Introduction

Antiphospholipid syndrome (APS) is a systemic autoimmune disease characterised by arterial, venous, or microvascular thrombosis, pregnancy morbidity, or nonthrombotic manifestations in patients with persistent antiphospholipid antibodies (aPL). Classification of APS, for the identification of homogeneous research cohorts, is currently based on the Sapporo criteria published in 19991 and revised in 2006.2 The revised Sapporo criteria for APS require clinical features (thrombosis or pregnancy morbidity) and laboratory tests (for lupus anticoagulant (LAC), IgG/IgM anticardiolipin antibodies (aCL), and/or IgG/IgM anti–β2-glycoprotein I antibodies [anti-β2GPI ]) with at least 2 aPL tests performed at least 12 weeks apart.2

Since the introduction of the Sapporo criteria, advancements in our understanding of APS include better characterisation of aPL-associated nonthrombotic clinical manifestations, identification of the role of traditional thrombosis risk factors in aPL-positive individuals, and risk stratification by aPL profile.3 4 Furthermore, the revised Sapporo criteria have been criticised for not incorporating evidence-based definitions, e.g., aPL positivity, microvascular disease, or pregnancy morbidity, resulting in the inclusion of a heterogeneous group of “aPL-positive” patients with different risk profiles for research.4 5 More stringent methodology, using data-driven and expert-based approaches to develop robust classification criteria in rheumatic diseases, is now available.6 Thus, new classification criteria can better ensure future high-quality, risk-stratified epidemiologic studies and clinical trials in APS, leading to improved patient care and management recommendations.

Given the limitations of the current criteria,7–9 an international effort, jointly supported by the American College of Rheumatology (ACR) and EULAR, was initiated with the goal of using rigorous methodology to develop a new APS classification system based on a more modern disease understanding, allowing for the weighting of individual criterion, and demonstrating excellent operating characteristics with the highest possible specificity. Maximising the specificity of the 2023 ACR/EULAR APS classification criteria was a major goal at the outset, as overly inclusive criteria may decrease the ability of investigators to understand disease pathophysiology and treatment effects in clinical trials and research.

MethodsMethodologic overview

Our 4-phase methodology (see online supplemental section 1) was similar to the methodologies used in the development of recent rheumatic disease classification criteria.10–16 The phases were as follows: Phase I, criteria generation; Phase II, criteria reduction; Phase III, criteria definition, further reduction, and weighting through a consensus-based multicriteria decision analysis (MCDA) methodology,17–19 as well as classification threshold identification; and Phase IV, validation.

The initiative was overseen by a 24-member international multidisciplinary Steering Committee, led by principal investigators from North America (DE) and Europe (SZ); 13 members were from the Americas, 9 from Europe, and two from New Zealand. Steering Committee members were selected based on their expertise in APS and/or methodologies; 3 patients (US and Europe) represented the patient experience. The Steering Committee assembled (1) a core planning group; (2) a master group of 54 international physician-scientists designated as “Collaborators” (40% from Europe, 40% from North America, and 20% from South America, based on their clinical and/or research APS interest); and (3) domain-specific subcommittees. Members of these relevant Steering Committee groups are listed below in the collaborators section.

Phase I (criteria generation) and Phase II (criteria reduction) overview

In Phase I, we generated a comprehensive list of candidate criteria, using both consensus-based and evidence-based methods. We e-mailed a survey with open-ended questions to the master group (n=54) to identify potential criteria and different APS subpopulations. We systematically clustered responses by organ systems to avoid duplication and improve interpretability, and reviewed the literature for additional items.

In Phase II, we reduced the generated list using systematic reviews, meta-analyses,20–24 and expert consensus. We administered two consecutive surveys (61 expanded master group members, 19 Steering Committee members) assessing the specificity of each Phase I item in differentiating APS from similar conditions. We ranked items by mean survey score, hierarchically organised them into domains by specificity, and eliminated low-specificity items by nominal group technique during an in-person meeting.25 26 Within each domain, the Steering Committee agreed that only the highest specificity item should be scored, consistent with classification criteria methodology.6 In addition, the Steering Committee discussed the need for “entry criteria,” that is, minimum criteria required to identify the relevant patient population to whom the classification criteria would be applied (for further details on Phases I and II, see Barbhaiya et al 8).

Phase III (criteria definition, further reduction, and weighting, and classification threshold identification) overview

In Phase III, we defined criteria generated during Phase I/II, as part of clinical (Phase III-A) and laboratory (Phase III-B) domains, further reduced the number of criteria using expert consensus and real-world patient scenarios (Phase III-C), and determined criteria weights and the threshold above which cases would be consistently classified as APS (Phase III-D). We also finalised the entry criteria. For details, see online supplemental section 3.

Phase IV (validation) overview

In Phase IV, using two separate validation cohorts, we compared performance characteristics of the revised Sapporo criteria to those of the new APS classification criteria against consensus by independent adjudicators, that is, representing the “gold standard.” We made an a priori decision to have two validation cohorts, in order to demonstrate consistency and validity. We assembled cohorts by asking Phase IV Collaborators (selected among the original 54 members and 20 additional members, none of whom were involved in Phase III) to contribute 30 cases evaluated for “APS suspicion,” that is, a clinical APS manifestation with any positive or negative aPL test result or no aPL test, or a positive aPL test result with no clinical APS manifestation. Of the 30 cases, half were considered “likely” and half “not likely” to be APS for research purposes. We collected clinical and laboratory data relevant to the revised Sapporo criteria and new classification criteria using a standardised form; cases were randomly assigned to two different cohorts.

APS classification for each case was verified by three independent adjudicators (a rheumatologist from North America (RR), an internist/clinical immunologist from Europe (ZA), and a haematologist from Europe (HC)), who were blinded with regard to the Phase III discussions and unaware of the proposed classification criteria. Adjudicators participated in a moderated discussion of discordant cases at the end of each validation cohort assessment until agreement was reached. Moderated discussions were aimed at focusing adjudicators on factors relevant for APS case classification, avoiding additional information to reduce bias.

Phase IV (validation) statistical analysis

Assuming a total discordance of 20%, that is, disagreement between expert consensus and new criteria, a power of 80%, and an alpha risk of 5%, an estimated sample size of 243 would be necessary to detect a difference in performance characteristics between the revised Sapporo and new APS classification criteria. We incorporated two validation cohorts (n=284 per cohort) in Phase IV. Sensitivity, specificity, and Wald 95% confidence intervals (95% CIs) for each validation cohort were comparatively evaluated for the revised Sapporo and new APS classification criteria, each against independent adjudicators’ consensus. Nonoverlapping CIs and a P value threshold of<0.05 denoted significance. For details, see online supplemental section 6.

Statistical analyses were performed using SAS version 9.4 (SAS Institute). This study was approved by the Hospital for Special Surgery Institutional Review Board, and by individual centres as needed.

ResultsPhase I (criteria generation) and phase II (criteria reduction)

Phase I generated 152 candidate criteria, expanded to 261 items with subgroups and candidate criteria with potential negative weights. Subsequent reduction methods resulted in 27 candidate criteria, hierarchically organised into six additive domains (macrovascular, microvascular, obstetric, cardiac valve, hematologic, and laboratory) (for details, see the Table in online supplemental section 2).

During an in-person meeting, the Steering Committee agreed that to maximise specificity, candidate clinical criteria must be interpreted in the context of a “clinically acceptable” aPL profile, emphasising the importance of “entry criteria.” Following this meeting, modified Delphi exercises were carried out, as follows: (1) all members voted in favour of entry criteria requiring at least 1 clinical and one laboratory criterion; and (2) the majority voted in favour of a time restriction between the clinical and laboratory criterion as part of the entry criteria.8

Phase III (criteria definition, further reduction, and weighting, and classification threshold identification)Phase III-A, clinical definitions

The Steering Committee developed clinical candidate definitions (Phase III-A) for the following features: (1) macrovascular thrombosis and traditional venous thromboembolism (VTE) and cardiovascular disease (CVD) risk factors (see online supplemental section 8); (2) microvascular disease; (3) pregnancy morbidity; (4) cardiac valve involvement; and (5) thrombocytopenia (tables 1 and 2) (details will be published elsewhere).27

Table 1

Definitions of the 2023 ACR/EULAR antiphospholipid syndrome (APS) classification criteria

Table 2

Definitions of high-risk venous thromboembolism (VTE) and cardiovascular disease (CVD) profiles based on current general population guidelines (refer to online supplemental section 8 for details)

Phase III-B, laboratory definitions

The Steering Committee agreed on the following criteria for laboratory items (Phase III-B): (1) International Society on Thrombosis and Haemostasis (ISTH) guidelines should be followed for LAC testing and interpretation (table 1)27; and (2) single (one-time) LAC positivity may be relevant when repeat testing is unavailable. Pending assessment and refinement during the subsequent Phase III-C, the Steering Committee recommended that (1) there should be 2 levels of aCL/anti-β2GPI positivity (“moderate” and “high” positivity) based on enzyme-linked immunoassay (ELISA) techniques; (2) IgG aCL and IgG anti-β2GPI positivity should be evaluated in combination; and (3) IgG and IgM isotypes should not be additively considered (details will be published elsewhere).

Finally, given the limited data for or against the definition of aPL “persistence” (ie, 2 positive tests for aPL at least 12 weeks apart),1 2 28 the Steering Committee decided not to change the definition.

Phase III-C, finalisation of the entry criteria

The Steering Committee agreed that an entry criteria time restriction of 3 years (vs 5 years in the revised Sapporo criteria) between a clinical criterion and a positive aPL test result improves certainty for APS classification; however, this decision was based on limited data29 and on primarily Steering Committee consensus (modified Delphi exercise). The final entry criteria, requiring the presence of at least one clinical criterion and one laboratory criterion (positivity for LAC, moderate- or high-level IgG/IgM aCL positivity, or moderate- or high-level IgG/IgM anti-β2GPI positivity) within 3 years of each other, are presented in figure 1.

Figure 1

2023 ACR/EULAR classification criteria. (a) Refer to table 1 for the definitions of clinical and laboratory criteria including the moderate- and high-titer anticardiolipin antibody (aCL) IgG/IgM or anti -β2-glycoprotein I antibody (aβ2GPI) IgG/IgM positivity. (b) Antiphospholipid antibody (aPL) positivity must be confirmed within +/ - three years of the documented (by medical records) clinical criterion. (c) Refer to table 2 for the definitions of high-risk profiles. (d) Suspected microvascular definition for each corresponding item should be first fulfilled. (e) For the purpose of laboratory domain scoring: 1) "persistent" aPL test results (at least 12 weeks apart) should be scored based on two consecutive positive lupus anticoagulant (LAC), two consecutive highest aCL, and/or two consecutive highest aβ2GPI results (two consecutive results with one moderate positive and one high positive aCL/aβ2GPI should be marked as "moderate positive" if there are no additional consecutive high results available); 2) for prospective data collection, two consecutive positive aPL results are required within three years of the clinical criterion; 3) for retrospective data collection, two consecutive positive aPL results and at least one positive aPL result within three years of the clinical criterion are required; 4) if there are multiple LAC assays performed on patients receiving anticoagulants (vitamin K antagonists, heparin, direct oral anticoagulants, indirect Factor Xa inhibitor), the results of the tests performed without anticoagulants should be included in the assessment unless the results of the tests performed with anticoagulants are reviewed/confirmed by an individual who has expertise in performing/interpreting the LAC assay (refer to table 1 for details); 5) moderate (40-79U) and high (>80U) level aCL/aβ2GPI are based on enzyme-linked immunosorbent assays (ELISAs) (refer to table 1 for details); and 6) for prospective studies, the most recent aPL test (LAC and/or moderate-high level aCL/aβ2GPI) should be positive to maintain homogeneity of research cohorts. BAL, bronchoalveolar lavage; CVD, cardiovascular disease; D1–D8, domains 1–8; exam, physical examination; lab, laboratory tests.

Phase III-C, real-world case collection and analysis

Of 314 potential APS cases in the derivation cohort (mean±SD age 44.7±14.6 years; 79% female) collected from 17 sites, including eight from Europe (47%), 7 from North America (41%), and two from South America (12%), case collectors rated 137 cases (44%) as “highly likely” and 177 cases (56%) as “equivocal or unlikely” to be APS. Duration between aPL positivity and any candidate clinical criteria was≤3 years in 89% of cases.30

Phase III-C, consensus discussions for further criteria reduction and final definitions

Discussions and decisions based on derivation cohort results, literature review, and expert consensus are summarised in online supplemental section 5. Eventually, the definitions and hierarchical order of items within each of 8 additive and independent domains (six clinical, 2 laboratory) were finalised (table 3). The Steering Committee also concluded that (1) patients with concomitant systemic autoimmune disease could be classified as having APS, but individual candidate criterion should not be scored if other “equally likely” or “more likely” causes for that criterion cannot be excluded, similar to other criteria sets12; (2) “persistent” aPL should be scored based on two consecutive results; and (3) “moderate” level aCL/anti-β2GPI positivity should be defined as 40–79 ELISA units and “high” level as≥80 ELISA units.

Table 3

Relative weights of additive classification criteria items based on 1000Minds analysis for antiphospholipid syndrome (APS) classification

Phase III-D, criteria weighting based on MCDA and classification threshold identification

During the in-person meeting, the MCDA exercise calculated weights based on 81 pairwise consensus-based decisions. Table 3 shows the resulting point-based classification system, with hierarchical levels in each domain identified based on their relative weights.

Following the in-person meeting, the minimum classification threshold was determined based on individual assessment of the 192 unique derivation cohort cases remaining after eliminating duplicates and cases not meeting the entry criteria. Of 192 cases, full agreement with APS classification was achieved for 116 cases (60%) (90 classified as APS, 26 as not APS). Agreement was relatively high for 37 cases (19%), with 80–93% agreeing with the classification (17 as APS, 20 as not APS). However, there were variable responses (50–80%) for the remaining 39 cases (20%).

Within each domain, descriptive analysis of the 192 cases showed that most respondents considered that 1) the presence of 1 “B” level clinical criterion, even with “C” or higher-level laboratory criteria, was insufficient for APS classification, but two or more “B” level (and/or one or more “C” or higher-level) clinical criteria were acceptable; and 2) the presence of 1 or 2 “B” level laboratory criteria, even with “C” or higher-level clinical criteria, was insufficient for APS classification (table 3).

During several teleconferences, all cases without 100% agreement were discussed with the guidance of the descriptive analysis until full consensus was achieved. Key conceptual issues addressed included the following: (1) the need to emphasise specificity over sensitivity to improve homogeneity of APS patients in research and to avoid enrolling misclassified patients in clinical trials with potentially toxic investigational medications; and (2) structuring the classification system to include an acceptable clinical criterion and an acceptable aPL laboratory criterion. Relative weights derived from 1000Minds analysis supported these ranking exercises, with one exception: combined weights of the “B” level macrovascular (VTE) and obstetric domains were low; the Steering Committee agreed that the combination would not meet the threshold for an acceptable clinical profile. As a result, consensus for the preliminary threshold for APS classification was achieved.

Despite Steering Committee agreement on the “APS” threshold, detailed analysis of the 39 cases with variable responses demonstrated the most frequently encountered controversial scenarios. These scenarios were (1) moderate- or high-titre IgM aCL/anti-β2GPI alone (“B” level) (table 3) with an acceptable clinical criterion (12 [31%] of 39 cases); (2) VTE or arterial thrombosis alone in patients with high-risk profiles for VTE or CVD (“B” level), with an acceptable laboratory criterion (9 [23%] of 39 cases); and (3) occurrence of 3 or more consecutive prefetal deaths (at<10 weeks) and/or early fetal deaths (at 10–16 weeks), or one or more fetal deaths (at≥16 weeks to<34 weeks) alone (“B” level) in the context of an acceptable laboratory criterion (8 [21%] of 39 cases).

The 2023 ACR/EULAR APS classification criteria are presented in figure 1. According to these criteria, patients should be classified as having APS if they fulfil the entry criteria (at least 1 clinical and one laboratory criterion within 3 years of each other) and accumulate at least three points from clinical domains and three points from laboratory domains.

Phase IV (validation)

We collected 568 potential APS cases from 29 international sites, including Europe (16 centers [55%]), North America (11 centers [38%]), South America (one center [3%]), and Asia (one center [3%]), to assess the performance characteristics of the preliminary classification criteria.

In the first validation cohort (n=284), independent adjudicators classified 98 cases (35%) as “APS” and 180 (63%) as “No APS.” Six cases (2%) were excluded—one case was excluded due to unresolved disagreement on classification, and 5 cases were excluded due to being unclassifiable because of incomplete data.

Following assessment of the first cohort, assessment of the second validation cohort was carried out based on adjudicators’ recommendations, as follows: (1) the definition of placental insufficiency was further characterised (table 1); and (2) each case was assessed using complete individual VTE/arterial thrombosis risk factor data, rather than the overall risk factor profile. In the second validation cohort (n=284), the adjudicators classified 113 cases (40%) as “APS” and 162 cases (57%) as “No APS”; 9 subjects (3%) were excluded as they were unclassifiable due to incomplete data.

Characteristics of the first validation cohort (n=278) and second validation cohort (n=275) are shown in table 4. Of the 553 patients, the age of the majority of them was 40 years or higher, and the cohort was predominantly White and female, consistent with APS demographics from other international cohorts.31 For both validation cohorts, the operating characteristics of the 2023 ACR/EULAR APS classification criteria, using the independent adjudicators’ consensus as the gold standard, demonstrated very high specificity of 99% in each cohort (95% CI 0.98 to 1.00 in cohort 1, and 95% CI 0.97 to 1.00 in cohort 2), whereas the revised Sapporo criteria for APS had a specificity of 91% (95% CI 0.86 to 0.95) in cohort 1 and 86% (95% CI 0.81 to 0.92) in cohort 2. The sensitivity of the new ACR/EULAR APS criteria was 83% (95% CI 0.75 to 0.90) in cohort 1 and 84% (95% CI 0.77 to 0.91) in cohort 2, compared with a sensitivity of 100% (95% CI 1.00 to 1.00) in cohort 1 and 99% (95% CI 0.98 to 1.00) in cohort two using the revised Sapporo criteria (table 5) (see online supplemental section 7 for further analysis).

Table 4

Demographic and clinical characteristics of 553 cases used in both validation cohorts, by independent adjudicators’ consensus for antiphospholipid syndrome (APS) classification

Table 5

Operating characteristics of the 2023 ACR/EULAR antiphospholipid syndrome (APS) classification criteria versus the revised Sapporo APS classification criteria compared against independent adjudicators’ consensus in two distinct validation cohorts

Discussion

The 2023 ACR/EULAR APS classification criteria comprise an additive, weighted system, assessing an individual’s relative probability of APS and defining a threshold for APS classification for research purposes. The new criteria were developed in four rigorous phases under the guidance of international physician-scientists experienced in the evaluation and management of APS patients, while utilising international cohorts totaling approximately 900 patients spanning the spectrum of APS. The new criteria are a paradigm shift in APS classification, given that: (1) these carefully defined clinical and laboratory criteria, based on literature review and expert consensus, improve the reliability and precision of classification; (2) the criteria are differentially weighted and organised into eight hierarchical domains; and (3) the criteria were validated based on two international cohorts of patients referred for suspicion of APS, demonstrating very high specificity (99%) relative to the revised Sapporo criteria (86%).

In contrast to making a diagnosis, which requires consideration of a broad range of features (including rare ones), available clinical tests, and differential diagnoses pertaining to the epidemiology in a specific region, the goal of classification criteria is to enrol individuals with a condition of interest manifesting key features of the disease to form relatively homogeneous cohorts for comparability across clinical studies and trials.32 Thus, classification criteria intentionally include standardised and stringent definitions32; very high specificity is required, even at the cost of sensitivity. Our goal was to achieve high specificity relative to the revised Sapporo criteria to improve homogeneity in APS research. While 99% specificity is a highly desirable performance characteristic of the new criteria for clinical trials and studies, the sensitivity of 84% captures a broad spectrum of patients referred for APS suspicion in whom the investigators are confident of APS classification.

The novel clinical features of the new APS classification criteria include the following: (1) risk stratification of macrovascular events by traditional thrombosis risk factors (although the revised Sapporo criteria acknowledged the need to recognise subgroups with and without thrombosis risk factors [2], our criteria are the first to offer a weighted assessment); (2) well-defined microvascular domain items thought to be mechanistically distinct from moderate-to-large vessel disease; (3) re-structured definitions of pregnancy morbidity to improve patient selection in obstetric studies; and (4) the addition of cardiac valve disease and thrombocytopenia, to capture and quantify the magnitude of heterogeneous APS manifestations.

The novel laboratory features of the new APS classification criteria include the following: (1) quantifying single-, double-, and triple-aPL positivity based on different domains and weights; (2) separating aCL/anti-β2GPI IgG and IgM isotypes, to avoid including aPL-positive patients with isolated aCL/anti-β2GPI IgM isotypes (ie, no other aPL positivity) in the same research studies as those with aCL/anti-β2GPI I IgG isotypes; and (3) defining 2 levels of aCL/anti-β2GPI positivity that will be interpreted as clinically relevant by most investigators. These decisions were based on literature reviews,33 Phase III-C relative risk analyses,30 and Steering Committee consensus (for details as well as the rationale for not including IgA isotypes or other solid-phase assay–based aPL tests, see online supplemental section 5). Although the Steering Committee agreed that only LAC assays and aCL/anti-β2GPI ELISAs should be included to ensure homogeneity, because automated laboratory systems are increasingly used in various countries, the Steering Committee suggested further studying the moderate/high thresholds in new automated platforms in association with clinical criteria from the new classification criteria (table 1).

During the development and validation phases, we identified 3 “controversial” clinical scenarios not meeting the APS classification threshold by Steering Committee consensus but rated as APS by outside adjudicators, that is, “false negatives” by the new criteria. These scenarios, and others below the threshold, are equivocal or uncertain for classification purposes, given the lack of strong literature support and physician agreement. Because the Steering Committee achieved a clear APS classification threshold above the controversial cases, as supported by the literature, and agreed that highly specific classification criteria are imperative for achieving homogeneity in APS research, along with ethical concerns of enrolling patients with controversial scenarios in the same clinical trials (e.g., trials of long-term anticoagulation therapy) as patients with highly likely APS, the Steering Committee deemed it acceptable to exclude these subgroups from current APS classification but to further study them independently (table 6).

Table 6

High-priority antiphospholipid syndrome (APS) research agenda to guide the future update of the 2023 ACR/EULAR APS classification criteria

While using the new classification criteria, researchers should pay attention to certain points. First, clinical expertise and attentiveness are required to attribute clinical criteria to APS; as this can be challenging when “equally or more likely” causes exist, the item in question should not be scored. For example, in the clinical scenario of an acceptable aPL profile and concomitant heparin-induced thrombocytopenia, the hematologic domain should not be scored. Similarly, in a patient with systemic lupus erythematosus (SLE) with an acceptable aPL profile and preeclampsia, the obstetric item should not be scored if the preeclampsia can be equally or more likely explained as attributable to SLE. Second, as the primary goal is to ensure high-quality prospective studies and clinical trials, complete information on patients’ VTE and CVD risk profiles is essential to evaluate the macrovascular domain. However, immediate real-world implementation of this concept may be challenging for retrospective studies, due to inadequate documentation including risk factor data. In this case, the Steering Committee agreed with taking a conservative bias approach such that the lowest possible non-zero weight should be assigned to macrovascular domain items with unknown risk factor data to avoid APS misclassification. Finally, the accurate assessment of “positive” aPL test results for APS classification is critical due to the following reasons: (1) despite LAC test limitations, i.e., false negative/positive results, the Steering Committee agreed that the test is extremely important if performed according to ISTH guidelines27; and (2) defined levels for “moderate” and “high” positivity apply to ELISA tests but not to other methodologies, e.g., automated platforms.33

The 2023 ACR/EULAR APS classification criteria have several strengths. First, international cases capturing the spectrum of APS contributed to its development, reducing the risk of selection bias and increasing generalizability. Second, to avoid the bias of circular reasoning,34 multidisciplinary participants in Phases III and IV were distinct. Third, use of 2 independent validation cohorts ultimately strengthened our results, demonstrating similar performance characteristics with overlapping confidence intervals. In fact, as the criteria were not changed between the two validation cohorts, our Phase IV results can be viewed as a single validation cohort with an interim analysis. Fourth, the new classification system allows for individual domain modification, allowing for future incorporation of additional clinical or new commercially available laboratory items if shown to be highly specific for APS, or new VTE and CVD risk factors based on future guidelines. Fifth, our criteria incorporate “entry criteria” to reflect current thinking that only cases with at least a minimal degree of clinical and laboratory suspicion for APS should be considered for classification. By intentionally collecting suspected APS cases for validation, we were able to test our entry criteria in a sensitivity analysis. Last, our model with absolute point requirements from both clinical and laboratory domains refines previous single-threshold models and more accurately reflects actual APS clinical decision-making.

As a limitation, our cohorts do not represent all possible subpopulations; however, we anticipate that investigators will test external validity in other cohorts, for example, aPL-positive SLE patients, non-White race/ethnicities, paediatrics, or nonacademic cohorts. The Steering Committee emphasised that large population-based studies in APS, accounting for social determinants of health and access to care, are needed to better establish disease prevalence overall and across sociodemographic groups. Absent a definitive gold standard, validating a classification system for an evolving disease definition such as APS can be challenging. As acknowledged by other classification criteria development efforts,10 although independent adjudicators may have inherent biases toward established criteria, our careful selection of individuals, based on the adjudicators’ expertise and with the adjudicators being blinded to relevant discussions about literature review and expert consensus–based decisions, reduced the bias of circular reasoning. Furthermore, the opportunity for adjudicators to discuss controversial cases in-depth, and to achieve consensus on all cases except one, ultimately strengthened their combined opinion.

In conclusion, using rigorous data-driven and expert-based methodology, including international multidisciplinary collaborators with APS expertise, methodologists, and patients, we have incorporated heterogenous aPL-related clinical and laboratory manifestations into a set of hierarchically clustered, weighted, and risk-stratified classification criteria reflecting current thinking about APS, providing high specificity and an improved foundation for APS research.

Acknowledgments

The APS Classification Criteria Steering Committee dedicates this work to the late Raymond P. Naden, MD. Dr. Naden helped lead the development of numerous rheumatic disease classification criteria, including ours; without him, the work presented in this manuscript would not have been possible. The APS Classification Criteria Steering Committee also thanks our ACR/EULAR liaisons, Ronald van Vollenhoven, MD (Amsterdam University Medical Center, Amsterdam, The Netherlands) and Ulf Muller-Ladner, MD (Immunologie, Kerckhoff-Klinik GmbH, Nauheim, Germany), as well as Amy Turner (ACR) for their help and guidance; David Berlin, MD (Weill Cornell Medicine, New York), Jessica Peña, MD (Weill Cornell Medicine, New York), Douglas Mintz, MD (Hospital for Special Surgery, New York), Corrine Sinnette (Brigham and Women’s Hospital, Boston), and Deanna Jannat-Khah (Hospital for Special Surgery, New York) for their contributions during Phase III; Yasaman Ahmedzadeh, MD (Hospital for Special Surgery, New York) for her contributions during Phases I, II, and III; and Joann Vega, CRCC (Hospital for Special Surgery, New York) and Cindy Force (ACR) for their administrative assistance. We also thank our patient representatives.