The Tissue Systems Pathology Test Predicts Risk of Progression in Patients With Barrett’s Esophagus

Abstract

Goals:

A systematic review and meta-analysis of published clinical validity studies was conducted to evaluate the predictive performance of the TSP-9 test.

Background:

Identifying patients with Barrett’s esophagus (BE) who will progress to high-grade dysplasia (HGD) or esophageal adenocarcinoma (EAC) is challenging. The tissue systems pathology (TSP-9) test can predict risk of progression to HGD/EAC in BE patients.

Study:

Databases were searched for studies that assessed the clinical validity of TSP-9, and data describing progressors, non-progressors, TSP-9 results, and hazard ratios (HR) with 95% confidence intervals (CIs) were extracted. Odds ratios (OR), sensitivity, specificity, and prevalence-adjusted positive and negative predictive values (PPV_adj/NPV_adj) were calculated and used for meta-analysis.

Results:

Six studies met eligibility criteria, comprising 699 patients. ORs and HRs for TSP-9 had mean common effect size estimates of 6.52 (95% CI: 4.40-9.66, P<0.0001, I ² =33%) and 6.66 (95% CI: 4.59-9.66, P<0.0001, I ² =0%), respectively, for predicting progression to HGD/EAC. Mean common effect size estimates were 61% (95% CI: 54%-68%) for sensitivity, 81% (95% CI: 78%-84%) for specificity, 28% (95% CI: 17%-42%) for PPV_adj (high risk), 14% (95% CI: 9%-21%) for PPV_adj (high/int risk), and 97% (95% CI: 96%-98%) for NPV_adj with minimal inter-study heterogeneity (I ² =79%, 21%, 0%, 0%, and 0%, respectively).

Conclusions:

Effect estimates of TSP-9 performance demonstrate that the test provides risk stratification for BE patients. The TSP-9 test can provide clinically impactful results to enable escalation of care for high-risk patients or to identify low-risk patients who can be safely managed with routine surveillance.

Barrett’s esophagus (BE) is the only known precursor to esophageal adenocarcinoma (EAC).^1–3 EAC has a 5-year survival rate of approximately 20% overall and 5% for metastatic disease.^4,5 Safe and effective endoscopic eradication therapy (EET), such as endoscopic resection, radiofrequency ablation (RFA), and cryotherapy, are available to prevent progression of BE to high-grade dysplasia (HGD) and EAC, and to treat dysplasia and early-stage EAC; in contrast, invasive or metastatic EAC has limited curative treatment options.^6–11 Thus, early identification of patients with BE who are at high risk for progression to HGD/EAC is critical for managing patients in a risk-stratified manner to optimize patient health outcomes.

Traditionally, society guidelines on clinical management practices for patients with BE recommend that physicians utilize the histologic grade of dysplasia and clinical factors such as segment length, to guide management decisions for treatment and surveillance of patients with BE.^1,10–15 The most recent society guidelines have acknowledged that EET may be appropriate for subsets of patients with non-dysplastic BE (NDBE) who are at increased risk for progression to EAC.¹¹ The approach of using histologic grade to guide management decisions has limitations due to the random nature of biopsy methods that sample a very small area of the esophagus, as well as significant inter-observer variability in the histologic grading of endoscopic biopsies.^1,14,16 Additional tools are needed for more accurate risk stratification in patients with BE to guide risk-aligned clinical management strategies. Patients at high risk for progression can benefit from early intervention with EET to reduce the incidence of EAC or short-interval surveillance to detect EAC at earlier stages and as such, reduce morbidity and mortality. Identification of patients at low risk for progression can reduce unnecessary procedures, improve quality of life for patients, and enable more effective use of health care resources.

The tissue systems pathology test (TissueCypher, TSP-9) is a clinically available risk stratification test for patients with BE. The TSP-9 test analyzes tissue biopsies collected during endoscopy using an artificial intelligence-driven computational pathology platform that quantifies 15 spatialomics-based features derived from 9 protein-based biomarkers and nuclear morphology in whole-slide images. The TSP-9 test integrates these quantitatively derived features to produce a risk score that ranges from 0 to 10 and risk class that risk-stratifies patients with BE as having a low-, intermediate-, or high-risk for progression to HGD/EAC within 5 years.^17,18 The clinical validity and utility of the TSP-9 test have been demonstrated in several multicenter studies^18–29 and a recent American Gastroenterological Association (AGA) Clinical Practice Update included a Best Practice Advice Statement that recognized that the TSP-9 test may be utilized for risk stratification of patients with NDBE,³⁰ and a recent AGA Guideline noted that the TSP-9 test may be used to identify patients with NDBE who are at increased risk for progression and may benefit from EET to prevent development of EAC.¹¹

This study aimed to conduct a systematic review and meta-analysis to evaluate the performance of the TSP-9 test for stratifying patients with BE for risk of progression to HGD/EAC within 5 years. For studies meeting eligibility criteria, predictive performance of the TSP-9 test was analyzed based on outcomes of progression or absence of progression during patient surveillance for BE patients diagnosed with NDBE, indefinite for dysplasia (IND), or low-grade dysplasia (LGD) at the time of TSP-9 testing. Since approximately 90% of patients with BE are diagnosed with NDBE,^31,32 and ∼50% of the patients who progress to HGD/EAC each year in the US are initially diagnosed with NDBE,^32–36 this study also aimed to evaluate the ability of the TSP-9 test to identify high-risk patients within the NDBE population who can be missed by current standard-of-care practices.

METHODS

Data Sources and Search Strategy

This study was designed and performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and Cochrane guidelines.^37,38 The methodology used for this review is summarized below.

A comprehensive search was performed of PubMed/MEDLINE, CENTRAL, WHO International Clinical Trials Registry Platform (ICTRP), and Clinicaltrials.gov portal databases from inception to September 27, 2023. The search included keywords and database-specific subject terms for BE and the tissue systems pathology (TSP-9) test (Supplemental Digital Content 1 http://links.lww.com/JCG/B277). After duplicate removal, 2 investigators (SSB and MA) independently reviewed titles and abstracts of all retrieved articles and excluded articles not addressing this review’s objectives/questions. The full texts of the remaining articles were reviewed for eligibility based on inclusion/exclusion criteria. A PRISMA flow chart was generated (Fig. 1) to illustrate article identification and screening steps for this systematic review to select studies to be included in the meta-analysis.

FIGURE 1.

PRISMA flow diagram demonstrating literature search results and study selection for meta-analysis.

Inclusion and Exclusion Criteria

For inclusion in this review, studies had to be primary, published, and peer-reviewed and had to describe patients with BE who had undergone TSP-9 testing and had a reportable result, had a pathology diagnosis of NDBE, IND, or LGD associated with the specimens tested by TSP-9, and had reported clinical outcome data on progression to HGD/EAC or no progression during endoscopic surveillance. Conference abstracts were not included. Studies were excluded if they were secondary research, if patients were treated with EET during the study period, or if the study evaluated only clinical utility and not clinical validity of the TSP-9 test. For duplicate/companion studies that related to the same patients from a previous study, data were combined to include all available data without duplication based on information available in each study. When different methods of evaluation were reported (such as when spatial or temporal analyses were reported), the TSP-9 performance metrics that represent how the TSP-9 is performed clinically (e.g., reporting the highest scoring specimen block when multiple specimen blocks were evaluated) were extracted for analysis to evaluate how the test would perform in a real-world clinical practice setting.

Data Extraction

Data were extracted using a modified version of the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies (CHARMS-PF checklist).^39,40 Qualitative data extraction of TSP-9 performance metrics and risk classes relative to disease progression or absence of progression during surveillance was performed independently by 2 investigators (CLL and SSB) and compared (Table 1). Disagreements were resolved through discussion between the 2 investigators who performed data extraction.

TABLE 1.

Summary of Data Used From Included Studies

Study name	Study number	Overlapping data	Data included in analysis
Critchley-Thorne et al18	1	Non-progressors in this study were the same as non-progressors in Study 2. This study also includes 13 progressors who were also evaluated in Study 4.	Individual data for OR, sensitivity, specificity, PPVadj (TSP-9 high/int-risk results and TSP-9 high-risk results), and NPVadj; HR estimate. Note: Individual data from progressors were combined for Study 1 and Study 2, due to non-progressors being the same patient population.
Critchley-Thorne et al19	2	Non-progressors in this study are the same as non-progressors in Study 2.	Individual data from progressors only for OR, sensitivity, specificity, PPVadj (TSP-9 high/int-risk results and TSP-9 high-risk results), and NPVadj. Note: Individual data from progressors were combined for Study 1 and Study 2 due to non-progressors being the same patient population.
Davison et al20	3	Not applicable.	Individual data for OR, sensitivity, specificity, PPVadj (TSP-9 high/int-risk results and TSP-9 high-risk results), and NPVadj; HR estimate. Note: Data include only 51 progressors who progressed within 5 years.
Frei et al21	4	This study includes 13 progressors who were also evaluated in Study 1.	Individual data for OR, sensitivity, specificity, PPVadj (TSP-9 high/int-risk results and TSP-9 high-risk results), and NPVadj; HR estimate.
Frei et al22	5	This study has the same cohort of patients as Study 6, but progression outcomes were measured at different times and different pathologists provided the diagnoses between studies.	HR estimate.
Khoshiwal et al24	6	This study has the same cohort of patients as Study 5, but progression outcomes were measured at different times and different pathologists provided the diagnoses between studies. In addition, 1 patient was excluded in this study compared with Study 5 due to insufficient histology slides for additional pathology review.	Individual data (TSP-9 high/int-risk classes combined for calculation of performance metrics) for OR, sensitivity, specificity, PPVadj, and NPVadj.

HR indicates hazard ratio; OR, odds ratio; NPV_adj, prevalence-adjusted negative predictive value; PPV_adj, prevalence-adjusted positive predictive value; TSP-9, tissue systems pathology-9.

Quality and Risk of Bias Assessment

Risk of bias assessment was performed independently by 2 authors (CLL and SSB) using the QUAPAS tool to assess risk of bias for 5 domains, including participants, index test, outcome, flow and timing, and analysis. The tool was also used to assess concerns regarding applicability of 4 domains, including participants, index test, outcome, and flow and timing.⁴¹ Discrepancies were resolved through discussion between the 2 authors who performed the risk of bias assessment.

Data Synthesis and Statistical Analysis

Data describing the number of progressors (defined as patients who were subsequently diagnosed with HGD/EAC) and non-progressors (defined as patients who were not diagnosed with HGD/EAC during the surveillance period) who scored high-, intermediate-, or low-risk by the TSP-9 test were extracted. These data were used to calculate the odds ratios (ORs) and 95% CI, sensitivity, specificity, prevalence-adjusted positive (PPV_adj), and negative predictive value (NPV_adj) of the TSP-9 test. Prevalence adjustment was performed as previously described.²⁵ Briefly, the adjustment was performed by determining the number of progressors that would be expected in the population (1.014%/year) to estimate the number of progression events that would have occurred in the sample using the observed sensitivity and specificity, as shown below.

$${\mathit{Progressors}}_{\mathit{adj}}=\mathit{Prevalence}\times n_{\mathit{cases}}$$

$${\mathit{True\; Positive}}_{\mathit{adj}}\times {\mathit{Progressors}}_{\mathit{adj}}\times \mathit{Sensitivity}$$

$${\mathit{True\; Negative}}_{\mathit{adj}}=\mathit{Specificity}\times \left(n_{\mathit{cases}}-{\mathit{Progressors}}_{\mathit{adj}}\right)$$

$${\mathit{False\; Positive}}_{\mathit{adj}}=n_{\mathit{cases}}-{\mathit{Progressor}}_{\mathit{adj}}-{\mathit{True\; Negative}}_{\mathit{adj}}$$

$${\mathit{False\; Negative}}_{\mathit{adj}}=n_{\mathit{cases}}-{\mathit{Progressor}}_{\mathit{adj}}-{\mathit{True\; Positive}}_{\mathit{adj}}$$

$${\mathit{PPV}}_{\mathit{adj}}=\frac{{\mathit{True\; Positive}}_{\mathit{adj}}}{{\mathit{True\; Positive}}_{\mathit{adj}}+{\mathit{False\; Positive}}_{\mathit{adj}}}$$

$${\mathit{NPV}}_{\mathit{adj}}=\frac{{\mathit{True\; Negative}}_{\mathit{adj}}}{{\mathit{True\; Negative}}_{\mathit{adj}}+{\mathit{False\; Negative}}_{\mathit{adj}}}$$

Sensitivity was defined as the percentage of progressors who scored high- or intermediate-risk, and specificity was the percentage of non-progressors who scored low-risk. PPV was defined as the percentage of patients who scored high-risk or high- and intermediate-risk who were progressors, and NPV was the percentage of patients who scored low-risk who were non-progressors. Hazard ratios (HRs) and 95% CI estimating the 5-year risk of progression to HGD/EAC for TSP-9 high-risk versus low-risk classes were also extracted.

Separate meta-analyses, using both common-effect (fixed-effect) and random-effects models, were performed for each measured effect estimate. For OR and HR, the inverse variance method was used, and for sensitivity, specificity, PPV_adj, and NPV_adj, the random intercept logistic regression model was used. Table 2 summarizes data from individual studies used in each meta-analysis. Publication bias was assessed using Egger's regression tests (Supplemental Digital Content 2 http://links.lww.com/JCG/B278). Statistical analyses were conducted using R statistical software (version 4.1.2 and version 4.3.1)⁴² and the R package meta (version 6.5-0).⁴³

TABLE 2.

Patient Cohort Characteristics for Studies Included in the Meta-analysis

			Pathologist diagnosis, n (%)				Segment length category, n (%) or cm, median (IQR)
First author (study number)	Study design	No. patients	Progressors	Non-progressors	Sex—male, n (%)	Age (y)	Progressors	Non-progressors
Critchley-Thorne et al18 (1)†	Nested case-control	Progressors: 38 Non-progressors: 145	NDBE=31 (81.6) IND=2 (5.3) LGD=5 (13.2)	NDBE=138 (95.2) IND=2 (1.4) LGD=5 (3.4)	Progressors: 33 (86.8) Non-progressors: 114 (78.6)	Progressors: Mean: 60.1 +/- SD: 11.3 Non-progressors: Mean: 61.0 +/- SD: 12.1	Short=10 (26.3) Long=27 (71.1)Unknown=1 (2.6)	Short=58 (40.0) Long=73 (50.3) Unknown=14 (9.7)
Critchley-Thorne et al19 (2)†	Nested case-control	Progressors: 30 Non-progressors: 145	NDBE=13 (43.3) IND=1 (3.3) LGD=16 (53.3)	NDBE=138 (95.2) IND=2 (1.4) LGD=5 (3.4)	Progressors: 28 (93.3) Non-progressors: 114 (78.6)	Progressors: Mean: 61.8 +/- SD: 9.5 Non-progressors: Mean: 61.0 +/- SD: 12.1	Short=9 (30.0) Long=19 (63.3)Unknown=2 (6.7)	Short=58 (40.0) Long=73 (50.3) Unknown=14 (9.7)
Davison et al20 (3)	Nested case-control	Progressors: 58 Non-progressors: 210	NDBE=43 (74.1) IND=5 (8.6) LGD=10 (17.2)	NDBE=184 (87.6) IND=18 (8.6) LGD=8 (3.8)	Progressors: 52 (89.7) Non-progressors: 152 (72.4)	Progressors: Median: 63.8 IQR: 57-69.1 Non-progressors: Median: 60.8 IQR: 52.6-68	Short=17 (29.3) Long=38 (65.5)Unknown=3 (5.2)	Short=92 (43.8) Long=113 (53.8)Unknown=5 (2.4)
Frei et al21 (4)‡	Nested case-control	Progressors: 38 Non-progressors: 38	NDBE=38 (100) IND=0 (0) LGD=0 (0)	NDBE=38 (100) IND=0 (0) LGD=0 (0)	Progressors: 31 (82) Non-progressors: 27 (71)	Progressors: Median: 64 IQR: 57-71 Non-progressors: Median: 62 IQR: 53-71	Median: 7 IQR: 5-8	Median: 6 IQR: 5-8
Frei et al22 (5)§	Retrospective cohort study	Progressors: 34 Non-progressors: 121	NDBE=0 (0) IND=0 (0) LGD*,§=34 (100)	NDBE=0 (0) IND=0 (0) LGD*,§=121 (100)	Progressors: 29 (85) Non-progressors: 94 (78)	Progressors: Mean: 64 +/- SD: 9 Non-progressors: Mean: 61 +/- SD: 11	Median: 5 IQR: 3-7	Median: 4 IQR: 3-5
Khoshiwal et al24 (6)§	Retrospective cohort study	Progressors: 24 Non-progressors: 130	NDBE=9 (36.8) IND/LGD=15 (63.2)	NDBE=96 (73.5) IND/LGD=34 (26.5)	Progressors: 20 (83.3) Non-progressors: 102 (78.5)	Progressors: Mean: 63.5 +/- SD: 9.5 Non-progressors: Mean: 61.0 +/- SD: 10.4	Median: 5.0 IQR: 3.0-6.3	Median: 4.0 IQR: 3.0-5.5
Totals∥		Progressors: 185 Non-progressors: 514 Total patients: 699	NDBE = 134 (71.3) IND/LGD = 54 (28.7)	NDBE=456 (87.2) IND/LGD=67 (12.8)
			Progressors and non-progressors: NDBE = 590 (83.0)​ IND/LGD = 121 (17.0)

^*Community-based diagnosis.

^†Study 1 and 2 utilized the same set of non-progressor patients as described in the Critchley-Thorne et al, 2017¹⁹ study publication.

^‡Study 4 included evaluation of additional spatial and temporal biopsies from 13 progressors that were also evaluated in study 1.

^§Study 5 and 6 evaluated the same cohort of patients (except for 1 patient), but progression outcomes were determined at different time points and diagnoses were provided by different pathologists.

^∥ Total number of progressors/non-progressors and total patients includes only unique patients. Totals for NDBE and IND/LGD include all patients.

NDBE indicates non-dysplastic Barrett’s esophagus; IND, indefinite for dysplasia; LGD, low-grade dysplasia; IQR, interquartile range; SD, standard deviation.

RESULTS

Search Strategy Yield and Quality Assessment

Literature search results are summarized in Figure 1. The initial search resulted in 30 articles, 1 of which was removed as a duplicate. After reviewing titles and abstracts, 23 of 29 articles that did not meet inclusion criteria were excluded. Six articles met inclusion criteria for full-text review for eligibility and were included in the meta-analysis.

Study and Patient Cohort Characteristics

Characteristics of the 6 studies and corresponding patient cohorts included in the meta-analysis are shown in Table 2, including study design, number of patients who progressed (progressors) and those who did not progress (non-progressors) to HGD/EAC, pathology diagnosis (NDBE, IND, or LGD), sex, and age. All studies were retrospective and 4 were case-control studies.^18–22,24 Patient overlap existed between all non-progressors in the study by Critchley-Thorne et al¹⁸ (Study 1) and the study by Critchley-Thorne et al¹⁹ (Study 2) and between 13 progressors in the study by Frei et al²¹ (Study 4) and the study by Critchley-Thorne et al¹⁸ (Study 1) as described in the evaluated publications (Table 1 and Table 2). Overall, this review synthesized data on a total of 699 unique patients (n=185 progressors, n=514 non-progressors). The median cohort size from data reported in the evaluated in publications with unique patient populations was 175 patients (range: 76 to 268). Mean ages ± SD were reported for 4 studies and ranged from 60.1 to 64 years for progressors and mean age was 61 years for non-progressors. Median ages were reported for 2 studies and ranged from 63.8 to 64 years for progressors and 60.8 to 62 years for non-progressors. The majority of patients included in all studies were male (>70%) and had pathology diagnoses of NDBE (83%) (Table 2).

Meta-analysis of TSP-9 Performance in All Included Studies

Five studies that met inclusion criteria reported results for the number of progressors and non-progressors with TSP-9 results for each risk class that were used in the meta-analysis of OR, sensitivity, specificity, PPV_adj, and NPV_adj. Data from the Critchley-Thorne et al¹⁸ and Critchley-Thorne et al19 studies were combined, as these studies shared the same nonprogressor patient population, and the progressor patient population was separated based on incident progression (diagnosis of HGD/EAC at least 1 year later) and prevalent HGD/EAC (subsequent diagnosis of HGD/EAC in less than 1 year). The mean effect estimates of the TSP-9 test performance for predicting risk of progression to HGD/EAC based on combined high/intermediate-risk versus low-risk class results across these studies showed an OR of 6.52, P<0.0001 (95% CI: 4.40-9.66) and 6.58 P<0.0001 (95% CI: 3.95-10.96) for the common-effect and random-effects model, respectively (Fig. 2A). Mean effect estimates for sensitivity and specificity were 61% (95% CI: 54%-68%) and 81% (95% CI: 78%-84%), respectively, for the common-effect model, and 62% (95% CI: 48%-75%) and 81% (95% CI: 77%-85%), respectively, for the random-effects model for the studies analyzed (Table 3, Supplemental Digital Content 3A-B http://links.lww.com/JCG/B279). The NPV_adj and PPV_adj were 97% (95% CI: 96%-98%) and 14% (95% CI: 9%-21%), respectively, for both models when TSP-9 low-risk class results and TSP-9 high/intermediate-risk class results were evaluated (Table 3, Supplemental Digital Content 3C-D http://links.lww.com/JCG/B279). When TSP-9 high-risk class results were analyzed, the PPV_adj was 28% (95% CI: 17%-42%) for both models (Table 3, Supplemental Digital Content 3E http://links.lww.com/JCG/B279). Mean effect estimates from common-effect and random-effects models were similar, as there was minimal heterogeneity between studies for all analyses except sensitivity, which had significant heterogeneity (OR: l²=33%, τ²=0.1019, P=0.21; sensitivity: l²=79%, τ²=0.2515, P < 0.01; specificity: l²=21%, τ²=0.0065, P=0.29; NPV_adj: l²=0%, τ²=0, P=0.69; PPV_adj (high/intermediate-risk vs. low-risk): l²=0%, τ²=0, P=0.97; and PPV_adj (high-risk vs. low-risk): l²=0%, τ²=0, P=0.41; Fig. 2, Supplemental Digital Content 3 http://links.lww.com/JCG/B279). No evidence of funnel plot asymmetry was detected, further demonstrating minimal heterogeneity in the evaluated performance metrics between studies (Supplemental Digital Content 3 http://links.lww.com/JCG/B279).

FIGURE 2.

Mean effect estimates of TSP-9 predictive performance metrics. A, Forest plot representing results from the meta-analysis of calculated odds ratios (ORs) for progression to HGD/EAC based on combined TSP-9 high-/intermediate-risk class results. ¹For the calculation of ORs, TSP-9 results were evaluated as high- or intermediate-risk versus low-risk. B, Forest plot representing results from the meta-analysis of hazard ratios (HRs) for progression to HGD/EAC in 5 years based on TSP-9 high-risk versus low-risk class results. ²HRs used in this meta-analysis compared TSP-9 high-risk versus low-risk results. Squares in graphical display represent the study weight from the common-effect model.

TABLE 3.

Summary of Mean Effect Estimates From Meta-analysis for Sensitivity, Specificity, PPV_adj, and NPV_adj for the TSP-9 Test [mean, (95% CI)]

	Sensitivity*	Specificity*	NPVadj (TSP-9 high/int risk)*	PPVadj (TSP-9 high/int risk)*	PPVadj (TSP-9 high risk)†
Common-effect model	0.61 (0.54; 0.68)	0.81 (0.78; 0.84)	0.97 (0.96; 0.98)	0.14 (0.09; 0.21)	0.28 (0.17; 0.42)
Random-effects model	0.62 (0.48;0.75)	0.81 (0.77; 0.85)	0.97 (0.96; 0.98)	0.14 (0.09; 0.21)	0.28 (0.17; 0.42)

^*For the calculation of sensitivity and specificity, NPV_adj, and PPV_adj, TSP-9 high- and intermediate-risk (high/int) results were combined.

^†The PPV_adj was also calculated using the TSP-9 high-risk result. For this analysis, the Khoshiwal et al, 2023²⁴ study was not included because the number of only high-risk results was not reported.

PPV_adj indicates prevalence-adjusted positive predictive value; NPV_adj, prevalence-adjusted negative predictive value; TSP-9, tissue systems pathology-9; CI, confidence interval.

Four studies that met inclusion criteria, which included results for 669 of the 699 patients, reported HRs for risk of progression in 5 years for high- versus low-risk TSP-9 results. Mean effect estimates of the TSP-9 test performance across studies showed an HR of 6.66, P<0.0001 (95% CI: 4.59-9.66) and 6.66, P<0.0001 (95% CI: 4.58-9.67) for the common-effect and random-effects model, respectively. There was no significant heterogeneity between studies (l²=0%, τ²=0.0006, P=0.48) (Fig. 2B), and no evidence of funnel plot asymmetry was detected (Supplemental Digital Content 3 http://links.lww.com/JCG/B279).

Subgroup Analysis of TSP-9 Performance in Patients With NDBE

Two of the studies that met the inclusion criteria for this review provided the necessary patient data in the original publication to perform a subgroup analysis of the TSP-9 performance metrics for patients with a pathology diagnosis of NDBE. These included the independent validation study of TSP-9 by Frei et al,21 which reported an HR of 5.9 (95% CI: 2.4-14.4) for patients who scored TSP-9 high- versus low-risk, and another independent validation study by Davison et al,20 which reported an HR of 5.1 (95% CI: 2.1-12.3) for a subgroup of patients with an expert pathologist diagnosis of NDBE. Both performance metrics are similar to the mean effect estimates determined through meta-analysis of all the included studies, indicating similar performance of TSP-9 in patients with NDBE when compared with the full intended use population of patients with NDBE, IND, or LGD (Fig. 2B).

Risk of Bias Assessment

As summarized in Figure 3, risk of bias assessment using the QUAPAS tool41 resulted in a consensus of low risk of bias when considering appropriate use of the index test (TSP-9), study flow and timing, and study analysis for all 6 studies. Risk of bias was determined to be high for participant selection for 4 (case-control) of the 6 studies18–21 (66.7%); however, these studies and this meta-analysis addressed potential participant bias by adjusting for prevalence for PPV and NPV values using the prevalence of HGD/EAC progression in the BE population. An unclear risk of bias was determined for study outcome descriptions for 3 (50%) of the studies.^18–20 All 6 studies were determined to have a low risk of bias for applicability concerns across the 4 QUAPAS domains assessed: participants, index test, outcome, and flow and timing.

FIGURE 3.

QUAPAS risk of bias assessment. A, Proportion of studies with high, low, or unclear risk of bias for each domain evaluated. B, Proportion of studies with high, low, or unclear concerns regarding applicability for each domain evaluated.

DISCUSSION

This systematic review and meta-analysis evaluated the performance of the TSP-9 test in stratifying patients with BE for risk of progression to HGD/EAC within 5 years. TSP-9 high- and combined high/intermediate-risk classes were associated with progression to HGD/EAC within 5 years. Analysis of ORs demonstrated that BE patients who scored TSP-9 high/intermediate-risk were at 6.52 times (95% CI: 4.40-9.66, P<0.0001) higher odds for progression to HGD/EAC than patients who scored low-risk across the evaluated studies. Common-effect and random-effects model estimates yielded similar results, as minimal heterogeneity was measured for key performance metrics, including ORs, HRs, specificity, and positive and negative predictive values for the TSP-9 test, indicating consistent performance of the test across evaluated studies, including in patients with NDBE. A subgroup analysis of individual studies that only included NDBE patients (Study 4)21 or included a subgroup analysis of NDBE patients (Study 3),20 reported HRs (HR 5.9, 95% CI: 2.4-14.4, and HR 5.1, 95% CI: 2.1-12.3, respectively) that were similar to the mean effect estimate reported across all studies included in this meta-analysis, indicating similar performance of TSP-9 in patients with NDBE when compared with all included BE patients. Taken together, these results demonstrate that the TSP-9 test can provide clinically impactful risk stratification that may increase detection of progressors, including at the early, non-dysplastic stage of BE, where management escalation strategies such as EET or short-interval surveillance can be effective in reducing the incidence and mortality of EAC.

The mean effect estimates for accuracy metrics for TSP-9 in predicting risk of progression to HGD/EAC in patients with NDBE, IND, and LGD reported in this meta-analysis were similar to those reported in a recently published pooled analysis by Davison et al25 that analyzed individual participant data from published clinical validity studies on the TSP-9 test metrics. The consistency between the key predictive performance metrics in the 2 independent analyses further supports the predictive power of the TSP-9 test. The weighting and measurement of inter-study heterogeneity incorporated in the meta-analysis approach adds additional strength to the mean effect estimates of the TSP-9 performance metrics.

The mean effect estimate of the TSP-9 test reported in this meta-analysis (OR 6.58, 95% CI: 3.95-10.96) exceeds the OR reported in a recent meta-analysis by Krishnamoorthi and colleagues of the predictive performance of a pathology diagnosis of LGD (OR 4.25, 95% CI: 2.58-7.00, I²=87%). However, the mean effect estimates of ORs reported by Krishnamoorthi et al⁴⁴ were determined by combining estimates of relative risk, hazard, and odds ratios from univariate or multivariate analyses performed in the included individual studies using the random-effects model, whereas univariate analysis of ORs for TSP-9 was performed using both common and random-effects models in this analysis. The meta-analysis by Krishnamoorthi et al⁴⁴ also analyzed the performance of several other variables, including age, male sex, smoking, alcohol, body mass index, segment length, and use of medications for predicting progression to HGD/EAC with ORs ranging from 0.48 to 2.16. In addition, the study by Frei et al²¹ (Study 4) reported that the TSP-9 high-risk results in the NDBE patient population were associated with a prevalence-adjusted progression rate of 6.9%/year. This rate exceeds the progression rate reported in another systematic review and meta-analysis for patients with pathology diagnosis of LGD (1.73%/year, 95% CI: 0.99-2.47),36 which is a clinically actionable progression per current guidelines for escalation of care to EET to prevent EAC or short-interval surveillance to detect dysplasia and EAC at the earliest possible stages.1,11,12,15 These data indicate that the TSP-9 test provides more accurate predictive information than the traditional clinical and pathology factors that have been associated with progression risk in patients with BE, resulting in clinically impactful results to better identify which patients are at risk for progression. Altogether, these findings further support the best practice advice from the American Gastroenterological Association that the tissue systems pathology test (i.e., the TSP-9 test) may be utilized for risk stratification of patients with NDBE.30

The mean predictive performance of TSP-9 that was demonstrated in the common effects model of this meta-analysis (OR 6.52, 95% CI: 4.40-9.66) was higher than the mean performance of p53 immunohistochemical staining (IHC) reported in a recent systematic review and meta-analysis (OR 3.84, 95% CI: 2.79-5.27),45 indicating that TSP-9 may perform better than p53 IHC in risk stratifying patients with BE. Furthermore, the meta-analysis of p53 IHC identified significant heterogeneity in ORs between studies (Q(7)=27.71, P<0.001), while significant heterogeneity was not detected in the meta-analysis of TSP-9 ORs (l²=33%, τ²=0.1019, P=0.21). The heterogeneity observed between p53 IHC studies may be due to variability in IHC staining reagents and protocols between different laboratories and/or variability in the manual scoring systems used, as well as inter-observer variability in the scoring of p53 IHC. p53 is one of the 9 protein-based biomarkers quantified by the TSP-9 test.17,18 In contrast to traditional IHC methods used to stain and qualitatively score p53 in many clinical laboratories, the TSP-9 test uses a quantitative, immunofluorescence-based digital pathology platform to quantify p53 automatically and objectively in a spatial context and in combination with additional biomarkers that have predictive value for neoplastic progression. The additional biomarkers evaluate abnormalities in epithelial cells such as p16, HER2, and AMACR, and also stromal factors that can contribute to malignant progression such as COX2, HIF-1ɑ, CD68, and CD45RO.17,18,46 The TSP-9 test also evaluates multiple quantitative measurements of nuclear morphology,17,18 and while nuclear morphology can be assessed qualitatively in histologic slides, it is limited by observer variability in the manual interpretation.16 An additional advantage of the TSP-9 approach is that samples are processed in a centralized laboratory,17 which reduces the impact of variability in immunostaining reagents and methods that have been associated with traditional IHC methods for biomarkers such as p53.1,47–49 Other biomarkers and testing methodologies have been investigated for risk stratification of patients with BE, including microRNA50- and methylation-based biomarkers.51–53 However, these biomarker-based approaches are still at early research and development stages and have not been validated for clinical use in BE patients or have limited evidence supporting their clinical validity and are undergoing further investigation, respectively. The limited available evidence for these biomarker-based approaches prevents comparison to the performance of the TSP-9 test in this review.

The results of this systematic review and meta-analysis indicate that the TSP-9 test can provide risk stratification for patients with NDBE. The overall rate of progression in patients with NDBE is low (0.63% per year);33 however, since these patients make up approximately 90% of the BE patients in endoscopic surveillance,31,32 this population likely harbors approximately half of the patients who develop HGD/EAC each year. The TSP-9 test may bring significant value to this population of patients by identifying high-risk patients who may benefit from escalation of care, including use of EET that is 92% to 99% effective in eradicating BE and preventing disease progression,54,55 as well as short-interval surveillance that can be effective in detecting progression at earlier stages.56 Both clinical escalation strategies have demonstrated effectiveness in reducing the incidence and mortality of EAC in patients with BE. Furthermore, the TSP-9 test identifies low-risk patients who may safely avoid overuse of endoscopy and EET, which can improve quality of life for patients and enable more efficient use of health care resources.

Strengths of the current systematic review and meta-analysis include the consistency with which the TSP-9 test was performed across studies due to the test utilizing automated analysis with locked parameters as described in the evaluated published studies. The minimal inter-study heterogeneity indicating consistent performance of the TSP-9 test between studies and the overall quality of the included studies increase the quality of evidence supporting the clinical use of this test, as both quality and consistency are components of the National Comprehensive Cancer Network (NCCN) Evidence Blocks™ that summarize the scientific rationale underlying guideline recommendations.57 Results of systematic reviews and meta-analyses are generally considered higher quality evidence than individual studies and are emphasized when making determinations regarding the level of evidence supporting the value of clinical tests.58,59 An additional strength of this study is the similarity between the percentage of NDBE patients included in this analysis (>80%) and NDBE patients in the general BE surveillance population (∼90%),31,32 indicating generalizability of the findings. Overall, risk of bias was low for TSP-9 testing, study flow and timing, study analysis, and applicability for all studies. Some unclear study outcome descriptions in 3 of the studies may have introduced bias.18–20 Also, the inherent nature of retrospective, case-control studies (3 of the 4 independent patient cohorts studied)18–21 has the potential to introduce bias due to the necessity of including an overrepresentation of progressors compared to what would be observed clinically. However, this was addressed by adjusting NPV and PPV to the prevalence rates of HGD/EAC observed in the general BE patient population. Given the low progression rate in the BE patient population with a pathology diagnosis of NDBE,33 cohort studies with consecutive patient enrollment or prospective randomized controlled trials are not logistically feasible to assess the performance of risk stratification strategies.

In conclusion, this systematic review and meta-analysis study demonstrated mean effect estimates for ORs and HRs that were consistent with previously published performance metrics for the TSP-9 test in individual studies, with minimal heterogeneity across studies from both common-effect and random-effects models. The analysis showed clinically actionable sensitivity, specificity, NPV_adj, and PPV_adj, further demonstrating the clinical validity of the TSP-9 test in identifying patients at high risk of progression to HGD/EAC at early, treatable stages of the disease, as well as low-risk patients who can avoid overuse of procedures. These findings further support medical society best practices for risk stratification of BE patients with the TSP-9 test,1,30 particularly in patients with NDBE, who comprise most of the patients in surveillance for BE. This study contributes an important additional level of evidence indicating that TSP-9 can improve risk stratification in patients with BE, enabling risk-aligned management to improve patient health outcomes.