Can Big Data Make Our Healthcare Systems Better?

"The corridor was empty, except for the two maintenance drones which tirelessly disinfected the walls, ceiling and floor 24/7. In room 109, a large Heart Emergency Unit was setting its electrodes on Mr. Doe's chest: Tammy, the automated brain behind the health monitoring systems, had detected the precursor signs of a potential cardiac arrest."

Fiction? Maybe. But two searchers have devised an approach that shows better results than treatment-as-usual performed by real doctors. The latter have to face multiple challenges: an always increasing complexity and costs, the numerous treatment options, with new ones added daily, the delay between researches and practice in the field, the multiplicity of information sources to cite but a few.

A computer can analyse hundreds of options in a short time, finding the costs and benefits of each and determining what course of action would be the best under a number of constraints. It can also take into consideration new treatments and discoveries faster than a human doctor would be able to.

As the therapeutic options expands, so do the specialization of doctors: training is a lengthy and costly process, and the limited human abilities force doctors to increasingly specialize. The result is that possible treatments out of a doctor's area of specialization might be ignored, resulting in a potential increase in cost and decrease in patient outcome.

Two searchers have used Artificial Intelligence techniques: machine learning, Markov Decision Process (MDP) and Partially Observable Markov Decision Process (POMDP) to model Dynamic Decision Networks (DDN) for therapeutic options, such as continuing a treatment or stopping it and so forth. The results are encouraging and the outcome was a modeled better patient outcome as defined by the Outcome Rating Scale (ORS) of the Client-Directed Outcome-Informed (CDOI). It also showed a number of improvement with lowered costs.

Remains the question on how these new techniques will be used: it can be for "good", i.e. to help doctors designs better treatment options, but also for "bad", for instance by insurance companies to set a maximal reimbursement based on the number calculated by the system.

The paper has been published on arXiv as arXiv:1301.2158v1 [cs.AI].