Intended for healthcare professionals

Clinical Review State of the Art Review

Linking risk factors and outcomes in autism spectrum disorder: is there evidence for resilience?

BMJ 2020; 368 doi: (Published 28 January 2020) Cite this as: BMJ 2020;368:l6880
  1. Mayada Elsabbagh, PhD
  1. Montreal Neurological Institute, Azrieli Centre for Autism Research, McGill University, Montreal, Canada
  1. Correspondence to M Elsabbagh mayada.elsabbagh{at}


Autism spectrum disorder (referred to here as autism) is one of several overlapping neurodevelopmental conditions that have variable impacts on different individuals. This variability results from dynamic interactions between biological and non-biological risk factors, which result in increasing differentiation between individuals over time. Although this differentiation continues well into adulthood, the infancy period is when the brain and behavior develop rapidly, and when the first signs and symptoms of autism emerge. This review discusses advances in our understanding of the causal pathways leading to autism and overlapping neurodevelopmental conditions. Research is also mapping trajectories of brain and behavioral development for some risk groups, namely later born siblings of children with autism and/or infants referred because of developmental concerns. This knowledge has been useful in improving early identification and establishing the feasibility of targeted interventions for infant risk groups before symptoms arise. However, key knowledge gaps remain, such as the discovery of protective factors (biological or environmental) that may mitigate the impact of risk. Also, the dynamic mechanisms that underlie the associations between risk factors and outcomes need further research. These include the processes of resilience, which may explain why some individuals at risk for autism achieve better than expected outcomes. Bridging these knowledge gaps would help to provide tools for early identification and intervention that reflect dynamic developmental pathways from risk to outcomes.


Various forms of autism affect 52 million people worldwide.1 Currently, the condition is defined based on social and communication impairment, and repetitive and restrictive behaviors that can vary in individuals along a continuum of severity, sometimes accompanied by additional disorders, eg, intellectual and/or language impairment.2 A diagnosis of autism can be made as early as 18-24 months of age, when characteristic symptoms begin to distinguish those affected from typical developing children and those with other developmental conditions.3

Our understanding of the causes of autism has evolved over time. In the 1950s, autism was blamed on “refrigerator mothers,” who did not show affection to their children. Since then, this view has been rejected and a biological framework adopted instead. Scientists initially searched for a single cause, for example an underlying gene or brain region that can explain all or most cases of autism. This search for underlying causes advanced the discovery of risk factors without producing “litmus tests” that can identify the condition at any point in development, or treatments that work for everyone. Instead, research has identified numerous genetic and non-genetic factors that interact over time leading to heterogeneous outcomes across individuals.

Consequently, a model of complex causes has been adopted to map risk factors to outcomes over the course of development:

(1) The same risk factors that increase susceptibility for autism also increase risk for a wider range of neurodevelopmental disorders.456 Some of these conditions emerge early on, eg, attention deficit/hyperactivity disorder (ADHD) and language disorders, while others emerge later in life, eg, schizophrenia and depression.

(2) Risk factors work together to modify brain development from very early on in life, resulting in the reorganization of neural networks that underlie cognition and behavior.7

(3) Altered development in neural systems alters sensitivity to and learning from environmental inputs.78

(4) Individuals with autism vary in their developmental trajectories across multiple behavioral dimensions, including diagnostic features, cognition, and adaptive skills. Consequently, the impact of autism also varies across their lifespan. Some individuals can lead independent and fulfilling lives, but many develop medical, educational, and social difficulties that have a serious negative effect on their quality of life from infancy to adulthood.9

This review integrates evidence from autism research linking underlying risk factors with developmental outcomes (neural and behavioral) in infancy. Most functional brain networks emerge before age 3 when brain plasticity is at its peak. During this developmental period the earliest signs and symptoms of autism can be directly identified and monitored as they emerge, and it corresponds to a time at which they are likely responsive to intervention. The review also covers what we know about early life resilience: a developmental process characterizing individuals who achieve better than expected outcomes.

Sources and selection criteria

We conducted a PubMed search for publications up to 2019, using the key terms “risk,” “risk factors,” “genetic risk,” “environmental risk,” “protective factors,” “resilience,” “development,” “infant,” “developmental pathways,” “outcomes,” “intermediate-/endo-phenotype,” and “early intervention” in combination with the terms relevant for autism spectrum disorder, including autistic disorder, Asperger’s syndrome, and pervasive developmental disorders. We prioritized articles based on their relevance to the topics covered in the manuscript and narrowed down to literature relevant for early development in autism before three years of age. Where available, we prioritized systematic reviews and meta-analyses for selection, and we selectively reviewed relevant references cited in those articles, guided by the topics relevant for the current review. For topics such as prodromal intervention where the number of original studies was relatively small, we reviewed all literature.


The last systematic review of global prevalence conducted in 2012 showed that autism spectrum disorders affect 1-2% of children, with a higher prevalence in boys.10 Estimates vary widely within and across geographic regions. Many more prevalence estimates have recently become available from previously under-represented regions,11 eg, Poland (0.35%, 95% confidence interval 0.34 to 0.69),12 Ecuador (0.19%, 95% confidence interval 0.1 to 0.38),13 Uganda (1.2%, 95% confidence interval 0.85 to 1.82),14 Qatar (1.14, 95% confidence interval 0.89 to 1.46),15 but again, with considerable variability.11

Since the first epidemiological survey in 1966,16 time trends of increasing prevalence have been well documented within and across diverse geographic regions.10 Most reasons for the increase in prevalence are uncontroversial and include the broadening of diagnostic boundaries,17 increased diagnosis of females (albeit still with an uneven ratio), and increased diagnosis in milder forms of autism,18 substantial increase in public awareness, and public health response globally.19 Diagnostic substitution, ie, relabeling of cases with the refinement of diagnostic criteria has also been proposed,20 but its impact on time trends remains uncertain.21

Although claims exist about possible biological or environmental risk factors to explain time and geographic variability in prevalence, they often involve non-causal associations. By contrast, epidemiological data have highlighted health disparities that account for at least some of the observed variation. The most comprehensive evidence comes from the United States Centers for Disease Control and Prevention, and shows that prevalence across ethnic groups in the US has been a moving target. The pattern of change suggests a “catch up” in diagnosis among minorities who were initially underdiagnosed.22 Further, a positive socioeconomic status gradient in autism, ie, higher income being associated with higher autism rates, has been found in some countries (eg, US,23 Australia24), but not in others (eg, Sweden,25 France26) and therefore cannot be attributed to differences in access to care. Although evidence suggests a variation of prevalence according to immigration status, both higher and lower prevalences among children born to immigrants have been reported.27 Many factors have been proposed to explain this variation but without strong evidence to support them. Such factors include race and ethnicity, region of birth, access to care, possible exposure to adverse events that led to migration, and experience of stigma in the new country.28

In summary, progress has been made, especially in recent years, in estimating the prevalence of autism and in surveillance of autism over time; knowledge that continues to be critical in informing the global public health policy response.29 More evidence is still needed to substantiate inference that variability in prevalence estimates might reflect contribution of biological and/or environmental risk factors. This would improve ascertainments of at-risk groups. More evidence is also needed to further address how social determinants modify help-seeking behavior, access to care, clinical presentation, and/or outcomes across the lifespan. This knowledge would inform the design of early identification strategies that have potential in reducing ethnic or socioeconomic disparities where they exist.

Defining risk and resilience in autism

In conditions with complex causes, including autism, a direct causal relationship between risk, protective factors, and outcomes is difficult to establish. Each individual’s biological, environmental, and social characteristics shape development over time, and outcomes are therefore variable across individuals. To account for this complexity, the relationship between risk/protective factors and outcomes can be conceptualized as a landscape, where two individuals who share the same attributes may respond in different ways and thus take alternative trajectories toward distinct outcomes.30

In this landscape, risk or protective factors are defined as measurable attributes that increase the susceptibility of an individual toward a given outcome. Available knowledge of risk factors associated with autism can facilitate the identification of at-risk individuals or subgroups during different periods of development. Various risk factors (fig 1, A) converge onto plausible biological mechanisms during pregnancy or around birth that can lead to autism in childhood.313233 The presence of protective factors can also modify trajectories; yet, much less is known about protective factors in autism and those proposed likely exert their influence by moderating risk factors, eg, female sex moderating genetic risk.34

Fig 1
Fig 1

Early developmental outcomes in autism are shaped by dynamic interactions over development. (A) Underlying risk factors increase susceptibility for autism and related neurodevelopmental conditions. Risk is expressed from early life, altering (B) brain and (C) behavioral development. Early identification and intervention can mitigate the impact of risk during the period of maximal brain plasticity (D)

Risk and protective factors interact over time in complex ways that can be captured by mapping early trajectories of brain and behavioral development (fig 1 B-C). In autism, risk is expressed in the brain very early on in life, thereby shaping outcomes.7835 Outcomes are usually defined as the incidence, prevalence, and severity of a condition. Wide ranges of behavioral outcomes have been investigated in the early developmental period, including the child’s developmental level, functioning, and family wellbeing. When families are asked about what outcomes matter to them, they prioritize areas that affect everyday life and functioning.36 Developmental outcomes that have been studied show increasing differentiation over time, and together give rise to variability between individuals, beyond what is captured by diagnostic categories. Although this differentiation continues across the individual’s lifespan, some outcomes are measurable in the early developmental period and can serve as risk indicators for subsequent development as soon as infants begin to diverge from typical development (fig 1 C).

When at-risk individuals achieve better than expected outcomes, we can infer processes of resilience.37 For example, if two individuals have the same risk factors for autism around birth but one goes on to develop the condition while the other does not, the latter individual would be described as resilient. However, inferring resilience can be imprecise and evidence is limited in autism because it can be difficult to identify and measure risk factors at the individual level and then track development over time to differentiate those who are resilient in the face of risk. Therefore, in this review, evidence from developmental trajectories is used to infer resilience processes where individuals exhibit high risk expression, ie, developmental perturbation, followed by a later “correction” in the trajectory toward typical outcomes.3538 Resilience has not yet been well investigated in autism despite increased interest in predictors of variable outcomes and the fact that environmental mediation through families, other care givers, and therapists, is actively used in intervention seeking to modify outcomes.

Figure 1 illustrates some of the known risk factors that converge into causal mechanisms for autism and related conditions, risk expression in the brain, and behavior in the first three years, and how emerging signs and symptoms can be modified through early identification and intervention. Each of these areas is discussed next.

Risk factors

Many risk factors have been identified for autism and converge into causal mechanisms for other neurodevelopmental conditions, with their onset occurring during different periods of development. Fig 1 (A) illustrates three groups of risk factors: genetic, maternal health, and neurological. Each factor has been independently associated with the risk of autism with some degree of consistency across studies.31 Each group of factors converges on to a plausible causal mechanism shared with other developmental conditions and/or is well grounded in animal models.313233 Such mechanisms are challenging to investigate directly in autism because of the time lag between their influence on fetal development and a confirmed diagnosis starting at age 3. Nevertheless, some factors can help to identify at-risk groups or individuals before symptoms emerge and can in some cases provide a plausible explanation for autism once it has been suspected or diagnosed.

Genetic factors

Numerous twin and family studies show that genetic and non-genetic factors contribute to an increased susceptibility to autism.394041 While estimates vary, there is general agreement that heritability is greater than 50%,42 with first degree relatives having a higher risk for developing autism relative to the general population. First degree relatives without a clinical diagnosis have a “broader autism phenotype” with subclinical traits similar to those seen in autism.4344 For example, later born siblings who do not have autism exhibit some autistic symptoms and lower levels of functioning compared with the general population.45

The most well defined and studied risk group is later born biological siblings of children with autism.46 Recurrence rates for later born siblings are around 3-7% when ascertained in the general population.4748 Higher estimates have been found in studies with prospective follow-up, where the recurrence rate is about 13.5% (95% confidence interval 8.4 to 20.9) in families with one affected child and 32.2% (95% confidence interval 21.8 to 44.7) in families with two or more affected children.46

Advances have been made in identifying genetic risk factors that are measurable at an individual level. These advances have been integrated into routine diagnosis to explain the causes of autism. Alterations to DNA that substantially increase the risk for autism are cataloged in the scientific literature and in medical genetics and counseling [US sp]guidelines.495051 Genetic testing, namely microarrays, can identify such variants and provide information on the underlying causes in roughly 10-20% of cases,52 a rate that is expected to increase with next-generation sequencing.5051 These variants are very rare (<1%) in the general population; however, their frequency is much higher among individuals diagnosed with autism5051525354 and other neurodevelopmental disorders.55 Such variants include CHD8, 16p11.2 and 15q11.2 deletions or duplications.

Several genetic syndromes overlap with autism in their phenotypic presentation, and a subgroup of these cases meets diagnostic criteria for autism well above expected rates in the general population.5657 Syndromes with the highest overlap with autism are Rett syndrome and Cohen syndrome, where more than half of patients also have autism. Others are Cornelia de Lange syndrome, tuberous sclerosis complex, Angelman syndrome, CHARGE syndrome, and fragile X syndrome, where more than 30% of patients also have autism. Lastly: neurofibromatosis type 1, Down’s syndrome, Noonan syndrome, Williams syndrome, and 22q11.2 deletion syndrome, where more than 10% of patients have autism.56 Prenatal screening, ultrasound findings during pregnancy, or genetic testing after birth can ascertain these genetic conditions. However, most are diagnosed later in childhood partly because of the variability in their clinical features across individuals. In cases where these genetic disorders are suspected or diagnosed, the risk of autism should also be considered.

Genetic risk factors lead to autism by modifying early brain development and functioning,58 including synaptic signaling.5859 It is thought that advanced paternal age increases the risk of autism (pooled adjusted odds ratio 1.55, 95% confidence interval 1.39 to 1.73) by increasing rates of de novo mutations and epigenetic alterations.60 Additionally, sex has been found to interact with heritability where female probands had a greater recurrence risk than male probands (16.7% versus 12.9%).6162 Female probands were also found to have an excess of deleterious autosomal copy number variants across several neurodevelopmental conditions including autism (odds ratio 1.46, calculated confidence interval 1.16 to 1.81).34 These findings account in part for the “female protective effect,” where females would require more causal factors to manifest the same degree of impairment as males.3461 Apart from monogenic X-linked neurodevelopmental disorders, the mechanisms responsible for the female protective effect are still largely unknown.

Maternal health factors

Several factors associated with immune system vulnerability in pregnant mothers are implicated in autism and are thought to interact with genetic factors to increase susceptibility.636465 Consistent risk factors for autism include a family history of autoimmune disease (odds ratio 1.28, 95% confidence interval 1.12 to 1.48),66 maternal infection during pregnancy (pooled adjusted odds ratio 1.12, 95% confidence interval 1.03 to 1.22),67 and maternal autoimmune disease (pooled odds ratio 1.34, 95% confidence interval 1.23 to 1.46).68 Further, in a cohort of about 2000 autistic children aged 4 to 18, the presence of rare genetic variants coupled with a history of maternal infection during pregnancy was associated with more autism symptoms.69

Vulnerability of the maternal immune system interferes with mechanisms of gene expression that affect fetal development.65 Such mechanisms are also strongly modulated by environmental factors that can compromise the integrity of the maternal microbiome, thus altering immune responses in some regions of the brain.65 One such factor is maternal nutrition, where the risk of autism is increased in children of overweight mothers (pooled adjusted odds ratio 1.47, 95% confidence interval 1.24 to 1.74) compared with normal weight mothers,70 and of mothers with diabetes compared with those without diabetes (pooled relative risk 1.48, 95% confidence interval 1.25 to 1.75).71 The interaction between immune system factors and nutrition affects prenatal brain development by increasing oxidative stress and cytokine response and their downstream signaling, which causes the activation of microglia and astrocytes in the brain.2863 These effects can be seen later in life in some of the children with autism by measuring cytokine concentration in peripheral blood72; in one study altered concentrations were associated with more behavior problems in children with autism.73 Maternal health offers important opportunities for augmenting protective factors during pregnancy, such as supplementing folic acid and vitamin D.32 However, given the equivocal evidence for their association with reduced risk of autism,31 these factors are best viewed as protective for neurodevelopment in general and not specific to autism.

Factors related to maternal mental health have also been investigated and show consistent association with the risk of autism, albeit indirectly. These include maternal depression, a known risk factor for a range of health conditions in children.74 Use of selective serotonin reuptake inhibitors (SSRIs) during pregnancy as a result of diagnosed depression increases risk for autism (pooled odds ratio 1.45, 95% confidence interval 1.15 to 1.82).75 This is because SSRIs can cross the placenta and block serotonin transporters, leading to the accumulation of serotonin, which in turn modulates cell division, neuronal migration, synaptogenesis, and other pre- and post-natal developmental processes in children. Atypical serotonin synthesis has been found to lead to higher levels of serotonin in the blood, which is sometimes found in those with autism.76

Other maternal health factors likely mediate the association between the risk of autism and some putative environmental factors by modifying epigenetic mechanisms; these include migration resulting in maternal stress28 and use of assisted reproductive technology.77 What is common in both cases is that environmental factors lead to epigenetic modifications of DNA. These changes are stable and heritable and affect methylation in offspring, thereby affecting mechanisms that induce or suppress genetic expression during development. 78

Neurological factors

Several perinatal risk factors lead to neurological vulnerability, increasing the risk for autism and other neurodevelopmental conditions.31 The same factors can account for the 12.1% of autism cases where epilepsy is also present.79 Some of the perinatal factors that increase the risk of autism are birth injury or trauma (relative risk 4.90, 95% confidence interval 1.41 to 16.94) and low birth weight (<2500 g) (relative risk 1.63, 95% confidence interval 1.19 to 2.33),80 caesarean section (pooled adjusted odds ratio 1.23, 95% confidence interval 1.07 to 1.40),81 and umbilical cord complications (pooled relative risk 1.50, 95% confidence interval 1.00 to 2.24).80 The higher frequency of such pregnancy and birth complications in older mothers is thought to account for the association between advanced maternal age and the risk of autism (pooled adjusted odds ratio 1.41, 95% confidence interval 1.29 to 1.55).60 One causal mechanism resulting from this group of risk factors is hypoxic-ischemic damage that induces inflammation, dysregulation of signaling pathways, and in turn neuronal damage and death.31

Taken together, this evidence suggests that autism has no unique cause. Instead, multiple mechanisms exist, including well validated ones shared with other conditions. Some cases of autism share causes with monogenetic conditions, while other cases have conditions affecting prenatal and/or perinatal environments, which result in neurological vulnerability. In the latter, autism can arise as a result of mutational burden and/or epigenetic modifications interfering with brain development and functioning; autoimmune activation modifying brain growth prenatally; or neuronal death or damage around birth. Future research may help identify biological subtypes emerging from distinct causal pathways where targeted therapies may be useful during specific periods of development. In all cases, including causal pathways yet to be discovered, alterations in cellular pathways affect further brain development and reorganization.

Trajectories of brain development

Causal mechanisms leading to a diagnosis of autism in childhood start altering trajectories of structural and functional brain development soon after birth.78 Our knowledge of these alterations comes from investigating intermediate phenotypes that have been used at the intersection of neuroscience and psychiatry because they more likely reflect the underlying cause of a disorder relative to complex diagnostic outcomes. Often misunderstood as biomarkers, intermediate phenotypes do not satisfy most conditions for clinical utility expected in biomarkers: they are not static indicators of the presence or absence of autism.82 Rather, investigating intermediate phenotype trajectories in early in life offers opportunities for understanding processes of risk and resilience that unfold over time. In the future, some intermediate phenotypes may prove to be useful biomarkers in conjunction with conventional risk and diagnostic assessments. Evidence for altered trajectories using these methods relates to two broad categories of intermediate phenotypes: neuronal connectivity intermediate phenotypes that reflect overall brain reorganization, and perception/attention intermediate phenotypes that reflect altered development of specific functional systems such as visual or auditory systems.

Neuronal connectivity—Distinct neuronal groups or brain regions are connected either through anatomical links or through statistical dependencies/causal interactions, known as functional or effective connectivity. Anatomical connectivity is optimally measured with magnetic resonance imaging, while functional connectivity is measured with electroencephalography (EEG)/magnetoencephalography (MEG). Connectivity has become increasingly popular as scientific studies have so far failed to identify focal brain abnormalities or atypicality in specific structural or functional brain systems as underlying autism or other neurodevelopmental disorders. A systematic review of all EEG/MEG functional connectivity research on autism concluded that different studies or techniques most likely measure distinct and unrelated underlying neural mechanisms and are sensitive to different artefacts.83 Few consistent patterns emerged across the life span of an individual. However, a general trend emerged toward long range underconnectivity.83

Atypical trajectories in brain development emerge very early on in infants who later develop autism, well before the onset of overt symptoms. Data from two independent cohorts showed atypical trajectories in EEG spectral power848586 and complexity87 beginning in the first year of life. Similarly, atypical trajectories in structural brain development have been reported in one cohort of at-risk infants followed from 6 to 24 months using various “intermediate” phenotypes, ie, measurable quantitative traits including variations in cortical surface area,88 extra axial fluid,89 corpus callosum size,90 measures of fractional anisotropy,91 and neural network efficiency.90

Perception and attention—Early emerging brain processes modify selection and guide action and learning from the external environment, and therefore underlie variation in typically developing and in and autistic individuals. Infants at risk for autism, including those who go on to develop the condition, exhibit enhanced perceptual skills, shown in measures of visual search92 and pupillary light reflex.93 Since these developmental precursors likely give rise to the strengths associated with the autism phenotype later in life, eg, heightened perception94 they may be conceptualized as protective factors, but further studies are needed to directly examine how they map on to variable developmental outcomes in individuals. However, as children who develop autism enter their second year of life, they begin to exhibit reduced flexibility in visual attention.959697

Early emerging perceptual and attentional differences modulate sensitivity to socially relevant information. On the one hand, infants at risk, including those who later receive a diagnosis, orient to faces98 and modulate their attention in response to different face cues, which predicts their expressive vocabulary in toddlerhood.99 By contrast, a developmental pattern of reduced attention to socially relevant information over the first two years of life is associated with a later autism diagnosis.8 This pattern is likely the result of atypical sensitivity to more complex, socially relevant information that requires integration of bottom-up and top-down information, such as the processing of eye gaze100 and biological motion.101 Possible mechanisms of resilience have been suggested, but have yet to be investigated.38 For example, one longitudinal study reported that infants at risk who do not develop autism showed a unique pattern of brain responses to specific social stimuli that was not found in infants who developed autism or in the control group, ie, this could indicate a possible alternative/compensatory mechanism.100

Taken together, strong evidence exists for the neural risk processes leading to autism early in life. Risk expression in infancy results in domain-general changes altering structural and functional networks leading to overall brain reorganization. In turn, atypical interactions within developing brain systems and the external environment result in domain-specific alterations prominently affecting brain systems that underlie social behavior in those who develop autism.7 By contrast, we know very little about underlying processes that may lead to resilience in infants at risk, or about possible protective factors that give rise to strengths associated with the autism phenotype later in life. Converging evidence from behavioral studies shows that risk expression is also seen in early development, but direct measures of brain structure and functions are more sensitive in detecting risk expression within the first year of life.102

Trajectories of behavior

Early signs of autism observed in high-risk groups are the developmental precursors to fully fledged social and communication symptoms of autism. The trajectories of these risk signs in infants who later develop autism compared with those who don’t are indistinguishable in the first year; however, the two groups become differentiated more reliably after the first year of life. Parents tend to be the first to detect early signs of autism, typically well in advance of diagnosis.103104105 Tools used for early identification of autism rely on parental observations of early emerging signs and symptoms. The most frequent concerns reported by parents relate to development of language and communication.104105 Other concerns relate to motor or sensory delay or atypicality, as well as problems with sleep or feeding.104 In prospective studies of infants at risk, parents also report extremes of temperament (eg, the infant is too passive and has difficulty with self-regulation) that are associated with a later diagnosed autism in some studies.106107108109110

Risk signs prospectively assessed in infants at familial risk include various forms of delays and atypicalities (fig 1 B).8111112 During infancy, cognitive113 and adaptive functioning trajectories that measure daily living activities 113114 of individuals who later develop autism also become increasingly impaired. One of the reasons emerging risk signs are not strong predictors of an autism diagnosis early on, is that at-risk infants who do not develop autism exhibit similar risk signs, which emerge as sub-threshold symptoms consistent with a broader phenotype in biological relatives.45115 Other studies prospectively contrasted early development of infants with fragile X with infants at familial risk and found that those with fragile X had overall lower levels of social communication116 and adaptive skills.117 Focusing specifically on response to name, one study contrasted developmental profiles of infants from 9 to 24 months old across multiple conditions. Compared with those with Rett and fragile X syndromes; infants later diagnosed with autism initially had similar scores that later declined in the second year of life.118

Few studies have modeled various developmental trajectories including symptoms, cognitive development, and adaptive functioning from the second year of life onward in at-risk infants suspected of autism and referred for evaluation. Autism symptom trajectories measured using the Autism Diagnostic Observation Schedule (ADOS) showed overall stability over time for subgroups with different levels of initial severity.119120121 Stability or change in ADOS trajectories was generally mirrored in other outcome measures, namely IQ.119120 These findings are consistent with a prospective study of IQ and adaptive functioning trajectories in younger infant siblings starting from the first year to 36 months old.114122 Over time, infants who later developed autism were over-represented in subgroups characterized by delays or a failure to acquire new skills, and those with decreasing adaptive skills over time.122

Despite the overall stability in symptoms, all of the above studies identified groups that improved or worsened over time across multiple domains that were overall correlated. A study that measured traits of ADHD found similar results where children with persistent autism symptoms also had stable or worsening ADHD traits over time, whereas improving subgroups showed a simultaneous decrease in both autism symptoms and in ADHD traits.119 Consistent findings from younger infant siblings at risk for autism suggest that temperamental characteristics related to later developing executive functions such as inhibition and working memory (which are impaired in autism and ADHD) may act as a protective factor, attenuating emerging symptoms.123

Other findings point to more dynamic interactions between developing systems; in some infants with worsening trajectories, delays in expressive language development starting in the second year of life were amplified and compounded over time, thus emerging as widespread delays and atypicality in other cognitive skills.119120 By contrast, one study suggested a pattern of domain specific resilience in a subgroup characterized by initially low but rapidly improving adaptive skills, especially skills in the communication and motor domains, during the same period when IQ remained stable.122 This suggests that although the correlations with IQ are generally strong in autism, a very high IQ does not guarantee high adaptive skills and a slightly lower IQ does not mean an individual cannot have good adaptive skills. Considering the overlap in causes and phenotypes between autism and other conditions, there is also a push for universal dimensions cutting across physical and mental health to characterize a person’s functional (dis)ability.124 In parallel, continued investigation of autism compared with overlapping conditions such as genetic syndromes can further identify how the autism phenotype might be unique across various outcome dimensions and despite the shared causes.

Early identification and intervention


Longstanding interest in early identification and intervention is justified based on fundamental neuroscience principles: the first years of life are the period of maximal brain plasticity; therefore, modification within this period is most effective in optimizing long term outcomes. Some of the factors in fig 1A can be used to flag at-risk groups, and standardized tools useful at various ages have been developed to support very early identification of risk signs and emerging symptoms. In turn, early identification allows for the possibility of early intervention.125

In reality, delays tend to occur between initial concerns, referrals, a confirmed diagnosis, and access to intervention.9899100125 The nature of emerging signs and symptoms reliably accounts for some of the variation in age at diagnosis.126127128 Several social determinants can modify the age of diagnosis by increasing or decreasing the likelihood of identification of signs, help-seeking, and/or access to services. For example, in the US, reports indicate a pattern of earlier age at diagnosis for children in families of higher socioeconomic status or in communities with more resources, which likely reflects access to services.126127 Similarly, in Canada, variation in diagnostic service delivery models across different provinces, ie, diagnosis by community physicians versus by specialist teams, was found to modify the timing of referral to early intervention programs.129 By contrast, other sociodemographic factors can heighten parental awareness and recognition of emerging risk. For instance, children with autism are identified earlier if they have typically developing or autistic sibling(s),105 and if they have frequent contact with their grandmothers.130 Variability in the factors affecting early identification is one of the reasons that a systematic approach to risk assessment and monitoring that actively involves families is essential.

Several standardized tools have been developed to facilitate risk assessment at various ages depending on risk status and on the nature of concerns; however, many require further validation, especially in general population samples.125131 For example, screening tools such as the Modified Checklist for Autism in Toddlers (M-CHAT)132 is recommended for use in primary pediatric care at around 18 months and has sensitivity near 0.83 (95% confidence interval 0.75 to 0.90) and specificity around 0.51 (95% confidence interval 0.41 to 0.61).133 In view of the complex causes, some issues are important to consider when using such tools. First, although we have limited knowledge on the performance of autism identification tools in unselected general population samples,133 they are very useful in children with risk indicators. Second, in risk groups where parents are already concerned, most tools have good sensitivity but poor specificity,125 which is not surprising considering the shared risk between autism and phenotypically overlapping conditions. Repeat use of such tools to monitor risk signs over time is recommended125134 where some infants missed at younger ages would be detected later.135

Indeed, several studies show that introducing standardized tools in a range of settings (healthcare, education, and community centers) does increase screening rates.127 Yet, concerns remain that the benefits of screening are diminished when subsequent monitoring and appropriate referrals are not made for children who screen positive. Therefore, additional capacity building strategies have been used to improve autism identification within existing routine systems of developmental surveillance and monitoring,29 including training for service providers to identify signs, use standardized protocols, and enhance care coordination.136137138 Further research is needed to examine the efficacy of such health systems interventions in lowering the age of diagnosis and increasing access to services, particularly in regard to early intervention programs.

Consistent with global policy recommendations, evidence also highlights the importance of integrating early identification tools into routine services, including country level developmental surveillance or other maternal and child health programs.29 Such programs have been substantially strengthened all over the world and offer ideal opportunities for leveraging existing knowledge and capacity in the area of child development. They also offer ideal opportunities for research to address current knowledge gaps in how social determinants modify help-seeking behavior, access to care, an/or clinical presentation, and/or outcomes, thereby improving early identification in the community.


Evidence for efficacy and effectiveness of interventions targeting children with overt symptoms of autism in toddlerhood is equivocal.139140141 Despite the evidence, a robust case exists from neuroscience and early childhood development in general for the importance of this period in optimizing long term outcomes. Therefore, guidelines (discussed below) recommend referral to such programs when they are available. One challenge in this area is to improve understanding of the “active ingredients” in such interventions, in order to enable rational selection of strategies based on each child’s characteristics and needs.142 Principles underlying several approaches, developed and tested independently, were recently grouped together under the umbrella term of naturalistic developmental behavioral interventions.143 Such interventions focus on supporting social interaction and play, and encouraging child initiated communication, all within naturalistic environments, where positive outcomes often include improvements in IQ, language, and communication.140141

The evidence base is more recent for very early targeted intervention, ie, those designed for at-risk groups and for indicated interventions, or those designed for children with identified autism symptoms.144145146147 Some early interventions rely on therapists, while others are “parent mediation,” ie, therapists work with parents on strategies that can improve interactions, engagement, and communication in individuals diagnosed with autism. Studies have started to directly test interventions with infants at risk within the prodromal period, before the onset of overt autism symptoms, but have so far revealed mixed results. One trial randomized 54 infants, 7 to 10 months old, who had an older sibling with autism to either a 12 session parent mediated social communication intervention or to no intervention.148149 In the intervention group, positive effects were observed immediately after the parent-child interaction intervention (effect size 0.29, 95% confidence interval 0.26 to 0.86),148 which was followed by a reduction in autism symptoms several months afterwards (effect size 0.32, 95% confidence interval 0.04 to 0.60; P=0.03).149 Positive effects of such interventions in modifying parent- child interaction were not replicated in another trial with 103 infants aged 9 to 14 months with observed risk signs (effect size –0.19, 95% confidence interval –0.63 to 0.25), although the follow-up study is still ongoing.150 Both studies also found mixed intervention effects on other outcomes. For example, care giver report, but not researcher rated measure of child language and communication, showed positive treatment effects.150 Another pilot study randomized 36 siblings aged 6 months into intervention versus monitoring, and reported intervention effects using direct brain function measures in response to social stimuli, namely a greater increase in frontal power, but did not report behavioral outcomes.151

Similar to interventions with older children diagnosed with autism, we are still in the early days of understanding the ways in which modifying an infant’s external environment shapes developing neural mechanisms, leading to gains in behavioral outcomes. For example, parenting skills and enhanced self-efficacy can act as general protective factors for early infant development, which maintains trajectories on typical paths. Alternatively, gains in synchrony of parent-child interaction may modify input to the child in a way that enhances resilience in the developing brain, resulting in a “correction” of the developmental course that has already been derailed. In both cases, gains can be either attained in multiple developing systems or alternatively in one area of development, such as early language, that then cascades to other developmental domains over time. The progress made in establishing feasibility and acceptability of such interventions is now paving the way for testing these different developmental accounts, potentially linking early intervention to improved outcomes.


Available knowledge of genetic risk factors is now integrated into guidelines recommending genetic testing to ascertain potential causes in a subgroup of those with autism.52152153154 Although there are differences in recommendations with respect to universal screening, guidelines consistently recommend surveillance for infants at risk and the use of standardized tools to assess risk signs and emerging symptoms, followed by appropriate referrals for diagnosis and early intervention.3112135155156 The high risk of recurrence among later born siblings makes them an important group for surveillance.404647


Advances in the discovery of risk factors for autism have opened up the possibility of identifying groups or individuals at risk for subsequent monitoring, starting from birth. Some factors can be used as risk indicators, while others may be modifiable in order to mitigate the impact of risk. In all cases, multiple factors converge on to a smaller group of causal mechanisms shared with other neurodevelopmental conditions that also constitute possible diagnostic outcomes or comorbidities in those at risk.

Our understanding of the complex causes of autism has also influenced how we define and measure developmental outcomes. There is a shift away from understanding autism as a narrowly defined categorical disorder. Multiple dimensions, diagnostic, cognitive and functional, more accurately reflect individual phenotypes. Progress is also ongoing in elucidating the dynamic developmental pathways of brain and behavioral development that link risk factors to outcomes. Very early brain development in autism is characterized by subtle hits across multiple early developing brain systems, giving rise to overt behavioral symptoms observed in the second year of life in a subgroup of those at risk.

Yet, research has disproportionally focused on risk factors and on characterizing problems and symptoms in infants who develop autism. We know much less about protective factors and surprisingly few studies have fully used prospective designs to understand variability in developmental outcomes around the time of diagnosis, including resilience, to give better than expected outcomes. Few studies suggest that there is indeed a subgroup with initial high expression of risk interfering with development, where trajectories are corrected over time. Further research in this area can potentially help identify the underlying brain mechanisms for the observed variability in developmental pathways.

An improved understanding of underlying mechanisms of risk and resilience can also advance the area of very early interventions before the emergence of overt symptoms in order to mitigate impact of risk on further development. Feasibility of such interventions has been demonstrated, but evidence for efficacy remains equivocal. Despite the promise of biological measures in the context of intervention research, we currently have no rigorous studies that carefully connect brain and behavior changes as a result of early interventions. This knowledge gap is important to address in future research when considering the various brain and behavioral mechanisms potentially underlying targeted intervention, and the possibility that different mechanisms might yield different intervention outcomes across individuals.

Research questions

  • What are the effective biological indicators of prodromal expression of risk?

  • What makes some individuals resilient in the face of risk?

  • How do we define and measure the most relevant outcomes for a child/their family?

  • What are the predictors of variability in outcomes?

  • Are there modifiable risk factors that can promote optimal/positive outcomes?

How patients and the public were involved in the creation of this article

No patients were directly involved.


  • Series explanation: State of the Art Reviews are commissioned on the basis of their relevance to academics and specialists in the US and internationally. For this reason they are written predominantly by US authors

  • Contributorship statement: N/A

  • Competing interests The BMJ has judged that there are no disqualifying financial ties to commercial companies. The author declares the following other interests: none.

  • Further details of The BMJ policy on financial interests are here:

  • Provenance and peer review: commissioned; externally peer reviewed.


View Abstract