PILOT - Unstructured Healthcare Data

Equitus.us's Knowledge Graph Neural Network (KGNN) helps the healthcare industry reduce Extract, Transform, and Load (ETL) costs by automating the data ingestion and contextualization process. Unlike traditional methods that require complex, manual pipelines to move and standardize data, KGNN is designed to ingest data directly from its source and automatically create a unified knowledge graph. This approach eliminates the need for expensive data engineering teams and simplifies a major bottleneck in healthcare analytics.

---

How KGNN Reduces Healthcare ETL Costs 

The healthcare industry is notorious for data silos. Patient records, medical images, lab results, insurance claims, and clinical trial data often exist in disparate, fragmented systems. Traditional ETL is ill-equipped to handle this volume and variety, leading to significant costs and delays. Here's how Equitus KGNN provides a solution:

• Automated, Schema-less Ingestion: Instead of building a new data pipeline for every data source, KGNN can ingest, It uses a semantic layer to understand the meaning of data from sources like Electronic Health Records (EHRs), medical IoT devices, and clinical notes. This "auto-ETL" functionality drastically reduces the labor and time required for data preparation.

• Eliminating Manual Contextualization: A major cost in healthcare data projects is the effort to manually link patient data to contextual information. For example, a doctor's note might mention a symptom, but it takes human effort to link that symptom to a specific diagnosis, a prescribed drug, and the patient's demographic information. KGNN automates this process by identifying entities and relationships, such as connecting a disease to its associated symptoms, treatments, and genetic markers. This contextualization is essential for downstream AI applications and is performed automatically, saving significant time and money.

• Decentralized Data Processing: Equitus KGNN runs natively on IBM Power servers, which have a built-in Matrix Math Accelerator (MMA).⁸ This allows healthcare organizations to perform complex AI computations and data integration at the edge, close to the data source.⁹ By doing so, they avoid the high costs and security risks associated with sending massive amounts of sensitive data to the cloud for processing. This on-premise capability reduces network and cloud computing expenses.

Key Benefits for the Healthcare Industry 

The reduction in ETL costs is a fundamental enabler for several high-value healthcare applications:

• Accelerated Research and Drug Discovery: By unifying data from disparate sources, KGNN allows researchers to quickly analyze molecular, genetic, and clinical trial data. This can accelerate the identification of new drug candidates and potential drug interactions, leading to faster research breakthroughs.

• Improved Clinical Decision Support: A unified knowledge graph that links patient records, lab results, and medical literature enables AI systems to provide more accurate and context-aware insights to clinicians. This can lead to better diagnoses, more personalized treatment plans, and reduced medical errors.

• Enhanced Fraud Detection: KGNN can model complex relationships between patients, providers, and billing codes. This makes it easier to spot fraudulent patterns that would be missed by traditional analytics, such as suspicious billing for services, reducing millions in fraudulent payments.

___________________________________________________________________________

 ACADIA, a nationwide hospital chain, 250 hospitals/ 250,000 employees,  encounters a vast amount of unstructured data every day, which is information that does not have a predefined format or is not organized in a traditional database. This data is critical for providing patient care, but it requires specialized tools like natural language processing (NLP) and machine learning to analyze and extract insights.

Here is a list of all the unstructured data a nationwide hospital chain might encounter, categorized for clarity:  Efficiently incorporating these disparate and uncollected information can help a healthcare organization better control costs, risks and compliance. 

Clinical & Patient Care Data

• Clinical Notes: This is a huge category and includes:

  • Physician's notes and progress notes

  • Nurses' notes

  • Therapists' and other specialists' notes

  • Discharge summaries

  • Consultation reports

  • Operative reports

• Medical Imaging: The images themselves are unstructured, though they may have some structured metadata attached. This includes:

  • X-rays

  • MRIs

  • CT scans

  • Ultrasounds

  • PET scans

• Pathology and Lab Reports: While lab test results themselves are often structured (e.g., a specific blood count value), the accompanying narrative and descriptive reports are unstructured.

• Audio and Video Recordings:

  • Transcriptions of doctor-patient interactions or consultations

  • Dictations by healthcare professionals

  • Recordings from speech therapy or other patient sessions

  • Video recordings of surgical procedures or physical therapy

• Patient Correspondence and Narratives:

  • Emails and messages between patients and providers

  • Personal accounts from patients about their symptoms or health history

  • Patient-reported data collected via surveys, free-text fields in patient portals, or communication logs.

• Physiological Monitoring Data: Data streams from wearable devices, IoT devices, and continuous patient monitors. This can be complex, high-volume data that doesn't fit a simple tabular format.

• Genomic Data: The vast amount of data from a patient's genome sequencing.

Administrative & Operational Data

• Correspondence and Communication:

  • Emails and memos between staff, departments, and external organizations

  • Internal communications, such as instant messages or chat logs

• Documents and Files:

  • Scanned copies of patient records, contracts, or administrative forms

  • PDF documents, such as informed consents, legal documents, or research papers

  • PowerPoint presentations

• Administrative Notes and Memos: Unstructured notes from administrative staff, meeting minutes, and internal reports.

Public & External Data

• Patient Feedback:

  • Online reviews of the hospital chain and individual facilities (e.g., on Google, Yelp, or other review sites)

  • Comments on social media about the hospital's services or reputation

  • Free-text feedback from patient satisfaction surveys

• Research and Literature:

  • Medical research papers and clinical trial reports

  • Case studies and publications

• Social Media: Public posts and comments about the hospital, its staff, or the patient experience.