance. We observed a positive association between the number of positive SLE attributes detected and the number of SLE encounters during the study period. Patients with three or more SLE encounters were identified with roughly 2.5–3 times the number of positive criteria than patients without any SLE encounters, suggesting that these algorithms may be useful in identifying criteria set attributes in cohorts of patients with SLE.10 Patients with three or more SLE encounters also displayed higher rates of algorithm-classified SLE when compared with the referent group in all three criteria sets during our subanalysis in patients with fragmented care, further supporting their efficacy in accurately detecting SLE attributes in EHR data. In our investigation of multisite EHR data distribution among patients with fragmented care, we demonstrated that a significant number of criteria and per cent of individuals with algorithm-classified SLE significantly increased when data were included from all CAPriCORN sites, suggesting that SLE-specific information is often missing from single healthcare sites.

We also noted important patterns among individual classification criteria that are important for understanding the algorithms’ performance in this dataset and their future use in population surveillance. First, we found a positive ANA in only 24% of patients with three or more SLE encounters. ANA is a laboratory test with high sensitivity and estimated to be positive in roughly 95% of patients with SLE.25 26 It is possible that many patients in the three or more SLE encounters partition did exhibit a positive ANA at other healthcare sites or third-party testing centre not included in CAPriCORN or prior to a site’s EHR implementation, as repeat testing of ANA is not recommended, nor has it been found to be clinically useful.25 27 Evidence of these positive tests would, therefore, be unlikely included in a structured data concept analysed by these algorithms and only found in provider assessments of patients. Specific to CAPriCORN, ANA tests may also not have been properly associated with a structured medical terminology, which would have also prevented its identification. Similar patterns of low detection were seen in other criteria such as alopecia, oral ulcers and arthritis. These attributes may be found more often in narrative documentation as opposed to standardised medical terminologies like ICD codes given that these attributes are often documented during the course of care but may not be coded for billing. Multimodal approaches for detection of concepts in EHR data that incorporate processing of unstructured data elements have been shown to perform better than using structured data alone.28 The low detection rates that we observed among some of the classification criteria suggest that these multimodal approaches may be a valuable addition to our algorithms to improve their sensitivity in detecting certain attributes. The opposite pattern was observed among other criteria, with extremely high rates of detection across all SLE partitions. For example, leucopenia, while detected significantly more often in patients with three or more SLE encounters, was still identified by the algorithms in 67% of patients without an SLE encounter. Leucopenia occurs in approximately half of all patients with SLE,29 with a multifactorial aetiology that can be difficult to attribute directly to SLE.30 Our algorithm for identifying leucopenia includes rules for excluding other aetiologies unrelated to SLE. However, the high rates of leucopenia detection we observed across the entire cohort suggest that this rules-based approach may not cover all possible confounding causes of leucopenia when implemented in a generalised data context like in CAPriCORN.

There are several strengths to our work and the data sources we used. First, the cohort retrieved from CAPriCORN, while containing a slightly higher proportion of white and female patients than recent census estimates of Chicago, represents a demographically diverse group of patients.31 Furthermore, the partition of patients with three or more SLE diagnosis encounters more frequently contained black or African American patients, consistent with recent epidemiological estimates of SLE in the USA32 and serving as a secondary data quality measure that supports accurate identification of patients with SLE. Finally, the data linkage strategies employed by CAPriCORN18 retained protected health information within each organisation while still achieving successful linkage across sites.

Our study is limited by certain characteristics inherent to EHR data and features of the dataset we used for this current work. Our ability to fully assess the algorithms’ performance in accurately identifying criteria set attributes is limited by the anonymisation of patients and the site from which data were obtained. Therefore, we were unable to manually review charts of patients retrieved from our query for disease activity, medications and laboratory procedures. As a result, some patients may have been unknowingly misclassified by our algorithms due to differences in the underlying elements used in the PCORnet data model from the data present at northwestern or variations in data between sites. Future work will focus on manually adjudicated cohorts from multiple healthcare organisations to comprehensively assess their performance in contexts like CAPriCORN. In relation to our proof-of-concept investigation into care fragmentation and information gain, we did not assess the degree to which missing SLE-related information at a patient’s topmost visited site impacted clinician decision-making and patient outcomes. While the presence of particular criteria can influence clinician recommendations for screening and intervention,33 we were unable to investigate the effect of missing information with the data available to us. Future work will focus on missing SLE information to investigate its role in poorer outcomes observed among patients with fragmented care.19

Our findings have several implications for surveillance and population health management of SLE. First, the higher rate of detection of all criteria among patients with more SLE diagnoses supports their potential use in identification and characterisation of patients for use in SLE registries and longitudinal surveillance systems. Normally, evaluation of a patient for the presence of SLE requires a sizeable amount of time and labour in identifying classification criteria and other clinical indicators of SLE. The algorithms implemented in this study could be used to supplement this process through automated detection of criteria for experts to assess without the need for reviewing patients’ entire medical record. Our proof-of-concept investigation into care fragmentation supports additional use cases for these algorithms beyond reduction of chart review burden. First, we were able to investigate a hypothesis that may explain poorer outcomes observed among patients with fragmented care19 and show that SLE-specific information was often missing at sites that patients visited most often. These findings and follow-up studies on the effects of missing information on patient outcomes could be used to inform healthcare organisation stakeholders on the effective design of EHR data architecture and interorganisational collaboration to optimise care for patients with SLE. We also demonstrated that a significant amount of SLE attributes were identified with multisite data inclusion even among patients without any documented SLE encounters. This may support future use of these algorithms in combination with coordination between local health networks to highlight patients for further evaluation and allow for earlier SLE detection, intervention and prevention of disease-related damage.33

In conclusion, our findings suggest that these algorithms maintained their ability to find true SLE attributes in broad data settings. We also observed criteria positivity rates that were different than expected among a select number of attributes, highlighting the need for multimodal data incorporation to allow for improved detection of attributes such as ANA that are nearly universally present in SLE and excluding confounding instances of others such as leucopenia. Finally, using structured data alone, we were able to demonstrate the utility of these algorithms by showing that a significant proportion of SLE-specific information is often missing at sites that patients visit most often, which may partially explain the disparity in outcomes observed among patients with fragmented care.19

RR-G and TLW are joint senior authors.
Twitter: @TheresaWalunas
Contributors: NF was involved in the analysis and interpretation of data, design of the project and drafting the manuscript. KLJ, AF and JAP were involved in the analysis and interpretation of data. ST was involved in the analysis and interpretation of data and critical review of the manuscript. VM and ANK were involved in the acquisition and management of data. TLW was involved in the conception and design of the project, acquisition and interpretation of data and critical review of the manuscript. RR-G was involved in the conception and design of the project, acquisition, analysis and interpretation of the data and critical review of the manuscript. All authors were involved in manuscript review and revision and approval of the final draft. TLW is the guarantor of this manuscript.
Funding: This work was supported by CAPriCORN and the National Institute of Arthritis and Musculoskeletal Disease (5R21AR072262 and P30AR072579); the National Human Genome Research Institute and Electronic Medical Records and Genomics (eMERGE) Network (U01HG008657); and PCORI contract (RI-CRN-2020-002).
Disclaimer: The content of this article is solely the responsibility of the authors and does not necessarily represent the official view of any of the authors’ funding sources.
Competing interests: ANK is a strategic advisor for Datavant. TLW receives research support from Gilead Sciences. RR-G has been a consultant to Exagen Diagnostics, ThermoFisher, Ampel Solutions, Calabetta Bio and Biogen (all less than $10 000) and none relevant to the current research project.
Patient and public involvement: Patients and/or the public were involved in the design, or conduct, or reporting, or dissemination plans of this research. Refer to the Methods section for further details.
Provenance and peer review: Not commissioned; externally peer reviewed.
Supplemental material: This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.

Data availability statement

Data are available upon reasonable request. Aggregated data used and analysed during the current study are available from the corresponding author on reasonable request. The row-level data used and analysed in this study are not to be shared with any individual or organisation not covered by the agreement made with CAPriCORN.

Ethics statements

Patient consent for publication:

Ethics approval:

This study involves human participants but all queries performed on CAPriCORN, including the one used to retrieve data for this current study, are governed by the Chicago Area Institutional Review Board (CHAIRb). The query in this study did not require the use or return of PHI and is considered a fully de-identified dataset and received a not human subjects research determination. All of the data in CAPriCORN are de-identified and linked via a secure hashing algorithm. Therefore, the dataset obtained for this study represents a low risk of re-identification and was deemed not necessary to obtain consent from patients as part of the CHAIRb protocol outlined in CAPriCORN.

Acknowledgements

The authors wish to thank the Chicago Area Patient-Centered Outcomes Research Network for their support of this project and data contributions.

Murphy G, Lisnevskaia L, Isenberg D, et al. Systemic lupus erythematosus and other autoimmune rheumatic diseases: challenges to treatment. The Lancet 2013; 382:809–18.
doi:10.1016/S0140-6736(13)60889-2•Google Scholar
Fava A, Petri M. Systemic lupus erythematosus: diagnosis and clinical management. J Autoimmun 2019; 96:1–13.
doi:10.1016/j.jaut.2018.11.001•Google Scholar
Bruce IN, O’Keeffe AG, Farewell V, et al. Factors associated with damage accrual in patients with systemic lupus erythematosus: results from the systemic lupus International collaborating clinics (SLICC) inception cohort. Ann Rheum Dis 2015; 74:1706–13.
doi:10.1136/annrheumdis-2013-205171•Google Scholar
Aggarwal R, Ringold S, Khanna D, et al. Distinctions between diagnostic and classification criteria. Arthritis Care Res (Hoboken) 2015; 67:891–7.
doi:10.1002/acr.22583•Google Scholar
Tan EM, Cohen AS, Fries JF, et al. The 1982 revised criteria for the classification of systemic lupus erythematosus. Arthritis Rheum 1982; 25:1271–7.
doi:10.1002/art.1780251101•Google Scholar
Hochberg MC. Updating the American college of rheumatology revised criteria for the classification of systemic lupus erythematosus. Arthritis Rheum 1997; 40.
doi:10.1002/art.1780400928•Google Scholar
Petri M, Orbai A-M, Alarcón GS, et al. Derivation and validation of the systemic lupus International collaborating clinics classification criteria for systemic lupus erythematosus. Arthritis & Rheumatism 2012; 64:2677–86.
doi:10.1002/art.34473•Google Scholar
Aringer M, Costenbader K, Daikh D, et al. European League against rheumatism/American college of rheumatology classification criteria for systemic lupus erythematosus. Arthritis Rheumatol 2019; 71:1400–12.
doi:10.1002/art.40930•Google Scholar
Aringer M, Costenbader K, Daikh D, et al. European League against rheumatism/American college of rheumatology classification criteria for systemic lupus erythematosus. Ann Rheum Dis 2019; 78:1151–9.
doi:10.1136/annrheumdis-2018-214819•Google Scholar
Walunas TL, Ghosh AS, Pacheco JA, et al. Evaluation of structured data from electronic health records to identify clinical classification criteria attributes for systemic lupus erythematosus. Lupus Sci Med 2021; 8.
doi:10.1136/lupus-2021-000488•Google Scholar
Fleurence RL, Curtis LH, Califf RM, et al. Launching Pcornet, a national patient-centered clinical research network. J Am Med Inform Assoc 2014; 21:578–82.
doi:10.1136/amiajnl-2014-002747•Google Scholar
Kaushal R, Hripcsak G, Ascheim DD, et al. Changing the research landscape: the New York City clinical data research network. J Am Med Inform Assoc 2014; 21:587–90.
doi:10.1136/amiajnl-2014-002764•Google Scholar
Courcoulas A, Coley RY, Clark JM, et al. Interventions and operations 5 years after Bariatric surgery in a cohort from the US national patient-centered clinical research network Bariatric study. JAMA Surg 2020; 155:194–204.
doi:10.1001/jamasurg.2019.5470•Google Scholar
Amin W, Singh H, Dzubinski LA, et al. Design and utilization of the colorectal and Pancreatic Neoplasm virtual Biorepository: an early detection research network initiative. J Pathol Inform 2010; 1.
doi:10.4103/2153-3539.70831•Google Scholar
Trick WE, Rachman F, Hinami K, et al. Variability in Comorbidites and health services use across homeless Typologies: multicenter data linkage between Healthcare and homeless systems. BMC Public Health 2021; 21.
doi:10.1186/s12889-021-10958-8•Google Scholar
Jones WS, Mulder H, Wruck LM, et al. Comparative effectiveness of aspirin dosing in cardiovascular disease. N Engl J Med 2021; 384:1981–90.
doi:10.1056/NEJMoa2102137•Google Scholar
Raiker R, Pakhchanian H, DeYoung C, et al. Short term outcomes of COVID-19 in lupus: propensity score matched analysis from a nationwide multi-centric research network. J Autoimmun 2021; 125:102730.
doi:10.1016/j.jaut.2021.102730•Google Scholar
Kho AN, Hynes DM, Goel S, et al. Capricorn: Chicago area patient-centered outcomes research network. J Am Med Inform Assoc 2014; 21:607–11.
doi:10.1136/amiajnl-2014-002827•Google Scholar
Walunas TL, Jackson KL, Chung AH, et al. Disease outcomes and care fragmentation among patients with systemic lupus erythematosus. Arthritis Care Res (Hoboken) 2017; 69:1369–76.
doi:10.1002/acr.23161•Google Scholar
Wright A, Pang J, Feblowitz JC, et al. A method and knowledge base for automated inference of patient problems from structured data in an electronic medical record. J Am Med Inform Assoc 2011; 18:859–67.
doi:10.1136/amiajnl-2011-000121•Google Scholar
Feldman CH, Hiraki LT, Liu J, et al. Epidemiology and Sociodemographics of systemic lupus erythematosus and lupus nephritis among US adults with Medicaid coverage, 2000-2004. Arthritis Rheum 2013; 65:753–63.
doi:10.1002/art.37795•Google Scholar
Bernatsky S, Linehan T, Hanly JG, et al. The accuracy of administrative data diagnoses of systemic autoimmune rheumatic diseases. J Rheumatol 2011; 38:1612–6.
doi:10.3899/jrheum.101149•Google Scholar
PCORI. The National Patient-Centered Clinical Research Network: Patient-Centered Outcomes Research Institute. 2018;
Available: here
Google Scholar
Services UDoHaH. Explanation of data standards for race, Ethnicity, sex, primary language, and disability. 2021;
Available: here
Google Scholar
Fritzler MJ. Choosing wisely: review and commentary on anti-nuclear antibody (ANA) testing. Autoimmun Rev 2016; 15:272–80.
doi:10.1016/j.autrev.2015.12.002•Google Scholar •PubMed
Choi MY, Clarke AE, St Pierre Y, et al. Antinuclear antibody–negative systemic lupus erythematosus in an international inception cohort. Arthritis Care Res (Hoboken) 2019; 71:893–902.
doi:10.1002/acr.23712•Google Scholar •PubMed
Yeo AL, Le S, Ong J, et al. Utility of repeated Antinuclear antibody tests: a retrospective database study. The Lancet Rheumatology 2020; 2:e412–7.
doi:10.1016/S2665-9913(20)30084-9•Google Scholar
Peissig PL, Rasmussen LV, Berg RL, et al. Importance of multi-modal approaches to effectively identify cataract cases from electronic health records. J Am Med Inform Assoc 2012; 19:225–34.
doi:10.1136/amiajnl-2011-000456•Google Scholar •PubMed
Keeling DM, Isenberg DA. Haematological manifestations of systemic lupus erythematosus. Blood Rev 1993; 7:199–207.
doi:10.1016/0268-960x(93)90006-p•Google Scholar •PubMed
Levine AB, Erkan D. Clinical assessment and management of Cytopenias in lupus patients. Curr Rheumatol Rep 2011; 13:291–9.
doi:10.1007/s11926-011-0179-5•Google Scholar •PubMed
QuickFacts: Chicago City, Illinois: US Census Bureau.
Available: here
Google Scholar
Stojan G, Petri M. Epidemiology of systemic lupus erythematosus: an update. Curr Opin Rheumatol 2018; 30:144–50.
doi:10.1097/BOR.0000000000000480•Google Scholar •PubMed
Izmirly PM, Parton H, Wang L, et al. Prevalence of systemic lupus erythematosus in the United States: estimates from a Meta‐Analysis of the centers for disease control and prevention national lupus registries. Arthritis Rheumatol 2021; 73:991–6.
doi:10.1002/art.41632•Google Scholar •PubMed

Overview

Abstract
Introduction
Methods
Results
Discussion
References

Supplementary files
Footnotes
Publication history
Rapid Responses

Article metrics

Altmetric

Dimensions