CMS pays surgeons by a number that's 20% self reported depression 🧪, Spring + Alma stitch 170M lives 🧵, (Yet another) Doximity launches.
An orthopedic surgeon and a spine surgeon published the same story from opposite ends on Friday - and the substrate underneath the year's most-watched MSK payment models is partly measuring the wrong
The measurement is the policy: CMS is about to pay MSK surgeons by a number that’s 20% self reported depression.
Two surgeon-builders published the same story from opposite ends on Friday. A spine surgeon in Reading, Pennsylvania walked through the Oswestry Disability Index, the most-used PROM in lumbar spine surgery, and put numbers on something most surgeons have felt in clinic for years — Carreon et al. (2016) found a single SF-36 item, “Have you felt downhearted and depressed?”, accounts for about 20% of variability in 1-year ODI scores after lumbar fusion. A separate orthopedic surgeon at Duke read the six current and proposed CMS MSK payment models — TEAM, ACCESS, LEAD, ASM, CJR-X, RISE — together as one architecture, and put another number on the table: 2000–2024 mean Medicare reimbursement for primary THA/TKA dropped 56%; modeling forecasts an 85–86% decrease by 2030.
The two pieces interlock. CMS’s response to MSK price compression is to bolt mandatory PROM-driven adjustments on top of episode-based payment, in five overlapping models, with ±9–12% individual-physician payment swings tied to scores like the ODI starting January 2027. The PROM defect piece says the ODI is a self-report about belief about physical capacity — not a measurement of physical capacity — and roughly a fifth of it is measuring depression. CMS adjusts for age, comorbidities, SES in some MSK models. It does not adjust for the fact that 20% of the ODI is measuring something other than disability. Adverse selection applied to psychology — and the substrate the year’s most-watched payment models depend on is partly measuring the wrong thing.
❤️ Love
Check out the primary substacks, as it can’t really be summed up here. Also, it’s amazing how (i assume) gpt-image-2 (or claude designer) is making infographic for ortho substacker publishers.
😤 Haters
“This is just MSK. Doesn’t apply to me.” The pattern is the news. CMS is going to do this in primary care PCMH redesigns, in oncology APMs, in behavioral health bundles. Payment models are evolving faster than the measurement instruments they ride on. The clinician who learns to read measurement substrates this way reads the next five payment models in 90 minutes instead of finding out about the variance attribution after their bonus shrinks.
“Surgeons should just screen for depression and adjust their patient selection.” That answer turns into adverse selection by another name — surgeons in rural areas with limited mental-health access, lower-SES populations, regions with higher chronic pain or OUD prevalence end up filtering out the patients who needed surgery most. The honest answer is the one DelSole writes: screen, refer, document the contemporaneous recognition, and push the professional societies to demand psychological-comorbidity risk adjustment in the comment windows that are open right now.
“PROMs are the best we have.” PROMIS short forms, computer-adaptive testing, fear-avoidance-aware scoring all exist. The technical answer is real. The question is whether CMS adopts it in CJR-X (comments through Jun 9) or whether the answer ships as a vendor product on top of an unchanged measurement substrate. Either path produces a procurement-grade column on slide three of the eval deck — variance attribution, the same way the past two weeks added calibration to slide three.
Spring Health closed Alma — AI-native plus clinician marketplace, 170M lives, on pace for 10M visits this year.
Spring Health is an employer-facing mental health platform that uses AI to route members to the right level of care — therapy, coaching, or medication. Alma is a membership platform that helps independent therapists accept insurance and run a private practice. The combination of the two closed Friday, four months after announcement. The interesting part is the vertical pairing — an AI routing layer now owns its own clinician supply network, covering 170M lives. That shape — predictive routing on top of independent-practice infrastructure — did not previously exist as a single company, and it is a shape primary care, endocrinology, and several specialty bands will see in the next 18 months.
😤 Haters
“AI consolidates again. We’ve heard this story.” The frame people will reach for. Not the story. The story is an AI-native router acquiring a clinician marketplace as the supply layer — that pairing is new, and the next vertical to see it is primary care.
“Alma’s clinicians lose their independence the moment Spring’s routing model decides who they see.” The legitimate concern. Watch April Koh’s “third chapter — continuity” framing closely; if Spring routes by outcome data, the clinicians whose panels look better in Spring’s metrics get more referrals, which is a different kind of consolidation pressure than the legacy MSO model. Whether Alma’s clinicians experience this as scale or as control is the early-tell to watch.
💡 80/20: AI-native router on top of clinician marketplace is the shape.
Roon launched an AI-native physician network — neurosurgeon and ex-Pinterest founders, MedTwitter’s intentional replacement.
Roon went live this week, founded by neurosurgeon Rohan Ramakrishna and former Pinterest exec Vikram Bhaskaran. Verified-physician-only, AI synthesis of research wired into the social architecture (not next to it), structured peer-to-peer organized around case conferences and journal clubs. Reported early traction: thousands of doctors across cardiology, oncology, infectious disease, plus former heads of NIH, NCI, and CDC.
😤 Haters
“This is the fifth physician-only network I’ve heard about and they all died.” True for Sermo, the Slack-style clones, and the directory plays. Roon’s bet is structurally different — AI synthesis inside the social architecture rather than next to it, with verification done seriously. Whether that is enough to clear the network-effects threshold is the open question, but the design instinct is the first one in this category that has felt coherent in 2026.
“Doximity owns this market.” 80% physician penetration, yes. Doximity’s response time is the variable. If they ship a Roon-style synthesis layer in 90 days, the upstart’s differentiation window compresses. If they treat it as a curiosity, Roon has 18–24 months to become the default for case-discussion content.
💡 80/20: The training corpus Roon is building — labeled clinical reasoning posts and labeled engagement, with no PHI exposure — is a non-trivial AI asset.
What are you building this week? Reply and tell me — I read every one.
— Kevin