Abstract
Background and aims
Updated diagnostic guidelines for eosinophilic esophagitis (EoE) have eliminated the requirement for a proton pump inhibitor (PPI) trial, but there are no models to identify patients with EoE based on these new criteria. We aimed to develop a predictive model for diagnosis of EoE based on the updated EoE diagnostic guidelines.
Methods
We performed a secondary analysis of a prospective study of adult patients referred for outpatient esophagogastroduodenoscopy at University of North Carolina who had symptoms of esophageal dysfunction; patients with prevalent EoE were excluded. We analyzed data from 206 EoE cases (mean age 40.1, 62.6% male, 93.2% white) and 306 controls (mean age 52.3, 37.9% male, 79.7% white). We built predictive models for case-control status, using clinical, endoscopic, and histologic features, and defining EoE by either the new or historical definition of PPI non-response. Model discrimination was assessed by the area under the receiver-operator characteristic curve (AUC).
Results
Before endoscopy, younger age, male sex, history of atopic condition or food allergy, and dysphagia identified patients with EoE with an AUC of 0.83. When we included endoscopy findings suggestive of EoE, the model identified patients with EoE with an AUC of 0.92; this increased to 0.99 when histology was included.
Conclusion
We developed a model to identify patients with EoE, without a trial of PPIs, based on updated diagnostic guidelines. Clinical features and endoscopic findings identified patients with EoE with an AUC of 0.92—even without histologic data and in the absence of dysphagia. This model can be used to select patients with upper gastrointestinal symptoms but without dysphagia for early diagnostic endoscopy. The model can also be used to identify cases of EoE when eosinophil counts are greater than 15 in biopsies but other causes of esophageal eosinophilia cannot necessarily be excluded.
Keywords: esophagus, prediction, modeling, inflammation
Introduction
Eosinophilic esophagitis (EoE) is a distinct clinicopathologic syndrome, but can share clinical, endoscopic, and histologic features with gastroesophageal reflux disease (GERD) and other causes of upper gastrointestinal symptoms.1–3 Because of this clinical overlap, the first iteration of EoE diagnostic guidelines required a high-dose proton pump inhibitor (PPI) trial (or reflux testing) to diagnose EoE.4 If there was PPI response, the diagnosis was GERD; if there was no PPI response, the diagnosis was EoE. However, this paradigm has evolved over time, as more has been learned about the condition. In the 2011 and 2013 guidelines, there was a recognition that some patients who appeared to have EoE (and who did not clinically have reflux) would respond to PPI therapy, and these patients were diagnosed with PPI-responsive esophageal eosinophilia (PPI-REE).5,6 In subsequent years extensive research efforts concluded that patients with PPI-REE had many similarities to, and in some cases were indistinguishable from, EoE cases.7–10 This observation brought into question the role of the PPI trial as a diagnostic criterion.11,12 In 2018, updated international consensus diagnostic guidelines were developed, which removed the PPI trial from the diagnostic algorithm,13 following the example of guidelines released in Europe in 2017.14 The definition of EoE has thus expanded to include subjects with a previous diagnosis of PPI-REE.15 However, the implications of this change in membership of populations with EoE for the conclusions of prior predictive modeling research is not known.
Given the overlap in upper GI symptoms between EoE and other conditions, we retrospectively developed a model that identified factors that were highly predictive of EoE.16 In a subsequent prospective study, we validated this model.17 However, these models were constructed using what is now considered a historical or “classic” EoE definition with PPI non-response. With the updated A Working Group on Proton-pump Inhibitor Responsive Esophageal Eosinophilia (AGREE) diagnostic guidelines, there are no predictive clinical models and little is known about the diagnostic performance of prior models when an EoE case group includes patients who were previously considered to have PPI-REE.9,18
Therefore, we aimed to build an optimally predictive clinical model to determine EoE case status according to the AGREE consensus definition using data available to clinicians at three clinically important time points: a clinic visit before diagnostic endoscopy, at a diagnostic endoscopy but before biopsy is performed and histology results are available, and after histology results are available. Our secondary aims were to build and compare similarly constructed predictive models for the historical definition of EoE with a PPI trial and for subjects previously defined as PPI-REE.
Methods
Study design and patient population
We performed a secondary analysis of a study of prospectively enrolled adults undergoing outpatient upper endoscopy at the University of North Carolina (UNC) from July, 2011 through September, 2015. The study was approved by the UNC Institutional Review Board. The methodologic details of the cohort have been previously published.17,19–22 In brief, subjects with symptoms of esophageal dysfunction, such as dysphagia, food impaction, heartburn, reflux, or chest pain were eligible if they did not have a history of EoE, other eosinophilic gastrointestinal disorder, gastrointestinal bleeding, current anticoagulant use, esophageal cancer, esophageal varices, or prior esophageal surgery. For the present study, cases of EoE were re-defined as per the AGREE guidelines, including evaluation of the contribution of other potential causes of esophageal eosinophilia, but were also classified as per their original diagnosis (“classic” EoE or PPI-REE). Controls were subjects who did not have EoE. Consistent with the AGREE guidelines, no subjects with an overall peak eosinophil count per high-power-field less than 15 could be diagnosed with EoE, but subjects with an eosinophil count of 15 or more were not diagnosed with EoE if they had an alternative cause of esophageal eosinophilia.
Predictor variables
Candidate predictors included demographics, medical history, esophageal endoscopic findings, and histologic measures (see Supplemental Material for full details). The eosinophil counts (eosinophils per high-power field; eos/hpf; hpf = 0.24mm2) were determined prospectively according to our previously validated protocol.23,24 Candidate predictors with p>0.05 in a Wilcoxon rank-sum test for continuous variables or in a Cochran-Mantel-Haenszel test for categorical variables were excluded from further consideration.
Parameterization of predictor variables
We performed three case-control analyses: the primary analysis that compared EoE cases by AGREE consensus definition with controls, a secondary analysis that compared EoE cases by the historical definition of non-response to a PPI trial with controls and excluded PPI-REE, and a secondary analysis that compared PPI-REE cases to controls and excluded subjects with the historical definition of EoE non-responsive to a PPI trial. The primary outcome definition was used for all parameterization decisions, and this parameterization was also used for the secondary outcome definitions. For full details on the parameterization, please see the Supplemental materials. In addition to assessing potential predictor variables for data missing values, potential predictor variables were assessed for collinearity and variables with a correlation coefficient greater than 0.3 were combined into a single predictor variable that was positive if either variable was positive, missing if both variables were missing, and otherwise was negative. If there was a negative correlation coefficient less than −0.3, the variable with the lower Akaike information criterion (AIC), a likelihood-based model selection criterion, was retained and the variable with the higher AIC was rejected.25
Predictive model building process and reporting
Model parameters were selected iteratively by adding the parameter that would result in the lowest AIC until none of the remaining parameters would improve AIC by a threshold of p less than 0.05 for the likelihood ratio test between nested models. Parameters that did not meet the significance threshold were then removed until all remaining parameters met the threshold. For predictor selection, missing values predictor variables were imputed assuming a multivariate normal distribution using the Markov Chain Monte Carlo method.26 The predictive model was fit by the same iterative method for the outcomes of EoE case status as defined by the AGREE consensus criteria, EoE case status as historically defined by a PPI trial, and PPI-REE as historically defined. For each case definition, predictive models were fit limited to clinical criteria that would be available before diagnostic endoscopy, limited to clinical and endoscopic criteria that would be available at a diagnostic endoscopy before biopsy, with histology available in addition to clinical and endoscopic potential predictor variables but excluding eosinophil count, and all histology available in addition to clinical and endoscopic potential predictor variables. The predictive models for AGREE EoE diagnosis were also fit limiting to clinical criteria, limiting to clinical and endoscopic criteria, and with all predictors (but including only those subjects with an eosinophil count greater or equal to 15 to estimate marginal and conditional misclassification of EoE diagnosis with definitions limited to various diagnostic criteria). For each model, we reported the results of the iterative fitting process as well as the receiver operator characteristic curve and its integral calculated as the area under the curve (AUC) and its 95 % confidence limits. The authors also report the cross-validated AUC to mitigate optimism from validating the model on the learning dataset. Model calibration was assessed by comparing deciles of the predicted case probability to the proportion of cases within each stratum. While the dataset with imputation of missing values was used to select predictor variables, the final selected model was fit in a complete case analysis for all its included variables. Participants with any missing data for candidate variables were excluded for all steps of model selection.
Results
Included subjects and baseline characteristics
Of 578 study subjects, 39 were excluded for missing case status, and 27 were excluded for missing histology variables, which yielded 512 subjects eligible for analysis. Subjects meeting the AGREE consensus guidelines for diagnosis of EoE had similar demographic, endoscopic, and histologic characteristics to previously described populations of patients with EoE (Table 1).
Table 1.
Demographic characteristics, pertinent past medical history, and symptoms at enrollment of 512 subjects included for analysis by case control status.
| Control subjects (N = 306) | Classic definition of EoE with PPI trial (N = 149) | PPI-responsive esophageal eosinophilia (N = 57) | AGREE consensus definition of EoE (N = 206) | |
|---|---|---|---|---|
|
| ||||
| Demographic characteristics | ||||
|
| ||||
| Age – mean (SD) | 52.3 (14.3) | 37.9 (12.9) | 46.0 (13.8) | 40.1 (13.6) |
| Male – N (%) | 116 (37.9) | 90 (60.4) | 39 (68.4) | 129 (62.6) |
| Race:* Non-Hispanic White – N (%) | 243 (79.7) | 141 (94.6) | 51 (89.5) | 192 (93.2) |
| Black – N (%) | 50 (16.4) | 4 (2.7) | 4 (7.0) | 8 (3.9) |
| Hispanic White – N (%) | 5 (1.6) | 2 (1.3) | 1 (1.8) | 3 (1.5) |
| Native American – N (%) | 4 (1.3) | 0 (0.0) | 0 (0.0) | 0 (0.0) |
| Multiple races – N (%) | 2 (0.7) | 1 (0.7) | 0 (0.0) | 1 (0.5) |
|
| ||||
| Past medical history | ||||
|
| ||||
| Any atopic disease† – N (%) | 171 (60.6) | 102 (72.3) | 38 (76.0) | 140 (73.3) |
| Food allergy‡ – N (%) | 47 (16.5) | 50 (35.2) | 12 (23.5) | 62 (32.1) |
| Endoscopic dilation – N (%) | 97 (31.7) | 53 (35.6) | 13 (22.8) | 66 (32.0) |
|
| ||||
| Symptoms | ||||
|
| ||||
| Dysphagia – N (%) | 224 (73.2) | 146 (98.0) | 54 (94.7) | 200 (97.1) |
| Gastroesophageal reflux symptoms – N (%) | 128 (41.8) | 18 (12.1) | 11 (19.3) | 29 (14.1) |
| Abdominal pain – N (%) | 32 (10.5) | 9 (6.0) | 3 (5.3) | 12 (5.8) |
| Nausea or vomiting – N (%) | 17 (5.6) | 3 (2.0) | 2 (3.5) | 5 (2.4) |
|
| ||||
| Endoscopy findings | ||||
|
| ||||
| Rings – N (%) | 38 (12.4) | 122 (81.9) | 30 (52.6) | 152 (73.8) |
| Furrows – N (%) | 20 (6.5) | 132 (88.6) | 27 (47.4) | 159 (77.2) |
| Edema/decreased vascular pattern – N (%) | 9 (2.9) | 73 (49.0) | 2 (3.5) | 75 (36.4) |
| White plaques – N (%) | 13 (4.2) | 72 (48.3) | 11 (19.3) | 83 (40.3) |
| Erythema§ – N (%) | 7 (2.3) | 1 (0.7) | 1 (1.8) | 2 (1.0) |
| Crepe paper – N (%) | 3 (1.0) | 11 (7.4) | 1 (1.8) | 12 (5.8) |
| Erosive esophagitis – N (%) | 46 (15.0) | 4 (2.7) | 6 (10.5) | 10 (4.9) |
| Erosions – N (%) | 10 (3.3) | 2 (1.3) | 0 (0.0) | 2 (1.0) |
| Stricture – N (%) | 56 (18.3) | 50 (33.6) | 9 (15.8) | 59 (28.6) |
| Diffuse narrowing – N (%) | 13 (4.2) | 48 (32.2) | 3 (5.3) | 51 (24.8) |
| Schatzki's ring – N (%) | 31 (10.1) | 11 (7.4) | 10 (17.5) | 21 (10.2) |
| Hiatus hernia – N (%) | 129 (42.2) | 21 (14.1) | 15 (26.3) | 36 (17.5) |
|
| ||||
| Histology findings | ||||
|
| ||||
| Peak overall eosinophil count – Mean (SD) | 2.9 (8.9) | 121.0 (110.0) | 65.6 (52.8) | 106.0 (101.0) |
| Peak overall eosinophil count – Range | 0.0 – 85.4 | 15.7 – 667.0 | 15.0 – 273.0 | 15.0 – 667.0 |
| Eosinophil degranulation| – N (%) | 34 (12.5) | 122 (94.6) | 22 (40.7) | 144 (78.7) |
| Eosinophil microabscess| – N (%) | 5 (1.8) | 88 (68.8) | 3 (5.6) | 91 (50.0) |
| Basal zone hyperplasia¶ – N (%) | 29 (11.0) | 61 (50.0) | 11 (20.4) | 72 (40.9) |
| Spongiosis** – N (%) | 80 (28.8) | 117 (88.6) | 19 (35.2) | 136 (73.1) |
| Lamina propria fibrosis†† – N (%) | 5 (6.3) | 33 (41.3) | 5 (19.2) | 38 (35.8) |
SD, standard deviation
1 missing
39 missing
35 missing
2 missing
56 missing
57 missing
48 missing
11 missing among 190 (37%) with evaluable lamina propria.
Predictive model limited to characteristics available prior to diagnostic endoscopy
When limited to predictor variables that were available prior to a diagnostic endoscopy, the procedure selected similar included predictor variables with both the broader consensus definition of EoE and with the historical definition of EoE confirmed by PPI trial (Tables 2 and 3). The identified variables for the primary outcome were younger age, male sex, presence of dysphagia, and White race/ethnicity. The predictive model had similarly very good discrimination in each case definition: EoE as defined by AGREE consensus (AUC 0.84, 95%CI 0.81–0.88, cross-validated AUC 0.84, 95%CI 0.80–0.88), subjects previously defined as PPI-REE (AUC = 0.76, 95%CI 0.70–0.83, cross-validated AUC 0.74, 95%CI 0.67–0.81), and subjects previously defined as EoE following non-response to a PPI trial (AUC 0.85, 95%CI 0.82–0.89, cross-validated AUC 0.84, 95%CI 0.81–0.88, Figures 1A/2A).
Table 2.
Multivariable adjusted case control odds ratios with 95% confidence limits and model fit and prediction statistics for cases defined by AGREE consensus definition for the predictive models built with factors available prior to diagnostic endoscopy, factors available at diagnostic endoscopy, and factors available after histology results.
| Factors available prior to diagnostic endoscopy | Factors available at diagnostic endoscopy | Factors available after histology results excluding eosinophil count | Factors available after histology results including eosinophil count | |
|---|---|---|---|---|
|
| ||||
| Age at biopsy (per year) | 0.94 (0.93 – 0.96) | 0.96 (0.94 – 0.97) | 0.96 (0.94 – 0.98) | 0.97 (0.94 – 1.00) |
| Male sex versus female | 2.86 (1.80 – 4.55) | 2.20 (1.29 – 3.76) | 2.99 (1.56 – 5.72) | |
| White race versus others | 3.51 (1.76 – 7.04) | |||
| History of atopic condition or food allergy versus neither | 1.72 (1.08 – 2.75) | |||
| Dysphagia at baseline | 15.01 (6.12 – 36.83) | 7.58 (2.7 – 21.27) | 6.14 (2.03 – 18.53) | 12.27 (1.37 – 110.01) |
| Rings, furrows, edema, white plaques, narrowing, or crepe paper appearance on endoscopy | 20.69 (11.95 – 35.82) | 9.74 (5.33 – 17.77) | 5.55 (2.15 – 14.38) | |
| Erosive esophagitis on endoscopy | 0.28 (0.1 – 0.74) | 0.36 (0.14 – 0.92) | ||
| Hiatus hernia on endoscopy | 0.53 (0.29 – 0.97) | |||
| Eosinophil degranulation | 3.80 (1.94 – 7.46) | |||
| Eosinophil microabscess | 9.05 (3.08 – 26.55) | |||
| Overall peak eosinophil count (per additional ten eosinophils per HPF) | 1.54 (1.12 – 1.22) | |||
|
| ||||
| Model Akaike information criterion* | 517.19 | 372.39 | 312.44 | 140.64 |
| Area under the receiver-operator characteristic curve | 0.84 (0.81 – 0.88) | 0.92 (0.89 – 0.94) | 0.95 (0.93 – 0.97) | 0.99 (0.99 – 1.00) |
| Cross-validated area under the receiver-operator characteristic curve | 0.84 (0.80 – 0.87) | 0.91 (0.88 – 0.94) | 0.94 (0.92 – 0.96) | 0.99 (0.98 – 1.00) |
Akaike information criterion measures model fit, with a lower value indicating a better fit. AGREE, a working group on proton-pump inhibitor responsive esophageal eosinophilia, HPF, high power field.
Table 3.
Predictor variables selected by stepwise selection on the Akaike information criterion for predictor variables limited to information available prior to diagnostic endoscopy, predictor variables limited to information available during diagnostic endoscopy, all predictor variables excluding eosinophil count, and all predictor variables depending on the definition of cases of EoE defined as A Working Group on Proton-pump Inhibitor Responsive Esophageal Eosinophilia (AGREE) consensus definition, as subjects previously diagnosed as PPI-responsive esophageal eosinophilia (PPI-REE), and as previously defined as EoE confirmed with a PPI trial.
| AGREE / PPIREE / PPI Trial -, negative predictor +, positive predictor 0, not included | Model limited to clinical criteria that would be available before diagnostic endoscopy | Model limited to clinical and endoscopic criteria that would be available at a diagnostic endoscopy before biopsy | Model with histology in addition to clinical and endoscopic potential predictor variables but excluding eosinophil count | Model with histology in addition to clinical and endoscopic potential predictor variable including eosinophil count |
|---|---|---|---|---|
|
| ||||
| More advanced age | − / − / − | − / − / − | − / − / − | − / 0 / 0 |
| Male sex | + / + / + | + / + / 0 | + / + / 0 | 0 / 0 / 0 |
| White race/ethnicity | + / 0 / + | 0 / 0 / + | 0 / 0 / 0 | 0 / 0 / 0 |
| Atopy or food allergies | + / 0 / 0 | 0 / 0 / 0 | 0 / 0 / − | 0 / 0 / 0 |
| Dysphagia | + / + / + | + / + / + | + / + / 0 | + / 0 / 0 |
|
| ||||
| Rings, furrows, white plaques, narrowing, or crepe paper appearance | + / + / + | + / + / + | + / 0 / + | |
| Esophagitis | − / 0 / − | 0 / 0 / 0 | 0 / 0 / 0 | |
| Hiatus hernia | − / 0 / − | 0 / 0 / − | 0 / 0 / − | |
|
| ||||
| Eosinophil degranulation | + / + / + | 0 / 0 / 0 | ||
| Eosinophilic microabscess | + / 0 / + | 0 / − / 0 | ||
| Spongiosis | 0 / 0 / + | 0 / 0 / + | ||
|
| ||||
| Higher peak overall eosinophil count | + / + / + | |||
EoE, eosinophilic esophagitis, AGREE, A Working Group on Proton-pump Inhibitor Responsive Esophageal Eosinophilia, PPI, proton-pump inhibitor, PPIREE, PPI-responsive esophageal eosinophilia.
Figure 1.
Receiver-operator characteristic curves for predictive variables limited to (A) information available prior to diagnostic endoscopy, (B) limited to information available at the time of diagnostic endoscopy, (C) for all potential predictive variables excluding overall peak eosinophil count, and (D) for all potential predictive variables including overall peak eosinophil count for prediction of case control status among cases of EoE as defined by A Working Group on Proton-pump Inhibitor Responsive Esophageal Eosinophilia (AGREE) consensus definition, as subjects previously diagnosed as PPI-responsive esophageal eosinophilia (PPI-REE), and as previously defined as EoE confirmed with a PPI trial.
Figure 2.
Decile calibration plot for predictive variables limited to (A) information available prior to diagnostic endoscopy, (B) limited to information available at the time of diagnostic endoscopy, (C) for all potential predictive variables excluding overall peak eosinophil count, and (D) for all potential predictive variables including overall peak eosinophil count for prediction of case control status among cases of EoE as defined by A Working Group on Proton-pump Inhibitor Responsive Esophageal Eosinophilia (AGREE) consensus definition.
Predictive model limited to characteristics available at diagnostic endoscopy
When limited to predictor variables that were available at the time of a diagnostic endoscopy but without histology results, a core set of highly correlated EoE findings: rings, furrows, edema, white plaques, or crepe paper appearance, as well as younger age, male sex, and dysphagia, were included in each model. The model building in subjects previously defined as PPI-REE did not include absence of erosive esophagitis and absence of hiatus hernia, which were included in the models constructed using other definitions limited to clinical and endoscopic characteristics. The predictive model had the best discrimination with cases as the historical definition of EoE non-responsive to a PPI trial (AUC 0.95, 95%CI 0.93–0.97, cross-validated AUC 0.94, 95%CI 0.92–0.96), followed by EoE as defined by AGREE consensus (AUC 0.92, 95%CI 0.89–0.94, cross-validated AUC 0.91, 95%CI 0.88–0.94), and slightly worse performance with subjects previously defined as PPI-REE (AUC 0.85, 95%CI 0.81–0.90, cross-validated AUC 0.83, 95%CI 0.78–0.89, Figures 1B/2B).
Predictive model with all clinical, endoscopic, and histologic predictors eligible
We performed the model selection procedure allowing all predictor variables, excluding eosinophil count. This produced predictive models with younger age, male sex, dysphagia, endoscopic EoE findings of rings, furrows, edema, white plaques, or crepe paper appearance, eosinophil degranulation, and microabscess. This predictive model’s performance was sensitive to case definition: EoE as defined by AGREE consensus (AUC = 0.95, 95%CI 0.93–0.97, cross-validated AUC 0.94, 95%CI 0.92–0.96), subjects previously defined as PPI-REE (AUC 0.87, 95%CI 0.82–0.91, cross-validated AUC 0.85, 95%CI 0.80–0.90), and subjects previously defined as EoE following a PPI trial (AUC 0.99, 95%CI 0.98–1.00, cross-validated AUC 0.98, 95%CI 0.97–0.99, Figures 1C/2C). A similar model was fit including eosinophil count, and this model was, unsurprisingly, nearly deterministic in its prediction regardless of the case definition: EoE as defined by AGREE consensus (AUC = 0.99, 95%CI 0.99–1.00, cross-validated AUC 0.99, 95%CI 0.98–1.00), subjects previously defined as PPI-REE (AUC = 0.99, 95%CI 0.99–1.00, cross-validated AUC 0.99, 95%CI 0.99–1.00), and subjects previously defined as EoE following a PPI trial (AUC = 0.99, 95%CI 0.97–1.00, cross-validated AUC 0.97, 95%CI 0.94–1.00, Figures 1D/2D).
Predictive Models in Subsets of Clinical Interest
There were 15 (6.8%) subjects with an eosinophil count greater than 15 per high-power field without a diagnosis of EoE per AGREE criteria. Among subjects with 15 or more eosinophils per high-powered field, a model limited to clinical predictors had an AUC of 0.88 (95%CI 0.79–0.96) and cross-validated AUC of 0.84 (95%CI 0.74–0.94) with overall peak eosinophil count, age, and presence or absence of dysphagia. A model with clinical characteristics, endoscopic characteristics, and overall peak eosinophil count had an AUC of 0.91 (95%CI 0.83–0.99) and cross-validated AUC of 0.86 (95%CI 0.76–0.96) with sex, rings, furrows, edema, white plaques, or crepe paper appearance, overall peak eosinophil count and absence of hiatus hernia. The full model with absence of hiatus hernia, absence of degranulation, presence of spongiosis, and higher eosinophil count had an AUC of 0.91 (95%CI 0.81–1.00) and cross-validated AUC of 0.86 (95%CI 0.73–0.99, Table 4).
Table 4.
Multivariable adjusted case control odds ratios with 95% confidence limits for 206 eosinophilic esophagitis diagnoses per AGREE criteria among 221 subjects with overall peak eosinophil count greater or equal to 15 per high-power field.
| Case control odds ratio (95% confidence limits) | |
|---|---|
|
| |
| Clinical predictors and peak overall eosinophil count | |
|
| |
| Age at biopsy (per year) | 0.94 (0.90 – 0.99) |
| Dysphagia at baseline | 15.7 (2.5 – 96.8) |
| Higher peak overall eosinophil count (per hpf) | 1.03 (1.01 – 1.06) |
|
| |
| Clinical and endoscopic predictors and eosinophils | |
|
| |
| Rings, furrows, edema, white plaques, narrowing, or crepe paper appearance on endoscopy | 3.9 (1.1 – 13.3) |
| Erythema | 0.1 (0.0 – 1.1) |
| Hiatus hernia | 0.1 (0.0 – 0.4) |
| Higher peak overall eosinophil count (per hpf) | 1.03 (1.01 – 1.06) |
|
| |
| All clinical, endoscopic, and histology predictors | |
|
| |
| Hiatus hernia | 0.16 (0.04 – 0.61) |
| Eosinophil degranulation | 0.07 (0.01 – 0.63) |
| Spongiosis | 6.1 (1.4 – 26.3) |
| Higher peak overall eosinophil count (per hpf) | 1.03 (1.00 – 1.05) |
AGREE, A Working Group on Proton-pump Inhibitor Responsive Esophageal Eosinophilia, HPF, high-power field.
There were 6 (6.8%) subjects with a diagnosis of EoE per AGREE criteria among the 88 without dysphagia at baseline. A model limited to clinical predictors had an AUC of 0.87 (95% confidence limits 0.77–0.96, cross-validated AUC 0.77 95%CI 0.63–0.92) with age and gender at baseline (Table S1). There were 5 (8.3%) subjects with a diagnosis of EoE per AGREE criteria among the 60 with GERD symptoms and without dysphagia at baseline. A model limited to clinical predictors had an AUC of 0.83 (95%CI 0.69–0.98, cross-validated AUC 0.81, 95%CI 0.66–0.96) with age at baseline.
Discussion
In this study, we used prospectively collected data to build predictive models to support clinical diagnosis of EoE using the AGREE consensus definition prior to diagnostic endoscopy, after diagnostic endoscopy, and after diagnostic endoscopy when histology results were available. These models had excellent performance in discerning EoE from control subjects for the new AGREE consensus definition, but also held if EoE was defined historically with non-response to a PPI trial. An astute clinician can use these predictors to arrive at a reasonably certain diagnosis, and these predictive models could guide the decision to initiate treatment in the admittedly short window before esophageal biopsies result. Furthermore, when a peak eosinophil count on esophageal biopsies of 15 or greater is observed, the AGREE consensus definition requires exclusion of alternative causes of esophageal eosinophilia, which may be uncertain. We fit models among only those participants with a peak eosinophil count of 15 or greater of endoscopic, histologic, and clinical features to help differentiate EoE from other causes of esophageal eosinophilia.
Several clinically relevant observations from the modeling process were made. Prior to endoscopy, the models had very high discrimination and could be useful to select patient for early diagnostic endoscopy. Dysphagia was the strongest predictor of EoE diagnosis among the included patients with various esophageal symptoms. The presence or absence of dysphagia itself had a greater impact on risk than a 44-year difference in age or the presence of male sex, white race, and any history of atopy or food allergies combined.
While the predicted risk of diagnosis with EoE was generally low in patients with esophageal symptoms but no dysphagia, this was sensitive to age. For example, an 18-year-old woman with esophageal symptoms but neither dysphagia, atopy, food allergies, nor White race had a 13.4% predicted chance of EoE (95%CI 3.7–48.1%), while the same scenario in a 50-year-old the predicted chance was 1.6% (95%CI 0.4%−5.8%). This suggests that while an endoscopy for EoE may have a modest yield in a middle-aged person with possible esophageal symptoms but no dysphagia,27 early diagnostic endoscopy may be considered in young adults or adults with other risk factors for EoE even with non-dysphagia esophageal symptoms. To increase the clinical utility of this model, we have posted a calculator online (https://gicenter.med.unc.edu/cedas/index.php).
There have been several prior attempts to use statistical models to differentiate EoE aside from our previous analyses.16,17 One such attempt built a model with younger age, endoscopic features of EoE, and absence of use of PPIs, finding an AUC=0.86.28 A predictive modeling study of 23 EoE patients and 20 controls found peripheral eosinophilia, food impaction, and PPI-refractory heartburn to have high sensitivity/specificity for EoE.29 Another study found similar risk factors to the ones we report here,30 while a study in children with EoE found dysphagia, stricture, other endoscopic abnormalities, basal zone hyperplasia, and degranulated eosinophils predicted EoE compared to controls.31 A modeling study in children achieved an AUC=0.86 using male sex, dysphagia, history of food impaction, absence of pain or heartburn, linear furrows, and white plaques.32 To our knowledge, none of these would be applicable to the new diagnostic guidelines.
The study has limitations that are important to consider. The subjects were enrolled at a single academic center and findings may be subject to referral and other selection biases. However, the overall characteristics of these patients were reflective of other reported large EoE populations and we enrolled all patients coming for outpatient endoscopy, not just referral patients. Since only adults were included, a similar study would be required to determine if these results are applicable to pediatric EoE patients. There is not a definitive gold standard test for EoE diagnosis, and clinical diagnosis as per consensus guidelines used as the reference standard. Some of the control patients were on PPI therapy at the time of endoscopy, so if they had esophageal eosinophilia that resolved with PPI use, they would not have been identified as possible EoE cases. However, this would bias the models towards the null. The models were also developed using predictive modeling techniques and their findings should not be interpreted as causal. EoE may occur with an eosinophil count less than 15, though this is a standard diagnostic criterion.33 Such cases would be excluded by our definition, and typically require more specialized staining than is done routinely to detect. Strengths of this study include analysis of a large, prospectively recruited population of cases and controls undergoing outpatient endoscopy for evaluation of upper GI symptoms, rigorous and standardized data collection and clinical phenotyping, ability to assess historical diagnoses, utilization of the most recent EoE diagnostic criteria, and sub-analyses focusing on subjects with elevated eosinophils regardless of case/control status.
In conclusion, as EoE can be difficult to distinguish from other causes of upper GI symptoms in adults, we built predictive models to support diagnosis of EoE using the AGREE consensus definition prior to diagnostic endoscopy, after diagnostic endoscopy but before histology results, and after diagnostic endoscopy and histology results were available. We found that such models had excellent predictive indices for EoE even without histologic information, and near perfectly discriminated EoE cases from non-EoE control when full clinical, endoscopic, and histologic data were available. Such models can help clinicians make decisions for patients with upper GI symptoms about whether to perform early diagnostic endoscopy or gastroenterology referral for suspicion of EoE. These tools may also help clinicians apply the new AGREE diagnostic criteria to exclude alternative causes of esophageal eosinophilia in cases where biopsy reveal esophageal eosinophilia but EoE diagnosis is uncertain.
Supplementary Material
What You Need to Know
Background
Updated diagnostic guidelines for eosinophilic esophagitis (EoE) have eliminated the requirement for a proton pump inhibitor trial, but there are no models to identify patients with EoE based on the new criteria.
Findings
We developed a model to identify patients with EoE, without a trial of proton pump inhibitors, based on updated diagnostic guidelines. Clinical features and endoscopic findings identified patients with EoE with an area under the curve of 0.92—even without histologic data and in the absence of dysphagia.
Implications for Patient Care
This model can be used to select patients with upper gastrointestinal symptoms but without dysphagia for early diagnostic endoscopy, and also to identify cases of EoE when eosinophil counts are greater than 15 in biopsies but other causes of esophageal eosinophilia cannot necessarily be excluded.
Acknowledgments
Financial support: This work was funded by K23 DK090073, K24DK100548, T32 DK007634, P30 DK034987, and R01 DK101856.
Potential competing interests: Dr. Dellon has received research funding from Adare, Allakos, GSK, Meritage, Miraca, Nutricia, Celgene/Receptos, Regeneron, and Shire; has received consulting fees from Adare, Aimmune, Alivio, Allakos, AstraZeneca, Banner, Bioras, Calypso, Celgene/Receptos, Enumeral, EsoCap, Gossamer Bio, GSK, Regeneron, Robarts, Salix, Shire, and educational grants from Allakos, Banner, and Holoclara. The other authors have no potential conflicts of interest.
Footnotes
Guarantor of the article: Evan S. Dellon MD MPH.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
References
- 1.Dellon ES. Diagnosis and Management of Eosinophilic Esophagitis. Clinical Gastroenterology and Hepatology 2012;10:1066–1078. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Straumann A, Katzka DA. Diagnosis and Treatment of Eosinophilic Esophagitis. Gastroenterology 2018;154:346–359. [DOI] [PubMed] [Google Scholar]
- 3.Lucendo AJ, Sánchez-Cazalilla M. Adult versus pediatric eosinophilic esophagitis: important differences and similarities for the clinician to understand. Expert Review of Clinical Immunology 2012;8:733–745. [DOI] [PubMed] [Google Scholar]
- 4.Furuta GT, Liacouras CA, Collins MH, et al. Eosinophilic esophagitis in children and adults: a systematic review and consensus recommendations for diagnosis and treatment. Gastroenterology 2007;133:1342–1363. [DOI] [PubMed] [Google Scholar]
- 5.Dellon ES, Gonsalves N, Hirano I, et al. ACG Clinical Guideline: Evidenced Based Approach to the Diagnosis and Management of Esophageal Eosinophilia and Eosinophilic Esophagitis (EoE): American Journal of Gastroenterology 2013;108:679–692. [DOI] [PubMed] [Google Scholar]
- 6.Liacouras CA, Furuta GT, Hirano I, et al. Eosinophilic esophagitis: updated consensus recommendations for children and adults. J Allergy Clin Immunol 2011;128:3–20.e6; quiz 21–22. [DOI] [PubMed] [Google Scholar]
- 7.Molina-Infante J, Ferrando-Lamana L, Ripoll C, et al. Esophageal eosinophilic infiltration responds to proton pump inhibition in most adults. Clin Gastroenterol Hepatol 2011;9:110–117. [DOI] [PubMed] [Google Scholar]
- 8.Dranove JE, Horn DS, Davis MA, et al. Predictors of response to proton pump inhibitor therapy among children with significant esophageal eosinophilia. J Pediatr 2009;154:96–100. [DOI] [PubMed] [Google Scholar]
- 9.Dellon ES, Speck O, Woodward K, et al. Clinical and Endoscopic Characteristics do Not Reliably Differentiate PPI-Responsive Esophageal Eosinophilia and Eosinophilic Esophagitis in Patients Undergoing Upper Endoscopy: A Prospective Cohort Study: American Journal of Gastroenterology 2013;108:1854–1860. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Lucendo AJ, Arias Á, Molina-Infante J. Efficacy of Proton Pump Inhibitor Drugs for Inducing Clinical and Histologic Remission in Patients With Symptomatic Esophageal Eosinophilia: A Systematic Review and Meta-Analysis. Clin Gastroenterol Hepatol 2016;14:13–22.e1. [DOI] [PubMed] [Google Scholar]
- 11.Molina-Infante J, Bredenoord AJ, Cheng E, et al. Proton pump inhibitor-responsive oesophageal eosinophilia: an entity challenging current diagnostic criteria for eosinophilic oesophagitis. Gut 2016;65:524–531. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Eluri S, Dellon ES. Proton pump inhibitor-responsive oesophageal eosinophilia and eosinophilic oesophagitis: more similarities than differences. Curr Opin Gastroenterol 2015;31:309–315. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Dellon ES, Liacouras CA, Molina-Infante J, et al. Updated International Consensus Diagnostic Criteria for Eosinophilic Esophagitis: Proceedings of the AGREE Conference. Gastroenterology 2018;155:1022–1033.e10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Lucendo AJ, Molina-Infante J, Arias Á, et al. Guidelines on eosinophilic esophagitis: evidence-based statements and recommendations for diagnosis and management in children and adults. United European Gastroenterol J 2017;5:335–358. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Wen T, Dellon ES, Moawad FJ, et al. Transcriptome analysis of proton pump inhibitor-responsive esophageal eosinophilia reveals proton pump inhibitor-reversible allergic inflammation. J Allergy Clin Immunol 2015;135:187–197. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Dellon ES, Gibbs WB, Fritchie KJ, et al. Clinical, endoscopic, and histologic findings distinguish eosinophilic esophagitis from gastroesophageal reflux disease. Clin Gastroenterol Hepatol 2009;7:1305–1313; quiz 1261. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Dellon ES, Rusin S, Gebhart JH, et al. A Clinical Prediction Tool Identifies Cases of Eosinophilic Esophagitis Without Endoscopic Biopsy: A Prospective Study: American Journal of Gastroenterology 2015;110:1347–1354. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Warners MJ, van Rhijn BD, Curvers WL, et al. PPI-responsive esophageal eosinophilia cannot be distinguished from eosinophilic esophagitis by endoscopic signs. Eur J Gastroenterol Hepatol 2015;27:506–511. [DOI] [PubMed] [Google Scholar]
- 19.Dellon ES, Rusin S, Gebhart JH, et al. Utility of a Noninvasive Serum Biomarker Panel for Diagnosis and Monitoring of Eosinophilic Esophagitis: A Prospective Study. Am J Gastroenterol 2015;110:821–827. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Dellon ES, Cotton CC, Gebhart JH, et al. Accuracy of the Eosinophilic Esophagitis Endoscopic Reference Score in Diagnosis and Determining Response to Treatment. Clin Gastroenterol Hepatol 2016;14:31–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Tappata M, Eluri S, Perjar I, et al. Association of mast cells with clinical, endoscopic, and histologic findings in adults with eosinophilic esophagitis. Allergy 2018;73:2088–2092. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Reed CC, Wolf WA, Cotton CC, et al. Optimal Histologic Cutpoints for Treatment Response in Patients With Eosinophilic Esophagitis: Analysis of Data From a Prospective Cohort Study. Clin Gastroenterol Hepatol 2018;16:226–233.e2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Dellon ES, Fritchie KJ, Rubinas TC, et al. Inter- and intraobserver reliability and validation of a new method for determination of eosinophil counts in patients with esophageal eosinophilia. Dig Dis Sci 2010;55:1940–1949. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Rusin S, Covey S, Perjar I, et al. Determination of esophageal eosinophil counts and other histologic features of eosinophilic esophagitis by pathology trainees is highly accurate. Hum Pathol 2017;62:50–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Akaike H Likelihood of a model and information criteria. Journal of Econometrics 1981;16:3–14. [Google Scholar]
- 26.Schafer JL. Multiple imputation: a primer. Stat Methods Med Res 1999;8:3–15. [DOI] [PubMed] [Google Scholar]
- 27.Katz PO, Gerson LB, Vela MF. Guidelines for the Diagnosis and Management of Gastroesophageal Reflux Disease: American Journal of Gastroenterology 2013;108:308–328. [DOI] [PubMed] [Google Scholar]
- 28.Prasad GA, Talley NJ, Romero Y, et al. Prevalence and predictive factors of eosinophilic esophagitis in patients presenting with dysphagia: a prospective study. Am J Gastroenterol 2007;102:2627–2632. [DOI] [PubMed] [Google Scholar]
- 29.von Arnim U, Wex T, Röhl F-W, et al. Identification of clinical and laboratory markers for predicting eosinophilic esophagitis in adults. Digestion 2011;84:323–327. [DOI] [PubMed] [Google Scholar]
- 30.Mackenzie SH, Go M, Chadwick B, et al. Eosinophilic oesophagitis in patients presenting with dysphagia--a prospective analysis. Aliment Pharmacol Ther 2008;28:1140–1146. [DOI] [PubMed] [Google Scholar]
- 31.Aceves SS, Newbury RO, Dohil R, et al. Distinguishing eosinophilic esophagitis in pediatric patients: clinical, endoscopic, and histologic features of an emerging disorder. J Clin Gastroenterol 2007;41:252–256. [DOI] [PubMed] [Google Scholar]
- 32.Mulder DJ, Hurlbut DJ, Noble AJ, et al. Clinical features distinguish eosinophilic and reflux-induced esophagitis. J Pediatr Gastroenterol Nutr 2013;56:263–270. [DOI] [PubMed] [Google Scholar]
- 33.Ravi K, Talley NJ, Smyrk TC, et al. Low grade esophageal eosinophilia in adults: an unrecognized part of the spectrum of eosinophilic esophagitis? Dig Dis Sci 2011;56:1981–1986. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.