onc brain

About · curated by Nick Boehling, MD · @nb2276
Confirmatory

ARACOG (AFT-47)

ForAdvanced prostate cancer (mHSPC, nmCRPC, mCRPC) starting an ARSI

Maximally Changed Cognitive Domain (CANTAB) at 24 weeks safety

daro -15.8 vs enza -36.1

median % change in MCCD at 24 wks, P=0.009

TL;DREnzalutamide MCCD decline -36.1 vs -15.8 for darolutamide at 24wks (P=0.009) in randomized phase 2, N=111.

Why it mattersRadiation oncology

The comparison is between each pt's own worst-hit domain, and those differed by arm (PALFAM for daro, SWM for enza), so the read is how far the worst domain falls, not which one. Crossover before 24wks was scored at crossover inside the randomized arm, which attenuates rather than widens the gap. Moves ARSI selection when cognitive burden matters, not efficacy.

Monday clinic

In a man starting an ARSI for mHSPC or nmCRPC where cognitive burden is a live concern, this supports darolutamide over enzalutamide on measured cognition; it does not speak to disease control, which was never compared head-to-head.

ARACOG (AFT-47)
MetricDarolutamide (N=48)Enzalutamide (N=47)
Maximally changed modulePALFAMSWM
DomainVisual memory / executive functionWorking memory / executive function
Median change, baseline to 24 wks-15.8-36.1
Between-arm PP=0.009P=0.009
+2 more figures
ARACOG (AFT-47)
Schema: N=111 with mCRPC, nmCRPC or mHSPC randomized to darolutamide (study-provided) vs enzalutamide (standard of care), stratified by age (<65, 65-80, >80); enrolled 8/17/2021 to 3/11/2025.
Schema: N=111 with mCRPC, nmCRPC or mHSPC randomized to darolutamide (study-provided) vs enzalutamide (standard of care), stratified by age (<65, 65-80, >80); enrolled 8/17/2021 to 3/11/2025.
8 details

Randomized open-label phase 2 from the Alliance for Clinical Trials in Oncology, N=111, stratified by age (<65, 65-80, >80). Enrolled 8/17/2021 to 3/11/2025, with CANTAB modules, PROMs and timed-up-and-go collected at 12 and 24 weeks.

Men with mCRPC, nmCRPC or mHSPC. Randomization was stratified by age across three bands (<65, 65-80, >80), so the design anticipated an older population. Baseline cognitive eligibility criteria not reported in source.

Darolutamide was provided by the study; enzalutamide was given through standard of care, so the two arms differed in drug supply as well as drug. Doses and concurrent ADT not reported in source.

Primary: % change from baseline to 24 weeks in the Maximally Changed Cognitive Domain (MCCD), drawn from 5 remotely delivered CANTAB modules (SWM, PALFAM, OTS, SSP, RVP) covering executive function, visual memory, attention and working memory. Blood for polygenic hazard score and AR testing, PROMs and timed-up-and-go were collected alongside.

Median MCCD change -15.8 with darolutamide (PALFAM) vs -36.1 with enzalutamide (SWM), P=0.009, across 48 and 47 pts in the primary analysis. No oncologic endpoint was reported in source.

The two drugs have never been compared head-to-head for efficacy, and cognition in ARAMIS and ARASENS (darolutamide) and PROSPER and ARCHES (enzalutamide) was captured as clinician-graded adverse events against placebo, not measured with a battery. Mild executive dysfunction is exactly the harm an AE table structurally misses.

men on darolutamide or enzalutamide across mHSPC, nmCRPC and mCRPC, tested to 24 weeks
Does not represent men on apalutamide or abiraterone, nor men followed beyond 24 weeks.

Pts crossing over before 24 weeks carried their crossover score into the randomized arm, which should attenuate the gap rather than widen it. CANTAB is a research battery, not a clinical diagnostic, and no link to function, falls or discontinuation is reported in source. 24 weeks says little about years of therapy.

This shifts an argument that has run on mechanism and on AE tables onto measured performance. Where the two drugs are treated as interchangeable for disease control, cognition becomes a defensible tiebreaker. What it does not settle is whether an MCCD gap of this size is felt by the pt.

Randomized, prespecified primary endpoint met against a named comparator; open-label design and the composite MCCD construct temper it. Consistent with darolutamide's known limited CNS penetration.

  • Whether the MCCD difference translates to function, falls, or discontinuation
  • Durability of cognitive divergence beyond 24 weeks
  • Whether apalutamide differs from enzalutamide on the same testing
📚 Sources · 🐦 2 tweets

The longer read

The registrational programs for these two drugs never ran against each other, and neither measured cognition directly: ARAMIS and ARASENS for darolutamide, PROSPER and ARCHES for enzalutamide, all captured CNS effects as clinician-graded adverse events against placebo or against a chemotherapy backbone. An AE table is a poor instrument for this particular harm, because mild executive dysfunction rarely gets recorded as an adverse event and almost never gets graded above 1. Replacing that with repeated performance testing in pts randomized between the two drugs is the design contribution here, and it is the reason a 111-pt phase 2 is worth the reader's time at all.

The endpoint deserves scrutiny before the result does. Maximally Changed Cognitive Domain takes each pt's largest change across five CANTAB modules and compares the arm-level medians of that quantity. It buys sensitivity: five modules with overlapping domain coverage would each be underpowered at this sample size, and testing them separately would invite multiplicity. It costs interpretability, because the arm-level answer no longer names a domain. In this trial the maximally changed module was PALFAM in the darolutamide arm and SWM in the enzalutamide arm, so -15.8 and -36.1 are not two readings of the same test. The defensible statement is that the worst-affected domain fell further on enzalutamide, not that working memory specifically was hit.

Two design features push in opposite directions on confidence. Crossover before 24 weeks was handled by carrying the score at crossover into the randomized arm, which should attenuate rather than inflate a between-arm difference, since the pts most likely to leave a drug early are the ones doing worst on it. Against that, the trial was open-label and enzalutamide came through standard of care rather than from the study, so assignment was known to pt and clinician throughout. Performance-based testing is harder to bias than a questionnaire, but it is not effort-independent, and repeat administration carries practice effects that only cancel if engagement is symmetric between arms.

What the trial does not do is compare the drugs on disease control, and the population, spanning mHSPC, nmCRPC and mCRPC, is too mixed at this size to say whether the gap holds equally for a man starting an ARSI with hormone-sensitive disease and one deep into castration resistance. Twenty-four weeks is also a short window for a toxicity pts live with for years, and nothing in the source ties the CANTAB change to function, falls, or treatment discontinuation, which is where a cognitive signal earns its clinical weight.

For a reader who already held the mechanistic prior, darolutamide's structure and limited blood-brain barrier penetration, this moves the question from plausible to measured, and it moves it in a usable direction: where two drugs are treated as interchangeable on efficacy, cognition becomes a legitimate tiebreaker rather than an anecdote. For the result to be wrong, the open-label design would have to have produced asymmetric test engagement large enough to generate the observed gap between -15.8 and -36.1, which is a strong claim to make about a computerized battery.