Clinical and endoscopic factors predict higher pathologic grades of Barrett dysplasia




Barrett esophagus is highly prevalent in the Western world; however, only a minority of affected individuals progress to esophageal adenocarcinoma. Whereas many studies have examined risk factors for development of Barrett metaplasia, few data are available on risk factors for progression to neoplasia. Identifying simple, reliable, clinical, and endoscopic predictors of high-grade dysplasia and adenocarcinoma would be helpful for risk stratification in screening and surveillance programs.


Clinical, endoscopic, and histologic data were reviewed for patients with a new Barrett diagnosis between 2002 and 2005. Patients were classified, by an expert gastrointestinal pathologist, as having intestinal metaplasia, indefinite-for-dysplasia, low-grade dysplasia, high-grade dysplasia, and esophageal adenocarcinoma. Gender, age, race, ethnicity, hiatal hernia presence and size, Barrett segment length, H. pylori status, alcohol, smoking, proton pump inhibitor (PPI) use and duration, and reflux symptom duration were evaluated by logistic regression analysis for their association with dysplasia severity.


In all, 109 patients (26 women, 83 men, mean age: 58.8) were newly diagnosed with Barrett metaplasia (n = 39), indefinite/low-grade dysplasia (n = 35), and high-grade dysplasia/esophageal adenocarcinoma (n = 35) over a 3-year period. On logistic regression analysis, duration of reflux symptoms for ≥20 years (odds ratio [OR]: 5.66, P = .012), longer Barrett segment length (OR for 3–6 cm vs. <3 cm: 9.05, P < .0001; OR for ≥6 cm: 8.374, P < .0001), hernia size ≥4 cm (OR: 10.63, P = .014), and male gender (OR: 4.03, P = .0019) were associated with higher pathologic grade. Duration of reflux symptoms and Barrett length were significant as both discrete and continuous variables. Absence of H. pylori (OR: 2.731, P = .060) approached significance in predicting dysplasia severity. In bivariate models, gender and Barrett length (continuous form) were significantly associated with grade when considered together (OR: 2.52, P = .0490 and OR: 1.30, P < .0001), as were gender and hernia size >4 cm (OR: 4.64, P = .0049 and OR: 12.18, P = .0197).


Male gender, longstanding gastroesophageal reflux disease, hiatal hernia size, and segment length are strongly associated with higher grades of dysplasia at index diagnosis. These factors along with H. pylori status warrant further prospective evaluation as predictors of risk for development of high-grade dysplasia and esophageal adenocarcinoma. Cancer 2007. © 2007 American Cancer Society.

Barrett esophagus (BE) has been shown to occur in 10% to 15% of patients with gastroesophageal reflux disease (GERD).1 Endoscopic surveillance has been recommended for patients with BE because of its association with esophageal adenocarcinoma, a cancer that has rapidly risen in incidence over the last 3 decades.2–4 Despite this, the majority of patients with Barrett metaplasia and, in fact, low-grade dysplasia will never progress to high-grade dysplasia or esophageal adenocarcinoma.5–7 Indeed, most patients with BE die from unrelated diseases.

Although the data remain insufficiently understood, several risk factors have been suggested to be associated with Barrett metaplasia. Smoking and alcohol consumption have been evaluated in several studies, but consistent associations have not been found.8, 9 Whereas some studies have shown that the presence and duration of reflux symptoms increases the risk of BE,8, 10 other studies have shown that <60% of patients with Barrett adenocarcinoma report frequent reflux symptoms.11, 12 Recent studies suggest that Barrett segment length5, 13 and male gender14 may be predictive of higher pathologic grades. Given the high population prevalence of BE15 and its relatively low risk of progression, the identification of simple, reliable, clinical, and endoscopic predictors of risk for development of high-grade dysplasia and esophageal adenocarcinoma would be of great value. A recent multicenter study revealed that the majority of dysplastic lesions are detected during the initial endoscopy with biopsy.7 Thus, initial risk stratification is critical in better identifying prevalent cases of dysplasia. Such risk stratification would allow for the selection of patients for more intensive endoscopic surveillance and promote more efficient utilization of healthcare services and resources.

The goal of this study was to identify clinical, endoscopic, and histologic factors that are associated with an index diagnosis of high-grade dysplasia or adenocarcinoma in subjects with a new diagnosis of BE.


Between January, 2002, and September, 2005, 160 patients with a new diagnosis of BE were evaluated at the M. D. Anderson Cancer Center. The majority of these patient were referrals to a cancer prevention screening program in BE. Patients were identified through an endoscopic database and their clinical and pathologic information obtained from a detailed, institutional registry. Data, including patient demographics, endoscopic findings, histology reports, and social and medical history were retrospectively reviewed.

Criteria for the diagnosis of BE were endoscopic identification of the squamocolumnar junction proximal to the gastroesophageal junction and targeted biopsies with pathology revealing columnar epithelium with goblet cells. For inclusion in the analysis, pathologic diagnosis and grade were determined by 2 pathologists, 1 of whom was an experienced gastrointestinal pathologist at the M. D. Anderson Cancer Center. For patients with a diagnosis of Barrett high-grade dysplasia, a second, confirmatory reading of high grade was required for inclusion. Thus, of the initial 160 patients seen at the M. D. Anderson Cancer, 109 met the pathologic criteria for inclusion in the study analysis.

Patients were classified into 3 groups: 1) Barrett metaplasia, 2) indefinite for dysplasia/low-grade dysplasia (LGD), or 3) high-grade dysplasia (HGD)/esophageal adenocarcinoma (EAC). Only index cases of BE were included in the analysis. Patients with a preexisting diagnosis of BE, esophageal or gastric surgery, or treatment to the esophagus for other esophageal diseases/malignancies (eg, squamous cell cancer) were excluded.

Factors that may be associated with an increased risk of Barrett dysplasia or adenocarcinoma were examined, including: age, gender, ethnicity, smoking, alcohol use, Barrett segment length, hiatal hernia presence and size, Helicobacter pylori status, presence and duration of GERD symptoms, and proton pump inhibitor (PPI) use and duration. Helicobacter pylori colonization was determined based on gastric biopsy specimens stained with either hematoxylin and eosin, Giemsa, or Diff-Quik stains. Patients were defined as smokers if they were currently smoking or had smoked at least 1 cigarette, cigar, or pipe/day for a minimum of 1 year in the past. Alcohol consumption was characterized by no use, moderate use (1–6 drinks/week), and heavy use (≥7 drinks/week).

Statistical Analysis

All data analyses were performed using SAS (Cary, NC). Demographic, endoscopic, and histologic factors including gender, race, age, Barrett segment length, reflux symptoms, hiatal hernia presence and size, H. pylori status, PPI use and duration, alcohol intake, and cigarette use were examined for potential association with pathologic grade using logistic regression models. The potential risk factors were first examined separately. A P-value of .05 was considered significant. Independent variables that had a P-value of .05 or less in the univariate analyses were evaluated for the multiple logistic regression models. Comparison of mean segment length and GERD duration between the 3 pathologic groups was carried out using nonparametric analysis of variance (Kruskal–Wallis test). Comparison between paired samples was performed by the Wilcoxon paired samples test.


The study population consisted of a total of 109 patients with a new diagnosis of BE seen at the M. D. Anderson Cancer Center between January, 2002, and September, 2005. Demographic, clinical, endoscopic, and histologic characteristics of these patients are given in Table 1. Mean age at diagnosis was 58.82 (standard deviation [SD], 10.1; range, 25–85). Of the patients, 83 (76%) were men and 26 were women; 94 (87%) were white, with the remainder being African-American (n = 4), Asian-American (n = 2), and Hispanic (n = 9).

Table 1. Demographic, Endoscopic, and Histologic Characteristics of Barrett Patients at Index Diagnosis
 No.Intestinal metaplasiaIndefinite/LGDHGD/EAC
No. (%)No. (%)No. (%)
  • LGD indicates low-grade dysplasia; HGD, high-grade dysplasia; EAC, esophageal adenocarcinoma; PPI, proton pump inhibitor; GERD, gastroesophageal reflux disease.

  • *

    Results expressed in mean± 1 standard deviation.

No. of patients10939 (35.8)35 (32.1)35 (32.1)
 Women2615 (58)9 (34)2 (8)
 Men8324 (29)26 (31)33 (40)
Age*10959.46 ± 11.8857.77 ± 10.3259.23 ± 10.07
 Non-white156 (40)2 (13)7 (47)
 White9433 (35)33 (35)28 (30)
Alcohol intake
 None3816 (42)9 (24)13 (34)
 1–6 drinks/wk206 (30)6 (30)8 (40)
 ≥7 drinks/wk4816 (33)18 (38)14 (29)
 No3514 (40)9 (26)12 (34)
 Yes7024 (34)24 (34)22 (32)
PPI use
 No6827 (40)21 (31)20 (29)
 Yes4012 (30)14 (35)14 (35)
PPI duration
 ≤5 y144 (28)5 (36)5 (36)
 ≥6 y194 (21)5 (26)10 (53)
GERD symptoms
 No2611 (42)11 (42)4 (16)
 Yes8027 (34)24 (30)29 (36)
GERD duration
 >0–≤5 yrs208 (40)10 (50)2 (10)
 6–19 y246 (25)8 (33)10 (42)
 ≥20 y123 (25)2 (17)7 (58)
Hiatal hernia
 Present7728 (36)26 (34)23 (30)
  1–3 cm5825 (43)18 (31)15 (26)
  ≥4 cm193 (16)8 (42)8 (42)
Barrett length
 < 3 cm5733 (58)17 (30)7 (12)
 3–5 cm174 (23)3 (18)10 (59)
 ≥6 cm322 (6)15 (47)15 (47)
H. pylori present
 No6220 (32)21 (34)21 (34)
 Yes1610 (63)2 (12)4 (25)

Smoking and alcohol histories were obtained from 105 and 106 patients, respectively. Twenty (19%) patients reported consuming 1–6 drinks/week and 48 (46%) reported consuming ≥7 drinks/week. Seventy (67%) patients were current or former smokers. Eighty (76%) patients reported a past or present history of reflux symptoms, either chronic or intermittent. Only 40 (37%), however, were on PPI medication before their Barrett diagnosis. The remainder were either on H2 blockers, antacids, or were not taking medications for their symptoms. Among the 80 individuals with a past or present history of reflux symptoms, 12 (22%) reported a ≥20-year history of reflux. Among subjects with a ≥20-year history of reflux symptoms, 58% presented with either high-grade dysplasia or esophageal adenocarcinoma.

Evaluation of endoscopic records revealed that 71% (n = 77) of study subjects possessed a hiatal hernia. Among these, 25% (n = 19) had a hernia ≥4 cm in size. Fifty-four percent (n = 57) of subjects possessed a short segment (SS) BE (<3 cm). The remaining 46% had a long (≥3 cm) segment of Barrett with 30% (n = 32) possessing a segment length of ≥6 cm. Among subjects with a segment length ≥6 cm, the vast majority (94%) had some degree of dysplasia on initial diagnosis, with 47% presenting with either HGD or adenocarcinoma. This is compared with the 57 subjects with SSBE, where prevalence of HGD/EAC was 12.2% (n = 7). In these 7 individuals with SSBE, 5 possessed HGD alone. Thus, in our population the prevalence of EAC in subjects with SSBE was 3.5%.

A univariate analysis of potential risk factors for the development of HGD or EAC was conducted and the results are summarized in Table 2. Among discrete variables, male gender (odds ratio [OR]: 4.028, P = .0019) and hiatal hernia size ≥4 cm (OR: 10.63, P = .014) were statistically significant predictors of HGD/EAC. GERD duration was categorized as ‘≤5 years,’ ‘6–19 years,’ and ‘≥20 years.’ A longer duration (>5 years) of reflux symptoms was associated with a higher pathologic grade, with the strongest association occurring in subjects with ≥20 years of reflux symptoms (OR: 5.664, P = .012). Longer Barrett segments were also associated with an increased risk of HGD/EAC (OR: 8.374, P < .0001, for segment length ≥6 cm). Absence of H. pylori (OR: 2.731, P = .060) the presence of a hiatal hernia (OR: 5.168, P = .061), and the presence of reflux symptoms (OR: 2.27, P = .072) approached statistical significance in predicting higher pathologic grades. All other factors, including PPI use and duration, age, and smoking and alcohol history did not achieve significance. Body mass index (BMI) was also assessed as a continuous variable but did not achieve significance (OR: 1.042, P = .266).

Table 2. Univariate Analysis of Potential Risk Factors for Predicting Higher Pathologic Grades of Barrett Esophagus at Index Diagnosis
VariableControlOdds ratio (95% CI)PN
  • GERD indicates gastroesophageal reflux disease; PPI, proton pump inhibitor.

  • *

    Continuous variables.

 MenWomen4.028 (1.68, 9.69).0019109
Hiatal hernia presentNo hernia5.168 (0.92, 28.74).06197
Hiatal hernia size
 ≤3 cmNo hernia3.933 (0.69, 22.30).12284
 ≥4 cmNo hernia10.630 (1.61, 70.16).014 
Barrett length
 3–5 cm< 3 cm9.050 (2.99, 27.34)<.0001106
 ≥6 cm< 3 cm8.374 (3.39, 20.67)<.0001 
Barrett length* 1.333 (1.181, 1.503)<.0001106
GERD symptoms 2.270 (0.93, 5.54).072 
 YesNo symptoms  81
GERD duration  .894 
 ≤5 yNo symptoms1.078 (0.36, 3.26).03081
 6–19 yNo symptoms3.293 (1.13, 9.63).012 
 ≥20 yNo symptoms5.664 (1.47, 21.81)  
GERD duration* 1.05 (1.012, 1.18).00881
H. pylori  .060 
 AbsentPresent2.731 (0.94, 7.98) 78
Alcohol intake 1.507 (0.56, 4.09).421 
 1–6 drinks/wkNone1.103 (0.50, 2.41).807106
 ≥7 drinks/wkNone   
Race 1.336 (0.49, 3.65).572 
 WhiteOther  109
Smoking 1.073 (0.51, 2.26).853 
 YesNo  105
PPI use 1.410 (0.69, 2.90).350 
 YesNo  33
PPI duration 1.798 (0.49, 6.56).375 
 ≥6yrs< 6 years  33
Age* 0.998 (0.97,1.03).897109

A bivariate logistic regression was carried out with pairs of factors that were significant in the univariate analysis. Male gender and larger hiatal hernia size (OR: 4.64, P = .0049 and OR: 12.18, P = .0197), and male gender and longer segment length (OR: 2.52, P = .0490 and OR: 1.30, P < .0001) when considered together were each independently associated with high grades of dysplasia.

As Table 2 reveals, Barrett length (OR: 1.333, P < .0001) and GERD symptom duration (OR: 1.05, P = .008) were also significant when analyzed as continuous variables. The association between Barrett segment length and pathologic grade is shown in Figure 1. There is a stepwise increase in mean segment length with advancing pathologic grade. Indeed, HGD/EAC was associated with a mean length of 6.75 cm compared with 1.99 cm for intestinal metaplasia alone. A similar trend was noted with GERD symptom duration (Fig. 2), with HGD/EAC subjects reporting mean symptom duration nearly 10 years longer than those with intestinal metaplasia or LGD (P = .0027 by Wilcoxon paired samples test). When subjects with indefinite/LGD were compared with subjects with intestinal metaplasia alone, there was no significant difference in GERD symptom duration (P = .84, Wilcoxon paired samples test).

Figure 1.

Mean Barrett segment length for each pathologic grade group. Results expressed as mean and SE. P < .0001 by Kruskal–Wallis test.

Figure 2.

Mean duration of gastroesophageal reflux disease (GERD) symptoms for each pathologic grade group. Results expressed as mean and SE. P = .0025 by Kruskal–Wallis test.


BE is highly prevalent in the US population, yet the majority of individuals affected never progress to dysplasia. In this retrospective study we sought to identify which clinical, endoscopic, and histologic factors were associated with an index diagnosis of high-grade dysplasia or adenocarcinoma. We chose to evaluate patients with a new diagnosis of Barrett both with and without dysplasia. Because all of the individuals included were newly diagnosed, behavioral changes which may be initiated by a new Barrett diagnosis (eg, decreased alcohol intake, PPI use, etc) should not have played a major role. Our study is 1 of a few studies with a relatively large population of women with BE. The majority of other studies evaluating such risk factors have almost exclusively evaluated men, often in a veteran population,13, 16 and were often underpowered to address the influence of gender. As this study reveals, male gender, duration of reflux symptoms, hiatal hernia size, and segment length were associated with higher pathologic grades. The absence of H. pylori approached significance in predicting dysplasia. Interestingly, age, PPI use, and alcohol intake did not show an association.

Length of Barrett epithelium has been previously identified as an important factor for both the prevalence of dysplasia and cancer as well as the risk for malignant progression.16, 17 In our study, individuals with a segment length ≥6 cm were far more likely to harbor HGD or EAC than those with a segment <3 cm (OR: 8.374, P < .0001). Indeed, nearly half of all subjects with HGD or EAC had a Barrett segment ≥6 cm, as opposed to subjects with intestinal metaplasia alone, in whom only 5.4% possessed a segment length ≥6 cm. Whereas it is known that SSBE can give rise to adenocarcinoma,18 the prevalence of HGD dysplasia and EAC in short Barrett segments has varied in the literature.5, 14, 19 In our study the prevalence of HGD in subjects with SSBE was 8.8%, and the prevalence of EAC was 3.5%. These numbers are slightly higher than what others have reported and likely reflect the referral patterns at our center. Whereas we do see a large number of subjects with metaplasia alone, we are more likely to be referred subjects with HGD or EAC than other centers. Nevertheless, our findings underscore that there is an appreciable risk of developing adenocarcinoma in short Barrett segments and these patients should be surveyed accordingly.

Prior studies have shown that hiatal hernias are more common in subjects with HGD than in controls without GERD or with intestinal metaplasia alone.5, 13 In our study the presence of a hernia approached but did not reach significance (OR: 5.168, P = .061) in predicting higher pathologic grade. We did not, however, evaluate normal controls. Only subjects with Barrett were evaluated and, across all pathologic grades, the majority of our patients possessed a hiatal hernia. Hernia size, however, was strongly predictive of higher pathologic grades, and 34% of subjects with HGD had hernias >4 cm, compared with 10% of subjects with intestinal metaplasia alone. This is consistent with other studies5, 13 and the concept that greater disruption of the antireflux barrier promotes greater reflux injury and subsequent neoplastic progression.

The use of reflux history in guiding Barrett surveillance is controversial. GERD is highly prevalent in the general population and it is estimated that 15% to 20% of adults have weekly symptoms.20 It is estimated that in a population of >100,000, over 10,000 subjects would have reflux symptoms with a corresponding incidence of EAC of only 2.3/100,000 per year.21 In our study, over 75% of subjects with Barrett reported weekly symptoms of reflux. Whereas the mere presence of reflux was not predictive of higher pathologic grades, the duration of reflux was. Our results suggest that a longstanding history of reflux symptoms, particularly for ≥20 years, was significantly associated with an initial presentation of HGD or adenocarcinoma. Indeed, subjects with HGD or EAC had a history of reflux symptoms nearly twice as long as subjects with Barrett alone (17.73 vs 7.93 years; P = .0027). Interestingly, there was no difference in symptom duration between subjects with intestinal metaplasia and those with indefinite/LGD. Given that there is a high degree of interobserver variability in establishing the diagnosis of LGD and that the vast majority of subjects never progress,22 the clinical relevance of identifying individuals with indefinite/LGD is uncertain at best.

The role of H. pylori in BE is controversial. An inverse association between Cag A+ H. pylori infection and esophageal adenocarcinoma has been reported secondary to decreased gastric acidity from atrophic gastritis.23, 24 In our study, absence of H. pylori approached significance in predicting higher pathologic grades; however, we did not evaluate for the presence of atrophic gastritis or for Helicobacter strain.

Smoking and alcohol consumption were not significant predictors of HGD/EAC. Several observational studies have similarly shown no association between alcohol and EAC.25, 26 The role of smoking is somewhat more controversial, but consistent associations have not been found.13, 14 In one population-based study where risk of EAC among smokers appeared to be more than doubled, a dose-response pattern was seen and little risk reduction was observed until smoking cessation for >30 years.25 In this analysis, we only classified subjects as nonsmokers or smokers, the latter of whom were defined as smoking daily for at least 1 year in the past. It is possible that with a more detailed stratification of smokers, based on history and quantity, an association may emerge. This will be better evaluated in an upcoming prospective analysis. Whereas increasing BMI is associated with Barrett metaplasia, its role in adenocarcinoma is less clear. Whereas some studies have shown an association, others have not.27–29 In our patient population, increasing BMI did not achieve significance (P = −.266). Our population had a fairly high number of individuals with Barrett-associated adenocarcinoma, several of whom had dysphagia-related weight loss preceding their diagnosis. In order to adequately address this issue, a prospective evaluation with longitudinal documentation of BMI, including the period preceding cancer development, is necessary.

In summary, the presence of HGD or EAC was associated with male gender, duration of GERD symptoms, hernia size, and segment length in subjects newly diagnosed with BE. This study suggests that a combination of endoscopic and clinical criteria may be helpful in guiding endoscopic screening and surveillance, and better utilization of health-care resources. Given the cost of endoscopic screening, and the low likelihood of progression to cancer in individuals with metaplasia and even low-grade dysplasia, these results suggest that precise and dichotomous criteria (length ≥6 cm, hernia size ≥4 cm, and symptom duration >20 years) can be prospectively evaluated in an attempt to create algorithms for dividing patients into high-risk and low-risk surveillance programs. Validation of these results in a larger, prospective trial is indicated and will be forthcoming.