A cure for the cost crisis: How healthcare providers can improve efficiency and care with a data platform

Artificial Intelligence (AI) has been one of the most promising and disruptive technologies in recent years, especially with the advent of large language models (LLMs) that can generate natural language from unstructured data. Healthcare executives are looking for ways to leverage this technology, especially given the harsh economic climate many provider organizations are facing post-pandemic.

But despite all its promise, AI has challenges that warrant careful consideration, especially around “hallucinations”, which is a euphemism for AI making things up that are not true but sound like they could be true. This can have serious consequences for clinical decision-making and patient safety, if not detected and corrected.

The attractiveness of LLMs is clear – Healthcare is characterized by a large portion of unstructured or uncoded data that is hard to interpret by machines. We know there would be a huge benefit from utilizing the knowledge hidden in unstructured data, such as provider progress notes and observations, but with conventional technology, this has been hard to do.

The quality challenge for LLMs

The quality problem for LLM is not new – AI in all its forms has always been dependent on the quality of data it has been trained on. From early AI computer vision projects in the medical imaging realm, we know it is not good to rely on a black-box model that provides results, without disclosure of what the results are based on. All in all, in computer vision we have made less progress than what was originally hoped, and to some extent that could be due to a lack of large pools of well-curated training data. Although studies showed in early 2009 high sensitivity and specificity for diabetic retinopathy (DR) in this highly specified field, it took another 9 years for the first AI-based solution to be FDA-approved, and even as of 2023 there were only two additional FDA-approved solutions available. This timeline shows that even in this specialized field it takes considerable time and determination to develop a solution that meets the rigorous FDA requirements.

EMRs do not require FDA approval for clinical decision support, yet they play a crucial role in aiding clinical decision making. Some EMR vendors promote built-in LLM that summarize medical records. So, although EMR LLMs deal with data that is more diverse and heterogeneous than the images that trained diabetic retinopathy tools, progress seems to be faster. On one hand, technology has advanced exponentially over the last decade, which fuels the progress in AI. On the other hand, the fundamental problem remains of limited access to large pools of good, clean training data – and this is why it is important to consider a healthcare data integration engine and platform.

The role and benefits of a healthcare data platform

One of the technologies that can accelerate the adoption of AI for specialized LLMs in healthcare is HL7 FHIR. HL7 FHIR allows data aggregation to a much better degree than older standards such as HL7 v2 and CDA, thereby enhancing clinical interoperability. HL7 FHIR can be integrated seamlessly with existing legacy infrastructure through data conversion technology such as Infor FHIR Bridge and help to extract clinical data from their data silos with technologies such as bulk FHIR. Once data from various sources is extracted and transformed into a canonical format, other advanced technologies such as NLP (natural language processing) and LLMs can be applied to provide a semantically harmonized database well suited for AI applications.

An excellent use of a healthcare data platform is with electronic clinical quality metrics (eCQM). ECQMs were traditionally based on claims data and used for retrospective reporting. However, they can now leverage almost real-time clinical data to identify actionable gaps in care. By proactively closing gaps in care, patients will experience better outcomes and providers can achieve higher quality measure scores – critically important given the dire economic situation in the traditional fee-for-service business.

Another way to use the healthcare data platform is combining clinical and operational data in a single analytical environment to perform predictive analytics, for example for supply needs based on a combination of physician preferences and scheduled procedures.

Additional value of a FHIR-based healthcare data platform is the support of real-time, API (application programming interface) based workflows. MultiCare, a multi-hospital system in the Seattle area, implemented a FHIR accelerator program called DaVinci for prior authorization and achieved an increase from 3-5 prior authorization requests per hour to 10-12. This remarkable productivity gain showcases how new technology can significantly boost efficiency.

Figure 1 Source: Anna Taylor, MultiCare

Conclusion

A healthcare data platform that can integrate clinical and operational data as a basis for ML model training and efficient workflows is within reach. The benefits will be clinical (improved patient outcomes), operational (higher throughput, better supplies, etc.), and ultimately increased provider satisfaction. By leveraging AI and ML for better patient care and optimized workflows, providers can gain a competitive edge in the challenging healthcare market and deliver value to their patients and stakeholders.

Learn more about related technology solutions and advancements shaping the healthcare industry.

Watch our on-demand webinar on additional ways interoperability is improving healthcare.