Meet Sara: The Dawn of AI in Healthcare in the Middle East

In recent years, the GCC region has experienced a significant transformation in healthcare, with Artificial Intelligence (AI) swiftly becoming a cornerstone of medical progress. The region's dedicated focus on merging cutting-edge technology with healthcare has led to promising innovative digital solutions—ones that can address challenges and improve patient care for a healthier future.

With the launch of Chat GPT in 2022, AI applied to healthcare has undergone remarkable progress, transitioning from basic predictive models to intricate generative AI systems. These systems can potentially reach billions of people by incorporating natural language in both input and output processes. Built upon probability algorithms, generative AI models excel in forecasting the succeeding sequence of words by leveraging an extensive corpus of training text

Addressing the Challenges of Generative AI in Healthcare

Deploying universal generative AI in healthcare introduces distinct hurdles that require resolution for the industry to fully harness the potential of this innovative tool. Even though the outcome of Generative AI is truly compelling, the effectiveness of its intelligence is not without some challenges The key hurdles to generative AI in healthcare are outlined below.

• The risk of AI “hallucinations” can occur when a large language model (LLM) detects patterns or objects not apparent to humans, potentially resulting in misleading outputs IBM highlights instances of such "hallucinations,"
• The caliber of training data can play a pivotal role in healthcare. The nuanced and ever-evolving healthcare domain requires sophisticated and timely information. The presence of high-quality healthcare data still remains one of the major challenges in AI. The extensive data AI models rely on may lack clinical validation, potentially introducing inaccuracies in medical reasoning if not thoroughly verified regularly
• Navigation of the intricacies of healthcare concepts poses a challenge for generative AI models. While these models excel at managing medical terminology, they struggle to interpret complex healthcare ideas. For instance, querying OpenAI or Google's Gen AI models about diabetes yields satisfactory responses. However, when provided with patient data in formats like C-CDA, HL7, or FHIR and asked to determine the presence of diabetes, these AI models falter significantly.
• The impractical demand for precise query formulation becomes evident when engaging with freely available proprietary models such as OpenAI or Google. It becomes apparent that the phrasing of questions holds equal or greater significance than the model's training data or quality. However, expecting millions of healthcare workers and patients to acquire query-coding skills is unrealistic. AI models need to be resilient and designed for universal usability, allowing individuals of varying expertise levels to effortlessly engage with the technology.
• The lack of transparency in healthcare has been persistent after decades of feedback data submitted to predictive models without the presentation of the underlying rationale. Clinicians remain hesitant to adopt or trust model outputs because of the lack of transparency in the provided answers. Employing a "black box" approach does not enhance confidence. To instill trust among clinicians, a predictive model forecasting a discharged patient's 30-day readmission, for instance, must articulate the reasons or factors influencing its results. Generative AI models, similarly, risk being perceived as more opaque compared to predictive models unless case details and citations are presented to clinicians.

Introducing Sara for Healthcare: Revolutionizing Digital Health and Bringing Back the Joy of Care with AI

In light of these challenges, Innovaccer Middle East developed a healthcare-tailored AI solution named "Sara" to address some of the most pressing operational needs in healthcare. Sara for Healthcare was unveiled at a Product Showcase Session at Arab Health 2024 on January 31st

Sara stands as a distinctive suite of healthcare solutions driven by AI, poised to revolutionize patient care, elevate financial performance, enhance population health, and bolster consumer engagement. Serving as a driving force, it empowers executives, clinicians, care coordinators, and contact center representatives to deliver crucial data precisely when needed, fostering improved health outcomes. Moreover, Sara is meticulously designed to meet the specific safety, precision, security, compliance, and scalability requirements unique to healthcare IT Let's explore how Sara tackles these challenges:

• Expertise in Healthcare Data Proficiency: Sara is an AI-driven suite of solutions on the Innovaccer data platform. Having successfully integrated records for more than 60 million individuals in the United States, Innovaccer's platform demonstrates a keen understanding of diverse healthcare data. Proficient in unifying, cleansing, and interpreting healthcare data, Innovaccer transcends conventional settings (like patient-provider conversations in exam rooms or call center interactions) to incorporate additional rich data sources, providing valuable inputs for generative AI models.
• Extensive Training on Clinical Terms and Healthcare Concepts and Terminologies: Sara possesses proficiency in more than 10 million clinical terms and a diverse range of administrative and reimbursement concepts, fostering a profound comprehension of the complexities within healthcare. The following delineates the extensive training Sara undergoes in the healthcare domain:
  • Clinical Terminology:  Sara is well-versed in over 10 million clinical terms encompassing diseases, medications, labs, vitals, problems, complaints, allergies, immunizations, and more. This knowledge spans across 50+ coding systems such as ICD, CPT, SNOMED, RxNorm, NDC, CDC, CMS, and others.
  • Administrative Concepts: Sara incorporates information on over 10 million healthcare providers, 200,000 healthcare facilities, and 1000 health plans. It comprehensively understands their identities, databases, and operational processes.
  • Reimbursement Concepts: Sara undergoes training on over 50 episodes of care, 300 cost centers, 1000 revenue center codes, 1000 DRGs, and 30,000 CPT codes. Additionally, it is familiar with more than 1000 reimbursement formulas like PMPM or IP/1000, comprehending the measurement of care costs and reimbursement processes in fee-for-service and value-based care models.
  • Quality of Care Concepts: With knowledge encompassing over 3,000 value-sets, including 200 ICD-10 codes describing diabetes and 400 NDC codes constituting diabetic medications, Sara comprehends quality of care measurements and their potential absence, along with awareness of over 500 quality measures.
  • Risk and Coding: Sara demonstrates an understanding of more than 30,000 ICD-10 codes and their interaction with 200 financial risk categories (HCCs) or 100 clinical risk categories. It also recognizes suspect conditions based on evidence found in clinical or claims data.
• Continuous Learning from User Feedback: In the context of healthcare in the Middle East, Sara undergoes continuous enhancement through user interactions, refining its capabilities with each user edit. If a SOAP note is generated by Sara and subsequently edited by a clinician before being incorporated into their EMR, Sara acknowledges this feedback, enhancing its intelligence over time.
• Enhanced Transparency: Diverging from other generative AI models, Sara ensures clear explanations for its outputs, fostering reliability and trustworthiness in clinical settings. For instance, if Sara generates a SOAP note based on a patient-provider conversation, it not only produces the note but also a complete transcript of the conversation, incorporating clinically relevant facts obtained from the dialogue and prior clinical data. Similarly, when providing an aggregated data report based on an English query, Sara shares insights into how it interpreted the query, including operands, logical operators, and filters, using an SQL query.
• Responsibility and Accountability: Ensuring that Sara avoids delivering misleading outputs or engaging in hallucinations is paramount. Consequently, it becomes crucial for Sara to ascertain whether it possesses all the necessary information to provide an accurate response. If not, it should confidently state either "no" or "I don't know." For instance, if a SOAP note is generated based on a patient-provider conversation in an exam room where the clinician omits discussion on various lab results or objective assessments, the "Objective" section of the SOAP note will remain empty. Similarly, if tasked with generating insights from a user query and encounters an unfamiliar clinical term, Sara will transparently communicate its inability to comprehend that particular term.

The Magic of Sara for Various Healthcare Practices

From the viewpoint of healthcare providers, Sara augments trust by abstaining from hallucinations, relying on factual data, and instilling confidence in clinicians. It also takes into consideration the background data utilization by integrating a patient's comprehensive clinical history, facilitating more nuanced and informed diagnoses. Sara also supports improved care and reimbursement assistance by enabling clinicians to navigate through the intricacies of patient care and reimbursement processes, offering insights into quality and coding gaps, among other crucial information.

From the standpoint of healthcare analysts, Sara enables precision in data queries. In comparison to alternative AI models such as Open AI GPT-4, Sara demonstrated superior accuracy (83%) in comprehending and executing intricate healthcare queries. Additionally, with a comprehensive grasp of healthcare intricacies, it excels in comprehending healthcare-specific terminology and formulas. Innovaccer has employed a dataset comprising over 1,000 English queries, predominantly healthcare-related input submitted by executives and analysts, alongside the corresponding desired SQL query output. These queries spanned diverse concepts such as cost, quality, risk, membership, and network. In comparison to proprietary all-purpose Gen AI models like Open AI GPT-4, Innovaccer integrated our data model and healthcare terminology with their models to establish an equitable playing field. The ensuing accuracy results represent the percentage of queries in which the desired SQL query output was successfully attained.

AI for healthcare has arrived, and the time to embrace it is now!

In summary, over the past 25 years, technological innovations have accelerated in the healthcare sector, and now AI has arrived The adoption of AI will play a crucial role in shaping the future of healthcare

For AI to truly enhance productivity, it must be tailored to healthcare needs, be user-friendly, and be a reliable tool. Without these crucial elements, any solution is unlikely to be embraced with enthusiasm. Thoughtful consideration is paramount when selecting models and tools to integrate Innovaccer is confident in its unparalleled capabilities, which is why we are excited to introduce Sara to the healthcare sector. , 

For a closer look at  Sara and how it can enable you to revolutionize operational, administrative, and clinical aspects of healthcare, book a personalized demo now!