AI-designed drug could slow aging; chatbots abandon referral advice when patients push back; virtual reality helped anxious patients finish cardiac scans — Morning Medical Update

AI-designed drug lowered patients' predicted biological age

Six proteomic aging clocks all predicted lower biological age in a 12-week phase 2a trial.

A drug candidate designed with generative artificial intelligence lowered patients' predicted biological age across six proteomic aging clocks, according to an analysis published in Nature Biotechnology.

Rentosertib, a TNIK inhibitor that Insilico Medicine is developing for idiopathic pulmonary fibrosis, was tested against placebo in a 12-week phase 2a trial conducted in China. Serum from 42 participants with a mean age of 67.1 years was profiled at baseline and at weeks 2, 4 and 12. All six clocks predicted lower biological age in the treated arms, while the placebo group showed little change or slight increases.

The 30 mg twice-daily regimen produced the most consistent signal, and effects peaked at week 4 before plateauing. The authors wrote that proteomic clocks alone cannot separate aging effects from the drug's anti-fibrotic activity, and that confirmation would require studies outside a pulmonary fibrosis cohort.

The analysis was led by Insilico Medicine; its founder and chief executive officer is the corresponding author, and several coauthors are company employees. Cardiologist Eric Topol, M.D., who was not involved in the work, told The New York Times the drug looks encouraging but that no definitive trial has been run.

Chatbots abandon referral advice when patients push back

Advice to seek a sleep study survived in 64% of conversations with patients who resisted referral.

Consumer artificial intelligence chatbots dropped correct advice to seek a specialist referral for obstructive sleep apnea in more than a third of conversations with patients who played down their symptoms, according to research presented at the European Respiratory Society Congress in Barcelona, Spain.

Deeban Ratneswaran, a research fellow at Guy's and St Thomas' NHS Foundation Trust in London, and colleagues built seven simulated patients who all met criteria for a sleep study, then ran 700 conversations against ChatGPT, Google Gemini, Claude, DeepSeek and Grok. With a cooperative patient, all 350 conversations ended with advice to seek specialist assessment. With a resistant patient presenting identical medical facts, that advice survived in 225 of 350.

The models gave way most in the most serious scenarios. Correct advice held in 22% of conversations built around a textbook severe case and in 32% with a patient who had dozed off at the wheel, and the chatbots usually left the driving risk unmentioned.

Depending on the model, a quarter to half of conversations with resistant patients ended in lifestyle tips rather than a referral. Io Hui, who chairs the society's group on m-health and e-health and was not involved in the research, attributed the pattern to AI sycophancy, a tendency to please the user. The abstract has not been peer reviewed.

Virtual reality helped anxious patients finish cardiac scans

Of 105 patients offered a headset during PET/CT, 95% completed the ordered study.

Offering a virtual reality headset to patients with anxiety or claustrophobia helped 95% of them complete cardiac PET/CT imaging, according to a quality improvement project presented at the American College of Cardiology Quality Summit 2026. Staff in the non-invasive cardiac testing department at UCHealth's Southern Region in Colorado Springs, Colorado, evaluated 105 patients who used the headsets during scans from July 2025 through March 2026.

All but four reported some degree of claustrophobia. Eighty percent kept the headset on for the full procedure, and more than 75% reported a positive experience; poor headset fit was the most common complaint.

Patients chose from calming nature and travel environments meant to draw attention away from the scanner. Those with a seizure history or undergoing fluorodeoxyglucose studies were not eligible. The system cost $4,387 up front and $168 a month, against an estimated $3,000 for each aborted study in reimbursement, supplies and staff time.

Lead author Kaitlyn Amidon, BSN, RN, MBA, said the goal was carrying claustrophobic patients through the scan without compromising diagnostic quality.

The team cautioned that this was not a randomized trial and that no comparable baseline completion rate exists, because a new PET/CT scanner was installed at the same time.