CardioCare by Purina

what does the science really say?

Christine King  BVSc, MANZCVS (equine), MVetClinStud

In this graph, I've shown each mean as a coloured dot (blue for CON and green for CPB), with the standard deviation (SD) for each diet group and evaluation point as a vertical line extending above and below its mean. (That is, these bars are SD rather than SEM.)

At baseline, we're also provided with the full range of values (minimum and maximum) for the diet group. We aren't given the ranges for the other evaluation points, so I've simply added a shaded bar behind the baseline means and SDs to represent the range for each diet group at the start.

As you can see, there is huge variation among dogs, even within the same diet group. I would have overlaid these SD bars onto the paper's own graph, but they extended way beyond the bounds of the graph.

With spreads like these, in such small and homogeneous groups of dogs, it is no wonder that there were no statistically significant differences between the two diets after 6 months. Seriously, how did this paper clear peer review?!

Regression toward the mean

I haven't yet talked about this important statistical phenomenon, regression toward the mean.

It's also called regression to the mean, although I like 'toward' better, because 'to' implies arrival at a fixed point.

Over time, in any study involving live subjects (in this case, dogs), there is naturally some 'drift' in individual values that has nothing to do with the intervention being studied, nor even with progression of the disease.

Unless something interferes, the drift is toward the population mean. That is what is meant by 'regression toward the mean'.

The population mean is a theoretical number, as it's impossible to take every dog in the world and measure its LA/Ao, for example, and calculate the mean LA/Ao for the entire population of interest (e.g., all dogs, or all dogs with/at risk of mitral valve disease). However, it is accepted as existing, of being a real number, even though it can never be measured and therefore cannot be precisely known. (It's a bit like measuring the total volume of water in the world's oceans... Doable, in theory, but...!)

The importance of accepting that there is a population mean is that we look at the group means at baseline and accept that they may not accurately represent the population mean. The groups of dogs in this study are merely a tiny sampling of the population they're meant to represent (all dogs, or all dogs with early mitral valve disease), and they are being sampled at just a single point in time, a mere snapshot in their lives. There will naturally be some shift or drift in individual values and the group mean over time, simply because living systems change. That's a fundamental characteristic of life: change.

From real-life data and from statistical modelling, we know that individual values and therefore group means naturally drift or 'regress' toward the population mean, unless something happens to alter their course.

In any well designed clinical study, the subjects (in this case, the Beagles and Min. Schnauzers) are meant to be good representations of the population of interest — in this case, domestic dogs, or more specifically dogs with/at risk of developing mitral valve disease. The degree to which the study subjects fairly represent the target population has a huge influence on whether the intervention (in this case, CardioCare) performs in the real world as it did in the study.

Regression toward the mean can be a particularly bothersome fly in the ointment with any small or short study. That's because individual values which may have begun substantially higher or lower than the population mean will invariably affect the group mean at baseline, and will then drift toward the population mean unless something alters their course.

If the statistical analyses — and, more importantly, their interpretation — fail to account for this natural phenomenon, then we are in danger of reaching false conclusions. This is one of the key reasons why so many promising clinical studies fail to live up to expectations when the intervention is put to the test in the real world.

Here's an interesting definition and explanation of this phenomenon: "the more extreme a variable is upon its first measurement, the more likely it is to be closer to the average the second time it is measured. People who are especially lucky at picking stocks one year are likely to be less lucky the next year. ... An athlete who does extremely well in one season is likely to perform more poorly the next season. This probably explains much of the Sports Illustrated cover jinx — many believe that appearing on the cover of Sports Illustrated will bring an athlete bad luck." From Intuitive Biostatistics, by Harvey Motulsky, MD.

Perhaps a simpler example is how much your heart rate and blood pressure vary over the course of a day. At any given point in time, your heart rate may be substantially higher or lower than your average (mean) heart rate. It'll be substantially higher when you're exercising or scared out of your wits; it'll be substantially lower when you're asleep (unless you're having a nightmare); and it'll be about average when you're sitting down having lunch. Relying on a single measurement can be misleading.

As for why individual values, and thus group means, may have begun substantially higher or lower than the population mean, there are many reasons related to the subject (the dog), the investigator (the human), the methods, and factors external to them all (time of year, weather, time of day, etc.).

For example, a dog may have been feeling particularly nervous at his first evaluation or something startled him and he remained unsettled; he was more calm during later evaluations because he knew what to expect. Or a dog was feeling a bit crummy the first day (as we all do from time to time), and felt fine at later evaluations. Or one of the investigators was in a bad mood that day or was feeling a bit nervous at the start of this big, expensive study designed to make their employers a ton of money...

Anything that alters the resting sympathetic tone, in either direction (up or down), has the potential to affect dynamic measurements such as left atrial diameter and aortic tone.

We see these flies buzzing around in the data for systolic arterial blood pressure (SAP) at baseline. For example, in the dogs with mitral valve disease assigned to the CPB diet, mean SAP at baseline was 171 mmHg, but it ranged from 133 to 218 mmHg!

Individual values and group means will naturally drift toward the population mean as the study progresses, unless something occurs to alter their course. The smaller the study group (in this case, 8 to 10 dogs each), the more likely this phenomenon is to occur and to be a problem, unless accounted for when analysing and interpreting the data.

Take another look at the coloured graph (smaller this time, so the trends are more obvious), as I think it illustrates this phenomenon rather well...

The asterisks suggest that the differences between diets at 3 and 6 months were statistically significant. However, the only significant differences were found in the dogs fed the CON diet; mean La/Ao at 3 months and 6 months were significantly greater than at baseline (P = 0.012 and 0.010, respectively).

There was no significant reduction in mean LA/Ao over time in the dogs on the CPB diet (P = 0.06 and 0.26 at 3 and 6 months, respectively, compared with baseline).

That bears reading again, because the title of the study explicitly states that this dietary intervention "reduces left atrial enlargement in dogs with early preclinical myxomatous mitral valve disease." Whether one uses LAD or LA/Ao, this claim is a gross overstatement of their findings.

The compound effect of diet over time ('diet x time') yielded a P value of 0.005, which is apparently what the asterisks are meant to represent, and perhaps is what they're using to support their claim.

But note that the error bars overlap at all three time points. Our confidence that these data reliably represent what will happen in our own dogs should be low. (See page 2 for a discussion of standard error bars.)

Here's what the data look like, plotted more realistically:

At baseline, the values for LA/Ao in the dogs on the CPB diet are higher than those of the dogs on the CON diet. (But this difference does not rise to the level of statistical significance; not even close: P = 0.54; that is not a typo.)

The values for the CPB group drop rather dramatically (or so it seems) between baseline and 3 months — to be more in line with those of the dogs on the CON diet at baseline — only to drift back up again and consolidate around the mean (smaller SD) at 6 months — which, again, is more like those of the dogs on the CON diet at baseline!

The importance of this natural phenomenon cannot be overstated. It must be considered when small and biologically incongruous or inexplicable changes such as this occur over time.

The 'biologically incongruous/inexplicable' part is a give-away: how else would one explain why LA/Ao dropped so much in the first 3 months, only to rebound in the second 3 months, resulting in a net reduction of only 4% in 6 months (mean change in LA/Ao in the CPB group between baseline and 6 months)?

By ignoring this phenomenon, one is in great danger of making too much of too little (small differences between groups or over time) — which is what seems to have happened with this study.

The same lens needs to be applied to the dogs on the CON diet. We see evidence of small drifts in their individual values and group means over time, as well. In fact, that could well explain the "crossover" effect we see in the paper's graphs, plotting LAD and LA/Ao over time.

The authors interpreted this crossing of artificially created lines as evidence that the dogs on the CON diet got worse, while the dogs on the CPB diet got better, over time. But an equally plausible explanation, and one supported by the lack of statistically significant differences between diets, is the drift that naturally occurs with regression toward the mean.

***

I've had just about enough of this nonsense, but before I shift gears and talk instead about the diets, I would be remiss if I didn't briefly cover one last thing that they relied upon rather too much to support their claim...

As with LAD, they did some "creative accounting" in an attempt to contrive a statistically significant difference in LA/Ao between the two diets. As with LAD, they failed. When they looked at changes in LA/Ao over 6 months as % change from baseline, there was no significant difference between the two diets (P = 0.049, where their own threshold for significance was P < 0.017).

I won't bore you with the details. They're all there in the paper and the supplementary tables if you're really keen.

Simply put, some dogs on the CPB diet had an increase in LA/Ao while others had a decrease.

Likewise, some dogs on the CON diet had an increase in LA/Ao while others had a decrease.

***

On the next (and final) page, I'll discuss the diets fed in this study and suggest some things we may want to consider adding to a wholesome, species-appropriate canine diet (meat-based, low/no starch) that may help slow disease development and progression in at-risk dogs.

next page

© Christine M. King, 2022. All rights reserved.

First published 20 September, 2022. Last updated 24 September, 2022.

back