Greetings again, my friends! These past four weeks have been pretty packed and my next three weeks will be no different. That said, I’ve managed plug in a lot more NMRs due to the experiment I’ve been conducting which involved another week of successive blood draws, bringing the total to a whopping 28! In so doing, it appears the “diversion” I wondered about in Part III of the series appears to be a one-off. (See graphic below)
Tonight I’ll be flying to San Diego to attend the Low Carb USA conference this weekend. I’m excited to update a number of the doctors I’m typically connecting with over Twitter and email. I already have five meetings and am getting worried about my bandwidth impacting my lecture attendance. It’s gonna be busy!
I have completed a fairly involved experiment a couple weeks ago which I’ll be previewing at the conference to a number of people, and for which I’ll have posted here at the blog before August 8th (as Part IV of the series) to coincide with my discussing it on Jimmy Moore’s Living La Vita Low Carb Show. Needless to say, I’m extremely excited to share the results.
Note from Dave: This post is actually a replacement due to the previous Twitter Prediction post being weirdly targeted by spambots.
If you’ve read my series up to this point, you know the pattern being identified should be not only be reproducible, but predictive. So for my 22nd blood test, I decided to have a little fun. Instead of predicting about where it landed myself, I tweeted…
I need a few twitter volunteers from the audience…
Reminder — all these tweets are dated July 7, 2016, the night before my blood test.
To replot the graph with their predictions:
The LDL-C result from my July 8th test just came in Friday and is as follows:
And thus, here’s the updated graph with the new 283 result plotted on the red line:
Here’s the close up:
Thus making Jeff Winkler @winkler1 the winner!
If you’re wondering how I could have three total strangers could predict where my cholesterol numbers would be in such a tight range, you haven’t been reading my series! (See Part I, Part II, and Part III)
In Part I, I shared the data of my first 15 cholesterol blood tests and how closely it correlates with dietary fat. To recap:
My cholesterol baseline jumped after starting a ketogenic diet, which led me to do close blood testing against a theory I had on the lipid system appearing as a network (more on that theory later)
If the theory had merit, there would be two tenants to assume:
The cholesterol transporting lipid system would prove to be agile – much more so than is typically believed (this appears to be accurate).
Discrete patterns should emerge between the diet and serum cholesterol now that disruptive inflammation is lower (this certainly appeared to be the case).
After the testing, my bloodwork showed the following:
The more fat I ate, the lower my Total Cholesterol. (87% inverted correlation)
The more fat I ate, the lower my LDL-C. (90% inverted correlation)
The more fat I ate, the lower my Triglycerides. (61% inverted correlation)
The more fat I ate, the higher my HDL-C. (74% correlation)
In Part II, we moved beyond LDL-C into total and small LDL-P and their differences in the findings.
LDL-P and small LDL-P appeared to have a correlation with a three day average of dietary fat as like LDL-C, but with a key exception of adding a two day gap between the dietary period and the blood test.
Small LDL-P appeared to correlate closely with LDL-P but at a higher gain to loss ratio.
Both LDL-P and small LDL-P proved to be extremely agile and easily ramped up or cleared when observing via daily blood tests (data points 9-14). In fact, shifts in the hundreds of particles per day were easily achieved in either direction.
The Ratio Problem
If I were playing devil’s advocate to my own research up to this point, I could make the case that we can’t be completely certain my LDL cholesterol was inverting with fat, since the same could be said for protein. After all, my protein intake ran very proportional to my fat intake, even if at a smaller fraction. When running Three Day Average Protein against the LDL-C, I also got a significantly close correlation at -0.789
Inverted…
Thus, we need to have at least one experiment to have a lopsided ratio of one against the other relative to the ketogenic ratios.
The Intended Divergence
Me being me, I decided to do two tests: one with a Super High Fat (SHF) ratio 95% fat / 4% protein / 1% net carbs for three days, and one with a Super High Protein (SHP) ratio 50% fat / 45% protein / 3% net carbs for three days. And finally, I added a third test in the following four days at closer to my usual ketogenic ratio with 75% fat / 21% protein / 4% net carbs to see if my numbers snapped back into the original correlation.
To meet the requirements for the Super High Fat sprint, I averaged 293g fat, 31g protein, and 4g net carbs.
Here’s a sampling of the food I was eating over those three days:
Here are the results:
Inverted…
Sure enough, at a preposterously high ratio of fat to protein, my inversion correlation appears to still follow the fat, not the protein. My blood ketone levels (BHB) over these days were 2.1, 2.1, and 1.4 respectively.
Then I switched to Super High Protein. Over these three days I averaged 119g fat, 233g protein, 13g net carbs.
I was actually looking forward to this part of the experiment given I’ve always been a big fan of meat. In fact, I wondered if I wouldn’t get hooked on the higher protein ratio due to how much more meat I was allowed to consume.
Here’s a sampling of the food I ate:
But then something unexpected happened…
In addition to all this biometric data I collect, I also keep regular notes on any unusual aches, nausea, or pretty much anything I feel that seems out of the ordinary. As I was getting to the end of the Super High Protein sprint, it was the only time I felt gastrointestinal distress that I associate with my days before the diet. It was a familiar feeling, but certainly not one I was missing.
I likewise felt heavier and less energetic, taking a nap on two of the three days. While I knew I was generating a higher glucose load via Gluconeogenesis due to all the protein, I actually rechecked all my food labels to make sure I didn’t accidentally eat something high in carbs.
Here are the results:
Inverted with overlays…
No question – we have a clear divergence where LDL-C does not follow dietary protein for the last two data points, Very High Fat (VHF) and and Very High Protein (VHP). Note it didn’t follow the dietary fat for this period either:
It’s worth taking a moment to point out two very important observations.
First, if my only goal was to reduce my “bad cholesterol,” this would appear as good news. Assuming this trend held, having higher protein and less fat would result in lower LDL-C. As you can see from the graph above, were the inverse correlation holding, the lower 119g of fat would likely push up the LDL-C to around 323 rather than the 263 we see instead.
However, the energy level and GI issues I was experiencing were certainly a drawback. It also seemed to fall in line with the second tenet of my theory if it was causing inflammation.
See, here’s where the dark side of my theory comes in. What if there are steps I can take which lower my LDL cholesterol but only because it increases inflammation and/or oxidative stress? I might interpret this as a good outcome when it’s happening because, as everyone well knows, lowering dangerous cholesterol is all that matters.
Yet what if my body is sending me the correct signals in the first place? High Carb = feel slower, with occasional GI issues. Low Carb, High Fat = feel great with little to no issues. High Protein, Moderate Fat = same as High Carb.
I decided to do my final data point to follow the VHF and VHP, which did snap back to the correlation envelope. I then went on the Low Carb Cruise and connected with a few more doctors to discuss the data. After getting back, I did my 19th and 20th NMR with the intent to bring all my data to this blog by the end of May.
But I was in for one more twist. The biggest yet, in fact.
The Unintended Divergence
Here are all 18 NMRs where I was on the normal ketogenic diet (removing the SHF and SHP data points) with inversion:
Spot anything unusual?
Yes, data point 18 at the far right shows the largest single divergence in the correlation of any other coupling in this graph. Was it a lab error with my blood? Was it something I ate or drank differently?
The only major change I made was cutting out diet soda the week before (primarily Coke Zero). So I first kept to no diet soda for another week and did one more test just to be sure it wasn’t a lab error. The divergence appeared to hold (see below with 19).
After that, I went to town with Coke Zero for three days to see if it would spike the correlation to the other side. It seemed outlandish to think aspartame could have kept my cholesterol artificially high, but I had to be sure. Again, the divergence held (see below with 20).
This new data suggests my LDL-C has been dropping all on its own (and by extension, Total Cholesterol).
But data points 17-20 did not significantly change the correlations of HDL-C, LDL-P, or small LDL-P. In fact, it improved them slightly.
Other Markers Improve Correlation
HDL-C had a -0.733 before, now it is -0.761.
LDL-P had a -0.812 before, now it is -0.845.
Inverted…
LDL-P had a -0.726 before, now it is -0.781.
Inverted…
Next Steps
Will data points 1-16 represent a temporary “phase” of my diet with regard to LDL-C, proving 17-20 as the New Normal? Or is it the other way around? Your guess is as good as mine given the 6/9/16 test was the last one I took as of this posting.
The next blood test I’m taking is this week and it will be a very large combo pack of CMP, CRP, A1C, and other goodies in addition to the NMR. I’ll need to as I’ll be making some large changes to my exercise schedule. Starting next weekend I’ll be training for half marathons in the coming months. Will it start to impact my numbers? Stay tuned…
Coming soon – The Lipid Network Theory (Or “What Led Me Down This Rabbit Hole in the First Place”)
In Part I, I shared the data of my first 15 cholesterol blood tests and how closely it correlates with dietary fat. To recap:
My cholesterol baseline jumped after starting a ketogenic diet, which led me to do close blood testing against a theory I had on the lipid system appearing as a network (more on that theory later)
If the theory had merit, there would be two tenants to assume:
The cholesterol transporting lipid system would prove to be agile – much more so than is typically believed (this appears to be accurate).
Discrete patterns should emerge between the diet and serum cholesterol now that disruptive inflammation is lower (this certainly appeared to be the case).
After the testing, my bloodwork showed the following:
The more fat I ate, the lower my Total Cholesterol. (87% inverted correlation)
The more fat I ate, the lower my LDL-C. (90% inverted correlation)
The more fat I ate, the lower my Triglycerides. (61% inverted correlation)
The more fat I ate, the higher my HDL-C. (74% correlation)
Particle Primer
A typical cholesterol value from a blood panel only tests the quantity of cholesterol contained in LDL and HDL particles, not the particle number itself. And even then, it often uses the Friedewald Equation to estimate LDL. Cholesterol is actually just the “cargo” of LDL and HDL particles, not the particles themselves (the boats carrying the cargo). For a much more in-depth discussion, I highly recommend Peter Attia’s Straight Dope on Cholesterol lecture. (Which is where I’m shamelessly stealing the boat/cargo analogy from. Thanks, Peter!)
Thus, when we refer to LDL-C and HDL-C, the “-C” refers to the amount of total cholesterol when added together (the cargo of all ships combined). But when using LDL-P and HDL-P, the “-P” refers to the total particles (a total count of all the boats that carry the cargo).
The current belief in Lipidology is that LDL-P is a good, independent marker for cardiovascular disease. But within the low carb community, many believe there needs to be a distinction between “large, fluffy” LDL-P and “small, dense” LDL-P. They would argue that the small, dense particles are the true cause for heart disease, whereas the large, fluffy particles are benign or even protective. I won’t go much further into this debate other than to say I’ve read a lot of the studies each side prefers, and I’m not fully convinced of either position at this time.
Since all my blood tests were NMRs (Nuclear Magnetic Resonance), I kept careful track of these particles as well as my cholesterol. And the results were quite interesting…
LDL-P Result Offset
Back when I had eight tests over three months, the LDL-P appeared to match the LDL-C and its inverted correlation to the three day average of total fat.
Now again, let’s flip the axis of the same graph to see the inversion better…
So LDL-P correlates with the three day average of dietary fat just like LDL-C, right? Not so fast.
Here’s what happened during the week-long experiment.
And inverted…
Obviously, this graph seems to suggest that a two day delay is occurring. But could it really be that simple? We’ll get to that in a second… but for now, take extra notice of the massive swing in my LDL-P over just eight days!
On Monday’s blood test, I’m coming in at 2622. Two days later I have the highest score I’ve ever gotten at 3391! Then, just two days after that, I’m landing over 900 points down at 2455. I’ve read article after article that suggested this system takes extensive time to change. And while that might be true of 1/2 to 2/3 of the baseline score, it certainly doesn’t seem to pertain to the rest.
The LDL-P Gap
I set up a new formula against my spreadsheet to see if I could capture the correlation.
To use the example from before for LDL-C, if I took a blood test on a Friday morning (Day 0), the relevant numbers appeared to be the average of Tuesday, Wednesday, and Thursday together (Day -3, Day -2, and Day -1).
In the LDL-P equation, however, I’d still average three days, but with a gap of two days in between. So if the test were on a Friday morning (Day 0), I’d average together Sunday, Monday, and Tuesday (Day -5, Day -4, and Day -3).
Below, I put together the 1-8 tests over three months, the week-long 9-14 tests, and the outlier test together using this equation.
And inverted…
Naturally, this is where my jaw hit the floor.
This obviously brings up a lot of powerful new questions regarding LDL-P synthesis of the liver in this counter-regulatory role. But this post is for the lay person, so I promise not to geek out on my lipid theories here. That will come in a later post.
Small LDL-P
What of the “small, dense” LDL-P mentioned above? Do they, likewise, track with the new formula?
Yes, they do.
And inverted…
This, likewise, brought forth a number of extremely curious observations:
Once again, we have massive shifts in a very short span of time with the week-long experiment. I go from a 484 to a 941 in two days, back down to a 378 two days later, and finally to <90 after a weekend of eating high fat three days straight.
And also once again, I practically custom-set a <90 on the intentional outlier at the end by eating massive amounts of fat.
My smLDL-P tracks very closely with my LDL-P, yet its ratio of gain-to-loss is much higher.
Final Thoughts
LDL-P and small LDL-P appeared to have a correlation with a three day average of dietary fat as like LDL-C, but with a key exception of adding a two day gap between the dietary period and the blood test.
Small LDL-P appeared to correlate closely with LDL-P, but at a higher gain to loss ratio.
Both LDL-P and small LDL-P proved to be extremely agile and easily ramped up or cleared when observing via daily blood tests (data points 9-14). In fact, shifts in the hundreds of particles per day were easily achieved in either direction.
Next Steps
After having enough of this data in hand, I then realized there was still a problem with proportionality of the other macronutrients. If I was good at holding to a ketogenic ratio throughout this period – 75% fat / 20% protein / 5% carbs – then, technically, one could argue it is actually possible that protein or even carbs are driving the cholesterol correlation. After all, even at a lower volume than fat, they were still proportionally going up and down roughly the same amount, right?
Coming up in Part III – The Divergence… Would testing different ratios of macronutrients change the results?
I’ve thought a lot about how best to get this information out and knowing it will have many possible audiences. So I decided to just put it in a multipart series. This first installment will have some extremely surprising data with regard to dietary fat and its effect on LDL, HDL, and Triglycerides on a low carb / ketogenic diet.
Before we get started I wanted to emphasize two things:
A big THANKS to all the doctors and engineers I swore to secrecy on the deeper data I was sharing. I really appreciate you keeping a lid on this until I had verified and re-verified. (With extra special thanks to the collaboration of Dr. Sarah Hallberg, Dr. Jeffry Gerber, and Ivor Cummins.)
My cholesterol numbers are certainly high, and this might be distracting. I’m well aware of this and it is part of what started this data journey in the first place. But they’ll be more to this story by the time I get to my latest data.
The Brief Backstory
In March of last year, I got a marker for prediabetes from my blood panel, an A1C of 6.1. I was intent on avoiding future Type 2 Diabetes, so I researched and then began a low carbohydrate diet the month after. I lost 35 pounds, it raised my energy levels and seemed to improve my health in almost every way.
However, seven months later in November, I got back my first bloodwork and saw my total cholesterol had gone from 186 to a whopping 357, with LDL-C cholesterol climbing from 137 to 271. Given how incredibly great I felt the entire time, I was in complete disbelief. I heard some low carb dieters had exploding cholesterol, but assumed I wouldn’t, given I didn’t have higher than usual to begin with. From there, I began learning everything I could about Cholesterol Homeostasis.
The Idea
And this is where it got interesting for me. As a senior software engineer, the more I learned about the lipid system, the more I saw a familiar pattern –> a distributed network of objects. I read with particular interest on the assignment and apportionment of apolipoproteins and their potential role as a proxy for an information feedback loop to the liver.
Anyway, I’ll lay the general theory in more detail in a later post. Just know that if true, there should be two key presumptions:
The cholesterol transporting lipid system would prove to be agile — much more so than is typically believed
Discreet patterns should emerge between the diet and serum cholesterol now that disruptive inflammation is lower
So I went about testing these two points with my own blood starting in late November taking regular NMRs (Nuclear Magnetic Resonance – an advanced cholesterol test that directly measures lipoprotein particles and their sizes/classifications). Likewise, I kept meticulous track of my diet and nutrition values, both logging and taking pictures of just about everything I ate and drank. With blood taken every one to two weeks, I would make key changes between tests to see the outcome. Of course I was told over and over again that this was too frequent and that I needed to give the body more time to see results.
Initial Data
Sure enough, in eight tests over just three months I saw rapid changes, enforcing presumption (1). And yes, very discreet patterns did emerge, bringing evidence to presumption (2). In fact, both these behaviors were more pronounced than I could have imagined.
The tightest correlation appeared with total fat in my diet and LDL-C cholesterol. Now hearing this, you might not be surprised given that conventional wisdom seems to agree with the assumption higher fat (especially saturated fat) tracks with higher cholesterol in the blood. But that’s the first big twist — this isn’t a positive correlation, it’s a negative one. And get this – the strongest time tracking appears to be not months or even weeks, but a three day rolling average.
Let’s use an example to illustrate. Imagine I took a blood test on a Friday morning, which we’ll call Day 0. Tuesday is Day -3, Wednesday is Day -2, and Thursday is Day -1 just before that Friday. If you took the dietary fat average I had between those three days, you get the score in the yellow line above. Put another way – all the days before that Tuesday (Day -3) appear to have very little to no relevance to my LDL cholesterol – I can practically ignore them!
To better show the correlation, I’ll flip the vertical axis of this same graph for the three day average of fat. (I’ll be doing this a lot moving forward)
At this point, I took my data to the Low Carb Vail conference and met with a number of great doctors there. I also had a particularly good conversation with Ivor Cummins who is likewise very versed in the subject as well as being a fellow engineer.
I explained the next obvious step was to do a “reproduction test” and see if I could replay the experiment to get anywhere close to the same results. Except this time, it was going to be even more controlled with a tighter timeline. (And frankly, very unenjoyable!)
Reproduction Test
The plan was to have my blood drawn every single day, five days in a row (Monday-Friday). On the preceding days (Sunday-Thursday), I’d have a specific day of meals that matched the two of the extremes in my testing for LDL-P. The highest LDL-P score (3073) was Dec 19th, when I had just 68g of fat the previous day, which we’ll call the “Low Meal Plan.” The lowest LDL-P score (1967) being on Feb 19th when I had 222g the previous day we’ll call the “High Meal Plan.” All day Friday, Saturday, and Sunday I didn’t have a prescribed meal plan and just ate generally “high” fat. We’ll call this period the “General Meal Plan.”
The pattern would be:
Sunday – Low Meal Plan
Monday – Low Meal Plan
Tuesday – High Meal Plan
Wednesday – High Meal Plan
Thursday – Low Meal Plan
Friday, Saturday, Sunday – General Meal Plan (Uncontrolled)
For the first five days I ate the meals in as close to the same times and in the same order as the days they were modeled after. I also timed the blood test to have the same fasting time since the meal of previous day as well.
I was flying to a developer conference that weekend and hadn’t planned to take another test, but finally decided to add one more after I got there on the following Monday. So the Friday, Saturday, and Sunday meals when I went back to higher caloric intake with higher fat would likewise show as well, though they weren’t as controlled as the previous week.
So what happened? See for yourself:
And again – let’s flip that left axis so you can see the reverse correlation better…
This was especially meaningful, not just because it replicated the same three-day average, but because now we could actually see the changes in successive time to the cholesterol payload. (As an aside, the LDL-P had a more interesting story with this run of days. But again, I’ll detail that in the next post.)
The Intentional Outlier
Any engineer reading this is probably thinking the same thing I was by this point. “If this is an algorithm, then it stands to reason I can crank it up to a new level to generate a new extreme.” Which is exactly what I did.
I scheduled a test for April 8th, and for the three days before, I ramped up my diet to an average of 4,274 calories and 349g of fat per day. Just a little heads up… this is extremely hard to do on a low carb diet.
Let’s display all 15 data points together now with this last one on the end…
And again with the reverse axis…
Sure enough, the April 8th result was the lowest LDL cholesterol score of all the tests. Moreover, it likewise tracked to the correlative envelope of the previous 14 data points, giving the entire series a jaw dropping -.905 Pearson score.
Other Data Points
I could write another five posts on the other data points, but I’ll let the graphs speak for themselves with just a brief statement for each.
Three Day Average Total Fat (inverse axis) vs Total Cholesterol
Generally, I don’t care that much for the TC metric, but I thought I’d include it for those who do. It is worth noting that data point 9 started on 3/7 with a 348 TC, just two days later I peak out at 422, then six days later I’m down to 364. Again, this is one more example of where we are told frequently that these numbers take a lot more time to change when clearly my data suggests otherwise.
Three Day Average Total Fat vs HDL-C
While the correlation at 0.742 isn’t as prominent as as the -0.905 of LDL above, there’s no question it is still a very strong relationship. The primary divergence is in the first data point going back to 11/24/15, but otherwise it tracks remarkably close – even during the day-after-day reproduction experiment.
Three Day Average Total Fat (inverse axis) vs Triglycerides
Lastly, we see the trigs are pretty close in correlation as well. Ironically, a 0.6 correlation on at least one metric was what I was originally hoping for when I started this experiment. Yet here’s trigs showing such a strong inverse correlation and it’s the runt of the pack at -0.61. Go figure.
Final Thoughts
While I’m glad both (1) and (2) of my theory above proved to open the door to these emerging patterns, I certainly didn’t think it would be such a perfect inversion of almost everything I’ve read to date with regard to Cholesterol.
My data over 15 data points suggest:
The more fat I eat, the lower my Total Cholesterol (87% inverted correlation)
The more fat I eat, the lower my LDL-C (90% inverted correlation)
The more fat I eat, the lower my Triglycerides (61% inverted correlation)
The more fat I eat, the higher my HDL-C (74% correlation)
Coming up in Part II – Particles, Particles, Particles. Results and analysis for my LDL-P, small LDL-P, and Pattern A/B.
My primary costs are the many frequent and expensive blood tests I take for this research and data. Any size donation is appreciated. Thank you for your support!
Recent Comments