Canagliflozin reality check

Big Pharma fails to replicate or replace mother nature

for horses genetically out of place in this modern world

Recently, a Swedish study was published which examined the effects of canagliflozin on the insulin response to feeding in 70 horses and ponies with severe insulin dysregulation (ID), also known as insulin resistance.

For ease of reading, I’ll call them all horses from here on out. Regardless of their small size and other distinctive differences, ponies are actually horses (Equus ferus caballus).

To be included, the horse needed to have a plasma [insulin] > 100 μIU/mL when sampled 60–90 minutes after an oral sugar test (glucose syrup in this study).

For some context, the upper limit of normal for healthy horses undergoing this provocative, sugar-overload test is a plasma [insulin] of 45 μIU/mL. (At least, that was the threshold cited in this study.) So, the participating horses really did have severe ID.

To better reflect real-world circumstances, the authors used a feed challenge test instead of an oral sugar test for the study itself. They confirmed severe ID using an oral sugar test, but then they used a feed challenge, which I’ll explain later, to examine each horse’s response to treatment.

Canagliflozin was given orally at a dosage of 0.4–0.6 mg/kg body weight (bwt) each afternoon or evening for 3 weeks. The insulin response to feeding was tested at the start and again after 3 weeks of treatment.

The headline was that canagliflozin significantly decreased the insulin response to feeding. The average post-feeding plasma [insulin] dropped from 304 to 171 μIU/mL after 3 weeks of daily treatment. This decrease was highly statistically significant: P < 0.001.

That sounds impressive. And it is. But the devil, as the saying goes, is in the details.

. . .

Let’s take a closer look at one of the graphs they include in the paper:

Classic bell-shaped curve formed by normally distributed data.

One standard deviation (1σ) from the mean (0) in each direction

accounts for about 68% of all individuals in the group.

But that isn’t what happened. At all.

Here is the shape their data actually made when I plotted the % change on the horizontal, or X, axis and the number of horses on the vertical, or Y axis:

Median % change in [insulin] after canagliflozin (green line).

It divides the group into two equal halves of 35 horses each.

That is, half of the horses experienced a decrease in post-feeding plasma [insulin] of at least 50% (-50) after 3 weeks on canagliflozin. In a few of these horses, it approached a 100% decrease (-100). These are the horses to the left of the green line I've drawn on the graph.

But the other half, to the right of the green line, experienced a decrease in post-feeding plasma [insulin] of less than 50%.

Let that sink in. In half of the horses in this study, canagliflozin decreased the post-feeding plasma [insulin] by less than 50% after 3 weeks of daily treatment.

Bearing in mind that all of the horses in this study had severe ID, with a starting value more than 2 times the upper limit for healthy horses, a decrease of 50% (the median value) after 3 weeks still leaves them well above the normal range — and still vulnerable to laminitis associated with high plasma [insulin].

Also note that 9 of these 35 horses had an increase in post-feeding plasma [insulin] after 3 weeks of daily canagliflozin. Those are the orange bars in the graph. In 2 horses, it was more than 100% higher — that is, more than double the pre-treatment value.

Interquartile range

Bear with me here, because this statistical measure, the interquartile range (IQR), is actually quite useful.

The IQR tells us the range of values for the middle 50% of the group. In the absence of a nice, neat bell curve, which tells us the average value in the middle and the amount of individual variation (standard deviation) on either side of it, the IQR tells us what we might expect for the average horse with severe ID.

Interquartile range for % change in [insulin] after canagliflozin (light blue box).

It represents the values for the middle 50% of the group, what we might expect for the average horse with severe insulin dysregulation.

The IQR for the horses in this study ranged from a 75% decrease (-75) in post-feeding plasma [insulin] to a mere 25% decrease (-25), after 3 weeks of daily canagliflozin.

That's quite a range.

The IQR tells us two other important things as well:

In the 25% of horses whose values were below the IQR — the blue bars to the left of the box — post-feeding plasma [insulin] decreased by 75% to 99% (-75 to -100) after 3 weeks of daily canagliflozin.

These are the horses who responded very well to this drug. Note that they represent only one-quarter (25%) of the study group.

In the remaining 25%, the horses whose values were above the IQR — the blue and orange bars to the right of the box — post-feeding plasma [insulin] ranged from a decrease of less than 25% (-25) to an increase of close to 125% (> 100). Yikes!

Testing the insulin response to feeding

In this study, the horse's insulin response to feeding was tested by a disturbingly simple and mundane method:

* feed was withheld for 6 to 8 hours, such as overnight

* the horse was then fed a meal of its regular hay or haylage (commercially fermented hay), approximately 2 kg of hay for a 500-kg horse (400 grams of hay per 100 kg bodyweight); that's just a slice or two of a well-packed bale

* a blood sample was collected about 2 hours later

For the baseline feeding challenge, the blood sample was collected 2 to 2.5 hours after feeding. After 3 weeks of canagliflozin, the blood sample was collected 90 minutes later. The authors explain the difference, and it's a valid reason, but it's not important to us here.

An important component of the study design is that the horses remained in their home environment, and the owners agreed to maintain consistent feeding, exercise, and turnout routines for the duration of the study. The only substantial change to their management was the addition of canagliflozin.

So, while the 70 horses in this study will have been managed somewhat differently, each served as his or her own 'control'. Plasma [insulin] in response to feeding was measured for each horse before treatment began and again at the end of 3 weeks of daily canagliflozin, and the two values were compared for each horse.

By the way, the study was conducted between November and April (autumn to spring) in Sweden, so it was a good time of year to put this drug to the test.

The authors provided further details (with my emphasis added):

❝ Only horses in which lifestyle modifications had already been implemented as part of the management of ID were included in the study. These management changes comprised the exclusion of grain- or cereal-based complementary feeds, high-fat feeds, and treats such as apples or carrots.

The diet was primarily forage-based, with a water-soluble carbohydrate (WSC) content of <10%; however, forage with a higher WSC content was used in exceptional cases when low-WSC forage was unavailable. ❞

In other words, the horses in this study responded to a low-WSC meal with a greatly exaggerated plasma insulin response. That's a good definition of severe insulin dysregulation.

Plasma [insulin] before and after

Now let's look at the actual blood test results, rather than % change, because here again the con of this drug class comes into stark relief. There is also a little funny business with how the measurements are reported.

First, the starting point, the baseline, 'before', or Day-0 measurements.

baseline

Before canagliflozin treatment began, the average or 'mean' plasma [insulin] after feeding was 304.4 μIU/mL. That's really high for a low-WSC meal. Healthy horses don't exceed 45 μIU/mL after a provocative oral glucose (sugar-overload) test.

But here's where things get a little funny. Rather than reporting the standard deviation (SD) of the mean, which represents the majority (about 68%) of the horses in the group, they reported the standard error of the mean (SEM).

The SEM is always much smaller than the SD. While there are legitimate reasons for reporting the SEM, it's also a great way to hide just how much individual variation there is in the study group, particularly when you set out to show how well your treatment works.

The SEM at baseline was 25.4 μIU/mL. That's a nice, small number compared with a mean of over 300. But let me convert it to SD…

The mean ± SD at baseline was 304.4 ± 212.5 μIU/mL. So, the SD ranged from 91.9 to 516.9 μIU/mL. That's a huge range of values even before treatment began.

And that range represents just the middle 68% of the group (1σ above and below the mean).

after 3 weeks of treatment

After 3 weeks of daily canagliflozin, the mean plasma [insulin] after feeding was 171.2 μIU/mL.

Cool! That is a remarkable decrease. It's an average decrease of 133 μIU/mL, and every little bit counts.

But that figure is only the average for this wildly diverse group. And as the individual values for % change are all over the shop (from nearly 100% decrease to 125% increase), and they look more like an iceberg than a bell curve, the mean is meaningless.

When a set of data is this far from being normally distributed, that's when we have to look at the median and IQR. Alas, we don't have those numbers for the blood test results, so let's continue with the means and SEMs as reported…

The SD of that mean is 216.7 μIU/mL (converted from a SEM of 25.9). The SD was 27% larger than the mean itself!

That's a dead giveaway that you don't have normally distributed data — which, of course, the graph makes abundantly clear.

Still, if we look at just one standard deviation from the mean, which represents about 68% of the group, post-feeding plasma [insulin] ranged from essentially 0 to about 388 μIU/mL. And there may still have been individuals above that range after 3 weeks of daily canagliflozin.

Which horses showed the greatest improvement?

Did the horses with the highest baseline values show the greatest reduction with treatment?

Or was it the horses with the lowest baseline values who responded the best?

We don't know, as the study didn't look for the who, what, where, when, or why of it all.

But they did present a brief case study in their supplemental files that is very interesting.

Case study

The case involved an 18-year-old Shetland pony gelding weighing 200 kg (440 lb), with a body condition score (BCS) of 7/9 and a cresty neck score (CNS) of 3/5, before treatment.