For more than a decade, artificial intelligence (AI), defined as machine intelligence that mimics human cognitive function, has been promised as the solution to business problems across many industries - healthcare included. It was touted as the answer to health inequities, a way to assist physicians with difficult clinical decisions, a tool to predict readmission risks and a means for lowering the overall cost of care.
Overall, AI has brought enormous positive change to the healthcare industry -- for example, it has been leveraged to assist in visual diagnosis of disease (e.g. diabetic retinopathy by Google) and to predict the risk of sepsis. However, it has not always lived up to its lofty expectations for at least two main reasons:
Difficult Integrations - For AI to be effective, it needs to know both what to do (intelligence) and how to do it (integration). While it is easy to be excited about superhuman intelligence, it is also easy to forget about the arduous task of integrating it into existing workflows. Clinical AI falls short when it is not properly integrated with existing systems of record (e.g. electronic medical records). For example, several clinical decision support systems have been developed over the years to assist physicians with real-time decision-making in acute care situations. However, adoption and usage of these systems requires them to be available to clinicians at the point-of-care in an easy-to-use manner. Too often, powerful solutions like this are deployed in third-party applications or external portals with poor user experiences, which is why they’re rarely used.
Another shortcoming is the over-reliance on point solutions, which are narrow solutions to discrete problems. Point solutions are effective in a vacuum, but too many of them results in a fragmented user experience and complicated IT integrations.
Increasing Workload - In what is often called the productivity paradox, AI and machine learning technologies that were supposed to enable healthcare providers to be more productive have actually made them far busier and have taken away time from patient care.
For example, EMR-based predictive alerts for sepsis risk and drug-drug interactions have often been cited as a cause of “alarm fatigue,” whereby frequent exposure to alarms causes physicians and care staff to become desensitized to them altogether. In other cases, AI systems that require providers to interpret complex data and recommendations (often false positives) quickly result in cognitive overload.
RPA: The secret to unlocking the true potential of AI
If AI is knowing what to do, Robotic Process Automation (RPA) is having the “hands” or ability to actually do it. The two are not at odds with each other, in fact quite the opposite is true -- they are most powerful when combined, as one is the intelligence (AI) and the other is the ability to act on it (RPA).
RPA is defined as the training of software bots to perform actions in a software application using the graphical user interface (GUI). In other words, the bots use a virtual keyboard and mouse to interact with the computer screen just like a human would, and as a result, they are able to automate manual, repetitive tasks with stunning accuracy and precision.
RPA is well-suited to enable EHR integration for two main reasons. First, RPA doesn’t require connecting to existing interfaces (APIs), which can be one of the most time-consuming and expensive aspects of deploying health informatics software. Second, RPA isn’t limited by which interfaces already exist, allowing for a broader set of capabilities to be automated. For example, at Notable we often use RPA to write data into sticky notes within a patient’s chart to serve as a just-in-time nudge to the provider -- a capability that isn’t accessible via API access in most EHRs.
It is worth noting that RPA can exist without AI, in which case the bot’s logic has to be hard-coded so it knows exactly what to do and when to do it. For example, a bot could be trained to add “Diabetes Mellitus” to the problem list if a patient’s hemoglobin a1c test is above a pre-set diagnostic threshold. Or, an AI-powered bot can learn to predict a hospitalized patient’s estimated discharge date based on data from thousands of similar patients who have been previously admitted; and use RPA to update the estimated discharge date field in the EMR daily.
At first, the bot may be occasionally incorrect in its predictions, but over time its supervised learning model will become highly accurate in discharge date estimations and improve their performance over time just like humans do. By augmenting AI with RPA, health systems can save their providers and staff valuable time.
Intelligent automation -- The Notable difference
Notable is the first platform to combine artificial intelligence and robotic process automation to solve healthcare problems across the care continuum, from the front desk to the back office. This is called intelligent automation.
Using AI, Notable’s bots can determine when and how to perform automated workflows within existing software applications including EHRs.
Automating registration work queues. Many health systems have large call centers and work queues focused on obtaining accurate registration information for new patients, such as the health plan, guarantor and insurance card in an effort to prevent downstream denials. Notable’s bots or ‘digital assistants’, use AI (patient intakes to verify missing data, computer vision to ingest the insurance card, and predictive classifiers to predict a health plan) to obtain this information, and then use enterprise-grade RPA to export this information into the EMR. In doing so, Notable’s bots serve as the perfect adjunct to human staff, allowing staff to focus on higher complexity tasks and less repetitive work.
Expediting clinical workflows. Many of our physician partners spend hours each day reviewing the EMR to find key pieces of information. This manual “chart review” is time consuming and error prone, but doing it leads to optimal care. Using NLP and computer vision, Notable’s digital assistants assist providers by reading hundreds of patient notes (both structured and unstructured) and surfacing only the most relevant insights before every visit. This curated care coordination helps improve the quality of care while off-loading the burden from already overworked medical providers.
The next era of AI
The graveyard of AI applications for healthcare is filled with high-potential applications that never captivated users due to ineffective integrations and frustrating user experiences.
AI holds tremendous potential for pushing healthcare forward, but it is time we recognize the importance of integration that is robust, rapid and repeatable. Intelligent automation, the combination of AI for intelligence and RPA for integration, provides tremendous promise for the next era of healthcare technology.
Click here to see how Notable can automate any administrative or clinical workflow to reduce costs and streamline operational complexities at your organization.