Artificial Intelligence in behavioral healthcare: Improving care and reducing cost

A couple of weeks ago, Indiana University researchers Casey Bennett (also of Centerstone Research Institute) and Kris Hauser(Computer Science Assistant Professor, Indiana University) published an article on Artificial Intelligence in Medicine detailing how a simple machine learning algorithm could simultaneously improve behavioral healthcare outcomes while reducing cost. While this is a stated goal of health reform, the healthcare industry has a poor track record of achieving that goal, particularly in the United States as illustrated in the following slide from Mary Meeker’s report, USA, Inc.

(See Figure 1: Healthcare spending per capita vs. average life expectancy among OECD countries, 2007)

The Bennett & Hauser study took a behavioral health population and used a computer simulation to compare costs and clinical outcomes obtained through actual care in real clinics with real patients. By optimizing the decisions made at each step in the process of care, patient outcomes could be improved by 42% while at the same time reducing cost by 58%.

(See Figure 2: Artificial intelligence (AI) vs. treatment-as-usual)

In economic terms, imagine that a weight loss clinic is able to achieve 10 pounds of weight loss for $500. The cost for each pound lost would be $50. In the Bennett & Hauser study, the cost per unit of clinical change in usual treatment was $497. The cost achieved by the artificial intelligence algorithm was $197, less than half the cost per unit of improvement.

Unfortunately, such improvements in value (i.e., outcomes divided by cost) are not incentivized in traditional fee for service payment models, which promote quantity, not quality. Changing this is one of the core ideas of health reform and accountable care – to shift from incentivizing volume to paying for performance. That said, the research on pay for performance is mixed at best. While it makes intuitive sense to pay for outcomes instead of volume, we have repeatedly seen that rational incentives often don’t bring the intended result, or worse yet, have unintended consequences.

The difficulty of reducing costs while improving outcomes actually represents an opportunity for behavioral healthcare organizations that are savvy enough to embrace clinical decision support technologies such as those described in the article above. In the coming era of accountable healthcare, those who can unequivocally demonstrate true value in their product will be rare and in high demand. Given what we know about the prevalence of behavioral health conditions and their impact on overall healthcare spending, behavioral healthcare providers have the specific behavior change expertise required to demonstrate significant value in a system that rewards performance.

The challenge we face is reengineering behavioral healthcare’s delivery model to figure out how to appropriately integrate the benefits of artificial intelligence with the power of a healing therapeutic relationship. The changes coming through health reform are likely to accelerate disruption of traditional models of care, and move healthcare toward adoption of AI and predictive modeling technologies just as we have seen in other industries, such as retail, marketing, and finance. Are we going to be ready?