Reuters reports five of six BVCT trial calls right; four NO-GO calls remain open
: Reuters reported that five of six BioinvestGPT calls on closely watched trials were right; the miss was pelacarsen. Four of the five are scored on the BVCT dashboard: HARBOR (del-desiran) True Negative (TN), CARDIO-TTRansform (eplontersen) True Negative (TN), ZEUS (ziltivekimab) True Negative (TN) and INTerpath-001 (intismeran autogene) True Positive (TP), on recurrence-free survival; overall survival is pending.
Reuters reported on that BioinvestGPT’s calls on closely watched drug trials were so far correct in five of six (Reuters). The story, by Reuters’ Deena Beasley, looks at companies that use AI to simulate clinical trials with virtual patients.
One of the five calls was on HARBOR, the Phase 3 trial of del-desiran, Novartis’ experimental drug for myotonic dystrophy type 1. BioinvestGPT’s BeatSoC Virtual Clinical Trial (BVCT) call, sealed on , was for an insignificant clinical benefit, and HARBOR did not meet its primary endpoint. Reuters wrote that Novartis’ chief executive had forecast $5 billion in annual peak sales, but that the result did not surprise BioinvestGPT.
As Reuters describes BioinvestGPT’s platform, it builds virtual humans to fit a trial’s entry criteria and then tests a model of the drug on them. BioinvestGPT co-founder Bragi Lovetrue told Reuters: We can pinpoint the reason why a drug is effective and safe, and in many cases, why not (Reuters).
The five calls Reuters counted as correct
Four of the five are scored on the dashboard. Their record cards show when BVCT sealed each call and how far ahead of the readout, from 24 days for INTerpath-001 to 859 days for ZEUS. The fifth, on Vaxcyte’s pneumococcal vaccine, is not yet scored. On the dashboard, a true negative (TN) is a call of weak or insufficient benefit that the readout confirmed; a true positive (TP), a call of sufficient benefit that the readout confirmed.
FAILURE (statistically insignificant clinical benefit in vHOT and 10MWT compared with placebo)
Commercial Prediction
FAILURE (commercially insufficient clinical benefit in vHOT and 10MWT slightly inferior to DYNE101 with RR > 1.2 and moderately inferior to PGN-EDODM1 with RR > 1.4)
Scored on the primary endpoint, vHOT; 10MWT is not yet adjudicable. Outcome: Reuters, . Full case, including BVCT’s earlier FAILURE call on the same asset at Phase 1/2 (No. 999): del-desiran (AOC 1001) case study.
SUCCESS (statistically significant additive clinical benefit in RFS and OS compared with placebo)
Commercial Prediction
SUCCESS (commercially sufficient additive clinical benefit in RFS weakly-to-moderately superior to pembrolizumab with RR < 0.7 and in OS weakly-to-moderately superior to pembrolizumab with HR < 0.7)
Reuters counted Moderna and Merck’s melanoma vaccine among the five; this is its Phase 3 trial. Scored on recurrence-free survival; overall survival data are still pending. Outcome: Reuters, . Full case: Moderna pipeline case study.
partial SUCCESS (statistically maybe significant yet weak additive clinical benefit in CV mortality/CV recurrence compared with SoC alone (tafamidis))
Commercial Prediction
partial SUCCESS (commercially maybe sufficient yet weak additive clinical benefit in 6MWT/KCCQ/all-cause mortality at most weakly superior to SoC alone (tafamidis))
Reuters summed up this call as a weak clinical benefit for AstraZeneca and Ionis’ Wainua (eplontersen), in line with both recorded calls above; the trial missed its primary endpoint. Outcome: Reuters, . Full case: eplontersen case study.
partial SUCCESS (statistically maybe significant yet weak additive clinical benefit to SoC in MACE reduction)
Commercial Prediction
FAILURE (commercially insufficient additive clinical benefit to SoC with inferior-to-semaglutide efficacy in MACE)
Reuters counted ziltivekimab’s first failed trial among the five: ZEUS, which read out before HERMES and ATHENA were discontinued (both also covered in the case study). BVCT’s NO-GO is the commercial call above. Outcome: Reuters, . Full case: ziltivekimab case study.
Commercial: SUCCESS — commercially sufficient benefit in OPT GMT for age between 18 and 49 years at least numerically superior for the age cohort greater than 50 years with RR < 0.9). SUCCESS —commercially sufficient clinical benefit in severe pneumonia prevention weakly superior to PCV20 / PCV21 with HR < 0.8 for both age greater than 50 years and age between 18 and 49 years due to pAMF.
Technical: SUCCESS — statistically significant benefit in OPT GMT at most numerically inferior to PCV20 / PCV21 for shared serotypes (1.0 < RR < 1.2) and superior for novel serotypes with RR < 0.8.
Ref: HANPL-V8SAM-PYCPA-TXRDM
Phase 3 · Pneumococcal Vaccine · vs. PCV20 / PCV21. Reuters reported success on . Its virtual-trials story names Vaxcyte’s pneumococcal vaccine but no trial; matching that call to No. 2158 is our reading.
Reuters also reported a miss: BVCT had expected Novartis’ lipoprotein(a)-lowering drug pelacarsen to succeed, yet in its late-stage trial it did not lower the risk of major heart attacks or strokes among people with an inherited risk factor. At publication on , pelacarsen was not on the dashboard, so the cohort frozen on does not include it; Reuters reported the result on . Lovetrue told Reuters that a later analysis showed we got the mechanism wrong (Reuters): the model had missed genetically set variation in lipoprotein(a) particle size.
Four open NO-GO calls
The story also covers four open BVCT calls, all still awaiting readout: Biogen’s TOPAZ-1 and TOPAZ-2, Phase 3 trials of litifilimab in systemic lupus erythematosus, and Takeda’s two Phase 2 zasocitinib trials, one in Crohn’s disease and one in ulcerative colitis. In Reuters’ summary, the model finds both drugs less than optimal for the populations these trials enrol. All four are NO-GO calls. For litifilimab, the technical call is partial FAILURE and the commercial call FAILURE; for zasocitinib, both are FAILURE. We first published these calls on (see our Top 10 readouts case study); Reuters printed them on .
Reuters noted that Takeda has recently asked US regulators to approve zasocitinib, a TYK2 blocker, for plaque psoriasis, and is testing it in psoriatic arthritis in Phase 3. According to Reuters, Takeda’s research chief said that human genetics pointed to TYK2 as a target and that machine learning helped fine-tune the pill’s chemical structure. The research chief expressed great confidence in the mechanism but doubted that AI tools are yet close to making definitive predictions.
Biogen’s head of clinical development for multiple sclerosis, immunology and Alzheimer’s said, according to Reuters, that the company uses all the tools it has, AI among them. She added that with only two biologics approved for lupus, AI algorithms built on historical data may lean towards predicting a negative result. Reuters notes litifilimab is also being studied in a different type of lupus (earlier report: Reuters, ).
Commercial: FAILURE — commercially insufficient benefit in clinical remission CDAI and endoscopic response SES-CD slightly inferior to deucravacitinib with RR > 1.2 and strongly inferior to tulisokibart with RR > 1.7.
Technical: FAILURE — statistically insignificant benefit in clinical remission CDAI and endoscopic response SES-CD vs. placebo.
Ref: MEFEG-XX7SH-JPDXH-OAOVN
Phase 2 · Moderately-to-Severely Active Crohn’s Disease.
Commercial: FAILURE — commercially insufficient benefit in clinical remission mMS numerically-to-slightly inferior to deucravacitinib with RR > 1.1 and moderately-to-strongly inferior to tulisokibart with RR > 1.5.
Technical: FAILURE — statistically insignificant benefit in clinical remission mMS vs. placebo.
Ref: K3STJ-EZQQ7-K9BF7-BHGXB
Phase 2 · Moderately-to-Severely Active Ulcerative Colitis.