Medical Tests: the more the better

I have two gripes with the medical profession. First, doctors should take more data not less. Also, medical professionals should recognize that most clinical trials assume that individuals are identical, which we know is untrue. I don’t direct these gripes at my doctors because I believe they are deeper thinkers than average and accommodate my eccentricities.

All tests have uncertainties and test results can also be affected by uncontrollable variables in people’s lives. As an example, physicians are reluctant to give the PSA (Prostate Specific Antigen) test to younger men under the pretense that anomalously high readings lead to unneeded tests (such as biopsies), which can have harmful side effects. But a paucity of data is the true culprit, where one bad reading in vacuum has little predictive value. The need for early testing is to set a baseline. It’s the change in PSA levels that should be of concern, not a single high reading.

The plot of my PSA data clearly shows a rise above the baseline starting in 2014. Such a slow rise could have many causes, such as an enlarged prostate – a common condition. In 2014, the data were too sparse to establish a trend, so my doctor suggested that we wait until the following year to see if the PSA levels had dropped. The level had not dropped the following year but was still below that magic value of 4, so we let it go another year. Once it hit 4, my doctor ordered an additional test that gives the ratio of free to bound PSA, which also came out too high, so he sent me to a specialist who ordered an MRI. Eventually this led to a biopsy, a diagnosis of prostate cancer and its removal in 2020. But this is an old story. 

See: https://unknownphysicist.blogspot.com/2020/06/i-was-relieved-i-had-cancer.html and https://unknownphysicist.blogspot.com/2021/01/can-major-surgery-cure-depression.html 

More recently, I got COVID while at a meeting in Seattle, then passed it along to my wife. My children had noticed that their resting heart rates had climbed when they had COVID last year. Since we all have Fitbits, I plotted the values for my wife and me, which appear in the plot here. The trends are remarkably similar.

During my travel days, presumably when I picked up the virus, my resting heat rate was at the baseline value of 47 beats per minute. The day after I returned, my resting heart rate began to rise and I tested positive for COVID on the second day after my return. Given that Omicron has a two-day incubation period, testing positive on the second day of my return is consistent with me being infected while traveling on trains and planes. My wife tested positive two days after I tested positive, consistent with me infecting her the night I returned. My resting heart rate at its peak was 53 BPS, which is 13% above my baseline, while my wife peaked at 17% above her baseline. Interestingly, the Fitbit records the average resting heart rate over the full day, and mine read 55 at its peak in the middle of the day before dropping to an average for the day of 53, so if I had used 55 as the peak, my increase too would have been 17% above the baseline. Here again, we can see that a simple measure of a common physiological metric provides useful health information, and used in combination with other tests could yield even better predictions of health risks that inform behavior. Having a nurse measuring your pulse once a year will only catch huge variation, which might miss important red flags. So, we need more testing!

Two family friends who I believe are both excellent physicians tell me that my metrics defy the conclusions of medical studies. My guess is that there are two factors involved. First, each human being has its own genetic makeup. There are many commonalities humans, such as infections making us sick and our immune system fighting back. But the details of how we each respond depend on our genetic programming. People like me have no problem with COVID, which I got in 2020 at a time when vaccines were not available, and the death rate was high. It’s genetic variability that gave me the strength to play ice hockey while infected during a time when others were on respirators and died.

Population studies take averages of metrics over people of varied genetic traits. If a certain diet is found to lead to a 12% decrease in mortality, it could be that 2/3 of the population experiences a 21% decrease in mortality while 1/3 of the population experiences a 6% increase in mortality. So, making recommendations based on an average value might help a majority of individuals, but could be harming a large minority of them. To take such effects into account, any large populations studies should include a breakdown by individual to determine if subgroups have adverse outcomes. Unfortunately, many studies do not have a large enough sample size for such fine-grained studies of the sort needed to get at the truth. In such cases, the studies are more than worthless to the unlucky minority.

Cancer treatments are an example where researchers and clinicians are now beginning to tailor treatments to an individual’s DNA, which results in much improved outcomes. One must ask about the ethics of choosing a medical treatment based on a population average when it is known that some treatments may not only lack efficacy but might do harm to the patient.

Another difficulty is ensuring that study participants adhere to the study’s parameters. For example, many studies use questionnaires to query participants about their habits. How many people on a self-proclaimed low carb diets cheat by eating a bit of sugar here and there. It’s only human nature to downplay our transgressions and exaggerate our virtues. While N=1 studies are unreliable for generalization to the population at large, they are useful to the important 1; the individual being tested.

In a past post, I described how eliminating fruit while keeping the rest of my diet fixed greatly reduced my blood glucose. (see: https://unknownphysicist.blogspot.com/2023/01/fruit-is-bad-for-you-at-least-it-is-for.html) Obviously, many people do well with fruit but not me. As an update to that study, I recently noticed that my blood glucose had drifted upwards during my bout with COVID. It never got into triple digits, but on average was elevated. Then I realized that I had been using lozenges to soothe my throat. They contain honey and sugars, so taking about a half dozen of them per day was enough to be noticeable. Within a few days of stopping, my blood glucose dropped.

For these reasons, I am an advocate for taking and analyzing more data. In aggregate, the data contains much more information than a yearly reading. Also, I believe that I’m more disciplined than the average survey taker, so I can trust my data to decide on a healthy lifestyle that suits my genetics. After my recent data with resting heart rate, I am motivated to cull my data in search of other interesting correlations that I can use to design new experiments. Until then, happy self-experimentation!

Self Test - Part II: the effect of ezetimibe on visceral fat

 

This post focuses on the power of smoothing, which is a process that picks out trends from noisy data.  We'll see that using a low-cost bathroom scale that has a visceral fat reading, we can see an observable effect that is correlated with the start of taking Ezetimibe, a cholesterol-reducing medication. 

My Yunmai bathroom scale provides a measure of visceral fat, that greasy stuff that lines our organs.  Higher percentages of it in our body increases risks of all sorts of diseases, so visceral fat is a good quantity to track and try to minimize.  The best way to reduce its percentage is to exercise to turn fat to muscle and to lose weight.

My scale has four electrodes, which are arranged in two pairs.  The left and right pair measure the resistance of the soles of each foot.  The resistance from one foot to the other measures the resistance of your body.  Combined with your weight and height, a formula is used to estimate the visceral fat.

I talked to an engineer at Cal Tech whose research was in the general area of biometrics, and she told me that such resistance measurements are related to the visceral fat, but they are not so accurate.  Furthermore, scales such as mine are biased towards the lower part of the body while those that use electrodes that you grasp in your hands is biased to the upper body.  So, corrections need to be made.

The bottom line is that I do not trust the absolute measurement but changes in the reading has some meaning provided that the scale is sensitive enough to detect the changes.  In my case, my reading ranges from 5 % to 9% percent as whole numbers.  The black points in the figure above show the visceral fat percentage as a function of date.  These points form horizontal lines at the whole numbers and alone give us very little useful information.

However, readings fluctuate between 8% and 9% at early times then between 7% and 8% and so on, implying that the visceral fat is falling with time.  Smoothing is the process of averaging adjacent points.  The red points show 50-point smoothing, where 25 consecutive points to the left of the data point and 25 to the right are averaged and plotted at the middle point of the range.  As a result, we get values between the whole numbers.  Suppose that my visceral fat is at 6.8%.  Then, the scale would read 7.0% more often that 6.0%, so the 50-point average would yield something around 6.8%.  Smoothing also eliminates day to day fluctuations that might hide the long terms trends.

The blue line shows 200-point smoothing, corresponding to over a half-year smoothing window.  This eliminates all but the most long-term trends.  The light blue vertical line in mid-2018 shows the date on which I started to take Ezetimibe, a medication that reduces cholesterol.  In my last post, I described how I used smoothing and a fit to a saturated exponential that found that my weight increased by 3 pounds after I started taking Ezetimibe.  The visceral fat data correlates with the observed weight gain.

To summarize the graph, smoothing shows my visceral fat falling from late 2016 and leveling off in early 2018 while I was on my high-fat diet.  This correlates with my weight loss.  Then, after taking Ezetimibe, my visceral fat started a long-term climb from 6% to about 6.5%.  Smoothing has allowed me to determine this rise to a precision that exceeds the whole numbers provided by the scale.

Finally, there is a dip in the 50-point smoothed curve that falls below 6%.  There is also one reading of 5% as seen by the black point.  The date corresponds to one week after my radical prostatectomy surgery and correlates with my initial weight loss then weight gain after the surgery.  Interestingly, the 200-point smoothed curve shows two plateaus – the first after starting the Ezetimibe and the second one after my prostate surgery.

As I have stressed in my past posts, this is a single experiment with potentially many confounding factors, so we cannot conclude that Ezetimibe increases visceral fat.  However, the result provides a hypothesis that could be tested with a larger number of participants.  Putting it all together, Ezetimibe drastically reduced my serum cholesterol (a huge effect that is consistent with controlled experiments) and is correlated with a small weight gain.  This added data shows that it might add to visceral fat.  This brings up the point that medications have complex effects on the body.  They work as intended to treat one condition, but the side effects might oppose some of the benefits.  In this case, the drop in cholesterol is far greater than the potentially ill effects of a small gain in visceral fat.

If there are others out there like me, pooling our data together could increase the confidence that the correlation is not a mere coincidence.  So give it a try!