Introducing Anomaly's Engineering Blog
Welcome to the inaugural post of the Anomaly engineering blog. Over $300B is lost every year to avoidable denials and payment errors, impacting the affordability of healthcare.
Anomaly is building novel software and working with some of the largest healthcare companies to enable fast, accurate healthcare payments—making the whole system more efficient and affordable.
Why is healthcare billing so complicated?
Healthcare is inherently complex and constantly changing; the billing system, in turn, has tens of thousands of codes and evolving rules. Healthcare providers and insurance companies struggle to keep up.
Every time you visit the doctor, or any healthcare provider, a claim is sent from the doctor's office to your insurance company. Billions of claims are sent between providers and insurance companies every year. A claim is basically an invoice with a list of services performed. Claims used to be filled out on a paper form that looks like this:
But today the vast majority of claims are transmitted electronically using a standard format called X12 EDI 837. There is a standardized procedure code for every medical service that you receive, from the baseline new patient consultation (99201) to the COVID PCR test (87635). Every possible diagnosis from the "common cold" (J00) to "bitten by pig" (W55.41XA) also has a standardized code.
While these coding systems allow a doctor's office to express very detailed information, there are tens of thousands of them, and there's often a lot of ambiguity and decision making that goes into choosing the correct options. Furthermore, every health insurance company has hundreds of pages of unique guidelines on when services are allowed to be billed based on the diagnosis and other patient facts, and these rules constantly change. This complexity inevitably leads to myriad errors on a massive scale.
What does this mean for Anomaly?
Everyone wants the right amount to get paid the first time. But that isn’t what happens in healthcare.
Over 15% of healthcare claims are reprocessed. This can be for a variety of reasons. The claim may be initially denied by the insurance company for a coding error, only to eventually get paid after the provider corrects and resubmits the claim. Or perhaps the denied claim is paid once the provider successfully challenges, or appeals, the insurer’s denial decision. These reprocessing scenarios happen to millions of claims every day.
Whatever the reason, reprocessed claims drive billions of dollars of cost and cause significant frustration for providers and payers. Most importantly, they impact patients, who often get caught in the middle of payer-provider disputes. They may even have the bill pushed onto them.
Anomaly is working with the country's largest healthcare companies to build the intelligent API layer streamlining healthcare payments.We are building predictive capabilities that analyze billions of transactions and integrate at key points in the claims lifecycle to predict and prevent payment issues at scale. Smart Response predicts denials and payment amounts with high precision, recall, and explainability—so providers and patients can avoid expensive denials before they occur. Instant Pay (coming soon) will further streamline payments by enabling providers to immediately get paid at claim submission
Engineering challenges in building a claim prediction engine
We'll have future posts that dive into these areas in more detail, but here's a high level overview.
- Data Engineering
- Billions of claim transactions
- Complex temporal relationships and fuzzy matches across transactions
- Many data sources and ingestion points, both batch and streaming
- Governance of sensitive healthcare data
- Machine Learning
- Very high dimensionality data
- Highly imbalanced data-- we're trying to predict relatively rare events
- Complex hierarchical relationships within the features
- Temporal and spatial dimensions
- ML ops from offline training to online serving of dozens of models
- API Infrastructure
- Thousands of predictions per second
- Tight latency and downtime SLAs
Building Anomaly
We've built up a small but mighty engineering team and laid the groundwork for our predictive products. We're excited to expand the team with additional engineers and data scientists with expertise and ambition in any of the areas above. Does that sound interesting to you? Join us!