# Autism

Autism across the lifespan: varied developmental pathways, limits of biological explanations, communication and practical support, and treatment of specific co-occurring difficulties.

Information, not medical advice. Discuss personal concerns with a qualified health professional.

[Watch the film](https://www.youtube.com/watch?v=Z3Zteqc_Y8c) · [All films and essays](https://evidencepress.org/why-minds-break/films/)

Autism describes a varied group of developmental patterns affecting communication, social interaction, interests, routines and responses to sensory experiences. The current diagnosis includes people who would previously have received an Asperger diagnosis. Some autistic people speak fluently and live independently; others have little reliable speech, intellectual disability or substantial lifelong care needs. A shared diagnosis does not imply one cause, one kind of brain or one appropriate treatment. [Overview of autism](https://doi.org/10.1038/s41572-019-0138-4)

Research supports a strong genetic contribution, involving several routes through early brain development, alongside some established prenatal influences. It has not traced a complete chain from cause to brain changes to everyday experience for every person. Support can nevertheless address communication, distress, health and participation. Its value should be judged by meaningful benefits to the person, rather than by whether someone appears less visibly autistic.

## Genes contribute in several different ways

A study of more than two million people across five countries estimated autism's heritability at about 81%, with an uncertainty range of roughly 73–86%. Heritability describes how much variation in susceptibility within the studied populations was associated with inherited differences. It does not mean that 81% of an individual's autism is genetic, that researchers can identify the cause in 81% of people, or that the remaining percentage must be environmental toxins. High heritability also does not make outcomes fixed or support pointless. [Five-country family study](https://doi.org/10.1001/jamapsychiatry.2019.1411)

Many common genetic differences each make a very small contribution. Rarer changes can have larger effects. Some are inherited; others arise in an egg, sperm or early embryo. Sometimes a stretch of genetic material has been deleted or duplicated. These influences can combine in the same person, and many also relate to other developmental conditions. There is no single autism gene. Identifying a genetic condition such as fragile X syndrome or tuberous sclerosis can help explain additional medical needs without providing a universal explanation of autism. [Research on common and rare genetic contributions](https://doi.org/10.1038/s41588-022-01064-5); [American Academy of Pediatrics clinical report](https://doi.org/10.1542/peds.2019-3447)

Some studies find a greater measurable burden of autism-associated genetic changes among diagnosed females. One possible explanation is that females, on average, require more of these influences to cross a developmental or diagnostic threshold. This is sometimes called a protective-effect model. It does not fully explain differences in diagnosis between sexes: recognition, referral and the way autism is measured also matter. A single hormonal or chromosomal explanation is not established. [Study of genetic contributions and sex differences](https://doi.org/10.1038/s41588-022-01064-5)

Clinical diagnosis and investigation of a possible cause answer different questions. Diagnosis draws on developmental history, current experiences and behaviour. Routine brain scans or biological tests do not establish it. Selected genetic or medical investigations may be appropriate when there are particular clinical indications; a negative genetic result does not rule autism out. [NICE guidance on assessment](https://www.nice.org.uk/guidance/cg142/chapter/Recommendations)

## Prenatal influences: separating causes from associations

Exposure to valproate during pregnancy is among the clearest established environmental contributors. A review of eight cohort studies found an adjusted rate of autism diagnosis about three times higher after prenatal exposure, with estimates ranging from about 2.2 to 4.3 times. Higher doses, analyses of valproate used alone and comparisons between siblings supported the concern. This is evidence about a particular exposure, not an explanation for autism generally. Anyone taking valproate who is pregnant or planning pregnancy needs prompt specialist advice; abruptly stopping prescribed treatment without medical guidance can also be dangerous. [Review of prenatal valproate exposure](https://doi.org/10.1111/acps.13797)

Relative comparisons need context. In one study of children whose mothers had epilepsy, autism had been diagnosed by age eight in about 10.5% exposed to valproate and 4.2% with no prenatal antiseizure medication exposure. That is a difference of about six children per hundred. These were unadjusted figures for particular groups, not a universal causal risk estimate. The study's adjusted analysis also found an elevated rate, about 2.7 times higher. [Study of prenatal antiseizure medicines](https://doi.org/10.1056/NEJMoa2309359)

Certain infections reaching the developing fetus, including congenital rubella and cytomegalovirus, can affect brain development. These are different from an ordinary maternal illness. A 2026 Swedish study examined 975 children with recorded congenital infections within a birth population of approximately 3.7 million. A sibling comparison still found a roughly threefold association with autism, but it concerned a small, distinct group. It cannot be extended to every infection during pregnancy. Very premature birth is also associated with autism; family comparisons only modestly weakened that association. Exactly how much reflects prematurity itself, its causes or accompanying complications remains uncertain. [Congenital-infection study](https://doi.org/10.1001/jamapediatrics.2026.4229); [study of prematurity](https://doi.org/10.1542/peds.2020-032300)

Older parental age, maternal diabetes, air pollution, low birthweight and severe birth complications have also been associated with autism. The causal evidence varies. Older parental age, for example, can involve genetic and reproductive factors rather than a purely environmental exposure. Families can share inherited characteristics, health conditions and circumstances that influence both an apparent exposure and a child's development.

A Danish study illustrates this problem. It examined 236 maternal diagnoses in more than 1.1 million children and initially found 30 associations with autism. Many became weaker when researchers compared siblings or used paternal diagnoses as a check on family-related explanations. That does not make all prenatal influences unreal; it shows why a correlation alone cannot establish what caused autism. Family comparisons are not perfect controls: siblings have different circumstances, exposure measurements may be inaccurate, and rare exposures can leave few informative comparisons. Similarly, a 2026 review of paracetamol in pregnancy found no increased autism association in sibling comparisons: the estimate was 0.98, with an uncertainty range of 0.93–1.03, where 1 means no difference. The better-controlled evidence does not support a clinically important increase from use as directed. [Study of maternal diagnoses and family influences](https://doi.org/10.1038/s41591-024-03479-5); [paracetamol review](https://doi.org/10.1016/S3050-5038%2825%2900211-0)

Vaccines do not cause autism. The World Health Organization's evidence review, published in September 2026 and covering research through August 2025, found that the strongest studies did not support a causal association, including for vaccines containing thiomersal. Parenting style is not an established cause either. A responsive, supportive environment can improve wellbeing without implying that parents caused their child's developmental differences. [WHO vaccine evidence review](https://www.who.int/publications/i/item/9789240123533); [WHO autism information](https://www.who.int/news-room/questions-and-answers/item/autism-spectrum-disorders-%28asd%29)

## What research suggests about the developing brain

Different genetic findings converge on several broad processes: controlling when and where genes are active, developing and maturing nerve cells, and building the connections through which cells communicate. A major sequencing study involving 35,584 samples, including 11,986 autistic people, identified 102 associated genes at its chosen statistical threshold. That was a study finding, not a complete current list or a set of genes that must be changed in every autistic person. The work suggests shared biological processes across different genetic routes, while leaving much of the path to particular experiences unexplained. [Large genetic sequencing study](https://doi.org/10.1016/j.cell.2019.12.036); [review of molecular mechanisms](https://doi.org/10.1002/ana.70013)

One proposed mechanism concerns the balance between signals that encourage nerve cells to become active and signals that restrain their activity. Glutamate and GABA are chemical messengers involved in these processes. A review of brain-spectroscopy studies found lower average GABA measurements in some autistic groups, particularly children and some regions involved in emotion. Results varied by age, region and method. Measuring the amount of a chemical across a volume of tissue is not the same as measuring moment-to-moment signalling between cells. These findings do not establish one chemical imbalance shared by all autistic people. [Review of brain chemical measurements](https://doi.org/10.1016/j.neubiorev.2024.105728)

Connections between brain regions also vary considerably. A 2026 study combined 1,824 people, including 796 autistic participants, from 32 sites. No single connection was unusually different in more than 4% of either group. Individual patterns nevertheless showed some convergence when considered at the level of broader regions or networks. That combination—considerable individual variation with some shared organisation—is more informative than a claim that all autistic brains are uniformly overconnected or underconnected. It does not provide a diagnostic scan. The broad genetic and developmental explanation would remain without these newest findings. They refine that account; they do not establish a single mechanism that every older, conflicting result must eventually fit. [Study of individual brain-connectivity patterns](https://doi.org/10.1038/s44220-026-00656-y)

Research on brain tissue after death has found differences in the activity of genes linked to nerve-cell connections and glial cells, which help support and regulate the brain's environment. Such findings raise questions about immune-related and support processes. But tissue collected after years of development cannot establish whether a difference was an early cause, a later adaptation or a consequence of other experiences. It does not show that autism generally is an inflammatory or autoimmune disease. [Study of gene activity across the brain's outer surface](https://doi.org/10.1038/s41586-022-05377-7)

The gut offers another example of the direction-of-causation problem. In a study of 247 people, selective eating and diet helped explain differences in gut microbes associated with autism. A later correction did not remove that central interpretation. The findings do not rule out every role for the gut, and gastrointestinal problems deserve care. They do show why discovering a different stool-microbe pattern does not establish that it caused autism or that changing it will treat autism. [Study of diet and gut microbes](https://doi.org/10.1016/j.cell.2021.10.015); [published correction](https://doi.org/10.1016/j.cell.2023.12.001)

## Thinking, perception and social understanding

Several psychological theories describe aspects of autistic experience: differences in interpreting other people's mental states, organising behaviour towards goals, or balancing details with wider context. None explains every autistic person. A newer approach examines how the brain combines expectations with incoming information. A review of 23 studies found a modest average difference in experimental measures of this process, about 0.37 standard deviations. A standard deviation describes the spread of scores: 0.37 is a comparison between group averages, not a 37% change in an individual. The studies varied substantially, and the result did not support one simple account of how all autistic people make predictions. [Review of prediction and perception](https://doi.org/10.1007/s11065-025-09672-8)

The “double empathy” idea draws attention to mutual misunderstanding between people with different experiences, rather than assuming that communication difficulty resides entirely within the autistic person. It is useful to distinguish this broad perspective from particular testable claims. In a preregistered study of 311 people, mixed autistic and non-autistic groups did not show the predicted disadvantage on a specific information-transfer task. This limits a strong claim about that task; it does not settle every question about understanding in daily relationships. Biological explanations and the social conditions of an interaction address different levels of the same experience. [Replication study of communication between groups](https://doi.org/10.1038/s41562-025-02163-z)

Researchers are also asking whether broad diagnoses can be divided into more informative developmental profiles. A 2025 study of 5,392 children described four groups with different patterns of traits and genetic associations, with support from additional analyses. This does not establish that exactly four natural kinds of autism exist. Genetic analyses in the study were also restricted in ancestry. Groups depend on which children, measures and statistical methods are used. To help care, a proposed grouping must reliably predict needs or responses to support in new populations. [Study of developmental subgroups](https://doi.org/10.1038/s41588-025-02224-z)

## Communication support and early learning

There is no single best programme for every autistic person, and no medicine has a generally reliable effect on autism's core developmental features. Effective care starts with specific goals: reliable communication, greater independence, less distress, improved health or easier participation in family, education and community life.

For young children, NICE recommends developmentally appropriate, play-based social-communication support involving parents, carers or teachers. The aim is to help communication partners recognise the child's signals and respond in ways that encourage shared attention, engagement and reciprocal communication. Developmental and behavioural approaches often overlap. Programmes described as naturalistic developmental behavioural interventions, including the Early Start Denver Model, combine learning principles with play and everyday interaction; they are not a completely separate category from all approaches labelled ABA. [NICE guidance for children and young people](https://www.nice.org.uk/guidance/cg170/chapter/recommendations)

The Project AIM review covered 252 studies and 13,304 children. Average benefits were generally small. In randomised trials, naturalistic developmental behavioural approaches improved social-communication measures by about 0.35 standard deviations, with an uncertainty range of 0.23–0.47. This describes a shift in scores, not children becoming “35% less autistic”. When the reviewers applied their strictest safeguards against biased measurement, convincing evidence narrowed substantially: among these approaches, only the effect on measured diagnostic characteristics remained statistically clear. That does not prove every other benefit absent; it limits how confidently broader promises can be made. [Project AIM intervention review](https://doi.org/10.1136/bmj-2023-076733)

Longer follow-up is encouraging in some respects but also specific. The PACT parent-mediated communication study followed 121 of its original 152 children approximately six years after treatment. Analysis across treatment and follow-up favoured the intervention on measured symptom severity. The estimate at follow-up alone was imprecise, and a composite language measure did not show a clear group difference. A result in one domain should not become a promise of improvement in all domains. [PACT follow-up study](https://doi.org/10.1016/S0140-6736%2816%2931229-6)

People whose speech is absent, unreliable or insufficient need a dependable way to communicate. Pictures, signs and speech-generating devices are forms of augmentative and alternative communication, often shortened to AAC. A review of 29 studies found promising language results when AAC was included in naturalistic developmental support, but only three studies made the most directly relevant comparison, and designs varied. The practical aim is communication that lets the person express needs, interests, choices and refusal, not speech at any cost. [Review of AAC within communication interventions](https://doi.org/10.1007/s10803-024-06382-7)

## Behavioural support, intensity and the person's goals

Applied behaviour analysis, or ABA, is a broad family of methods based on how behaviour changes with experience. Teaching a way to request a break, providing a comprehensive early-learning programme and trying to suppress harmless movements are different interventions with different goals. A verdict on the label alone is therefore too broad.

A review using individual data from ten studies, covering 491 children, found that intensive ABA-based programmes improved everyday adaptive-skill scores by about seven points after two years, with a plausible range of roughly two to twelve points. Some cognitive scores also improved. However, all studies had important risks of bias, results varied and longer-term evidence was limited. These findings cannot identify one universally superior programme. A separate review of 144 studies and 9,038 children found no robust relationship, within intervention types, between better outcomes and more daily hours, longer duration or more total treatment. Since amounts were not generally randomly allocated, that does not prove intensity never matters; it does undermine a universal prescription for a fixed number of hours. [Review of intensive behavioural programmes](https://doi.org/10.1177/1362361320985680); [review of intervention amount](https://doi.org/10.1001/jamapediatrics.2024.1832)

Targeted parent training can help with disruptive behaviour. In a trial of 180 children aged three to seven, 68.5% receiving the training were judged improved by clinicians unaware of allocation, compared with 39.6% receiving parent education. Yet differences on the two main parent-rated scales did not reach the trial's prespecified threshold for clinical importance. This supports possible benefit while preserving uncertainty about its magnitude. Useful support should be timely: uncertainty about a universal programme is not a reason to leave an identified communication or daily-living need unmet. [Parent-training trial](https://doi.org/10.1001/jama.2015.3150)

Before treating a behaviour as a problem in itself, assess what may be driving it: pain, constipation, anxiety, communication barriers, sensory overload or disruption to a familiar routine. Helping someone communicate discomfort or escape an overwhelming situation may address the need more directly. Support should respect agency; suppressing a harmless repetitive movement or increasing eye contact is not, by itself, evidence of improved health or quality of life. [NICE priorities for individual support](https://www.nice.org.uk/guidance/cg170/chapter/key-priorities-for-implementation); [Australian supporting-children guideline](https://www.autismcrc.com.au/best-practice/supporting-children/guideline)

## Treating anxiety, sleep problems and ADHD

Specific accompanying conditions have their own treatments. Adapted cognitive behavioural therapy, or CBT, can help anxiety in autistic young people who can engage with its language and thinking tasks. Adaptations may include concrete explanations, visual or written materials, predictable sessions, breaks and involvement of a chosen supporter. Across 19 randomised trials involving 833 young people, immediate benefits were larger on clinician ratings than parent ratings and smaller again on the young people's own ratings: approximately 0.88, 0.40 and 0.25 standard deviations. Sustained benefit at follow-up was not established. The differences between raters and limited follow-up matter when setting expectations. [Review of adapted CBT for anxiety](https://doi.org/10.1186/s40359-021-00658-8)

Depression, obsessive–compulsive disorder and other mental health problems also deserve appropriate care, adapted to communication and sensory needs. Some recommendations draw on evidence from the wider population, so they should not be presented as though every adaptation has extensive autism-specific trial support. [NICE guidance for autistic adults](https://www.nice.org.uk/guidance/cg142/chapter/Recommendations)

For sleep difficulties, guidance recommends checking health conditions and medicines, then developing an individual sleep plan. Melatonin may be considered with professional supervision when these steps are insufficient; product quality, unwanted effects and uncertain long-term safety deserve attention. In a trial of 125 children whose sleep had not improved sufficiently with behavioural support, prolonged-release melatonin added about 32 minutes of sleep a night compared with placebo and improved time to fall asleep. Around 69% met the trial's response definition, compared with 39% on placebo: about one additional responder for every three to four treated. These were short-term, largely caregiver-reported results from a particular, industry-supported trial, not a guarantee for every child. [Sleep-management guideline](https://doi.org/10.1212/WNL.0000000000009033); [melatonin trial](https://doi.org/10.1016/j.jaac.2017.09.414)

When an autistic person also has ADHD, methylphenidate or atomoxetine may help ADHD symptoms; there is a smaller evidence base for guanfacine. Certainty in these trials is low or very low, and individual dose adjustment and monitoring remain necessary. Treating attention or impulsivity difficulties is different from treating autism itself. [Review of ADHD medication in autism](https://doi.org/10.1111/jcpp.13305)

## Medicines for severe irritability

Risperidone and aripiprazole have the most replicated short-term medication evidence for severe irritability in autistic children. A Cochrane review included 17 trials with 1,027 participants, but only one small adult study. On a commonly used irritability scale, average improvements relative to placebo were about eight points for risperidone and six for aripiprazole. The reviewers rated the evidence as uncertain. Findings on the broad irritability scale do not establish equally strong effects on aggression or self-injury considered separately, and paediatric results cannot simply be extended to adults. [Cochrane antipsychotic review](https://doi.org/10.1002/14651858.CD014965.pub2)

These medicines can cause sleepiness, weight gain, metabolic changes and movement problems. NICE recommends specialist consideration when other interventions are insufficient or the behaviour is too severe for them to be delivered. Treatment needs a defined target and a way to measure benefit and adverse effects. Guidance calls for review after three to four weeks and stopping under professional supervision if there is no clinically important response by six weeks. Guideline use and a medicine's licence are different: current UK aripiprazole product information does not establish a paediatric autism-irritability dose. [NICE treatment and review recommendations](https://www.nice.org.uk/guidance/cg170/chapter/recommendations); [UK aripiprazole product information](https://www.medicines.org.uk/emc/product/13576/smpc)

## Daily life, adulthood and sensory needs

Clear communication, predictable arrangements, suitable sensory environments and practical help with daily activities can make participation easier. Adults may benefit from structured skills support or supported employment matched to their goals. Guidance combines research with clinical judgement and accessibility principles; not every sensible adjustment has been tested in a large randomised trial. [NICE adult support guidance](https://www.nice.org.uk/guidance/cg142/chapter/Recommendations)

Adult intervention research is growing but remains uneven. A 2026 review of 35 studies and 1,631 participants found a small average mental-health benefit, about 0.35 standard deviations, and encouraging findings for social cognition and quality of life. Quality-of-life estimates were variable and imprecise, while pooled effects on adaptive functioning and employment were not statistically clear. Another review of 69 studies found 42 were pilot studies, and only 29 used measures designed or standardised for autistic people. These mixed interventions and preliminary designs cannot establish one best adult programme or which adaptations produce the benefits. [Adult intervention meta-analysis](https://doi.org/10.1037/ccp0001003); [review of adult psychosocial support](https://doi.org/10.1177/23969415261436238)

Accommodating painful noise or light is also different from claiming that a sensory therapy improves broad functioning. The SenITA trial recruited 138 children, with 106 in its main analysis. Sensory integration therapy did not outperform usual care on its main behavioural outcomes at six or twelve months. Families reported progress towards individual goals, but corresponding goal change was not measured in the comparison group, so it could not establish a treatment effect. This finding does not make sensory distress unimportant; it limits claims for that particular therapy and its measured outcomes. [SenITA trial](https://doi.org/10.3310/TQGE0020)

## Approaches that are not established treatments

NICE advises against chelation, secretin, hyperbaric oxygen and exclusion diets such as gluten-free or casein-free diets for autism's core features. A medically indicated diet for coeliac disease or another diagnosed condition is a separate matter. [NICE recommendations on unsupported treatments](https://www.nice.org.uk/guidance/cg142/chapter/Recommendations)

Leucovorin needs a particularly careful distinction. In March 2026, the US Food and Drug Administration approved it for cerebral folate transport deficiency associated with a confirmed FOLR1 genetic variant. This is a rare condition affecting transport of folate, a vitamin, into the brain; some affected people have autistic features. The approval is not for autism generally. The American Academy of Pediatrics does not recommend routine leucovorin use in autistic children: studies are small, selected groups may differ, and who benefits, treatment protocols and longer-term safety remain uncertain. [FDA approval and its specific scope](https://www.fda.gov/news-events/press-announcements/fda-approves-first-treatment-patients-cerebral-folate-transport-deficiency); [AAP guidance on leucovorin](https://www.aap.org/en/patient-care/autism/use-of-leucovorin-in-autistic-pediatric-patients/)

Facilitated communication, where another person physically guides or influences pointing or typing, is also distinct from legitimate AAC. Its claimed authorship has not been reliably established, and facilitator influence is a serious concern. Communication support must preserve the autistic person's own authorship, choices and ability to refuse. [Royal College of Speech and Language Therapists position](https://www.rcslt.org/members/delivering-quality-services/facilitated-communication/)

## Measuring progress honestly

Learning a taught skill, improving a language score, managing daily life and feeling better are different outcomes. Study groups may differ before treatment begins. Parents and therapists often know which intervention is being used, which can influence ratings; measures closely tied to taught activities may overstate improvement elsewhere. Strict analyses reduce some of these problems but leave fewer studies. Neither ignoring bias nor dismissing all imperfect research helps families make useful decisions.

Harms also need better measurement. A review of 150 intervention reports found only 11 mentioned adverse events. This does not show that all interventions are harmful or safe; it shows that the question was inadequately recorded. Time, fatigue, distress, cost and effects on autonomy matter alongside measured benefits. Programme developers may evaluate their own interventions, and short follow-up or missing participants can further distort estimates. [Review of adverse-event reporting](https://doi.org/10.1177/1362361320965331)

Much research underrepresents older adults, people with intellectual disability, people who do not use speech, and families with limited access to services. Findings from verbally fluent adults cannot automatically describe people requiring substantial support. Respect for autistic identity must coexist with recognition of severe disability and the need for intensive practical care when appropriate, including help with communication, eating, personal care and safety. [Review of severe developmental disabilities](https://doi.org/10.1136/bmj-2025-086562)

Better biological research needs to follow development from before diagnosis, linking genetic, prenatal, cellular, brain and behavioural measurements in the same people. It needs improved representation across ancestry, sex and age, and independent tests of proposed subgroups. Stronger tests of prenatal influences should measure dose and timing carefully and compare results from designs with different weaknesses. Small samples, flexible analysis and selective publication also need attention. Laboratory models are useful only when their predictions can be checked in people. Better support research needs fair comparisons of approaches and amounts of support, with comparable contact time, independent assessors, longer follow-up, meaningful everyday outcomes, costs, and systematic attention to benefits and harms. A promising biological test or subgroup needs independent evidence that it reliably predicts benefit before it can guide treatment. Its central question should remain concrete: does this help this person communicate, manage pain, sleep, learn, work, participate or live with less distress? An individual plan should specify a few such goals, match each to appropriate support and set a review point. Age, communication, daily-living abilities, accompanying conditions, previous responses and preferences all matter, and needs can change. Review should ask whether benefits appear outside sessions and whether the demands on the person and family are acceptable. Progress deserves recognition when it improves life, whether or not it changes how autistic someone appears.

---

AI-assisted adaptation of supplied research; not independently clinically reviewed. Original essay text: CC0 1.0. Published 7 November 2026.