Health care systems can leverage generative AI to enhance patient care delivery in numerous ways. The technology can be used to analyze unstructured data such as clinical notes, radiology reports and patient histories to extract valuable insights that can aid in diagnosis and treatment.
For example, AI can identify patterns and correlations that might be missed by human clinicians, leading to more accurate and timely diagnoses.
A report from McKinsey noted that while disparate sources of unstructured data can be leveraged as assets to power generative AI, risks abound, requiring a comprehensive integration strategy around data security, meaningful use and monitoring.
Dr. Harvey Castro, a physician and health care consultant, explains unstructured data sources in health care are vast and varied, including clinical notes, radiology and pathology reports, electronic health records (EHRs), patient histories, genomic data and even data from wearable devices.
“These sources contain a wealth of information that can be harnessed to power generative AI applications,” he says. “However, harnessing these data sources effectively requires sophisticated data processing and machine learning techniques.”
Natural language processing (NLP) can be used to extract meaningful information from text-based sources like clinical notes and EHRs.
“Image recognition algorithms can analyze radiology and pathology reports,” he adds. “Predictive modeling can be used to forecast health outcomes based on genomic data and patient histories.”
Manoj Saxena, founder and chairman of the Responsible AI Institute, points to manifold applications for GenAI in health care, from clinical decision support to drug discovery and development.
“GenAI models predict the efficacy of potential drug candidates by analyzing research papers and chemical databases, as well as vast volumes of research data to identify trends and gaps for further medical exploration,” he says.
He adds GenAI can develop personalized educational content for medical students and health care professionals to enhance knowledge and understanding.
“There’s a copious amount of unstructured data in health care,” he says. “Generative AI can help wrangle these data sets to help medical professionals be more efficient and productive.”
For example, unstructured clinical notes and patient data from EHRs can be processed by generative AI to assist in clinical decision-making and personalized treatment recommendations.
“Unstructured data from medical images, such as X-rays and MRIs, can be run through generative AI models to aid radiologists in diagnosing and detecting abnormalities,” Saxena says.
Castro points out patient data protection is a critical concern when using unstructured data for AI applications in health care, noting there are several strategies health care organizations can employ to ensure data security.
“Firstly, data anonymization techniques can be used to remove personally identifiable information from the data used in AI applications,” he says. “This can help protect patient privacy while still allowing for the analysis of valuable health data.”
Secondly, robust data security measures– including secure data storage, encryption and strong access controls–should be in place to protect against data breaches.
Lastly, Castro recommends health care organizations consider the use of privacy-preserving AI techniques, such as differential privacy and federated learning.
“These techniques allow for the analysis of data in a way that protects individual privacy and can be particularly useful in the context of generative AI applications in health care,” he says.
Heather Lane, senior architect of data science for Athenahealth, agrees it is imperative to design AI-based features to integrate with existing security mechanisms and to remain vigilant around healthcare data privacy.
“For example, when negotiating contracts with third party AI vendors such as OpenAI or Google, providers must ensure strong contractual guarantees that any data that passes through their AI systems will not be stored or released, and they will handle it with appropriate security precautions,” she says.
Beyond that, providers need to take caution when fine-tuning models, for example, when considering what data they can and can’t use for tuning.
“They must maintain those decisions over the lifecycle of the trained models,” Lane adds.
Overall, there are numerous ways in which the health care sector can utilize AI technology, as long as proper caution is taken during and after implementation.