Last year I attended an online seminar hosted by a large New England hospital. The speaker summarized some of the operational research and analytics that he and his team had developed using the hospital’s data. By pulling a variety of data from the hospital’s EHR – everything from demographic data to vital signs – and running fairly sophisticated models, they hoped to predict clinical outcomes in a specific population of pediatric patients. Although I left the seminar with an admiration for the speaker’s attention to detail, I also had a question on my mind: how scalable is the work?
The question of scale is not necessarily new or original. Plenty of academic initiatives pioneer innovative approaches to treatment, care management and coordination, or operations management. Major journals then publish the eye-catching results. But what happens next? Often, nothing. There’s no money in academia for replication, so others don’t try to reproduce the results. And even those who attempt to implement a similar program run into roadblocks; for example, an approach that worked with significant funding at a major academic medical center doesn’t always translate to rural safety net hospitals with razor-thin margins.
To a certain degree, the scalability of analytics depends on the scope of several dimensions. First and foremost is the population of interest: how many people does this affect? In the case of the seminar speaker and his team, the analytics focused on a narrow pediatric population. Hospitals without a significant number of similar patients – and without the consequent operational need – might be hesitant to invest in the technology.
Second, the scope of data involved must match the availability of IT resources and the availability of data. With the plethora of software hospitals use, IT staff at hospitals often specialize in certain systems. An analytics solution that requires a laundry list of data also requires collaboration with multiple IT stakeholders, a challenge given typical IT workloads. And even if IT resources are available, the data might not be.
Third, the predictive models and analytic methods must balance utility and transparency. Although deep learning models and neural nets might have high predictive value, their reputation as “black boxes” hinders physician trust relative to more run-of-the-mill regression models with easily interpretable results. Similarly, a PhD in applied mathematics can’t be a requirement for someone to be able to understand and successfully use an analytic solution at a hospital.
None of this is to say that we should prefer general solutions to specific solutions. Rather, scalability depends on knowing how to balance the two. At Hospital IQ, we’ve developed an AI-enabled operations management platform to solve problems facing hospitals across the country. But we also know that no two hospitals are the same; that’s why we embed opportunities for customization and configuration across our suite of solutions. With the increasing financial pressures facing hospitals, the question of scalability should continue to be at the forefront of every executive’s mind when evaluating strategic initiatives and innovative technology.