In this issue, the article by Olsson et al* gives an example of the use of Design of Experiments (DOE) applied to a person who is having problems sleeping. These authors have worked very hard to present this example in a way that the reader can follow along and repeat on his/her own. Design of experiments is an old idea in statistical process control and is described in introductory textbooks for engineers. It has been used in health care for the manufacturing of drugs, but is not apparently in use for individual patients.

This approach could be relevant for people with chronic conditions that must be self-managed over long periods of time, including hypertension, diabetes, stress, arthritis, asthma, and insomnia. It can also be used to monitor cigarette smoking, diet, or exercise. If you are not familiar with this method as applied to individuals, this may be one of the most important articles you will ever read.

Why would I risk my reputation with such a bold statement? This kind of reasoning and approach is at the core of the coming era of individualized medicine. To greatly oversimplify human history, in the first stage of medicine, disease had a common cause and therefore a common treatment. For example, prayer and pilgrimage might cure any illness. Bad air or miasmas caused illness and the cure was to live somewhere else. Or the combination of the 4 humors caused all illness. In that era, no physical examination of the patient was needed. Around 1820, patients began to be systematically differentiated by type of disease: tuberculosis, cancer, heart disease, etc. This required examining the patient. Autopsy results were linked to prior symptoms to create disease classifications. Treatments matched the disease. The third stage is the individualized treatment of patients, based in part on their unique genetic make up. Even for 2 patients with the same disease, for example skin cancer, the therapy will be unique to each. This could include adding parts of the patient's own DNA to the drug. Design of experiments will be an essential companion to this approach or be used on its own. (Actually all 3 stages occur at the same time, so it is a matter of historical emphasis. It is not easy, after all, to summarize the history of medicine in 1 paragraph.)

To see what a difference individualized evaluation and care would make, consider the following example using hypothetical numbers. Randomized clinical trials that compare classes of patients are stage 2 medicine, while DOE is part of stage 3 medicine. Consider a typical randomized trial for a chronic condition. One hundred patients receive the new drug and 100 similar, blinded patients receive a placebo. In the experimental group, 25 patients benefit and for the rest there is no difference in outcome. In the control group, only 5 patients benefit and 95 are not affected. This difference is unlikely to have occurred by chance and the difference is statistically significant. The manufacturer is ecstatic. Their new drug is on average better. It is widely advertised. The price is set at $1000 for a year's treatment. One in 10 patients has mild side effects. Randomized trials are often too small to measure these second-order effects. These side effects are, on average, about one tenth of the size of the beneficial impact of this drug. That is to say the quality of life for 1 benefited patient would be cancelled out by 10 patients with side effects.

Now consider the consequences of this evaluation on 1000 patients in your health plan who now take this drug. The total cost is $1,000,000, and 250 of these patients actually benefit. One hundred have side effects, which is equivalent to 10 patients who benefit. These 10 benefit equivalents need to be subtracted from the total benefit. (250 - 10 = 240 patients who benefit) The cost per patient who benefited is 1 million dollars divided by 240, or $4166.67. Now consider the use of the DOE. Each patient carries out his or her personal evaluation following the method described by Olsson et al to see if the drug works for them. Of the 1000 patients, the 250 who benefit take the drug and 750 do not. Of the 250, 25 have side effects, which is equivalent to 2.5 patients who benefit. The net patients who benefit is therefore 247.5. The total cost is $250,000, or $1010.10 per patient who benefits. This cuts the cost by 75% and increases the benefit by 3%. Feel free to change any of these assumptions and rerun this analysis to suit yourself.

If this speculation is at all realistic, it affects a lot of health care players. For example, the drug companies will sell fewer drugs. For entrepreneurs, the field of individualized diagnostics is a place to be. The Food and Drug Administration (FDA) will have to rethink what it does. Some drugs that have failed in clinical trials, and therefore failed FDA approval, may still be beneficial to a few patients. For patients, participation in care will be more essential than ever. Third-party payers need to encourage this approach to gain savings. Patients are not going to become personal quality engineers, but they could have access to computer software in many forms to help do their own DOEs. Such personal DOEs will not be carried out with the statistical rigor that a true statistician would admire, but the basic logic of DOE would drive these personal efforts. Quality managers will increasingly be asked questions like the following: "What percentage of your patient population are getting drugs that do not benefit them? How are you set up to coach your patients who want to do their own DOEs?"

But, do not take my word for this. Read Olsson et al. Share it with some colleagues. Do one of these experiments on your own, and share and discuss the results. Then decide for yourself. If you do this now you will get the basic idea and you can have bragging rights about your new knowledge.

Duncan Neuhauser, PhD

Associate Editor

^*Olsson J, Terris D, Elg M, Lundberg J, Lindblad S. The one-person randomized controlled trial. Qual Mang Health Care. 2005;14(4):206-216. [Context Link]