A Second Pooled Analysis Just Found Hours Did Almost Nothing. Age at Start Did.

Published in Molecular Autism on August 4, 2026, Mandelli and colleagues pooled individual participant data from 11 datasets — 582 autistic children, 121 of them female — and after propensity score matching analyzed 484. The datasets aggregate into nine studies: one randomized controlled trial, five controlled-group designs, and three uncontrolled pre-post designs.

What it found

• Cumulative intensity was largely null. The authors report it "had largely no effect on most MSEL outcome variables," with one exception, a significant effect on MSEL receptive language.
• On one measure, more was associated with less. For adaptive behavior they report a small main effect on VABS Socialization, in which scores decreased as intensity increased.
• Two things did predict outcome. "Earlier age at intervention start predicted more positive outcomes, while higher pre-intervention DQ predicted accelerated growth."
• Several expected moderators did not. Sex showed no effect surviving correction. Nor did intervention type, ESDM against non-ESDM.

The limits the authors set on it

They are unusually direct. Data "was contributed on a voluntary basis and only 1 study was contributed from an RCT," so findings "may be prone to self-selection bias that may limit generalizability." The non-ESDM comparison lumps together heterogeneous approaches. With roughly 60 females per intervention type, power to test sex differences is limited. And on intensity specifically: "These null effects must be taken with caveat that intensity was not a variable that could be systematically manipulated and evaluated."

That last sentence is the one to carry. Nobody randomized children to more or fewer hours. The analysis observed the hours children happened to receive, and children who receive more hours differ from those who receive fewer in ways that are hard to remove.

Why this matters more than either paper alone

In January, an individual participant data meta-analysis in Autism Research reported effect sizes from 0.66 to 1.36 and named treatment intensity the strongest group-level predictor of outcome across measures. In August, a mega-analysis using the same broad method found intensity almost entirely null. Both state serious design limitations. Neither is a randomized test of dosage.

Seven months apart, the field now holds two pooled analyses pointing opposite ways on the single variable that authorizations are written in. A payer's medical director looking for evidence that high-hour authorizations are not supported now has a paper. So does the practice arguing the opposite.

What you must know or do

Know that both papers exist, and what separates them. If a denial or a policy cites "the evidence" on intensity, the useful question is which pooled analysis, on which outcome measures, and whether dosage was randomized. In both cases it was not. That is not a debating point; it is the actual state of the evidence, and it is a stronger position than asserting a number.

Treat the finding both papers can support as the operational one. Earlier age at intervention start predicted better outcomes here. That is a question about intake, referral response and waitlist order — things a practice controls directly — rather than about dosage, which is increasingly decided by someone else. If children wait months between referral and first session, that delay is now an evidence question and not only a scheduling one.

Do not let your own record depend on the literature. The two papers will be cited against each other for years. Progress data showing movement against a baseline for the specific child at the specific intensity is what survives either citation.