INTRODUCTION

Barrett's esophagus (BE) is a precursor to esophageal adenocarcinoma (EAC), and gastroenterology societies recommend considering screening endoscopy in patients with multiple risk factors such as age 50 years or older, male sex, smoking, obesity, and family history of BE/EAC (1). After diagnosis of BE, endoscopic surveillance with systematic biopsies is recommended to detect dysplasia and early EAC at endoscopically treatable stages (1–5). However, clinical management of BE can be challenging due to the limited predictive accuracy of currently available clinical and pathology variables and the limitation of sampling of only a small area of the esophagus with biopsies despite following Seattle protocol of 4-quadrant sampling at 1- to 2-cm intervals, which can miss areas of dysplasia and EAC (6,7). While clinicopathologic variables have some prognostic value, none of these are sufficient to effectively risk-stratify patients with BE to enable personalized management (8). The histologic finding of dysplasia is the main factor used by physicians to determine management plans for surveillance or endoscopic eradication therapy (EET), and guidelines recommend review of dysplasia by a gastrointestinal (GI) subspecialist/expert pathologist. However, there is significant variability among both community/generalist and expert pathologists in the diagnosis of dysplasia in BE (9,10). In clinical practice, this means that a patient's management plan, and thus their health outcome, can vary significantly depending on which pathologist reviews their slides. There is also significant variability in the surveillance and treatment recommendations by physicians in the management of BE (11,12). Increasing age, segment length, and male sex all have predictive value, and the latest guidelines from the American College of Gastroenterology recommend the use of segment length to guide surveillance intervals for patients with nondysplastic BE (NDBE) (1,13). However, in practice, these variables have limited value because patients with clinically low-risk features such as NDBE, short-segment BE, or female sex can present with prevalent high-grade dysplasia (HGD)/EAC or progress to HGD/EAC during their surveillance interval. Of note, 25% of patients with BE who later developed EAC were diagnosed within 1 year after index endoscopy and are therefore considered as “missed” cancers (6). Improved risk stratification is needed to enable early identification and treatment of patients with a pathologic diagnosis of NDBE or indefinite for dysplasia (IND) but who are harboring prevalent or develop incident HGD/EAC and to identify low-risk patients with LGD, IND, or NDBE who can be effectively managed by surveillance. A tissue systems pathology (TSP-9) test (TissueCypher Barrett's Esophagus Test) has been validated in 5 multicenter studies to predict the risk of incident progression of BE to HGD/EAC and to detect the presence of prevalent HGD/EAC (14–18). The aims of this study were to evaluate the predictive performance of the TSP-9 test in comparison with that of standard clinicopathologic variables and in clinically relevant subsets of patients in a large, diverse cohort of patients with BE with known outcome data to further characterize how the test may be used to improve the clinical management of patients with BE.

METHODS Study design and patients

A pooled analysis of data from 5 published studies (summarized in Table 1) was completed. Patients in the training study performed to develop the TSP-9 test were excluded (n = 183), and specimens from additional time points for 13 patients who were evaluated in more than 1 study were also excluded. Age, sex, segment length, presence or absence of hiatal hernia, original/real-world pathology diagnoses abstracted from health records, expert review pathology diagnoses provided by GI subspecialist pathologists as part of research studies, HGD/EAC–free surveillance time for nonprogressors, time to diagnosis of HGD/EAC for patients with prevalent HGD/EAC or incident progression to HGD/EAC, and TSP-9 risk classes (high, intermediate, and low) were evaluated. Prevalent HGD/EAC and incident progression data were collected from the previous publications in which prevalent HGD/EAC was defined as a diagnosis of HGD/EAC ≤1 year after the baseline endoscopy, and incident progression was defined as detection of HGD/EAC >1 year after the baseline endoscopy. The predictive performance of the TSP-9 test and clinicopathologic variables was compared, and the TSP-9 test was evaluated in subsets of patients defined by clinicopathologic variables as high risk or low risk. The TSP-9 test parameters are described in the see Supplementary Digital Content (see Supplemental Methods, https://links.lww.com/CTG/B7). Institutional Review Board approvals were obtained for each of the previously published studies.

Table 1. - Summary of studies included in pooled analysis Study Design Diagnoses Patient outcomes Key findings Critchley-Thorne et al, Cancer Epidemiology Biomarkers and Prevention, 2016;25:958–68 (14) Nested case-control
n = 366 NDBE = 329
IND = 8
LGD = 29 Nonprogressors = 287
Incident progressors = 79 1st validation study demonstrating that TSP-9 independently predicts incident progression from ND/IND/LGD to HGD/EAC Critchley-Thorne et al, Epidemiology Biomarkers and Prevention, 2017;26:240–8 (15) Nested case-control
n = 145 NDBE = 151
IND = 3
LGD = 21 Nonprogressors = 145a
Patients with prevalent HGD/EAC = 30 2nd validation study demonstrating that TSP-9 independently predicts presence of prevalent HGD/EAC in patients with Dx ND/IND/LGD Davison et al, American Journal of Gastroenterology, 2020;115(6):843–52 (16) Blinded, nested case-control
n = 268 NDBE = 175
IND = 49
LGD = 44 Nonprogressors = 210
Incident progressors = 58 3rd validation study demonstrating that TSP-9 independently predicts incident progression from ND/IND/LGD to HGD/EAC Frei et al, Clinical & Translational Gastroenterology, 2020;11(10):e00244 (17) Blinded, case-control study in ReBus Registry
n = 76 NDBE = 76 Nonprogressors = 38
Incident progressors = 38 4th validation study demonstrating that TSP-9 independently predicts incident progression from ND to HGD/EAC Frei et al, American Journal of Gastroenterology, 2021;116(4):675–82 (18) Blinded cohort study in the screening cohort of randomized controlled trial
n = 155 LGD = 155 Nonprogressors = 121
Progressors = 34 5th validation study demonstrating that TSP-9 independently risk stratifies patients with community-based LGD

EAC, esophageal adenocarcinoma; HGD, high-grade dysplasia; IND, indefinite for dysplasia; LGD, low-grade dysplasia; NDBE, nondysplastic Barrett's esophagus; TSP-9, Tissue Systems Pathology-9.

a145 nonprogressors were the same patients in the validation set described in Critchley-Thorne et al, Cancer Epidemiology Biomarkers and Prevention, 2016;25:958–68, and only data from the 2016 study were evaluated in this pooled analysis (14).
Statistical analyses

Kaplan-Meier curves were used to evaluate the risk of progression to HGD/EAC in the TSP-9 risk classes and in the diagnostic groups determined by pathology. Log rank test was used to assess the equality of progression curves from the Kaplan-Meier analysis. An accelerated failure time regression model with an exponential linkage was used to compare the performance of the TSP-9 risk classes with that of the clinical and pathology variables using univariable and multivariable analyses. Hazard ratios (HRs) with 95% confidence intervals (CIs) were calculated from the accelerated failure time regression model coefficients. Pathology diagnoses (LGD vs NDBE and IND vs NDBE), sex (male vs female), segment length (long [≥3 cm] vs short [<3 cm]), hiatal hernia (present vs absent), and TSP-9 risk classes (high vs low and intermediate vs low) were included as discrete variables. Age was assessed as a continuous variable.

Performance metrics for the TSP-9 test were sensitivity (% of progressors scored intermediate/high risk because both progressed at a significantly higher rate than patients who scored low risk), specificity (% of nonprogressors scored low risk), negative predictive value (NPV, % of patients scored low risk who did not progress), and positive predictive value (PPV, % of patients scored intermediate/high risk who progressed). The same metrics were calculated for both real-world and expert pathology diagnoses: sensitivity (% progressors with LGD), specificity (% of nonprogressors with NDBE or IND because patients with these diagnoses progressed at similar rates), PPV (% of patients with LGD or IND/LGD combined who progressed), and NPV (% of patients with NDBE who did not progress). NPV and PPV were adjusted for prevalence based on published estimates of progression rates from NDBE, IND, or LGD to HGD/EAC (19–21). Annual progression rates were calculated from 5-year NPV and PPV. The number needed to predict (NNP) was calculated using the prevalence-adjusted PPVs and prevalence-adjusted percentages of patients scoring in risk classes and diagnostic groups.

RESULTS Patient characteristics

We evaluated 699 patients from 5 published clinical studies, of which 40 (5.7%) had prevalent HGD/EAC, 150 (21.5%) were incident progressors, and 509 (72.8%) did not progress to HGD/EAC during follow-up (Table 2). The time to diagnosis of HGD/EAC was a median of 0.4 years (interquartile range [IQR], 0.17–0.62) for patients with prevalent HGD/EAC and 3.2 years (IQR, 2–5) for incident progressors. Nonprogressors had an HGD/EAC–free surveillance time of 6.7 years (IQR, 5–9). The median age of patients was 61 years (IQR, 53.1–69.1, range 19–88), 78.5% were male, and 65.7% of patients had long-segment BE (≥3 cm). The real-world pathology diagnosis was NDBE in 56.1% of patients, 9% had IND, and 34.9% had LGD. After an expert pathology review, 81.1% of patients had NDBE, 7.2% had IND, and 11.7% had LGD. The TSP-9 test scored 16% of patients as high risk, 13.7% intermediate risk, and 70.2% low risk for progression to HGD/EAC within 5 years.

Table 2. - Patient characteristics Nonprogressors Patients with missed prevalent HGD/EAC Patients with incident progression to HGD/EAC n (%) 509 (72.8%) 40 (5.7%) 150 (21.5%) Age, median (IQR and range) 61 (IQR, 53–69, range 19–88) 62 (IQR, 55–68, range 47–79) 63 (IQR, 56–70, range 19–87) Sex, n (%)  Male 385 (75.6%) 35 (87.5%) 129 (86.0%)  Female 124 (24.4%) 5 (12.5%) 21 (14.0%) Segment length, n (%)  Long 311 (61.1%) 27 (67.5%) 121 (80.7%)  Short 148 (29.1%) 9 (22.5%) 27 (18.0%)  Unknown 50 (9.8%) 4 (10.0%) 2 (1.3%) Hiatal hernia, n (%)  Absent 77 (15.1%) 32 (80.0%) 116 (77.3%)  Present 357 (70.1%) 7 (17.5%) 19 (12.7%)  Unknown 75 (14.7%) 1 (2.5%) 15 (10.0%) HGD/EAC–free surveillance timea, median yrs (IQR) 6.7 (5–9) NA NA Time to diagnosis of HGD/EACb NA 0.41 (0.17–0.62) 3.2 (2.2–4.6) Real-worldc pathology Dx, n (%)  NDBE 290 (57.0%) 11 (27.5%) 91 (60.7%)  IND 51 (10.0%) 3 (7.5%) 9 (6.0%)  LGD 168 (33.0%) 26 (65.0%) 50 (33.3%) Expert review pathology Dx, (n) (%)  NDBE 438 (86.1%) 19 (47.5%) 110 (73.3%)  IND 40 (7.9%) 1 (2.5%) 9 (6.0%)  LGD 31 (6.1%) 20 (50.0%) 31 (20.7%) TissueCypher risk results, n (%)  High, n (%) 32 (6.3%) 21 (52.5%) 59 (39.3%)  Intermediate, n (%) 61 (12.0%) 10 (25.0%) 25 (16.7%)  Low, n (%) 416 (81.7%) 9 (22.5%) 66 (44.0%)

EAC, esophageal adenocarcinoma; HGD, high-grade dysplasia; IND, indefinite for dysplasia; IQR, interquartile range; LGD, low-grade dysplasia; NDBE, nondysplastic Barrett's esophagus; TSP-9, Tissue Systems Pathology-9.

aTime between baseline endoscopy from which biopsies were tested with the TSP-9 test and last follow-up.

bTime between baseline endoscopy from which biopsies were tested with the TSP-9 test and the diagnosis of HGD/EAC at a subsequent endoscopy.

cReal-world pathology Dx is the original diagnosis abstracted from health records; these diagnoses were provided by a mix of generalist and expert pathologists at the clinical sites that participated in this study.

The TSP-9 test provides significant risk stratification in patients with BE that outperforms current clinicopathologic variables

The predictive performance of the TSP-9 test and the pathology diagnoses was compared. The TSP-9 test provided significant risk stratification in patients with NDBE, IND, or LGD (P < 0.0001, Figure 1a). Patients with BE who scored high risk by the test were 7.8× more likely to progress (HR = 7.8 95%, CI 5.6–10.9, P < 0.0001) compared with patients who scored low risk. Patients who scored intermediate risk were also at significantly increased risk for progression vs those who scored low risk (HR = 2.7, 95% CI 1.7–4.2, P < 0.0001) (Figure 1a). The real-world pathology diagnoses of NDBE, IND, and LGD did not provide significant risk stratification (HR = 1.4, 95% CI 1–1.9, P = 0.0525) (Figure 1b). The expert pathology review diagnosis showed improved risk stratification compared with the real-world diagnosis (Figure 1c). Patients with expert diagnoses of LGD were 4.2× more likely to progress (HR = 4.2, 95% CI 2.9–6.3, P < 0.0001) compared with patients with NDBE; however, patients with IND progressed at a similar rate to patients with NDBE (Figure 1c). The TSP-9 test detected 62.3% of progressors compared with 28.3% for the expert review diagnosis of LGD (see Supplementary Table S1 and S2, https://links.lww.com/CTG/B8, https://links.lww.com/CTG/B9). When used in combination, the TSP-9 test and the expert review diagnosis of LGD detected 67.9% of progressors.

Figure 1.:

TissueCypher (TSP-9) provides significantly improved risk stratification vs pathology diagnoses. Kaplan-Meier (KM) analysis of probability of progression to HGD/EAC in patients with BE stratified into the following: (a) low-risk, intermediate-risk, and high-risk classes by the TSP-9 test; (b) NDBE, IND, and LGD subsets by the original/real-world diagnoses abstracted from health records; (c) NDBE, IND, and LGD by expert review diagnoses provided as part of research studies. (d) Multivariable analysis comparing prediction of progression by the TSP-9 test vs expert review diagnosis. n = 699 patients with BE in panels a–d, including 150 incident progressors, 40 patients with prevalent HGD/EAC, and 509 nonprogressors. Five-year hazard ratios (HRs) with a 95% confidence interval (CI) were calculated from an accelerated failure time (AFT) regression model. EAC, esophageal adenocarcinoma; HGD, high-grade dysplasia; IND, indefinite for dysplasia; LGD, low-grade dysplasia; NDBE, nondysplastic Barrett's esophagus.

Univariable analysis of clinicopathologic features and TSP-9 risk results demonstrated stronger predictive power of TSP-9 risk results than the clinicopathologic predictors including segment length, hiatal hernia, age, sex, and pathology diagnoses (see Supplementary Table S3, https://links.lww.com/CTG/B10). In multivariable analysis, TSP-9 high-risk (HR = 6.82, 95% CI 4.79–10.50, P < 0.0001), intermediate-risk (HR = 2.34, 95% CI 1.40–3.97, P = 0.0006), male sex (HR = 1.84, 95% CI 1.19–3.35, P = 0.0196), and diagnosis of LGD real-world (HR = 1.62, 95% CI 1.16–2.40, P = 0.007) or expert review (HR = 2.47, 95% CI 1.54–4.03, P < 0.0001) were independent predictors of progression (Table 3). In an additional direct comparison in a multivariable model, the TSP-9 risk result was a stronger predictor of progression (HR = 6.2, 95% CI 4.4–8.9) than the expert diagnosis (HR = 2.2, 95% CI 1.5–3.4) (Figure 1d). The number of patients in the target population who would need to be tested to identify a patient who will be diagnosed with HGD/EAC within 5 years (i.e., the prevalence-adjusted NNP) was also evaluated. The NNP was 45 for the TSP-9 test when using the high-risk class alone or 32 when using intermediate-risk and high-risk classes combined, and the NNP for expert pathology was 70 (see Supplementary Table S4, https://links.lww.com/CTG/B11).

Table 3. - Multivariable analysis of predictive performance of clinical factors and pathology diagnoses with and without TissueCypher (TSP-9) Clinical factors and real-world Dx (without TSP-9) Clinical factors, real-world Dx with TSP-9 Variable HR (95% CI) P Variable HR (95% CI) P Segment length–long vs short 1.46 (0.98–2.3) 0.0617 Segment length–long vs short 1.29 (0.85–2.00) 0.2102 Hiatal hernia–present vs absent 1.03 (0.68–1.84) 0.8937 Hiatal hernia–present vs absent 0.94 (0.60–1.55) 0.816 Age (per yr) 1.02 (1.00–1.84) 0.0231 Age (per yr) 1.01 (1.00–1.03) 0.1371 Sex–male vs female 2.10 (1.32–4.25) 0.0042 Sex–male vs female 1.84 (1.19–3.35) 0.0196 Real-world Dx–IND vs NDBE 1.01 (0.46–1.80) 0.9661 Real world Dx–IND vs NDBE 1.10 (0.51–2.12) 0.7679 Real-world Dx–LGD vs NDBE 1.75 (1.22–2.48) 0.0015 Real world Dx–LGD vs NDBE 1.62 (1.16–2.40) 0.007 Harrell's C-index: 0.6437 (0.6079–0.6941) TSP-9–intermediate vs low risk 2.34 (1.40–3.97) 0.0006 TSP-9–high vs low risk 6.82 (4.79–10.50) <0.0001 Harrell's C-index: 0.7574 (0.7214–0.8019) Clinical factors with expert review Dx (without TSP-9) Clinical factors, expert review Dx with TSP-9 Variable HR (95% CI) P Variable HR (95% CI) P Segment length–long vs short 1.56 (1.05–2.39) 0.028 Segment length–long vs short 1.32 (0.85–2.13) 0.1812 Hiatal hernia–present vs absent 1.04 (0.67–1.76) 0.863 Hiatal hernia–present vs absent 0.96 (0.62–1.63) 0.8616 Age (per yr) 1.02 (1.00–1.03) 0.053 Age (per yr) 1.01 (0.99–1.03) 0.2138 Sex–male vs female 1.85 (1.17–3.68) 0.018 Sex–male vs female 1.68 (1.08–3.12) 0.0493 Expert review Dx–IND vs NDBE 1.52 (0.57–2.97) 0.227 Expert–IND vs NDBE 1.26 (0.52–2.63) 0.5147 Expert Review Dx–LGD vs NDBE 4.90 (3.38–7.45) <0.0001 Expert–LGD vs NDBE 2.47 (1.54–4.03) <0.0001 Harrell's C-index: 0.6899 (0.6546–0.7411) TSP-9–intermediate vs low risk 2.21 (1.30–3.71) 0.0014 TSP-9–high vs low risk 5.26 (3.52–8.13) <0.0001 Harrell's C-index: 0.7717 (0.7365–0.8154)

CI, confidence interval; HR, hazard ratio; IND, indefinite for dysplasia; LGD, low-grade dysplasia; NDBE, nondysplastic Barrett's esophagus; TSP-9, Tissue Systems Pathology-9.

The predictive power of the TSP-9 test and clinicopathologic variables was compared using multivariable regression models in which progression to HGD/EAC was evaluated in reduced models in relation to age, sex, segment length, presence of hiatal hernia, and real-world or expert review pathology diagnoses and in full models in relation to the same clinicopathologic variables and the risk class results produced by the TSP-9 test.

The TSP-9 test predicts incident progression and detects missed prevalent HGD/EAC

The performance of TSP-9 and expert review diagnosis in predicting incident progression and predicting the presence of prevalent HGD/EAC was evaluated. The TSP-9 test demonstrated 57.1% sensitivity in detecting incident progressors compared with 20.7% for the expert diagnosis of LGD (Figure 2a). Patients who were scored high or intermediate risk by the test were 7× (HR = 6.6, 95% CI 4.5–9.9, P < 0.0001) or 2× (HR = 2.1, 95% CI 1.1–3.4, P = 0.0079), respectively, more likely to develop incident HGD/EAC than patients who scored low risk (Figure 2a). Patients who scored high or intermediate risk were 31× (HR = 30.9, 95% CI 13.4–87.6) or 8× (HR = 7.6, 95% CI 2.9–20.8), respectively, more likely to be harboring prevalent HGD/EAC (P < 0.0001, Figure 2b). Furthermore, the TSP-9 test demonstrated 77.5% sensitivity in detecting missed cases with prevalent HGD/EAC (Table 2), whereas the sensitivity of the expert diagnosis of LGD was 50.0%, indicating that the TSP-9 test is a significantly more sensitive method for detecting missed prevalent HGD/EAC. The 5-year probability of progression to HGD/EAC increased continuously as the TSP-9 risk score increased (Figure 2c).

Figure 2.:

TissueCypher (TSP-9) test predicts incident progression and detects the presence of missed prevalent HGD/EAC. (a) Kaplan-Meier (KM) analysis of probability of progression to HGD/EAC in nonprogressors (n = 509) and incident progressors (n = 150) stratified into low-risk, intermediate-risk, and high-risk classes by the TSP-9 test. (b) Box and whisker plots of the TSP-9 risk score in nonprogressors (n = 509) and patients with prevalent HGD/EAC (n = 40). (c) Prevalence-adjusted probability of progression to HGD/EAC within 5 years as a continuous function of the TSP-9 risk score in all 699 patients. Dashed curves indicate a 95% confidence interval (CI). EAC, esophageal adenocarcinoma; HGD, high-grade dysplasia.

The TSP-9 test provides significant risk stratification in the diagnostic subsets of BE

The predictive performance of the TSP-9 test was evaluated in the subset of patients with NDBE and IND/LGD. The IND and LGD subsets were combined due to the limited sample size in the IND (n = 50) and LGD (n = 82) subsets diagnosed by expert review. The TSP-9 test provided significant risk stratification in the NDBE and IND/LGD subsets (Figure 3). The test identified 59% of the progressors with a real-world pathology diagnosis of NDBE, and 56.7% of progressors with expert pathology diagnoses of NDBE, demonstrating that the test can identify most patients with NDBE who harbor missed prevalent HGD/EAC or develop incident HGD/EAC (see Supplementary Table S2, https://links.lww.com/CTG/B9). Patients with NDBE who scored TSP-9 high risk progressed at a similar rate (3.2%/yr) to patients with expert-confirmed LGD (3.7%/yr), and at a higher rate than patients with expert-confirmed IND/LGD (2.3%/yr) indicating that the test identifies a subset of patients with NDBE who may benefit from EET (see Supplementary Tables S1 and S2, https://links.lww.com/CTG/B8, https://links.lww.com/CTG/B9). In the subset of patients with expert diagnoses of IND/LGD, patients who scored TSP-9 low-risk progressed to the combined end point of HGD/EAC at a lower rate (0.45%/yr, see Supplementary Table S2, https://links.lww.com/CTG/B9) than the patients with expert-confirmed NDBE (0.78%/yr, see Supplementary Table S1, https://links.lww.com/CTG/B8).

Figure 3.:

TissueCypher (TSP-9) provides significant risk stratification in diagnostic subsets of NDBE and IND/LGD. Kaplan-Meier (KM) analysis of probability of progression to HGD/EAC in patients stratified into low-risk, intermediate-risk, and high-risk classes by the TSP-9 test in the following subsets of patients with: (a) real-world pathology diagnosis of NDBE (n = 290 nonprogressors, 91 incident progressors, and 11 patients with prevalent HGD/EAC); (b) Real-world pathology diagnosis of IND/LGD (n = 219 nonprogressors, 59 incident progressors, and 29 patients with prevalent HGD/EAC); (c) Expert review diagnosis of NDBE (n = 438 nonprogressors, 110 incident progressors, and 19 patients with prevalent HGD/EAC); (D) Expert review diagnosis of IND/LGD (n = 71 nonprogressors, 40 incident progressors, and 21 patients with prevalent HGD/EAC). Five-year hazard ratios (HRs) with a 95% confidence interval (CI) were calculated from an accelerated failure time (AFT) regression model. EAC, esophageal adenocarcinoma; HGD, high-grade dysplasia; IND, indefinite for dysplasia; LGD, low-grade dysplasia; NDBE, nondysplastic Barrett's esophagus.

The TSP-9 test detects progressors in clinically high-risk and low-risk subsets of patients with BE

The predictive ability of the TSP-9 test was evaluated in patient subsets stratified by sex and segment length. The test provided significant risk stratification in the female and male subsets (Figure 4a,b) and in the subsets with short-segment and long-segment BE (Figure 4c,d). Consistent with published literature, male sex or patients with long-segment BE harbored more progressors (29.9% and 32.2%, respectively) than the female sex or patients with short-segment BE (17.3% and 19.6%, respectively, Table 2). The TSP-9 test demonstrated clinically impactful sensitivity in all evaluated subsets, identifying 71.4% of female progressors, 60.9% of male progressors, 54.5% of progressors with short-segment BE, and 65.8% of progressors with long-segment BE, demonstrating that the test can risk stratify independently of clinical variables (see Supplementary Table S2, https://links.lww.com/CTG/B9).

Figure 4.:

TissueCypher (TSP-9) provides significant risk stratification in subsets of male, female, short-segment, and long-segment BE patients. Kaplan-Meier (KM) analysis of probability of progression to HGD/EAC in patients stratified into low-risk, intermediate-risk, and high-risk classes by the TSP-9 test in the following patient subsets: (a) Female patients (n = 124 nonprogressors, 21 incident progressors, and 5 patients with prevalent HGD/EAC); (b) Male patients (n = 385 nonprogressors, 129 incident progressors, and 35 patients with prevalent HGD/EAC); (c) Short-segment BE (n = 148 nonprogressors, 27 incident progressors, and 9 patients with prevalent HGD/EAC); (d) Long-segment BE (n = 311 nonprogressors, 121 incident progressors, and 27 patients with prevalent HGD/EAC). Five-year hazard ratios (HRs) with a 95% confidence interval (CI) were calculated from an accelerated failure time (AFT) regression model. EAC, esophageal adenocarcinoma; HGD, high-grade dysplasia.

DISCUSSION

This study compared the predictive performance of the TSP-9 test and that of currently used clinicopathologic variables to assess the effectiveness of the test in stratifying patients with BE for risk of progression to HGD/EAC. The TSP-9 test demonstrated significantly improved risk stratification compared with real-world pathology or expert pathology or clinical variables. The test predicted incident progression and predicted the presence of missed prevalent HGD/EAC in patients with BE. Both the high-risk and intermediate-risk classes were at significantly increased risk of progression compared to the low-risk class, indicating that both the high-risk and intermediate-risk results may be clinically actionable in identifying patients who can benefit from escalation in management to prevent EAC. The TSP-9 test risk stratified patients with BE with overall higher sensitivity (62.3%) in detecting progressors than the expert diagnosis of LGD (28.3%). The higher sensitivity of the TSP-9 test led to a significantly lower NNP (32) when compared with the expert pathology diagnosis of LGD (70). The improved sensitivity was due to the TSP-9 test's ability to detect 57% of the progressors with expert-confirmed NDBE. In patients with NDBE, the test identified a high-risk subset that progressed at a similar rate (3.2%/yr) to the patients with expert-confirmed LGD (3.7%/yr). The TSP-9 high-risk class was a stronger predictor of progression than clinicopathologic variables and identified progressors within patient subsets that are considered to be at very low risk based on their clinical variables. Collectively, these results demonstrate that the TSP-9 test is an independent predictor of progression in patients with BE that may increase early detection of progressors at the NDBE stage and can guide risk-aligned management decisions to improve health outcomes.

Approximately 90% of the population with BE is diagnosed with NDBE on surveillance endoscopy. The overall rate of progression to HGD/EAC is low in these patients, and the reported annual progression rate of 0.63%/yr indicates that 1.9%–3.2% of patients with NDBE will progress during the 3- to 5-year surveillance interval that is recommended by societal guidelines (21,22). Considering the size of the patient population with NDBE in the United States, this group likely harbors many patients who are at high risk of progression. A key finding of this study was that TSP-9 identified most progressors in the subset of patients with expert diagnoses of NDBE. In this diverse cohort, patients with expert diagnoses of NDBE who scored TSP-9 high risk progressed at a rate of 3.2%/yr, which was similar to the rate of progression observed in patients with expert-confirmed LGD (3.7%/yr) in this study. The 3.2% annual rate of progression in patients with NDBE who scored TSP-9 high risk was higher than published estimates of progression from LGD (1.7%/yr) from a meta-analysis that included studies of community and expert diagnoses and within the 1.8%–9.1% per year range of progression estimates for confirmed LGD in the United States and Europe (20,23,24). Current GI societal guidelines recommend EET or close surveillance for patients with confirmed LGD based on the increased risk of progression in this group of patients (1). The findings reported here indicate that there is an additional group of patients within the population with NDBE who, based on their significantly increased progression risk, will benefit from being managed in a similar manner to patients with LGD. While patients with NDBE lack observable morphologic changes consistent with dysplasia, their Barrett's mucosa can harbor molecular and cellular changes associated with increased risk of progression (25), which can be detected by TSP-9. Patients at high risk of progression can benefit significantly from EET, which is highly effective at preventing neoplastic progression (26). Radiofrequency ablation, the most commonly used EET, is 98%–99% effective in eradicating dysplasia in BE and has been shown to reduce progression from LGD to HGD/EAC from 12.6% to 1.7%, which can significantly improve patient health outcomes (27,28). Close surveillance can also be an effective strategy to detect dysplasia and EAC at early, treatable stages, leading to improved health outcomes. An additional important finding of this study was the ability of the TSP-9 test to detect 77.5% of patients with prevalent HGD/EAC, indicating that the test could provide an effective method to reduce the clinical impact of post-endoscopy esophageal adenocarcinoma, which accounts for approximately one-quarter of HGD/EAC in patients with BE (29). Earlier detection by the TSP-9 test in these patients has the potential to significantly reduce the burden of missed HGD/EAC.

While guidelines recommend escalation of management for confirmed LGD, the diagnosis can be challenging due to significant interobserver variability. Diagnoses of IND and expert-confirmed LGD are associated with significantly increased risk of progression compared with that of NDBE. However, the estimated 1.5% annual rate of progression from IND and 1.8%–9.1% (range of published rates) annual progression rate from confirmed LGD indicate that most patients with IND or LGD are not at increased risk and may be effectively managed by surveillance (19,20,23,24). This study demonstrated that patients with expert diagnoses of IND/LGD who scored TSP-9 low risk progressed at a lower rate (0.45%/yr) than patients with expert-confirmed NDBE (0.78%/yr). This finding suggests that the TSP-9 test may be able to identify lower-risk patients with IND or LGD who may be effectively managed by surveillance. Objective identification of patients who can safely avoid this intervention can lead to quality-of-life improvements and more efficient use of health care resources (30,31).

Recent guidelines recommend use of segment length to guide surveillance intervals for NDBE, and male sex has also been associated with higher risk of progression (1,32). While patients with short-segment BE and female patients represent lower-risk patients, subsets of these patients will progress during surveillance intervals, and these patients would benefit from additional risk stratification. This study demonstrated that the TSP-9 test stratifies risk independent of clinical and pathology variables, indicating that the test can provide valuable risk stratification in the full population of patients in surveillance for BE.

These results extend findings from a recent study that showed significant improvement in the prediction of progression to HGD/EAC when the TSP-9 test results were used adjunctly with clinicopathologic variables (33). Novel findings reported here include demonstration that the TSP-9 test has significantly higher sensitivity for detection of incident progressors and patients with prevalent HGD/EAC when compared with pathology diagnoses and that the test can risk stratify within subsets considered to be at low risk based on clinicopathologic factors such as patients with NDBE, short-segment BE, and female patients. This study also reported the number of patients who need to be tested with TSP-9 to predict true progressors and reported prevalence-adjus