Saturday, February 20, 2010

Chess, Computers, and the Future of Clinical Research

I just read an outstanding article titled "The Chess Master and the Computer" by Garry Kasparov in The New York Review of Books (Vol. 57, No. 2, 2/11/10). In this article, he recounts the Man vs. Machine battle of chess in the late 1990s and then explores the implications, psychology, and meaning of artificial intelligence overtaking humans in one of the greatest, timeless, and most complex games invented. I highly recommend googling and reading the article.

I love chess. As mentioned by Kasparov in the article, it is an amazing game because all the rules are defined and all the relevant information is available to players. Chess is predominantly a battle of strategy, reasoning, and planning whereas a game like poker also employs elements of psychology. But because chess is a transparent game, the information can be fed into a computer and with the right algorithm, brute force tactics can be employed; computers can try every single possible move and look at millions of possible scenarios to determine the optimal move. Interestingly, Kasparov argues that there has been much technology but little innovation; newer chess programs simply take advantage of faster processing speeds without changing the base algorithm.

I wonder what we can take from this in terms of computers aiding clinical decision making. The problem becomes even more complex with medicine because the information is incomplete and the rules are undefined. Take something as simple as a baby aspirin for primary prevention of strokes or heart attacks. Aspirin is well defined in secondary prevention - it reduces morbidity and mortality in patients who have cardiovascular disease such as coronary artery disease. But aspirin's risk-benefit balance is unclear in primary prevention - the effectiveness in healthy patients with no medical problems. Aspirin's risk is major bleed including gastrointestinal bleed and hemorrhagic stroke.

Could a computer help clinical decision making here? Perhaps. Randomized clinical trials, the gold standard of clinical research, have enrolled tens of thousands of patients, but the bottom line for most practicing physicians is the conclusion that mortality does not seem to change and the benefits and risks balance each other out. Computers, on the other hand, can process so much more information. What if computers knew unlimited information about each patient enrolled? Could the computer find trial subjects that were similar to the patient? Could patients be matched to research participants in terms of age, gender, frequency of doctors' visits, other medical problems, family history, ethnicity, alcohol intake? Could we compare even more obscure factors: socioeconomic status, job, marital status, impulsive or high-risk behaviors? Could a computer then identify a similar person and make an accurate prediction about whether aspirin would help? This level of processing would be impossible for a human, but for computers, the limiting factor is a database.

I think this may be the future of research. It reeks strongly of genomics, which has influenced me greatly. Computers can help us move towards personalized medicine by taking enormous bits of information - gene expression of 20,000 genes or a database of a hundred thousand research subjects - and extract from that a conclusion specific to the patient. It also departs from the dogma of clinical research that randomized controlled trials (RCTs) are the best source of data. I know this is heresy, but someday we might say RCTs are limited because they lump large numbers of people together when instead we should be individualizing and personalizing medicine.

Image from Wikipedia, shown under GNU Free Documentation License.

No comments: