Prevalence of Regression within AUTISM Spectrum Disorders: A Quantitative Synthesis

Background: Developmental regression occurs in a sizable minority of individuals with autism spectrum disorders (ASD); however, a wide range of prevalence rates of regression is reported in the literature with a range of 12 – 50%. The literature also varies regarding types of regression reported, with language and/or social regression commonly reported and other regression types (e.g., motor, cognitive) reported less often. Authors have hypothesized that variation in the reported prevalence rates of regression depend in part on sample characteristics, such as sample size, with higher prevalence rates reported for smaller samples. Authors have also hypothesized that reports from clinic based samples yield higher rates of developmental regression when compared to population based samples.

Objectives: Organize the literature according to common reported types of regression and provide an aggregated estimate of regression for individuals with ASD. Examine the relationship between sample size, sample type (i.e., clinical or population-based) and prevalence rates. Provide an aggregated estimate of mean age of regression reported in the published literature.

Methods: Data from 61 published studies representing 14,647 participants were coded for four regression types: language, social, language/social, and mixed. Language regression was coded if the study operationally defined regression solely as language loss. Social regression was coded if the study operationally defined regression solely in terms of social behavior, such as social smiling. Language/social was coded if regression was operationally defined as a combination of both language and social loss. Mixed regression was coded if regression was defined by (a) language and/or social regression and another type of regression, such as motor and cognitive loss, or (b) was not operationally defined in the study. Prevalence rates of regression and age of onset of regression were coded and unweighted averages for prevalence and age of onset calculated. The relationship between sample size and prevalence rates was evaluated via correlation; differences in prevalence rates between clinic and population samples were evaluated via t test.

Results: Inter-rater agreement for coding decisions was excellent (κ = 1.0 for clinic/population coding; κ = .90 for regression type). Mean prevalence rates were: (a) mixed regression, 30.27% (SD = 13.3; Mdn = 25.9%); (b) language regression, 23.38% (SD = 10.38; Mdn = 26%); (c) social regression, 14.15% (SD = 7.2%; Mdn = 15%); and (d) language/social regression, 29.40% (SD = 10.95; Mdn = 32.7%). Sample size and prevalence rates were significantly related for mixed regression (r[51] = 0.25, p<.000), language regression (r[25] = -0.30, p<.05), and social regression (r[3] = 0.93, p<.000), but not language/social regression (r[1] = 0.120, p=0.067). Clinical samples yielded higher rates of mixed (d = 1.21) and language (d = 2.14) regression when compared to population studies.

Conclusions: Prevalence rates for mixed regression and combined language/social regression converge on estimates of between 29-30% of individuals with ASD. Reported rates of language regression or social regression in isolation yield lower prevalence rates. Sample type impacted reported rates of regression, with clinical samples yielding higher rates of regression.