Physical activity has become one of the most widely recommended interventions for people with neurodevelopmental disorders, a family of conditions that includes autism spectrum disorder, attention-deficit/hyperactivity disorder, intellectual developmental disorder, and specific learning, communication and motor disorders. Clinical guidelines across the world now endorse exercise as a cost-effective, scalable component of comprehensive care. Yet a sweeping new analysis suggests that the safety evidence underpinning these recommendations is far shakier than the enthusiasm would imply, not because exercise is dangerous, but because researchers have largely failed to look for harm in any systematic way.
The study, published in Sports Medicine – Open, represents the first systematic effort to quantify how adverse events are monitored and reported in randomized controlled trials of physical activity interventions for people with neurodevelopmental disorders. A research team led by Jinrong He and Xueping Wu of Shanghai University of Sport searched PubMed, Web of Science, Scopus, the Cochrane Library and multiple EBSCO databases, including SPORTDiscus and APA PsycInfo, completing their formal search on November 11, 2025, followed by three rounds of supplementary snowball searching. After screening 41,852 deduplicated records with the help of an active-learning screening tool, they included 309 randomized controlled trials encompassing 322 study arms and 13,228 participants with neurodevelopmental disorders.
The headline finding is stark: not a single one of the 309 trials fully complied with the CONSORT-Harms recommendations, the international standard for systematically monitoring and reporting potential harms in randomized trials. Only 58 studies, roughly one in five, described any adverse event monitoring at all, and most of those relied on passive, non-prespecified procedures such as spontaneous participant self-reports. Just 14 trials implemented prespecified monitoring plans, and only one study assessed adverse events during post-intervention follow-up to detect delayed harm. In other words, the vast majority of trials that established the benefits of exercise for this population did so without any structured mechanism for detecting what might have gone wrong.
Among the studies that did report monitoring results, approximately one quarter documented at least one adverse event or adverse effect. The events were overwhelmingly mild. The most common were delayed-onset muscle soreness following resistance and strength training, minor injuries such as falls, a foot sprain, bruising around the knee, and hypoglycemic episodes during combined neuromuscular training, rashes, blisters or hip pain associated with increasing aerobic exercise volume, and psychological discomfort linked to sensory stimulation or environmental adaptation, including distress from wearing sports headphones, overly rapid session progression, and difficulty adjusting to a swimming pool environment. A smaller number of studies reported health problems such as fever, seizures or respiratory infections, but explicitly judged these unrelated to the intervention.
The withdrawal data told a similarly troubling story. Of the 201 studies that reported participant dropout, 133 gave reasons that did not involve adverse events, but 38 used descriptions so vague that the reviewers could not determine whether the withdrawals were harm-related, with phrases like withdrew for personal reasons offering no diagnostic value. Sixteen studies reported withdrawal reasons involving adverse events such as fractures, general health problems, COVID-19 disruptions, or in one case a child’s fear of horses. Critically, none of the included studies described a systematic procedure for adjudicating withdrawal reasons, and the reports generally lacked the temporal information, severity grading and clinical detail needed to link an event causally to the intervention. Most strikingly, four studies reported withdrawals due to health problems while simultaneously stating that no adverse events had occurred, an internal contradiction that illustrates how easily harm data can slip through the cracks.
To estimate risk quantitatively, the team pooled 45 trials that provided usable adverse event data, using a random-effects Mantel-Haenszel model with a treatment-arm continuity correction to handle the many studies in which neither group reported any events. The pooled relative risk was 1.14, with a 95 percent confidence interval of 0.67 to 1.94, indicating that physical activity interventions did not significantly increase the risk of reported adverse events compared with non-exercise controls. Heterogeneity was minimal, at an I-squared of just 2 percent, and sensitivity analyses using alternative pooling methods, including the Battaglia continuity correction and inverse-variance models with restricted maximum likelihood and DerSimonian-Laird estimators, produced nearly identical results. Meta-regression found no significant moderation by participant age, and dose-response analyses across weekly frequency, session duration, total sessions and total training time revealed no statistically significant association between intervention dose and adverse event risk.
But one subgroup analysis produced a result that reframes the entire field. Trials that used prespecified adverse event monitoring procedures showed a significantly higher risk estimate, with a relative risk of 4.32 and a confidence interval of 1.36 to 13.68, and this moderator alone explained 90 percent of the between-study variance in effect sizes. The authors are careful to interpret this correctly: the elevated estimate does not mean that monitored trials carry genuinely greater risk. Rather, it demonstrates that when researchers actively look for harm, through structured training diaries, proactive querying at prespecified time points, physiological monitoring of heart rate and blood pressure, and explicit stopping rules, they find it. Passive monitoring, by contrast, systematically misses mild events, particularly in a population where individuals may have difficulty recognizing, interpreting or communicating internal states such as pain, fatigue or emotional distress.
This detection gap has particular significance for neurodevelopmental populations. Research on pain perception in autism, for example, suggests atypical pain profiles and underestimation of others’ pain, while people with intellectual disabilities may struggle to report subjective symptoms. Relying exclusively on spontaneous self-reports in such groups, the authors argue, likely leads to missed or misclassified events and systematic underestimation of true harm rates. The descriptive patterns also hint at subtype-specific safety profiles that deserve targeted monitoring: adverse events in intellectual developmental disorder predominantly involved physical and medical problems, autism-related reports centered on sensory and environmental adaptation difficulties such as aquatic and equine settings, and the limited ADHD evidence pointed to psychological burden from intervention arrangements. These patterns remain exploratory, but they suggest that a one-size-fits-all monitoring framework may be insufficient.
The authors also identify deeper structural forces behind the underreporting. Some trials never established monitoring frameworks at the protocol stage, leaving data collection without consistent standards. Others collected harm information but failed to document it completely. The literature on conflicts of interest suggests that when favorable conclusions confer academic or professional benefits, researchers may selectively disclose or downplay unfavorable findings, and social desirability bias in exercise research may dull investigators’ sensitivity to risk signals. The absence of mandatory journal standards for harm reporting allows these problems to compound across the publication cycle. The team proposes concrete remedies: prespecified active monitoring embedded in trial protocols, including pre-session health screening, objective physiological indicators during training, and proactive post-session follow-up; intervention-specific surveillance such as tracking sensory responses in equine therapy; and transparent reporting of planned versus delivered intervention dose, since roughly 70 percent of the included studies failed to report exercise intensity at all, making dose-response analysis impossible.
The bottom line for clinicians and families is cautiously reassuring but conditional. The available evidence does not show that exercise increases overall harm risk for people with neurodevelopmental disorders, and the events that are detected are typically mild and manageable. But the authors emphasize that this conclusion reflects a lack of evidence for excess risk rather than evidence of the absence of risk, given pervasive underreporting and methodological heterogeneity. They recommend that guideline developers give greater weight to trials that used prespecified, active harm surveillance, and that future studies adopt standardized monitoring, structured withdrawal adjudication, and complete descriptive reporting of adverse events. Until the field routinely looks for harm with the same rigor it applies to measuring benefit, the true safety profile of exercise for millions of people with neurodevelopmental disorders will remain partly invisible.
Subject of Research: Adverse event monitoring and reporting in physical activity interventions for people with neurodevelopmental disorders
Article Title: Invisible Harms, Visible Benefits? Adverse Event Reporting in Physical Activity Interventions for Neurodevelopmental Disorders: A Meta-analysis and Critical Appraisal
Article References: He, J., Peng, C., Zhang, L., Wang, D., Tan, X., Wen, X., Shen, X., & Wu, X. (2026). Invisible Harms, Visible Benefits? Adverse Event Reporting in Physical Activity Interventions for Neurodevelopmental Disorders: A Meta-analysis and Critical Appraisal. Sports Medicine – Open, 12(1), Article 129. https://doi.org/10.1186/s40798-026-01095-w
Image Credits: AI Generated
DOI: 10.1186/s40798-026-01095-w
Keywords: adverse events, neurodevelopmental disorders, physical activity interventions, meta-analysis, autism spectrum disorder, ADHD, intellectual disability, CONSORT-Harms, randomized controlled trials, exercise safety, harm monitoring, participant withdrawal
Cite Scienmag News
APA MLA Chicago
Ophelia Keating. (September 12, 2026). Exercise Is Safe for Neurodevelopmental Disorders, But Only If Trials Actually Look for Harm. Scienmag. https://scienmag.com/exercise-is-safe-for-neurodevelopmental-disorders-but-only-if-trials-actually-look-for-harm/
Ophelia Keating. “Exercise Is Safe for Neurodevelopmental Disorders, But Only If Trials Actually Look for Harm.” Scienmag, 12 September 2026, https://scienmag.com/exercise-is-safe-for-neurodevelopmental-disorders-but-only-if-trials-actually-look-for-harm/. Accessed 12 September 2026.
Ophelia Keating. “Exercise Is Safe for Neurodevelopmental Disorders, But Only If Trials Actually Look for Harm.” Scienmag. September 12, 2026. https://scienmag.com/exercise-is-safe-for-neurodevelopmental-disorders-but-only-if-trials-actually-look-for-harm/
Copy citation Download RIS
Tags: ADHDadverse event reporting in clinical trialsadverse eventsautism spectrum disordercomprehensive analysis of exercise safety evidenceCONSORT-Harmsevaluating harm detection in physical activity researchexercise guidelines for neurodevelopmental conditionsexercise safetyharm monitoringintellectual disabilitymeta-analysismethodological gaps in neurodevelopmental disorder trialsmonitoring harm in neurodevelopmental disorder researchneurodevelopmental disorder exercise safetyNeurodevelopmental Disordersparticipant withdrawalphysical activity interventionsrandomized controlled trialsresearch quality in neurodevelopmentrisks and benefits of exercise for autism spectrum disordersafety assessment in physical activity studiessystematic review of adverse events in neurodevelopmental disorder interventionssystematic review of exercise interventions for autism and ADHD
