The Next Generation of Intelligent Healthcare Information Technology

This post is the third in a series of guest blog posts by some of our sponsors sharing their perspective on Convergence.

Author: John Glaser, Senior Vice President, Population Health, Cerner

Over the course of the last twenty-five years, there have been three major leaps in information technology.

The Internet and World Wide Web were the first of these leaps; debuting for commercial use in the 1990s. Connected mobile devices (the second leap) took a big step forward with the introduction of the iPhone in the late 2000s.

We are now in the early stages of the next major leap; the next generation of intelligent information technology (IT).

While still in its early stages, the next generation of intelligent IT is already prevalent in our everyday lives. Few of us use a paper map anymore to guide our travels. The temperature and security of our homes can be controlled from anywhere. Self-parking cars are common. Alexa can order pizza and tell you the weather in Sydney. These leaps have had, and will continue to have, a very significant impact on our personal lives and introduce new business models and product and service opportunities.

The leap in intelligence is being driven by opportunities and demands across a wide range of industries including transportation, retail, financial services and manufacturing. All leaps involve an “ecosystem” of technical capabilities and companies coming together. For example, the mobile device leap resulted from new devices, high speed wireless networks, improvements in batteries, an abundance of third party applications and sensor advances such as location awareness.

While an ecosystem is necessary, at the center of each leap was a specific class of technology innovation. Machine and deep learning. Artificial intelligence. These are some of the high-level labels applied to this class of advances in intelligence. Behind these high-level labels are specific analytic methods such as Singular Value Decomposition, Restricted Boltzmann Machines and Random Forests. 

Four Focus Areas

Electronic health record products have embraced intelligence for many years. Clinical decision support is an integral component of CPOE. Surveillance technologies are used to perform real time monitoring of potential sepsis. Predictive algorithms are used to identify patients at high risk of readmission. Care guidelines and protocols remind caregivers to order certain tests or schedule appointments.

Due to mounting pressure to improve the quality, safety and efficiency of care delivery, health care is joining other industries in an effort to capitalize and accelerate the next generation intelligence leap. This leap will focus on four areas.

Extraction of data and structures. Intelligence will be used to advance our ability to process natural language, images and video. Increasingly these capabilities will be used to extract quality measures from clinical data, suggest encounter activity and resource codes based on visual and audio analyses of the encounter and auto-reconcile inconsistencies, gaps and errors in clinical data.

This processing will not only extract data but also identify patterns in the data. These patterns will include determination of different patterns of medication non-adherence (with suggestions of different interventions), evidence that a particular medication is causing harm or has a potential beneficial off-label use, and identification of social determinants that are having a disproportionate impact on the health of people.

Cognitive interaction. Intelligence will help shape provider and consumer interactions with the technology to reflect their interests, needs, cognitive style and tasks. For example, intelligence will recognize that a provider is documenting a visit and will dynamically structure the documentation to reflect the patient’s condition, treatment best practices, and patient and clinician preferences.

Advance visualization approaches can assist decision makers in understanding what the data is telling them.

As our understanding of behavioral economics and means to motivate people becomes more sophisticated, cognitive interaction will also expand to encompass emotional interaction and the interaction will expand from a transaction to an experience.

Operational process modeling. Intelligence will help create, orchestrate and monitor clinical and operational processes. For example, population health will lead to the development of “plans for health” for patients and consumers. These plans incorporate a clinical plan, e.g., medications for managing a chronic disease, a social plan, e.g., patient needs food assistance, and patient personal preferences.

Any plan will evolve over time. A well-managed disease may drift into poor management. A long-term spouse may pass away leading to concerns about a patient’s depression. A spike in pollen count will impact the activities of a child with asthma.

Intelligence will help form and integrate plans, suggest potential alterations to a plan based on situations such as above, identify individuals that need attention because a of a material deviation from a plan and monitor overall plan performance across a population.

Clinical models. Intelligence will be applied to create a wide range of clinical models and decision aids. Clinical models have been a fixture of electronic health records for many years; for example, clinical decision support has warned of medication interactions and health maintenance reminders have prompted patients to schedule a quarterly appointment.

The extension of intelligence to support clinical models will occur across several dimensions as we shift from encounter-orientation to longitudinal care orientation, from an almost exclusive focus on clinical data to a focus that includes social determinants of health, and environmental, genomic and behavioral data     and from a mode of reactive sick care to proactive management of health.                                                                                                                `

For example, a model that assessed risk of treatment non-compliance might note that the patient is homeless, doesn’t have reliable transportation to the clinic or is on food stamps, hence there is a higher risk of them not adhering to the treatment plan. In addition to leveraging an expanded set of data for the model, the model will also suggest approaches to addressing the risk.

Considerations

Ensuring that the next leap – the leap of intelligence – is optimally effective requires that the work in the four areas described above takes into account several considerations.

First, existing health care information technology will need to be adapted to incorporate the technology and methods of the next generation of intelligence. These adaptations can include:

• The utilization of FHIR-based APIs that support leveraging third party applications that execute novel algorithms,
• Integrating workflow technologies into core transaction systems,
• Standardizing the data models and vocabularies used by existing systems, and
• Surrounding existing applications with adaptive user interfaces.

Second, there will be a tension, at times, between the rigor of a model and our ability to achieve broad use. For example, a model might have great predictive power but limited prescriptive power (the rationale for a risk score is opaque) or have limited use because it requires too much manual entry of data.

Third, we will need to evaluate whether the use of the intelligence leads to higher quality and reduced costs of care and improved clinician work satisfaction. Models are rarely optimal when they are born and evaluation is required to guide model evolution. Moreover, intelligence is used in a broader socio-technical context of real people in real care settings in a real reimbursement environment. Determining the best way to configure the various pieces of this context to fully take advantage of the intelligence requires thoughtful analyses.

Four, all major technology leaps take years, even decades, to have a mature and long lasting impact. Twenty years after the arrival of World Wide Web use by business, 12% of retail revenue is generated through ecommerce. Ecommerce revenue is growing faster than in-store revenue and ecommerce growth is one of the reasons that several major department stores are facing insolvency. However, it took 20 years to get to 12%.

This third leap will be impactful but its impact will play out over the course of years.

Conclusion

The third major leap in information technology – the next generation of intelligence – is here. While there is a lot of work to be done (and many years) to fully incorporate these advances into health care, there is no doubt that the care that is delivered will be safer, more efficient and of higher quality.

The word “compute” in computers will have achieved a new meaning and a more consequential meaning.