How semantic interoperability can help optimize healthcare data

Data quality and usability are still challenges, and advanced interoperability solutions can help organizations derive value from their information stores.

Healthcare organizations collect and process vast amounts of information, ranging from structured data generated by machines like laboratory tests and vital signs to unstructured data, often provided by clinicians in the form of narrative clinical notes, patient interactions and self-reported patient outcomes, expressed in natural language.

This healthcare data contains valuable insights into a patient's health status, medical history and treatments if the data makes sense to the person viewing it. Often, though, it lacks essential data quality (completeness, accuracy, structure, conformity and consistency). Furthermore, when data is aggregated from multiple sources, it frequently contains duplications. These duplications are not always obvious literal copies and can be references to the same information from different standpoints and may even contain contradictions. Such redundancies add a layer of complexity to data aggregation, presenting significant obstacles to leveraging the information effectively.

To use healthcare data effectively, the industry must address these challenges. Semantic interoperability — the ability of healthcare systems to exchange data by way of mapping different terms to shared meanings — is the key to building a solid foundation for usable healthcare data. This can ultimately lead to improved quality of care and better patient outcomes.

Common clinical data language challenges

Because clinical decisions can mean life or death, healthcare applications should not guess the meaning of a clinical note or concept, so speaking the same clinical data language is critical. However, a significant hurdle arises because overlapping and diverse health specialties, code systems and standards create a multitude of different languages that coexist. A lot of structure must be in place before machines can correctly interpret healthcare data.

Mapping one-to-one equivalences between different terminologies is often impractical. Instead, it requires a more specific approach, which may involve adding terminology constructs or tools. An example is value sets that combine concepts from various terminologies in groups that are logically connected (often in the context of a specific use case), for instance, concepts related to social determinants of health.

As healthcare organizations connect to more information sources, the need for efficient data standardization and seamless integration has never been more critical. While standards like Fast Healthcare Interoperability Resources (FHIR), RESTful APIs, and terminologies like SNOMED and LOINC are advancing interoperability, the industry needs more. They serve as building blocks, providing standardized ways to encode and share healthcare data and are a step in the right direction, with limitations.

For example, the need for more guidance on code usage and terminology overlap in standards like the USCDI creates obstacles. Terminologies like SNOMED may express the same clinical meaning in different ways, leading to disparities. This poses challenges for software applications that cannot recognize the many options for expressing the same notion and creating mappings and value sets may require expertise in clinical terminologies.

Furthermore, existing systems often rely on techniques like lookup tables for data translation, which can be problematic. Lookup tables have limitations in handling the dynamic nature of healthcare terminologies. They are static and often fail to address the constant evolution of these terminologies. Translation using lookup tables can be intrusive, replacing one code with another and potentially losing the original information. This approach may be suitable for temporal data but is less effective for carrying information over time. Another challenge of lookup tables is that they do not take into account the context in which concepts are used. Substitution typically would depend on the use case, purpose and profile that sets such context.

Semantic interoperability: Bridging the gap

To address this, healthcare organizations must adopt more sophisticated tools to keep up with the evolving nature of healthcare terminologies.

Semantic interoperability, powered by AI and machine learning (ML), solves these challenges. It serves as the bridge between data producers (clinicians and other healthcare professionals) and data consumers (healthcare systems, applications and decision support tools). It focuses on the meaning of the conveyed information and its context. It establishes a common framework for healthcare data interpretation, ensuring that information is not just transmitted but also comprehended accurately on both ends.

Rather than focusing solely on standardization, which may not always be feasible or desirable because of the diversity of healthcare specialties and use cases, semantic interoperability emphasizes the constant transformation and adaptation of conversations that occur within systems and between humans. It is about managing the multitude of clinical terminologies, facilitating data exchange across various specialties and domains and enabling data to be understood in the proper context.

Terms can be mapped and updated in real-time, ensuring that translations remain accurate and augmented with current content, like commonly used provider-friendly terms. Additionally, value sets, which maintain the integrity of important concept groupings over time, can be created with the help of advanced semantic interoperability tools. This approach enables various ways of processing incoming data and working with it, going beyond the limitations of static lookup tables.

The value of semantic interoperability

Semantic interoperability makes healthcare data meaningful to software applications and ensures it is usable by both downstream systems and humans. It facilitates such interactions in the context of diverse use cases, enabling the exchange of data that conveys meaning and concepts relevant to these use cases.

The technology simultaneously promotes standardization and interoperability and streamlines data exchange between healthcare providers, laboratories, diagnostic services, payers, patients, public health systems and other entities. Key benefits can be measured.

Usability. “Interoperability” in some cases is only between two specific systems. If the source system is relatively unique — a lab information system, for example — it may have its data encoded in a particular manner. In this case, the connection only works between that system and the upstream provider. This can create difficulties when working with other providers, as the process must be repeated with each unique system.

By using semantic interoperability and mapping data to a national standard like LOINC, it’s easier to connect to multiple upstream providers. Ultimately, it helps create a seamless network of data that becomes more usable across the healthcare ecosystem.

Efficiency. The amount of data generated per patient will soon overwhelm clinicians. An average hospital now produces around 50 petabytes of data each year. While healthcare professionals have worked with incomplete information for centuries and became accustomed to it, they are now drowning in data (from IoT, remote patient monitoring, genomics and other sources).

Making sense of all the information, determining what is relevant to a specific problem and establishing meaningful connections to improve clinical outcomes are new challenges. Semantic interoperability increases the productivity and efficiency of healthcare professionals by making data more actionable and ready for use by technologies such as AI and ML algorithms.

Computability. The healthcare landscape is rapidly evolving, with a growing number of clinical decision support systems and AI tools being integrated into the EHRs that clinicians use or plugged into the clinical workflows behind the scenes. For these innovative software tools to function properly, data must be computable and machine-interpretable. Semantic interoperability ensures that data is served to these digital tools in the form that they expect, enabling their results to become usable across a network of multiple EHRs and information systems.

Real-world use cases

Clinical decision support tools. Clinical decision support tools are bound to a rule system based on certain conditions to guide the user to a possible decision path. As a result, they rely on high-quality data to navigate complex medical conditions and suggest treatment options based on clear and codified healthcare information. As more clinicians rely on these tools more often, the challenge of effectively exchanging data becomes more pronounced.

Semantic interoperability ensures clinical decision support tools have structured inputs with a high degree of data quality so relevant conditions are recognized and understood, enabling accurate decision support. Healthcare organizations can seamlessly integrate these new technologies and tools into their EHR systems, avoiding lengthy and complex data translation processes. The system automatically understands the data flow and starts reasoning, making the tool onboarding process more efficient.

Building AI models with health data. Healthcare organizations are increasingly turning to AI for clinical data analysis and precision medicine. However, AI models are only as good as the data they rely on. AI model creation requires structured, clean, reliable, and bias-free training data.

The problem is that when the source data is very disorganized, ​data scientists need to spend most of their time preparing it for AI model training​. Clinical data is messy, EHR records are not standardized, and nobody uses the same codes. Before data scientists can build AI algorithms, they must create data pipelines that enable data enrichment and augmentation, map text strings to standard codes, group and filter codes through value sets for efficient feature extraction and prepare the results for model training. This is very time-consuming, and, with the amount of training data AI models need, it becomes impractical to do manually.

An example is creating an AI-based precision medicine platform with predictive analytics capabilities. This platform ingests real-world diverse healthcare data from countless unique data sets, including structured and unstructured data with various codes and formats from medical and pharmacy claims, EHRs, medical devices, genetic sequencing, molecular data and more. Some datasets include ICD, SNOMED, CPT and LOINC codes, while other data ​are​ semi-structured or unstructured text. Semantic interoperability is used to enrich and properly classify the data collected so that it feeds the AI development pipelines and dramatically simplifies the process by automating the data preparation.

Patient summaries. If a diagnostic service provider such as a radiologist is performing an interpretation of a study and wishes to see additional information, it is far more efficient to have quick access to a summary of only the findings that are relevant to the current study being interpreted than to comb through the complete patient record in the EHR. In this case, the data needs to be filtered only to emphasize what is relevant and important to the study. However, relevance is highly contextual and dynamic and, therefore, difficult to model without semantic interoperability.

Practical steps for adopting semantic interoperability

Semantic interoperability is not just a technical solution; it's a fundamental shift in how healthcare data is managed and leveraged. It promises a future where quality care is the norm, and clinicians can trust external data included in longitudinal patient records and cohorts for clinical research.

Healthcare organizations should consider the following practical steps when preparing to adopt semantic interoperability:

Assess standard terminologies. Evaluate the terminologies currently in use within the organization. Understand the extent to which proprietary data is employed in place of standard codes. Develop a strategy for the effective use of codes across your organization.

Evaluate data usability. Assess how easily other healthcare facilities, systems and applications can use your data. Identify the most valuable use cases and any barriers to data exchange in the context of these use cases.

Identify tools for improving data usability. Develop a practical approach to overcome the barriers to extracting value from data. Become familiar with the existing tools that help solve challenges – from reducing the impact of duplicates to automation for improving staff productivity to applying technologies such as AI and machine learning.

Look at the full interoperability picture. Understand the role semantic interoperability plays in a complete data interoperability picture, centered around use cases and existing goals of extracting value from data and improving outcomes, efficiency and productivity. Develop an interoperability strategy that aligns with business goals and develop partnerships that aid in execution.

Evgueni Loukipoudis is vice president of research and emerging technologies for Rhapsody.

More for you

Loading data for hdm_tax_topic #better-outcomes...