The next frontier of AI in healthcare

A recent OptumIQ annual survey of major healthcare organizations on AI in Healthcare shows an average of $32.4M investment per organization over the next 5 years. 91% of the 500 healthcare leaders surveyed are confident they will see an ROI on AI – in the next 4 years for hospital executives, and in 3 years or less for 38% of employers and 20% of health plan executives. 75% of those surveyed are either actively implementing or planning to execute an AI strategy.

In planning an AI strategy, It would help to understand how AI may be added into the current IT mix. AI may be included in an existing application or integrated with applications in a workflow. Or in the lesser-known, process-centric approach, AI may encapsulate the workflow, which arguably would take us to the next frontier.

The application includes AI

EHR vendors, consistently blamed for interfering with the patient-provider relationship for their applications’ subpar UI/UX, strive to innovate by adding AI in their applications. Using voice assistants for documentation, and Natural Language Processing (NLP) to summarize free-text notes are two of the examples. “We want to help tailor the system to pick out the most interesting information available, as well as the tasks they’re most likely to want to perform and place them at the user’s fingertips. That will allow the clinician to spend more time with the patient,” said Epic’s Seth Hain, R&D Division Manager for Analytics and Machine Learning. The jury is still out whether AI will solve the EMR UI/UX problem – clinicians who have been burned by big promises before may not be rushing to believe that AI will solve all their EHR problems.

The workflow integrates AI

The Westchester Center Health Network (WMCHealth) case study is a good example of adding AI to an existing workflow. WMCHealth uses both its EHR’s risk model, and a third-party vendor, Health Catalyst’s predictive model in prioritizing discharged patients for readmission-reduction efforts. They add Health Catalyst’s risk scores, and EHR data, on a dashboard with the discharge lists that organize the case managers’ work to help them prioritize the patients who need to be engaged. The new risk scores from the integrated AI help to identify more true positive cases (8%) and reduce false positives (30%) vs. either EHR risk model or LACE.

Another example of applying AI in a healthcare workflow is Beth Israel Deaconess Medical Center’s use of TensorFlow on Amazon SageMaker to scan pre-surgical document packages to identify and insert consent forms into the corresponding electronic medical records. The tool delivers a notification to the EHR, if the consent form is missing, to trigger the follow-up workflow action.

AI encapsulates the workflow

Business Process Management (BPM) practitioners across industries have long been codifying a workflow process into a series of tasks, completion of which would produce the work result. A codified workflow could interface with multiple systems and workers, and its performance be monitored and analyzed.

Up until recently, BPM tools were cumbersome, and BPM projects expensive. So, it isn’t surprising that BPM projects have primarily been implemented in enterprises, and for cost-reduction of complex backend processes, e.g. order fulfillment, and supply chain management. But recent interests in using BPM for customer experience and digital transformation have brought BPM out of the backroom. Coincidentally, improving customer experience is also one of the main use cases of AI. This intersection spurs BPM vendors to race toward AI-enablement of their platforms.

The shift of focus to customer (or patient) experience increases BPM’s relevance in Healthcare. The codified workflow, in essence, is a digital version of what a care team currently performs manually. It enables the healthcare organization to monitor the care workflow, react to adverse conditions promptly, and continually improve the process, as illustrated in a BPM project at Ottawa Hospital.

AI, operating on the codified workflow, essentially includes the workflow in its predictive model, and can automate not just the workflow and its tasks, but also modifications of the workflow to continually improve the process. What is unique to this process-centric approach is that the workflow may get smarter over time as the AI is considering how we do things and trying to do things for us.

A pathway to the future

Healthcare, however, is missing a key ingredient that has been driving the adoption of the process-centric approach in other industries: wide availability of APIs to systems a workflow typically touches. While recent low-code BPM tools have greatly eased workflow codification with drag-and-drop integration with Salesforce.com, Dropbox, Google apps, and the like, such convenience is limited in Healthcare. However, over the years EHR vendors like AllScripts and Athenahealth as well as a list of others have exposed APIs to access their data. Firms like Redox and Sansoro Health have also pushed their proprietary APIs, leveraging standards like HL7 v2 and FHIR, to shield healthcare organizations and developers from the complexity of integration across multiple systems.

Workflow codifications lead to a new crop of “business process applications”, which can interface with healthcare workers via a form specific to the workflow context (e.g. inpatient discharge), or some voice assistant. As these business process applications are workflow aware, they are poised to simplify workers’ interactions with multiple systems, and among themselves by automating workflows, tasks, and process optimizations. For example, an AI-driven, codified workflow could be doing the magic of coordinating the work among the care team, an external testing center, and the patient such that with a simple command “refer the patient to Eastlake for testing”, it sees the care process through such that the patient is back to the office, post-test, for the next step within a reasonable timeframe. Not to mention there would be no clicking around in the EHR to enter data into appropriate screens.

3-pronged strategy for competitive advantage

Waiting for an existing application to add AI and integrating an AI application or service into a workflow cause minimal disruption to the current IT environment. But they also greatly reduce AI’s ability to improve how we do things, and to do things for us.

The aforementioned OptumIQ survey revealed top areas of AI investments being automating business processes (administrative operations or customer service), at 43%, and fraud, waste and abuse detections 36%. In addition, the top two expected benefits are increased efficiency, and more accurate diagnosis. A third of the respondents also expected improved patient experience and decreased per-capita cost of care. Automating business processes, and improved patient experience, in particular, make the case for leveraging the process-centric approach, which has customer/patient experience in its focus.

In summary, a balanced, three-pronged strategy will enable a healthcare organization to minimize risk of disruption where necessary, but not limit its ability to innovate to its current workflows or existing applications. Gearing up the organization with capabilities, and practice to allow “AI to encapsulate the workflows” is ultimately an opportunity to gain a sustainable competitive advantage in an era that sees continued pressure from bountiful consumer choices, profit margin pinches, and assumptions of risk in the patients’ long-term wellbeing.