Introduction

Some of the text in this article has been reproduced from Essat et al.,1 Ren et al.,2 Thokala et al.3 and Wren et al.4 These are Open Access articles distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.

Rationale for research and background

Acute aortic syndrome (AAS) is a life-threatening emergency condition affecting the thoracic aorta. The ‘syndrome’ is an umbrella term for acute aortic dissection (including type A, involving the ascending aorta; and type B, limited to the descending aorta), intramural haematoma and penetrating ulcer. Without treatment, AAS can progress to aortic rupture, with rapid deterioration and death.

Chest pain is the most common presenting symptom of AAS (80%), although back pain (40%) and abdominal pain often occur.5 These symptoms account for over 2 million emergency department (ED) attendances per year in England6 and are overwhelmingly due to causes other than AAS. The incidence of AAS has been estimated as one in every 980 ED attendances with atraumatic chest pain,7 thus creating a substantial diagnostic challenge.

Computed tomographic angiography (CTA) scanning of the aorta has high sensitivity and specificity for diagnosing AAS but incurs significant costs and risks of ionising radiation. Clinicians therefore need to use CTA selectively in patients presenting to the ED with symptoms that could be due to AAS. Imaging techniques other than CTA, such as electrocardiogram (ECG)-gated CTA, echocardiography and magnetic resonance angiography, can accurately diagnose AAS,8 but these also require careful patient selection. Our analysis focuses on CTA as the imaging modality recommended in UK and international guidelines.5,9–11

Patients with successful treatment for AAS can have good life expectancy and return to full health, while misdiagnosis can lead to avoidable death. Around 25% of patients with AAS are not diagnosed with the condition until 24 hours after presenting to the ED,12 and the misdiagnosis rate during the initial ED visit for AAS is estimated to be as high as 38%.13 Optimal patient selection for urgent CTA is challenging with misdiagnosis affecting between one in three and one in seven patients with AAS,14,15 leading to worse outcomes, while CTA overtesting leads to diagnostic yields of 2–3%.12,16

Clinical scores and biomarkers can be used to select patients with suspected AAS for CTA. The Aortic Dissection Detection Risk Score (ADD-RS) uses information on high-risk conditions (such as Marfan syndrome or known aortic disease), pain features (abrupt onset, severe intensity or ripping/tearing) or examination features (perfusion deficit, new aortic insufficiency murmur or hypotension/shock) to identify patients at risk of AAS.17 The Canadian clinical practice guideline uses a clinical decision aid to stratify patients into low, moderate and high risk of AAS.9 The AORTA score uses six clinical features to stratify patients to low or high risk.18 D-dimer is the most extensively studied biomarker for AAS.19 A low D-dimer level in a patient with a low clinical probability of AAS could rule out AAS.20

Current guidelines reflect the uncertainty in the existing evidence.10 Canadian Heart Association, American Heart Association (AHA) and European Society of Cardiology (ESC) guidelines all recommend estimating clinical probability of AAS, but there are inconsistencies in the clinical features used. All recommend CTA for high-risk patients but provide different recommendations for low- and intermediate-risk patients. The Canadian guidelines recommend D-dimer for intermediate-risk patients,9 ESC guidelines recommend D-dimer for low-risk patients5 and AHA guidelines do not identify a role for D-dimer.21 Best Practice Guidance from the Royal College of Emergency Medicine (RCEM) and Royal College of Radiologists (RCR) recommend CTA if any high-risk clinical features are present and further research to determine the role of D-dimer in suspected AAS.11

Objectives

We aimed to identify an optimal diagnostic strategy for suspected AAS and measure the expected value of information from future primary research. Our specific objectives were:

1. To estimate the accuracy of clinical scores, models or algorithms, and/or biomarkers (including D-dimer) for diagnosing AAS.
2. To estimate the effectiveness [in terms of quality-adjusted life-year (QALY) gained], cost-effectiveness (in terms of net benefit and incremental cost per QALY gained) and practical implications (in terms of the burden of radiological investigations) of using alternative diagnostic strategies for AAS.
3. To estimate the expected value of perfect information (EVPI) to highlight the amount healthcare decision-makers could spend on future primary research to reduce the uncertainty in diagnostic strategies for AAS.

Methods for data collection and analysis

We undertook a systematic review, meta-analysis, decision analysis modelling and value of information analysis. The systematic review identified studies of index tests evaluating the accuracy of clinical scores, models, algorithms or biomarkers for a reference standard diagnosis of AAS in patients attending the ED with symptoms suggesting AAS (new-onset chest, back, or abdominal pain, syncope or symptoms related to perfusion deficit). Meta-analysis was used to generate summary estimates of accuracy for AAS for index tests evaluated with sufficient data of acceptable quality. We then developed a decision-analytic model to estimate the cost-effectiveness and practical implications of using alternative diagnostic strategies based on index tests identified in the systematic review to select patients with suspected AAS for CTA. The model was also used to estimate the EVPI associated with future research into diagnostic strategies for suspected AAS.

Full details of our methods and findings have been published.1–4 This synopsis provides an overview of the methods and results.

Systematic review and meta-analysis

We undertook a systematic review to identify studies estimating the diagnostic accuracy of clinical scores and biomarkers for detecting AAS. The systematic review was undertaken in accordance with the general principles recommended in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement,22 the guidelines published by Cochrane Screening and Diagnostic Test Methods Group23 and was prospectively registered on the International Prospective Register of Systematic Reviews (PROSPERO) database (CRD42022252121).

We identified potentially relevant studies through electronic searches of key electronic databases, including MEDLINE (OvidSP from 1946), EMBASE (OvidSP from 1974) and the Cochrane Library (www.cochranelibrary.com). All database searches were conducted from inception to February 2024. Searches were supplemented by hand-searching the reference lists of all relevant studies (including existing systematic reviews), forward citation searching of relevant articles, contacting key experts in the field (including professional and academic research groups) and undertaking targeted searches of the World Wide Web using the Google (Google Inc., Mountain View, CA, USA) search engine. No date or language restrictions were applied on any database.

We included all cohort studies that evaluated the diagnostic accuracy of clinical scores (including models and algorithms), and biomarkers against a reference standard for AAS in a population with suspected AAS. The reference standard used a definitive imaging modality (CTA, ECG-gated CTA, echocardiography or magnetic resonance angiography), operation or autopsy to confirm AAS. Studies including people with AAS following major trauma or as incidental findings were excluded. Studies using a case-control design (i.e. studies in which patients were selected on the basis of the results of their reference standard test) were also excluded due to the potential for design-related bias which tend to lead to overestimation of diagnostic accuracy24,25 and are not generally representative of a test’s accuracy in a clinical setting24,25 (a post hoc change – for further details, see Changes from the proposed project).

Potentially relevant articles were selected using a two-step process. (1) One reviewer examined all titles and excluded any citations that clearly did not meet the inclusion criteria (i.e. non-human, unrelated to AAS). (2) All abstracts and full-text articles were then examined independently by a minimum of two reviewers. Any disagreements in the selection process were resolved through discussion or, if necessary, arbitration by a third reviewer and included by consensus. Data were extracted by one reviewer and independently checked for accuracy by a second. Where we identified multiple publications of the same study, we extracted and reported data as a single study, seeking clarification from the study authors where appropriate. We assessed the methodological quality of each included study using the quality assessment of diagnostic accuracy studies-2 (QUADAS-2) tool.26

We planned to undertake meta-analysis of studies evaluating the diagnostic accuracy of clinical scores and biomarkers (e.g. the ADD-RS alone, D-dimer alone and the ADD-RS in combination with D-dimer) and to present descriptive results for any index tests with insufficient data for meta-analysis. In each meta-analysis of tests using a single threshold, the diagnostic test data were analysed using a bivariate hierarchical model,27 where, at a lower level, sensitivities and specificities of each study are modelled using binomial distributions and, at a higher level, logit-transformed sensitivities and specificities are modelled using a normal distribution to allow for possible (negative) correlation, as suggested in Deeks et al.23 For multiple threshold, we estimated the accuracy of the ADD-RS at thresholds of > 0 point and > 1 point using a multinomial meta-analysis model of Jones et al.28 As heterogeneity between studies was generally expected in studies of diagnostic test accuracy, a random-effects model was used to allow for the heterogeneity beyond chance between studies. Model descriptions and prior specifications can be found in Appendix 1.

All the analyses were conducted using a Bayesian framework via Markov chain Monte Carlo simulation and implemented in the R software (R Foundation, Vienna, Austria) environment using JAGs and rjags package.29 Different options of prior distributions were explored. Convergence to the target posterior distributions was assessed using the Gelman–Rubin convergence statistic.30 A total of 1,000,000 iterations with a burn-in of 100,000 and thinning of 10 were used to estimate the model parameters. Results were presented as forest plots and receiver operating characteristic (ROC) plots of sensitivity versus 1 – specificity. Estimates of sensitivity and specificity with 95% credible intervals (CrIs, also known as Bayesian confidence intervals) were plotted individually against each threshold to illustrate the variations among the synthesised studies. Ninety-five per cent prediction intervals (PrIs) were also reported, illustrating the between-study heterogeneity and a range of values that might be expected in a future study.

We estimated the accuracy of D-dimer at the 500 ng/ml threshold, where this was reported. We also undertook a sensitivity analysis of D-dimer that also included studies that did not report the 500 ng/ml threshold, using the threshold reported in the study that was closest to 500 ng/ml. Moreover, we identified six studies18,31–35 that reported the accuracy of the ADD-RS and D-dimer. We contacted the authors who agreed to share data to allow analysis of multiple alternative combinations of ADD-RS with D-dimer. We selected combinations for analysis that have previously been reported as potential strategies for selecting patients with suspected AAS for computed tomography (CT) scanning. This resulted in the analysis of the following index tests: ADD-RS > 0; ADD-RS > 1; ADD-RS > 0 or D-dimer > 500 ng/ml; ADD-RS > 1 or D-dimer > 500 ng/ml; and a strategy combining the ADD-RS with D-dimer based on the Canadian guideline (ADD-RS > 1 or ADD-RS = 1 with D-dimer > 500 ng/ml). We also performed sensitivity analysis for ADD-RS > 0 and ADD-RS > 1 limited to studies that also evaluated D-dimer to assess the impact of study selection.

We were unable to perform meta-analysis of biomarkers other than D-dimer due to the limited number of studies per biomarker and variable reporting of items. As a result, a narrative synthesis approach was undertaken, with data being summarised in tables with accompanying narrative summaries that included a description of the included variables, statistical methods and performance measures (e.g. sensitivity, specificity).36,37

Decision-analytic modelling

We developed a decision analysis model in Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA) that applied diagnostic strategies to a hypothetical population of 1000 patients attending hospital with symptoms suggesting AAS. Cost-effectiveness of the diagnostic strategies, measured as the incremental cost per QALY gained by each strategy compared with the next most effective alternative, was estimated using a lifetime horizon and NHS healthcare perspective. Probabilistic analysis incorporated uncertainty in the parameter estimates to provide more robust estimates of the mean costs and QALYs.

The structure of the model is shown in Figure 1. Each diagnostic strategy is applied to the patient cohort to determine the proportions classified as true positive (TP), false positive (FP), true negative (TN) or false negative (FN), depending upon the prevalence of AAS and sensitivity or specificity of the diagnostic strategy. We assume that true and FNs would not receive further testing for AAS, while true and FPs would receive confirmatory testing using CTA as the reference standard for AAS (with TPs receiving treatment for AAS).

FIGURE 1 Structure of the decision-analytic model. Reproduced from Thokala et al.3 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Tables 1 and 2 show the parameters used in the model. The prevalence of AAS was estimated from the Diagnosis of Acute Aortic Syndrome in the Emergency Department (DAShED) study,33 and the mean age of the patients, the proportion of type A and type B AAS and the proportion of males were estimated from the International Registry of Acute Aortic Dissection (IRAD).38 We assume that the population excludes patients whose frailty and/or comorbidities preclude surgery or thoracic endovascular aortic repair (TEVAR), because this would substantially increase the complexity of the model and reduce applicability to the population most likely to benefit from investigation for AAS. The base-case analysis used the estimate of AAS prevalence from the total DAShED population presenting with any possible symptoms of AAS (0.26%). Secondary analyses used AAS prevalence estimates for populations in the DAShED study selected on the basis of clinician estimates of AAS likelihood being greater than zero, > 1/10 and > 2/10 (0.61%, 1.0% and 1.74%, respectively) and the population in the DAShED study who received CTA (2.95%). The QALYs for those without AAS are estimated based on life expectancy of general population from the Office of National Statistics, and the general population utilities estimated from Ara et al.49

We evaluated strategies that use ADD-RS and/or D-dimer to select patients for CTA, including strategies recommended in guidelines. Current UK guidance recommends CTA for any patient who has risk features for AAS, which is similar to the ADD-RS > 0 strategy. Sensitivity and specificity of the diagnostic strategies were estimated by meta-analysis of diagnostic cohort studies. We also tested two hypothetical diagnostic strategies in the model: no investigation for AAS (0% sensitivity, 100% specificity) and CTA for all (100% sensitivity, 0% specificity). The costs of applying each strategy were estimated assuming 2 minutes of consultant time to calculate the ADD-RS and a reference cost for D-dimer.

We assumed that the diagnostic strategy only influenced outcomes among patients with AAS. Patients with TP type A AAS received surgical treatment, with 80% survival at 2 months estimated from IRAD data,38 while FN patients were assumed to die or have delayed treatment, with 50% survival at 2 months estimated from studies by Matthews et al.39 and Pourafkari et al.40 Patients with TP type B AAS received TEVAR (10%) or medical management (90%) with overall survival of 87.4% at 2 months estimated from IRAD data,38 while FN patients did not benefit from blood pressure control, resulting in 74.8% survival at 2 months, based on the relative risk for blood pressure control.50 The estimated annual mortality risk was 2.5% for type A AAS patients,38 5.5% for type B patients managed medically and 3.8% for those receiving TEVAR, based on data from Sa et al.41

The model included costs of diagnostic strategies, CTA, CT for incidental findings, treatment for AAS, costs of death in ED and long-term costs associated with AAS management. The costs of CTA, open repair and TEVAR were estimated from the NHS reference costs,46 while the costs of medical management for type B AAS were estimated using expert clinical opinion. The annual costs assumed to be the same for all AAS survivors.

In the long term, all survivors had an annual probability of reintervention which was estimated based on IRAD data,38 and it was assumed that all reinterventions are TEVAR. The model also included a small risk of cancer associated with CTA based on the study by Huang et al.,47 and modelled the impact of cancer as one-off lifetime cost and QALY loss estimated from Goodacre et al.,48 at the mid-point of life expectancy estimated from UK life tables.

The utilities for patients with AAS were sourced from studies identified in the recent systematic review by Carbone et al.51

Results summary

Systematic review

Figure 2 shows the PRISMA flow chart for the systematic review. We screened 2017 titles, 767 abstracts and 117 full-text articles, and included 39 studies in the review: 13 studies of the ADD-RS,18,31–35,52–58 25 studies of D-dimer,18,31–35,54,59–76 13 studies of other biomarkers52,63,68,69,73,76–83 and 3 studies of other clinical scores18,33,84 (several studies evaluated more than one index test).

FIGURE 2 The PRISMA flow chart for each review. Reproduced from Ren et al.2 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Studies of Aortic Dissection Detection Risk Score alone and in combination with D-dimer

We included 13 studies18,31–35,52–58 investigating ADD-RS alone or ADD-RS in combination with D-dimer. Of these, 12 studies18,31–33,35,53–58 contributed to the meta-analysis for ADD-RS alone (one reported the same data as another included study) and 6 studies18,31–35 contributed to the meta-analysis for ADD-RS in combination with D-dimer (sample size N = 162 to N = 22,075, prevalence of AAS 0.26–64%). The range of prevalence across the studies reflects variation in patient selection. Studies that used clinical presentation (e.g. symptoms suggesting AAS) to select patients tended to have lower prevalence than those that selected patients who received imaging for AAS. The methodological quality of the included studies was variable, with most studies having low or unclear risk of bias and applicability concerns in at least one item of the QUADAS-2 tool (see Figure 3). The proportion of patients receiving reference standard imaging varied between studies.

FIGURE 3 Quality assessment of diagnostic accuracy studies-2 assessment summary graph for studies of ADD-RS. Reproduced from Ren et al.2 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Figure 4 shows the summary plots for ADD-RS alone in panels A and B, and the summary plots for ADD-RS with D-dimer in panels C and D. Individual sensitivity and specificity from the same study are linked with lines. In panels A and C, ROC plots are displayed. In panels B and D, pooled sensitivity and specificity, along with the 95% CrI and 95% PrI, are plotted. The 95% CrI are marked by the shaded areas around the summary estimates, showing the range of likely values for average diagnostic accuracy. The 95% PrI are marked by the wider and lighter-shaded areas around the summary estimates, indicating a range of values that might be expected in a future study. Figure 5 shows the pooled estimates and 95% CrI for each strategy. Details on the exploration of different prior distributions can be found in Appendix 1.

FIGURE 4 Summary plots for ADD-RS and ADD-RS in combination with D-dimer. Reproduced from Ren et al.2 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

FIGURE 5 Pooled estimates and 95% CrI for each strategy. Reproduced from Ren et al.2 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Table 3 shows the results of meta-analysis, each with a 95% CrI and PrI. ADD-RS > 0 has high sensitivity and low specificity, while ADD-RS > 1 has low sensitivity and high specificity. Combinations of ADD-RS with D-dimer provide a range of trade-offs between sensitivity and specificity. There is a large amount of heterogeneity between studies, as illustrated by the wide 95% PrI. A sensitivity analysis of ADD-RS > 0 and ADD-RS > 1 limited to the six studies that also evaluated D-dimer showed very similar estimates to the base-case analysis with 10 studies.

Studies of D-dimer

We included 25 studies of D-dimer18,31–35,54,59–76 (sample size N = 41 to N = 1848, prevalence of AAS 0.9% to 64.8%). The methodological quality of the included studies was variable, with most studies having low or unclear risk of bias and applicability concerns in at least one item of the QUADAS‐2 tool (see Figure 6). The diagnostic threshold was pre-specified, apart from the studies of Fan et al.,62 Peng et al.69 and Zhang et al.,76 where they were determined from the study data, and thus had an uncertain risk of bias in quality assessment. A D-dimer threshold of 500 ng/ml was reported in 17 studies,18,31–35,59,63–68,71–74 and the threshold of 490 ng/ml reported by Fan et al.62 was considered equivalent to 500 ng/ml, so 18 studies18,31–35,59,62–68,71–74 were included in the base-case analysis and 25 studies18,31–35,54,59–76 in the sensitivity analysis.

FIGURE 6 The QUADAS-2 assessment summary graph for studies of D-dimer. Reproduced from Essat et al.1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Figure 7 shows the forest plot and Figure 8 shows the summary plot for the 18 studies reporting data for the 500 ng/ml threshold. Table 4 shows the summary estimates for the base-case meta-analysis and the sensitivity analysis. The wide PrI, especially for summary specificity, reflects substantial heterogeneity in the results from individual studies. Specificity varied markedly from 33%63 to 86%.71 The summary estimates for the sensitivity analysis were similar to the base-case analysis, suggesting that inclusion of studies with alternative thresholds did not markedly change our estimates of accuracy. A non-informative prior distribution is used for the analysis. Details on the prior specifications can be found in Appendix 1.

FIGURE 7 Forest plot for studies of D-dimer using the 500 ng/ml threshold (N = 18). Reproduced from Essat et al.1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

FIGURE 8 Summary plot for studies reporting D-dimer using the 500 ng/ml threshold (N = 18). Reproduced from Essat et al.1 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Studies of biomarkers other than D-dimer

We included 13 cohort studies of other biomarkers52,63,68,69,73,76–83 (sample size N = 76 to N = 999, prevalence of AAS 1–51%). A team of researchers in Italy undertook five of the studies, evaluating the following biomarkers in consecutive cohorts: matrix metalloproteinases 8 and 9,63 lactate dehydrogenase,81 white blood cell count, platelet count, fibrinogen,79 copeptin80 and soluble suppression of tumourigenicity-2 (sST2).78 The other studies were undertaken in China (three studies),69,73,76 Germany,83 Japan82 and Canada,68 evaluating the following biomarkers: troponin,68,73 alpha-smooth muscle actin,69 smooth muscle myosin heavy chain,69 soluble elastin fragments in serum,69 polycystin-1,69 acidic and basic calponin at six and 24 hours after presentation,82 sST2,73 neutrophil-lymphocyte ratio76 and leukocyte count.83 The methodological quality of the included studies was variable, with most studies having unclear or high risk of bias and applicability concerns in at least one item of the QUADAS-2 tool.

The area under the receiver operating characteristic (AUROC) for the biomarkers was generally modest, with sensitivity sufficient to rule out AAS only when a threshold was used that resulted in low specificity. The exception was the study of Wang et al.,73 reporting 99.1% sensitivity, 84.9% specificity and AUROC 0.97 for sST2, which differed markedly from the findings reported by Morello et al.78 for sST2 (sensitivity 58%, specificity 70.8%, AUROC 0.675). Accuracy improved when the biomarkers were combined with D-dimer but was not clearly superior to D-dimer alone.

We excluded 38 case-control studies85–122 that evaluated a wide variety of biomarkers using a variety of different control groups. Only three used control groups consisting of patients presenting with suspected AAS.103,104,115 Case-control studies may identify biomarkers for future research but do not provide reliable estimates of accuracy to inform clinical practice.24,25

We identified three studies18,33,84 that reported clinical scores, models or algorithms other than the ADD-RS. Morello et al.18 developed the AORTA score based on six clinical features (thoracic aortic aneurysm, severe pain, sudden pain, pulse deficit, neurologic deficit, hypotension) and reported an area under the curve (AUC) of 0.729. McLatchie et al.33 evaluated the AORTA score and reported an AUC of 0.689. McLatchie et al.33 also evaluated the Canadian guidelines (results reported above) and the Sheffield score, which was developed through clinical expertise (AUC 0.628). Duceau et al.84 used machine learning to develop an algorithm for pre-hospital triage of suspected AAS with AUC of 0.73 in a validation cohort.

Decision-analytic modelling

Table 5 shows the number of CTA performed and number of cases of AAS detected and missed in the base-case analysis at a typical hospital with 3281 cases of possible AAS per year, based on extrapolating from the incidence of cases in the DAShED study,33 and across the NHS, based on extrapolation from the incidence of AAS across the NHS in England and Wales (https://digital.nhs.uk/data-and-information/data-tools-and-services/data-services/hospital-episode-statistics). The low prevalence of AAS in this unselected population means that most strategies result in numbers of CTA that would markedly exceed current rates of CTA at a typical hospital, estimated to be 298 per year by extrapolation from the DAShED study. The exceptions (no testing, CTA if ADD-RS > 1) resulted in high rates of missed AAS.

Table 6 shows the deterministic results for the base-case analysis. CTA for those with ADD-RS > 1 was cost-effective at the £20,000/QALY threshold, although Table 3 suggests that over half of cases of AAS would be missed. CTA for those with ADD-RS > 1 or ADD-RS = 1 with D-dimer > 500 ng/ml was cost-effective at the £30,000/QALY threshold, although Table 3 suggests this would require tripling or quadrupling CTA capacity.

Figure 9 shows the cost-effectiveness acceptability curves from the base-case probabilistic sensitivity analysis. It shows that as the maximum acceptable incremental cost-effectiveness ratio (MAICER) increases, the strategy with the greatest probability of being cost-effective changes from no testing to ADD-RS > 1 and then ADD-RS > 1 or ADD-RS = 1 with D-dimer > 500 ng/ml.

FIGURE 9 Cost-effectiveness acceptability curves from the base-case probabilistic sensitivity analysis. Reproduced from Thokala et al.3 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Table 7 shows the results of secondary analysis assuming 0.61% prevalence of AAS, as seen in the DAShED cohort when patients with a clinician-estimated probability of AAS of zero were excluded. If clinicians can use their judgement to exclude patients they consider to have zero risk of AAS, then CTA for those with ADD-RS > 1 or ADD-RS = 1 with D-dimer > 500 ng/ml is cost-effective at the £20,000/QALY and 30,000/QALY thresholds, and would require 465 CTA per year in a typical hospital (≈60% increase).

Table 8 shows the results of sensitivity analysis assuming 2.95% prevalence of AAS, as seen in the DAShED study cohort who received CTA. This suggests that CTA for all is not cost-effective compared to strategies using ADD-RS and D-dimer. CTA for those with ADD-RS > 1 or D-dimer > 500 ng/ml is cost-effective at the £20,000/QALY or £30,000/QALY threshold and would approximately halve the number of CTA required compared to CTA all.

The EVPI analysis showed that at a threshold of £20,000/QALY, the individual EVPI was £4.46 per patient. Using a population size of 796,538 suspected AAS patients to the NHS each year and a 5-year horizon, the population EVPI was estimated as £17.75M.