The new lifeline for overwhelmed healthcare systems

As the world’s population continues to grow and age, the healthcare system in several regions is moving closer to the brink of collapse. According to the World Health Organisationthe current number of health professionals, including doctors, radiologists and other professionals, is not sufficient to cope with the increasing workload. Additionally, increased stress and burnout due to the increase in cases is causing many to leave the field, further reducing the number of practicing workers. Becker Health estimates show that nearly 72,000 U.S. physicians will have left the workforce between 2021 and 2022, and that some 30,000 physicians joining the workforce will not be enough to meet growing demand.

In essence, both of these challenges – increasing workload and decreasing workforce – leave one major impact: a reduced quality of patient care. This is where the much-discussed generative AI can come into play, saving healthcare providers valuable time and resources and allowing them to focus on improving clinical outcomes.

Understanding the potential of generative AI

First, it’s important to understand that AI is not new to healthcare. Organizations have been experimenting with predictive and computer vision algorithms for some time, specifically to predict the success of treatments and diagnose dangerous diseases earlier than humans. However, when it comes to generative AI, things are still quite new as the technology only came to prominence a few years ago with the launch of ChatGPT. Gen AI models use neural networks to identify patterns and structures in existing data and generate new content such as text and images. They are applicable across industries, including healthcare – where organizations cumulatively generate approximately 300 petabytes of data every day.

Now that the generation of AI has the ability to learn from data and create something new, it can’t completely replace doctors or do the work they do, but it can ease the strained healthcare pipeline by expanding certain aspects of the system. This can be anything from simplifying patient journeys and teleconsultation to processing clinical documentation and providing relevant information when the doctor is in surgery.

Let’s take a look at the most feasible applications with gen AI possible today.

AI assistants for medical guidance

After COVID-19, most organizations launched remote consultation services, where patients could contact the doctor without actually visiting the hospital. The approach worked, but left doctors overworked as they dealt with both online and offline patients. With gen AI, healthcare organizations can launch LLM-enabled AI assistants to address this. Essentially, they could refine models like GPT-4 based on medical data and build assistants that can handle basic medical cases and guide patients to the best treatments based on their systems. If a particular case seems more complicated, the model can refer the patient to a doctor or the nearest healthcare provider. In this way, all cases would be addressed without putting enormous pressure on doctors. Several organizations, including Sanofi, Bayer and Novartis, have taken this approach and launched AI assistants on their respective platforms.

Agents for streamlining administrative work

In addition to assessing conditions and providing guidance, generative AI chatbots can also be built to perform basic healthcare operations such as booking appointments and reminding patients of their scheduled visits. This could reduce the hours human operators have to spend handling an increasing number of calls and messages in healthcare systems.

Among the providers who use conversational AI agents are Mercy Health, Baptist Health and Intermountain Healthcare. They’ve all launched bots to automate tasks like patient registration, routing, scheduling, FAQs, IT helpdesk ticketing, and prescription refills. Additionally, many have even begun deploying Gen-AI copilots that listen to the conversation between the patient and physician and generate summarized clinical notes, saving physicians the effort of manually documenting and storing the information in an EHR . Nabla, one of the providers of such copilots, even uses these notes to generate a series of patient instructions on behalf of the physician. This capability can be further developed into a gene AI system that sits next to the doctor and creates personalized treatment and therapy plans based on current conditions and previously recorded parameters, including genetic makeup, health history and lifestyle.

Retrieve data in workflow

One of the biggest strengths of LLMs is that they can be extended with Retrieval Augment Generation (RAG) to tap additional data sources without retraining. This allows healthcare organizations to build in-house smart assistants or search systems that can provide the most relevant, contextual answers for any given question. For example, RAG-based systems can help physicians with decision support by providing evidence-based recommendations for a specific condition.

In other cases, they may produce evidence-based medical reports/patient data from EHR systems or share the latest clinical treatment guidelines. Based in San Diego Nanome used this technique to develop an assistant that leverages large language models (LLMs) and access to real-time internal data and molecular simulation systems to assist pharmaceutical teams in their drug development workflows.

Data analysis, report generation

Another notable application of generative AI would be data analytics, specifically the analysis of medical images such as CT scans, MRIs, and X-rays. Even after rapid digitalization, most diagnostic agencies today rely on human experts to study medical images and write reports for patients. The work takes a lot of time and effort and is even prone to errors arising from inherent biases or simply human fatigue.

With a generational AI-powered approach, teams were able to refine models like GPT-4 vision and use them to study medical data and generate reports, automating and accelerating the entire process for good. Yes, the idea is still fresh, but early experiments show it to be a promising application of gene AI in healthcare. In fact, a JAMA Network study found that AI-generated reports for chest x-rays had the same level of quality and accuracy as those produced by human radiologists.

Drug development

Finally, thanks to its ability to understand intricate patterns and structures in complex medical data, generative AI can also aid in drug development. The technology can assess unique markers of a particular disease and come up with new combinations of chemicals or new molecular structures that could lead to potential drug candidates. It can even screen the generated compounds based on their characteristics and predict side effects and drug interactions.

Just last year, Insilico Medicine’s gene AI-generated INS018_055 drug for idiopathic pulmonary fibrosis, which affects about 100,000 people in the US, has entered clinical trials in humans and is now moving closer to wider release.

Caution is a must

Despite these potentially transformative applications, healthcare organizations must understand that generative AI will only be as good as the data it is trained/tuned on. If the data is not properly prepared or contains any form of bias, the models’ outputs will also reflect these issues, which will damage the company’s reputation. This means that organizations must first prepare the data in the best possible way – and remove personally identifiable information (PII) from it – and then move to the next phases of the project lifecycle, including training and inference, to leverage the efforts of clinicians and administrative staff to some extent. Less intensive.