After experiencing ER, autistic children display a developmental plateau (32, 33) that generally resolves into a late recovery of communicative language. Certain atypicalities are overrepresented in individuals for whom ER is reported by caregiver, mainly in communication atypicalities (“pronominal reversal – ever”, “never shakes head at age 4 to 5”, “hand leading – ever”, “little use of pointing to express interest at age 4 to 5”, “never nods at age 4 to 5”, and “limited comprehension of simple language at age 4 to 5”), as well as perception-based behaviors (“stereotypic use of objects or interest in parts of objects – ever”). Other atypicalities of the communicative and social areas were also overrepresented when the verbal level of the individuals was not considered, the most significant of which were "little attention to voice at age 4 to 5", "generally no reciprocal smiling at age 4 to 5", and "rarely responds to the approaches of other children at age 4 to 5". ER thus initiates an atypical developmental sequence (12, 32–34) characterized by certain atypicalities to a greater extent than that observed in non-regressive individuals. Four of these atypicalities (“hand leading – ever”, “pronominal reversal – ever”, “never shakes head at age 4 to 5”, “stereotypic use of objects or interest in parts of objects – ever”) tended to occur concomitantly more frequently among individuals with ER than among non-regressive individuals.
The combination of three or more of these four atypicalities was > 5 times more likely among autistic children who had experienced an ER. The multiple combinations of these four atypicalities, which are mostly in the domain of communication, were associated with a more pronounced autistic phenotype, both in terms of social skills and in terms of repetitive behaviors and restricted interests retrospectively and at the time of assessment. This cluster of atypicalities was not associated with worse phonological memory deficits on the non-word repetition task, although they tended to be associated with more limited vocabulary knowledge. Phonological memory is usually preserved in autistic children (35). These deficits are instead found mainly in non-autistic neurodevelopmental disorders (36, 37).
Critically, the combination of these atypicalities increased the odds of a clinical diagnosis of autistic disorder rather than another diagnosis on the PDD spectrum. The combination of a minimum of two of these atypicalities, as opposed to the single manifestation or absence of one of these four atypicalities, increased the odds of obtaining a diagnosis of autistic disorder by a factor of three. This group of atypicalities, mostly retrospectively reported in the period of 4 to 5 years of age, therefore cluster with an overall phenotypic presentation that remains frank at the age of enrollment in our sample (6 to 18 years). Overall, the co-presence of at least two atypicalities in this cluster of four atypicalities constitutes a semiological pattern, which is confirmed by the ADI and ADOS scores, and the clinician's expertise in assessing the overall phenotypic presentation later in development.
Clinical importance of reported autistic early regression
Overall, the aggregation and contingency of the selected atypicalities, even transient, show their semiological importance in the developmental context that accompanies ER, despite their actual equivalent value in the current system of polythetic criteria. Approximately 40% of the verbal autistic children in our sample shared at least two of the four selected atypicalities.
The strategy used in this study emphasizes the interdependence between a developmental trajectory, usually marked by ER, and the transient presentation of cross-sectional signs. Certain behavioral manifestations may emerge around the same period as the identification of ER, such as the atypical use of objects (17, 33, 38), which is strongly associated with regression in our study. Other atypicalities emerge later in development, accentuating the oddity of language and communicative development, without affecting the final adaptive outcome. By granting a “specifier” status to "language impairment" and ignoring ER in the criteria in the DSM 5, the current diagnostic formulation underscores the semiological and nosological value of the codependence between the developmental period and the manifestation of signs and may undermine the validity of the diagnostic construct (39). Better integration of developmental trajectories with autistic atypicalities could therefore address the increasing heterogeneity and phenotypic ambiguity of the current criteria (40).
Limitations
The sample on which this research was conducted is likely to underrepresent children with autism with an intellectual disability. Autistic children with a low nonverbal IQ are less likely to become verbal (41), which amplifies the underrepresentation of these children in some of the analyses of this study.
This is a cross-sectional study, which assumes an equivalent value for the retrospective data used to characterize the participants. However, the time period between the event and its recall is likely to influence the reliability of the measure (8, 42). Knowledge about the diagnosis is able to influence recall, with a well-informed parent reporting more difficulties in their child's development (43). This bias is likely to amplify the observed combination of ER-associated atypicalities, but does not invalidate the association found between the copresence of signs .
The choice of using the ADI-R criteria as a definition of ER is questionable. This study does not address the phenotypic distinction that could be associated with different types of regression. Information about ER is obtained retrospectively, which, although not very sensitive, is conservative.
The use of a stepwise analysis to identify the atypicalities best associated with ER is a data-driven type of analysis. As such, the results are dependent on the sample used and the strength of the associations and their validity may be influenced by selection bias. This study was limited to a single cohort; replication in another cohort would add validity to the selected group of atypicalities. The bootstrap sensitivity analysis is reassuring with respect to the consistency of the variable selection.
The heterogeneity of individuals included in the autism spectrum as currently defined may seem over-inclusive or abstract in clinic and research. The integration of a relatively specific longitudinal dimension, such as regression or plateau, to the diagnostic criteria could represent a way to isolate a homogeneous group manifesting key atypicalities within the current autism spectrum to which more heterogeneous individuals can be secondarily compared. Based on the recent arguing of the critical role of ER in autism, we have used stepwise regression to identify four atypicalities of which their presence is both associated with a higher severity of the autistic areas and the categorical evidence of the diagnosis. This retrospective study thus represents a first step in identifying the combination of atypicalities associated with the regressive phenomenon. This method, applied to the whole inventory of autistic signs, could contribute to limit the continuous increase of the heterogeneity of the clinical pictures accepted within the spectrum.