Ngan MacDonald

HTI-2 is here. Laboratories will want to take note as there are requirements for use of Laboratory Orders Interfacing (HL7 version 2.5.1) and Laboratory Results Interfacing (HL7 version 2.5.1) for public health reportables (ELR). If you don't yet have HL7 2.5.1 interfaces with trading partners (paper, fax, HL7 version 2.3.1, etc) you should make plans to update interfaces soon. This impacts both senders and receivers of laboratory data. If you are a pathology laboratory, this may impact you if you perform any public health reportables like Creutzfeldt-Jakob Disease(CJD) as it is reportable in many jurisdictions. ONC has webinars. EHR vendors, providers, and those performing specimen collections will want to ensure they send complete information such as Specimen Type, Specimen collection procedures, Specimen Source site and other Ask at Order Entry (AOE) details required by the performing laboratory and/or public health for reportables.

16

1 Comment

"A new survey conducted [...] revealed that healthcare stakeholders view data analytics as key to meeting many of their strategic goals, but struggle with issues like data quality and consolidation. [...] Over 80 percent of those surveyed indicated that features to improve data quality are must-haves in a data platform, whereas 65 percent highlighted the need for a comprehensive enterprise data solution. Over three-fifths prioritized features that enable clinical staff to use data to enhance their productivity, and 60 percent wanted the ability to consolidate data across multiple disparate platforms." https://lnkd.in/gFVmrgSr

3

1 Comment

Does your data foundation support your AI strategy? Join our discussion next week!

25

The Secret to Driving AI Impact Isn't in the Algorithms – It's in the Execution. Over my 10 years in the AI/ML industry, leading teams at IBM Watson, Centene Corporation, and Datavant, I've seen one common thread that separates successful AI initiatives from failures: It’s not just about having the smartest data scientists/MLEs or the most cutting-edge algorithms – it's about bridging the gap between AI potential and business value. Here’s what I've learned about translating AI into real business impact: 1. Start with the Problem, Not the Tech Too often, I see companies enamored with the latest AI trends without a clear understanding of the business problem they aim to solve. The most successful projects I’ve led started with a well-defined pain point. AI is only effective when it’s solving a real, high-priority issue. 2. Build a Cross-Functional Team AI isn't just for data scientists. You need domain experts, product managers, and operational leaders who understand the intricacies of your industry. When I led a project that saved a healthcare company millions, it wasn’t just AI magic – it was a collaborative effort between our tech team and healthcare professionals. 3. Establish a Robust MLOps Framework Early Deploying a model isn't the finish line – it's just the beginning. One of the proudest achievements in my career was building an MLOps platform that allowed my team to monitor, iterate, and scale models efficiently. This ensured our solutions adapted as the business evolved, rather than becoming outdated and irrelevant. 4. AI Governance is Non-Negotiable Data privacy, ethical considerations, and transparency are not just buzzwords. Establishing an AI governance framework should be a top priority. When I implemented AI governance at scale, it not only safeguarded us from potential risks but also built trust with stakeholders, accelerating AI adoption across the organization. 5. Communicate the Value to Non-Technical Stakeholders One of the biggest challenges in AI leadership is getting buy-in from non-technical stakeholders. Early in my career, I struggled with this, but now I focus on framing AI outcomes in terms of revenue, cost savings, or customer experience improvements. This shift in communication led to multiple successful AI initiatives gaining executive sponsorship. 💡 Pro tip: Stop obsessing over “explainability” in purely technical terms. Instead, illustrate how the model’s insights translate into actionable business strategies. I’d love to hear your experiences – what’s the biggest challenge you’ve faced in implementing AI? How did you overcome it? #AILeadership #AIExecution #MLOps #BusinessImpact #AIProductManagement

6

This figure represents a human-centered, health-data-driven ecosystem that places the patient at the center, surrounded by four key quadrants. These quadrants include administrative and financial data (who am I?), logistics and facility services data (where am I?), medical data (am I healthy?), and paramedical data (how do I recover?). The figure highlights how these different data types interact and are integrated to ensure a seamless connection between the patient and healthcare providers. The goal is to improve the quality and continuity of care through the effective management and flow of data across various stakeholders. ✍🏻 Stevens, G., Hantson, L., Larmuseau, M. et al. (2022). A human-centered, health data-driven ecosystem. Discov Health Systems 1, 10. DOI: 10.1007/s44250-022-00011-9

176

12 Comments

https://lnkd.in/eb_ZRqVC Progress on integrating Public Health into the healthcare data ecosystem. Goal: 'Rather than having each public health agency build and maintain cumbersome interfaces with each health care provider in their jurisdiction, the Trusted Exchange Framework and Common Agreement enables public health agencies to securely interoperate with provider organizations across the country using a single interface and user agreement.'

12

🔍 Exploring LLM as a Judge: Enhancing RAG Evaluation Continuing our deep dive into Enterprise RAG architecture & design best practices from the week before (https://lnkd.in/eSggTNyE), and building on our discussion of evaluation-driven development from last week (https://lnkd.in/eAsiprjH), let’s dive into the interesting concept of using LLMs as judges in your Enterprise RAG applications. Considered as a good trade-off between eyeballing and having human intervention on every response, this approach offers scalability and consistency. Quick recap on why Use LLMs as Judges? 1/ Scalability: LLMs can process vast amounts of data quickly, making them ideal for large-scale evaluations. 2/ Consistency: Unlike human evaluators, LLMs provide uniform assessments, reducing variability in evaluations. 3/ Comprehensive Analysis: They can evaluate multiple dimensions such as relevance, coherence, and factual accuracy. But how does it work? The effectiveness of LLM judges relies on high-quality training data provided by your most priced assets (i.e. expert human intervention). Here is a high level overview of how these LLM Judges are trained by humans: > Human-Curated Datasets: LLMs are fine-tuned on carefully curated datasets containing: High-quality Q&A pairs. E.g. {"Q": "Coverage limit for out-of-network PT under Plan X?", "A": "Plan X covers 60% of allowed amount, up to 20 visits/year."} Good/poor response examples. E.g. {"Good": {"Q": "Appeals process for denied claim?", "A": "Submit written request within 180 days. Include member ID and claim number."}, "Poor": {"Q": "Appeals process for denied claim?", "A": "Contact customer service."}} Expert annotations. E.g. {"type": "coherence", "Q": "Submit out-of-network claim?", "A": ["1. Get itemized bill", "2. Download form", "3. Fill & attach bill", "4. Send to claims dept"], "annotation": {"rating": 5, "comment": "Clear steps"}} > Diverse Training Data: Datasets include a wide range of topics, question types, and difficulty levels to ensure broad applicability. It is best practice to leverage transcripts from prior customer interactions and/or training manuals for contact center employees to help train your LLM judge. > Iterative Refinement: Continue evaluation-driven development. Your LLM judge may need multiple training rounds, with human experts validating and refining performance. Stay tuned for my next post, where I’ll dive deeper into evaluation criteria and how to effectively utilize LLM Judges in your RAG systems! 💬 How are you leveraging AI to evaluate your RAG systems? ♻️ Subscribe to my newsletter & repost if you find value in these insights: https://lnkd.in/g3bdneR7 #enterpriserag #evaluationtechniques #aievaluation #genai #datascience #machinelearning #aistrategy #cto #cdo #aicouncil #aws #enterpriseai #aiadoption #digitaltransformation #healthcarepayers #healthcareai #insurtech #cio

13

As technological innovations continue to unfold and collaborative efforts intensify, we can expect a transformative impact on how healthcare information is shared and accessed. #fhir #interoperability #api #hl7 #digital #wellola https://lnkd.in/efzR3Nub

4

Great discussion of how federated data sharing and a unified ecosystem for healthcare research can lead to rapid advances. Recommend this if you want to see advances in patient care and treatments for the toughest diseases. #AYDIJ #datainnovation #healthcareanaltyics #researchanalytics #pivotpointconsulting

6

Our #RapidEvaluation conference brings the community together to discuss challenges and opportunities in health and care. In this blog, Senior Data Analyst Tatjana Marks shares overarching themes from this year's conference. Read more and watch session recordings ⬇️ https://lnkd.in/ejzTY5Rk [Alt text: 'It's through collaboration that we can address the problems that matter for most patients and ensure the use of evidence in decision making.' Tatjana Marks, Senior Data Analyst at the Health Foundation]

9

1 Comment

Interesting perspective on #TechnologicalDebt in health care by Dave Lamar focused on the provider market, but we seem the same issues within state depts and payers. While good data governance practices are key, there must be a better way to solve this common - if not universal - challenge in health care, if only by offloading some of the administrative responsibilities that require the individual management and administration of no-value-add administrative functions. #healthdata

7

4 Comments

🚨 🚨 The decision over cost to obtain medical records using FHIR API in #claimsmanagement and #litigationmangement is happening. 1. Seeking comment on threshold condition for Attorney requests. (P. 531) 2. Seeking comment on fees for TEFCA Sub participants. (P.561) 3. Info session July 17 (Not likely on fees for injury claims and litigation.)

Member health information is critical to ensuring better care, improved outcomes, and lower costs, but how should health plans be collecting and organizing this information? Costa Cassis, our VP of Data, is starting this discussion with an overview of Member 360 Data Models. Swipe below + follow us to learn more about how Member 360 can transform how your organization serves its members.   #member360 #datamodels #healthcare #healthcareconsulting #healthdata #patientinformation #patientdata #memberinformation #memberdata #dataanalytics #improvedoutcomes #lowercosts #quintupleaim #healthplans #healthinsurance #cms #medicare #medicaid #medicareadvantage

14

2 Comments

I will be very interested to see where this goes. Changing human behavior around health is difficult, particularly with the senior population. Additionally, affordability of healthcare is becoming more challenging. Healthcare organizations might follow this model after seeing the results. #ai #artificialintelligence #machinelearning #gpt4 https://lnkd.in/gyw8p-QD

15

1 Comment

If you're looking for an informative session to wrap up your week, consider registering for this Technology Community of Practice session that will talk about driving better outcomes, overcoming data obstacles, and applying trauma-informed principles to your data practices. See you there!

1

The Office of the National Coordinator for Health Information Technology (ONC) has proposed new rules aimed at enhancing data interoperability within the healthcare industry. These regulations will significantly impact health plans by imposing new data standards, encryption requirements, and API certifications. Key Provisions: - Data Standards: Health plans must adopt the newer USCDI version 4 for data exchange, replacing version 3. - SMART on FHIR: Plans must upgrade to SMART v2.2 to enable third-party app access to EHR data. - Data Encryption: Stricter encryption standards are mandated to safeguard patient data. - API Certifications: Voluntary certifications for payer-to-payer, prior authorization, and provider directory APIs are introduced to streamline processes. Learn more: https://hubs.ly/Q02K6qlc0

1

It’s not enough say that Electronic Medical Records are failing us and we just need to force clinicians into collecting better data up front. It’s deeper than that.  Especially if you want your data to power research, precision medicine, population health, or your shiny new generative AI strategy, you need to be strategic about ensuring you have the right data. Lots of talk in our industry about data being a strategic asset, but data is strategic only if you give it the appropriate strategic thought and time. The very first step in developing a data strategy is framing the most important questions. I strongly advocate for my clients to develop what I call North Star use cases. North Star use cases represent the questions that are most important to your organizational strategy and that resonate deeply with stakeholders. An example North Star use case developed with an NCI-designated Cancer Center: For patients with [Disease] at [Stage] with/without the following [Comorbidities] and/or [Gene Mutations], what [Treatments] were provided, what were the [Responses] to those treatments, and what were the [Outcomes]? Hard to imagine advancing cancer care without that fundamental question. It applies to important operational, clinical, finance, research, and health equity questions. Data exists in the medical record, but it is organized around a patient and their encounters with the health system. Outside records are either difficult to access and analyze or nonexistent. Important relationships between North Star data elements are not represented or difficult to discern. An example: The patient has multiple primary cancers. What are the subsites for those primaries? Which cancer is their chemotherapy treating? Which course of treatment is it? Which imaging studies measure the treatment response for this treatment? Because of the difficulty of getting at the data, manual data curation is performed to fill the gaps, often repetitively across departments. This isn’t sustainable! The benefit of a data strategy is in helping to articulate your goals (i.e. answering the North Star-type use cases) and developing a systematic plan for addressing barriers to those goals. Is always instructive how many stakeholders don’t know what kinds of questions they are supposed to be able to answer with data and analytics. The good news: you can chart a strategic path to better data (and better AI on top of it). Develop a North Star. Deconstruct the North Star, identify your critical data elements. Find and articulate related data quality issues and missing relationships. Develop interventions (both EMR and post-EMR manual/AI curation) to fill those gaps while measuring and communicating the results. Communicate your plan and win the hearts and minds of your data consumers by giving voice to their hopes and desires with a solid, relatable, non-tech-speak plan. Just Health Technologies LLC #healthcareai #datastrategy

65

10 Comments