Stat counter


View My Stats

Saturday, January 30, 2010

The trap of small numbers

Toyota has a problem. While Toyota is generally referred to in medical circles in reference to its lean processes and potential application to delivery of health care, there are parallels in other areas as well. Toyota has announced it will stop production of a large segment of their product line in order to address problems with uncontrolled acceleration. In 2008, there were 52 complaints to Consumer Reports relating to this problem. This is in the background of the sale of almost nine million cars, about two thirds of these in the US. To make the math simple, let us assume that Toyota sold about five million cars in the US. That means this issue appears to have been reported in approximately 0.0001% percent of sales.

When you start to look at low frequency events like this and try to make sense out of causality, it is problematic. It is hard to to separate random events from linked events, signals vs. noise. This would not be a problem except that low frequency and often spectacularly bad outcomes tend to drive some sort of response and responses mean investment of time, effort, and resources. Again, all of this would be fine if that investment yielded some some of result where the inputs were exceeded by the outputs.

I know I run the risk of critics who might claim that when lives are at stake, money is no object. That is nonsense. If that were the case the only thing we should ever invest in is to save lives. We utilize resources to further other ends, resources which could be allocated to save someone's life. We need to recognize that there is more at stake in living our lives than life and death (see previous blog).

Reaction to low frequency events is not limited to the automobile industry. These types of responses are common to virtually every realm of human activity. The TSA is a huge investment in response to low frequency events. In this case there is an unambiguous link between terrorists blowing up or crashing planes and human lives lost. Whether there is a link between the huge sums invested in TSA and the frequency of lives lost is an open question.

Medicine is absolutely riddled with dogmatic recommendations based upon responses to low frequency events. Read any drug monogram and you will see exhaustive lists of reported symptoms and outcomes which occur in vanishingly small percentages of patients which may or may not occur with any greater frequency than in any control group. Rare disease associations with mundane symptoms or non-specific clinical findings are highlighted in such a way that we respond as if the outlier events were the norm. We develop surveillance programs for drug and disease monitoring which are driven by risk avoidance, even if the risk is vanishingly small or non-existent and the components of the surveillance program may create even more risks than they alleviate.

There is a tendency to blame the fear of lawsuits as the driver of this culture. While this may be true to some degree, I have been in medicine long enough to see that  perhaps we have gotten our sequence backwards. Even before malpractice concerns were front burner we adopted a culture within medicine that no amount of effort or resources expended were too great. An inexpensive intervention or approach which yielded a good outcome in 98% of the time would be ditched for something that costs 10x more for a good outcome 98.5% of the time. At least part of the driver might have also been that more expensive approaches yielded higher margins as well. Everyone came out ahead...sort of.  As the medical profession  adopted expensive approaches to deal with vanishing small improvements in patient outcomes and these approaches became standard of care, the legal profession was all to happy to use this as a business opportunity.

While my experience with being an expert witness is limited, what is common to the dozen or so cases I have participated in is all of them deal with low frequency events. The gut response to such rare and catastrophic occurrences is to look for changes in how one deals with mundane clinical encounters which may result in spectacularly bad outcomes. What to do when a patient presents with a headache, or fever and a rash, or a productive cough? There are unquestionable anecdotes where each of these scenarios is followed by the tragic death of those affected, despite what appears to be appropriate conservative management.

Why not maintain a full court press all of the time and do as much for anyone who might be at risk for a bad outcome? The problem is we live in a work were resources are not infinite and use of resources for a non-productive end means that they cannot be used for something with a more productive outcome. It seems obvious from 20,000 feet but is never obvious when functioning one patient at a time.

So, what to do? Much of the decision as to what to do will boil down to needing data. When we are tempted to act to forestall some bad outcome which occurs very rarely, we need to identify just how likely the bad event might be. I am not talking about defining possibilities down to precise numbers but trying to begin define them in terms of orders of magnitude. Do these things happen 50% , 0.5%, or .0005% of the time.  Also, there is a tendency to track surrogate endpoints because measuring what is really important is so difficult or the events which we are trying to avoid are so rare. If the endpoints we are trying to avoid are so rare that we have a hard time measuring them, perhaps that says something about where they should be viewed as priorities.

Ultimately is comes down to culture change. We must change the culture within medicine where physicians begin to understand numbers, recognize their limitations, and convey these concepts to patients. We must change the expectation of patients where they realize there are always risks and that all interventions carry potential unintended consequences. It is a tall order.

No comments: