The clinical implementation of Artificial Intelligence (AI) is the most viable means of uniting the interests of the healthcare industry’s capital constituents: the patient, the payer, and the provider. AI can improve healthcare outcomes while reducing costs when used to address patient compliance, chronic care management, genome sequencing, and physician diagnostics by classifying treatment options. Its widespread clinical deployment is poised to transform the healthcare industry into one that maintains wellness instead of merely combating illness.
Maximizing AI’s clinical value depends on the proper execution of four interrelated steps, each of which represents emerging developments within the industry:
- Remote Patient Monitoring: Tracking patient behavior and providing timely feedback based on individualized data requires mainstream adoption of clinically accurate, medical-grade wearable devices. These tools will serve as the endpoint for remote patient monitoring (RPM) systems connected to care facilities.
- Continuous Connectivity: Connecting medical-grade wearable devices to clinics requires uninterrupted transmission of constantly generated patient data, the likes of which are projected to be typical of the Internet of Things.
- AI Training: The clinical value of AI will be derived through the training of its advanced machine learning algorithms on large datasets suitable to chronic medical conditions. These algorithms require significant training periods to analyze data for specific afflictions, similar to the prolonged education period of attending medical school.
- Comprehensive Business Model: AI’s capacity to unify the interests of the three healthcare partisans mandates a comprehensive business model which successfully aligns the economic realities of each. Central to this model is the economy of scale for providers from whom healthcare service originates.
The proper implementation of each of these steps will ensure a future in which AI substantially contributes to decreased costs of chronic care and patient non-adherence, while achieving patient objectives in accordance with contemporary physician economics. Their implementation will also provide physicians with a vital support tool for conducting remote diagnostics, treatment classifications and accelerated care management.
Step One: Remote Patient Monitoring
The initial step for implementing AI in clinical settings is incorporating it into RPM systems. Such systems are only as reliable as the endpoint devices on which they’re predicated. Medical-grade wearable devices provide clinically relevant data when sanctioned by industry approved regulatory entities such as the Food and Drug Administration and HIPAA. These entities validate the consistency and accuracy of the device’s data prior to adoption by care facilities or patients using a rigorous set of criteria. The cross-industry sanctioning of these tools is the primary distinction between medical-grade devices and the plethora of consumer-based, software-driven mobile applications that generate data without regulatory approbation.
Infusing these devices with AI augments the advantages of each. Together, they support the holistic lifestyle management required for chronic conditions, such as diabetes, within a clinical context. When suitably trained in the various medical disciplines necessary for this condition (including aspects of nutrition, kinesiology, and endocrinology), AI algorithms can issue reminders for patients to check their blood sugar levels, increase their exercise, or visit their endocrinologist based on the device’s readings—in this case, a connected glucometer. The glucometer itself becomes more meaningful to the healthcare process because of the feedback generated by AI, which is only possible because of its integration with the device.
Regardless of the ailment, the key is to empower medical-grade wearable devices with AI for coaching and feedback. AI can issue these same benefits for other chronic conditions such as cardiovascular disease. For managing cardiovascular disease, an EKG equipped with AI yields the same benefits as a connected glucometer for diabetes.
Step Two: Connectivity
Maintaining constant connectivity is an implicit prerequisite to positioning medical grade wearable devices as the touch point for RPM systems. The IoT’s distributed paradigm has emerged as the preferred method of facilitating ubiquitous connectivity from a multiplicity of endpoints. This model requires connected devices to constantly transmit data to the cloud for integration, aggregation, and analytics before sending results back as needed.
The two main issues in this process are security and data offloading. Security is eminent for the devices at the edge of the network—which are outside centralized security measures such as protective firewalls—and for the centralized locations in which most computations occur. Virtual Private Networks (VPNs) alleviate some of these concerns by adding an additional protective layer beyond that of the centralized data center. End users of remote patient monitoring systems get the additional benefit of protecting their data within a private network of an external network. This approach reinforces privacy and regulatory benefits too, which are integral aspects of managing healthcare data.
A fundamental boon of employing the IoT’s distributed cloud model is the continual data offloading required for AI’s computations in a centralized location. In this respect, AI’s chief value proposition is its ability to expediently aggregate, cross-reference, and learn from data to determine its correlation to patient objectives. The cloud functions as the medium in which this process occurs. The more data deep learning algorithms and neural networks analyze, the more refined their results become for specific healthcare applications. If the algorithm is exclusively analyzing a patient’s blood sugar level, it’ll provide basic information. But when analyzing this data in tandem with those pertaining to exercise regimens, specific eating habits, and weather conditions, it can deliver a host of predictive and preventative feedback to improve diabetes care management. If a patient hasn’t eaten in several hours and has walked more than customary, AI could predict a drop in their blood sugar level and prescribe reminders for eating or taking medication to prevent this occurrence.
Step Three: Training AI
AI needs to learn the myriad healthcare disciplines to provide specialized coaching and feedback for chronic conditions like cardiovascular disease and diabetes. The distributed IoT paradigm is essential in this regard because it provides a centralized location for endpoint devices to offload the data needed to compile factors most apposite to care, such as the potential effect of high humidity on blood sugar levels.
Suitably training AI in the care management of chronic issues implicitly mandates offloading data for two didactic reasons. First, transferring remote patient data to the cloud is the most accessible way of combining that data with other datasets related to the affliction. The vast amounts of data pertaining to disease specific parameters, health risks, patient history, sugar levels, food intake, and other facets of nutrition cannot be contained solely in endpoint devices. Even if it were possible, it would still be necessary to aggregate such data with those for other diabetic concerns such as the influence of alcohol, age, and environmental factors. The cloud’s storage capabilities and elastic provisioning of computational resources on demand is ideal for the diverse learning required of AI’s algorithms. Second, it’s necessary to initially train AI for chronic diseases in the cloud because of the scale involved. When applying that learning to individual patient behavior, AI will need to offload patient data to see how it compares to the information already learned.
When learning the specifics of a new RPM user with a chronic condition, such as cardiovascular disease or diabetes, AI will take a series of measurements to establish a baseline (i.e. heartbeat consistency or blood sugar levels). It will then constantly compare a patient’s results—at specified intervals—to deliver feedback impacting care. Determining relevant care methods such as whether to increase or decrease the intake of carbohydrates, will be gleaned from offloading data in the cloud pertaining to that area of specialization. The results of this learning are then used to issue predictive action to attain patient objectives, such as reminding users they’ve been in a bar for two hours and maximized their alcohol intake for their blood sugar level. Thus, AI is able to create a baseline measurement, track a patient’s progress, determine if they’re adherent with practitioner’s guidelines, recommend action to increase adherence, offer a means of patient engagement, and provide predictive action to avert negative health outcomes.
Step Four: A Comprehensive Business Model
The business model for the clinical implementation of AI must align the economics of the patient, the payer, and the provider. The physician economics of the provider form the basis of that model since that’s where health services begin. Those economics are based on three factors: reimbursement, efficiency, and clinical relevance. Accounting for reimbursement with the clinical implementation of AI necessitates doing so with existing billing policies. There are certain tools embedded within the job functions of practitioners (such as stethoscopes) which directly affect their billing. Practitioners regularly use these tools because they ensure the completion of basic necessities , without which they can’t bill the payer. Additionally, there are certain tools healthcare facilities purchase because they allow them to perform more complex tasks that can still be billed to insurance companies. Hospitals invest in x-ray machines because of the payer’s reimbursement for their use. The deployment of clinically accurate wearable devices equipped with AI must become billable as basic tools that are required for practitioners to use in patient check-ups and diagnostics, just like stethoscopes and x-ray machines.
The efficiency component of physician economics is measured in terms of decreasing time spent on necessary tasks. Materials conserving time and money are desirable because they increase organizational return on investment (ROI). Typical use cases involve the automation of manual processes, such as dispensing medication or pills. Organizations will invest in automatic pill dispensing mechanisms for routine medication because it positively impacts their ROI and clinical efficiency. AI’s aptitude for quickly analyzing disparate and large quantities of data produces this advantage. Ultrasound imaging is similar to AI’s utility in this regard. Those contemplating investing in ultrasound imaging must calculate its impact on practitioner efficiency—how much time will personnel spend on such analysis without it versus its impact on their ROI. In situations in which they’re paid to use ultrasound equipment, ROI is more readily calculated and the investment is justifiable, especially if it increases productivity.
The implementation of ROI in clinical use cases must also complement the final facet of physician economics: clinical relevance. By increasing engagement, offering feedback, issuing predictions to heighten patient compliance with chronic care management, and providing clinically accurate data in the process, AI infused RPM fulfills this requirement of physician economics. Its adoption will enable practitioners to focus on rare or difficult healthcare use cases deserving of their time, because AI will be in the background doing routine analysis for efficient practitioner interpretation.
The widespread adoption of AI will improve the overall efficiency and effectiveness of the healthcare system. Its implementation will provide demonstrable value for each of the three traditional participants in that system, drastically improving it as a whole. The critical first step in doing so is integrating clinically accurate RPM devices with AI which will improve practitioners’ efficiency and effectiveness, particularly if they’re paid to use them.
These devices must have continuous connectivity and constantly offload data to the cloud to learn from the sizable datasets relevant to chronic disease management. This step is necessary to properly train AI and expand its clinical value. Still, the provider’s satisfaction is just one facet of uniting the interests of the three central healthcare participants. The payer should also promote the uniform adoption of these technologies because of their ability to reduce long-term costs. AI produces this effect by diagnosing patients quicker so the care process can begin before treatment becomes dire—and costly. Eventually, practitioners will send patients toolkits every three months for remote checkups that will provide an initial form of expedient diagnostics. The cumulative impact will be a focus on preventative care, as opposed to reactionary care.
Patients are perhaps the ultimate beneficiary of AI deployments in clinical practice because they allow patients to improve their adherence, health, and quality of life. Today, diabetes patients must remember to manually record glucometer readings, visit their endocrinologist every three months, and take their medication without feedback or incentive for adherence between visits. Soon, an AI connected glucometer will record their blood sugar level readings in its software, disseminate feedback and reminders based on correlations between current measurements and their baselines, and function as a mobile, virtual healthcare coach. The projected effects are incomparable in their capacity to bring positive healthcare results for patients with chronic conditions.