Sunday, January 16, 2011

As I Was Saying

I meant to gripe in more specifics about what bothered me about the cursory health screening that the health insurance company ran, but then I got distracted by trying to make sense of things like this:

and with diagramless crossword puzzles, popular culture of 20-30 years ago, and various other puzzles.

So where were we. Yes, bothered by the health insurance screening. As I do love the format of the numbered list, let's go with that.
  1. The measured weight with what looked like a fancy version of a digital bathroom scale, not a "doctor's office" scale. How accurate is this machine? Does it every need re-calibrating after they drag it around to workplace health screenings all over the region? How accurate is it to weigh people in the winter, wearing heavy clothes? Oh, none of that matters, as they then rounded the weight to the nearest 10 pounds and looked that number up in a BMI chart that was rounded to the nearest integer. This seems like a really, really bad way of assessing an individual's risk of health problems based on BMI (especially since the BMI is most useful on the population level). It is less bad if this data were being used in the aggregate by the actuaries at the insurance company to set the rates for my employer. But I hope that people don't get nagged by the insurance company over their BMI numbers from this measurement. (Mine came out as 18, when I'm more of an 18.5.)

  2. For things like cholesterol, it's unclear to me what "risk" is being predicted based on levels and how medical science defines and calculates this risk. I find it highly suspicious that every cut-off number from the entire health screening is a "nice number" in our base 10 number system. Would a cholesterol level of 0xC8 be as catchy as one of 200?

  3. So I'm wondering where these various nice numbers came from. Did they take their initial population and bin them based on some metric (with the bin cut-offs all at nice numbers) and then track them and then do some sort of alarming ANOVA on the groups? (I hope not!) Did they do some sort of categorical measure of their population (had heart attack within study time period: yes/no)? If they did a logistic regression, what probability of heart attack (or other bad outcome) did they decide was too high of a risk for setting the levels? And then how much did they round to get nice numbers like 200 and 250? If they only tracked one bad outcome, like heart attack, then they really could have missed a meaningful effect of having a level that is too low.

  4. More on rounding and "nice numbers." There was also no sense of how much things were rounded and how much small differences mean something. Also no sense of the accuracy of the tests. If they ran the test twice in a row, how close would the numbers be? For the blood glucose test, the normal range was described as 70-99 and with various levels of diabetes risk being over 100. Mine came out as 68. Are two points meaningful here? Or is that difference in the rounding noise? Is lower than normal good or bad?

  5. There was no context provided for the numbers. For examples, my blood lipid numbers came out as:



















    TestScoreRecommended
    Cholesterol139less than 200
    LDL72less than 130
    HDL54greater than 60
    Triglycerides62less than 150
    My HDL is too low. But no advice was forthcoming about whether you needed to be on the "good" side of all these measures or if three out of four is good enough. Additionally there was no advice about how to raise HDL. Other than their all-purpose advice -- eat modest amounts of foods generally accepted as "good for you" and get plenty of exercise. Already doing that, thanks.

  6. So, in summary: I found that the health screen was probably more useful to the insurance company (in setting rates) than to the individuals on the plan. No matter what your numbers were, you were told to eat less and exercise.