At the recent HL7 conference, I had the opportunity to connect with peers who, like me, work at the intersection of healthcare and technology, bridging the crucial gap between clinicians and developers. One clear consensus emerged: the use of AI in healthcare is here to stay. In this blog, I’d like to share my reflections on the near-term role of AI in healthcare and its impact on our industry.
The dual-nature of AI
I always say that AI in healthcare is a double-edged sword. On one hand, it promises operational efficiencies that can revolutionize how healthcare providers deliver care. On the other, it poses risks when applied to patient diagnostics without proper safeguards.
One of the most immediate benefits of AI is in automating routine tasks that consume valuable time for healthcare professionals. Consider clinical documentation – a process that can be laborious and time-consuming. AI-powered transcription tools can capture doctor-patient interactions in real-time, generating accurate summaries and freeing up clinicians to focus more on patient care.
Natural language processing (NLP) algorithms can transcribe and summarize consultations in a matter of seconds – this can significantly reduce the administrative burden on clinicians. These tools can integrate seamlessly with Electronic Health Record (EHR) systems so documentation is completed promptly. In addition to a more seamless and timely discharge process of patients, hospitals also save time with an expedited process which directly improves cash flow and operations.
Unlocking insights
Another area where AI and machine learning (ML) unlock unparalleled values is in analyzing large datasets to identify patterns that would be nearly impossible for humans to detect. Patients with chronic conditions often have years' worth of medical data – lab results, imaging studies and treatment histories. Machine learning algorithms can sift through this data to find trends, correlations and anomalies that could inform treatment plans.
For instance, AI can analyze longitudinal lab results to detect subtle changes over time, potentially identifying early indicators of disease progression or treatment efficacy. This capability is invaluable for personalized medicine, where treatments are tailored based on individual patient data.
Diagnostic bias and misapplication
While AI has immense potential, it is not without risks, particularly in diagnostics. AI models are trained on datasets that may not be representative of the diverse patient populations we serve. This can lead to biased algorithms that produce inaccurate or even harmful recommendations.
For example, an AI diagnostic tool trained predominantly on data from Caucasian populations may not perform accurately when applied to patients of different ethnic backgrounds. This is not just a theoretical concern; there have been documented cases where AI models failed to identify conditions in certain demographic groups due to biased training data.
Furthermore, the current LLMs still lack the contextual understanding and clinical judgment that healthcare professionals bring to patient care. It cannot account for nuances such as a patient's socioeconomic status, cultural background or personal preferences – factors that are often critical in clinical decision-making. While AI shows promise in healthcare, it currently faces significant technical limitations and cannot fully replace human clinicians. The technology serves better as an augmentation tool rather than a replacement for human medical expertise. AI can provide recommendations or highlight areas of concern, but the final decision must rest with the healthcare professional who can interpret these suggestions within the broader clinical context.
Interoperability and AI integration
Integrating AI into healthcare systems is not just about implementing algorithms; it's about ensuring these tools can communicate effectively within existing healthcare IT infrastructures. This is where interoperability standards like HL7 and FHIR become critical.
From a technical standpoint, AI can significantly streamline interoperability projects. One of the most time-consuming aspects of these projects is data mapping—aligning data fields from one system to another. AI can automate a substantial portion of this process.
For instance, machine learning algorithms can learn that the "Patient ID" field in one system corresponds to the "Patient Identifier" in another. By automating up to 75% of the mapping work, AI allows IT teams to focus on the more complex mappings that require human judgment. This not only accelerates project timelines but also reduces the likelihood of human error.
AI in SaMD
For software medical device companies, AI offers opportunities to enhance device functionality and interoperability. The AI in the medical devices market is projected to reach $97.07 billion USD by 2028 with opportunities of algorithms embedded within devices to provide real-time analytics, predictive maintenance or adaptive user interfaces.
Currently, all AI-powered medical device systems must operate as “locked systems,” meaning their algorithms are fixed and cannot adapt post-deployment, even though AI is inherently capable of continuous learning. Moving toward adaptive AI systems – those that can learn from new data and adjust accordingly – requires a robust technical architecture that meets both regulatory and performance standards.
FDA’s Total Product Lifecycle (TPLC) approach addresses this complexity by combining premarket evaluation with ongoing post-market monitoring. This framework allows for proactive management of AI’s evolving nature, ensuring that adaptive systems remain compliant and effective throughout their use in healthcare.
The integration of AI in healthcare is both inevitable and transformative, yet it must be approached responsibly. Regulatory bodies like the FDA are actively developing guidelines to ensure that AI is adopted safely and ethically. HCPs and medical device manufacturers should proactively understand these evolving regulations and embed compliance into their digital products and processes from the outset, laying a foundation for trustworthy, patient-centered innovation.
A 2025 view
AI holds great promise for improving healthcare delivery, but it is not a panacea. Its immediate value lies in automating routine tasks and unlocking insights from vast amounts of data. However, healthcare organizations must urgently address their technical debt, as approximately 73% still rely on legacy systems for core clinical operations. 1.3 billion people around the world already use digital health in 2024 and that number will continue to increase, HCPs cannot afford to maintain outdated systems that lag 15-20 years behind other sectors.
Success in AI implementation requires not just financial investment, but also strategic allocation of human resources, modernization of legacy infrastructure and careful avoidance of siloed solutions that could hinder interoperability. At Star, we are committed to bridging the gap between healthcare and technology, ensuring that AI and other digital tools are integrated seamlessly and effectively. Get in touch and let us help you to alleviate administrative redundancies and build a more connected care for your patients.
