Blog
Field notes on payer analytics, clinical informatics, and healthcare AI.
Filter by topic
Subscribe
New analysis, when it publishes.
Occasional notes on claims data, clinical informatics, and healthcare AI. No newsletter filler, no sequences, unsubscribe in one click.
Or follow the RSS feedThe boring way to read the drug pipeline beat the clever one
Two percent of the drug pipeline carries roughly 40% of its cost-weighted exposure. I went looking for that risk with three signals, one pointed the wrong way, one could not rank what mattered, and the one that worked barely needed a model.
Two percent of the drug pipeline carries roughly 40% of its cost-weighted exposure. I went looking for that risk with three signals, one pointed the wrong way, one could not rank what mattered, and the one that worked barely needed a model.
- Payer Analytics
- Health-Data Methods
- Research Notes
Eight authoritative ways to say “diabetes”
Ask eight recognized stewards for the ICD-10-CM codes that identify diabetes and you get eight different answers, spanning 87 to 633 codes. Which one you grab quietly rewrites your cohort.
Only 86 codes appear in all eight definitions, 120 appear in exactly one, and the broad diabetes sets are up to 82% non-type-2. Silent substitution of one value set for another moves cohort size and prevalence by multiples.
- Health-Data Methods
- Research Notes
The four-star cliff: how a moving quality line reprices Medicare Advantage
At 4.0 stars a contract earns a 5% benchmark bonus; miss the line and that bonus is gone, while the rebate share it keeps sits on a separate, wider schedule.
The share of contracts at or above 4.0 swung from 68% to 42% in two years, and roughly two in five were downgraded in a single cycle. Stars are not a moat, they are a repricing event, and they arrive on a two-year delay.
- Medicare Advantage
- Payer Analytics
The zero-premium ceiling: what a saturated premium lever signals for MA margins
Nearly half of individual Medicare Advantage plans were free in 2019; two in three are today, but the share stopped climbing around 2023, and the plans that still charge held their price.
The $0 premium is a lever the market has largely spent. The free share plateaued near 67%, the average premium fell on mix rather than price, and competition has moved into supplemental benefits and the Part B giveback.
- Medicare Advantage
- Payer Analytics
LLMs in clinical documentation and coding: separating real capability from demo-ware
The demo shows a language model turning a discharge summary into a clean claim. The peer-reviewed literature shows something very different. Here is how to tell which one you're buying.
Autonomous ICD-10-CM assignment lands near 34% exact match, while LLM validation of already-assigned codes hits roughly 94% against a human baseline of 85%. The gap is not a bigger model, it is a smaller problem, and it tells buyers exactly which products are deployable today.
- Healthcare AI
- Health-Data Methods
The healthcare AI that outlasts AGI: own what stays scarce
A founder's map of which moats survive cheap intelligence, and which dissolve in it.
If general-purpose intelligence keeps getting cheaper, the cognition at the center of most healthcare AI pitches becomes the commodity, not the moat. The durable opening is the regulated result, proprietary physical and biological data, in-person presence, payment and compliance rails, and the trusted patient relationship.
- Healthcare AI
- Market Thesis
Build healthcare AI where no one's looking: the back office, not the bedside
A founder's map of where to build, and where not to.
The pitches everyone is funding, ambient scribes and imaging triage, sit in the most crowded squares on the board. The durable opening is upper-right on a need-versus-market map: denials and revenue integrity, nursing workflow, trial operations, perioperative ops, and LLM-ready data infrastructure.
- Healthcare AI
- Market Thesis
They moved south. Did costs go up?
A workforce's health-cost risk changes the day its footprint changes. The direction is rarely the one you'd guess, and you can model it from ZIPs and headcount, long before any claims arrive.
A mid-market firm acquires a 1,100-person Sunbelt company and closes its Hartford office. Expected chronic-disease cost rises $3.43M, but $4.12M of that is headcount and -$0.68M is a geography credit. On a per-head basis, the blended cost falls 4.5%.
- EPHP
- Payer Analytics
Two Atlanta offices, 8 miles apart, $1,466/EE in projected chronic disease cost
What ten chronic conditions already explain about your renewal, before you even open the claims file.
State and national averages hide the variance entirely. Two 5,000-EE Atlanta workforces eight miles apart sit on opposite sides of the US chronic-disease cost benchmark, a $7.3M annual gap on chronic conditions alone.
- EPHP
- Payer Analytics
Let's Connect
Open to thoughtful collaborations.
Whether you're shaping a clinical strategy, evaluating a product roadmap, or exploring an opportunity in population health - I'd be glad to hear from you.