In an analysis of a birth cohort of children born in California between January 1990 and December 2002, conception in December, January, February, and March was associated with a modest increased risk of autism, ranging from 8% to 16%, after adjusting for maternal education, child’s year of conception and ethnicity. Conception in March was associated with the highest risk of autism.
Time of conception can provide clues about environmental factors that could be associated with autism. Environmental agents that predominate in California during December – March include virus infections and agricultural applications of certain pesticides.
According to the Centers for Diseases Control and Prevention, February is the month with the highest percentage of influenza cases (nearly 50% of all cases), followed by January (20%), and March and December (about 15% each).24
The California Influenza Surveillance Project data likewise show that influenza activities are at their peak from about December to March.25
These are the specific months in which conception was associated with increased risk of autism in our data.
There has been little epidemiologic research to address a possible association of viral infection with increased risk of autism. In the 1970s, two independent research groups reported an association between prenatal congenital rubella infection and increased risk of autism.2,3
Clinical infection of pregnant women with measles, mumps and chicken pox has also been reported to be associated with increased risk of autism.26
However, the prevalence of those viruses is very low in the US due to effective vaccination programs, and thus unlikely to play a strong role in the etiology of current cases of autism.
Influenza, in contrast, is much more common. One earlier case-control study reported that mothers were more likely to have had influenza or to have been exposed to influenza during their pregnancies with offspring having autism than with their other offspring.26
In contrast, Dassa et al.27
did not find an association between high probability of maternal exposure to influenza during pregnancy and autism. However, the manner in which exposure was characterized in the latter study could have resulted in misclassification, leading to biased results.
Prenatal influenza infection in mice leads to offspring with behavioral deficits that are similar to the deficit in social interaction and restricted interest observed in children with autism. Similar results have been observed in the absence of viral infection by injecting the double stranded RNA poly (I: C) in pregnant dams, which produces an inflammatory response. 28
The heterogeneity of the viruses that have been associated with risk of autism, in addition to the fact that maternal exposure to the double stranded RNA poly (I: C) can produce in animals the same effect as viruses,28
has led researchers to suggest that maternal reaction to infections is a more plausible mechanism of pathology than a direct viral effect on fetuses. 29
Agricultural and other pesticide applications also vary by season. Data from California’s Pesticide Use Report indicate that between 1990 and 2003, about 50% of the total amounts of 13 agricultural organophosphate and organochlorine pesticides were applied in the summer.30
Previous studies have suggested an association between both organophosphate and organochlorine pesticides and neurodevelopmental disorders.5,6,31
In the present analysis, we did not find an association between conception in the summer and risk of autism. However, children conceived in the higher-risk months in our study would have been in their second or third trimester during the summer season. If the excess seasonal risk is related to organophosphate or organochlorine pesticides, this would imply that the sensitive period for a pesticide effect could be in the second or third trimester.
If agricultural pesticides were causes of autism, one might not expect every analysis of seasonality to find the same risk months because agricultural practices (and thus pesticide application) vary by location. Home and garden applications of pesticides also vary by season and deserve consideration.
Other seasonal factors include maternal allergies during pregnancies32
and mild maternal nutritional deficiencies for such factors as vitamin D.33
We know of no epidemiologic studies that have explored the association of such factors with autism.
Our results could be subject to some minor misclassifications of outcome or exposure. We derived months of conception from the date of birth and the length of gestation. The latter was estimated based on the self-reported date of last menstrual period from the birth file. These dates can be subject to recall errors due to uncertainty in the reported dates, bleeding during early pregnancy not associated with menses, menstrual irregularities and variation in the length of the follicular phase of the menstrual cycle. 34
These uncertainties can lead to either differential or nondifferential misclassification of the month of conception. For example, low fertility can be associated with irregularity in menstruation. If low fertility is associated with autism,35
there could be a differential misclassification of exposure (month of conception). In this case, it would be difficult to determine the direction of bias. However, since our findings are clustered in consecutive months, the impact would likely be low.
We relied on diagnoses reported on the Client Developmental Evaluation Report and Early Start Report, which are not systematically confirmed. However, a previous study confirmed that about 98% of cases of autism reported in the Department of Developmental Services database were at least on the autism spectrum, based on either an Autism Diagnostic Interview-Revised or an Autism Diagnostic Observation Schedules-Generic, and 64% were confirmed autistic-disorder cases based on both instruments.36
Thus, we can assume that false-positive cases will be minimal among our cases, if we consider all autism-spectrum disorders as cases.
It is possible that autism or autism spectrum disorders are present among the controls. Not all children with autism in California are registered in the Department of Developmental Services database.21
A previous study reported that about 75–80% of autism cases in California 22
are registered. We do not expect these missed cases to affect our results because they would be a very small fraction of the controls. If we assume that 25% of cases are misclassified as controls, and also assume that this misclassification is independent of month of conception, our ORs and 95% confidence intervals are virtually unchanged ().
Effect of nondifferential exposure misclassification bias on the estimated ORs for the association between selected months of conception and autism before 6 years of age.
A recent study in the UK found increased risk of autism associated with conception in summer.16
However, neither age at diagnosis nor the duration or definition of the follow-up period was provided. Moreover, the study had a small number of cases (86 compared with 19,238 cases in our report). Atladotir et al 37
reported no association between season of conception and risk of autism, although, the follow-up time for a chance of a diagnosis of autism differed by birth cohort in their study.
Previous studies on season of birth have reported increased risk of autism in infants born in March or August. We found that November births (corresponding to February conception) had the highest risk after controlling for year of birth, maternal education, and child ethnicity.
Increases in the rate of autism have been reported.21
In order to remove the effect of such increase on the estimated rates, we controlled for time trend.37
Year of conception was a strong confounder of the association between month of conception and risk of autism (and also between month of birth and risk of autism), as evidenced by changes greater than 10% in the magnitude of the beta coefficient for some months or seasons of conception. By adjusting for the time trend on a continuous scale, we minimize residual confounding that can occur when time is controlled on a categorical scale. The relation between the rate of autism and calendar time in our data was approximately linear ().
Most previous studies have not controlled for calendar time.9,11–14,18,38
One study controlled for a birth-cohort effect by grouping years of birth into 4-year categories.15
Such analysis could be subject to residual confounding due to wide categories. Kolevzon et al.17
reported controlling their results for year of birth, although they did not state precisely how the variable “year of birth” was parameterized.
We observed modestly higher risks of autism among children born in winter months. Month of conception can represent a surrogate for seasonal causal factors (e.g., potentially, viral infections, pesticides, and so forth). It is possible that by using month of conception, we introduced nondifferential misclassification relative to the actual exposure, for which season is acting as a proxy. Nondifferential misclassification can lead to an underestimate of the true measure of association. Thus, if these findings are reliable, the true underlying causal factor probably has a stronger association with autism than observed here for month of conception.
The importance of small effects has been questioned.39
Odds ratios or relative risks are measures of effect that provide information on the profile of individual cases of diseases but are not informative about the determinants of overall disease incidence in the population. 40
Small changes in the overall population risks can have important effects on population disease burden, particularly when exposures are commonplace.41
Our results suggest the presence of some environmental factors that vary by season are involved in the etiology of autism.