ChatGPT’s potential impact on preventative care and emergency visits

ChatGPT has been making an impact on enduring healthcare challenges. Many providers and patients are reporting the artificial intelligence helping with preventative care and preventing non-emergent emergency department visits.

Imran Qureshi is chief technology officer at b.well Connected Health, a FHIR-based interoperability platform vendor. He has expertise in AI and a lot of advice to share on the subject of health systems harnessing ChatGPT and other large language model-based technologies.

We interviewed Qureshi for a deep dive into how hospitals and health systems can harness ChatGPT and other LLM technologies to improve patient care, how provider organizations can use LLMs to ease clinician workload, and how providers can deploy ChatGPT and LLMs to empower patients.

Q. How can hospitals and health systems harness ChatGPT and other LLM-based technologies to improve patient care?

A. In the current healthcare landscape, hospitals and health systems are navigating challenging times, with financial pressures exacerbated by the COVID-19 pandemic’s lasting effects. Notable health systems, including Ascension Health, Trinity Health and Providence Health, reported a combined loss of $8 billion in 2022.

Concurrently, the healthcare workforce is facing significant attrition; approximately 20% of doctors are planning to leave the profession and more than 170,000 healthcare workers departed their roles in 2021, according to JAMA.

This exodus is compounded by the fact that 85% of primary care physicians express a desire to devote more time to patient interactions, per a 2017 survey by Ipsos, yet an analysis by the Journal of General Internal Medicine showed that administrative tasks consume 55% of their workload.

Patients, too, are expressing dissatisfaction with their healthcare experiences, with 67% reporting negative healthcare encounters within a three-month period in a 2021 survey by Accenture, and 34% are either switching providers or hesitating to seek future care.

The crux of patient discontent lies in the desire for more meaningful interactions with their healthcare providers as expressed in a survey by Deloitte in 2016, emphasizing the importance of quality time and attention.

A promising solution to these multifaceted issues lies in the strategic deployment of artificial intelligence, particularly large language models like ChatGPT. By leveraging LLM-based technologies, healthcare can be made more accessible and efficient for both providers and patients.

Non-physician healthcare workers, including nurses, aides and therapists, can use these technologies to navigate patient charts, automate responses to common inquiries and facilitate patient care coordination, effectively extending the capabilities of physicians.

Moreover, LLMs empower patients and caregivers to engage in self-service for basic healthcare queries, reducing the demand on physicians’ time for routine questions. This technology also holds the potential to guide patients through the healthcare system, ensuring timely and appropriate care interventions.

With approximately 250,000 primary care physicians and an additional 250,000 non-physician primary care providers in the United States, according to the National Center for Healthcare Workforce Analysis, the redistribution of routine tasks to non-physician staff and directly to patients can significantly enhance the efficiency of patient care.

The vast workforce of nine million non-physicians and the collective engagement of 350 million patients and caregivers presents a tremendous opportunity to transform healthcare delivery.

The integration of LLM-based technologies into healthcare workflows offers a beacon of hope for addressing the current challenges faced by the industry. By facilitating a more efficient division of labor and enabling patients to take an active role in their healthcare journey, these technologies promise to enhance patient satisfaction and allow healthcare professionals to focus more on direct patient care.

This shift toward a more patient-centered approach, supported by advanced AI tools, heralds a new era of improved healthcare outcomes and experiences.

Q. How can hospitals and health systems use LLMs to ease clinician workload?

A. In today’s healthcare environment, primary care physicians are striving for more meaningful patient interactions but find themselves hindered by the extensive administrative tasks required of them.

Currently, 85% of these professionals, in a 2017 survey by Ipsos, express a desire to allocate more time to patient care, yet the reality is that less than half of their appointment time is spent on actual patient interaction per the Journal of General Internal Medicine, with nearly two hours weekly dedicated to after-hours documentation, according to JAMA.

The root of this issue often lies in the cumbersome nature of EHRs, which are primarily designed for data entry rather than efficient information retrieval.

Clinicians frequently spend a significant portion of patient encounters inputting data into these systems and then additional time outside of appointments for data entry. This process is further complicated by the difficulty in extracting information from EHRs, as clinicians must navigate through previous clinical notes due to the inefficiency of the system’s design.

Implementing LLM interfaces with EHRs could revolutionize this process, enabling clinicians to interact with the system using natural language queries. Such an interface would allow for straightforward questions like “Does this patient have a family history of diabetes?” or “What is the trend of this patient’s blood pressure over the past eighteen months?” to be answered efficiently, drawing upon the comprehensive data within the EHR.

Furthermore, LLM technologies could automate the documentation process, allowing clinicians to input notes verbally and having the system update the EHR accordingly, thereby minimizing manual data entry.

Beyond improving EHR interactions, LLMs could also alleviate clinician workload by handling routine inquiries about health and healthcare logistics. These technologies can engage directly with healthcare staff, patients and caregivers, identifying the nature of inquiries and providing accurate responses or directing them to appropriate humans.

This level of triage can significantly reduce the demand on clinicians’ time, ensuring they are consulted for more complex issues that require their expertise.

By simplifying data entry, enhancing access to patient information, and delegating routine questions to LLM-based technologies, we can substantially lighten the administrative burden on clinicians.

This shift not only makes healthcare delivery more efficient but also allows clinicians to concentrate on what matters most – providing attentive, personalized care to their patients.

Q. How can hospitals and health systems deploy ChatGPT and LLMs to empower patients?

A. Hospitals and health systems have already invested heavily in the development of EHRs, data warehouses and analytical platforms. The advent of LLM technologies, such as ChatGPT, does not necessitate discarding these valuable resources.

Instead, these technologies can be integrated into the existing infrastructure, enhancing its capabilities and providing significant benefits to patients.

By translating the extensive data within these systems into a format that LLM technologies can understand – essentially creating a knowledge store in plain English – healthcare organizations can overcome the challenges associated with data variability.

This approach allows LLMs to interpret and process information from diverse sources and formats, such as different conventions for recording dates of birth, without the need for complex data mapping or standardization.

Using LLM technologies offered by leading platforms like OpenAI, Microsoft Azure, AWS or Google Cloud, healthcare providers can build upon this knowledge base to deliver precise and accessible answers to a wide array of patient inquiries, drawing directly from the data within the knowledge store.

Patients can ask questions and receive answers in plain English instead of having to understand clinical terminology.

This LLM architecture can reuse the existing infrastructure in healthcare and add three new pieces: a knowledge store, a language interface and a risk management layer.

An LLM architecture starts with all the existing data sources, but instead of spending time and money on converting these to rigid data warehouse schemas, we can convert them to plain text and store them in a knowledge store.

In healthcare we’ve spent more than two decades trying to map all our data to our data warehouses, but very few organizations can claim all their data is available in their data warehouse. How much longer do we continue down the path of rigid data warehouses if 20 years has not been enough?

An LLM architecture allows people to query for answers in plain English (or Spanish or any other language). Patients, their caregivers, doctors, nurses, administrators and other workers without technical abilities can now get answers without waiting for data analysts or data engineers to translate their question into database code like SQL.

Today, most organizations have a backlog of months or years to provide these answers. A language interface, powered by LLMs, can convert questions asked in plain English into SQL code and extract answers without any involvement by data analysts.

To ensure the accuracy and appropriateness of information provided, a risk management layer can be introduced. This layer would involve the use of carefully selected data to screen, refine and answer questions, ensuring sensitive or inappropriate queries are managed correctly.

The process includes generating variations of the original question, evaluating the consistency of answers, and incorporating feedback from healthcare professionals to optimize the system’s performance.

A risk management layer consists of eight steps:

Curate – Choose trusted data sources as input to the LLM.
Instruct – Instruct the LLM to use only the curated content.
Filter – Filter out categories of questions that LLM should not answer.
Ask – Query the LLM using the prompt created from the above steps.
Evaluate – Rephrase the question multiple ways and check for consensus in answers.
Fact Check – Check answers against underlying data.
Notify – Notify users that AI was used to answer their question.
Learn – Collect feedback from users for reinforcement learning so the LLM gets better over time.

This innovative use of LLM technology in an LLM architecture can empower patients and their caregivers with the ability to obtain immediate responses to their questions, ranging from logistical concerns about medical appointments to specific health-related inquiries.

Additionally, it enables them to perform straightforward healthcare tasks, such as ordering medication refills or seeking more affordable medication options, without needing to navigate complex healthcare systems.

Patients and their caregivers can ask questions like “Am I allowed to drink water before my appointment?”, “How much will I have to pay out of pocket for Lipitor (based on my insurance documents)?”, “What doctor can help me with my foot pain?”, “Should I go to the emergency room or urgent care or my doctor?” and “Is it normal for my mom to spit blood?”

Patients and their caregivers can also use the LLM technology to do basic tasks in healthcare without having to learn complex systems: “Order my next refill for Lipitor,” “Find me a doctor who treats foot injuries” or “Where can I get this medication cheaper?”

There are 350 million patients and their caregivers in the country. Even if we empowered them to do a small part of healthcare, imagine the impact we can have.

When patients can self-service some of their questions and requests, we can also reduce the expense of call centers for hospitals and insurance companies.

This will also reduce the burden on the doctors, nurses and other healthcare workers. They can focus on personalized attentive care rather than routine tasks.

By integrating LLM technologies with existing healthcare infrastructures, hospitals and health systems can significantly enhance patient engagement and self-management capabilities. This not only improves the patient experience by providing instant access to information and simplifying healthcare tasks but also optimizes the use of healthcare resources and professionals’ time.

Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki
Email him: bsiwicki@himss.org
Healthcare IT News is a HIMSS Media publication.