Statistical analysis

Results

Baseline characteristics

There were 1044 patients who received stress echocardiography from the larger PROMISE cohort and 986 patients whose studies were interpretable (Figure 1). Their mean age was 59.6 +/- 8.17 years, 52.8% were women, and 1014 (97.1%) had one or more cardiovascular risk factors. Mean atherosclerotic cardiovascular disease ASCVD risk was 13.2(11.19). Detailed baseline demographic characteristics are summarized in Table 1. Providers felt that CAD risk was intermediate in 537 (51.4%) patients, which was consistent with an intermediate Diamond and Forrester score in 556 (53.3%). Chest pain was the primary symptom in 798 (76.6%) patients but was felt to be typical angina in 9.3%. There were 208 (21.1%) patients using a beta-blocker and these patients were more often sent for pharmacologic stress than exercise stress (36.7% vs 19.2%; p<0.001). The majority (99.0%) of stress echocardiograms were done using exercise as the stress modality (97.3% treadmill, 2.7% bicycle) .
X number of tests did not achieve target heart rate and Y number were felt to have non-diagnostic image quality.

Concordance and discordance between site and core laboratory interpretation

A small number of patients had resting left ventricular dysfunction, n of which [improved] with stress.
Agreement between site and core lab interpretation was moderate (kappa 0.46; 95% CI 0.35-0.57). Among discordant studies, more studies were determined to be abnormal by site but normal by core lab (n=51) but a small number were determined to be abnormal by core lab and normal by site (n=15). On multivariate analysis, patient variables predictive of a discordant interpretation were […]. Echocardiographic variables predictive of a discordant interpretation were […]. Accreditation status of the study site was/was not predictive of discordant stress echo interpretation ().
Compared to those with agreement, patients with discordant interpretations were older (63.5 [8.3] vs 59.3 [8.0] years), had higher mean ASCVD score (16.6 vs 13.0), and were more likely to be treated with beta-blocker (35.0% vs 20.1%) (all p ≤0.01). There was no difference in body mass index, chest pain characteristics, or risk factors between the concordant and discordant groups (all p>0.05). Echo characteristics associated with discordance were limited image quality as determined by sites (19.6 vs 8.7%, p=0.02) and any wall motion abnormality at rest or stress (by territory: left anterior descending [53.0 vs 2.7%], circumflex [22.7 vs 1.4%], and right coronary [42.4 vs 1.9%]) (all p<0.01). Variables not associated with discordance included stress type (exercise vs pharmacologic) or echo contrast use (both p>0.05).

Factors associated with concordant and discordant interpretations

Outcome events

The overall event rate among stress echocardiogram patients in the PROMISE trial \cite{disease2015}was low. In the small number of SE patients who had a clinical outcome, those with discordant SE results were more likely to undergo referral for a second non-invasive test or coronary angiography. A discordant stress echo result was not associated with 90-day revascularization, cardiovascular death, myocardial infarction, all-cause death, and hospitalization for unstable angina.

Discussion

Our study showed that among outpatients with chest pain who underwent stress echocardiography for risk stratification, agreement between between site and core lab interpretation was moderate, with disagreement mainly due to positive results by site interpretation that were normal by core lab. Predictors of discrepancy between site and core lab interpretations were [...]. 
Other studies have shown disagreement between site and core lab interpretations of cardiac testing. A substudy of the Global Registry of Acute Coronary Events (GRACE) registry found significant and clinically important discrepancies in local site and core laboratory interpretation of electrocardiograms in acute coronary syndrome patients \cite{Yan_2007}. In another substudy of the PROMISE trial, modest disagreement between site and core lab interpretations resulted from obstructive stenosis by visual site interpretation that was not hemodynamically significant by blinded quantitative assessment \cite{Shah_2017}. Previous echocardiographic studies have shown that reader experience contributes to the diagnostic accuracy of stress echocardiogram interpretation and accuracy of interpretation improves with adequate training \cite{Picano_1991}. A smaller substudy of the ISCHEMIA trial \cite{health2015} cohort found agreement between local site and core lab interpretation of myocardial ischemia by stress echocardiogram in 81% of cases and similar for both exercise and pharmacologic stress tests. In cases of disagreement, the majority were adjudicated as no or mild ischemia by the core lab and more extensive ischemia by the local site \cite{Kataoka_2015}. A multicenter study using stress echocardiograms from six institutions found an average kappa coefficient of 0.48 between institutions and percentage agreement was highest in cases of three-vessel disease or left anterior descending artery disease \cite{Peteiro_2006}. Factors that have been previously associated with disagreement include minor wall motion abnormalities \cite{Hoffmann1998}\cite{Hoffmann_2002}, low peak rate-pressure product \cite{Hoffmann1998}, low image quality\cite{Hoffmann1998}\cite{Hoffmann_2002}, and  Harmonic imaging has also improved interpreter agreement \cite{Hoffmann_2002}.
Because stress echocardiography is a commonly used imaging modality to discriminate between cardiac and non-cardiac chest pain and has implications for patient management, features which may influence non-expert interpretation of stress echo is of clinical importance. Previous studies have found a suggestive clinical history to significantly bias interpretation of electrocardiograms among physicians \cite{van_den_Berge_2013}. Multiple previous studies have shown that cognitive bias is associated with therapeutic or management errors \cite{Saposnik_2016}. Fortunately, the majority of cases with disagreement were due to a site positive result of ischemia with a core lab result of normal. While this may lead to an increased use of medical therapy and further testing, it is likely that functionally significant coronary artery disease is not being missed by non-expert sites. [Insert statement about specific variables that influenced discordance].