Objective: For the first time, we use a longitudinal population-based autism cohort to chart the trajectories of cognition and autism symptoms from childhood to early adulthood and identify features that predict the level of function and change with development.

Method: Latent growth curve models were fitted to data from the Special Needs and Autism Project cohort at three timepoints: 12, 16 and 23 years. Outcome measures were IQ and parent-reported Social Responsiveness Scale (SRS) autism symptoms. Of the 158 participants with an autism spectrum disorder at 12 years, 126 (80%) were re-assessed at 23 years. Child, family and contextual characteristics obtained at 12 years predicted intercept and slope of the trajectories.

Results: Both trajectories showed considerable variability. IQ increased significantly by a mean of 7.48 points from 12 to 23 years while autism symptoms remained unchanged. In multivariate analysis, full-scale IQ was predicted by initial language level and school type (mainstream/specialist). Those with a history of early language regression showed significantly greater IQ gains. Autism symptoms were predicted by Social Communication Questionnaire scores (lifetime version) and emotional and behavioral problems. Those attending mainstream schools showed significantly fewer autism disorder symptoms at 23 than those in specialist settings; this finding was robust to propensity score analysis for confounding.

Conclusion: Our findings suggest continued cognitive increments for many across the adolescent period, but a lack of improvement in autism symptoms. Our finding of school influences on autism symptoms requires replication in other cohorts and t settings before drawing any implications for mechanisms or policy.