Golden age of population studies from health IT

You start to isolate cases and causes, and your conclusions become much firmer, as the population you study grows. It's the law of large numbers in action. This is what health It can deliver.
Written by Dana Blankenhorn, Inactive

In automating health records insurers and the government have two key goals in mind.

  1. By keeping records together patients save money and time. You don't have to fill out a paper form each time you see a new doctor. Tests follow you and don't have to be retaken. Everyone in your "circle of care" can see all your relevant data, which reduces mistakes. Billing is simpler.
  2. By collecting and sharing large amounts of data, we can find what works and what is cost-effective. Comparative effectiveness may be controversial but it is sound science with clear benefits.

As numbers get bigger small trends get clearer.

One of the most active areas of medical research today is the population study. I get press releases on these every day, often with scary headlines.

Like this one. Study says two sodas a week raises pancreatic cancer risk. Soda industry calls study flawed.

What happened is that Noel Mueller of Georgetown looked at data from the Singapore Chinese Health Study, a 14-year survey of of about 60,000 people. Of those people 140 got pancreatic cancer. From those cases, Mueller found an 87% greater risk of cancer from drinking just two soft drinks a day.

(The image above is from the National Institutes of Health, and combines the two colleges that worked on this study, the National University of Singapore and the University of Southern California in the U.S.)

The soda industry says hold on a minute. Of those 140 cases 110 drank no soda, and 12 more drank less than two a week. Mueller drew his conclusion from 18 cases.

It's a fair criticism. But what happens when you start collecting records routinely, using health IT.

Now instead of 60,000 cases from Singapore you have tens of millions of cases from around the U.S. From those you can strip out all the variables you want and still have thousands of cases where one variable made a big difference.

This is true across the board.

Take a story I did here 14 months ago, on hypertension, a subject close to my own heart.

Two different drug combinations were tested on 11,500 people. Over three years 679 on one combination had a heart attack, stroke or bypass, while 552 had such an emergency on the other combination.

The first combination pill included a diuretic, along with a standard ACE inhibitor, the second an ACE and a jack called a calcium channel blocker.

Do some math and you find 11.8% had trouble on the first set of meds, 9.6% had trouble on the other. That's a difference of 2% -- two people out of every hundred.

From this data the test was stopped and a recommendation made. I took my own advice, got with my doctor, and made the change to the ACE and the jack.

I can't talk about risk of death, but I feel a lot more like 21 again. Exercise and coffee both have diuretic effects. I was falling asleep in mid-afternoon regularly. Now I can stay at my desk and sleep better at night, too.

Your mileage will vary. But what happens if you can effectively perform such a study on data from 6 million people, or 60 million? You start to isolate cases and causes, and your conclusions become much firmer, as the population you study grows. It's the law of large numbers in action.

This is what health IT can deliver, large numbers. It opens up spectacular new avenues of research, because smaller differences can be isolated, and larger pools of data can be analyzed. Small percentage differences are magnified when they occur across millions of cases instead of thousands.

Getting from here to there will not be easy. All the data has to get into the same format. But we are not talking here of any risk to privacy. We are talking of analyzing numbers, not cases.

A lot is going to be learned. The next decade is going to be a boom time for population studies, and as a result medical practice is bound to be quite different in 2020 from what it is in 2010.

Editorial standards