Why haven't we always done research this way? We treat patients one at a time, using all available knowledge. The early statisticians and trial design specialists were explorers who valiantly sifted through data to generate hypotheses and test them. They used all the data they could find. Then frequentist statistical thinking began to dominate. It arose from agriculture research in which large fields could be planted in various ways, manured, and, much later, the crops harvested and assessed. The frequentist would then make inferences about the population from which the samples were drawn and about future samples that could be drawn from the same population. These slow, plodding, smelly trials may be suitable for crop cycles, but patients are at your door needing care. You can't send them away until the study is harvested. You must treat them using all your knowledge now.

The second problem with traditional frequentist statistics is that it focuses on the central tendency—the mean. For approval, the U.S. FDA demands two studies that have shown the means of two samples, for example, your test drug versus placebo—to vary enough that the alpha error is .05 or less. But we don't treat patients in herds. Unfortunately, the U.S. FDA thinks of patients in just that way. What is the proper dose of digoxin if you treat 100 patients each with the same dose? The answer might be 0.5 mg per day orally, but some with hyperthy-roidism and atrial fibrillation will be undertreated and others with hypomagnese-mia or hypokalemia or renal failure will be overdosed.

This is what is so malignant about the U.S. FDA restricting access to drugs until their frequentist statistical studies of herds has resulted in data that pleases them, rather than focusing on the patients in need now. It also leads the U.S.

FDA to believe that their crisply defined sets of diseases are correct, when every practitioner knows that patients come in very fuzzy sets of joint pain, not classical Still's Disease, or cancer, not the stage IIIB lymphopenic tumors that made up the herd that won drug approval. No wonder so many cancer patients are treated off-label—their practitioners are classical Bayesians who use all information to select the best treatment regimen and view patients as individuals, not members of herds.

Today, with hyperlinked data and high-dimensional multivariate analyses, patients can be studied as though they were individuals with some common features. By studying such patients thoroughly, we can learn the predictive characteristics that will allow us to choose the best therapy. We may not always be correct, but we can always make the best decision based on all the available information. That is why it is so critical that the U.S. FDA stop its embargoes on data that do not fit its definition of approved, crisp diagnostic sets in its approved labels. Under any circumstances such censorship is insulting to health practitioners charged with the best care of their patients. When the patient is suffering and dying, such censorship is immoral, unethical, and reprehensible.

Why use half a brain if you have both halves? We are the victims of the left-brain, frequentist, post-hoc thinking. After the study is over, we analyze it to death and we become the world's best Monday morning quarterbacks (perhaps Tuesday morning, after the Monday night game). But we've lost the resilience of our ancestors who learned to cope, predict the best they could, innovate, try, and make it work. Of course, those primitive Bayesians must have known how to bet when in the past they sailed thousands of miles in Polynesia with the stars and a few sticks or today open a hot belly and master whatever they find.

Tomorrow our newborns will be problem solvers in virtual reality, surfing libraries, sampling historical CD-ROMs, invoking sophisticated analysis programs, and consulting online oracles all for that third-grade term paper. They will visit museums where there are pens, dictionaries, newspapers, and linearly scheduled TV programs, and they will wonder how the Nowanderthals ever survived without real-time translators, voice input, and attention-focusing knowbots. Too futuristic for you? Remember how shocked you were at 2001: A Space Odyssey, when it appeared, and how contemporary it seems today.

We can leap forward toward tomorrow, enhancing the ethics of our research by reducing needless patient exposures and reducing the time and cost of therapeutics development, through elegant braiding of chemistry, toxicology, and clinical studies supported by comprehensive informatics and a relevance diagram map. There will still be risk and failure and expense, but more patients will live through earlier access to innovative treatments, and more patients will be able to afford them.

0 0

Post a comment