A new study of over 2.5 million children provides evidence that receiving the measles, mumps, and rubella vaccine before age two is not associated with an increased risk of developing autism. The findings, published in The Pediatric Infectious Disease Journal, indicate that the precise timing of the immunization was not linked to a child’s likelihood of being diagnosed with the developmental condition.
Autism spectrum disorder is a developmental condition that affects how a person communicates, interacts with others, and experiences the world. The condition is currently estimated to affect about 3.2 percent of children in the United States. The measles, mumps, and rubella vaccine is a standard early childhood immunization that protects against three highly contagious and potentially severe viral infections.
Fears linking this routine vaccine to autism persist among some parents, frequently leading to delayed or declined immunizations. This anxiety largely stems from a 1998 case series, a report on just 12 children, that was later retracted due to scientific misconduct. Numerous large studies over the years have found no link between the vaccine and the developmental condition.
Even with this extensive body of research, observational studies on vaccine safety face built-in design challenges. Factors like a family’s healthcare habits, early parental concerns about a child’s development, and the child’s age can influence both when they get vaccinated and if they receive an autism diagnosis.
Todd Burstain, a physician at UW Medicine and the lead researcher of the study, explained why a new approach was necessary. “The prior studies on MMR and autism were mainly done in Europe where the vaccine and the schedule for MMR are slightly different,” Burstain told PsyPost. “The European study was older (data from 2000-2012). During this time autism was not being diagnosed as frequently and the prevalence in these studies was about 1/3 of current prevalence in the US.”
If an analysis does not adjust for the fact that children receive vaccines at different developmental stages, the results might be skewed. To address these complexities, Burstain and his colleagues used a more precise study design.
The authors utilized an approach known as age-anchored landmark analysis. This method groups children by the age window in which a vaccination decision is typically made, comparing vaccinated and unvaccinated children within those defined age windows to reduce bias.
This method avoids what researchers call immortal time bias, a statistical distortion that can arise when the time before a child is vaccinated is counted as if it belonged to the vaccinated group. The design also counts only autism diagnoses made after each age window begins, so a diagnosis can never come before the vaccination it is compared with. Burstain argued that prior European studies, which compared vaccinated and unvaccinated children without anchoring them to specific ages, “could potentially introduce an immortal bias into the data.”
To carry out the research, the scientists analyzed records from Cosmos, a massive electronic health record database spanning hospitals and clinics across the United States. “The Cosmos database is a tremendous resource for clinical observation studies,” Burstain said. “More researchers should become familiar with Cosmos and learn how to use it responsibly based on the best scientific principles.”
The dataset included 2,560,035 children who had at least one routine primary care visit between 12 and 24 months of age. The prevalence of childhood autism in this sample was 2.55 percent. The researchers focused on a specific diagnostic code for childhood autism, a narrower category than the full autism spectrum, which partly explains why the prevalence in the sample was lower than the national estimate.
The researchers excluded any children diagnosed with autism at or before 11 months of age to ensure the vaccine came before the diagnosis. The children were followed up to age eight or until December 2024. By 24 months of age, 91.4 percent of the children in the sample had received their first dose of the measles, mumps, and rubella vaccine.
The authors divided the children into defined age windows, such as 11.5 to 12.5 months, 12.5 to 13.5 months, and up to 24 months. Within each window, they compared children who received their first dose during that window with children who had not been vaccinated by the start of the window and remained unvaccinated throughout follow-up.
To ensure the comparison was as fair as possible, the researchers adjusted for factors that might influence both vaccination and autism diagnosis, choosing them with a causal diagram, known as a DAG, that maps how different factors could affect each other. These included the mother’s age, race and ethnicity, gestational age at birth, whether the baby was small for its gestational age or one of multiple births, diabetes or genital or urinary tract infections during pregnancy, neighborhood disadvantage, and region of the country. They avoided adjusting for factors that happen after the vaccination, which can distort the results.
Across all the age windows analyzed, the first dose of the vaccine was not associated with an increased risk of autism. In the primary analysis, covering vaccination between 11.5 and 24 months, the adjusted hazard ratio was 0.97, with a 99 percent confidence interval of 0.91 to 1.03. A hazard ratio compares how quickly new diagnoses occur in two groups, and because this range includes 1.0, the result is consistent with no difference in risk. The 0.97 estimate translates to an estimated 3 percent lower rate of diagnosis among vaccinated children, a difference that is not statistically distinguishable from zero.
The upper end of the confidence interval rules out an increase in risk of more than about 3 percent. “The size of the study matters only in that it increases the power of the study to detect any small differences,” Burstain noted. “The power of this study was significantly greater than the power of the prior European studies and so even very very small increases in risk would have been seen if they were present.”
“For the average person, this study can now say with very strong certainty that children in the US who receive the MMR vaccine between 11.5 and 24 months of age do not have any increased risk of developing autism,” he added.
In two of the later windows, 14.5 to 17.5 months and 17.5 to 24 months, vaccinated children had slightly lower rates of autism diagnosis, with hazard ratios of 0.90 and 0.92. The authors attribute small deviations like these to leftover bias in the data rather than to any protective effect of the vaccine.
The findings held up under different analyses. Results were essentially unchanged with more birth-related factors added (hazard ratio 0.97) and without the neighborhood measure, which was missing for 53 percent of children (hazard ratio 0.96). They also held after excluding children with only one recorded autism diagnosis.
To double-check their methodology, the researchers also tested a negative control exposure. They looked at the pneumococcal conjugate vaccine booster, which is given around 12 to 15 months to prevent bacterial infections but has no theoretical link to autism. A negative control is a comparison where no effect is expected, used to check whether a method produces false signals.
The similar result for the pneumococcal booster suggests the analysis was not generating spurious links tied to routine checkups, supporting the authors’ view that leftover bias is unlikely to explain the measles, mumps, and rubella findings.
“Using DAG adjustments, a negative control, and a landmark longitudinal design resolves any potential bias issues present in prior studies,” Burstain emphasized.
As with all research, there are a few things to keep in mind. The study relied on observational medical records, meaning the researchers could not control for every single variable in a child’s life. The authors acknowledge in the paper that some leftover confounding cannot be fully ruled out.
“This is an observational trial so we can say that we did not see any association between autism and MMR vaccination in this age group, but we cannot say MMR does not cause autism,” Burstain said. “Causation can only be claimed with randomized prospective trials. Doing this however would be unethical. As a result this is the best that science can do.”
Notably, the dataset did not include information on a family history of autism. If parents who already have an older child with autism are more hesitant to vaccinate their younger children, this missing detail could skew the data. Such a pattern would tend to make the vaccine look protective rather than harmful. A 2015 study the authors cite, which examined U.S. children with older siblings who had autism, also found no link between the measles, mumps, and rubella vaccine and autism.
“The most common misinterpretation of the study is what it doesn’t say,” Burstain clarified. “The study does not say that in a small population of patients with a particular diagnosis or family history etc, that there is no increased risk. This was not studied in this paper but the family history question has been studied in other papers looking at siblings.”
Additionally, the children who received the vaccine tended to have much longer follow-up periods in the medical system compared to the unvaccinated children, with a median of 853 days for vaccinated children versus 366 for unvaccinated. Typically, spending more time interacting with healthcare providers creates more opportunities to receive a formal diagnosis, yet the researchers still found no link between the vaccine and autism.
The requirement that children have linked birth-parent records also excluded a substantial proportion of otherwise eligible children from the analysis. This means the findings might not capture the experiences of all children receiving medical care.
Future studies could incorporate external data on family history and socioeconomic factors to provide an even sharper picture. Scientists might also test if these findings hold true for broader autism spectrum disorder diagnoses rather than the specific childhood autism diagnostic code used in this analysis.
Burstain and his team are planning related work. He noted that they have a paper in review “looking at 88 other factors that may or may not be associated with an increased risk of childhood autism.” The researchers “are also planning on repeating the MMR study for Asperger’s (F84.5) and for ADHD,” he said.
The study, “Association Between First MMR Vaccination Before Age 2 Years and Childhood Autism in a U.S. EHR Cohort of 2.5 Million Children,” was authored by Todd L. Burstain, Dale Jacques, and Jennifer M. Burstain.