How Does Relaxing the Algorithm for Autism Affect DSM-V Prevalence Rates?
- First Online:
- Cite this article as:
- Matson, J.L., Hattier, M.A. & Williams, L.W. J Autism Dev Disord (2012) 42: 1549. doi:10.1007/s10803-012-1582-0
- 1.6k Views
Although it is still unclear what causes autism spectrum disorders (ASDs), over time researchers and clinicians have become more precise with detecting and diagnosing ASD. Many diagnoses, however, are based on the criteria established within the Diagnostic and Statistical Manual of Mental Disorders (DSM); thus, any change in these diagnostic criteria can have a great effect upon children with ASD and their families. It is predicted that the prevalence of ASD diagnoses will dramatically decrease with the adoption of the proposed DSM-5 criteria in 2013. The aim of this current study was to inspect the changes in prevalence first using a diagnostic criteria set which was modified slightly from the DSM-5 criteria (Modified-1 criteria) and again using a set of criteria which was relaxed even a bit more (Modified-2 criteria). Modified-1 resulted in 33.77 % fewer toddlers being diagnosed with ASD compared to the DSM-IV, while Modified-2 resulted in only a 17.98 % decrease in ASD diagnoses. Children diagnosed with the DSM-5 criteria exhibited the greatest levels of autism symptomatology, but the Mod-1, Mod-2, and DSM-IV groups still demonstrated significant impairments. Implications of these findings are discussed.
Autism is one of the most prevalent and severe disorders evident in children and adults (Matson and Kozlowski 2011). Generally considered neurodevelopmental in origin, considerable research on etiology, diagnosis, and treatment have appeared in recent years (Fernell et al. 2011; Matson and LoVullo 2009). For some time, the disorder has been considered to be on a continuum and has been defined by a triad of impairments: impaired social interaction, compromised social communication and restricted or repetitive behavior (Gould 1982; Wing 1981; Wing and Gould 1979). In the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (DSM-IV-TR), the continuum of pervasive developmental disorders, often referred to as autism spectrum disorders (ASDs), has been comprised of five separate diagnoses: autistic disorder, Asperger’s disorder, Rett’s disorder, childhood disintegrative disorder, and pervasive developmental disorder not otherwise specified (PDD-NOS).
Marked changes in the criteria for ASD have been proposed and have produced considerable controversy (Ghaziuddin 2010). With the proposed changes to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) the triad of impairments will be collapsed into a dyad, combining social communication and social interaction deficits into a single category. Additionally under the DSM-5 the current separate diagnoses under the umbrella of autism spectrum disorders will disappear, subsumed under the single proposed autism spectrum disorder. Thus, under the proposed DSM-5 changes, individuals who would have been diagnosed with Asperger’s or PDD-NOS under the DSM-IV-TR may possibly be given a diagnosis of autism spectrum disorder under the DSM-5. However, the changes to the criteria essentially raise the bar for an autism spectrum disorder diagnosis, requiring more numerous and severe symptoms in order to qualify. To qualify for a diagnosis of autistic disorder in DSM-IV-TR requires at least six of 12 symptoms, which are divided into the aforementioned autism triad: deficits in social interaction; deficits in communication; and repetitive and restricted behaviors and interests. In contrast, the proposed DSM-5 criteria divides seven symptoms of ASD into two main groups: deficits in social communication and social interaction; and restricted, repetitive behaviors and interests. Wing et al. (2011) and others have questioned the nature and magnitude of the proposed modifications to the diagnostic criteria, fearing that some individuals with autism may be unrecognized or misdiagnosed under the new manual.
A dimension which has considerable importance for the diagnosis of autism is the algorithm used to establish cutoffs for the disorder. The type and number of symptoms clinicians look for when diagnosing autism often determines access to social and educational services. McPartland et al. (2012), Matson, Belva et al. (2012), Matson, Kozlowski et al. (2012), and Worley and Matson (2012) have completed four studies comparing DSM-IV-TR and DSM-5 criteria for autism/autism spectrum disorders (ASDs). All of these studies show that according to the proposed algorithm, 30–45 % of children, adolescents, and adults classified with ASDs according to DSM-IV-TR criteria will not meet DSM-5 criteria for ASD. Given the cost of services for this group of persons, and the failure to meet mandates for educational and insurance coverage under the new diagnostic criteria, there will be a dramatic impact on many families worldwide as a result of this change. A study by Frazier et al. (2012) suggests the increased specificity of proposed DSM-5 criteria relative to DSM-IV-TR may reduce false positive diagnoses. Nonetheless, these researchers found that without a relaxed algorithm, as many as 12 % of ASD-affected individuals, particularly females, will be missed. Frazier et al. (2012) conclude that phase II testing of DSM-5 should consider a relaxed algorithm, which may improve specificity of ASD identification over the DSM-IV-TR without missing as many affected individuals as the unmodified proposal.
The purpose of the current study was to explore different modifications of the proposed DSM-5 algorithm to evaluate the impact on prevalence of diagnosis relative to DSM-IV-TR. This approach seems prudent since previous studies have shown that changing criteria can dramatically affect prevalence rates (Frazier et al. 2012; Hertz-Picciotto 2009).
This is an extension of a study by Matson, Kozlowski et al. (2012) which showed a significant decrease in the number of children diagnosed with ASDs under the proposed DSM-5 criteria. Accordingly, the same sample pool was used. Participants in this sample were identified through Louisiana’s EarlySteps program. Physicians and other healthcare professionals are mandated to refer to EarlySteps any children between the ages of birth to three with a suspected developmental delay; additionally, anyone may make a referral to EarlySteps. Under the Individuals with Disabilities Education Act, Part C, this program provides early intervention services to children with developmental delays/disabilities, and their families, from birth to 36 months of age. Because of this broad, state-wide screening process, it is believed that this sample in this study is representative of toddlers with developmental delays in this state, and that the diagnoses within this sample is representative of prevalence rates of autism spectrum disorders in this population. The sample in this study included 2,493 caregivers and their toddlers 17–36 months of age (M = 25.88, SD = 4.77) who were receiving services through the State of Louisiana’s EarlySteps program. Most of the caregivers who participated were biological mothers, but the sample also included biological fathers, foster or adoptive parents, grandparents, or other relatives. A wide variety of medical diagnoses were present in the children receiving services, including but not limited to: cerebral palsy, infant diabetes, epilepsy, arthrogryposis, bronchial-pulmonary dysplasia, neurofibromatosis, deafness, blindness, asthma, tubular sclerosis, hypoplastic left heart syndrome, and premature birth. Diagnoses also included Klinefelter’s syndrome, developmental delay, mild intellectual disability, ASDs, and pervasive developmental disorder not otherwise specified (PDD-NOS).
Participant demographics by diagnostic group
DSM-5 (n = 404)
Mod-1 (n = 512)
Mod-2 (n = 634)
DSM-IV (n = 773)
No ASD (n = 1720)
Mean age months (SD)
Other reported diagnoses (%)
Baby and Infant Screen for Children with aUtIsm Traits-Part 1 (BISCUIT-Part 1)
The BISCUIT is a three part battery of assessments used to assess ASD symptomatology, comorbidity, and challenging behaviors (Matson et al. 2007). This study utilized Part 1 (which is used to diagnose autism and PDD-NOS), in which informants use a 3-point Likert scale to rate 62 items assessing how the child compares to typically developing children of the same age: 0 (not different; no impairment), 1 (somewhat different; mild impairment), or 2 (very different; severe impairment). On the BISCUIT-Part 1, total scores below 17 fall into the “No ASD/Atypical Development” range, scores between 18 and 34 fall into the “Possible ASD/PDD-NOS” and scores at or above 35 fall into the “Probable ASD/Autistic Disorder” range (Matson et al. 2009b). A factor analysis of the BISCUIT-Part 1 revealed three distinct factors: socialization/non-verbal communication, repetitive behaviors/restricted interests, and communication (Matson et al. 2010). The BISCUIT-Part 1 was found to have an overall correct classification rate of .89, with an internal reliability of .97 (Matson et al. 2009a, b). While intellectual disability or other delays are not uncommon in individuals with pervasive developmental disorders, those identified as having a PDD/autism spectrum disorder exhibited symptomatology above and beyond what children with intellectual disability alone would exhibit, as assessed using a combination of frequency and severity to provide a total score of impairment. When differentiating children without an ASD diagnosis from those with PDD-NOS, the BISCUIT-Part 1 has been found to have high sensitivity and specificity at 0.847 and 0.864, respectively. Sensitivity and specificity were also high (0.844 and 0.833, respectively) when differentiating between those with PDD-NOS from Autistic Disorder (Matson et al. 2009b).
Modified Checklist for Autism in Toddlers (M-CHAT)
Developed as a quick screener to be easily administered by pediatricians and other health care professionals, the M-CHAT is an informant-based ASD assessment. Caregivers report “yes” or “no” to 23 questions regarding the child’s typical functioning (Robins et al. 2001). If three or more items are failed, then the screen is considered positive indicating that further assessment may be needed. The M-CHAT also includes six critical items, and a failure of at least two of these six critical items also results in a positive screen. An internal reliability of .85 has been reported, with an internal reliability of .83 for the critical items alone (Robins et al. 2001). Sensitivity and specificity were found to be .87 and .99, respectively (Robins et al. 2001).
Battelle Developmental Inventory, Second Edition (BDI-2)
The BDI-2 was developed to assess personal/social, adaptive, motor, communication, and cognitive development in children from birth to 7 years 11 months (Newborg 2005). The 450 items use a Likert scale in which 0 = no ability in this skill, 1 = emerging ability, and 2 = ability at this skill. A developmental quotient (DQ) is calculated for each domain; the inventory also yields a total DQ with a mean of 100 and standard deviation of 15. This measure has been found to have sound psychometric properties. Test–retest reliability was tested 2–25 days after the initial assessment in a group of 4 years old and a group of 2 years old children; test–retest reliability was above .80 for all domain and total scores (Alfonso et al. 2010). Overall DQ stability was excellent: .93 for the 2 years old group, and .94 for the 4 years old children (Baton and Spiker 2007). Internal consistency coefficients ranged from .98 to .99 (Newborg 2005).
Validity has been established in several populations, including those with ASD and other developmental delays (Newborg 2005). A series of studies compared mean differences in BDI-2 scores and effect sizes groups of children with identified disabilities or at risk for developing disabilities including autism, cognitive delays, developmental delays, motor delays, premature birth, and speech/language delays. These groups were compared to matched children from the standardization sample, whose members demonstrated more typical developmental patterns. Sensitivity in detecting those with the expected delay or disability at the one standard deviation below the mean level ranged from .75 to .91, with sensitivity below .80 for groups with more heterogeneously developing populations (developmental, motor, and speech/language delay groups). Specificity to avoid falsely identifying a typically developing child ranged from .75 to .91, with the motor delay and speech/language delay groups having lower specificity (at or below .80) (Baton and Spiker 2007). A study by Elbaum et al. (2010) found sensitivity and specificity values which matched or exceeded those of many other developmental screening tests commonly reported in literature. Overall, the test was found to have adequate psychometric properties and to serve as a suitable measure of childhood development (Athanasiou 2007; Bliss 2007).
Testers and Test Administration
Prior to test administration, this study was approved by the Louisiana State University Institutional Review Board and the state of Louisiana’s Office for Citizens with Developmental Disabilities (OCDD). Informed consent was then obtained from informants, all of whom were parents or legal guardians of the participating children. Assessment included one-to-one parent interviews followed by child observations. All assessments and observations occurred in the child’s home or daycare setting. Approximately 175 assessors served as test administrators, each of whom had an existing caseload and held an appropriate degree and certification or licensure to qualify for service provision in the State of Louisiana’s EarlySteps program. Licensures and certifications included the fields of occupational therapy, physical therapy, special education, social work, speech-language pathology, and psychology; degrees ranged from bachelor’s degrees in early childhood education to doctoral degrees in psychology. Additionally, each tester had experience in assessment and intervention with young children and with the assessment scales previously described.
All ASD diagnoses were made by a licensed doctoral level psychologist with over 30 years of experience in the field of developmental disabilities (the first author). The clinician had access to the information garnered from the assessment administrations as well as from record review, but was blind to the actual BISCUIT-Part 1 scores and made diagnoses based on clinical judgment using the DSM-IV-TR and the DSM-5 criteria for Autistic Disorder, DSM-IV-TR rubric for PDD-NOS, M-CHAT scores, and developmental profile scores from the BDI-2. Data were coded to obscure identity of the children, and the two DSM diagnoses were made months apart with the diagnoses for DSM-5 made blind to the DSM-IV-TR diagnoses. On DSM-IV-TR, a second Ph.D. level clinical psychologist with experience in the assessment and treatment of children with developmental disabilities provided diagnoses for a subset of the sample (n = 97) to measure the reliability of diagnoses. This clinician was also made blind to any previous diagnoses and assigned diagnoses based upon the same aforementioned information. Interrater reliability was found to be high (kappa value of .98, p < .001; percent agreement = 98.97).
As previously stated, all participants with missing or improperly coded data were excluded from the sample. A priori analyses were first conducted to determine if diagnostic groups significantly differed on demographic variables, including gender, ethnicity, and age. Chi-square analyses determined that there were no significant relationships between gender and diagnostic group, χ2 (4) = 6.68, p = .154, or between ethnicity and diagnostic group, χ2 (12) = 19.46, p = .078. An analysis of variance (ANOVA), however, found diagnostic groups to significantly differ on age, F(4, 2488) = 3.84, p = .004. After an inspection of the Bonferroni post hoc tests, the No ASD group (M = 25.67, SD = 4.78) was found to be significantly younger than the DSM-5 group (M = 26.66, SD = 4.58). This was the only statistically significant difference amongst all group comparisons on age. To test if age was significantly correlated with the dependent variable of the main analyses (i.e., BISCUIT-Part 1 total score), a correlation analysis was conducted. There was no significant relationship between the ages of the participants and their BISCUIT-Part 1 total score, r = 0.013, p = 0.513; hence, age was not entered as a covariate in the main analyses.
Means and standard deviations of BISCUIT-part 1 factor scores by diagnostic group
Repetitive behavior/restricted interests
Autism is a neurodevelopmental disorder which has seen an evolution in how it is defined. Previous modifications of diagnostic criteria (e.g., DSM-IV to DSM-IV-TR) have resulted in marked effects on prevalence rates for autism (King and Bearman 2009). Thus, studies showing large changes in prevalence rates from the DSM-IV-TR to the DSM-5, which are arguably more radical than previous changes in DSM autism criteria, should not be surprising. If other proposed changes to the DSM-5 remain, it is likely that many children now diagnosed as having an autism spectrum disorder will end up in the communication diagnostic category. This number may be even larger with older children where Aspergers is more likely to be a diagnosis. However, this is an empirical question which awaits further study; even if individuals previously categorized as having an ASD fall into this new category, it is impossible to surmise what treatments one would qualify for with this diagnosis, and the possibility remains that individuals with ASD may go undiagnosed. One potential cause of these major differences in prevalence rates between the DSM-IV-TR and DSM-5 could be the proposed conceptual changes in the substance of the definition (Wing 1981; Wing et al. 2011). A second method to moderate the dramatic estimated changes in prevalence rates may be to relax the DSM-5 algorithm. Based on this hypothesis, we have relaxed the algorithm slightly (i.e., Modified-1 criteria) and then more dramatically (i.e., Modified-2 criteria) to assess the effects of these changes on ASD prevalence rates.
It was found that if the DSM-5 criteria were slightly relaxed (Modified-1) and applied to a sample of children currently meeting DSM-IV-TR diagnostic criteria for ASD, a 33.77 % decrease in the prevalence of ASD can be estimated. DSM-IV-TR ASD diagnoses were also estimated to drop only 17.98 % should the new DSM-5 criteria be loosened even further (Modified-2). While these rates may still have adverse effects on some children with an ASD diagnosis and their families, these changes are much less dramatic than currently proposed DSM -5 criteria (Matson, Kozlowski et al. 2012; McPartland et al. 2012; Worley and Matson 2012). Nevertheless, despite the fact that greater autism symptomatology was exhibited by the DSM-5 group, children who meet criteria for ASD according to the Modified-1 criteria, the Modified-2 criteria, and the DSM-IV-TR still exhibit significant impairment.
Requirements for service provision are often times reliant on diagnoses made using DSM criteria. Multiple studies have now shown that many persons currently with an ASD diagnosis will no longer qualify for the diagnosis when the DSM-5 is adopted. The current study extends upon these findings by developing two possible more lax modifications to the DSM-5 criteria. The decrease in ASD prevalence declined, especially when using the less stringent modification (Modified-2). Nevertheless, it is of considerable importance that individuals who will no longer qualify for an ASD diagnosis continue to receive services since they still exhibit significant impairments.
Conflict of interests
The authors report no conflicts of interests. The authors alone are responsible for the content and writing of the paper.