Physical Restraint Disparity Among Autistic Youth Grows With Age

A new letter published in the Journal of Autism and Developmental Disorders is challenging how researchers and clinicians interpret one of the most troubling findings in pediatric emergency medicine: the disproportionate use of physical restraint on autistic children and young people. The letter, authored by Ferah Hira, an independent researcher based in Balıkesir, Turkey, and Serdar Hira of the Department of Clinical Biochemistry at Royal Hospital in Bandırma, Turkey, takes aim at the statistical logic underpinning a recent brief report that claimed the restraint disparity between autistic and non-autistic youth grows steadily as children get older. The critique is technical, but its implications are far-reaching, touching on how evidence of inequity in clinical settings is quantified, communicated, and ultimately acted upon.

The original study at the center of the debate, conducted by McGaughey and colleagues and published in the same journal, examined physical restraint patterns among pediatric emergency department patients with autism spectrum disorder. The researchers stratified their data by age group and reported that the association between autism and restraint differed across those strata, interpreting the pattern as evidence of a progressively increasing disparity across development. That interpretation carries significant weight. If the gap between autistic and non-autistic youth truly widens with age, it would suggest that something about adolescence and young adulthood—perhaps changing behavioral expectations, diminished tolerance for autistic communication styles, or the fading of child-specific accommodations—intensifies the risk of restrictive interventions in emergency settings. The finding resonated because physical restraint is not a neutral procedure; it is associated with physical injury, psychological trauma, and erosion of trust in healthcare systems, particularly among neurodivergent patients.

Hira and Hira’s central argument is that the original analysis committed a well-known but persistent statistical error: inferring that an association differs between groups because it is statistically significant in one group and not in another. The letter points out that separate significance tests within individual age strata cannot establish that the magnitude of the autism-restraint association actually differs between those strata. This distinction matters because a nonsignificant result in one age group is not evidence that no disparity exists in that group. It may simply mean the study lacked the statistical power to detect a real effect, a problem that is especially acute when sample sizes shrink after stratification. The authors draw particular attention to the wide confidence interval reported in the youngest age group of the original study, arguing that this width signals imprecision—an estimate that could plausibly range from a modest to a very large effect—rather than genuine evidence of an absent disparity.

To ground their critique, the letter’s authors invoke a methodological principle that has been drilled into epidemiology and clinical research for decades but is still routinely violated in practice: to test whether an effect differs across levels of a third variable, researchers must formally test the interaction itself. In this case, that means modeling an autism spectrum disorder-by-age interaction term and evaluating whether it is statistically significant, or equivalently, directly comparing the effect estimates and their confidence intervals across age groups. The letter cites the classic expositions of this principle, including Altman and Bland’s 2003 BMJ piece “Interaction revisited: The difference between two estimates” and Matthews and Altman’s 1996 companion paper “Interaction 2: Compare effect sizes not P values,” both of which distilled the lesson that comparing p-values across subgroups is a fundamentally flawed inferential strategy. A p-value reflects both the size of an effect and the precision with which it is estimated, so two subgroups can share nearly identical effect sizes while producing dramatically different p-values simply because one has more data than the other.

The distinction is not merely academic pedantry, the authors argue, because it changes what the study can legitimately claim. If the interaction test has not been performed, then the claim of a progressively increasing disparity across development remains untested. The observed pattern in the stratified estimates could reflect a real age-dependent escalation of the restraint gap, or it could reflect nothing more than sampling variability and differing precision across strata. Until the autism-by-age interaction is formally evaluated, Hira and Hira contend that the idea of an age-dependent progression should be treated as a hypothesis-generating observation rather than an established finding. This framing is careful and, in its way, conciliatory: the letter does not dispute that a restraint disparity exists overall, and it explicitly credits the original study with providing important evidence of restraint inequity in pediatric emergency care. What it disputes is the developmental narrative layered on top of that headline finding.

A second, subtler methodological point in the letter concerns the difference between population-level age patterns and within-person developmental trajectories. The original study was cross-sectional in the relevant sense: it did not follow the same individuals over time, and therefore its age-stratified results describe heterogeneity across the population of emergency department patients at different ages, not the trajectory any individual child experiences as they grow older. This distinction, often framed as the difference between age, period, and cohort effects, is easy to overlook but consequential. Cohort effects could arise, for example, if today’s adolescents were diagnosed and supported under different clinical norms than today’s young children will be a decade from now, or if service structures differ for different age bands in ways that have nothing to do with development per se. A cross-sectional snapshot cannot separate these possibilities. The letter argues that the findings therefore represent population-level age heterogeneity rather than evidence about how restraint risk evolves within individuals as they develop.

The stakes of getting this right are considerable. Physical restraint in emergency departments is among the most contentious interventions in pediatric care, and autistic youth are disproportionately subjected to it, often in the context of sensory overload, communication barriers, co-occurring psychiatric presentations, or behaviors misunderstood as aggression. Advocacy organizations and health services researchers have increasingly called for trauma-informed, neurodiversity-affirming approaches to emergency care, and robust evidence about when and for whom restraint risk is highest is essential for targeting those interventions. If the disparity genuinely escalates with age, adolescent and young adult emergency services would warrant particular scrutiny, and transition-age supports might become a priority. If, however, the apparent escalation is a statistical artifact of subgroup analysis, resources might be misdirected, and a falsely reassuring conclusion—that the disparity is stable or even smaller in young children—could take root despite imprecise estimates suggesting otherwise.

The exchange also illustrates a broader and recurring problem in the translation of statistical results into public health narratives. Stratified analyses are seductive because they generate seemingly concrete, humanly legible stories: the gap doubles in teens, the gap is absent in young children. But each stratum typically contains a fraction of the total sample, and precision collapses accordingly. Wide confidence intervals in small strata are frequently misread as “no effect,” when the honest reading is “we do not know, and the data are compatible with a large effect.” The letter’s authors emphasize that the wide confidence interval in the youngest group indicates imprecision rather than evidence that no disparity is present—a reading that directly reverses the most consequential interpretation in the original report. In this sense, the critique echoes a long tradition of methodological commentary in the BMJ and other journals warning that the difference between “significant” and “not significant” is not itself statistically significant, a phrase that has become something of a mantra in quantitative research training.

It is worth noting that the critique arrives in the form of a letter to the editor, a genre that plays an underappreciated role in scientific self-correction. Letters of this kind do not present new data; the Hiras state explicitly that no new data were created or analyzed for their work. Instead, they scrutinize the inferential logic of published claims, and journals publish them when the logical flaw is significant enough to affect how the field should read the original findings. The letter reports no funding, declares no conflicts of interest, and was handled through the journal’s standard editorial process, received on August 6, 2026, accepted on September 1, 2026, and published on September 10, 2026. Its brevity belies its potential influence: if the original authors respond with a formal interaction analysis, the field will either gain a properly tested, potentially important finding about age-dependent restraint risk, or a corrected understanding that the disparity, while real, does not yet have demonstrated age-dependent progression.

For clinicians on the front lines of pediatric emergency medicine, the practical takeaways of the exchange are twofold. First, the overall message of restraint inequity stands: autistic youth presenting to emergency departments are restrained at higher rates than their non-autistic peers, and this alone justifies urgent attention to communication supports, sensory accommodations, and de-escalation training. Second, claims about how that inequity changes with age should be held more loosely than initial coverage of the original study may have suggested. The Hiras’ letter does not diminish the urgency of the problem; if anything, it sharpens the scientific standard that evidence used to justify interventions must meet. As the authors conclude, the study provides important evidence of overall restraint inequity, while the proposed age-dependent progression should remain hypothesis-generating until formally tested. In a field where statistical claims can shape clinical guidelines, training priorities, and public understanding of a vulnerable population, that standard is not a technicality—it is the difference between evidence and narrative.