The power of big data and cognitive computing technology is lost on no one. Its areas of application are almost unlimited and range from the analysis of biochemical networks to management consulting and the Twitter-sphere.
IBM’s Watson is 1 of the most conspicuous players in the arena. It is an advanced question-answering computer system that is perhaps most famous for winning the quiz show Jeopardy in 2011.1 It works by taking questions formed in simple English, parsing them into keywords and phrases, and then applying hundreds of language analysis algorithms simultaneously to find phrases in its database that are associated with the question. The more algorithms that find the same answer, the more likely Watson is to be correct.
This simplicity belies the sophisticated technology responsible for its power, composed of advanced natural language processing, knowledge representation, and machine learning capabilities.2 2011 was also the year that Watson broke on to the healthcare scene, when IBM partnered with Columbia University in New York, and the University of Maryland in College Park to identify the best way to exploit its clinical decision-making technology.3
In October 2016, IBM and Quest Diagnostics, a genomic sequencing and oncology diagnostics company, in collaboration with Memorial Sloan Kettering Cancer Center (MSKCC) in New York, and the Broad Institute of MIT and Harvard in Cambridge, Massachusetts, launched Watson Genomics, promising to revolutionize personalized cancer care.
Researchers at IBM believe that oncology can uniquely benefit from the powers of Watson, with specific promises to make treatment recommendations and connect patients with ongoing clinical trials. Its procedure goes as follows: first, Quest will sequence the genome of the tumor. Watson Genomics then takes this product and compares the specific genetic mutations in the tumor to those at its disposal. This now includes not only the medical literature available publicly, but also MSKCC’s and the Broad Institute’s databases.4
There are, however, several obstacles that may prevent contributions by Watson to clinical care.
Clinical trials are carefully designed to expose 1 or many variables to establish causal links in oncology. But there are confounding variables and hidden biases that complicate the studied phenomena, though researchers take many steps to limit their influence. They publish research protocols and statistical goals before the research is undertaken. They randomize subjects and gather and compare ethnographic and clinical information. Even with these techniques, however, there remain elements that pervert the validity of a trial.
“Controlling” for certain variables exists on a spectrum, not in a binary fashion where they are either “controlled for” or “not controlled for”. There are unforeseen factors and inherent biases in design and implementation that influence study outcomes. It is unclear whether considerations of external validity, or the extent to which a study bears on a given patient or population, also pose a problem for Watson.