Generative AI will be hailed as the superhero of healthcare technology in the not-too-distant future. Its most important mission: saving lives. The list of lifesaving technologies is long and formidable—pasteurization, blood transfusion, anesthesia, pacemakers, insulin, and vaccines are among the front runners.
However, Generative AI is in a class of its own. The technology promises to shake up the entire healthcare industry value chain, from drug research and manufacturing to MedTech, patient intake, examination, diagnosis and treatment, payer services, regulatory oversight, and the general efficacy of public health agencies.
Overcoming disease, enhancing the quality of life, and extending life expectancy have always been humanity’s greatest challenges. The World Health Organization reports that global healthcare spending increased from US$9 trillion in 2020 to US$9.8 trillion in 2021 (10.3 percent of global GDP). A worldwide survey of adults in 2023 revealed that 61 percent of respondents believed healthcare was unaffordable in their country.
However, the potential of Generative AI in the industry is enormous. This technology can enhance healthcare and make it affordable for a wider population. According to the McKinsey Global Institute estimates, it could generate $60B to $110B a year in economic value for the industry, accounting for between 1.8 and 3.2 percent of industry revenue.
Generative AI is already making its mark across the healthcare industry. It is significantly speeding up the R&D for drug discovery, which, as per researchers, costs billions and takes years. A prime example is Hong Kong-based Insilico Medicine, a Generative AI-driven biotech company. In just 18 months, the company identified a preclinical drug candidate to combat idiopathic pulmonary fibrosis, a respiratory disorder causing an irreversible decline in lung function. The research cost just $2.6M, showcasing the cost-effectiveness and efficiency of Generative AI in healthcare. It took another 12 months to get the preclinical drug candidate to Phase 1.
Canada-based DiagnaMed used Generative AI to analyze electroencephalography signals and predict cognitive decline in patients with mental health and neurodegenerative disorders. Generative AI will, in all likelihood, also be used to ensure these—and other similar –pathbreaking solutions meet regulatory approval quickly and at minimal cost. These are just some early examples of the interventions that Generative AI can create in healthcare. Once they go past the process of clinical trials and regulatory approvals, they will create unimaginable long-term value.
However, Generative AI can be put into more immediate service in the industry, making life simpler for caregivers and patients. We’ll look at three relatively common day-to-day challenges in the industry and how Generative AI can help solve them.
Three Healthcare Challenges Generative AI Helps Solve
Clinical Note Summaries: Physicians spend 57% of their time on documentation, which is often annoying but necessary. High-quality documentation is necessary to ensure that the medical team and future providers have accurate patient histories for rapid diagnosis, arrive at the most effective treatment plans, stay aligned with changing regulatory requirements, aid legal investigations, and meet reimbursement guidelines. In most cases, the clinical notes quickly turn into a mountain of fragmented information in various formats and diverse locations. Data in the intake forms and information gathered during doctor-patient conversations could be considered initially irrelevant and omitted.
However, summarizing clinical notes, extracting information from transcripts, and identifying what is important is critical. Reducing the notes, structuring them, and making the information usable deeply influence how medical teams make decisions related to prognosis and treatments. Generative AI can quickly and easily produce neatly structured, context-specific, and accurate clinical note summaries, discharge summaries, regular patient progress reports, and hand-off summaries. In the process, Generative AI can also save time and reduce the risk of errors.
Image courtesy: John Snow Labs
Real-life deployments of Generative AI to address the challenges of clinical note summaries show that the technology is 30% more accurate than other commonly used processes, such as Bidirectional and Auto-Regressive Transformers (BART), Flan-T5, and Pegasus.
Data De-Identification: While the healthcare industry is rich in data pivotal to providing better care and progressing medical research, the data must be processed to protect patients’ privacy. Using and sharing the data as-is violates legislation and regulations around healthcare. Before the data can be used, all personal identifiers—PII and PHI—must be removed. The identity of patients must be kept confidential without reducing the insights the data contains.
Despite the advances in automation, most data masking and abstraction depend on hand-crafted rules and pre-compiled tables, leading to inaccuracies. Unstructured text also presents the challenge of maintaining medical correctness, consistency, and usability of the de-identified data. Traditional methods require additional time and cost-intensive manual review by human experts.
Using specialty LLM models, designed with built-in mechanisms for adherence to compliance standards like HIPAA and GDPR while enhancing patient privacy, helps overcome many challenges. Generative AI can ensure the data is optimized and de-identified to support collaborative research and improve diagnosis and the effectiveness of treatment regimes.
Databricks and John Snow Labs, leaders in Natural Language Processing-based solutions, have jointly created a scalable industry-specific PHI De-Identification process that makes half the errors that ChatGPT does. This highlights the need for specialized solutions.
Named Entity Recognition: NER helps extract vital patient information from unstructured intake records, doctor-patient transcripts, lab reports, EHRs, and other sources. It extracts the presence and relationships between diseases, drugs, symptoms, and events.
Most NER methodologies are trained on a limited set of clinical entities and rely on vast amounts of (unstructured) data. This makes NER the ideal candidate for Generative AI intervention. Research shows that task-specific prompt frameworks incorporating medical knowledge and training samples enhance the ability of LLMs to improve NER.
These examples indicate the ability of Generative AI to revolutionize the healthcare industry with a long and deep impact. The path to the revolution begins with healthcare enterprises deploying the technology to solve everyday problems that make life simple for caregivers.
Learn more at Trigent’s Website.