Utility of two-view vs. three-view abdominal radiography in canines presenting with acute abdominal signs

Transcription

1 Received: 19 June 2017 Revised: 13 November 2017 Accepted: 19 December 2017 DOI: /vru ORIGINAL INVESTIGATION Utility of two-view vs. three-view abdominal radiography in canines presenting with acute abdominal signs Marina V. Mavromatis 1 Mauricio Solano 1 Michael Y. Thelen 2 1 Department Clinical Sciences, Tufts University Cummings School of Veterinary Medicine, North Grafton, MA SAS Institute Inc, Cary, NC Correspondence Mauricio Solano, Clinical Sciences Department, Cummings School of Veterinary Medicine, 200 Westboro Rd, North Grafton, MA Mauricio.Solano@tufts.edu Abstract There are clear differences in the normal radiographic appearance of the abdominal organs between a left lateral vs. a right lateral view. While a few veterinary academic institutions have transitioned to a three-view abdominal radiographic study protocol, obtaining only two views of the canine abdomen remains the current standard in veterinary medicine. In this combined retrospective and prospective, case-controlled study, 48 canine patients presenting with signs of acute abdomen were recruited. Four board-certified veterinary radiologists and four general practice veterinarians with greater than 3 years of experience in small animal practice were asked to determine if 10 predetermined findings were present within the set of images and if surgery was recommended based on those findings. Image readers were unaware of the clinical history. Threeview studies did not yield statistically significantly greater accuracy than two-view studies when evaluating all readers together. No statistically significant associations between the availability of the third view and increased accuracy or confidence were found in evaluations of general practitioners specifically. Evaluation of three-view radiographic examination, as compared to two-view examination, did not have perceived or statistically significantly increased diagnostic utility. Based on our findings, there is no statistically increased utility to justify a standard three-view abdominal radiographic examination over a two-view study for canines presenting with signs of acute abdomen. KEYWORDS abdomen, abdominal radiographs, dog, ileus, mechanical obstruction, X-ray 1 INTRODUCTION Abdominal radiography is a useful tool to aid in the diagnosis of patients presenting with clinical signs of acute abdominal disease (vomiting, inappetence, anorexia, diarrhea, constipation, and abdominal pain). While three-view radiography of the thorax has been accepted as having increased utility in diagnosis of disease, 1 the typical convention for imaging the canine abdomen involves a ventrodorsal projection and single orthogonal view. 2 A collection of seven case studies involving canine and feline patients with various gastrointestinal diseases demonstrated the utility of obtaining both right and left lateral abdominal radiographs to aid in obtaining an accurate diagnosis. 3 There are clear differences in the normal appearance of the abdominal organs between a left lateral vs. a right lateral view. This is in part due to the effect gravity has on the fluid and gas within the gastrointestinal tract. Gas within the gastrointestinal tract provides negative contrast for evaluating the mucosal and intraluminal characteristics of these structures and will locate in different areas of the stomach and intestines in the left lateral vs. right lateral projections. Variability in the appearance of the stomach and particularly the pylorus with regard to laterality of the acquired images has been demonstrated. 4,5 A recent publication demonstrated that the order in which radiographs are obtained also changes the typical distribution of gas within the gastrointestinal tract and hence would be implied to affect the ability to identify lesions and arrive at a correct diagnosis. 6 Differences in the appearance of organs, such as the spleen, have also been noted. The spleen is closer to the X-ray detector in a left lateral resulting in a more accurate depiction of the size and shape of the organ when compared to the right lateral view. While a few veterinary academic institutions have transitioned to a standard three-view abdominal radiographic study protocol (personal communication with veterinarians familiar with imaging sections at University of Florida, University of Georgia, University of Illinois, and others), obtaining only two views of the canine abdomen remains the most common abdominal protocol in veterinary practices. The authors Vet Radiol Ultrasound. 2018;59: wileyonlinelibrary.com/journal/vru c 2018 American College of Veterinary Radiology 381

2 382 MAVROMATIS ET AL. speculate this change from two- to three-view studies of the abdomen is partly based on available data, 3 6 but mostly on anecdotal information as a recent online search (Pubmed and Google Scholar using the key words comparison, accuracy, 2-view, two view, 3-view, three view, abdom*, radiograph*, canine, dog, left lateral, right lateral, thoracic, and thorax from 1963 to present) revealed no statistical analysis conducted to prove the increased accuracy or confidence of the reviewer in diagnosing gastrointestinal disease, or any other disease, with three-view abdominal radiography. If three-view abdominal radiography increases the diagnostic accuracy of detection of common gastrointestinal diseases, this may decrease the need for more expensive and less available alternative imaging techniques, such as computed tomography or ultrasound, or potentially avoidable surgical intervention to achieve a diagnosis. In this study, the authors hypothesize that (1) three-view radiography of the canine abdomen will have increased diagnostic accuracy for detection of three common gastrointestinal diseases (mechanical obstruction, abdominal masses, and pancreatitis) as compared to twoview radiography, (2) reviewers of three-view radiographic studies of the canine abdomen will have increased confidence in their assessments and recommendation for or against surgical intervention, and (3) there will be increased utility of evaluation of three-view radiographic studies as compared to two-view radiographic studies for general practitioners as compared to board-certified radiologists. 2 METHODS 2.1 Subject inclusion criteria The study was a combined retrospective and prospective, casecontrolled design. A total of 48 canine patients presenting to the Tufts Foster Hospital for Small Animals Emergency Service were prospectively and retrospectively recruited over an 18-month period from June 2015 to December Patients presented with signs of acute or acutely worsened gastrointestinal disease including vomiting, inappetence, anorexia, diarrhea, constipation, and abdominal pain; for which abdominal radiographs were desired. 2.2 Imaging techniques For all included dogs, three radiographic projections (ventrodorsal and right and left lateral; appropriate mas and kvp determined for each individual patient) of the abdomen were obtained (Canon Digital Radiography Systems, CXDI flat panel detector and image post processing software Sound Smart DR, Carlsbad, CA) with no more than 1 h difference between the first and last images obtained. Images were stored in Digital Imaging and Communications in Medicine (DICOM) format on a local Patient Archiving and Communication System (Carestream Vue Version 11.4, Rochester, NY) to be reviewed at a later date by Diplomates of the American College of Veterinary Radiology (at least 3 years in practice with board certification) and licensed general practitioners of veterinary medicine (3 years postgraduation) that were not affiliated with the hospital. Because all recruited cases were client owned patients within the hospital, images were evaluated first by attending residents and clinicians. In the case of patients for which surgery was the radiographic recommendation, patients had to have support of the radiographic findings via an ultrasonographic, surgical, or postmortem examination to be included in the study. In cases for which no surgical intervention was recommended based on radiography, patients were included if clinical signs of gastrointestinal disease resolved with medical management or supportive care only, without recurrence for at least 2 weeks following presentation. The study design was approved by the Tufts Cummings School of Veterinary Medicine Clinical Studies Review Committee. 2.3 Data recording techniques The number of data points necessary to show statistical significance was determined via power calculation. Assuming 50% sensitivity for all study questions, powering a study for an odds ratio increase of 1.75 when having the third view, we found that 272 data points had a statistical power of for logistic regression analysis. Therefore, approximately 34 patients needed to be recruited (power 0.9, alpha 0.05). At the end of the recruitment period, only 16 patients meeting all inclusion criteria outlined above had been recruited and the study design was amended to include retrospectively recruited patients that met all inclusion criteria and presented to the authors hospital over the same period. These patients were identified by performing a search for three-view abdominal studies within the Patient Archiving and Communication System and subsequently reviewing medical records for those cases with diagnostic studies to be sure all inclusion criteria were met. An additional 32 cases met the inclusion criteria this way, for a total of 48 cases. All decisions for subject inclusion or exclusion were made by two of the authors: M.V.M., a senior diagnostic imaging resident, and M.S., a Diplomate of the American College of Veterinary Radiology. The 48 cases were randomly assigned a number between 1 and 48 and subsequently assigned to two groups of 24 patients each. Group A consisted of two-view radiographic studies (ventrodorsal and right lateral) of the cases randomly assigned numbers 1 through 24 and three-view studies of the cases numbered 25 through 48; Group B consisted of the opposite configuration of two- and three-view examinations (Figure 1). The prospectively recruited cases were evenly split between the two groups with eight of these cases each in groups A and B. Two radiologists and two general practitioners were assigned to each group using a randomized block design for a total of eight readers. Each reviewer received images, based on their group assignment, on a digital versatile disc containing DICOM viewer software (efilm Lite V3, Merge-IBM, Milwaukee, WI). All reviewers were blinded to signalment and clinical signs. Reviewers were asked to determine if 10 predetermined findings were present within the set of images without knowledge of the clinical history. These 10 findings included: normal abdomen, focal small intestinal dilation, diffuse small intestinal dilation, small intestinal plication, gastrointestinal wall thickening, gastric dilation, hepatomegaly, loss of abdominal serosal detail, opaque gastrointestinal foreign

3 MAVROMATIS ET AL. 383 FIGURE 1 Right lateral (A), ventrodorsal (B), and left lateral (C) projections of the abdomen of a dog representing a typical case presented to the readers. Four readers were presented with images A and B while four readers were presented with images A, B, and C. Abnormalities for this case included focal small intestinal dilation of several loops of jejunum (asterisk), jejunal bunching (arrow heads), and a metallic gastric foreign body (curved arrow). The loss of detail was considered normal for the age of the patient. Final diagnosis: mechanical obstruction due to a linear foreign body and a gastric foreign body (wire). Removal of the offending material was performed at surgery material, and mass effect. Answers were submitted based on the degree of certainty or confidence that the finding was present using a Likert scale (1, I am certain this finding is not present; 2, I am somewhat certain this finding is not present; 3, I cannot determine whether this finding is present or not; 4, I am somewhat certain this finding is present; and 5, I am certain this finding is present). Reviewers were given an opportunity to submit additional comments separately and finally asked to determine if surgery was indicated for the patient using a similar Likert scale of certainty. Time to evaluate the images was not limited and reviewers were asked to enter their responses for each individual case in an online survey as each case was reviewed. In order to develop the standard for statistical analysis, two of the authors (a senior diagnostic imaging resident (M.V.M.) and Diplomate of the American College of Veterinary Radiology (M.S.) with greater than 15 years experience) evaluated the three-view radiographic studies and finalized records from the time of presentation for each case to determine if an abnormality was present or absent. Image interpretation in DICOM format were performed using medical grade monitors (EZIO Rediforce RX440, Cypress, CA) on a Patient Archiving and Communication System software (Carestream Vue Version 11.4, Rochester, NY). If they disagreed, a consensus had to be reached for each finding. 2.4 Data analysis techniques To determine whether associations between each predictor and response combination were statistically significant, survey data were analyzed using the Cochran Mantel Haenszel) test. A follow up Cochran Mantel Haenszel analysis was also performed to determine whether radiologists match the final consensuses more so than general practitioners. To ascertain whether reviewing three-view abdominal studies statistically significantly increased diagnostic accuracy above two-view studies, a logistic regression model was implemented. The second portion of the statistical analysis examined the potential

4 384 MAVROMATIS ET AL. increase in diagnostic confidence under three-view studies. An increase in diagnostic confidence is defined as trending towards Likert items 1 or 5 in the online survey. We investigated whether the third view increased confidence in the diagnostic finding and utilized the Cochran Mantel Haenszel approach to stratify by general practitioner and radiologist. Lastly, we examined whether general practitioners, specifically, were more confident when reviewing a three-view study than a two-view study using Fisher's exact test. All calculations were performed by one of the authors (M.Y.T.) using standard statistical software (SAS 9.4, Cary, NC; R Version 3.2.3, R Foundation for Statistical Computing, Vienna, Austria). Each data point was analyzed as an independent observation. For all analyses, values of P < 0.05 were considered significant. TABLE 2 Estimates for the four finding areas where radiology diplomates were found to match the final consensus statistically significantly more often than general practitioners Proportion estimate Odds ratio estimate Diffuse small intestinal dilation [ ] Focal small intestinal dilation [ ] Opaque gastrointestinal foreign material [ ] Small intestinal plication [ ] Notes. Radiology diplomates were found to match the general consensus more statistically significantly in these four finding areas only. Odds ratio estimates and 95% Wald confidence intervals are provided. 3 RESULTS Four cases were excluded from statistical analysis as a consensus could not be reached on at least one finding. Using the Shieh-O'Brien approximation, the probability of detecting a true threefold increase in the odds ratio between two- and three-view diagnostic accuracy and confidence was 84.3% with 44 patients and eight reviewers, or 352 total data points. Review of the online survey data showed six data points were unusable due to nonresponse and 14 data points were unusable due to undecipherable patient information. One duplicate entry was found and was reconciled by retaining the latest submitted record for the patient/reader combination. Test blocking by diplomate or general practitioner did not find that three-view studies yielded statistically significantly greater accuracy, than two-view studies (Table 1). The radiologists matched the final consensus more than general practitioners in four of 11 finding areas; areas under which statistical significance was found are listed in Table 2, along with odds ratio estimates calculated using logistic regression. In addition, we focused specifically on general practitioners to see if the availability of TABLE 1 Cochran Mantel Haenszel test statistics and P-values for comparing three-view studies to two-view studies Statistic P Diffuse small intestinal dilation Focal small intestinal dilation Gastric distension Gastric or intestinal wall thickening Hepatomegaly Loss of peritoneal detail Mass effect resulting in displacement of any intra-abdominal organs Normal abdomen Opaque gastrointestinal foreign material Small intestinal plication Surgical intervention indicated TABLE 3 Diffuse small intestinal dilation data analyzed using the Cochran Mantel Haenszel test to determine whether a third view increases diagnostic confidence in test cases Study group Confidence Two views Three views Diplomates Low High % High 72.4% 84.9% Non-Diplomates Low High %High 84.0% 63.0% Notes. High Confidence is defined as the evaluator being certain that a finding is present or not present (Likert items 1 or 5). Low Confidence is defined as the evaluator being somewhat certain or not able to determine the same (Likert items 2, 3, or 4). The diffuse small intestinal dilation question is the only study question where the Cochran-Mantel-Haenszel test shows a nonzero correlation between having the third view and diagnostic confidence, with P = a third-view increased accuracy. The chi square/fisher's exact test procedure did not show any statistically significant associations between the availability of the third-view and increased accuracy for general practitioners. The only finding area that showed a statistically significant increase in diagnostic confidence was Diffuse Small Intestinal Dilation (P = ), for which there was only one patient with this finding included in the study. Summary measures compared are shown in Table 3. Using Fisher's exact test, the authors did not find any finding areas where general practitioners were statistically significantly more confident if a third-view was available in their review. On the contrary, for the Diffuse Small Intestinal Dilation finding, alone, a logistic regression model showed a statistically significantly negative effect for the third-view with an estimated odds ratio of with a 95% confidence interval of and 0.685; the point estimate for the probability of having increased confidence for the Diffuse Small Intestinal Dilation finding if you were a general practitioner was 24.5%. Again, only a single patient was determined to have this finding based on consensus review.

5 MAVROMATIS ET AL DISCUSSION The goal of this study was to determine if having three views of the canine abdomen inherently resulted in increased accuracy and/or confidence in reviewing the images of patients presenting with acute abdominal signs and determining if surgical intervention was necessary for those patients. A second goal was to attempt to justify obtaining three views as a standard in veterinary medicine. Evaluation of three-view radiographic examinations, as compared to two-view examinations, of the abdomen did not have perceived or statistically significantly increased diagnostic utility in correctly identifying key abnormalities within the images or determining whether surgical intervention was needed. This was true for both boarded radiologists and general practitioners. This is particularly surprising as most subjects presented here had gastrointestinal foreign material with or without mechanical ileus and a prior collection of case reports demonstrated the increased diagnostic utility of the left lateral view, in addition to the right lateral view, for a selection of similar cases. 3 Authors of a recent study comparing the distribution of gas within the duodenum based on initial right- vs. left- lateral recumbency speculated that the new twoview standard abdominal radiographic study could include left lateral and ventrodorsal views for cases where gastrointestinal obstruction is suspected or needs to be ruled out. 6 Our findings would support this as it seems any previously inferred increased diagnostic accuracy for such cases is more likely related to having obtained a left lateral view at all (i.e., as part of a two-view examination), rather than in addition to initial right lateral and ventrodorsal views. We also hypothesized that reading three-view studies would result in reviewers overall responding to the survey with their findings more confidently as they would have more imaging information for each case. If this alone had been statistically significant, it may have justified obtaining three-view radiographic studies of the abdomen by decreasing the need for subsequent diagnostics that may not be necessary or readily available. This was only determined to be true in identification of diffuse small intestinal dilation, which was present in a single patient. Each reviewer saw every case only once as a two- or threeview examination. It is unknown if there would have been a measurable increase in confidence level if reviewers had been asked to determine their findings for two-view studies first and then subsequently been given a third-view or seen the same case with all three views available to them at a later date. A crossover study design was not chosen as the authors were concerned this may have confounded the data as some reviewers would likely recognize the case upon seeing it a second time and have a preconceived determination of whether findings were present or not. The goal was to determine if in isolation, having three views of the canine abdomen inherently resulted in increased confidence in reviewing the images and determining if surgical intervention was necessary. Thus, it was beyond the scope of the present paper for the study design to include analyzing crossover data. There are several limitations to this study. First, the reviewers were not given any information on any of the cases, including the presenting complaint or reason for obtaining abdominal radiographs. This is not the case in most clinical settings where knowledge of the patient's history and examination findings may aid in the evaluation of images and the confidence level of the evaluator in any findings they may identify. The goal of this study was to compare the performance of three-view studies as compared to two-view studies of the abdomen and not to determine how accurate the evaluators were overall. Second, as part of the inclusion criteria, all patients had to present with clinical signs of abdominal disease leading the primary clinician to order routine two-view abdominal radiographs. In the authors hospital, radiographs are not required prior to performing abdominal ultrasonography. This led to a strong selection bias for cases with foreign bodies with or without mechanical ileus as compared to cases that, for example, had palpable abdominal masses and did not have radiographs requested at all. Hence, to determine if a three-view examination of the canine abdomen has increased diagnostic utility for patients with abdominal masses or surgical disease other than obstructive foreign bodies, a study looking at a larger population of patients with more variable diagnoses is needed. Third, it was not possible to randomize the two-view studies being reviewed due to the study design and number of patients recruited. The authors already needed to retrospectively recruit cases in order to meet the power to discriminate between the utility of a two-view study vs. three-view study. Many more patients would have needed to be recruited in order to have the power to additionally randomize the two-view study being evaluated (one of three possible combinations). Alternatively, a crossover study design could have been selected to meet power with the number of cases available; the reasoning behind not choosing a crossover design has already been stated. The decision to define the two-view studies as right lateral and ventrodorsal views was made based on the practice within our hospital where twoview studies standardly are obtained with right lateral and ventrodorsal views, the expert impression that many other hospitals also practice this way, and information published in a diagnostic imaging textbook. 2 Lastly, reviewers were not given specific guidelines on how to determine if specific findings were present or not, as the authors wanted them to analyze images as they would normally in their practice. This may have introduced cognitive response bias as individuals may have interpreted the description of the findings differently in addition to using different criteria for determining if a finding was present or not. In conclusion, findings from the current study indicated that there is no statistically increased utility (for diagnosis or confidence) to justify a three-view abdominal radiographic examination over a two-view study for canines presenting with signs of gastrointestinal disease. Radiologists and general practitioners alike were not more accurate or confident with a third view added to a standard two-view abdominal radiographic examination. LIST OF AUTHOR CONTRIBUTIONS Category 1 (a) Conception and Design: Mavromatis MV, Thelen MY, Solano M (b) Acquisition of Data: Mavromatis MV, Solano M, Thelen MY (c) Analysis and Interpretation of Data: Mavromatis MV, Thelen MY

6 386 MAVROMATIS ET AL. Category 2 (a) Drafting the Article: Mavromatis MV, Solano M, Thelen MY (b) Revising Article for Intellectual Content: Mavromatis MV, Solano M Category 3 (a) Final Approval of the Completed Article: Mavromatis MV, Solano M, Thelen MY ACKNOWLEDGMENTS The authors would like to acknowledge the image reading expertise provided by Drs M. Lora-Michelis, M. Holowinski, J. Hobbs, L. Schwarz, J. Heitzman, K. Sizer, K. Payton, and T. Gaffrey. The authors would also like to acknowledge the Fund for Companion Animal Health for their assistance in funding this project. REFERENCES 1. Ober CP, Barber D. Comparison of two- vs. three-view thoracic radiographic studies on conspicuity of structured interstitial patterns in dogs. Vet Radiol Ultrasound. 2006;47: Graham JP, Berry CR, Thrall DE. Textbook of Veterinary Diagnostic Radiology. 5th ed. St Louis, Missouri: Saunders Elsevier; 2007: Armbrust LJ, Biller DS, Hoskinson JJ. Case examples demonstrating the clinical utility of obtaining both right and left lateral abdominal radiographs in small animals. J Am Anim Hosp Assoc. 2000;36: Love NE. The appearance of the canine pyloric region in right versus left lateral recumbent radiographs. Vet Radiol Ultrasound. 1993;34: Grandage J. The radiological appearance of stomach gas in the dog. Aust Vet J. 1974;50: VanderHart D, Berry CR. Initial influence of right versus left lateral recumbency on the radiographic finding of duodenal gas on subsequent survey ventrodorsal projections of the canine abdomen. Vet Radiol Ultrasound. 2015;56: How to cite this article: Mavromatis MV, Solano M, Thelen MY. Utility of two-view vs. three-view abdominal radiography in canines presenting with acute abdominal signs. VetRadiolUltrasound. 2018;59: