Abstract
Background
The potential role of structured clinical pretest probability assessment in the diagnostic workup of giant cell arteritis (GCA) is not well defined.
Patients and methods
We applied four clinical prediction rules to patients enrolled in the ongoing prospective PREDICT GCA study. All patients underwent temporal artery biopsy in addition to the detailed clinical and sonographic workup. The final diagnosis was based on the 6 months follow-up and ambiguous cases were judged by an independent expert panel. The diagnostic accuracy of the four models was determined using 2×2 contingency tables and receiver operating characteristic analysis. Patients with and without discordant results in different scoring systems were compared using univariate significance tests.
Results
Seventy patients were analysed. The highest diagnostic accuracy (area under the curve (AUC) 0.84) and the highest negative predictive value (NPV, 86.7%) were provided by the Ing et al model. The other scores performed worse, with AUC/NPV of 0.74/78.6% (Southend GCA Probability Score), 0.75/73.7% (Bhavsar-Khalidi Score) and 0.70/77.8% (PREDICT score). In 37.5% of patients, clinically significant discordances in categorisation (classification as both low and non-low risk by different scoring systems) were evident. These patients less frequently presented with clinical symptoms and signs of cranial GCA (all p<0.01), but the rate of visual impairment did not differ from that in patients with consistent risk estimation results.
Conclusion
Clinical prediction models for the diagnosis of GCA differ in their diagnostic accuracy. Clinically relevant discordances of individual risk prediction are common.
Trial registration number
DRKS00031293.
Keywords: Classification, Giant Cell Arteritis, Sensitivity and Specificity, Systemic vasculitis, Vasculitis
WHAT IS ALREADY KNOWN ON THIS TOPIC
Giant cell arteritis (GCA) is a chronic disease with acute onset, and cranial ischaemic complications typically occur before diagnosis.
Sonography-based fast-track clinics have shown promise in reducing the risk of blindness, but structured clinical estimation of pretest probability has not been implemented in routine clinical practice so far.
WHAT THIS STUDY ADDS
Comparing four different prediction rules solely relying on clinical information, our study highlights clinically relevant between-score heterogeneities in individual estimation of GCA risk.
We identified subgroups of patients in whom structured risk estimation is particularly challenging.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
Our study informs ongoing research projects on structured clinical risk estimation in suspected GCA.
The study results build the basis for integration of clinical prediction rules in stepwise diagnostic algorithms to be applied in primary care.
These might facilitate accurate and timely identification of individuals who require sonographic fast-track diagnostics and those in whom a diagnosis of GCA may be safely excluded even without further diagnostic imaging.
Introduction
Giant cell arteritis (GCA) is the most common systemic type of vasculitis, with an estimated lifetime risk of 1% and 0.5% in women and men, respectively.1 GCA affects individuals over 50 years of age with a peak incidence between 70 and 79 years.2 If left untreated, the most common early complication is permanent loss of vision, which nowadays still affects about one in six patients.3 4 As early ocular ischaemic complications typically occur prior to therapy initiation, fast-track clinics are important to avoid such ischaemic events.5 6
No single clinical or laboratory feature allows confirmation or refutation of a diagnosis of GCA,7 making imaging, in particular ultrasound, indispensable in the diagnostic workup of suspectedGCA. However, with the increasing availability of high-resolution ultrasound technology, overdiagnosis due to false positive ultrasound results (eg, detection of mild or moderate wall temporal artery wall thickening in patients with a high cardiovascular risk profile) has emerged as an important issue.8–12 Moreover, assessment of clinical pretest probability appears to be underused in the process of clinical decision-making in the setting of suspected GCA so far. Numerous clinical prediction rules/scoring systems aimed at structured estimation of the probability of GCA in individuals with suggestive symptoms have been published, including some that take only clinical variables (±laboratory variables) into account13–19 and some that additionally or solely rely on imaging information.20–24
In this study, we present the comparative evaluation of four clinical prediction models not involving sonographic imaging data in order to evaluate these models with regard to their potential of estimating clinical pretest probability before imaging studies are performed.
Patients and methods
Selection of clinical prediction models for analysis
Based on a selective literature search and a recently published systematic review,25 we identified published clinical GCA prediction models based solely on clinical variables, with or without laboratory information, that provided cutoff-values for categorisation of clinical pretest probability (low vs moderate vs high).13–16 Prediction models including imaging information, solely relying on laboratory information or providing no clear categorisation of estimated clinical probabilities, were excluded from analysis.17–24 The 2022 American College of Rheumatology/European Alliance of Associations for Rheumatology (EULAR) classification criteria for GCA were not considered in this work, as this classification system, intended for clinical research, includes imaging information and offers no categorical description of clinical probability for diagnostic purposes.26
Characteristics of the four clinical prediction models included in the analysis
The main characteristics of the four clinical prediction rules included are briefly outlined here and summarised in online supplemental table 1.
The prediction model by Ing et al13 is based on neural network and logistic regression prediction models. The associated multivariate GCA risk calculator gives the predicted probabilities as percentages for the neural network, logistic regression and an average risk estimation based on both models. The model was retrospectively validated based on temporal artery biopsy (TAB) results of 1201 patients from 14 medical centres in Canada. Predictors were age, gender, jaw claudication, new-onset headache, temporal artery abnormality on physical examination (palpatory tenderness, decreased pulse and scalp nodularity), diplopia, ischaemia-related loss of visual acuity or field or vision loss (ischaemic optic neuropathy, retinal artery occlusion or stroke), platelet level, C reactive protein (CRP) and erythrocyte sedimentation rate (ESR) prior to glucocorticoid administration. For the present analysis, we used the average risk estimation of the Ing et al model. Risk is classified as being low, moderate and high when the estimated probability equals <14%, 14%–42.9% and >43%, respectively.13 27
The Southend GCA Probability Score (SGCAPS)15 assigns weighted point values to various clinical and laboratory findings, yielding a maximum of 32 possible points. Notably, the SGCAPS accounts for alternative diagnoses (infection, cancer, systemic rheumatic disease, head and neck pathology, others) with subsequent deduction of points (−3 points). Classification of risk as low, moderate and high is based on a sum score of <9 points, 9–12 points and >13 points, respectively.28
The Bhavsar-Khalidi Score (BKS)14 was published as a conference abstract in 2019. Based on the presence or absence of cranial symptoms, visual symptoms, systemic symptoms, temporal artery physical examination findings and inflammatory markers, the score stratifies patients into three risk categories: low risk (≤4 points), moderate risk (5–7 points) and high risk (≥8 points), with an achievable maximum of 20 points. Sensitivity and specificity for the diagnosis of GCA were reported to be 87% and 65%, respectively. In 2020, a study with small sample size (n=20) suggested good interrater reliability.29 Full publication of the score is pending.
Our group developed a clinical prediction rule, which is currently undergoing prospective validation in the PREDICT-GCA study.16 This simple-to-calculate score includes only four clinical items, resulting in a sum score that categorises patients into a low (0–1 point), moderate (2–3 points) and high (4–6 points) clinical probability of suffering from GCA. In a previous retrospective study, the clinical score was incorporated into a stepwise diagnostic algorithm, further integrating CRP levels and high-resolution compression sonography (hrCS). Subsequently, the score was slightly modified (incorporation of scalp tenderness and temporal artery swelling) in order to improve the negative predictive value (NPV).30 For the purpose of this comparative study, a variant of the score incorporating CRP values (awarding one additional point when CRP values were >2.5 mg/dL; identical cut-off values remained for patient categorisation) was calculated.
Study population
All scoring systems were applied to 71 patients included in the PREDICT-GCA study, an ongoing prospective, monocentric observational study enrolling patients aged >50 years with a substantiated clinical suspicion of GCA, as determined by experienced rheumatologists, angiologists or ophthalmologists. The study protocol and the baseline characteristics of the first 50 patients have been published previously.30
Besides detailed clinical assessment, patients undergo structured multimodal vascular sonography, including hrCS of the cranial arteries, by experienced, blinded operators and unilateral TAB. The final reference diagnosis takes the 6 months follow-up into account, with ambiguous cases evaluated by two independent GCA experts not involved in the study procedures (C-JB, VSS). Indications for external expert review include (i) a potential clinical diagnosis of GCA but negative TAB, (ii) a potential alternative diagnosis but positive hrCS and (iii) all cases with borderline TAB. For the present analysis, clinical, laboratory, sonographic and histopathological data of the anonymised ambiguous cases were presented to the external experts during a remote consensus meeting held on 21 July 2025.
Statistical analysis
Statistical analyses were performed using SPSS V.30.0 (IBM, Armonk, New York, USA). Group comparisons were conducted using univariate significance tests. Two-sided p values <0.05 were considered significant. To assess the diagnostic accuracy of the clinical GCA prediction models, 2×2 contingency tables were used to calculate positive predictive values (PPV) and NPV. Receiver operating characteristics (ROC) analysis was applied to analyse and visualise the balance between sensitivity and specificity, as determined by the area under the curve (AUC, given as absolute number and with 95% CI).
Registration
This study was registered with the German Clinical Trials Register (ID: DRKS00031293).
Results
Study cohort and final clinical diagnosis
One patient was excluded from analysis because TAB did not contain any arterial segment, leaving 70 patients for analysis.
The diagnostic adjudication process is illustrated in online supplemental figure 1. Sixteen patients with a final diagnosis based on independent expert review are characterised in online supplemental table 2. Among these, five patients exhibited positive hrCS but negative TAB and eight patients had borderline histopathological results of TAB samples. A final diagnosis of GCA was established in six of 16 patients. Noteworthy, in three patients, the final reference diagnosis differed from the clinical diagnosis at our centre (patients #3, #4 and #12 in online supplemental table 2).
Table 1 provides an overview of the clinical characteristics of the overall cohort and patients with (n=37) and without (n=33) a final diagnosis of GCA. While SGCAPS, BKS and the PREDICT score were calculated for all 70 patients, the Ing et al score could be calculated only for those 56 patients with available ESR values.
Table 1. Baseline characteristics of the overall cohort and of patients with and without a final diagnosis of GCA.
| Variables | Overall cohort (n=70) |
Patients with a final diagnosis of GCA (n=37) |
Patients without a final diagnosis of GCA (n=33) |
|---|---|---|---|
| Female sex, n (%) | 38 (54.3) | 24 (64.9) | 14 (42.4) |
| Age (mean±SD) | 73.3±9.6 | 74.9±8.6 | 71.6±10.5 |
| New-onset headache, n (%) | 49 (70) | 30 (81.1) | 19 (57.6) |
| Jaw claudication, n (%) | 33 (47.1) | 24 (64.9) | 9 (27.3) |
| Scalp tenderness, n (%) | 21 (30) | 17 (45.9) | 4 (12.1) |
| Temporal artery tenderness, n (%) | 29 (41.4) | 20 (54.1) | 9 (27.3) |
| Diplopia, n (%) | 14 (20) | 9 (24.3) | 5 (15.2) |
| Amaurosis fugax, n (%) | 13 (18.6) | 2 (5.4) | 11 (33.3) |
| Permanent visual impairment, n (%) | 35 (50) | 13 (35.1) | 22 (66.7) |
| Polymyalgia rheumatica, n (%) | 10 (14.3) | 7 (18.9) | 3 (9.1) |
| Constitutional symptoms, n (%) | 28 (40) | 19 (51.3) | 9 (27.2) |
| Extremity claudication, n (%) | 4 (5.7) | 3 (8.1) | 1 (3.0) |
| Cardiovascular disease, n (%) | 15 (21.4) | 5 (13.5) | 10 (30.3) |
| Pre-existent rheumatic disease, n (%) | 15 (21.4) | 11 (29.7) | 4 (12.1) |
| History of cancer, n (%) | 6 (8.6) | 2 (5.4) | 4 (12.1) |
| Temporal artery abnormality, n (%) | 42 (60) | 26 (70.3) | 16 (48.5) |
| Pre-existent treatment with low-dose glucocorticoids and/or glucocorticoid-sparing agents, n (%) | 7 (10.1) | 4 (11.1) | 3 (9.1) |
| Glucocorticoid treatment started for suspected GCA, n (%) | 47 (68.1) | 24 (64.9) | 23 (71.9) |
| Daily (methyl-)prednisolone dose (mg), mean±SD | 272±388 | 265±424 | 280±349 |
| C reactive protein (mg/dL), mean±SD | 4.9±6.2 | 7.0±6.6 | 2.4±4.9 |
| Erythrocyte sedimentation rate (mm/1 hour), mean±SD | 47±31 | 58±33 | 34±23 |
| Patients with abnormal hrCS, n (%) | 42 (60) | 31 (83.4) | 11 (33.3) |
| Patients with positive temporal artery biopsy, n (%) | 29 (41.4) | 28 (75.7) | 1 (3) |
GCA, giant cell arteritis; hrCS, high-resolution compression sonography.
Diagnostic accuracy and predictive values
The rate of patients categorised as having low, moderate or high clinical probability for GCA differed between the models, with low clinical probability assigned from 12.9% (PREDICT) to 27.1% (BKS). In patients considered to have low clinical probability, abnormal hrCS findings were found in every third patient (PREDICT and Ing et al models) to every second patient (SGCAPS), whereas a positive TAB was found in only one patient, in each case. High clinical probability estimates ranged from 39.3% (Ing et al model) to 57.1% (SGCAPS) (figure 1).
Figure 1. Proportion of different clinical probability estimates with the different prediction models. Red, yellow and green colours indicate high, moderate and low clinical probability. *14 cases excluded due to missing erythrocyte sedimentation rate values. BKS, Bhavsar-Khalidi Score; SGCAPS, Southend GCA Probability Score.
ROC analysis with the final clinical diagnosis as the reference revealed that the best balance between sensitivity and specificity was achieved with application of the Ing et al model (AUC 0.84, 95% CI 0.73 to 0.95). The SGCAPS (AUC 0.74, 95% CI 0.60 to 0.87), BKS (AUC 0.75, 95% CI 0.61 to 0.88) and PREDICT score (AUC 0.70, 95% CI 0.56 to 0.84) performed worse (figure 2A). When CRP values were integrated within the calculation of the PREDICT score, the AUC increased to values similar to those of the BKS and the SGCAPS (AUC 0.75, 95% CI 0.63 to 0.87) (figure 2B).
Figure 2. Receiver operating characteristics analysis visualising the diagnostic accuracy of the prediction models. (A) Comparison of the four models. (B) Change of overall diagnostic accuracy of the PREDICT score with inclusion of CRP values (addition of 1 point in case of values exceeding cut-off values of 2.5 mg/dL). BKS, Bhavsar-Khalidi Score; CRP, C reactive protein; SGCAPS, Southend GCA Probability Score.
The NPV of the scores in low clinical probability scenarios and the PPV of the moderate and high clinical probability scenarios are given in table 2. The score with the highest NPV in the low clinical probability setting (86.7%) was the Ing et al model, resulting in a misclassification rate of only 13.3%. The NPV values of the other three scores were quite similar, with correct classification of patients without a diagnosis of GCA in about three out of four patients. Noteworthy, the two patients misclassified in the low clinical probability category of the PREDICT score both had CRP values >2.5 mg/dL and thus would have undergone sonographic testing in the dedicated stepwise algorithm. The NPV of the PREDICT score combined with a CRP cut-off value of <2.5 mg/dL was 100%.
Table 2. Predictive values and misclassification rates of the four clinical probability scores.
| PREDICT (n=70) | SGCAPS (n=70) | BKS (n=70) | Ing et al* (n=56) | |
|---|---|---|---|---|
| Low clinical probability | ||||
| Proportion of patients (n, %) | 9 (12.9) | 14 (20) | 19 (27.1) | 15 (26.8) |
| Patients with abnormal hrCS, n (%) | 3 (33.3) | 7 (50) | 8 (42.1) | 5 (33.3) |
| Patients with positive temporal artery biopsy, n (%) | 1 (11.1) | 1 (7.1) | 1 (5.3) | 1 (6.7) |
| Patients with a final diagnosis of GCA (misclassification rate, n, %) | 2 (22.2) | 3 (21.4) | 5 (26.3) | 2 (13.3) |
| Patients without a final diagnosis of GCA (NPV, n, %) | 7 (77.8) | 11 (78.6) | 14 (73.7) | 13 (86.7) |
| Moderate clinical probability | ||||
| Proportion of patients (n, %) | 31 (44.3) | 16 (22.9) | 18 (25.7) | 19 (33.9) |
| Patients with abnormal hrCS, n (%) | 17 (54.8) | 6 (37.5) | 9 (50) | 11 (57.9) |
| Patients with positive temporal artery biopsy, n (%) | 7 (22.6) | 3 (18.8) | 5 (27.8) | 5 (26.3) |
| Patients without a final diagnosis of GCA (misclassification rate, n, %) | 16 (51.6) | 10 (62.5) | 10 (55.6) | 10 (52.6) |
| Patients with a final diagnosis of GCA (PPV, n, %) | 15 (48.4) | 6 (37.5) | 8 (44.4) | 9 (47.4) |
| High clinical probability | ||||
| Proportion of patients (n, %) | 30 (42.9) | 40 (57.1) | 33 (47.1) | 22 (39.3) |
| Patients with abnormal hrCS, n (%) | 22 (73.3) | 29 (72.5) | 24 (72.7) | 14 (63.6) |
| Patients with positive temporal artery biopsy, n (%) | 18 (60) | 23 (57.5) | 20 (60.6) | 12 (54.5) |
| Patients without a final diagnosis of GCA (misclassification rate, n, %) | 13 (43.3) | 12 (30) | 10 (30.3) | 7 (31.8) |
| Patients with a final diagnosis of GCA (PPV, n, %) | 17 (56.7) | 28 (70) | 23 (69.7) | 15 (68.2) |
14 cases excluded due to missing ESR values.
BKS, Bhavsar-Khalidi Score; ESR, erythrocyte sedimentation rate; GCA, giant cell arteritis; hrCS, high-resolution compression sonography; NPV, negative predictive value; PPV, positive predictive value; SGCAPS, Southend GCA Probability Score.
Differences between scores were only modest in the non-low (moderate and high) probability scenario, with the PPV ranging from 57.4% (PREDICT Score) to 65.9% (Ing et al model). Thus, about four of 10 patients without a final diagnosis of GCA were falsely classified as having moderate or high risk of GCA.
In the subset of patients with available ESR values and a calculable Ing et al score, the NPV of the low-risk categories remained stable across PREDICT, SGCAPS and BKS (75.0%, 76.9% and 76.5%, respectively), as did the PPVs of the corresponding non-low-risk categories (56.3%, 60.5% and 64.1%, respectively).
Discordant results between the models
The clinical probability of GCA assigned to any individual patient by the four models is given in figure 3. When analysing the 56 patients for whom calculation of all four scores was possible, different categories of estimated clinical probability with different scores were observed in as much as 39 patients (69.6%), with two different clinical probability categories assigned in 36 patients (64.3%) and three different clinical probability categories assigned in three patients (5.4%). Noteworthy, out of the 17 patients with completely consistent clinical probability adjudication over all scores, 12 patients (70.6%) were categorised as having a high clinical probability, 11 of whom received a final diagnosis of GCA. All 21 patients categorised as low risk in at least one score had a non-low-risk estimate in one or more of the other scores.
Figure 3. Discordance of clinical probability estimates between different scores in individual patients. Colour of patient numbers indicates final clinical diagnosis (red, giant cell arteritis; green, alternative diagnosis). BKS, Bhavsar-Khalidi Score; ESR, erythrocyte sedimentation rate; SGCAPS, Southend GCA Probability Score.
When compared with patients with consistent categorisation as non-low risk, those patients with clinically relevant inconsistent estimates were insignificantly younger (70.4±10.0 years vs 74.4±9.5 years, p=0.16) and significantly less often exhibited classical symptoms and signs of cranial GCA, for example, jaw claudication (9.5% vs 62.9%, p<0.01) and temporal artery abnormalities by physical examination (33.3% vs 77.1%, p<0.01). However, the rate of patients with permanent visual impairment did not differ significantly between groups (52.4% vs 48.6%, p=1.00). Mean values of humoral inflammatory markers were significantly lower in patients with inconsistent classification (eg, CRP: 2.8±5.9 mg/dL vs 5.8±6.5 mg/dL, p<0.01). Positive TAB results were less common (14.3% vs 45.7%, p=0.02), and a final diagnosis of GCA was established less often in these patients (28.6% vs 65.7%, p<0.01) (table 3).
Table 3. Comparison of the clinical characteristics of patients with and without discordant classification as both low risk and non-low risk in different scores.
| Variables | Patients with available ESR values (n=56) |
Patients with discordant classification (n=21) |
Patients without discordant classification (n=35) |
P value |
|---|---|---|---|---|
| Age (mean±SD) | 72.9±9.8 | 70.4±10.0 | 74.4±9.5 | 0.16 |
| Female sex, n (%) | 31 (55.4) | 11 (52.4) | 20 (57.1) | 0.79 |
| Final diagnosis of GCA, n (%) | 29 () | 6 (28.6) | 23 (65.7) | 0.01 |
| New-onset headache, n (%) | 40 (71.4) | 11 (52.4) | 29 (82.9) | 0.03 |
| Jaw claudication, n (%) | 24 (42.9) | 2 (9.5) | 22 (62.9) | <0.01 |
| Scalp tenderness, n (%) | 19 (33.9) | 2 (9.5) | 17 (48.6) | <0.01 |
| Temporal artery tenderness, n (%) | 23 (41.1) | 3 (14.3) | 20 (57.1) | <0.01 |
| Diplopia, n (%) | 10 (17.9) | 1 (4.8) | 9 (25.7) | 0.07 |
| Amaurosis fugax, n (%) | 10 (17.9) | 5 (23.8) | 5 (14.3) | 0.48 |
| Permanent visual impairment, n (%) | 28 (50) | 11 (52.4) | 17 (48.6) | 1.0 |
| Polymyalgia rheumatica, n (%) | 8 (14.3) | 2 (9.5) | 6 (17.1) | 0.70 |
| Constitutional symptoms, n (%) | 25 (44.6) | 6 (28.6) | 19 (54.3) | 0.10 |
| Manifest cardiovascular disease*, n (%) | 12 (21.4) | 6 (28.6) | 6 (17.1) | 0.33 |
| Known rheumatic disease, n (%) | 10 (17.9) | 3 (14.3) | 7 (20) | 0.73 |
| History of cancer, n (%) | 6 (10.7) | 3 (14.3) | 3 (8.6) | 0.66 |
| Temporal artery abnormality by physical examination, n (%) | 34 (60.7) | 7 (33.3) | 27 (77.1) | <0.01 |
| Pre-existent treatment with low-dose glucocorticoids and/or glucocorticoid-sparing agents, n (%) | 6 (10.7) | 1 (4.8) | 5 (14.3) | 0.40 |
| Glucocorticoid treatment started for suspected GCA, n (%) | 37 (66.1) | 14 (66.7) | 23 (65.7) | 1.00 |
| Daily (methyl-)prednisolone dose (mg), mean±SD | 267±384 | 242±352 | 282±406 | 0.93 |
| C reactive protein (mg/dL), mean±SD | 4.6±6.4 | 2.8±5.9 | 5.8±6.5 | <0.01 |
| Erythrocyte sedimentation rate (mm/1 hour), mean±SD | 47±31 | 36±29 | 53±32 | 0.02 |
| Patients with abnormal hrCS, n (%) | 33 (58.9) | 10 (47.6) | 23 (65.7) | 0.26 |
| Patients with positive temporal artery biopsy, n (%) | 19 (33.9) | 3 (14.3) | 16 (45.7) | 0.02 |
Inclusion of only patients for whom all four scores could be calculated (n=56).
ESR, erythrocyte sedimentation rate; GCA, giant cell arteritis; hrCS, high-resolution compression sonography.
Discussion
This study aimed to analyse the feasibility and diagnostic accuracy of available clinical prediction models for the diagnosis of GCA. The scores differed in their ability to exclude a final diagnosis of GCA in patients classified as being at low risk of disease, as reflected by their NPVs. By contrast, the capability of including a final diagnosis of GCA in patients by classifying them as having a non-low (moderate or high) clinical probability (PPV) was rather similar across all tests. It is noteworthy that no single patient in the present study was consistently categorised as low risk by all four scores.
Positive results of hrCS were found in a substantial proportion of patients considered to have low clinical probability of GCA across all scores in the present study. Even more important is the fact that we found a substantial rate (more than two-thirds) of patients to be classified in different clinical probability categories when applying the different scoring systems, with clinically important inconsistencies (categorisation of an individual subject in both low and non-low category by different scores) in more than one-third of patients. Those patients were characterised by a lower rate of classical symptoms and signs of cranial GCA, whereas the rate of permanent visual impairment did not differ significantly from consistently classified patients. Ultimately, patients with clinically important inconsistencies in risk estimation were less frequently diagnosed with GCA.
Clinical probability assessment is key before performing a diagnostic test, as the pretest probability, which depends on disease prevalence in a given population, influences the PPV and NPV of the diagnostic test.31 While structured clinical probability assessment has been well established in some fields of clinical medicine, it is rather neglected in others. Probably, the most prominent examples of structured pretest probability assessment are venous thromboembolism (most notably the Wells score) and chronic coronary syndrome (for which current guidelines recommend the risk factor-weighted clinical likelihood model).32 33
In suspected GCA, the variable clinical presentation and the risk of ocular ischaemic complications, which mostly occur before treatment is initiated, create pressure for immediate decision-making: whether to test or not and whether to treat or not, even before testing is performed in cases with a high index of clinical suspicion. While a diagnosis of GCA should not be missed to prevent sight loss and other ischaemic complications, misdiagnosis of GCA should also be avoided given the potential complications of immunosuppressive treatment. The latter problem has become increasingly important in recent years with the advent of high-resolution ultrasound techniques, which carry a risk of false-positive findings.8–12
The combination of clinical symptoms and signs, as pursued by standardised clinical probability scoring systems, might offer greater predictive values compared with single clinical features.7 These considerations underscore the importance of structured clinical pretest probability assessment. While the EULAR guidelines do not discuss aspects of test theory in the diagnostic workup of GCA,34 the current EULAR recommendations on the use of diagnostic imaging in large vessel vasculitis take clinical probability into account for (post-test) interpretation of imaging and TAB results.35 The 2020 published guidelines of the British Society of Rheumatology explicitly consider estimation of pretest probability as relevant for interpretation of ultrasound and biopsy results and provide a stepwise diagnostic algorithm based on initial clinical pretest probability assessment.36 However, how to assess clinical pretest probability and how to discriminate between clinical probability categories (low vs moderate vs high) have not yet been specified in guideline recommendations.
We are aware of only one study by Sargi et al, which compared the performance of the SGCAPS, the BKS, the Ing et al model and sonography-based scoring systems in a prospective cohort of 200 patients with suspected GCA (TAB performed in 15 patients).27 PPVs were rather low for all clinical scores in that study, with the best PPV achieved with application of the BKS (55%). By contrast, NPVs were excellent for all clinical scores, and the SGCAPS provided the highest NPV (98.9%). As halo count and halo score offered both very high NPV and PPV, Sargi et al concluded that combining a clinical score with a quantitative sonographic method, particularly SGCAPS and halo count, represents the most accurate approach for the fast-track diagnostic workup of GCA. The authors did not stratify predictive values according to the calculated clinical probabilities and did not analyse inconsistencies in classification between scores.
As mentioned in the ‘Patients and methods’ section, we excluded scores that include sonographic parameters in order to assess clinical pretest probability. Consequently, we did not analyse sonographic scoring systems such as the halo score and its modifications, as well as the OGUS.20 21 Since the best-evaluated score in this field to date, the HAS-GCA score, is based on a combination of a clinical score (SGCAPS) and the sonographic halo score, we excluded this score from the analysis as well.22 It must be emphasised that the HAS-GCA score has demonstrated excellent diagnostic accuracy in a prospective multicentre validation study, with a C-statistic of 0.969 (95% CI 0.952 to 0.990).22 Although developed with stringent methodology, its feasibility in clinical practice, particularly in primary care, may be limited, as the HAS-GCA requires quantitative high-resolution ultrasound in all cases, regardless of clinical pretest probability. Results are expressed as a likelihood percentage rather than in operationalisable categories.
The PREDICT score was developed with the intention to avoid ultrasound imaging in selected patients with low clinical pretest probability in order to prevent false-positive ultrasound findings, which are increasingly recognised as a diagnostic problem. For this purpose, the clinical score was integrated into a stepwise diagnostic algorithm together with a sensitive biomarker.16 As a consequence, and in contrast to the three other scores evaluated in our study, the PREDICT score by intention does not include an inflammatory marker as an implicit test item. However, when CRP values are integrated as an additional item into the score, as done in the present study in order to enhance comparability across the single tests, its diagnostic accuracy was shown to be comparable to that of the other composite scores (SGCAPS, BKS). The stepwise diagnostic algorithm based on the PREDICT score is currently undergoing prospective validation in the ongoing PREDICT-GCA study.30 If the safety of this diagnostic approach can be proven, the algorithm could show promise for the management of patients with suspected GCA in primary care and may have the potential to avoid overdiagnosis due to false-positive ultrasound findings in patients with low clinical pretest probability.
It must be emphasised in this context that the PREDICT score, the BKS and the Ing et al model exclusively address the cranial disease pattern of GCA.13 14 16 The SGCAPS includes a broader spectrum of symptoms, including some closely related to extracranial GCA, but cannot be regarded as sufficiently validated for extracranial GCA.15 As the clinical spectrum of extracranial large-vessel GCA is so heterogenous, clinical prediction rules in our opinion appear to be rather unsuitable for the diagnostic workup of the extracranial disease manifestation when it is not accompanied by symptomatic cranial GCA.
The major strengths of the study presented here include the prospective collection of very detailed clinical data together with mandatory TAB in all patients and independent external validation of the final clinical diagnosis in ambiguous cases. We acknowledge some limitations. First, by contrast to the routine care scenarios of prediction rule development in retrospective studies, the rate of patients with non-low clinical probability (including patients presenting with visual impairment) and, subsequently, a final diagnosis of GCA was substantially higher in our cohort.13 15 16 This may have been a result of a selection bias due to reluctance to perform TAB in low clinical probability patients in the setting of a prospective clinical study. Second, beyond ultrasound imaging of the temporal and axillary arteries and TAB, patients did not undergo routine cross-sectional imaging of the large vessels. Therefore, we cannot fully rule out that cases of large-vessel GCA (ie, isolated aortitis) may have been misdiagnosed. However, during the follow-up required for determination of the final diagnosis, we did not encounter any patient with a delayed diagnosis of extracranial GCA. Third, items in different scoring systems are not fully interoperable (online supplemental table 1). Due to missing ESR values, 14 patients had to be excluded from the analysis of the prediction model published by Ing et al, which in the analysis of the limited sample of 56 patients exhibited both the highest overall diagnostic accuracy and the highest NPV.13 Determination of the ESR is not mandatory in the PREDICT GCA study, as the PREDICT algorithm relies on CRP values. In view of the numerous confounding factors that may influence ESR values (eg, age, sex and chronic comorbidities), we consider CRP to be more robust biomarker and better suited to routine clinical care.
In summary, clinical prediction models differ in their diagnostic accuracy, and clinically relevant discordances in risk prediction between different scores are common. The role of explicit assessment of clinical pretest probability using prediction models in the diagnostic workup of suspected GCA remains to be determined.
Supplementary material
Footnotes
Funding: The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Provenance and peer review: Not commissioned; externally peer reviewed.
Patient consent for publication: Consent obtained directly from patient(s).
Ethics approval: The study protocol was approved by the Ethics Committee of the Medical Faculty of Ludwig-Maximilians-Universität, Munich, Germany (reference number: 22-0267). This study was conducted in accordance with the Declaration of Helsinki. All patients provided written informed consent before inclusion in the study.
Data availability statement
Data are available on reasonable request.
References
- 1.Crowson CS, Matteson EL, Myasoedova E, et al. The lifetime risk of adult-onset rheumatoid arthritis and other inflammatory autoimmune rheumatic diseases. Arthritis Rheum. 2011;63:633–9. doi: 10.1002/art.30155. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.González-Gay MÁ, Heras-Recuero E, García-Fernández A, et al. Revisiting the epidemiology of giant cell arteritis. Clin Exp Rheumatol. 2025;43:742–8. doi: 10.55563/clinexprheumatol/xc46ld. [DOI] [PubMed] [Google Scholar]
- 3.Soriano A, Muratore F, Pipitone N, et al. Visual loss and other cranial ischaemic complications in giant cell arteritis. Nat Rev Rheumatol . 2017;13:476–84. doi: 10.1038/nrrheum.2017.98. [DOI] [PubMed] [Google Scholar]
- 4.Bosch P, Espigol-Frigolé G, Cid MC, et al. Cranial involvement in giant cell arteritis. Lancet Rheumatol. 2024;6:e384–96. doi: 10.1016/S2665-9913(24)00024-9. [DOI] [PubMed] [Google Scholar]
- 5.Diamantopoulos AP, Haugeberg G, Lindland A, et al. The fast-track ultrasound clinic for early diagnosis of giant cell arteritis significantly reduces permanent visual impairment: towards a more effective strategy to improve clinical outcome in giant cell arteritis? Rheumatology (Oxford) 2016;55:66–70. doi: 10.1093/rheumatology/kev289. [DOI] [PubMed] [Google Scholar]
- 6.Monti S, Bartoletti A, Bellis E, et al. Fast-track ultrasound clinic for the diagnosis of giant cell arteritis changes the prognosis of the disease but not the risk of future relapse. Front Med. 2020;7:589794. doi: 10.3389/fmed.2020.589794. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.van der Geest KSM, Sandovici M, Brouwer E, et al. Diagnostic accuracy of symptoms, physical signs, and laboratory tests for giant cell arteritis: a systematic review and meta-analysis. JAMA Intern Med. 2020;180:1295–304. doi: 10.1001/jamainternmed.2020.3050. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Czihal M, Köhler A, Prearo I, et al. Hyperechogenic intimal lesions and wall thickness of the temporal and facial arteries in elderly patients with arterial occlusions of the eye. RMD Open. 2021;7:e001688. doi: 10.1136/rmdopen-2021-001688. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Czihal M, Lottspeich C, Köhler A, et al. Transocular sonography in acute arterial occlusions of the eye in elderly patients: Diagnostic value of the spot sign. PLoS One. 2021;16:e0247072. doi: 10.1371/journal.pone.0247072. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.De Miguel E, Beltran LM, Monjo I, et al. Atherosclerosis as a potential pitfall in the diagnosis of giant cell arteritis. Rheumatology (Oxford) 2018;57:318–21. doi: 10.1093/rheumatology/kex381. [DOI] [PubMed] [Google Scholar]
- 11.Fernández-Fernández E, Monjo-Henry I, Bonilla G, et al. False positives in the ultrasound diagnosis of giant cell arteritis: some diseases can also show the halo sign. Rheumatology (Oxford) 2020;59:2443–7. doi: 10.1093/rheumatology/kez641. [DOI] [PubMed] [Google Scholar]
- 12.Seitz P, Lötscher F, Bucher S, et al. Ultrasound intima-media thickness cut-off values for the diagnosis of giant cell arteritis using a dual clinical and MRI reference standard and cardiovascular risk stratification. Front Med. 2024;11:1389655. doi: 10.3389/fmed.2024.1389655. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Ing EB, Lahaie Luna G, Toren A, et al. Multivariable prediction model for suspected giant cell arteritis: development and validation. Clin Ophthalmol. 2017;11:2031–42. doi: 10.2147/OPTH.S151385. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Hu A, Beattie K, Bhavsar S, et al. 090. A clinical scoring system for risk stratification of giant cell arteritis. Rheumatology (Oxford) 2019;58:30. doi: 10.1093/rheumatology/kez058.030. [DOI] [Google Scholar]
- 15.Laskou F, Coath F, Mackie SL, et al. A probability score to aid the diagnosis of suspected giant cell arteritis. Clin Exp Rheumatol. 2019;37 Suppl 117:104–8. [PubMed] [Google Scholar]
- 16.Czihal M, Lottspeich C, Bernau C, et al. A diagnostic algorithm based on a simple clinical prediction rule for the diagnosis of cranial giant cell arteritis. J Clin Med. 2021;10:1163. doi: 10.3390/jcm10061163. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.González-López JJ, González-Moraleja J, Rebolleda G, et al. A calculator for temporal artery biopsy result prediction in giant cell arteritis suspects. Eur J Intern Med. 2014;25:e98–100. doi: 10.1016/j.ejim.2014.07.010. [DOI] [PubMed] [Google Scholar]
- 18.Weis E, Toren A, Jordan D, et al. Development of a predictive model for temporal artery biopsies. Can J Ophthalmol. 2017;52:599–605. doi: 10.1016/j.jcjo.2017.04.001. [DOI] [PubMed] [Google Scholar]
- 19.El-Dairi MA, Chang L, Proia AD, et al. Diagnostic algorithm for patients with suspected giant cell arteritis. J Neuroophthalmol. 2015;35:246–53. doi: 10.1097/WNO.0000000000000234. [DOI] [PubMed] [Google Scholar]
- 20.van der Geest KSM, Borg F, Kayani A, et al. Novel ultrasonographic Halo Score for giant cell arteritis: assessment of diagnostic accuracy and association with ocular ischaemia. Ann Rheum Dis. 2020;79:393–9. doi: 10.1136/annrheumdis-2019-216343. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Dejaco C, Ponte C, Monti S, et al. The provisional OMERACT ultrasonography score for giant cell arteritis. Ann Rheum Dis. 2023;82:556–64. doi: 10.1136/ard-2022-223367. [DOI] [PubMed] [Google Scholar]
- 22.Sebastian A, van der Geest KSM, Tomelleri A, et al. Development of a diagnostic prediction model for giant cell arteritis by sequential application of Southend Giant Cell Arteritis Probability Score and ultrasonography: a prospective multicentre study. Lancet Rheumatol. 2024;6:e291–9. doi: 10.1016/S2665-9913(24)00027-4. [DOI] [PubMed] [Google Scholar]
- 23.Moudrous W, Visser LH, Yilmaz T, et al. A new prediction model for giant cell arteritis in patients with new onset headache and/or visual loss. Ann Med. 2022;54:2770–6. doi: 10.1080/07853890.2022.2130971. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Junek M, Hu A, Garner S, et al. Contextualizing temporal arterial magnetic resonance angiography in the diagnosis of giant cell arteritis: a retrospective cohort study. Rheumatology (Oxford) 2021;60:4229–37. doi: 10.1093/rheumatology/keaa916. [DOI] [PubMed] [Google Scholar]
- 25.Junek ML, Okaj I, Patel S, et al. Multivariable models to predict a diagnosis of giant cell arteritis: systematic review and meta-analysis. J Rheumatol. 2025;52:1081–8. doi: 10.3899/jrheum.2022-1204. [DOI] [PubMed] [Google Scholar]
- 26.Ponte C, Grayson PC, Robson JC, et al. 2022 American College of Rheumatology/EULAR classification criteria for giant cell arteritis. Ann Rheum Dis. 2022;81:1647–53. doi: 10.1136/ard-2022-223480. [DOI] [PubMed] [Google Scholar]
- 27.Sargi C, Ducharme-Benard S, Benard V, et al. Assessment and comparison of probability scores to predict giant cell arteritis. Clin Rheumatol. 2024;43:357–65. doi: 10.1007/s10067-023-06721-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Sebastian A, Tomelleri A, Kayani A, et al. Probability-based algorithm using ultrasound and additional tests for suspected GCA in a fast-track clinic. RMD Open. 2020;6:e001297. doi: 10.1136/rmdopen-2020-001297. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Junek M, Beattie K, Bhavsar S, et al. Inter-rater reliability of the Bhavsar-Khalidi clinical scoring system for risk stratification of giant cell arteritis. J Rheumatol. 2020;47:1059–60. [Google Scholar]
- 30.Thielmann L-C, Findik-Kilinc M, Füeßl L, et al. A clinical probability-based, stepwise algorithm for the diagnosis of giant cell arteritis: study protocol and baseline characteristics of the first 50 patients included in the prospective validation study with focus on cranial symptoms. J Clin Med. 2025;14:2254. doi: 10.3390/jcm14072254. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Uy EJB. Key concepts in clinical epidemiology: estimating pre-test probability. J Clin Epidemiol. 2022;144:198–202. doi: 10.1016/j.jclinepi.2021.10.022. [DOI] [PubMed] [Google Scholar]
- 32.Ceriani E, Combescure C, Le Gal G, et al. Clinical prediction rules for pulmonary embolism: a systematic review and meta-analysis. J Thromb Haemost. 2010;8:957–70. doi: 10.1111/j.1538-7836.2010.03801.x. [DOI] [PubMed] [Google Scholar]
- 33.Vrints C, Andreotti F, Koskinas KC, et al. 2024 ESC Guidelines for the management of chronic coronary syndromes. Eur Heart J. 2024;45:3415–537. doi: 10.1093/eurheartj/ehae177. [DOI] [PubMed] [Google Scholar]
- 34.Hellmich B, Agueda A, Monti S, et al. 2018 Update of the EULAR recommendations for the management of large vessel vasculitis. Ann Rheum Dis. 2020;79:19–30. doi: 10.1136/annrheumdis-2019-215672. [DOI] [PubMed] [Google Scholar]
- 35.Dejaco C, Ramiro S, Bond M, et al. EULAR recommendations for the use of imaging in large vessel vasculitis in clinical practice: 2023 update. Ann Rheum Dis. 2024;83:741–51. doi: 10.1136/ard-2023-224543. [DOI] [PubMed] [Google Scholar]
- 36.Mackie SL, Dejaco C, Appenzeller S, et al. British Society for Rheumatology guideline on diagnosis and treatment of giant cell arteritis. Rheumatology (Oxford) 2020;59:e1–23. doi: 10.1093/rheumatology/kez672. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data are available on reasonable request.