Something is wrong with the model — take 2

Nearly 4 years ago I wrote a post about the disconnect between the increasing longevity of the US population and its increasing obesity. You can read the whole thing after the ****. The post was titled “Something is Wrong with the Model”. Indeed something is. It doesn’t fit with a lot of data. Those proposing the model don’t like this at all. You can read all about the brouhaha in the 23 May issue of Nature (pp. 428 – 430).

In general I tend to skip medical articles involving meta-analysis under the garbage in garbage out theory. The most egregious example was Women’s Health Initiative when 3 separate meta-analyses of a bunch of uncontrolled studies concluded that estrogen replacement therapy decreased the risk of coronary heart disease by 35 – 50%. The gigantic (161,100 women followed for 12 years, with 1,000,000 clinic visits) Women’s Health Initiative trial of hormone therapy to prevent coronary disease was halted earlier than planned when it was found estrogen based therapies increased the risk of coronary heart disease, stroke and breast cancer.

The excitement was over a paper [ J. Am. Med. Assoc. vol. 309 pp. 71 – 82 ’13 ] which performed a meta-analysis on 97 studies of body weight and mortality which in aggregate involved nearly 3 million people.

A popular measure of weight is the body mass index (BMI) which is weight in kiloGrams divided by your height in Meters squared. Not something which is obvious. If you want to figure yours know that a kiloGram is 2.2 pounds, and a meter is 39.37 inches.

At any rate a BMI over 25 is considered overweight, and one over 30 is considered obese. At 6 feet 1+ (which I used to be) a weight of 190 puts me at 24.69. To be obese (BMI over 30) I’d need to weight 228 (which I almost did 50 years ago).

When you plot BMI vs. probability of death you get a U shaped cure, with the very thin and the very fat showing increased risk of dying (mortality). The Nature paper is interesting as it shows 6 curves for people at ages 20, 30, 40, 50, 60, 70. As one might expect the curves for each age lie below the next oldest. All of them rise with BMIs under 20 and over 30, so there’s no argument about whether obesity is bad for longevity.

Well, if the curve is U shaped, it has a minimum. The excitement comes in because the healthiest weight (the minimum) is a BMI of just over 25 for those in their 60s and around 26 for those in their 70s. Also in ALL 6 age groups the curve is pretty flat between 25 and 30, rising on either side of the range.

Naturally people who’ve invested their research careers in telling everyone to diet and that weight is bad, don’t like this, and a symposium involving 200 unhappy people convened 20 February at the Harvard School of Public Health is described, along with a lot of the back and forth between the authoress of the study (Flegel) and Willett of Harvard who didn’t like it one bit. The best comment IMHO is from Robert Eckel “We’re scientists. We pay attention to data, we don’t try to un-explain them.” Read the article, it’s well written and there’s a lot more.

One final point, which might explain why the minima of the curves shift to higher BMIs at older age — which the article didn’t contain. People lose height as they age, yet the BMI is quite sensitive to it (remember the denominator has height squared). The great thing about BMI is that it’s easily measured, and doesn’t rely on what people remember about their weight or their height. Well as a high school basketball player my height was 6′ 1”+, now (at age 75) its 6’0″. So even with constant weight my BMI goes up.

Well it’s time to do the calculation to see what a fairly common shrinkage from 73.5 inches to 72 would to to the BMI (at a constant weight). Surprisingly it is not trivial — (72/73.5) * (72/73.5) = .9596. So the divisor is 4% less meaning the BMI is 4% more, which is almost exactly what the low point on the curve does with each passing decade after 50 ! ! ! This might even be an original observation, and it would explain a lot.

****
Something is wrong with the model

Back in grad school when a theory came up with a wrong prediction, we all clapped hands because it showed us exactly where a new theory was needed, and just how it failed. No casting about for something to work on. A program that crashes intermittently is very hard to fix. Once you’ve found input that consistently makes it crash the job becomes much easier.

The Center for Disease Control released new data for 2007 (based on 90% of all USA death certificiates) showing that mortality rates dropped again (by over 2%) to 760/100,000 population. It’s been dropping for the past 8 years, and viewed longer term is half of what it was 60 years ago. Interestingly death rates from heart disease dropped a staggering 5% and even cancer dropped 2%.

But the populace is fat and getting fatter. This has been going on for 30 years. You can Google NHANES for the gory details, but the following should be enough. [ Science vol. 299 pp. 853 – 855, 856 – 858 ’03 ] The data from a recent NHANES (’99 – ’00) shows that the percentage of obese (as opposed just overweight) increased from 23% in the surveys from ’88 to ’94 to 31%. This is based on the body mass index (BMI). Someone 6′ 1″ would have to weigh 225 pounds to be obese.

We are told to be prepared for an epidemic of diabetes, high blood pressure, elevated blood lipids because of this. Every doc has seen blood sugar drop, blood pressure lowered, lipids come down in people with any/all of the above when they are able to lose a significant amount of weight. These diseases are significant only if they kill people, which they certainly seem to do in my experience. The next time you’re visiting a friend in the hospital, look at what’s lying in the beds. Very likely, many more than 31% of them are obese.

So why are death rates dropping and people living longer? Something must be wrong with the model — it’s pretty hard to quarrel with the data as being inadequate. Certainly the increased incidence of obesity should have produced something by this time (it started 30 years ago).

Well, the self serving answer for the drug developers is that their drugs are better. MDs would like to think it’s due to better care. Possibly. Here’s some detail.

#1: More people are exercising than they used to. How many joggers and walkers did you see on the streets 20, 30 years ago?

#2: Fewer people are smoking. Forget lung cancer (if you can). The big risk for smokers is premature vascular disease. Normally we all have carbon monoxide in our blood (it comes from the breakdown of hemoglobin). [ Brit. Med. J. vol. 296 pp. 78 – 79 ’88 ] Natural carbon monoxide production would lead to a carboxyhemoglobin level of .4 – .7%, but normal levels in nonsmokers in urban areas are 1 – 2%. Cigarette smoke contains 4% carbon monoxide, so smokers have levels of 5 – 6%. This can’t be good for their blood vessels.

#3: Doctors know more than they did. My brother is a very competent internist. He took over the practice of a similarly competent internist after his very untimely many death years ago. Naturally he got all the medical records on the patients. He found letters (now over 25 years old) from the late MD to his patients informing them of their lab results, and assuring them that their cholesterol was just fine at 250 mg%.

#4: The drugs are better. In addition they may be working in ways that we have yet to fathom. Consider the statins — their effect on vascular disease is far greater than their effect on blood lipids (cholesterol, triglyerides) — particularly when compared to other agents that lower blood lipids to the same extent.