OpenRounds Editorial

Healthcare AI Daily Briefing

Thursday, October 29, 2026

New to OpenRounds? Get the free daily email or see how we work.

Policy & Governance

Interview: Health‑system leaders say they lack leverage over AI vendors and need model‑change transparency.

Podcast from Lifers (Second Opinion)

The interview describes a convening of health‑system C‑suite and clinical leaders who reported that hospitals feel they have limited leverage with major EHR vendors and new AI companies, and they require visibility into any model changes to manage risk and liability. [1]

Operations & Workflow

Interview: ShiftMed uses AI agents for nurse onboarding.

Podcast from Healthcare AI Pioneers

The interview describes ShiftMed moving from human recruiters managing credentialing via email and PDFs to a platform where nurses keep documents in a cloud‑based app, with AI agents conducting onboarding conversations around the clock for the roughly thirty thousand monthly app downloads. [2]

Diagnostics & Pathology

AI screens blood smears to flag inpatient mortality risk within five days.

Podcast from AI and Healthcare

The interview describes a lab effort to automate peripheral blood smear screening for sepsis, clinical deterioration, and mortality risk in hospitalized patients. The system combines smear morphology — such as immune cell activation visible even in immunocompromised patients — with vital signs and labs, then calibrates a risk score; for example, a concerning slide can yield a ninetieth‑percentile risk of ICU mortality in the next five days. [3]

Medical Imaging

Explainable AI cuts under‑reliance and boosts accuracy on head CT hemorrhage reads, especially on hard cases.

Journal article (The Lancet. Digital health)

In a randomized crossover trial of 50 radiologists and 75 novices each interpreting 50 head CT scans, correct explainable AI advice (AI prediction with annotations) reduced under‑reliance and improved diagnostic performance (OR 1.84, p<0.0001) and confidence (b=0.15, p<0.0001) versus no advice, with the greatest benefit on difficult intracranial hemorrhage cases. Explainable AI did not significantly shorten reading time (b=‑1.81, p=0.071) and was rated more useful than basic AI, particularly by novices. Both over‑reliance on incorrect advice and under‑reliance on correct advice occurred. [4]

Medical Imaging

3D CNN detects cervical spinal cord injury on CT with stable AUCs across multicenter validation.

Journal article (BMC medical imaging)

Researchers built a standardized 3D CNN using cervical spine CTs from two independent centers, with ROIs delineated via ITK-SNAP and stratified random splits for training, internal validation, and external validation. The model showed stable discriminative performance across all three sets: AUC-ROC 0.943, 0.953, and 0.948; AUC-PR 0.936, 0.939, and 0.943, with minimal variation and no evident overfitting. Threshold sensitivity analysis indicated flexibility for clinical decision-making. The authors note prospective validation is required before clinical deployment and position the framework as a preliminary research tool for settings where MRI is unavailable. [5]

Diagnostics & Pathology

Glycated peptide model predicts cardiovascular disease in peritoneal dialysis patients with 0.822 AUC in external validation.

Journal article (Clinical and translational medicine)

In a prospective multicenter study, researchers combined two glycated peptides — GEAFTLK(g)ATVLNYLPK and K(g)QHLFVK — with six clinical variables (phosphate, diabetes, total Kt/V, albumin, troponin T, glycated albumin) using elastic-net logistic regression. The model achieved ROC-AUC of 0.822 in an external validation cohort of 102 patients, outperforming three published cardiovascular risk models. A training-derived Youden cutoff of 0.373 identified a high-risk group with higher 1-year CVD incidence in both internal (n=240) and external validation sets. [6]

Sources

A hospital AI deal done in under two months | Dr. Ilya Burkov, Nebius & Thomas Knox, VitVio · Lifers (Second Opinion) Original source
Partnering to Meet Evolving Workforce Needs at SSM Health · Healthcare AI Pioneers Original source
How AI Reads Blood Cells to Help Diagnose Cancer with Gregory Goldgof, MD, PhD · AI and Healthcare Original source
Examining reliance patterns on AI advice in medical imaging: a mixed-methods, randomised, crossover experiment. · The Lancet. Digital health Original source
Construction and validation of a unified 3D convolutional neural network framework for automated identification of cervical spinal cord injury based on CT imaging. · BMC medical imaging Original source
Predicting cardiovascular disease in peritoneal dialysis patients using glycated proteomics and machine learning: A prospective multicentre study. · Clinical and translational medicine Original source