The abstract for this study was presented at the 2009 ACVIM Forum and Canadian Veterinary Medical Association Conference, Montreal, Quebec, Canada
A Multi-Institutional Study Evaluating the Diagnostic Utility of the Spec cPL™ and SNAP® cPL™ in Clinical Acute Pancreatitis in 84 Dogs
Version of Record online: 7 JUN 2012
Copyright © 2012 by the American College of Veterinary Internal Medicine
Journal of Veterinary Internal Medicine
Volume 26, Issue 4, pages 888–896, July-August 2012
How to Cite
McCord, K., Morley, P.S., Armstrong, J., Simpson, K., Rishniw, M., Forman, M.A., Biller, D., Parnell, N., Arnell, K., Hill, S., Avgeris, S., Gittelman, H., Moore, M., Hitt, M., Oswald, G., Marks, S., Burney, D. and Twedt, D. (2012), A Multi-Institutional Study Evaluating the Diagnostic Utility of the Spec cPL™ and SNAP® cPL™ in Clinical Acute Pancreatitis in 84 Dogs. Journal of Veterinary Internal Medicine, 26: 888–896. doi: 10.1111/j.1939-1676.2012.00951.x
- Issue online: 13 JUL 2012
- Version of Record online: 7 JUN 2012
- Manuscript Accepted: 24 APR 2012
- Manuscript Revised: 27 MAR 2012
- Manuscript Received: 9 JUL 2011
Pancreas-specific lipase is reported to aid in diagnosing acute pancreatitis (AP) in dogs but has not been rigorously evaluated clinically.
To describe variability of disease in dogs with suspected clinical AP, and to evaluate accuracy of 2 pancreatic-specific lipase immunoassays, Spec cPL (SPEC) and SNAP cPL (SNAP), in diagnosing clinical AP. We hypothesized that SPEC and SNAP provide better diagnostic accuracy than serum amylase or total lipase.
A total of 84 dogs; 27 without AP and 57 with clinical signs associated with AP.
Multicenter study. Dogs were prospectively enrolled based upon initial history and physical examination, then retrospectively classified into groups according to the likelihood of having clinical AP by a consensus of experts blinded to SPEC and SNAP results. Bayesian latent class analyses were used to estimate the diagnostic accuracy of SPEC and SNAP.
The estimates for test sensitivities and specificities, respectively, ranged between 91.5–94.1% and 71.1–77.5% for SNAP, 86.5–93.6% and 66.3–77.0% for SPEC (cutoff value of 200 μg/L), 71.7–77.8% and 80.5–88.0% for SPEC (cutoff value of 400 μg/L), and were 52.4–56.0% and 76.7–80.6% for amylase, and 43.4–53.6% and 89.3–92.5% for lipase.
Conclusions and Clinical Importance
SNAP and SPEC have higher sensitivity for diagnosing clinical AP than does measurement of serum amylase or lipase activity. A positive SPEC or SNAP has a good positive predictive value (PPV) in populations likely to have AP and a good negative predictive value (NPV) when there is low prevalence of disease.
Comparative Gastroenterology Society
controls (at enrollment)
canine pancreatic-specific lipase
not suspect acute pancreatitis (after enrollment)
negative predictive value
possible acute pancreatitis (at enrollment)
positive predictive value
suspect acute pancreatitis (after enrollment)
SNAP cPL Test1
Spec cPL with 200 μg/L cutoff
Spec cPL with 400 μg/L cutoff
Diagnosing AP in the dog can often be difficult. The pancreatic inflammation is variable, ranging from mild edematous changes to severe pancreatic necrosis, and the lesions can be multifocal or diffuse. The clinical signs associated with AP are also quite variable, the diagnostic serum tests amylase and lipase both have variable diagnostic accuracy, and imaging findings can be inconsistent for the definitive diagnosis. The accepted definitive diagnosis of AP remains histopathologic, but due to the invasiveness involved in tissue biopsy, it is not commonly performed. Consequently, the diagnosis of AP is generally a clinical diagnosis. Furthermore, the systemic manifestations of AP might be confused with other diseases making the diagnosis difficult.[1-7] Recently an enzyme-linked immunosorbent assay (ELISA) was developed and validated to quantitatively measure serum concentrations of canine PL. It is a sensitive indicator of exocrine pancreatic disease in the dog reflecting the release of PL into the serum as the result of pancreatic acinar cell damage. The clinical performance of the commercial test SPEC has not been extensively scrutinized. Also, a point-of-care commercial test SNAP is available as a colorimetric assay with analytical sensitivity of 200 μg/L for pancreatic-specific lipase and is considered to have an abnormal result if the assay pancreatic lipase test spot is equal to or more intense than the control test spot. To date it is not known how well these tests correlate with a clinical diagnosis of AP or if other organ or systemic metabolic conditions might also be associated with an elevated PL.,
To correctly interpret results of any diagnostic tests, the potential for false-positive and false-negative results must be understood. Evaluating the diagnostic accuracy of tests has traditionally been performed by referencing the results of a new test to those obtained by another test that is considered perfect (ie, a “gold-standard”). However, when perfect or near perfect tests are not available, then the reference-based analysis yields inherently biased results. In contrast to reference-based diagnostic test evaluation, newer methods for evaluating diagnostic tests have been developed which use advanced algebraic methods for estimation of the true, unbiased parameters for the study population regarding test sensitivity and specificity without presuming to know the true disease classification of individual study subjects. They also produce estimates of true prevalence of the condition being evaluated in the study population. Because these methods estimate classification probabilities that are hidden from direct observation, they are sometimes called latent class analysis. Bayesian latent class analysis of diagnostic tests is currently recognized as a best-practice method for evaluation of diagnostic test accuracy when a perfect method of diagnosis is unavailable for reference-based evaluations.[11, 12]
The aim of this study was twofold: first, to evaluate and describe the variability of disease in a cohort of dogs with suspected clinical AP using an expert panel that was blinded to the PL results of the population evaluated. The 2nd aim of the study was to then retrospectively evaluate the diagnostic accuracy of the commercially available PL tests, SPEC and SNAP, when performed on the patients at the time of hospital admission. We hypothesized that an elevated SPEC and an abnormal SNAP will identify dogs with AP with a better sensitivity and specificity than either serum amylase or lipase.
Materials and Methods
Dogs in which acute pancreatitis was considered a possible differential diagnosis based on presenting clinical signs and physical exam findings at the time of presentation (possible acute pancreatitis [PAP]) (n = 57), and dogs without acute pancreatitis having no clinical signs or physical findings associated with AP (control dogs at enrollment [CO]), (n = 27), were prospectively enrolled. A standard set of data was collected from all dogs, and each case was retrospectively reviewed. Consensus of an expert panel evaluating these data place the dogs from the 2 subgroups (PAP and CO), as either not having clinical AP (NAP) or being suspected of having clinical AP (SAP). Bayesian latent class analysis was used to estimate the sensitivity and specificity of SPEC, SNAP, amylase, and lipase for the clinical diagnosis of AP. This study was approved by the Institutional Animal Care and Use Committees of authors involved in the study.
Members of the CGS were recruited to participate in the study by the study coordinators by invitation, then solely on individual interest of the participants in contributing dogs to the study. Solicitation did not occur from the laboratory performing the tests, nor were the recruits paid for their participation. A total of 15 CGS members (either generalists or internal medicine specialists) from 14 institutions enrolled cases into the study. Study materials were prepared and distributed to each participant to ensure consistent and thorough data acquisition that included patient inclusion criteria, an owner consent form, a diagnostic check-list, and shipping materials for laboratory testing. Clinicians were asked to enroll any dogs in which AP was included in their initial differential diagnosis at the time of admission to the hospital but before diagnostic testing (PAP, n = 57). The inclusion criteria for PAP was based on historical complaints such as hyporexia or anorexia, vomiting, diarrhea, and lethargy, and physical exam abnormalities commonly reported with AP such as cranial abdominal pain and dehydration. Dogs with no clinical evidence of AP (CO, n = 27) were also enrolled if acute pancreatitis was not remotely suspected in the initial differential diagnosis with a lack of historical and physical exam findings consistent with the diagnosis of AP, including gastrointestinal signs. The majority of CO cases were presented for lower urinary tract evaluation.
Each dog was required on the day of admission to have a complete blood cell count, standard biochemical profile, urine analysis, abdominal ultrasound (performed either by the attending clinician or a radiologist), and a SNAP assay. Serum was also submitted at the same time for amylase, lipase, and SPEC to a commercial laboratory.1 That laboratory reports amylase normal ≤1240 U/L, lipase normal ≤750 U/L, and SPEC normal ≤200 μg/L and abnormal >400 μg/L. It also states SPEC results of <200 μg/L being not consistent with pancreatitis, ranges 200 – 400 μg/L as being possibly consistent with pancreatitis, and >400 μg/L being consistent with pancreatitis. For data analysis both the 200 μg/L (SPEC200) and 400 μg/L (SPEC400) cutoff values were evaluated. Clinicians were allowed to use in-house or referral diagnostic lab analyzers for the CBC, biochemical profiles, and urinalyses as long as verified normal reference intervals were reported with the results. Ultrasound images of the abdomen were recorded and printed or stored on compact disk. Patient progress reports including physical exam findings, diagnostic test results, written imaging interpretations, patient assessments, clinician diagnoses, treatments administered, and outcome for every 24-hour period were required. Any additional materials that could be provided by the clinicians' ancillary departments (such as radiology, clinical pathology, histopathology reports) were also requested. All study materials and samples were shipped to the study coordinators for compilation after resolution or hospital discharge of the case.
Each case was reviewed by the study coordinators and all objective information was compiled into a spreadsheet, including laboratory results, biopsy results (if present), and physical exam findings. The subjective information was copied verbatim from the clinicians' records and also organized for review. The ultrasound images were identified by patient study number and sent to a boarded veterinary radiologist (DB) for review and interpretation. No case information was given to the radiologist, including whether the dog was PAP or CO. The radiologist was asked to place each dog in one of the 5 following categories, based on a subjective assessment of the image or images of the pancreas present: (0) definitely not pancreatitis; (1) probably not pancreatitis; (2) possibly pancreatitis; (3) probably pancreatitis; (4) definitely pancreatitis. The category was chosen based on the presence or absence of findings such as a hypoechoic pancreatic parenchyma, peripancreatic hyperechogenicity, free abdominal fluid, pancreaticomegaly, gas in the duodenum, and a corrugated duodenum. The ultrasound scores were only used to aid in clinical scoring and were not evaluated statistically against the clinical scores because of the wide variability in the quality of the ultrasound images.
A panel of 4 boarded internists (DT, KS, MF, and JA) was presented summaries of each case (PAP and CO) in a Power Point format, blinded only to the SPEC and SNAP results. None of the panelists participated in compiling case summaries before the presentation. A dog's complete medical record, laboratory findings, and ultrasound interpretation were retrospectively reviewed by each panelist, with particular attention paid to the 1st day of admission when all initial diagnostic tests were performed; however, the dog's entire hospital course was also evaluated including the primary clinician's final diagnosis at discharge, if written in the record. Each dog was then independently assigned to 1 of 5 groups by each panelist. The cases were stratified into the following clinical disease categories: Group 0, definitely not AP where there is no clinicopathologic evidence of AP; or, there is convincingly normal ultrasound of the pancreas with a primary alternative primary disease present, not known to be associated with AP, to explain the clinicopathologic signs; Group 1, probably not AP where there is weak or inconsistent clinicopathologic or ultrasonographic evidence for AP with a convincing primary alternative disease present to explain clinical signs and the therapy would be directed at the nonpancreatic disease but recognized that there could be some secondary pancreatic involvement; Group 2, possibly AP where there is compatible clinical evidence of AP but inconsistent clinicopathologic or ultrasonographic evidence, and no cytohistopathologic evidence of AP; a clinical suspicion of AP is present without the presence of a convincing alternative disease to explain clinical signs and that AP therapy would be instituted; Group 3, probably AP where there is compatible clinicopathologic evidence with convincing ultrasonographic evidence of AP but cytohistopathologic evidence of AP is not provided or was not obtained for the patient and pancreatic therapy would be instituted; Group 4, definitely AP where there is clinicopathologic and cytohistopathologic evidence of AP, independent of ultrasonographic findings, that support pancreatic therapy would be instituted.
The panelists were presented and evaluated each case together in a group setting; however, the group number assigned to each case was individually determined by each panelist. If a consensus was reached (3 of the 4 panelists revealed the same group number for the case), then that group number was given to that case. If a consensus could not be reached, the case was excluded from the study. All cases (PAP and CO) were evaluated as independent cases, blinded to the panelists as suspect AP cases or control cases, and given group numbers accordingly.
In the absence of a diagnostic test with perfect accuracy, traditional reference-based methods for evaluating test sensitivity and specificity will yield biased estimates. Latent class analysis has been recognized as being superior to reference-based analysis as it does not require a perfect test for comparison and yields theoretically unbiased estimates of diagnostic sensitivity and specificity for the tests being evaluated as well as unbiased estimates of disease prevalence. Bayesian latent class analysis was used to characterize the sensitivity and specificity for concentrations of 5 different serum assays when used to diagnose AP (amylase, total lipase, SNAP for pancreatic lipase, and SPEC for pancreatic lipase using 2 different cutoff values [SPEC200 and SPEC400]). Standard methods for Bayesian latent class analysis for simultaneous evaluation of 3 diagnostic tests (2 of which have correlated results) in 2 subpopulations were used in these analyses and were performed using Markov-Chain Monte Carlo (MCMC) simulation modeling using freely available software (Win-BUGS 1.4.2, available at: http://www.mrc-bsu.cam.ac.uk/bugs/welcome.shtml; code available upon request).[11, 12] Two different methods were used for stratifying subjects into the subpopulations; one stratification split the subjects based on the history of vomiting (yes versus no) and the other method of stratification was based on the clinical scores produced by the consensus of experts, including not clinical AP (NAP) which were groups 0–1, and suspect clinical AP (SAP) which were groups 2–4. Two different stratification methods were used because amylase and total lipase values were considered as part of all available clinical information when the expert panel was determining consensus clinical scores. All models included serum amylase and total lipase, which were modeled as conditionally dependent tests. The serum PL assays (SNAP, SPEC200, and SPEC400) were investigated separately (ie, in parallel) as conditionally independent tests. Different modeling approaches were evaluated before settling on a final approach for analyzing these data (ie, which tests would be modeled as being conditionally independent). Some of the different permutations resulted in models that could not be solved mathematically (ie, they did not converge), and the final modeling approach was selected as it provided the most stable and precise estimates, and also controlled for the greatest amount of conditionally correlated results from different tests.
For the Bayesian analysis, prior assumptions about test sensitivities and specificities and disease prevalence were obtained from consensus opinion from the panel of experts (Table 1). The panel was not aware of results of any of the PL tests in the study population. Given the limited published information about the true values of these parameters, prior probability distributions were purposefully weighted so that they were weakly informative to avoid undue biasing of the posterior estimates. Covariance parameters for test dependency were modeled using uniform distributions for prior probabilities.
|Parameter||Beta Distribution (a, b)||Mode (%)||95% Probability Interval (%)|
|Serum Amylasea||Sensitivity||(3.48, 5.61)||35.0||11.6–69.8|
|Serum Lipaseb||Sensitivity||(3.26, 3.26)||50.0||15.8–84.2|
|SNAP cPL||Sensitivity||(2.66, 1.55)||75.0||19.1–96.0|
|Pancreatitis Vomiting Dogs||Prevalence||(1.55, 2.66)||25.0||4.0–80.9|
|Pancreatitis Nonvomiting Dogs||Prevalence||(1.30, 8.00)||4.0||0.8–41.0|
|Pancreatitis Clinical Score 2–4||Prevalence||(1.73, 1.24)||75.0||10.0–96.9|
|Pancreatitis Clinical Score 0–1||Prevalence||(1.24, 3.18)||10.0||1.8–72.0|
For each model, an initial burn-in of 5,000 iterations was discarded and node estimates were based on the subsequent 50,000 iterations generated in 3 separate chains. Convergence for each model was assessed by evaluation of convergence in the separate chains, and by use of widely differing initial values. A sensitivity analysis was run for each model by changing the prior beta distributions for all test variables to ensure that the results were repeatable. Average estimates of sensitivity and specificity for tests obtained from MCMC modeling were then used to calculate probable estimates for sensitivity and specificity that would be achieved through series interpretation when using tests in combination (ie, using both serum lipase and SPEC tests) to diagnose AP. Positive and negative predictive values were also calculated for serum lipase in series with SPEC200 and plotted across a range of true prevalence values.
A total of 84 dogs originally enrolled in the study were categorized as either SAP (n = 57) or NAP (n = 27). Each dog was retrospectively evaluated and assigned group scores of 0–4 as follows: group 0, n = 51 (n = 26 SAP; n = 25 NAP); group 1, n = 13 (n = 11 SAP; n = 2 NAP); group 2, n = 9 (all SAP); group 3, n = 8 (all SAP); group 4, n = 3 (all SAP). All dogs enrolled as CO were found to be NAP. Some dogs enrolled as PAP were determined to be NAP (groups 0 and 1). There was a high level of agreement in the assignment of cases to different groups by the evaluators (kappa 0.87) and none of the dogs was excluded from the study because the panelists were unable to reach a majority consensus. One SAP group 4 dog was excluded from data analysis for the SPEC because of an aberrant laboratory result (the mishap was not reported by the laboratory) and 2 SAP group 2 dogs were also excluded from the data analysis for the SPEC because samples were not submitted. The SNAP tests were excluded from analysis in 1 SAP group 0 dog and 1 group 2 dog because test results were not recorded by the clinicians.
Table 2 summarizes the clinical groups, the final diagnosis, and the results of SNAP (positive or negative) and SPEC (>200 or >400 μg/L). For SAP group 0 dogs, the most common diagnosis included gastrointestinal disease, gastrointestinal neoplasia, hepatobiliary disease, and endocrine disease (hypo- and hyperadrenocorticism). For SAP group 1 dogs, the most common diagnosis included gastrointestinal disease, renal disease, cardiac disease, and endocrine disease (diabetes mellitus and hyperadrenocorticism). SAP dogs in groups 2, 3, and 4 were thought to have primary AP but some also had concurrent diseases such as gastrointestinal disease, cardiac disease, and pancreatic neoplasia.
|“Positive” SNAP Result (n)||Groups 0–1 = primary disease (n); Groups 2–4 = clinical pancreatitis with concurrent disease (n)||SPEC > 200 and <400 (n)||SPEC > 400 (n)|
|Group 0 (n = 50)a||(15)||Gastrointestinal (6)Hepatopathies (2)Renal failure (1)Undetermined (0)Urolithiasis (5; all enrolled as known NAP cases)||(1)||(8)|
|Group 1 (n = 13)||(7)||Septic (2)Gastrointestinal (2)Renal failure (1)Acute hepatopathy (1)Exposure to furosemide, no other explanation of clinical signs (1)||(4)||(4)|
|Group 2 (n = 8)a||(8)||Gastrointestinal (3)Unexplained illness (2)Pancreatic neoplasia with diabetes mellitus (1)Abdominal mass (1)Renal failure (1)||(1)||(6)|
|Group 3 (n = 8)||(8)||Gastric ulcerations (1)Postpericardectomy (1)Disseminated intravascular coagulopathy (1)Only pancreatitis (3)No diagnosis listed (1)||(3)||(5)|
|Group 4 (n = 3)||(2)||Pancreatitic neoplasia (1)Possible azathioprine-induced pancreatitis[15-17] (1)Pancreatic cyst with cytologically confirmed inflammation (1)||(0)||(2)|
Median estimates from latent class analysis (Table 3) suggest that the true prevalence of pancreatitis among dogs assigned to groups 0–1 were 18.0–23.3% (limits of 95%PI for these estimates ranged from lower bounds of 5.8% to upper bounds of 41.8%) while median estimates of true prevalence for dogs assigned to groups 2–4 were 83.4–93.0% (min lower 95%PI bound = 54.9%, max upper 95% PI bound = 99.6%). Because serum amylase and total lipase concentrations were considered by the expert panel when assigning dogs to these clinical groups, history of vomiting was used to create a second dichotomization of the study population. The median estimates of true prevalence for AP among dogs without a history of vomiting ranged from 7.7 to 8.6% (min lower 95% PI bound = 0.5%, max upper 95%PI bound = 30.5%). The wide range for bounds of the 95% PI were likely attributable to the modest sample size.
|Parameter||Modelc||Iterationsd||Median (%)||95% Probability Interval (%)|
|True Pancreatitis Prevalence in Vomiting Dogs||V-1||150,000||47.8||24.7–73.2|
|True Pancreatitis Prevalence in Nonvomiting Dogs||V-1||150,000||8.6||0.6–30.5|
|True Pancreatitis Prevalence in Dogs with Clinical Group 2–4||CG-1||150,000||93.0||70.2–99.6|
|True Pancreatitis Prevalence in Dogs with Clinical Group 0–1||CG-1||150,000||19.8||6.5–37.0|
|Serum Amylasea Sensitivity||V-1||150,000||55.8||35.6–76.9|
|Serum Amylasea Specificity||V-1||150,000||78.8||64.9–91.4|
|Serum Lipaseb Sensitivity||V-1||150,000||47.4||27.3–74.5|
|Serum Lipaseb Specificity||V-1||150,000||89.3||77.5–97.1|
Systematic differences were not apparent in the estimates of test sensitivity and specificity that were obtained (Tables 3 and 4) when population data were dichotomized using clinical group scores (CG-1 to CG-3) versus when they were modeled based on a dichotomy of vomiting history (V-1 to V-3). This suggests that the estimates obtained from these 6 different models can be considered as a group. When considering the study hypothesis about differences between traditional serum tests (amylase and total lipase) and the PL assays (SNAP, SPEC200, and SPEC400), results indicated that the SPEC200 and SNAP tests were significantly more sensitive than the other assays (Tables 3 and 4). The averages of estimates for sensitivity of serum amylase and total lipase were 54 and 49%, respectively (Table 3), in contrast to the averages of estimates for sensitivity of the PL tests (SPEC200, SPEC400 and SNAP tests) which were all much higher (90, 75, and 93%, respectively; Table 4). The lower bounds of 95%PI for sensitivity of the SPEC200 and SNAP tests did not include median estimates for amylase and total lipase, nor did the upper bounds of the 95%PI for amylase and total lipase include median estimates for the PL tests.
|Test||Parameter||Model||Iterationse||Median (%)||95% Probability Interval (%)|
The average of test specificity estimates for serum amylase and total lipase were higher than sensitivity estimates for those tests (78 and 91%, respectively; Table 3), and were comparable to the average estimates for the PL tests (SPEC200 = 72%, SPEC400 = 78%, and SNAP = 74%; Table 4). Median estimates for the specificity estimates of the various tests were included within the 95%PI among the other tests, suggesting that the test specificity for the different assays were not significantly different. In a clinical setting, it seems likely that data from standard serum chemistry panels might be available in addition to results from a PL assay, and that both sources of data might be considered in the diagnostic process. The combined test sensitivity and specificity when interpreting the serum lipase in series with the SPEC200 test was estimated to be 44.4 and 97.2%, respectively (Fig 1). The positive predictive value (PPV) for the SPEC200 interpreted in series with total lipase is considerably higher than for the total lipase alone, SPEC200 alone, or SNAP, and this difference would be especially notable when the true prevalence of AP is low to moderate (eg, 2–60%; Fig 1). In contrast, the negative predictive value (NPV) for SPEC200 and SNAP would be much higher than for total lipase or SPEC200 interpreted in series with lipase at all true prevalences of AP (Fig 1).
The study confirms our hypothesis that the SPEC and SNAP assays for canine pancreatic-specific lipase had significantly higher sensitivity than did serum amylase and lipase for diagnosis of acute pancreatitis in dogs. Because of the high sensitivity of the PL assays, a negative test result will be highly predictive of the absence of disease in the general population of dogs where the prevalence of AP is low. The higher sensitivity of the PL assays also makes them more useful than traditional serum amylase and total lipase tests when used as screening tests; however, the modest specificity of all of the tests evaluated limits their value (as independent tests) in definitively confirming the true presence of the disease in the absence of other diagnostic information (when the pretest probability of disease is low to moderate). Interpreting these results of these tests in series dramatically improves the specificity and positive predictive value in these same circumstances (Fig 1).
In most cases a presumptive clinical diagnosis of AP is made based on historical, clinical, and laboratory findings. However, the clinical diagnosis of AP in the dog is often difficult to confirm. The most commonly accepted definitive diagnosis requires histopathology but this is rarely performed in most clinical situations. A clinical diagnosis is hampered by the variability in the severity of the pancreatic inflammation and the subjectivity of both clinical signs and physical examination findings. Acute vomiting associated with abdominal pain is thought to typify most cases having acute pancreatitis but these findings are neither consistent nor specific for the disease. Routine laboratory testing and imaging studies may also fail to confirm the diagnosis and serum amylase and lipase lack both sensitivity and specificity.[3, 5, 7] The PL assays have been validated to detect pancreatic-specific lipase in serum but have not been extensively scrutinized for their clinical accuracy in diagnosing AP.[8, 13]
It was the consensus of the panelists that most cases of AP could be diagnosed using the clinical information provided and after ruling out other diseases. The panelists also concurred that obtaining histology of the pancreas was unnecessary for most clinical cases; however, the lack of histological conformation was a limitation to this study and having knowledge of the histology may have changed the categorization of the cases. It was also believed by the panelists that not every clinical case could be diagnosed with 100% accuracy and there is no single diagnostic test or combination of tests that can yield 100% confidence in a diagnosis. The panel further suspected that the PL tests could also be abnormal as a result of secondary pancreatic damage from other disease conditions and that some of the cases could have also had an underlying occult pancreatitis affecting the PL tests. The purpose of this study, however, was to determine the diagnostic accuracy of the SPEC and SNAP of dogs with a clinical diagnosis of AP.
All dogs were required to have a pancreatic ultrasound examination. Because of the variability in practice protocol of the participating clinics, we could not require that ultrasound imaging be performed specifically by boarded radiologists, which is a shortcoming of the study. However, each ultrasound examination was reviewed and scored post hoc by a single boarded veterinary radiologist to help standardize this modality as much as possible for the panelists before case review. For this study the diagnostic accuracy of the imaging could not be determined because image acquisition was quite variable due to technique, machine characteristics, and operator experience. Most images submitted were still-framed and the pancreas could not be assessed from all directions or in its entirety. Occasionally, surrounding structures were also not visible. Due to these variables it became difficult to assign rigid quality control for interpretation purposes. Consequently, the boarded radiologist reviewed all the images, graded their diagnostic quality, and provided an ultrasound score for the dog on likelihood of AP. The panelists were free to either accept or reject the ultrasound interpretation and to weight the significance of the ultrasound findings in their ranking process. However, if an ultrasound examination was not performed on a dog, or if the pancreas was not evident in the images submitted, the case was not included for review by the panelists as it did not meet the inclusion criteria.
The Bayesian latent class analysis used in estimating the accuracy of these diagnostic tests is considered the best-practice method for evaluating diagnostic tests when there is not a perfect or near perfect test to use in reference comparisons. Clearly this is the case regarding antemortem diagnosis of AP in dogs. Unlike traditional reference-based evaluations, results of the latent class analysis can provide unbiased estimates of test sensitivity, specificity, and disease prevalence. However, latent class analysis requires relatively large sample sizes to obtain precision in estimates, which is likely the reason that 95%PI for the median estimates were relatively wide. The model diagnostics and similarity of median estimates obtained from the different models suggest that this was not caused by model instability. We chose to estimate test accuracy for the 3 PL tests using 3 separate models because of the strong likelihood that test results may be correlated (ie, the tests were not conditionally independent). Our preliminary investigations suggested that the most reliable model results were obtained when modeling the amylase and lipase as correlated tests and the PL assays as independent of these 2 tests. However, it seems unlikely that this issue would differentially affect the results for the canine PL assays.
Although the authors cannot rule out selection bias in favor of AP for recruitment of cases for this study, we feel doubtful that recruitment affected the statistical analysis. Any case in which the clinician could not rule out AP was allowed to be included, including dogs that had a low probability of disease at the time of admission. The inclusion of dogs with variable severity of clinical signs likely added diversity to the population evaluated. Furthermore, the majority of enrolled PAP cases were eventually placed into the groups 0 or 1 (NAP) which were not considered to have clinical AP.
Although SNAP has a colorimetric threshold of 200 μg/L, disparity was observed in a few cases between the results of the SNAP and SPEC tests. This variability may partially be due to the fact that SNAP requires a visual assessment by the operator, and serum PL levels close to the 200 μg/L threshold in which the SNAP assay may appear abnormal may have been read as either abnormal or normal. There were 4 cases in which the SNAP assay was abnormal but the SPEC value was below 200 μg/L, and 5 cases where the SNAP assay was normal but the SPEC value was above 200 μg/L. An explanation for this discrepancy cannot be determined from our study but future studies involving PL should consider the influence of concurrent diseases on the PL levels. Interestingly, all the disparate SNAP and SPEC tests involved cases in groups 0 and 1 which were not thought to have clinical AP. Previous studies indicate that renal function or hepatic disease may increase serum amylase and lipase levels. Evaluation of concurrent renal dysfunction and hepatopathy with elevated pancreatic lipase levels may help elucidate the effects of disease on these diagnostic tests.
In this study, the average of median estimates of sensitivity and specificity that were obtained in these analyses were 52.7 and 79.1%, respectively, for serum amylase, and similarly were 47.2 and 91.4%, respectively, for total serum lipase. The poor sensitivity of these tests has been reported previously in several studies and appears to be due to the variable physiologic factors that influence the levels of these enzymes in circulation.[2-5] Concurrent diseases such as compromised renal function due to renal tubular failure, decreased GFR from dehydration, and hepatic disease likely influence the levels of these enzymes in patients and may have played a role in the poor utility of these diagnostic tests. Three- to fourfold increases of serum amylase and twofold increases in total lipase have been reported to support a diagnosis of pancreatitis; however, this assumption has not been clinically scrutinized. We, therefore, did not calculate our statistics based on this vague assumption but simply included those outside the normal reference range for the laboratory used.
We were unable to determine the influence of concurrent diseases on the SPEC and SNAP values and it is very possible that other or concurrent disease may have influenced the SPEC and SNAP values. The premise of this study, however, was to evaluate a patient at presentation, and use the information acquired during the hospitalization period to support the presence or absence of clinical AP. Although this was a prospective study for the participating clinicians, our panelists used a retrospective approach in evaluating the cases based on information acquired after patient admittance. The purpose of this approach was to have an “omnipotent eye” to the patient's status at admittance in hopes of more accurately determining the likelihood of disease. We attempted to determine the most likely underlying condition for each dog and these conditions were highly variable. We were unable to make any associations between the cause and effect relationship of these concurrent or primary illnesses and pancreatitis. Additional studies may help elucidate this information but must be interpreted with caution as many of the dogs in groups 0 and 1 in our study had abnormal/positive SNAP and SPEC tests.
Limitations of this study exist. The ultrasound examinations were not ideal for critical evaluation. Still-framed images were submitted for evaluation and therefore were limited to two-dimensional pictures only. Complete organ evaluation (Doppler blood flow, cross-sectional size, peripancreatic tissue evaluation, etc) was therefore not possible. Ultrasound has been shown to be a modestly accurate technique for diagnosing pancreatitis with a sensitivity of up to 68%. It is possible that if thorough ultrasound examinations had been available to the radiologist who reviewed the images, a different clinical group number may have been assigned to some of the cases by the panelists. A better protocol for the use of ultrasound in future multi-institutional studies might include standardized evaluation techniques (ie, incorporating video for dimensional analysis, including Doppler flow images, and performing a systematic and consistent evaluation) to help make the use of this test more beneficial. An additional limitation of this study includes poor compliance with follow-up. Although we intended to recheck SPEC values within a week of discharge, very few dogs were returned for this diagnostic assessment. Including PL results after a designated time from the acute illness may be useful for following the progression and resolution of pancreatic disease. However, this has not been determined nor proven yet and we do not currently know the influence of other illnesses on the test results. Future studies should focus on addressing this issue.
In this study we conclude that the SNAP and SPEC assays have a higher sensitivity than measurement of serum amylase or total lipase for the diagnosis of clinical acute pancreatitis in dogs when taken in conjunction with a full clinical evaluation of a dog. If the SPEC is above 200 μg/L, the specificity was 77.0%, indicating that there are false-positives with this test. Because of the relationship between prevalence and predictive values, it should be recognized that using laboratory tests to aid in diagnosis of AP in situations when the index of suspicion is low could result in false-positive SPEC or SNAP results and lead to potentially inappropriate treatment (Fig 1). In contrast, if the SPEC200 or SNAP is positive/abnormal when clinical suspicion is moderate to high, there is a high PPV of true disease (Fig 1); but additional clinical and imaging studies should be used to support that diagnosis. Similarly, the predictive value of a negative test result is highest (ie, most likely to correctly predict true disease status) when the index of suspicion is low, and drops when dogs are showing clinical signs consistent with this disease. For dogs in the general population with a low clinical suspicion of AP, a negative result is very likely to be accurate. For dogs considered to most likely have AP as defined in this study, a negative SNAP or SPEC is likely to be accurate a majority of the time in ruling out AP; however, based on the NPV of these tests as shown in Figure 1, a small percentage of dogs having AP can have a false-negative test. In either situation additional appropriate clinical testing and other explanations for the clinical signs should always be explored.
The authors acknowledge and thank Dr Leif Lorentzen at IDEXX Laboratories, Inc for coordinating and organizing the study, Ms Erin Wood at IDEXX Laboratories, Inc for compiling and organizing study participant material, and Dr Jennifer Davis and Dr Fernando Leyva from Colorado State University for organizing case material. Funding for the laboratory testing, travel, and statistical analysis was provided by IDEXX Laboratories, Inc.
IDEXX Laboratories, Inc; North Grafton, MA
Determined using the average of sensitivity and specificity estimates reported in Table 3 (90 and 72%, respectively, for SPEC200; 93 and 74% for SNAP; 49 and 91% for serum lipase).
Interpreting test results in series (both individual tests need to be positive for result to be interpreted as positive versus any individual test is negative and the result is interpreted as negative). Assumed specificity and sensitivity of series interpretation = 97.2 and 44.4%, respectively.