The Apixio Platform
Every year, 1.2 billion clinical documents are generated in the U.S., but most of this data goes unused in the provision of healthcare. Without the right technology to transform PDF and image documents into machine-readable text, and well-trained models to mine these documents for relevant patient and clinical insights, organizations aren’t making the most of their data.
At Apixio, we recognized an opportunity for artificial intelligence to unlock unstructured text and help plans and providers use the data they already have to improve risk capture, care quality, and clinical documentation. The Apixio Platform is the foundational technology suite that allows us to surface targeted information from healthcare documents to inform operations and care delivery.
Step 1: Data Acquisition by InfoStream
We use a secure extraction process to pull and encrypt PDF, image, and EHR documents such as patient charts, medical and pharmacy claims, eligibility data, and more. These documents are then securely transmitted to our Data Loader for processing and analysis.
Step 2: Data Processing
Our ETL workflow loads, validates, and processes documents so they can be run through our proprietary machine learning models. During this stage, we use our optical character recognition (OCR) pipeline to translate the data in image files into machine-readable text.
Step 3: Data Analysis
Processed data is stored in individual Apixio Patient Object Models (APOMs), which include patient diseases, medications, procedures, biometric values, and derived information from prior analyses. We then apply proprietary data classifiers and predictive models to generate insights.
Step 4: Results Sorting
Insights from our machine learning models are sorted and sent through configurable workflows to our applications. Results are tuned per client, per project to ensure maximum relevancy.
Step 5: Application Review
Results are served up for review in our risk adjustment, quality, and prospective solutions. Feedback from expert users is fed back into our machine learning algorithms to continually refine our approach and improve results.
User annotation and data labeling (via automated and manual methods) are used to continuously update our models. We employ supervised and unsupervised techniques to train healthcare machine learning models. There are mechanisms built into our proprietary science infrastructure to deal with noisy annotations and labeling errors. We specially configure our workflow applications to reduce errors and improve the accuracy of expert annotation, which is essential for crafting and maintaining high performing algorithms.
By using Apixio, we’re improving our auditing bandwidth and enhancing the ability of our coders to focus on other chart audits and other projects that we couldn’t do before.
In the past we relied on manual review of our charts. This meant lost manpower and physician time in tedious processes.
Implementing HCC Identifier has allowed us to mine EHR and scanned chart data for valid, risk adjusting conditions with improved transparency and efficiency.
The coding team was wonderful to work with, and incorporated our specific coding guidelines when working on our project. When they noticed a trend in what we were rejecting, they shared a Coding Clinic we were unaware of for clarification on how to code the scenarios.
Apixio & Magna Health Plan
Fast Facts 40,633 lives reviewed 2,755 HCC deletes found 95.3% agreement with…View Case Study ⟶
How a Veteran Coder Used Apixio’s HCC Profiler to Eliminate Data Entry Error and Double Productivity
Fun Stats Increased productivity from 3 charts per hour to 10 charts…View Case Study ⟶