
Our CARE-PF analysis is out in Respirology. n=4236 fibrotic ILD. Transplant-free survival 91%/72%/57% at 1/3/5 yr (IPF 39%; CTD-ILD 70% at 5 yr). @UofT @UHN @UofT_DoM doi.org/10.1002/resp.70305

🇨🇦Canadian I Respirologist🫁| PH-ILD Doctor @uhn| Alumnus @UNB @WesternU @mcmasterU @uniofeastanglia. Founder of coreSR.ai.

Our CARE-PF analysis is out in Respirology. n=4236 fibrotic ILD. Transplant-free survival 91%/72%/57% at 1/3/5 yr (IPF 39%; CTD-ILD 70% at 5 yr). @UofT @UHN @UofT_DoM doi.org/10.1002/resp.70305

In 1904, we called this Clever Hans [00:00]. Today, we call it shortcut learning. Here is why high accuracy and external validation aren't enough to prove an AI model is looking at the right signal: youtube.com/watch?v=r07455dR… #AI #ML #NVIDIA #CleverHans

TITAN-SR started on ~700K examples and is now on v6 at up to 6M unique examples, delivering higher specificity across diverse review types without compromising high recall targets. Watch the breakdown: youtube.com/watch?v=LdM3lN68… Start here: coresr.ai

Most survival meta-analyses are stuck reading a number off a figure. We rebuilt the 320 patients from the HYPERION KM curves Published HR: 0.24 Reconstructed Cox: 0.238 Same answer, from the picture alone. Full demo: youtu.be/22WY8DSxujY #SystematicReview #KM #CoreSR

Launching SynthesisBench: an open screening benchmark across 85 systematic reviews, 2,327 records and 26 systems. Top Youden's J: TITAN-SR v6 0.879, Claude Opus 5.5 0.831, GLASS-SR 0.827. Run by CoreSR. synthero.ai/screening

New CoreSR models: TITAN-SR v6 screens, GLASS v3 appraises with quotes from the paper, ROB BERT v6 fills in RoB 2 while a second model checks it. On 60 held-out reviews, TITAN kept 932 of 933 includes and screened out 78% of the rest. youtu.be/uTbnVkTiOnE

We’re excited to spotlighting past week publications from researchers across the University of Toronto and the Toronto Academic Health Science Network (TAHSN) #TAHSN #TorontoPainResearch #UTCSP

Canada records 60 per cent spike in arrests, detentions of Canadians in United States ctvnews.ca/canada/article/ca…

Is JEV a new architecture? It might just be a smallish LLM with a muzzle...#jev #llm #ai youtube.com/watch?v=UdHV216g…


Which AI should screen 10,000 abstracts? Chatbots write one word at a time. BERT reads both ways and gives a score you can tune for high recall. TITAN-SR kept 99.8% of included studies in 22 later Cochrane reviews. sciencedirect.com/science/ar… coresr.ai #SystematicReview #Epi

Takeaway: pulse dosing didn't beat a lower-dose regimen, so it's reasonable to question routine pulse use in AE-IPF. Caveats: open-label, n=100, and not powered to exclude a modest benefit. doi.org/10.1093/ajrccm/aamag… #IPF #ILD 5/5

Result: no survival benefit from pulse dosing. 28-day mortality 10.0% vs 14.6% (P=0.55) 90-day mortality 24.5% vs 29.8% (P=0.65) Wide CIs, and secondary outcomes (ICU, ventilation/ECMO, adverse events) didn't differ either. 4/5

Design: open-label RCT, 8 South Korean university hospitals, 100 adults with AE-IPF (51 pulse, 49 non-pulse). Pulse: methylpred 10 mg/kg x 3 days Non-pulse: 1 mg/kg x 7 days Both then got a protocolized taper to low-dose prednisolone. 3/5

Why it matters: AE-IPF has high short-term mortality. Guidelines only weakly recommend steroids and don't define dose or duration, yet pulse methylprednisolone is widely used on thin evidence. 2/5

Pulse-dose steroids for acute exacerbation of IPF: do they actually save lives? 🧵 SAFER just published in AJRCCM, a multicenter randomized trial testing this head-on. Short answer below. 1/5 #IPF #ILD #PulmTwitter

Not all ILA progresses. Can CT AI flag who will? COPDGene+AGES: MIL-UIP. Score>0.5 → ~13–14%/yr fibrosis↑; death HR~1.6 both cohorts. Takeaway: automated UIP-like CT may risk-stratify ILA. Caveat: observational; needs validation. #ILD #IPF #PulmTwitter doi.org/10.1093/ajrccm/aamag…