Northeast Georgia Health System is a non-profit on a mission of improving the health of our community. Their mission would not succeed without the generosity of donors who contribute their time, talents, and treasures through the NGHS Foundation. Because of the donation and voluntary services, they can fund projects and programs that enhance the exceptional care already provided by NGHS.
The Customer partner was a lead generated from the HealthLake webinar held in Nov 2021. They were interested in a proof of concept with integrated natural language processing (NLP) and analytics to improve hospital operational efficiency and provide better patient care. They wanted to extract information such as medication, diagnosis, and past medical conditions from doctors’ clinical records and enrich patient records to examine access to health care and provide researchers with additional data points for advanced analysis to improve patient care.
The proposed design, called the Serverless Data Lake Design Pattern, transformed the unstructured, semi-structured, and structured data into a single structured format and stored it in S3 for prediction and processing via AWS Athena and AWS QuickSight and made data accessible for end-user consumption.
To implement the POC, sample FHIR data of patient records were ingested into the S3 raw zone. Data was transformed into a structured format using HealthLake (service) and Lambda (event notification). Structured data was stored in the refined zone from the refined zone using the Glue crawler we created. The metadata allowed Glue and Athena to view the refined data information. To retrieve specific information for analysis and prediction purposes, we further query the refined data using the Lambda function with Athena. The query information provided better insight into the data for building QuickSight dashboards and ML predictions in SageMaker.
The QuickSight Analysis identifies the Demographic shifts of hospital patients. This is essential for providing specialized care to different subgroups, as information helps to tailor care to patient needs.
Information such as a patient’s location may affect their ability to receive prompt treatment, and a patient's ethnicity and medical history may predispose them to the disease. For example, ACE inhibitors have been less effective than other classes of drugs such as calcium channel blockers and thiazide diuretics for treating hypertension in African Americans.
Dashboard analytics also help identify metrics and make vendor decisions to personalize patient interactions and conversations. This enables healthcare professionals to be culturally competent to put patients at ease, address patient-specific concerns, and make them feel respected. At the population level, this PII complaint information will help the County level at NCHS and more broadly the CDC at the Central level to assess and develop practitioners, predict trends and identify outbreaks.
A holistic view of patient clinical data via AWS Health Lake helps in treatment coordination, especially multidisciplinary treatment coordination that requires extensive health information management.
Information such as allergies, their records, and appropriate dates can help identify seasonal trends, outbreaks, etc. Thus, early detection and appropriate intervention can help to reduce the negative impact of allergies on quality of life.
This information would also help the regulatory body to closely evaluate the identified sources and types of allergens and plan for mitigating them by region. This will also be helpful for identifying allergy risks within their zip code, implementing care plans, and stocking anti-allergy medications which are more assertive in that area.
Timely immunization information data can help reduce logistical challenges in the supply chain of vaccines and medicines available to the public and help better manage immunization programs.
Hospitals would be better equipped to identify trends and numbers of vaccines and drugs administered over time and make decisions.
It will help immunization registries and regulatory bodies track individual immunization records, help health workers identify defaulters, and also track vaccine inventory and cold chain conditions, to ensure that vaccines are being kept in good condition and made available when they are needed.
Financial interactions illustrate the inpatient treatment amount and the health insurer who provided it. It helps identify the number of claims per medical condition, which allows organizations to determine areas of clinical procedures that require better insurance packages.
This also helps identify self-pay and uninsured patients who need charitable assistance through financial assistance and medical cost sharing.
A machine learning model using AWS Sagemaker to predict the insurance amount based on the extracted data such as medical condition, gender, age, etc. A regression model is used to identify estimates that can be used to create actuarial tables representing annual premium prices based on expected treatment costs.
This also helps identify key variables that affect insurance coverage.