Published online before print July 17, 2003, 10.1148/radiol.2283021162
(Radiology 2003;228:609-613.)
© RSNA, 2003
Schatzki Ring, Statistically Reexamined1
John C. Pezzullo, PhD and
Ann M. Lewicki, MD, MPH
1 From the Departments of Clinical Pharmacology and Biostatistics (J.C.P.) and Radiology (A.M.L.), Georgetown University, Washington, DC. Received September 9, 2002; revision requested October 28; revision received November 13; accepted January 6, 2003. Address correspondence to J.C.P., 1711 Liggins Ave, Kissimmee, FL 34744-4091 (e-mail: johnp71@aol.com).
 |
ABSTRACT
|
|---|
In the article by Schatzki published in 1963, data about the lower esophageal ring relate ring diameter to presence of dysphagia. Statistical analysis of these measurements was performed to quantify conclusions of Schatzki and to extract additional information from the data. Ring diameters in 332 patients with and without dysphagia are described in a histogram in the original article of Schatzki. Data were evaluated with analysis of variance, logistic regression, and receiver operating characteristic (ROC) analysis to quantify the relationship between ring diameter and dysphagia. Follow-up information was available in 36 symptomatic and 30 asymptomatic patients of Schatzki. Logistic regression indicated that there was a highly significant difference in ring diameter between the asymptomatic group and patients with recurrent dysphagia (P < .001) but not in patients who had a single episode of dysphagia at presentation (P = .229). Odds ratio of 0.686 indicated that a 1-mm increase in ring diameter decreased the likelihood of dysphagia by 31%; conversely, a 1-mm decrease in ring diameter increased the likelihood of dysphagia by 46%. ROC curve of sensitivity and specificity of ring diameter and symptoms showed that the 20-mm cutoff of Schatzki had a 96% (104 of 108) sensitivity and a 58% (130 of 224) specificity, with area under the ROC curve of 0.888. Retrospective statistical analysis of original data of Schatzki validated his major conclusions about the data. Some important questions remain unanswered because of missing data in the study of Schatzki.
© RSNA, 2003
Index terms: Deglutition disorders, 71.74 • Esophagus, diseases, 71.74 • Esophagus, function, 71.74 • Esophagus, stenosis or obstruction, 71.74 • Statistical analysis
|
INTRODUCTION
|
|---|
In March 1953, Ingelfinger and Kramer (1) described a contractile ring in the distal esophagus, which they observed radiographically. The ring was the cause of dysphagia in their group of six patients, and the dysphagia was characteristic. Ingelfinger and Kramer also reported the endoscopic findings in these patients and presented the histologic features in one patient in whom the esophagus had been resected as treatment of the dysphagia.
In December of the same year, Schatzki and Gary (2) described a similar finding at barium esophagography performed in five patients, and their patients had similar symptoms. However, they also reported that this contractile ring could be observed in patients without dysphagia. Whether the ring caused symptoms seemed to be related to the diameter of the narrowing (2). Today, this contractile ring in the distal esophagus is usually called the Schatzki ring, and it is described in major textbooks that deal with diseases of the esophagus (3–6).
A decade after the initial description of the lower esophageal ring, Schatzki (7) published a report about a larger group of 332 patients with this ring, and he correlated the presence and severity of dysphagia to ring diameter. The measurements from this 1963 study have withstood the test of time and continue to be quoted in current textbooks and other references (3,4). We decided to reexamine the data of Schatzki with statistical methods that are now commonplace and that would be applied to such data today.
|
STATISTICAL EVALUATION
|
|---|
Data and Tests Used
The 1963 article of Schatzki (7) was the sole source of data used in this analysis. Most of the quantitative data are contained within figure I of that article (Fig 1), and the figure shows the ring diameter and dysphagia category (ie, asymptomatic, one episode, or multiple episodes) in 332 patients. On this histogram, each patient is represented as a small box, which is shaded to indicate the category, and is positioned according to the ring diameter. On this graph, diameters were read to the nearest millimeter and were used to calculate descriptive statistics, such as the following: Student t test, analysis of variance, and nonparametric tests; logistic regression; and receiver operating characteristic (ROC) curves.
Statistical Analysis
Limited 5-year follow-up data about changes in ring diameters and incidence of symptoms in patients with symptomatic and asymptomatic rings were obtained from tables I–III of the article by Schatzki (7). These changes were tested for significance by using the sign test, in which the number of "unchanged" patients (ie, those in whom the diameter of the rings was unchanged) is ignored and the numbers of "increased" and "decreased" patients (ie, those in whom the diameters were increased or decreased) are compared as one would compare the number of heads and tails in the flipping of a fair coin. Also applied was the Fisher exact test with a 2 x 2 cross-tabulation.
Schatzki did not address the problem of errors in his ring size measurements because of magnification from imaging techniques and patient size variability. He also did not control and/or correct for intra- and interreader variation in ring measurements.
All statistical calculations were performed by using a statistical software package (SPSS, version 11.0; SPSS, Chicago, Ill).
|
RESULTS
|
|---|
Descriptive Statistics
The distributions of ring diameters for the groups without and with symptoms (ie, classified according to frequency of dysphagia) are summarized in Table 1. The diameters are not normally distributed but appear to be sufficiently close to normality to permit the application of the common parametric statistical tests (eg, Student t test, analysis of variance).
Comparison of Diameters among Groups without and with Symptoms
Table 2 shows the results of comparisons of ring diameters among groups without and with symptoms. There was a highly significant difference in ring diameters between the asymptomatic group and the group with two or more episodes of dysphagia. The single-episode group was intermediate; there was no significant difference in ring diameter between the asymptomatic and single-episode groups, and there was only a borderline difference between the single-episode and recurrent-episode groups.
Diameter as Predictor of Dysphagia
Categories for the two symptomatic groups (ie, those with patients who had single or two or more episodes) were combined to produce a dichotomous classification for a logistic regression and ROC curve to assess the effectiveness of ring diameter as a predictor of dysphagia. The results of the logistic regression are shown in Table 3. As expected, the ring diameter was a highly significant predictor of dysphagia in this group of patients (P < .001).
The odds ratio of 0.686 indicates that every increase of 1.0 mm in ring diameter decreased the odds of a patient having dysphagia (ie, decreased the predicted ratio of symptomatic to asymptomatic patients) by about a third (decrease, 31%). Equivalently, it can be stated that every decrease of 1.0 mm in ring diameter increased the odds of a patient having dysphagia by about a half (increase, 46%).
The Hosmer-Lemeshow goodness-of-fit P value of .047 indicated a borderline inconsistency of the data to a logistic model. From a visual inspection of figure I in the 1963 article by Schatzki (7), it would appear that the major departure of these data from the logistic model would be found in the sudden and near complete disappearance of dysphagia as the ring diameter changed from 20 to 21 mm.
Sensitivity and Specificity
An ROC curve that shows the trade-off between sensitivity (ie, ability to detect symptomatic rings) and specificity (ie, absence of false-positive findings) as a function of ring diameter is shown in Figure 2. If the symptomatic and asymptomatic groups had identical distributions of ring diameters, the curve would be a diagonal straight line. The extent to which the ROC curve departs from the diagonal straight line and approaches the left and top axes is a measure of the effectiveness of ring diameter in the prediction of dysphagia. The curve is actually a curved scale that shows the sensitivity and specificity that would be obtained for any particular value of ring diameter. Interpolating the ring diameters from this curve, one can infer that use of a ring diameter of about 18 mm as a classification cutoff provides about 80% sensitivity (ie, use of this cutoff will lead to detection of 80% of symptomatic rings) and 80% specificity (ie, use of this cutoff produces 20% false-positive findings). Larger or smaller values will improve either sensitivity or specificity, respectively, at the expense of the other. With the ring diameter cutoff of 20 mm implied by Schatzki, 96% (104 of 108 [sensitivity]) of symptomatic rings will be detected, but the false-positive rate will be 42% (94 of 224), with specificity of 58% (130 of 224); conversely, with a ring diameter cutoff of 13 mm, no false-positive findings (specificity, 100%) are produced, but 63% (sensitivity, 37%) of symptomatic rings are missed.
View larger version (21K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2. ROC curve, derived from the original histogram of Schatzki, shows the sensitivity and specificity that result from any specific choice of cutoff for ring diameter as a predictor of dysphagia.
|
|
The mean area under the ROC curve is 0.888 ± 0.018 (standard error of the area under the curve) (95% CI: 0.852, 0.924), which is significantly different from the area under the ROC curve of 0.500 that would be obtained if there were no association at all between ring diameter and dysphagia (P < .001), and this value is consistent with that obtained previously in the group comparisons.
Because the selected sample of patients did not include all asymptomatic individuals encountered, it is not meaningful to construct a prediction equation from the results of the logistic regression, nor is it meaningful to quote positive or negative predictive values from these data.
Follow-up Data
The data in regard to 36 symptomatic and 30 asymptomatic patients are summarized in Table 4 and were analyzed as follows: First, we tested asymptomatic and symptomatic patients separately for whether there was a general tendency toward decreasing or increasing ring size during the follow-up period. The results determined with the sign test were that ring diameters decreased significantly more often than they increased in both the asymptomatic (P = .039) and symptomatic (P = .001) patients. Second, we tested whether the change in ring diameter during the follow-up was different for asymptomatic versus symptomatic patients. An analysis of the 2 x 2 cross-tabulation of presence or absence of symptoms according to ring size change (ie, decrease, no change, increase) with the Fisher exact test showed that there was no significant association between the presence of symptoms and the subsequent behavior of the ring (P = .25). Comparison of decreased and nondecreased (ie, unchanged or increased) diameters according to the Fisher exact test with a 2 x 2 cross-tabulation also indicated no significant association (P = .23).
We could not determine from data in the original article how initial ring size correlated with change in ring size over time; the article does not contain patient-level data with original and follow-up ring sizes. We also could not determine how magnification factors and intrareader variations affected the data and conclusions of Schatzki.
|
DISCUSSION
|
|---|
When Schatzki (7) published the data about the lower esophageal ring, the application of statistics to the examination of information from clinical practice was still a topic for debate (8,9). As the development of medicine progressed, it became clear that good clinical practice could be based on sound scientific facts and that conclusions from clinical observations needed to be scrutinized with statistics. Today, statistical analysis is an integral part of patient-oriented medical and patient-oriented radiologic research. Our major scientific journals have consultants responsible for statistics and sections devoted to this discipline (10,11). Current issues of the New England Journal of Medicine list their statistical consultants by name on the title page of the journal. In publications that hold scientists to a statistical standard, raw data that may be observational and descriptive have to be collected in a scientific reproducible way so that the conclusions drawn from such studies are valid. That requires that in some studies consideration be given to intrareader and interreader variability. The computer has made statistical calculations much easier, and user-friendly software has placed statistical analysis within the realm of scientifically oriented clinicians. As Proto (12) noted in a recent editorial, statisticians must be an integral part of a study if errors—errors such as selection of incorrect statistical tests—are to be avoided.
Analyzing the data of Schatzki retrospectively, we have been able to confirm his conclusions, in particular the association between ring diameter and the presence of dysphagia; we have also confirmed statistically the tendency for ring diameter to decrease rather than increase over time and to show that this happens in both asymptomatic and symptomatic patients. We were also able to quantify the qualitative conclusions of Schatzki by showing that a 1-mm decrease of ring diameter corresponds to a 46% increase in the likelihood that a patient has dysphagia. Results of our analysis also showed that an 18-mm cutoff for ring diameter provides a more even balance between sensitivity and specificity rather than the 20-mm cutoff that Schatzki chose. Schatzki chose that cutoff to denote that almost all patients who had a ring size greater than 20 mm were asymptomatic.
This exercise provides an illustration of the kind of additional quantitative information that can be extracted from a set of numbers that appear in an article published many years ago by applying the kinds of statistical analyses that are commonplace nowadays. This exercise also demonstrates how the quantitative analysis of data has changed during the past 40 years and how we now tend to draw more precisely quantifiable conclusions and look for more subtle effects than we did before.
Statistical validation of many of the data of Schatzki demonstrates that much can be done retrospectively, but we also must emphasize that integration of clinical data with statistics is much better performed prospectively. For instance, we could not determine predictive values, positive and negative, from this retrospective statistical analysis because critical data (ie, prevalence rate) were missing in the published report of Schatzki. Also missing were data that allow correction for magnification and reader variation. Therefore, we could not say how changes in ring diameter over time may be explained by these two factors. In a well-designed study, statistical input is essential from the outset so that all the necessary information will be collected. Such prospective planning also helps in the determination of the sample size to detect clinically important effects.
Schatzki chose 13 mm as a lower cutoff; in patients with diameters smaller than 13 mm, dysphagia was always present. Our analysis showed that this cutoff has a sensitivity of 37% and a specificity of 100%. In 1994, Rohrmann (13) stated that he believed that the critical diameter at which symptoms appear was 11 mm. He thought that this had been confirmed with findings of a prospective study in which the effects of a single esophageal dilatation on symptoms and changes in ring diameter were assessed (14).
Barium esophagography remains an effective tool for the detection of a lower esophageal ring. Schatzki and Gary (2) and Schatzki (7) emphasized the importance of good luminal distention for demonstration of the ring. More recently, Ott et al (15) showed that full-column barium esophagography is superior—more sensitive—than is air-contrast esophagography in the discovery of a lower esophageal ring. In a prospectively designed study, these investigators showed that solid barium esophagography is more sensitive than is endoscopy for the detection of this ring in the lower esophagus (15). When a patient has dysphagia at presentation, that seems characteristic of a lower esophageal ring; radiologists, therefore, should not hesitate to perform barium esophagography in the way Schatzki recommended and should not be deterred from doing so when no abnormalities are found at prior endoscopy.
Perhaps a well-designed study for the evaluation of the Schatzki ring is already in progress or planned; if so, many of the unanswered questions regarding this entity may soon be further elucidated.
|
ACKNOWLEDGMENTS
|
|---|
We are indebted to Rebecca L. Fisher, MLIS, Dahlgren Memorial Library, Georgetown University, for editorial assistance.
|
FOOTNOTES
|
|---|
Abbreviation: ROC = receiver operating characteristic
Author contributions: Guarantors of integrity of entire study, J.C.P., A.M.L.; study concepts, A.M.L.; study design, J.C.P., A.M.L.; literature research, A.M.L.; data acquisition, J.C.P.; data analysis/interpretation, J.C.P., A.M.L.; statistical analysis, J.C.P.; manuscript preparation, definition of intellectual content, editing, revision/review, and final version approval, J.C.P., A.M.L.
|
REFERENCES
|
|---|
- Ingelfinger FJ, Kramer P. Dysphagia produced by a contractile ring in the lower esophagus. Gastroenterology 1953; 23:419-430.[Medline]
- Schatzki R, Gary JE. Dysphagia due to a diaphragm-like narrowing in the lower esophagus ("lower esophageal ring"). Am J Roentgenol Radium Ther Nucl Med 1953; 70:911-922.[Medline]
- Levine MS. Abnormalities of the gastroesophageal junction. In: Gore RM, Levine MS, eds. Gastrointestinal radiology. 2nd ed. Philadelphia, Pa: Saunders, 2000; 484-493.
- Stewart ET, Dodds WJ. Radiology of the esophagus. In: Freeny PC, Stevenson GW, eds. Margulis and Burhenne’s alimentary tract radiology. 5th ed. St Louis, Mo: Mosby–Year Book, 1994; 192-263.
- Boyce HW. Hiatus hernia and peptic diseases of the esophagus. In: Sivak MV, eds. Gastroenterologic endoscopy. 2nd ed. Philadelphia, Pa: Saunders, 2000; 580-597.
- Tobin RW. Esophageal rings and webs. In: Castell DO, Richter JE, eds. The esophagus. 3rd ed. Philadelphia, Pa: Lippincott Williams & Wilkins, 1999; 295-300.
- Schatzki R. The lower esophageal ring: long term follow-up of symptomatic and asymptomatic rings. Am J Roentgenol Radium Ther Nucl Med 1963; 90:805-810.[Medline]
- Cimmino CV. Statistics and the physician (editorial). Radiology 1961; 76:128-129.
- Newell RR. Letter to the editor. Radiology 1961; 76:831.
- Proto AV. Radiology 2001—the upcoming year. Radiology 2001; 218:1-2.[Free Full Text]
- Proto AV. Radiology 2002—continued progress. Radiology 2002; 222:1-2.[Free Full Text]
- Proto AV. Radiology 2002—statistical concepts series. Radiology 2002; 225:317.[Free Full Text]
- Rohrmann CA, Jr. When is a Schatzki ring clinically significant, and what is the best maneuver to demonstrate it on barium swallow? does the abnormality progress if it is not treated? AJR Am J Roentgenol 1994; 163:215.[Medline]
- Eckhardt VF, Kanzler G, Willems D. Single dilation of symptomatic Schatzki rings. Dig Dis Sci 1992; 37:577-582.[CrossRef][Medline]
- Ott DJ, Chen YM, Wu WC, et al. Radiographic and endoscopic sensitivity in detecting lower esophageal mucosal ring. AJR Am J Roentgenol 1986; 147:261-265.[Abstract/Free Full Text]
TOP
ABSTRACT
INTRODUCTION
= multiple episodes, 
= single episode, 
= asymptomatic patients. (Reprinted, with permission, from reference