Artificial Intelligence in Healthcare: Balancing Clinical Decision Support, Cost, and Safety

Recent developments in healthcare technology highlight both the potential and the current limitations of artificial intelligence in clinical settings. As health technology companies and medical institutions introduce new AI-driven tools, recent studies and software deployments underscore the critical need for careful validation, safety checks, and cost awareness.

Evaluating AI Recommendations and Healthcare Costs

A study recently published in NEJM AI evaluated how general-purpose and specialized medical large language models handle real outpatient and emergency visits. Researchers from the Technion, Stanford University, and collaborating institutions tested 24 AI systems on medical notes from 200 real visits. The study found that AI systems recommended more up-front testing, specialist referrals, and treatments than the physicians who actually treated the patients, resulting in care that was more than three times as costly.

Using 2026 Medicare rates, the average cost of AI-suggested care was $314 per visit, compared to $97 for physician-recommended care. Diagnostic testing and specialist referral rates were both significantly higher in the AI-generated plans. However, the study authors noted that the notes used came from a convenience sample of medical transcription reports rather than a representative draw of U.S. ambulatory practice. When the researchers tested cost-aware prompting, they found that instructing models to prioritize parsimonious and cost-effective care reduced costs closer to physician baselines, though diagnostic agreement sometimes decreased.

New Safety Features for Medication Dosing

In response to known vulnerabilities in probabilistic models, digital health platform Doximity launched DoseCheck, a deterministic, rule-based verification layer within its clinical decision support tool. Testing by the company across nine widely used AI models revealed that a small percentage of dosing claims were incorrect, with a notable portion classified as serious or potentially lethal. DoseCheck is designed to verify AI-generated dosing claims against structured information such as FDA labeling and professional guidelines to mitigate these risks.

Education, Workforce Management, and Specialty Care

Other recent developments across the digital health landscape include:

• Healthcare AI Training: Google launched an educational certificate program designed for healthcare professionals, accredited for continuing education credits and adopted by several medical and nursing schools to teach responsible AI use in administrative and communication tasks.
• Specialty Compounding: Following federal guidance permitting pharmacies to manufacture certain unapproved peptides under enforcement discretion, platforms like Hims & Hers announced preparations for a peptides specialty backed by independent providers and clinical support.
• Stethoscope Innovation: Eko Health received FDA clearance for its CORE 300 stethoscope, which blends traditional auscultation with digital amplification and AI-enabled structural murmur detection.
• Operational Tools: Companies such as RevSpring and symplr introduced conversational AI payment agents and workforce management solutions aimed at streamlining administrative workflows and nurse scheduling.
• Institutional Oversight: The Cleveland Clinic Cancer Institute established a dedicated AI task force to continuously evaluate model accuracy, screen for algorithmic bias, and monitor long-term safety.

As health systems navigate the rapid integration of machine learning tools, experts emphasize that ongoing longitudinal assessment and rigorous oversight remain essential to ensure that technology enhances rather than complicates patient care.