"You Don't Know What You Don't Know" - so does AI.
Many companies are trying to develop and deliver Healthcare AI (Artificial Intelligence). AI is going to change how healthcare will be delivered by making it personalized, detecting and preventing and by continuously evolving. Data plays a crucial role in the development and effectiveness of healthcare AI systems. To make it real, AI algorithms require large amounts of diverse and high-quality data to train and learn patterns and gain the required accuracy.
You Don't Know What You Don't Know" - so does AI, is also true for AI models. The less the model knows, the less effective it is; feed it wrong data, the less its accuracy. Scientists are busy building various healthcare AI models and healthcare companies have years of patient data sitting somewhere on storage. The data available is in various formats, captured in different timelines and collected from various sources. It is like navigating multiverse.
Here are some key challenges HiPaaS is trying to solve:
Scale and Performance: AI needs data in petabytes and current solutions to load data are not scalable and take lot of time. HiPaaS is highly scalable and we have successfully converted large sets of healthcare data like medical records, lab results, encounters, and patient demographics of Terabytes size each. HiPaaS converter is auto scalable using AWS serverless fargate architecture. Performance is key, we are continuously optimizing to convert data faster (e.g. 5-10 GB in few minutes). Recently, we were able to transforms 10+TB or observations data in few days.
Sources and Formats: There are various formats, standards , types and sources of healthcare data. Mainframe systems, public healthcare domain data and old EHR systems data is often available in custom csv files. EHR's like EPIC, Cerner and others provide CCDA, FHIR and/or HL7 data feeds. Payers and Insurance companies often have EDI X12 data related to claims and authorizations. To further complicate things, the data sets can be a point in time or over series of timeframes. HiPaaS has inbuilt mappers for standard formats and a mapping tool for custom data formats.
Accuracy & Quality: Quality data is vital for achieving accurate and reliable outcomes in healthcare AI applications. Robust and well-curated datasets help in reducing errors, false positives/negatives, and bias within AI models. At HiPaaS, we have developed required validations, lookups, crosswalks which will check, clean the data before converting. Codes like ICD10, ICD9, Snomed, Ionic are mapped and can be configured based on source data. Furthermore, HiPaaS also handle data entry mistakes (e.g. decimal "." can be entered in 5 different ways).
Errors and Work-queues: The data which errors out is equally important and cannot be ignored. It might have some valuable information that might train the model. HiPaaS FHIR Converter tool puts all errors in various work queues for your staff to review, fix and then convert. The error queues contain missing dependencies (like patient not found or Medication code missing), actual errors (like wrong codes or values), duplicates, and missing mandatory fields required to medical resource.
Compliance and Security: It's important to note that while data is valuable for healthcare AI, ensuring privacy, security, ethical use and compliance takes precedence. HiPaaS is HIPAA compliant, secure and SOC2 certified. We continuously conduct security testing of our software. Our team is trained and certified to follow the law of the land to handle PHI and healthcare data. HiPaaS FHIR Converter can also be installed in customers cloud account or data center.
HiPaaS partnered with AWS to deliver "FHIR Data Converter for Amazon Healthlake" to help companies unlocking healthcare data and feed it to the models. The data from various formats is converted to common FHIR® (Fast Healthcare Interoperability Resources) format and becomes interoperable. Once the data is loaded, it can then be used by models to train itself.
More about HiPaaS at www.hipaas.com. HiPaaS is Trademark of HiPaaS Inc.