Patterns of Developmental Milestone Delay Identified Using Latent Class Analysis in a Finnish Population Based Sample of Autism Cases and Controls

Poster Presentation
Saturday, May 4, 2019: 11:30 AM-1:30 PM
Room: 710 (Palais des congres de Montreal)
K. Cheslack-Postava1, S. Hinkka-Yli-Salomäki2, I. W. W. McKeague3, A. Sourander2 and A. S. Brown4, (1)Columbia University Medical Center, New York, NY, (2)University of Turku, Turku, Finland, (3)Biostatistics, 455 Central Park West, Apt 8D, New York, NY, (4)Columbia College of Physicians and Surgeons, New York, NY
Background: Autism spectrum disorders (ASD) are characterized by atypical developmental trajectories in language and social domains. However, delays in non-verbal cognitive and in fine and gross motor domains are also common. Latent class analysis (LCA) may be useful to empirically identify potentially complex patterns of delayed development across domains.

Objectives: To empirically identify developmental classes characterized by shared patterns of delayed milestone achievement, using population-based data from subjects with autism and with typical development.

Methods: Data on the age at achievement for 21 developmental milestones with expected age at attainment between birth and 24 months were abstracted in Finland from well-child clinics for 846 cases with childhood autism identified in the Finnish Hospital Discharge Register and 925 controls matched on sex, date of birth, and place of birth. For each milestone, a dichotomous indicator of delay was coded as positive if the age at achievement was greater than the 90th percentile of control values. Latent class models with 2-10 latent classes were fit based on the dichotomous variables for each milestone and the optimum number of classes was selected based on the Bayesian Information Criterion (BIC). An LCA model with the optimum number of classes was then used to estimate the class membership probabilities for each individual and the item-response probabilities for each type of delay. Subjects were classified based on their maximum probability of class membership, and the distribution of developmental class membership was compared between autism cases and controls using multinomial logistic regression.

Results: A 5-class model best fit the data based on the minimum BIC. 72.5% of subjects (61.5% of cases and 82.6% of controls) belonged to a “non-delayed” class with low probabilities of delay for all milestones. A class containing 10% of subjects had high probabilities of gross motor delays (crawling and standing). An “early delay/smile” class (6.7%) exhibited delays in early motor (i.e. lifting head) milestones and in smiling responsively, but not in later skills. A “late delay/interactive” class (4.5%) showed delays in social and receptive language milestones (i.e. pointing, collecting objects upon request, playing peek-a-boo). The remaining class was characterized by overall delays. All classes other than “non-delayed” were significantly associated with increased risk of autism. The highest odds ratios (ORs) for autism were associated with the “late delay/interactive” class (OR (95% CI)=7.13 (3.86, 13.2); p<0.0001) and the overall delays class (OR (95% CI)=6.38 (3.73, 10.9); p<0.0001), after adjustment for pre-term birth and low birth weight.

Conclusions: While the majority of cases with autism did not show apparent patterns of delay based on developmental milestones before age 24 months, specific patterns of delay were associated with highly increased autism risk relevant to a subset of cases. Characterizing the heterogeneity of autism based on shared patterns of early development may assist etiological and clinical research by identifying more homogenous groups of cases, for targeted early intervention and identification of correlations with other developmental events and neurobiological processes.