New achievement unlocked!
New achievement unlocked!
December 2024
In today’s VETgirl online veterinary CE podcast, we’re going to talk about one of my favorite breeds, the super sweet Cavalier King Charles Spaniel (what we’ll Cavaliers from now on, since it’s a mouthful!). Unfortunately, we all know that this breed has horrible myxomatous mitral valve disease. So, if you see Cavaliers in your clinic, when should you decide to put these dogs on heart medications? After all, we know that the initiation of pimobendan can significantly delay the onset of congestive heart failure in dogs with stage B2 myxomatous mitral valve disease (or “mitral valve disease”), and that detection of the transition from stage B1 to B2 in this population of dogs is important so that therapy can be initiated expediently.
As of the 2019 ACVIM consensus statement on this disease, stages B1 and B2 have been modified to match the parameters used in the 2016 EPIC trial, which provided the aforementioned evidence for the benefit of pimobendan administration in stage B2. Stage B1 includes dogs with no cardiomegaly, or very mild cardiomegaly that does not meet the EPIC criteria. B2 includes dogs with cardiomegaly that meets or exceeds the EPIC criteria (in patients who remain asymptomatic for their heart disease). The gold standard for determination of disease stage is echocardiography.
As echocardiography may not always be readily available in the clinical setting, a useful prediction model incorporating other parameters such as history, physical examination, radiographs, and blood biomarkers would be ideal to enable practitioners to confidently stage dogs with mitral valve disease and begin therapy where appropriate. This would be helpful especially for our pet owners living in areas where Cardiologists are scarce, are far away, or have a lengthy wait time for new clients, and also this option may not be feasible for all pet owners. A recent model demonstrated validity to the concept of using these other tools to stage mitral valve disease, but only a small number of dogs in that study had thoracic radiographs performed, which is one of the primary tools available to most practitioners. Prediction models such as this, operate under the premise that a combination of variables will be more predictive than any one single variable alone. This is important when considering parameters like vertebral heart size (VHS) or N-terminal pro-B-type natriuretic peptide (NT-proBNP), which have been demonstrated to have significant variability among breeds, even within the normal range.
So, Wesselowski et al wanted to evaluate this concept in a study entitled “Use of physical examination, electrocardiography, radiography, and biomarkers to predict echocardiographic stage B2 myxomatous mitral valve disease in preclinical Cavalier King Charles Spaniels.” This study aimed to 1) develop breed-specific cutoffs for numerous cardiac tests among Cavaliers and 2) develop predictive models utilizing these cardiac tests in combination to identify echocardiographic stage B2 mitral valve disease in this breed.
This was a cross-sectional study that prospectively recruited healthy Cavaliers over a three-year period from four sites. Dogs were excluded if they were < 1 year of age, had concurrent systemic illness that could impact the cardiovascular system, or were diagnosed with any form of cardiac disease other than mitral valve disease. Elevated creatinine or symmetric dimethylarginine (SDMA) was also cause for exclusion. Mild elevations (defined as < 20% increase above the normal reference range) of blood urea nitrogen (BUN) were acceptable in isolation.
Owners were asked to answer five questions as to whether their dog had a cough, difficulty breathing, abnormal activity level, history of fainting spells or abnormal appetite. If owners answered yes, additional information was solicited. Owners also filled out a functional evaluation of cardiac health (FETCH) questionnaire. The Cavaliers then underwent the following diagnostics:
1) A physical exam to assess for presence/absence of murmurs and/or arrhythmias, and murmur grading when present
2) A right lateral thoracic radiograph (to obtain VHS and Vertebral Left Atrial Size, or VLAS)
3) A 30-second 6-lead ECG at 100 mm/s with digital analysis capability for precise measurement of the typical P-QRS-T intervals as well as mean heart rate, vasovagal tonus index, and a novel measurement named summed P-wave and QRS duration (P + QRS).
4) Doppler blood pressure, although patient acclimatization period was not standardized
5) Standard echocardiogram for classification of disease stage (A, B1, or B2), and
6) Venipuncture for NT-proBNP, cardiac troponin I (cTnI), and renal panel analysis (renal panel included BUN, creatinine, SDMA, albumin, phosphorus, potassium).
Now if you love statistics, this paper is your jam! We tend not to focus on stats too much during our VETgirl podcasts, so we don’t all fall asleep. But if you are like me and Justine and Garret, and find yourself muddling through the stats analysis section of manuscripts, I’ll try to keep the analysis as simplified and understandable as possible, sticking just to the parameters that are of most value or interest. Remember the goals here: 1) Establish Cavalier-specific cutoffs for the cardiac screening tests performed and 2) evaluate for viable predictive models to correlate to echocardiographic disease staging. Predictive models would incorporate two or more parameters, so a large variety of combinations exist and the authors explored this, including comparison of two-parameter, three-parameter, and four-parameter models. Individual variables were assessed for high sensitivity, optimized combined sensitivity and specificity, and high specificity. The prediction models were created via a method whereby the model begins with no variables, and the most significant variables (based on P < 0.05) are then added sequentially until a pre-specified stopping rule is reached or all of the variables make it in. Individual variables were evaluated in clusters as one of four categories: 1) history/physical examination 2) blood tests 3) ECG findings and 4) radiographic measurements. Within these clusters, the candidate variables determined to be statistically significant were: history of cough, murmur grade, NT-proBNP, SDMA, P+QRS duration, and VHS, so these were then input into the models. Once the model analysis had been completed, they were re-run with exclusion of dogs in stages A and B1 that lacked heart murmurs to assess the impact of this parameter. Finally, a decision tree analysis was used to develop a single three-test tree algorithm using murmur, VHS, and NT-proBNP. For this latter portion, murmurs were subdivided into soft (grades 0-2) and louder (grades 3-6).
Ok, so what’d they find in this study? The study population consisted of 226 Cavaliers. It’s worth mentioning that 15.5% of dogs had a history of cough, and 5.9% had a history of owner-reported labored breathing, but when the latter group was questioned further there generally was either an alternative medical etiology (e.g., laryngeal paralysis) or owner misinterpretation of things like panting or snoring.
Before digging into the prediction model results, here are some of the highlights of the individual variable analysis. Interestingly, 27 Cavaliers in stage B1 did not have heart murmurs. One of our VETgirl cardiologists says he sees this regularly as well with highly proactive breeders or owners who bring young Cavaliers in for screening – a good number of them already have a little mitral valve disease on echocardiogram despite not yet having an audible murmur. By comparison, 91% of stage B2 Cavaliers had a murmur grade 4/6 or higher. That’s a notable finding. Although murmur grade is a crude indicator of disease severity in totality, it’s clear that if the murmur is a grade 4 or louder, we should really be encouraging even the most reluctant of owners to perform cardiac diagnostics. While renal panel analytes remained similar across groups, telling us that renal disease was not likely to be contributing to changed in the cardiac biomarkers, they found that NT-proBNP and cardiac troponin I increased with stage of mitral valve disease. Radiographic VHS and VLAS increased with stage as expected. NT-proBNP, VHS, VLAS, VHS + VLAS all demonstrated good discriminatory ability between stage B2 and earlier stages. Among ECG findings, only the novel parameter P + QRS duration showed this ability. Of the individual variables, area under the curve was best for NT-proBNP at 0.855. Remember, the highest possible area under the curve is 1.0 – so 0.855 is nothing to sneeze at! Radiographically, VHS proved to be slightly better than VLAS, while VHS + VLAS was slightly better than VHS alone. However, significant overlap occurred among the confidence intervals. The areas under the curve for both NT-proBNP and VHS in this study were also superior than previously reported non-breed specific data, suggesting that – spoiler alert!- using breed-specific data is likely to be more accurate in predicting stage of disease.
In terms of specific cutoffs and their various degrees of sensitivity and specificity for all variables, I would refer you to Table 3 for the complete data. To highlight a few important items, an NT-proBNP of 670 pmol/L or higher provided the highest sensitivity for detecting B2, at 91%, but the specificity was low at 50.8%, which makes sense as the upper end of the reference interval for dogs as a whole is 900 pmol/L. An NT-proBNP cutoff of 1138 provided high specificity at 90.1% but at the expense of a low sensitivity at 68.9%. A VHS of 10.6 or greater carried similar results to the NTproBNP of 670, while a VHS of 11.0 carried a sensitivity and specificity of 80.0% and 74.0%, respectively. A cutoff of 11.5 provided excellent specificity at 90.6% but at the expense of a drop in sensitivity to 51.1%. The upper normal reference range of VHS for Cavaliers has been reported to be slightly higher than the “all-dog” upper normal of 10.7, so again these results do make sense with what we already know.
Let’s talk about the models. We mentioned earlier that the individual parameters that retained significance from the four predetermined “clusters” were as follows: history of cough, murmur grade, NT-proBNP, SDMA, P+QRS duration, and VHS. You’ll note that cardiac troponin I and VLAS did not make that cut. When the models were created, within each type of model – meaning a two-variable, three-variable, or four-variable model – all of the various permutations were analyzed and compared within and between model types. For example, they compared two-variable models that used physical examination and radiographs, vs NT-proBNP and radiographs, vs physical examination and P + QRS duration on ECG, and so on and so forth. Within each type of model, there was no significant difference in predictability of disease stage regardless of the combination of parameters used, and the predictabilities were all quite good. The lowest predictability for any combination was the two-variable model combining P + QRS on ECG with radiographic VHS, and even that carried an area under the curve of 0.895, which was still higher than the highest area under the curve for any single variable, which as we said before was NT-proBNP at 0.855. Not surprisingly, the model with the highest predictability area under the curve was the four variable model whereby each variable came from one of the four previously mentioned “cluster” sets. Specifically, a four variable model using NT-proBNP, radiographic VHS, P+QRS duration, and murmur grade produced an area under the curve of 0.971. Impressive!
I do want to take a moment to provide clarity from the fine print on the novel ECG measurement of P+QRS duration used in this study. The investigators used ECGs run at 100 mm/s, which is faster than many entry-level machines are capable of running, and they measured the intervals using digital software which would be inherently more precise than a hand measurement off printed paper. This doesn’t invalidate the results, but more speaks to the potentially low practicality of doing this measurement regularly and accurately in practice compared with a radiographic VHS or NT-proBNP.
Lastly, remember that the authors intended to create a decision tree algorithm for us using 3 test variables, specifically heart murmur, VHS, and NT-proBNP. If you have time, pull the paper so you check out Figure 2 in the manuscript to fully appreciate this algorithm. Heart murmur is the first decision step in the tree, followed by radiographic VHS and NT-proBNP, in sequence. Step 1 provided high confidence in a low likelihood of stage B2 for those dogs with murmurs less than a grade 2, as we mentioned earlier. As one proceeds stepwise through the tree, the confidence in classifying a stage B2 increases as the appropriate cutoff in each step is attained. Importantly, the authors identified a definitive gray zone in Cavaliers with a VHS between 10.6-11.4 as correlated with echocardiography which has the potential to create notable false positives or negatives if VHS is used in isolation… this also means that it’s important we are measuring the VHS in these patients!
Wow – that was intense. With that, what can we take away from this VETgirl podcast? The study results support the following
basic premises:
1) Using breed-specific data, where available, for cardiac screening tests is likely more accurate than all-dog data
2) In situations where echocardiography is likely to be pursued, using a single screening test such as NT-proBNP or VHS at its highest sensitivity cutoff is preferred, as the eventual echocardiogram data will allow for definitive classification and the removal of false positives provided by that variable
3) In situations where echocardiography is not available or unlikely to be pursued, using a single screening test such as NT-proBNP or VHS at its highest specificity cutoff is preferred, to reduce false positives that will result in prescribing pimobendan to dogs where it is not yet appropriate
4) In situations where echocardiography is not available or unlikely to be pursued, using multiple variables in a prediction model or decision-tree algorithm is optimal vs a single variable, and although a four-variable model is optimal, the predictability for three- and two-variable models was also relatively reliable. Simply adding NT-proBNP to your radiographic VHS, for example, will significantly increase your likelihood of accuracy in predicting stage of disease
5) The prediction models clearly indicate that the murmur grade on physical examination should not be overlooked. Adding VHS, NT-proBNP, etc., thereafter will increase your accuracy of prediction.
6) It should be said that radiographic VLAS, which is newer and less widely studied or used in practice at the moment, did not provide any advantage over VHS, which is of course well studied and widely trained. At some point, we may just be coming up with new ways of saying the same thing – in this case, “this heart is enlarged” – without providing any greater accuracy.
7) The recent reclassification of stage B1 in 2019 to include some dogs with mild degrees of cardiomegaly that are not sufficient to meet the EPIC trial criteria for initiation of pimobendan does make distinction between B1 and B2 more challenging than before, further supportive of using multiple variables to improve accuracy if echocardiogram is not available.
8) Finally, the utility of such prediction models (as opposed to a decision tree) is not high if they are too complex and inaccessible for the everyday practitioner. The authors indicate that based on the data from this study, they are actively working on prospective validation of these models and development of a web-based interface for practitioners that would presumably allow for input of cardiac screening test data and subsequent calculation of a probable disease stage with associated degree of accuracy/confidence. How cool would that be!?
Reference:
1. Wesselowski S, Gordon SG, Fries R et al. Use of physical examination, electrocardiography, radiography, and biomarkers to predict echocardiographic stage B2 myxomatous mitral valve disease in preclinical Cavalier King Charles Spaniels. J Vet Cardiol 2023;50, 1-16.
Today’s VETgirl podcast is sponsored by Antech. As a veterinary professional, you deserve a choice. Now, you have it with Antech. Experience North America’s largest reference laboratory network, comprehensive in-house analyzers and solutions, the most trusted diagnostic imaging equipment, advanced telemedicine and AI-powered solutions, and the flexibility to practice medicine your way. The right diagnostics, at the right time — all from Antech.
Today’s VETgirl podcast is sponsored by Antech. Please note that the opinions in this podcast are expressed by the speaker/sponsor, and not directly endorsed by VETgirl.
Not a VETgirl ELITE member? Get your learning on with the #1 veterinary CE resource – clinically relevant, practical, and accessible 24/7 on your own time! Don’t miss out – click HERE to check out your options, including our new 14-day, no-commitment ELITE Trial.
Webinar
Jan. 23, 2026 / Available On-Demand
Podcast
Dec. 8, 2025 / Available On-Demand
Podcast
Mar. 23, 2026 / Available On-Demand
Webinar
Mar. 4, 2026 / Available On-Demand
Webinar
Feb. 13, 2026 / Available On-Demand
Webinar
Feb. 10, 2026 / Available On-Demand
New achievement unlocked!
Only VETgirl members can leave comments. Sign In or Join VETgirl now!