The $12M NuSI/Ludwig Study ~ Part III: Some "Early" Lessons


Continuing on with discussion of:
Effects of a low carbohydrate diet on energy expenditure during weight loss maintenance: randomized trial

In Part I, I discussed some issues with methodology, mostly focusing on the reduced Run-In Phase that likely compromised the outcomes irreparably.

In Part II, I highlighted a serious issue with the Run-In Phase, the purpose of which was to produce a somewhat homogeneous "reduced weight state" to test various diets in maintenance of that state.

  • randomization to the various test diets occurred after weight loss (PWL)
  • all subjects lost weight on the same 45% Carb / 30% Fat / 25% Protein diet targeting a weight loss of 12% ± 2% initial body weight.
The researchers do not appear to have made many adjustments in the weight loss phase to produce a more uniform weight loss.   Rather than 12% ± 2%  (from 10% to 14%)  losses, the actual outcome was roughly 10.5% ± 5% (actual range from  5.6% to 16.0%.  

To repeat, the amount of weight loss is irrespective of the test diet.  Everyone had baseline insulin measures taken, everyone was put on the same weight loss diet.  Thus we have an "accidental" test-within-a-test of the CIH/TWICHOO from these "early" results.

I offer these scatter plots for the105 subjects who successfully completed the study and for whom complete data for insulin measures and energy expenditure were available at all time points.

The Carb-Insulin Hypothesis (aka TWICHOO) predicts that weight loss will vary inversely with insulin levels:  The higher the insulin levels, the lesser the weight loss.  The Run-In Phase data supports no such relationship (indeed, if anything, absolute weight loss was greater for those with higher baseline insulin measures.

Meanwhile, differences in weight loss are easily explained by variation in caloric deficit during the calorie restricted Run-In due to coarse estimation of baseline energy expenditure (vs. rigorous measure).

This post expands on some relationships of baseline (BSL) and post-weight loss (PWL) measures as observed during the weight loss portion of the Run-In Phase.

Bottom Line:  Baseline insulin status seems to be irrelevant to weight loss on a "high carb" calorie-restricted diet.  

The study in question had two phases, a weight loss Run-In Phase, and a dietary intervention weight maintenance Test Phase.  As such, there are four major time points for data collection, involving two different "baseline" measures.  In the data files from which I obtained the data for the plots in this and other posts, these are:
  • BSL = Baseline = Pre-Run-In.  As the doubly labeled water assessment of total energy expenditure began on "Day 1", it is fair to say that these measures are of the subjects "off the street".  
  • PWL = Post Weight Loss = last two weeks of Run-In.   Following 10 weeks on a calorie restricted (60% of estimated energy requirements)  45%C/30%F/25%P composition diet, subject intakes were increased to stabilize the reduced weight.  DLW and other measures were made during these two weeks.
The Test Phase involved 20 weeks consuming weight-maintaining caloric levels of one of three test diets, 20% protein and varying in carb from 20% to 40% to 60%.  Data was collected at:
  • MID = Midpoint = 10 weeks.  DLW-TEE data was gathered in weeks 9 and 10.
  • END = End (grin) = 20 weeks.  DLW-TEE data was gathered in weeks 19 and 20.
The major "finding" of the study was a reported average increase of ~210-280 cal/day in total energy expenditure (TEE), depending on whether intent-to-treat (ITT) or per-protocol analysis.  In a late hour change, these findings use PWL as "baseline" instead of the originally slated BSL measures.  When subjects were divided into tertiles (thirds) of the BSL measure of Ins-30 (described later in this post), a comparison between the LOW and HIGH carb diet groups in the highest tertile of Ins-30 produced a "metabolic advantage" of some ~310-480 cal/day.   That latter figure being most *unbelievable*! 

So let's take a look at some of these tertiles -- both at true baseline, BSL, and the post-weight loss, PWL, test phase start point.

A note about all plots in this post

Subjects:   My analyses are for only those subjects who completed the trial successfully (maintained weight to within ± 2 kg of PWL weight) for whom there is complete 4 time point data for all energy and insulin measures.  This involves:  105 Total , 40 LOW carb, 35 MOD carb, 30 HIGH carb.  

Colors:    Gold = LOW , Blue = MOD , Green = HIGH , Gray = All
Diamonds are means, Turquoise diamond lines represent mean of all subjects

Tertiles:  (split to balance high and low for uneven sized tertiles)

T1 = Tertile 1 = Lowest Third:  lowest 13 for LOW, 12 for MOD,10 for HIGH , lowest 35 for ALL
T2 = Tertile 2 = Middle Third:  middle 14 for LOW, 11 for MOD,10 for HIGH , middle 35 for ALL 
T3 = Tertile 3 = Lowest Third:  highest 13 for LOW, 12 for MOD,10 for HIGH , highest 35 for ALL 

There are two plots side-by-side.  

  • Left displays tertiles within diet intervention groups, each grouping is T1, T2, T3 and ALL for each diet.
  • Right compares each tertile between diet interventions, groupings are T1, T2, T3 and ALL 

A note about statistical significance

As one can imagine, this endeavor has already been rather time consuming, and at this time I am not in a position to assess the statistical significance of any of the trends in the various plots I'm sharing.  In many cases, the lack of trend supporting TWICHOO  (Taubes Wrong Carb-Insulin Hypothesis Of Obesity) is more than sufficient, there is no need to statistically quantify it!

It's The Insulin, Stupid?

The remainder of this post will deal with the "I" part of the original hypothesis.  Insulin regulates fat mass, and higher insulin, as stimulated by carbohydrates, traps fat in the fat cells so that the rest of your cells are starved of energy and cannot burn the fat.  

At baseline, the investigators measured insulin in two contexts:
1.  Fasted insulin levels -- basal insulin levels.
2. Insulin-30 (Ins-30):  insulin response at 30 minutes after a standard oral glucose tolerance test (OGTT). --measure of insulin secretion in response to dietary carbohydrate.

As the hypothesis goes, repeatedly raising insulin with dietary carb results in elevated insulin virtually around the clock -- your fat is locked up and can never get out.   Eliminate the carbs and your insulin levels will finally drop low enough for your fat cells to give up their stores.  Atkins even once described this lipolysis to be "as delightful as sex and sunshine"!  

The increased energy expenditure being touted more of late is hypothesized to be in response to this glut of fat available to burn!   So isn't it odd that, for everything measured ...
Neither fasting insulin levels, nor Ins-30 are measured at any point after BSL baseline.
At LEAST fasting insulin should have been measured, especially considering that the hormones ghrelin and leptin WERE measured.

Tertiles of Baseline Insulin-30

Below are the Ins-30 levels (BSL) plotted according to the tertiles of those same Ins-30 levels.  The mean Ins-30 for the highest tertile of all subjects was roughly four times that of the lowest tertile.  Also of note, there is a far wider range of Ins-30 levels within the highest tertile than within the lowest two tertiles, either individually or combined.

BSL Ins-30 levels by Tertile of BSL Ins-30 levels

While it is unlikely that differences of the means rise to statistical significance, given that most comparisons in the analysis are between LOW (gold) and HIGH (green), there are some distribution differences worth noting.  By the way, at any time you can click on images to zoom in and take a closer look.   Some in the middle tertile of the LOW group have higher than all-mean Ins-30 levels, while all of the middle tertile of the HIGH group are "safely" below this mean.  Further, some of the highest tertile HIGH group are still below all-mean Ins-30 levels.    While these two groups have comparable hyperinsulinemic outliers, it is fair to note that the subjects randomized to the LOW carb intervention were more metabolically "damaged" than those randomized to the HIGH carb intervention. 

It would have been nice to have this measure as a post weight loss "baseline", as most would agree that weight loss will have the greatest effect on insulin dynamics in those who have some degree of metabolic dysregulation.  Ludwig has now argued strongly for why the pre-weight loss BSL energy expenditures are less appropriate as an "anchor" with which to compare the responses to test diets, and indeed his analysis uses the post weight loss (PWL) values.  I would argue that whether you agree or not, one can't have it both ways.  If PWL energy expenditures are a more appropriate baseline anchor, then so too would be PWL Ins-30 levels.  Only we don't have those!  [snarkasm] So much for that meticulous study design? [/snarkasm]

TWICHOO, or the Carbohydrate-Insulin Hypothesis, as even put forth in Ludwig's 2016 mass media diet book Always Hungry, predicts that carb-spiked insulin will sweep energy out of circulation and shuttle it off to the fat cells, leaving other cells starving for energy.  This cellular starvation -- sometimes called "internal starvation" -- will cause a person to become hungry and eat more.  Sounds somewhat plausible, but as such, TWICHOO predicts that those in the highest tertile of carb-stimulated insulin production would have the most trouble adhering to a calorie restricted, high carb diet, and would thus lose the least weight.   Let's see how that shook out, shall we?

Below are the plots of absolute weight lost (in kg) during the 10 week Run-In according to BSL Ins-30 tertiles.

Weight loss (kg) during Run-In by Tertiles of BSL Ins-30

Despite secreting roughly four times the insulin in response to carbohydrates, there is a trend towards increased absolute weight loss in the higher tertiles compared to the lowest ones.  For whatever reason, this randomly shook out to be most profound in those who were randomized to the LOW carb group.  This does not seem to persist, however, when weight loss is expressed as a percentage of initial body weight as shown below.

Weight loss as percent initial bodyweight during Run-In by Tertiles of BSL Ins-30
It is fair to say that there is no favorable trend in Ins-30 between highest and lowest tertiles that supports TWICHOO.  Meanwhile the fact they all lost weight on a high-ish carb reduced calorie diet regardless of insulin status provides evidence against it.   Both the biggest loser (16.0%) and the the smallest loser (only 5.6%) had Ins-30 responses near the median for all subjects.

Tertiles of Baseline Insulin-30 & Fasting Insulin (Ins-Fast)

Critical to TWICHOO is not only an exaggerated insulin response to dietary carbohydrate, but that somehow this leads to chronically elevated insulin levels -- basal insulin secretion measured in the fasted state.  Again, the idea here is that not only do postprandial insulin spikes lock away fat, but insulin remains elevated virtually 24/7 making it practically impossible to burn stored body fat for fuel.

I have repeated the above graphical analyses for the tertiles of BSL Ins-Fast, but to segue to that, I have compiled two sets of plots illustrating the relationship between BSL Ins-30 and BSL Ins-Fast for the subjects in this study.   First, below, are Fasting Insulin levels for the tertiles of BSL Ins-30.

BSL Ins-Fast levels by Tertiles of BSL Ins-30

As expected, there is a trend towards higher mean fasting insulin from lowest-to-highest tertile, but there is also wide overlap between Ins-Fast levels across the three tertiles of BSL Ins-30, particularly at lower levels.  This is consistent with the etiology of both fasting and postprandial Ins-30 having related but causes but great individual variation in the manifestation of basal vs. postprandial hyperinsulinemia.

I would note that, again by random "luck", this relationship appears to be strongest in the LOW carb group and is dominated by what is seen in the highest tertile of BSL Ins-30.   While the BSL Ins-30 distributions seem to indicate more metabolically dysregulated subjects among the LOW carb group, the elevated  BSL Ins-Fast in the third BSL Ins-30 tertile are a stronger indication that the LOW carb group has more metabolically dysregulated subjects.  What effect this had on the primary outcome (total energy expenditure in reduced-weight maintenance) is a topic for a future post.

And now, I flipped the insulin measures.  Below are the BSL Ins-30 levels grouped by tertiles of BSL Ins-Fast.

BSL Ins-30 levels by Tertiles of BSL Ins-Fast
I'm not sure there is much to add with this "flip", similar trends can be seen from lowest-to-highest tertiles, but again, there is wide overlap at the lower levels of BSL Ins-30 across the tertiles.

Perhaps the best way to put a bow on this are the scatter plots below of Fasting Insulin vs. Insulin-30 at baseline.  The plots are for each diet group separately, and all subjects.  Horizontal and vertical dashed lines are means for all data, and diagonal dotted line is the linear regression fit line for all data.

Scatter plots of BSL Fasting Insulin vs. BSL Insulin-30
By diet group and all subjects , Mean and Fit lines are for all subjects in each plot

Tertiles of Baseline Fasting Insulin

I'm going to round out this presentation of insulin-related plots by presenting the same plots using tertiles for BSL Ins-Fast = Baseline Fasting Insulin, in the same manner as I did previously with BSL Ins-30.

Below are the Fasting Insulin levels (BSL) plotted according to the tertiles of those same Ins-Fast levels.  Here there is an even wider range of Ins-Fast levels within the highest tertile than within the lowest two tertiles, either individually or combined, where the range of Ins-Fast levels is strikingly narrow.  Despite the greater variation, the highest tertile mean is not quite double that of the mean of all values.

BSL Fasting Insulin levels by Tertile of BSL Fasting Insulin levels

We again see a trend where those randomized to the LOW Carb group have a more pronounced difference in Fasting Insulin levels in the third tertile compared to other diet groups, and a higher third tertile mean Ins-Fast level.

So how did weight loss on a high carb diet shake out vs. these baseline values and tertiles?  Below are the plots of absolute weight lost (in kg) during the 10 week Run-In according to BSL Fasting Insulin tertiles.

Weight loss (kg) during Run-In by Tertiles of BSL Ins-Fast
Despite the higher basal insulin levels trapping fat hopelessly away, there's an even more pronounced trend of increased liberation and oxidation of fat as across the tertiles from lowest to highest.   And for weight loss as a percentage of initial weight ...

Weight loss as percent initial bodyweight during Run-In by Tertiles of BSL Ins-Fast

Once again, any trend tends to be ameliorated when measuring weight loss as a percentage of initial weight, however the lack of effect is still evidence against TWICHOO.  The wide ranging elevations in fasting insulin seen in the third (highest) tertiles should have been sufficient to limit weight loss.  Rather, it was a caloric deficit that produced weight loss.

A Final Smattering of Scatterings

Since randomization to maintenance test diet occurred after weight loss on a common diet, we can look simply at various  PWL "baselines" vs. true baseline, BSL values for all subjects.  Plotting weight loss on the y-axis (as absolute kg or percent) vs. insulin on the x-axis (as fasting or 30 min post OGTT), I generate the following.

Once again I repeat and reiterate ... The Carb-Insulin Hypothesis (aka TWICHOO) predicts that weight loss will vary inversely with insulin levels:  The higher the insulin levels, the lesser the weight loss.

I titled this blog post "Early" Lessons, because there was an early test of the CIH / TWICHOO that goes unaddressed by the researchers.  That is, early in the game ... during Run-In.  They fed (to improve compliance) a reduced calorie, constant macro composition diet to ALL participants during this Run-In phase.  As discussed in my last post, the researchers apparently did not adjust these diets much following baseline estimates. so that rather than producing a narrow range of weight loss by percentage (target 12% ± 2%) a rather wider range was achieved:  10.5% ± 5%.  While this introduced possibly fatal error into the Test Phase of the study (stay tuned!) , it still provided an unintentional test of CIH/TWICHOO.

There is NO relationship between baseline insulin measures -- INS-30 or INS-Fast -- and weight loss -- either as a percent or absolute amount -- during the Run-In/Weight Loss Phase of this current study that supports the Carb-Insulin Hypothesis (aka TWICHOO).  If anything, in absolute weight loss amounts, a slight correlation exists between higher insulin and greater absolute weight loss (statistical significance not determined).

Thank you NuSI, for putting some final nails in the coffin of a hypthesis that was never viable once your organization was formulated.  At least if we can further cement the basic lessons, the $12 MILLION dollars (not including the cuts for NuSI officers, employees and consultants) will not be for naught.