A good decision is based on knowledge and not on numbers.   Plato

     Few people have a natural liking for or intuitive understanding of statistics, a subject that influences your thinking on a daily basis whether you are aware of it or not. Statistical numbers have migrated out from professional medical journals to appear in the TV ads, talk shows and headlines that greet you with your morning coffee. In the current environment of sound bite reporting, some understanding of statistics is beneficial in sorting through medical information.  Statistical knowledge is also a good antidote to fear mongering and sensationalism – unfortunately common in the age of instant communication.

   The media barrage of medical stories and advertising coincides with huge expansion of the pharmaceutical and medical device industries over the last several decades, and with the decline of paternalism in medical practice which has encouraged patients to become active participants in health decisions formerly made by their doctors. Because statistics is not an easy subject to master, statistic-laden stories can be superficial.  Often they present only enough information to make the casual reader think the story’s topic is clear cut and relevant to him.  The first question for the reader or listener to ask himself is how to interpret the numbers presented. The second question is why the story is being presented. Who benefits besides the reader – and does the reader really benefit?

    Studies employing statistics begin with a scientific hypothesis. Will drug X prevent kidney failure in diabetics? Using well-demonstrated methods, researchers design a study which will test the drug in some people, but not others, and they will compare what happens over a specified time period, looking for certain “endpoints” such as abnormal blood tests, need for dialysis, progression to transplantation. Information collected becomes the “raw data” fed into statistical analysis.  Since medicine involves biological systems, natural variance between individuals affects results. Significance of findings is stated as a probability (P-value) that the results are not due to chance. That value is usually between .01 and .05.

   The numbers that emerge from statistical manipulation of data often look more respectable than the data itself. Take a headline headline that warned of esophageal cancer rates rising eight-fold with rising temperature of tea consumed. A thinking reader would immediately wonder how the researchers found enough people willing to measure tea temperatures, consume their tea before any cooling occurred, and do it the same way over the many years it takes to develop cancer. In fact, the report came from subjects’ responses to words describing their preferences – very hot, hot, medium or lukewarm. Selection methods, measurement accuracy and study design, all very important in research, are not necessarily reflected by headlines or in short reports.

    Very few people have the time or inclination to track down and analyze the original studies and raw data behind a medical headline.  The headlines are most often presented in as percentages- percent improved, percent prevented, percent cured.  A good rule of thumb is to assume that  a number  presented as a percentage is derived from raw data that is not as impressive as the derived percentage. Raw numbers do not have the same emotional impact on the reader that percentages do.  Take for example, a headline about a triumph for a statin drug, Crestor: “Cholesterol drug cuts vein clot risk by 40 percent.”  Forty percent sounds impressive. This story is long enough to show some raw data – if you are patient enough to keep reading through many paragraphs:

    “The clinical test was carried out on 17,802 subjects, both men and women in good health. During follow-up, 34 participants in the rosuvastatin (Crestor) group and 60 in the placebo group developed symptomatic VTE (venous thromboembolism), a 43 percent reduction.”

     These are numbers that give you some sense of proportion.  Fewer than 1 out of every 200 patients in this study had a chance of developing a leg clot in two years. And if you track down the original study, you find that the data comes from the now famous Jupiter study, which was designed to test the hypothesis that lowering LDL cholesterol helps prevent heart attacks in people who have high C-reactive protein levels (a laboratory marker of inflammation). Leg vein clots were not an endpoint to the study, but an interesting positive finding in the statistical analysis (a retrospective analysis that mined the study data for things other than the original endpoints).  None of this negates the fact that 34 is 26 fewer than 60 – a 43% reduction. But it does make you realize that a lot of people took a lot of medicine over a long time without having leg clots or being prevented from having leg clots. You also note that while they were said to be in good health, they were selected for the study because of elevation in a laboratory test that is a marker for inflammation. What would the rate of leg clots be in people with normal CRP levels?  Would Crestor make any difference in them?   That, we do not know.

    Who benefits from a story like this one? Without any media hype at all, the original study accomplishes its main purpose –  contributing a few more data points to the accumulating body of knowledge about how these drugs work, and inching medical progress forward. But in the age of health anxiety, medical stories sell newspapers and magazines and attract advertisers on TV, radio and the internet. And in the age of competing statins, manufacturers are happy to have any edge they can get in promoting their drug.  A healthy percentage of something positive related to a drug or a procedure makes the news.

    The careful, educated reader will benefit from reading with a critical eye and understanding that percentages – whether cures or risks or mortality – all describe groups of people, not individuals. Each patient, though,  is an individual in whom the percentages are always 50/50 –a cure will happen or not; a risk will play out or not; a life will end, or not. Each medical decision should be an individual one, based on knowledge of the whole person and on the most knowledge possible about the undertaking being contemplated – not just on data points in statistical analyses. and each medical.

    One last point – all studies of patients’ responses to medical treatment are confounded by the placebo effect – a very real phenomenon related to the mind’s ability to help fix problems when it believes that the treatment will work. Some estimates are that the placebo effect accounts for 30% of positive results,  no matter what the treatment is. That is a statistic worth remembering.