Assessment of Non-Invasive Diagnostic Imaging Modalities Eciency for Detecting Myocardial Ischemia in Patients Suspected of Having Stable Angina

Purpose: This study aimed to assess and compare the detection eciency of non-invasive diagnostic imaging modalities as easy-to-understand indices for patients with myocardial ischemia. Methods: We included 1,000 patients with chest pain and possible coronary artery disease (CAD), based on their clinical condition. The modalities to be assessed were as follows: cardiac magnetic resonance imaging (CMRI), single-photon emission computed tomography, positron emission computed tomography (PET), stress echocardiography (SE), and fractional ow reserve derived from coronary computed tomography angiography (FFRCT). We used the decision tree simulation analysis to assess and compare the following: (1) the number of true positive (TP), false positive (FP), false negative (FN), and true negative (TN) test results per 1,000 patients, (2) positive predictive value (PPV), (3) negative predictive value (NPV), (4) post-test probability (post-TP), (5) diagnostic accuracy (DA), and (6) the number needed to diagnose (NND). Results: In the basic settings (pre-test probability: 50%), PET generated the highest TP (450), NPV (89%, 95% condence interval [CI]: 86%-92%), DA (87%, 95% CI: 85%-89%), and NND (1.35, 95% CI: 1.26-1.48). In contrast, CMRI produced the highest TN (435), PPV (87%, 95% CI: 84%-90%), DA (87%, 95% CI: 85%-89%), and NND (1.35, 95% CI: 1.26-1.48). In addition, FFRCT generated the highest FP (120). SE produced the highest FN (155) and post-TP (29%, 95% CI: 25%-33%). Conclusion: PET and CMRI were considered more ecient than other modalities. The results of our study will be useful for both physicians who order the examination and patients who undergo it.


Introduction
Numerous cardiac imaging methods are currently available for detecting myocardial ischemia associated with ischemia-causing coronary artery disease (CAD), including coronary computed tomography angiography (CCTA), cardiac magnetic resonance imaging (CMRI), single-photon emission computed tomography (SPECT), positron emission computed tomography (PET), stress echocardiography (SE), and fractional ow reserve derived from CCTA (FFRCT) [1,2]. Moreover, they are widely performed as a non-invasive diagnostic imaging modality, and several researchers have reported on their diagnostic ability [3][4][5][6]. Recent guidelines recommend using CCTA for workup in patients with suspected stable angina, based on their initial test results [7,8]. However, CCTA generates exclusively morphological information about coronary arteries. To diagnose the presence of myocardial ischemia irrespective of the presence coronary artery stenosis, it is recommended that myocardial perfusion should be assessed with another modality [8][9][10]. Studies that have detected myocardial ischemia by diagnostic imaging on stable angina primarily use sensitivity and speci city as indices, thus indicating their ability. However, patients without prior knowledge might nd it di cult to understand the meaning of the examination contents, despite being directly informed about the numerical values described in the literature. Therefore, it is desirable to clarify these indices not only as values obtained from the literature, but also as indices that are easy to understand. Additionally, this practice might be bene cial not just the patient but also the medical experts associated with their examination. Therefore, we aimed to assess and compare the e ciency of detecting myocardial ischemia using non-invasive diagnostic imaging modalities as easy-to-understand indices by simulation.

Study design
We included 1,000 patients with chest pain and any of the following clinical conditions: High clinical likelihood of obstructive CAD expected from basic testing [11]; Requirement for myocardial perfusion evaluation; Suspected CAD on CCTA.
A simulation analysis was performed to assess the e ciency. As a basic setting, we set the pre-test probability (PTP) of CAD to 50% by referring to past reports [8,11]. We used decision analysis [12] to calculate the e ciency, assuming the aforementioned group of patients would undergo the following ve types of examinations: We performed a literature search for the data analysis. We searched for meta-analysis articles on non-invasive diagnostic imaging modalities that used invasive FFR as a reference standard and investigated the diagnostic ability (sensitivity and speci city) of CAD on a patient basis. This literature search was performed using the PubMed database to identify articles published between January 2015 and October 2018. The search items were as follows: (i) diagnostic accuracy of coronary artery disease and (ii) diagnostic performance of coronary artery disease. In case of multiple results, we extracted the top two articles with the highest number of target studies described in the meta-analysis. In contrast, we selected the relatively new article if the number of target studies included was similar. Subsequently, we conducted a qualitative evaluation of the literature. Referring to the method reported by Chong et al. [13], the contents of each literature were evaluated using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Diagnostic Test Accuracy (PRISMA DTA) checklist [14]. The PRISMA DTA Checklist contains 27 items that assess the quality of meta-analyses. We categorized each checklist item of the candidate literature as follows: "su ciently described", "insu ciency described", and "not described". While one point was assigned to each checklist item with "su ciently described", zero points were assigned to other items. Moreover, we calculated the total score of each candidate literature. We eventually selected the literature with the highest total score for the analysis. In case of same scores, the relatively newer literature was selected. We eventually extracted the sensitivity and speci city from the selected literature and used them for the data analysis. Calculation of e ciencies In the aforementioned patient group, we assumed that a work-up examination had been performed to assess the presence of myocardial ischemia. Based on the sensitivity and speci city, we conducted a decision analysis using the Bayes' theorem. We calculated the PPV, NPV, and the probability of a positive or negative result from the PTP, sensitivity, and speci city. Moreover, we calculated the probabilities of nally arriving at the endpoint of each branch of the decision tree ( Fig. 1). Each probability was used to calculate the TP, FP, FN, and TN per 1,000 patients. The NND was calculated simultaneously. It indicates the number of patients to be tested to correctly detect the disease in one of them [16]. For calculation of e ciencies, we used the method published by Hsu et al. [17] to calculate the number of people. E ciencies were calculated and compared for each imaging modality. Table 1 summarizes the method used to calculate each e ciency. We simultaneously calculated the 95% con dence interval (95% CI) as the value of the point estimates from b)-f). We eventually compared the e ciencies of the estimated ve imaging modalities. The absence of an overlap between each 95% CI indicated a statistically signi cant difference while comparing the aforementioned e ciencies (p<0.05).

Sensitivity analyses
The PTP was set at 50% in the basic analysis settings. However, the PTP of CAD depends on the background factors of patients, such as sex, age, and the presence or absence of risk factors in individual patients [8,11]. Therefore, we conducted sensitivity analyses to assess the e ciencies, considering the uncertainties associated with the hypothesis-based analysis. The PTP was changed from 10% to 90%, centering on the intermediate PTP [8], which reportedly requires imaging tests to detect myocardial ischemia. The change in e ciencies in each imaging modality were evaluated and compared. The e ciencies targeted for the sensitivity analysis were limited to one, which varied with changes in the PTP. Their post-test probabilities were calculated using various PTPs and each e ciency.
We calculated each e ciency and 95% CI using R version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria, package: epiR) and Microsoft Excel for Mac 2016 Ver.16.16.27.
E ciencies at the basic settings Table 3  With an increase in PTP, estimates in DA for PET and FFRCT were increased (up to 4% and 11%, respectively), and SE was decreased (up to 6%).

Discussion
We evaluated and compared e ciencies of ve non-invasive diagnostic imaging modalities for the detection of myocardial ischemia in patients with stable angina. In addition, we conducted a sensitivity analysis to account for the variation in PTP due to differences in patient background factors. Our study ndings that may be useful to patients were as follows: Among the ve types of modality in basic settings (PTP: 50%), The maximum and minimum probability of a positive result and ischemia was 87% (CMRI) and 75% (SE), respectively.
The maximum and minimum probability of a negative result and no ischemia was 89% (PET) and 71% (SE), respectively.
Despite a negative result, the minimum and maximum probability of ischemia, i.e., the probability of missing detection was 11% (PET) and 29% (SE), respectively.
PET generated the best TP, NPV, and least FN among the ve imaging modalities.
CMR generated the best DA, PPV, TN, and least FP among the ve imaging modalities.
FFRCT produced more false-positive cases than PET, CMR, and SPECT.

SE was inferior to all modalities.
In addition, the following information may be useful to physicians who order the examinations: In the sensitivity analysis, PET generated the highest TP, NPV, and lowest FN in all PTPs.
The TPs and FNs of FFRCT were almost similar to those of PET.
The NPVs of FFRCT were almost similar to those of CMR.
PET is considered best for patients or physicians who focus on an accurate detection and less missed diagnoses of CAD. However, it has slightly higher FPs than CMR. This can be attributed to the relatively lower speci city, compared to that of CMR. The DA and NND of PET and CMR were almost similar in basic settings. However, the number of FP in PET was about 20 cases more than that in CMR. PET may be slightly inferior to CMR, in terms of its role as a gatekeeper for CAG or revascularization. In contrast, CMR is considered best if it focuses on higher DA, PPV, and less FP. The TPs and FNs of FFRCT are almost similar to PET. Thus, FFRCT should be added to CCTA when the results of CCTA are equivocal. However, the number of FPs in FFRCT was the highest among the ve modalities due to the lowest speci city of FFRCT (Fig. 2-b and Table 2). The FP and FN results can lead to an inaccurate diagnosis. In addition to unnecessary psychological distress, FP test results in patients with no disease can increase their medical risk due to additional examinations [24]. Moreover, FN test results can cause late diagnosis or misdiagnosis [24].
Among non-invasive diagnostic imaging modalities, researchers have primarily conducted studies to evaluate the e ciency of detecting myocardial ischemia in stable angina by economic analysis, such as costeffectiveness analysis, cost-bene t analysis, and cost-bene t analysis [25][26][27]. However, an interpretation of the indicators of e ciency obtained from the results, such as cost-effectiveness ratio and cost-utility ratio requires a certain degree of specialized knowledge. Therefore, patients might nd it di cult to understand these indicators, despite being presented directly with the information. This is the rst study that used currently available evidences to assess the e ciency of each modality to detect myocardial ischemia by simulation. Therefore, we could elucidate the number of TP, FN, FP, and TN per 1,000 patients as e ciencies. In addition, by comparing them, we could elucidate the difference in e ciency as a speci c index. Similarly, the e ciency of indices, such as PPV, NPV, DA, and post-test probability were also elucidated and compared with each other. The aforementioned calculations require setting the PTP. However, we were able to assess e ciencies at different PTPs using sensitivity analyses. Besides, physicians might easily understand the indicators using NND than standard diagnostic accuracy expressions, such as sensitivity and speci city [16]. Thus, it is conceivable that our results would help patients to understand the ability of each examination and undergo the appropriate one. Apart from sensitivity and speci city, the aforementioned indices would be needed not only by patients but also by physicians who order the examinations during busy practices. Furthermore, physicians can determine the degree of an inaccurate CAD diagnosis by the percentage and number of people. Therefore, our ndings might contribute to reviewing diagnostic strategies and improving the work ow for diagnosis in patients with suspected CAD. The primary purpose of using non-invasive imaging modalities was to select patients who were likely to bene t from invasive coronary angiography and revascularization [1,28,29]. Therefore, the importance of non-invasive imaging is increasing [1]. Each imaging modality has a good ability to detect CAD. Moreover, they contribute to the reduction of unnecessary revascularization and optimization of the diagnosis and treatment costs. However, physicians should refer to the various economic evaluations while considering the e ciencies of each examination, based on diagnostic costs.

Limitations
Our study has several limitations. First, each index calculated as the e ciency was a value calculated by simulation. Therefore, our results may not be appropriate in alternative situations. Hence, we considered citing the results of meta-analyses for determining the diagnostic ability. Moreover, we performed a sensitivity analysis to enable the application of the aforementioned method in different cases. Second, the diagnostic abilities of each modality used to calculate e ciencies were cited from the meta-analyses. We failed to obtain any literature data from the same patient population. In addition, we could not consider the difference in the diagnostic ability, depending on sex. Thus, it might have introduced bias. Third, we de ned the e ciency as "indices that are easily understood by patients". However, our results have not yet been used to explain to actual patients. Therefore, we failed to verify whether patients could understand the calculated e ciencies. This necessitates further veri cation by taking measures, such as hearing patient opinions.

Conclusion
We assessed and compared the e ciency of non-invasive imaging modalities for detecting myocardial ischemia in patients with suspected stable angina. PET and CMRI have a superior e ciency, compared to other methods. Our results shed light on the e ciency of the aforementioned modalities using the "easy-tounderstand index". Hence, they might prove useful for both physicians and patients.   Table 3. A summary of e ciencies at the basic settings (Pre-test probability = 50%).