Can AI improve recurrence prediction in thyroid cancer? Our systematic review found promising discrimination, but major gaps in calibration, external validation, and methodological quality. High AUC alone is not enough for clinical use.
https://t.co/hFqHwqLjTr
@MayoClinicEndo
Our team has received a $3.6M @NIHAging R01! 🎉 We’ll study whether #LT4 can be safely discontinued in older adults with subclinical #hypothyroidism. Grateful to our patients, collaborators & research teams who made this possible! @DeptVetAffairs@MayoClinicEndo@UFHealth
Thyroid overdiagnosis costs $25B annually. In @clinicalthyroid , I review how an AI adapter refines ultrasound risk via pixel-level analysis, matching experts (AUC 0.794) and preventing 55.7% of unneeded biopsies to cut costs
@mayoclinicendo
https://t.co/QBmf5SzWaE
How to read an AI study like a peer reviewer. A practical eight-question framework helps clinicians identify trustworthy #AI studies. #MedTwitter#EndoTwitter https://t.co/idOg0hzurB @TheAACE#EndocrineAI
Just like a resident memorizing practice exams, an AI that memorizes training data will fail in the real world. In my latest for @TheAACE Endocrine AI, I explore overfitting, data leakage, and the critical skepticism clinicians need
https://t.co/phEEXo1O8d
@MayoClinicEndo
Time to challenge the "lifelong" levothyroxine dogma? In Clinical Thyroidology, I review new data: 64% of older adults on ≤50 µg/day successfully stopped at 1 yr. But relaxed TSH targets & long-term unknowns require a careful approach
https://t.co/yIaTjV48bi
@MayoClinicEndo
Current management shifts patients toward invasive procedures for subclinical disease unlikely to cause harm. We need standardized cross-sectional imaging reporting & far more selective follow-up protocols.
Our latest @MayoClinicEndo study uses AI to expose a major driver of thyroid cancer overdiagnosis: incidental findings on routine CT/MRI scans. We built an NLP pipeline to analyze 115,683 patients over 6 years. The data: 🧵
https://t.co/rwQbnFGyTM
@doctorjuanpa
• The diagnostic cascade is massive: an incidental finding led to 46.8x higher odds of biopsy, 55.8x higher odds of total thyroidectomy, and 61.7x higher odds of a cancer diagnosis.
• 88.1% of these detected cancers were small, low-risk papillary carcinomas. 🧵
Our recent commentary examines AI-enhanced molecular diagnostics for thyroid nodules. While combining AI & genetics could improve precision, small cohorts and NIFTP classification remain hurdles to clear before clinical reality
@MayoClinicEndo@TheAACE
https://t.co/GAwabWp58C
Safe AI requires shifting from bedside vigilance to system-level safeguards. This relies on three pillars:
- Enterprise accountability and shared risk
- Institutionalized stakeholder governance
- Continuous automated monitoring.
Responsibility belongs upstream
The "clinician in the loop" model is a flawed solution for AI oversight. It shifts accountability from developers to doctors, making clinicians the "moral crumple zone" for system failures.
Our new BMJ analysis: 🧵 https://t.co/dfZV6XWfys
@MayoClinicEndo@doctorjuanpa
Regulators and health systems increasingly rely on human oversight at the point of use as a primary safeguard for AI.
“Clinician in the loop” oversight shifts responsibility for AI safety from developers to doctors
https://t.co/Spb8RwG95J
Build practical diabetes management skills with the Essential Diabetes Care online course—recorded lectures and case studies in a self‑paced 9+ hour online format for providers.
Learn more: https://t.co/LLlcWi60kF
#CME#Diabetes
At @MayoClinicEndo, we are evaluating next-gen therapies targeting the root causes of thyroid autoimmunity. Explore our active clinical trials for Graves' disease and TED here: https://t.co/791vvd12ib
#Endocrinology#ClinicalTrials