Life insurance coverage corporations depend on correct medical underwriting to find out coverage pricing and danger. These calculations come from specialised underwriting corporations that analyze sufferers’ medical information intimately. As healthcare digitization has surged from 10% in 2010 to 96% in 2023, these corporations now face overwhelming volumes of complicated medical paperwork.
One main life settlement underwriter discovered their course of breaking beneath new pressures. Their two-part workflow — an inner group labeled paperwork earlier than docs reviewed them to calculate life expectancy — was struggling to maintain up as their enterprise grew and healthcare documentation grew to become more and more complicated. Medical specialists have been spending extra time sorting by paperwork as an alternative of analyzing medical histories, making a rising backlog and rising prices.
This bottleneck threatened their aggressive place in an business projected to develop at twice its historic charge. With correct underwriting straight impacting coverage pricing, even small errors might result in hundreds of thousands in losses. Now, because the medical business concurrently faces worsening workforce shortages, they wanted an answer that would rework their doc processing whereas sustaining the precision their enterprise will depend on.Â
It is a story of how they did it.
When medical report volumes get out of hand
Processing 200+ affected person case recordsdata weekly may sound manageable. Nevertheless, every case contained a affected person’s whole medical historical past — from physician visits and lab outcomes to hospital stays and specialist consultations. These recordsdata ranged from 400 to 10,000 pages per affected person. However quantity wasn’t the one problem for the medical underwriting supplier.
Their enterprise confronted mounting strain from a number of instructions. Rising business volumes meant they’d extra instances to course of. On the flip aspect, the healthcare business staffing shortages meant they needed to pay docs and different medical specialists high {dollars}. Their current guide workflow merely could not scale to fulfill these calls for. It was made worse by the truth that they needed to keep near-perfect doc classification accuracy for dependable life expectancy calculations.
The enterprise impression was evident:
- Slower processing occasions meant delayed underwriting choices
- Inaccurate life expectancy calculations resulted in hundreds of thousands in mispriced insurance policies
- Probably dropping enterprise to extra agile opponents
- Increased processing prices straight affected profitability
- Rising prices as docs frolicked on paperwork as an alternative of study
Their medical specialists’ time was their most beneficial useful resource. And but, regardless of the 2-step workflow, the sheer quantity of paperwork pressured these extremely educated professionals to behave as costly doc sorters fairly than making use of their experience to danger evaluation.Â
The maths was easy: each hour docs spent organizing papers as an alternative of analyzing medical situations price the corporate considerably. This not solely elevated prices but in addition restricted the variety of instances they may deal with, straight constraining income progress.
What makes healthcare doc processing difficult
Let’s break down their workflow to know why their medical report processing workflow was notably difficult. It started with doc classification — sorting tons of to hundreds of pages into classes like lab studies, ECG studies, and chart notes. This vital first step was carried out by their six-member group.
Every member might course of ~400 digital pages per hour. Which means, a single case file of two,000 pages would take over 5 hours to finish. Additionally, the velocity tends to range closely based mostly on the complexity of the paperwork and the potential of the worker.
The method was labor-intensive and time-consuming. With digital medical information coming from over 230 completely different methods, every with its personal codecs and constructions, the group needed to cope with numerous variation. It additionally made automation by conventional template-based knowledge extraction practically inconceivable.
The complexity stemmed from how medical data is structured:
- Crucial particulars are unfold throughout a number of pages
- Info wants chronological ordering
- Context from earlier pages is usually required
- Dates are generally lacking or implied
- Duplicate pages with slight variations
- Every healthcare supplier makes use of completely different documentation strategies
After classification, the group would manually determine pages containing data related to life expectancy calculation and discard irrelevant ones. This meant their workers wanted to have an understanding of medical terminology and the importance of varied check outcomes and diagnoses. There was little or no margin for error as a result of even the slightest errors or omissions might result in incorrect calculations downstream.
The paperwork would then be despatched to docs for all times expectancy calculation. Docs principally did this throughout their non-clinical hours, which already made them a scarce useful resource. To make issues worse, regardless of having staff to deal with preliminary classification, docs have been nonetheless pressured to spend vital time extracting and verifying knowledge from medical paperwork as a result of solely they possessed the specialised medical data wanted to appropriately interpret complicated medical terminology, lab values, and scientific findings.
Some case recordsdata have been enormous — reaching past 10,000 pages. Simply think about the sheer persistence and a focus to element required from the group and docs sifting by all that. That is why when the agency was on the lookout for automation options, there was a powerful emphasis on reaching practically 100% classification accuracy, self-learning knowledge extraction, and lowering person-hours.Â
How the underwriter carried out clever doc processing for medical information
Medical report volumes have been rising, and physician overview prices have been mounting. The underwriting group knew they wanted to automate their course of. However with life expectancy calculations depending on exact medical particulars, they could not danger any drop in accuracy through the transition.
Their necessities have been particular and demanding:
- Skill to course of hundreds of pages of medical information each day
- Understanding of complicated medical relationships throughout paperwork
- Classification accuracy needed to be near-perfect
- Fast and safe processing with out compromising high quality
- Combine out-of-the-box with Amazon S3
That’s when their VP of Operations reached out to us at Nanonets. They found that we might assist classify medical information with excessive accuracy, present a filtered view of great pages, extract knowledge key factors, and guarantee seamless knowledge flows inside the workflow. This satisfied them we might deal with their distinctive challenges.
This is what the brand new automated medical information automation workflow regarded like:
1. Doc preparation
- The interior workers combines all medical information— lab studies, ECG, chart notes, and different miscellaneous paperwork — for every affected person right into a single file
- Every affected person is assigned a singular quantity
- A folder with this quantity is created within the S3 enter folder
- 7-10 such instances are uploaded each day
Word: This strategy ensures safe dealing with of affected person data and maintains clear group all through the method.
2. Doc import
- The system checks for brand spanking new recordsdata each hour
- Every case can comprise 2000-10,000 pages of medical information
- Recordsdata are readied for secured processing by our platform
Word: This automated monitoring ensures constant processing occasions and helps keep the 24-hour turnaround requirement.
3. Doc classification
Our AI mannequin analyzes every web page based mostly on rigorously drafted pure language prompts that assist determine medical doc sorts. These prompts information the AI in understanding the precise traits of lab studies, ECG studies, and chart notes.
The classification course of entails:
- Figuring out doc sorts based mostly on content material and construction
- Understanding medical context and terminology
- Sustaining doc relationships and chronological order
- Recognizing when context from earlier pages is required
Word: The prompts are constantly refined based mostly on suggestions and new doc sorts, making certain the system maintains excessive classification accuracy.
4. Information extraction
Our system handles three most important doc sorts: lab studies, ECG studies, and chart notes. We now have two specialised extraction fashions to course of these paperwork – one for lab/ECG knowledge and one other for chart notes.
Mannequin 1 extracts roughly 50 fields from lab studies and ECG knowledge, together with affected person title, blood glucose stage, creatinine worth, glomerular filtration charge, hemoglobin worth, prostate particular antigen, white blood cell rely, hepatitis worth, ldl cholesterol worth, and plenty of different vital lab measurements.Â
Mannequin 2 processes chart notes to extract 13 key fields together with blood strain, heartbeat charge, O2 supply, O2 circulation charge, temperature, date of beginning, gender, top, weight, and smoking standing. Every knowledge level is linked to its supply web page and doc for verification.
5. Information export
The extracted data is exported as three separate CSV recordsdata again to the S3 Bucket — one every for doc classification, lab outcomes and ECG, and chart notes.
The classification CSV incorporates file names, web page numbers, classifications, and hyperlinks to entry the unique pages. The lab outcomes and ECG CSV comprise extracted medical values and measurements, whereas the chart notes CSV incorporates related medical data from docs’ notes.
In every file title, an identifier, like ‘lab outcomes’ and ‘ECG’ or ‘chart notes’, will likely be routinely added to determine the content material kind. And for consistency, CSV recordsdata are generated for all classes, even when no related pages are present in a case doc. Every affected person’s knowledge will likely be saved within the Export folder on the S3 bucket beneath the identical figuring out quantity.
6. ValidationÂ
The CSV outputs are imported into their inner software, the place a two-member validation group (diminished from the unique six) opinions the automated classifications. Right here, they’ll examine the extracted knowledge in opposition to the unique paperwork, making the verification course of fast and environment friendly.
As soon as the information is validated, the docs are notified. They will go forward to research medical histories and calculate life expectancy. As an alternative of spending hours organizing and reviewing paperwork, they now work with structured, verified data at their fingertips.
Word: For safety and compliance causes, all processed recordsdata are routinely purged from Nanonets servers after 21 days.
The impression of automated medical report processing
With structured knowledge and an environment friendly validation course of, the underwriting supplier has been capable of reduce the operational bottlenecks concerned within the course of.
Right here’s a fast overview of how a lot they’ve been capable of obtain inside only a month of implementation:
- 4 members on the information validation group have been reassigned to different roles, so validation now runs easily with simply 2 individuals
- Classification accuracy maintained at 97-99%
- Automated workflow is dealing with ~20% of the entire workload
- Full knowledge classification and extraction for every case file inside 24 hours
- Obtain a 5X discount within the variety of pages docs must overview per case to compute life expectancy
- Freed medical specialists to give attention to their core experience
These numbers do not inform the entire story. Earlier than automation, docs needed to sift by hundreds of pages as a result of they have been the one ones with the mandatory context to know affected person knowledge. Now docs get precisely what they want – detailed medical histories sorted chronologically which can be prepared for evaluation. It is a full shift from sorting papers to doing precise medical evaluation.Â
This variation means they’ll deal with extra instances with out having to rent dearer docs. That is an enormous benefit, particularly with healthcare dealing with workers shortages whereas the business continues to develop.
Wanting forward
This profitable implementation has helped the underwriting supplier perceive what’s doable with clever doc processing. They now wish to scale their medical report processing to cowl all ~200 instances weekly. That is not all. They’re already exploring how one can automate different document-heavy workflows, like belief deed processing.
Fascinated about what this implies on your group? The time to modernize doc processing is now. Healthcare documentation is changing into extra complicated, with a 41% progress in high-acuity care and rising continual situation administration. Add to this the rising staffing challenges in healthcare, and it is clear— when you do not modernize, your group will wrestle to maintain up.
Wish to see related outcomes together with your medical report processing? Let’s discuss how Nanonets might help. Schedule a demo now.
