Tisson Mathew

Duplicate patient records are a common occurrence in healthcare data. FHIR attempts to address this by providing a detailed spec for how to implement the Patient $merge operation. If you haven’t looked into this yet, it probably means your FHIR project is still early stage. There is simply no way to avoid duplicate patient records in real world scenarios. Examples of why duplicate patient records might exist: - Combining multiple datasets - Emergency admissions, where patient details might be incomplete - User error, name changes, misspellings, etc. Support for advanced FHIR operations across the various FHIR server providers is weak. Users are often left to “roll their own”. I have yet to find a server provider that supports /Patient/$merge out-of-the-box. The $merge operation accepts two Patient resources: - The Source: the old Patient about to be merged. - The Target: the new or “primary” Patient resource. The merge process updates the Patient link element of both resources. The Source Patient is given a Link.type of “replaced-by” and references the Target. The Target Patient is given a Link.type of “replaces” and references the Source. The Source Patient is then set to inactive. In some cases it may be deleted. A “housekeeping” step is then performed to update all FHIR resources that reference the Source so that they now reference the Target Patient. This process is asynchronous and may take some time, as years of data might need to be updated. During that time, results can be erratic and incomplete. What to do after the merge in the event of a GET request involving the Source Patient is very much up to the FHIR server. - Return the inactive Source Patient - Return the Target Patient - Return both The FHIR documentation in this area is comprehensive, but does raise a lot of questions that need to be addressed during implementation. https://lnkd.in/gXraAHxk A final word of caution. How /Patient/$merge is implemented is as much a business decision as it is a technical decision. Don’t let your devs have free reign on this one!

304

34 Comments

Healthcare AI models also fail because the “lab” (training) data is "too clean" 📉 If your underlying healthcare AI model is trained on curated, standardized datasets in the lab- it is a liability. 🛑 #GRIF Post 3: The Data Heterogeneity Gap. A core pillar of the Global Readiness & Implementation Framework (#GRIF) is addressing the “Domain Gap”—the chasm between clean training data and the messy, fragmented reality of the clinical and global bedside. 🏥 Consider a skin-based infectious disease detection AI app on a smartphone. 🩺 Trained on high-res lab images, it performed flawlessly. But in a trial setting, it hit a wall of "Out-of-Distribution" (OOD) data: -Blurred images from legacy hardware 📸 -Multi-device inconsistencies -Divergent Android vs. Apple image pre-processing 📱 The result? Sensitivity and specificity plummeted when it mattered most. 📉 #Takeaway: You must architect for the "noise." GRIF protocols demand stress-testing models against "dirty" data—missing fields, low-res inputs, and non-standard formats—long before market entry. If your AI can’t handle the mess, it can’t handle the market. 💼 How do you prepare your models for "data in the wild"? Let’s talk strategy below. ✍ #DataScience #GlobalHealth #AI #RealWorldEvidence #HealthTech #GRIF #HealthInnovation Gentrac Labs

6

Healthcare operations aren’t broken. They’re overloaded. AI can reclaim 20–40% clinician time and drive measurable efficiency but only if implemented right. This guide covers: • High-impact use cases • Data & governance essentials • A clear pilot → scale roadmap • Mistakes that delay ROI Operational AI isn’t about automation. It’s about clarity, speed, and sustainable systems. Save this for your leadership team. What’s your biggest operational bottleneck today?

39

12 Comments

OpenAI just launched AI for Healthcare. HIPAA compliance. Evidence retrieval. Clinical reasoning benchmarks. It's impressive work for clinicians. But there's a gap no one's talking about. 63 million Americans are family caregivers. They're doing healthcare at home - unpaid, often untrained, frequently alone at 2am when something goes wrong. When they turn to AI for help, what happens? We tested 10 frontier models on caregiver crisis detection. The results: Best safety score: 49% AI misses crisis signals in 1 out of 2 conversations OpenAI's benchmarks (HealthBench, GDPval) measure clinical reasoning for doctors. Important work. But they don't measure what happens when a daughter asks an AI if her exhaustion is normal, and the AI misses that she's describing burnout so severe it's become a crisis. Healthcare AI is optimizing for the exam room. Caregiving happens in the living room. That's a different kind of safety - and right now, almost no one is measuring it. bench.givecareapp.com

8

Big ideas in critical care are coming to SCCM Join NKDHS on March 23rd at 11:30 PM for a session diving into how predictive analytics is reshaping bedside decision-making in real time. We’ll cover: 🔹 How machine learning models—validated across multiple sites—can help predict hemorrhage before it becomes critical 🔹 How heart rate variability (HRV) can signal risk for post-CABG atrial fibrillation using advanced predictive platforms From early warning to smarter interventions, this session is all about turning data into actionable insights when it matters most. Let’s push the boundaries of what’s possible in critical care—hope to see you there! #PredictiveAnalytics #UnitingData #RevolutionizingCare #NKDHS

37

Why Small Language Models (SLMs) Are Powering High-Precision Prior Authorization in Healthcare Prior authorization remains one of the most expensive and friction-heavy workflows in healthcare. The challenge isn’t lack of data — it’s precision. Large, general-purpose language models are impressive, but prior authorization demands something different: deep domain understanding, deterministic behavior, and consistent outcomes across payer-specific rules. This is where Small Language Models (SLMs) are proving to be far more effective. SLMs trained specifically on prior authorization workflows, payer policies, CPT and ICD mappings, medical necessity rules, and historical authorization outcomes can operate with significantly higher precision. They reduce hallucination risk, handle edge cases more reliably, and align better with compliance and audit requirements. More importantly, SLMs enable intent-driven automation. They can identify authorization requirements early, extract the exact clinical evidence needed, assemble payer-specific submissions, and flag gaps before a request ever reaches the payer. The result is fewer delays, fewer resubmissions, and faster patient access to care. Another advantage is deployability. SLMs are lighter, faster, and easier to operate in private or hybrid environments — a critical requirement for healthcare organizations that care about data control, latency, and regulatory boundaries. At Contiinex, we see prior authorization evolving from a manual, reactive process into a predictive, AI-assisted workflow powered by specialized SLMs — not generic models. #PriorAuthorization #HealthcareAI #SLMs #RevenueCycleManagement #DigitalHealth #Contiinex

11

TODAY's HEALTH-AI BUZZ BEYOND THE HYPE: FOUR STRATEGIES FOR REALIZING HEALTHCARE AI'S POTENTIAL The latest in Healthcare AI The promise of AI in healthcare is immense, but successful implementation remains a challenge for many organizations. 💡 At HIMSS24, experts shared critical insights to move from pilot to meaningful impact, emphasizing a strategic approach over simple technological adoption. ▶︎ Focus on Value First: Don't just implement AI because it's new. Identify clear, measurable problems AI can solve, starting with smaller, focused projects that demonstrate tangible ROI before scaling. This ensures technology serves a purpose, not the other way around. ▶︎ Prioritize High-Impact Use Cases: Target areas where AI can significantly reduce burden or improve outcomes. Think administrative automation, enhancing clinical documentation, or optimizing patient flow. These quick wins build momentum and confidence within the organization. ▶︎ Uphold Data Quality and Governance: AI is only as good as the data it learns from. Robust data infrastructure, quality control, and strict governance are non-negotiable for accurate, reliable, and ethical AI applications. This foundational work prevents costly errors down the line. ▶︎ Engage End-Users from Day One: Clinician buy-in is paramount. Involve doctors, nurses, and staff in the development and deployment process. Their feedback is vital for creating user-friendly, effective tools that truly meet their needs and alleviate their daily pressures. These strategies highlight a shift from theoretical AI to practical, patient-centric solutions. The emphasis on collaboration and measurable impact underscores the need for a holistic approach, integrating technology with human expertise and organizational goals. Successfully navigating these points will be crucial for healthcare systems looking to truly leverage AI. What is your organization's biggest hurdle in adopting AI, and how are you addressing it? Share your insights! 👇 #HealthcareAI #DigitalHealth #AIinHealthcare #HealthTech #HIMSS24 #Innovation #PatientCare #FutureofMedicine We hope you liked the buzz or today Connect with us for more on health-AI #AI #healthcare #news

1