Introduction

There is a longstanding discussion on the conceptualization of autism, including whether and how subdomains of symptoms should be differentiated (e.g. Constantino 2009; Happé et al. 2006) and whether clinically defined subtypes (i.e. Autistic Disorder, Asperger’s Syndrome, PDD-not otherwise specified; PDD-NOS) can validly and empirically be distinguished (e.g. Kamp-Becker et al. 2010; Walker et al. 2004; Stevens et al. 2000). With the fifth edition of the Diagnostic Statistical Manual (DSM-5) on the way, debates have been revitalized and the current conceptualization of the DSM-IV-TR (APA 2000) category of pervasive developmental disorders (PDD) is being adapted. Recent research has challenged the validity of the existing symptom domains and subtypes (e.g. Mandy et al. 2012; Frazier et al. 2012; Mandy et al. 2011), and has underscored the need for a more empirically based conceptualization. Evidence that DSM-IV-TR PDD subtypes cannot be distinguished reliably (Lord et al. 2011) and are similar in terms of prognosis and treatment needs (Witwer and Lecavalier 2008; Lord et al. 2006) has led to the proposal to merge these subtypes into a single autism spectrum disorder (ASD) category in DSM-5. Factor analytic evidence (e.g. Gotham et al. 2007, 2008) has shown the value of replacing the current triad of autistic symptoms with a dyad, comprising of persistent deficits in social communication and interaction (SCI) as well as restricted, repetitive behaviors (RRB). According to the proposed DSM-5 criteria (American Psychiatric Association 2012), a diagnosis of ASD would only be made in the presence of clinically significant difficulties in both symptom domains.

The proposed DSM-5 criteria have been subjected to various empirical examinations (e.g. McPartland et al. 2012; Mandy et al. 2012; Frazier et al. 2012) and have attracted large media attention (e.g. Carey 2012). Concerns have arisen that the DSM-5 criteria may exclude a substantial proportion of individuals from a diagnosis who currently do meet DSM-IV-TR criteria for a PDD. In particular, cognitively able individuals who are currently usually classified as having PDD-NOS or Asperger’s Syndrome often show deficits in SCI, but few symptoms of RRB (Mandy et al. 2011) and a considerable proportion of these individuals may be excluded from an ASD diagnosis in DSM-5 (e.g. McPartland et al. 2012). Many patients, parents and professionals worry that cognitively able individuals with PDD-NOS or Asperger’s Syndrome will no longer receive a formal diagnosis under DSM-5, which could exclude them from eligibility for funding for clinical and/or educational support (Skuse 2012; Kaland 2011).

In an attempt to contribute to this debate, the current study used the empirically based approach of latent profile analysis to further clarify phenotypic profiles within PDD. We performed this study in a sample consisting of mainly cognitively able individuals that received a DSM-IV-TR diagnosis of PDD-NOS, since biggest worries concern this relatively understudied population. We want to emphasize that we did not assess actual DSM-5 criteria, but we considered empirically based phenotypic profiles in the light of the proposed DSM-5 conceptualization of ASD. Since individuals on the autism spectrum vary not only with regard to their level and type of core autistic traits and concerning cognitive ability (e.g. Stevens et al. 2000; Witwer and Lecavalier 2008; Eagle et al. 2010), but also in terms of co-occurring behavioral and emotional problems (e.g. Leyfer et al. 2006; Pearson et al. 2006; De Bruin et al. 2007; Simonoff et al. 2008; Gadow et al. 2004), we felt it would be useful to empirically consider all these characteristics from a dimensional standpoint. Co-occurring behavioral and emotional problems can severely influence daily functioning (Matson and Nebel-Schwalm 2007), thus charting these co-occurring problems can inform us on whether people on the ‘less extreme end of the autism spectrum’ have lesser or rather other clinical needs than people with full-blown ASD, which is relevant to the question of whether they should be eligible for a diagnosis and/or services. Also, information on co-occurring problems will help to elucidate whether there are meaningfully distinct groups, and whether certain individuals should be ‘split off’ into other diagnostic categories showing certain phenotypic overlap with ASD, such as the proposed category of social communication disorder (SCD; Bishop and Norbury 2002), or attention deficit hyperactivity disorder (ADHD; e.g. Nijmeijer et al. 2011). Finally, investigating co-occurring psychiatric problems enabled us to make an initial examination of the external validity of the groups we derived using latent profile analysis, by testing whether they differed on variables independent of those used to assign group membership (Robins and Guze 1970).

Taken together, the current multicenter study aimed to provide further empirical insight in the broad range of phenotypic characteristics of individuals currently diagnosed as PDD in a modest attempt to add to the discussion on whether a DSM-5 diagnosis of ASD would be suitable for certain individuals or whether perhaps alternatives might be more appropriate. For this purpose, we empirically distinguished classes with distinctive profiles of autistic traits within a sample of 949 6–18 year old individuals diagnosed with PDD according to the DSM-IV-TR using latent profile analysis (aim 1). Subsequently, as a way of further validation, we investigated how these classes with distinct autistic traits profiles differed with regard to original PDD and axis II classification, demographic characteristics, and in particular co-occurring behavioral and emotional problems (aim 2).

Methods

Sample

The current sample consisted of 949 individuals with DSM-IV-TR diagnosis of PDD aged 6–18 (82 % male, mean age 9,3 [SD = 2.8]), who were all diagnosed at the outpatient clinics of two specialized academic departments of child- and adolescent psychiatry in the Netherlands. Both centers have expertise in the assessment of ASD and as such draw referrals from large parts of the country. Differences in ASD symptoms between cases from the two centers were of negligible effect size (Appendix 1), therefore data from both centers were combined.

All children were assessed according to the criteria of the DSM-IV-TR (please see exact diagnostic procedure below). Taken the special interest in cognitively able individuals with PDD-NOS, the majority of our sample (80 %; n = 756) had a DSM-IV-TR diagnosis of PDD-NOS (remaining 10 %; n = 91 had Asperger’s Syndrome and 11 %; n = 102 had Autistic Disorder) and the majority was cognitively able (74 %; n = 701, remaining 26 %; n = 248 had an axis II classification of intellectual disability; 317, 318, 319, V62.89, or V71.09).

Materials

Diagnostic and Demographic Characteristics

Information on age, gender and DSM-IV-TR classifications on axis I and II was obtained from two large databases that contained coded file information. In these databases, primary axis I classifications and overall axis II classifications were stored.

In both centers, a similar consensus diagnostic procedure was used, in which a multidisciplinary team decided upon classifications according to the DSM-IV-TR, based on extensive diagnostic assessment that consisted of: (1) a two-hour-session parental interview on the child’s early developmental history, medical history, and the child’s current functioning; (2) an observation of the child during a semi-structured situation. Note; in this applied clinical research project, as a basis for the axis I DSM-IV-TR classification, participants were assessed by clinicians using the Autism Diagnostic Observation Schedule (ADOS; Lord et al. 1999). Clinicians were trained by a certified ADOS professional; yet, the clinicians themselves did not all attend the official research training for the ADOS, therefore actual ADOS scores were not stored in the databases; (3) intelligence testing of the child. Note; for this, Wechsler Intelligence Scales (Vander Steene et al. 1986) were used. Only for a negligible proportion of cases scores were available in the databases. Therefore, in the current study, IQ scores were not used for analyses; rather, the axis II classifications based on these scores were used; and (4) diagnostic information and questionnaire obtained from the teacher. Note; this entailed information from telephonic interviews and the Teacher’s Report Form (TRF; Achenbach and Rescorla 2001). In the databases, TRF scores were only available for a limited proportion of cases, therefore these data are not reported on, or used in, the current investigations.

Autistic Traits

To assess the large variety of autistic traits seen in children with PDD, we used a parent-report questionnaire; the Children’s Social Behavior Questionnaire (CSBQ; e.g. Hartman et al. 2006; Luteijn et al. 2000a). This measure was developed to quantify the various problem dimensions on which children with PDD tend to differ, thus tapping the heterogeneity in this group (Luteijn et al. 1998). A second aim during the development of this instrument was to specifically include the milder part of the PDD score distribution along with the more severe autistic behaviors in one and the same instrument, thus tapping the entire autism spectrum. Parents are asked to rate behaviors that refer directly to DSM-IV-TR criteria for autism, behaviors that represent less severe operationalizations of these criteria, and ASD-associated problems. Thus, the CSBQ does not strictly assess DSM criteria of ASD, but rather specific behaviors that are more concrete exemplars of these criteria to improve understandability for the parent. For the exact item content of the CSBQ, please see Table 1.

Table 1 Item content of the CSBQ subscales
Full size table

In its original form, the CSBQ contained 96 items. Revised in 2006, the 49 item CSBQ gained in specificity by removing problem items that were only marginally characteristic of ASD (Hartman et al. 2006). The 49 items are scored on a three-point scale (0 = not true, 1 = somewhat or sometimes true, 2 = very true or often true). Items are clustered into six empirically derived subscales (see Table 1). These CSBQ subscales assess specific ASD problem domains, i.e. they each tap into a particular conceptual construct, which can be summarized as follows: The scale ‘Reduced contact and social interest’ (i.e. Social) assesses aspects related to social contact, social interest, and social reciprocity. It refers to limited initiation of contact as well as reduced response to social overtures by others. The scale ‘Difficulties in understanding social information’ (i.e. Understanding) assesses difficulties in understanding the rules of communication as well as own social use of language. In other words, it charts problems in the perception of pragmatic aspects of language and communication, as well as the production of pragmatically finely tuned communication. The scale ‘Fear of and resistance to changes’ (i.e. Change) assesses difficulties with flexibly responding to changes, i.e. behaviors shown when confronted with changes, expressed as fear, panicking, resistance, or freezing. Such behaviors are regarded as higher-level RRB (e.g. Mooney et al. 2009). The scale ‘Stereotyped behavior’ (i.e. Stereotypies) assesses classical stereotyped behaviors such as flapping and swaying, as well as sensory interests such as smelling and touching objects, which are all regarded as lower-level RRB (Mooney et al. 2009). The scale ‘Not optimally tuned to the social situation’ (i.e. Tuned) measures behaviors related to daily adaptation to social situations. As such, it assesses difficulties with the regulation of emotions and behaviors, such as emotional overreacting and stubbornness/disobedience. While instances of such behavior may also be seen in typically developing children, the Tuned subscale depicts the more extreme forms manifested by children with PDD. The scale ‘Orientation problems in time, place, or activity’ (i.e. Orientation) refers to the ability to keep an overview of what goes on, what one is doing, and where one is heading. It assesses difficulties with the planning aspects of activities as well as with information processing, which are regarded as problems related to executive functioning.

Scores on the six subscales can be depicted in a profile, revealing which problem domains are predominantly present in the child and which are less prominent. Such a profile thus reveals the child’s major PDD problem areas as well as the domains that are less/not problematic, which can aid the diagnostic process as well as guide further treatment choices. The subscales Tuned and Orientation are less specific for PDD, with similar scores in children with ADHD. Thus, when focusing specifically on the most differentiating PDD core symptoms, scores on the subscales Social, Understanding, Change and Stereotypies are most insightful.

Scores on the subscales can be interpreted in relation to norms, which are available for several norm-groups (i.e. general community population, general psychiatric population, separate PDD subtypes and with/without mental retardation), for boys and girls separately, and for different age groups (Hartman et al. 2007). Based on these norms, scores can be categorized into 7 categories: very low, low, below average, average, above average, high and very high.

The CSBQ has shown good validity and a reliable factor structure (e.g. Hartman et al. 2006; Luteijn et al. 2000a; de Bildt et al. 2009; Excoffier et al. 2007). Several factor analytic studies with varying item pools indicated that the six PDD problem dimensions (i.e. subscales) that are differentiated by the CSBQ are firmly anchored in the data (Luteijn et al. 2000a; Hartman et al. 2006). That is, the PDD problem dimensions emerging from the original item pool of 96 items and from the revised version with the 49 items are highly similar. Additionally, the PDD problem dimensions also emerged from a simultaneous factor analysis of the CSBQ with Child Behavior Checklist (CBCL) items (along with additional non-PDD dimensions from the CBCL). The consistency in factor structure across different item-pools speaks to the construct validity of the problem dimensions. Indices of criterion validity are an association of the CSBQ with the Autism Behavior Checklist of .75 (Hartman et al. 2007) and associations of around -.40 for relevant CSBQ subscales with Theory of Mind ability (Blijd-Hoogewys et al. 2008).

Multiple studies have shown that the CSBQ has good psychometric properties with regard to test–retest and inter-rater reliability, internal consistency of the subscales (all reliability indices at least .75) and good criterion validity, both for high functioning children and for children with mild to moderate mental retardation (de Bildt et al. 2009; De Bildt et al. 2005; Hartman et al. 2006; Luteijn et al. 2000a, c.). Also in the current data set, internal consistency of the subscales was good (see α’s in Table 1).

For research purposes, the CSBQ has proven to be useful in genetic (Nijmeijer et al. 2010, 2011), neurocognitive (Geurts et al. 2008; Rommelse et al. 2009), behavioral (de Bildt et al. 2005; Luteijn et al. 2000b; Jaspers et al. 2012), and treatment outcome (de Bruin and Verheij 2012) studies, thus adding to its validity. The CSBQ has also aided in characterizing (sub-threshold) PDD problems in populations other than PDD, such as ADHD (Nijmeijer et al. 2008) and delinquent groups (‘t Hart-Kerkhoffs et al. 2009).

Co-occurring Behavioral and Emotional Problems

To assess co-occurring behavioral and emotional problems, the Child Behavior Checklist (CBCL; Achenbach and Rescorla 2001) was completed by the mother. The CBCL contains 118 problem items that are scored on a three-point scale (0 = not true, 1 = somewhat or sometimes true, 2 = very true or often true). These items are clustered into eight empirically based syndrome scales: ‘Anxious/Depressed’, ‘Withdrawn/Depressed’, ‘Somatic Complaints’, ‘Social Problems’, ‘Thought Problems’, ‘Attention Problems’, ‘Rule Breaking Behavior’, and Aggressive Behavior’. The good psychometric qualities of the original US CBCL were confirmed for the Dutch translation (Verhulst et al. 1996). Also in the current data set, internal consistency of the syndrome scales was sufficient (‘Anxious/Depressed’: α = .86; ‘Withdrawn/Depressed’: α = .74; ‘Somatic Complaints’: α = .72; ‘Social Problems’: α = .62; ‘Thought Problems’: α = .61; ‘Attention Problems’: α = .68; ‘Rule Breaking Behavior’: α = .63 and ‘Aggressive Behavior’: α = .89).

Statistical Analyses

As data preparation, CSBQ and CBCL total scale scores were computed by summing the item scores. For descriptive purposes, these total scale scores were related to the norm scores. For the CSBQ, age and gender specific general population norms were used, and a dichotomous variable was constructed, separating the individuals that scored in the very high range (≥ P95) from other categories. Also for the CBCL, a dichotomous variable was computed, separating individuals in the clinical range from the other categories. Total scale scores were then divided by the number of completed items on a scale, resulting in an average item score ranging from 0 to 2. This common item-level metric across scales and instruments allows for easier comparisons and was used for graphic illustrative purposes.

To determine whether classes with distinct profiles of autistic traits could be identified (aim 1), Latent Profile Analysis (LPA) was performed, using Mplus version 4.21 (Muthén and Muthén 2007). LPA is a technique that tests whether groups of individuals with similar responses on a series of scores can be identified. The primary objective of LPA is to find the smallest number of classes of individuals with distinct endorsement profiles. Models that fit a one class model, a two class model, a three class model, and so on, were analyzed in a stepwise fashion until specification of an additional class did no longer improve the fit to the data. To identify the model with the according (number of) classes that best fitted the data, the log likelihood value, Bayesian Information Criterium (BIC; Kass and Wasserman 1995), the Akaike Information Criterion (AIC), the entropy (S), and the results of the Vuong-Lo-Mendell-Rubin Likelihood ratio tests were evaluated. When considering LPA, the BIC is considered to be superior to the other information criteria (Nylund et al. 2007). In general, models with more classes have more parameters and can therefore provide a better fit to the data. This is reflected in lower log-likelihood values for models with more classes. The information criteria BIC and AIC give penalties to models with more parameters and therefore protect against unnecessary model complexity.

For descriptive purposes, CSBQ scale scores of the classes of the best fitting LPA model were compared among the classes regarding the dichotomized CSBQ variable based on the norms and differences were expressed in Cohen’s d effect sizes.

To investigate whether classes with distinct autistic trait profiles were also distinct with regard to several other characteristics (aim 2; further validation), the LPA classes were subsequently compared with regard to information that was obtained independently from the information that was used to obtain the classes, such as age, gender, cognitive ability, PDD subtype, and the CBCL syndrome scales. These comparisons were made by calculating the effect sizes (Cohen’s d) of the differences between the classes.

Results

Phenotypic Profiles

As shown in Table 2, in the LPA, the log likelihood, BIC, adjusted BIC, and AIC parameters all pointed in the same direction, favoring a six-class model. Going from a five-class to a six-class solution, the log likelihood, BIC, adjusted BIC, and AIC values all dropped significantly (p < 0.01). This means that adding a sixth class improved the model. A seven-class model did not further improve the model fit (p > 0.05). Hence, a six-class solution provided the best fit for the current data. The average latent class probability—an indicator of latent profile distinctiveness in this model—was 0.82 for class 1; 0.84 for class 2; 0.89 for class 3; 0.84 for class 4; 0.82 for class 5; and 0.79 for class 6—thus all around the value of 0.80 that is considered fitting.

Table 2 Fit indices of the Latent Profile Analyses
Full size table

Figure 1 shows the average item scores per CSBQ subscale for the six classes. Table 3 compares the CSBQ total subscale scores of the classes to the norm scores, and Table 4 reports the effect sizes of the differences in CSBQ subscale scores among the classes.

Fig. 1
figure 1

Classes revealed in the Latent Profile Analyses; Level of problems per type of trait for each class. Y axis: mean item score (range = 0–2), X-axis: CSBQ scales

Full size image
Table 3 Comparison of classes to norms scores on the CSBQ
Full size table
Table 4 Summary of differences between classes on the CSBQ subscales by effect size Cohen’s d
Full size table

Class 1 showed relatively high scores on all autistic domains (mean scores all above 1, within a range of 0–2). When relating class 1 scores to the norm scores (Table 3), between 85 and 99 % scored in the very high range of the core ASD CSBQ subscales. Differences between class 1 and classes 4, 5, and 6 were in the range of large to medium effect size (Table 4). Class 1 had most in common with classes 2 and 3 (small effect size or no difference).

Relative to the norm scores (Table 3), class 2 scored very high on Orientation (97 % very high range), Understanding (96 % very high range), Tuned (63 % very high range), and Stereotypies (58 % very high range). There were fewer cases—although still a substantial number—with very high scores on the Social subscale (49 % very high range), and very few with high scores on the Change subscale (3 % very high range). Relative to the other five classes (Table 4), class 2 was a typical “in between class”; in comparison to the highest scoring class (class 1) differences with a large effect size were those on the core autistic traits Social, Stereotypies, and Change; in comparison to the lowest scoring class (class 6), all differences were also of a large effect size.

Class 3 mostly paralleled class 1 (Fig. 1), with slightly lower scores on most subscales. The exception was the markedly lower score on Stereotypies (Table 3: only 28 % very high). Relative to the other five classes (Table 4), class 3 was second highest compared to class 1, indicating substantial autistic traits, but the RRB characterized by mainly rigidity (i.e. Change subscale), and not so much by stereotyped motor behaviors or unusual sensory behaviors (i.e. Stereotypies subscale).

Class 4 showed high scores on the subscale Change (84 % very high range, Table 3). In addition, this group showed heightened Tuned subscale scores (41 % very high range), and moderate scores on the Social subscale (Fig. 1, 40 % very high range). Very high scores on Understanding and Stereotypies were not so frequent in this class (3 and 11 % in the very high range respectively).

When considering against norm scores (Table 3), class 5 scored highest on the Orientation subscale (46 % very high), and scored moderate on the Understanding and Social subscales (37 % very high on Understanding; 34 % very high on Social). Scores were relatively low on RRB (only 16 % very high on Stereotypies and 7 % very high on Change). Although class 5 had the next to lowest scores overall, in comparison to class 6 (Table 4) differences in scores on the Social and Understanding subscales were still of a large effect.

Class 6 showed relatively low scores on all subscales. As illustrated in Fig. 1, the means across the six problem domains were relatively stable, with a mean of approximately 0.35, indicating that on average, only one out of six items was rated as a 2 (or alternatively, that 2 out of six were rated as 1), showing that according to parent-report autistic traits were only marginally present in this class.

Comparison on Diagnostic, Demographic Characteristics and on Co-occurring Behavioral and Emotional Problems

Table 5 shows the differences among the classes regarding age, gender, intellectual ability, proportions of DSM-IV-TR PDD subtypes, and proportions of scores in the clinical range on the CBCL syndrome scales. Figure 2 shows the average item scores on each of the CBCL syndrome scales for each of the classes. Tables 6 and 7 show the differences between classes on diagnostic, demographic characteristics and on co-occurring emotional and behavioral problems as indexed by effect size.

Table 5 Further characteristics of the classes
Full size table
Fig. 2
figure 2

Levels of co-occurring problems for each class. Y axis: mean item score (range = 0–2), X-axis: CBCL scales

Full size image
Table 6 Summary of differences between classes on diagnostic and demographic characteristics by effect size Cohen’s d
Full size table
Table 7 Summary of differences between classes on co-occurring emotional and behavioral problems (i.e. CBCL scale scores) by effect size Cohen’s d
Full size table

There were no differences with a large effect size between classes in terms of age, gender, cognitive ability, or DSM-IV-TR PDD subtype (Table 6). In terms of medium effect sizes, class 4 stood out as including the highest proportion of individuals diagnosed with PDD-NOS. Class 4 and 6 stood out as including the largest percentage of cognitively able individuals. Class 2 was clearly of younger age, with medium (and small relative to class 1) differences from the other classes. Finally, class 3 differed most from the other classes regarding gender ratio, with more females in this class.

As for co-occurring behavioral and emotional problems, class 1 overall had high scores on all problem domains (Fig. 2), with especially high scores on ‘Thought Problems’ (Table 5; 89 % in the clinical range), ‘Attention Problems’ (84 % clinical range), and ‘Social Problems’ (82 % clinical range).

Class 2 also had relatively high scores on all co-occurring problems (Fig. 2), but compared to class 1 there was more contrast between behavioral problems and emotional problems, with relatively higher behavioral problems (Table 5; 56 % clinical range ‘Aggressive Behavior’, 60 % clinical range ‘Rule-Breaking Behavior’, Table 7; comparable scores with class 1) than emotional problems (Table 5; 29 % clinical range ‘Anxious/Depressed’, 26 % clinical range ‘Withdrawn/Depressed’, Table 7; comparable scores with class 5, which was a moderate-to-low scoring class). In class 2 ‘Social Problems’ were also high (86 % clinical range), and ‘Thought Problems’ were moderately high (73 % clinical range).

Class 3 mainly paralleled class 1 with regard to behavioral and emotional problems (Fig. 2), with no difference on most scales (Table 7), but small differences in ‘Attention Problems’ and ‘Social Problems’ (both slightly lower), and a medium difference in ‘Thought Problems’ (Table 5; 72 % clinical range).

Class 4 distinguished itself from the other classes in such a way that problems on almost all emotional and behavioral domains were ‘moderate’ and rather equal (ranging from 25 to 57 % in the clinical range), while in the other classes behavioral problems were more profound than emotional problems.

Class 5 mainly paralleled class 1 (Fig. 2), but on a lower level with moderate scores ranging from 17 to 51 % in the clinical range.

Finally, class 6 also paralleled classes 1 and 5, but on an even lower level (10–28 % in the clinical range).

Discussion

Within a large sample of individuals with DSM-IV-TR defined PDD, of whom most were cognitively able and had a diagnosis of PDD-NOS, the current study examined which groups with distinct profiles of autistic traits could be distinguished using LPA, and what the nature of these groups was in terms of a broad range of phenotypic characteristics.

Six classes with distinct phenotypic profiles were discerned. Three of these classes (classes 4–6, together 57 % of the sample) showed phenotypic profiles that—when interpreted in the light of the proposed DSM-5 conceptualization of ASD—were not fully in line with the proposed ASD concept of clinically significant difficulties of SCI as well as RRB: One class (class 4, 15 %) showed a lot of resistance to change, but only moderate problems in social communication (i.e. Social and Understanding subscales), one class (class 5, 30 %) showed moderate social-communication problems, but only slight RRB, and one class (class 6, 12 %) showed overall low levels of parent-reported autistic traits. The other forty-three percent (classes 1–3) of our sample did exhibit a profile alike the DSM-5 conceptualization of ASD: One class (class 1, 12 %) with markedly high scores on all core ASD domains, one class (class 2, 8 %) with social-communication problems and stereotyped behaviors, but very little resistance to change, and one class (class 3, 23 %) with social-communication problems and resistance to change, but fewer stereotyped behaviors.

The phenotypic profiles of the individuals of whom the profiles were not fully alike the DSM-5 conceptualization ASD are interesting in the light of the fractioning approach towards autistic traits (Happé and Ronald 2008). The fractioning perspective on ASD clearly predicts the presence of individuals with merely social deficits (class 5) versus individuals with merely non-social deficits (class 4). In their work, Ronald et al. (2005) did not find a substantial group of individuals with merely non-social deficits. A group with such a profile could be distinguished in the current sample (i.e. class 4). Interestingly, similar to in the findings of McPartland et al. (2012), the individuals with a phenotypic profile that was not fully alike the DSM-5 conceptualization of ASD (classes 4, 5 and 6) were mainly cognitively able (75–84 %) and had a diagnosis of either PDD-NOS or Asperger’s Syndrome (93–99 %). These findings again stress the importance of being aware of the putative impact of the alterations in the DSM for cognitively able individuals with PDD-NOS or Asperger’s Syndrome and their social environment. However, we want to stress that for these individuals, alternative, perhaps more suitable diagnoses could be considered when taking a closer look at their overall phenotypic characteristics. For example, thirty percent of our sample (class 5) mainly showed social communication problems without convincing RRB. In many of these cases, the new proposed classification of Social Communication Disorder (SCD) might be an appropriate fit, given moderate difficulties in understanding pragmatic communication, with 37 % in the very high range (i.e. Understanding subscale). Bishop and Norbury (2002) also showed that there was a subset of children with mainly pragmatic difficulties that appeared rather sociable—relative to children with autism—and who had few abnormalities outside their oddities in the social communication domain. Our data also distinguished a substantial group of such individuals, for whom SCD might be a suitable alternative diagnosis. Although the proposed category of SCD has also been under scrutiny because of the danger of it becoming another residual category of ‘not quite’ ASD (Skuse 2012), if more clearly operationalized, it might be a useful category with a growing empirical basis. In the current study, almost one-third (30 %) of the total PDD sample fell into the class with a SCD-like phenotypic profile. If future studies replicate that indeed such a substantial proportion of individuals currently diagnosed with PDD fall into the SCD category, this would implicate a substantial shift from PDD to SCD classifications instead of ASD classifications. Thus, our findings certainly merit further investigations regarding this new diagnostic category.

Conversely, fifteen percent of the children (class 4) mainly showed resistance to chance without convincing SCI. As these children often also showed inadequately tuned behavior along with moderate levels of emotional and behavioral problems, it is questionable whether the emphasis should lie on their autistic traits or whether this group should be regarded as more generally ‘dysregulated’ when being under changing, unexpected, or unwanted circumstances. Since problems occur in the regulation of behavior as well as emotions, this profile may fit better in the mood disorders category. Although there is also a lot of controversy on these proposed DSM-5 categories (Axelson et al. 2011), our data support the notion that a group of children with a phenotypic profile of combined disruptive and mood regulation problems can be distinguished (class 4), and that in this group, symptoms cannot satisfactorily be explained as ASD as conceptualized in DSM-5. Clearly, more research is needed to underpin the empirical evidence for including these children in a mood disorder category such as the proposed category of Disruptive Mood Dysregulation Disorder (DMDD), but the current findings suggest that for some cases currently classified as PDD, this description could be suitable.

Our empirically derived latent class 6 showed below clinical threshold ASD traits and these children were better defined based on their attention and disruptive problems. Indeed, ASD and ADHD have been shown to have substantial phenotypic overlap (Nijmeijer et al. 2011). Thus, in a proportion of cases currently diagnosed as PDD-NOS, in the future, a primary diagnosis of ADHD may be suitable. In cases where there are many questions regarding diagnosis, especially when still young, close monitoring over time is crucial to make sure that these children do receive timely and suitable clinical attention.

Interestingly, within the individuals with a phenotypic profile in line with the DSM-5 conceptualization of ASD (classes 1–3), profile differences were revealed relating to lower-level and higher-level RRB. Twenty percent (classes 1 and 2) of these ASD cases showed significant stereotyped behaviors (lower-level RRB), and in these classes 40–44 % had a classification of intellectual disability on axis II. In contrast, twenty-three percent primarily showed resistance to change (higher-level RRB; class 3), but fewer stereotypies. In this class, a smaller percentage (26 %) had an intellectual disability. Stereotyped and motor-sensory behaviors (lower-level RRB) therefore more often seemed to go along with intellectual disability than resistance to change (higher-level RRB). This finding corroborates with previous research showing a negative association between intellectual developmental level and lower-level RRB, as well as a positive association between intellectual developmental level and higher-level RRB (for a review see Mooney et al. 2009). This finding illustrates that, given that within DSM-5 ASD subtypes will no longer be distinguished, it will become all the more important to get a full diagnostic picture of the profile of the full variety of symptoms as manifested within ASD (Ronald et al. 2005). Furthermore, our finding of classes of children with varying levels of social-communication problems and either high or low levels of RRB corroborates with the recent findings of Frazier et al. (2012) and Mandy et al. (2012), and with the proposed DSM-5 model.

With regard to the relation between autistic traits and co-occurring behavioral and emotional problems, our data generally showed a tendency of more co-occurring behavioral and emotional problems in individuals with more severe ASD traits. Previous findings in this area have been conflicting, some in line with the current findings (Kanne et al. 2009), some showing more co-morbid diagnoses in individuals on the less extreme end of the autism spectrum (e.g. Gadow et al. 2004; Pearson et al. 2006) and some showing few differences in co-morbid diagnoses between the subtypes of ASD (Snow and Lecavalier 2011). Since our investigations were cross-sectional, the relation between ASD profiles and CBCL scores can be explained in two directions; i.e. emotional and behavioral problems might exacerbate autistic traits or vice versa, autistic traits might exacerbate comorbid features. As our cross-sectional findings cannot be conclusive on this matter, longitudinal studies are needed to clarify this issue. Interestingly, although classes were discerned based on the CSBQ, and profiles of CSBQ scores of the different classes did not parallel each other, classes did mainly parallel each other when plotting CBCL scores. In other words, when considering co-occurring behavioral and emotional problems in groups with distinct ASD profiles, these co-occurring traits seem to vary along a severity dimension. Levels of attention problems (i.e. easily distracted), social problems (i.e. getting bullied), and thought problems (i.e. strange or obsessive thoughts and odd, nervous behaviours) in our sample of individuals with PDD were usually as high as the levels of core autistic traits. Such high levels of co-occurring problems warrant clinical attention, and awareness of putative co-occurring problems in ASD seems essential for further therapeutic decisions.

Again, we want to emphasize that no actual DSM-5 ASD criteria were assessed. Instead, we used data on parent-reported behaviors. This means our findings were solely based on how parents observe and judge their child’s current behavior in its natural environment, and not on clinical judgment based on observation of the child’s behavior plus other relevant multi-informant diagnostic information, and importantly, including judgment regarding the degree of impairment, i.e. the impact of the behaviors on daily functioning. Low scores on a parent-reported scale do not necessarily mean that the behaviors assessed are not associated with impairments regarding daily functioning. Perhaps, children in our study with a parent-reported phenotypic profile that is not regarded as fully in line with all DSM 5 criteria might meet DSM-5 criteria at a low severity level (i.e. ASD with a level 1 severity; DSM-5, APA: www.dsm5.org/ProposedRevisions) when based on clinical judgment also regarding degree of impairment in daily functioning. Therefore, for results based on clinical judgment of DSM-5 criteria, the DSM-5 field trials must be awaited. Since in the current study scores on diagnostic assessments such as the ADOS (Lord et al. 1999) and the Autism Diagnostic Interview-Revised (ADI-R; Rutter et al. 2003) were not available, future research including such measures will also be important.

All in all, the current multicenter study showed that a considerable amount of individuals classified as PDD in the DSM-IV-TR showed a parent-reported phenotypic profile that might not fully fit the proposed conceptualization of ASD in the DSM-5. Without a formal diagnosis, such individuals could no longer be eligible for funding for clinical and/or educational support, which could have substantial societal and public health ramifications (McPartland et al. 2012). However, when taking a closer look at the broader phenotypic characteristics of these cases, a clinical diagnosis of a mental disorder still seems warranted, and other classifications may serve the patient’s needs better. In fact, such alternative diagnoses may also result in treatment that is more appropriately aimed at the particular individual profile of symptoms. In younger cases where there is a remaining doubt and debate on a suitable diagnosis, close monitoring of development over time is recommended. Our data thus suggest that the proposed DSM-5 conceptualization of ASD may not be not overly narrow, allowing for inclusion of individuals with variation across RRB, intellectual ability, and social communication problems. We hope our findings will offer some optimism for many patients, parents, and clinicians who worry about the access to treatments and services for individuals who met criteria for PDD in the DSM-IV and may not meet ASD criteria in the proposed DSM-5.