Simple Identification of Complex ADHD Subtypes Using Current Symptom Counts

Introduction

Current diagnostic criteria set forth by the Diagnostic and Statistical Manual of Mental Disorders, Version IV (DSM-IV) divide ADHD into three subtypes: predominantly inattentive, predominantly hyperactive-impulsive and combined, based on the presence of at least 6 of 9 inattentive symptoms, 6 of 9 hyperactive/impulsive symptoms, or both.¹ Additional diagnostic criteria include impairment at school or home before age 7 and absence of other diagnoses that could explain the ADHD symptoms, such as autism. The definition of these DSM-IV thresholds incorporated measures of academic and social impairment, Children's Global Assessment Scale score and clinician DSM-III-R ADHD diagnosis.^{2, 3}

Original psychometric analyses of the 18 ADHD symptoms with factor analysis identified two highly correlated underlying factors (inattention and hyperactivity/impulsivity).⁴ In contrast to factor analysis, which assumes a continuum of disease, latent class analysis (LCA) is a version of mixture modeling that aims to separate individuals into discrete, phenotypically homogenous subgroups. From the standard 18 ADHD symptoms, LCA consistently identifies 7 to 8 highly heritable and distinct classes in several population-based samples^5-7: mild and severe inattentive subtypes, mild and severe combined subtypes, mild and severe hyperactive-impulsive subtypes, a group of subjects with few symptoms, and a final, less defined group. Three of these latent classes (severe inattentive, severe hyperactive-impulsive, and severe combined) demonstrate modest (severe inattentive) to excellent (severe combined) agreement with the similarly named DSM-IV ADHD subtype.^5-9

Compared to DSM-IV ADHD subtypes, the population based LCA subtypes could offer several advantages, in particular in being more phenotypically homogeneous and genetically distinct.^{8, 9} Agreement is high across maternal, paternal and teacher reports, especially for the severe classes.¹⁰ Recently, we showed that LCA-derived subtypes are more stable at five-year follow up than those in DSM-IV¹¹ and that the DSM-IV age of onset before 7 years criterion is unreliable.¹¹ The severe inattentive subtype is more homogeneous, averaging fewer than 2 endorsed hyperactive-impulsive symptoms, while the DSM-IV definition of the predominately inattentive type allows up to 5. As with their DSM-IV counterparts, the LCA severe inattentive and severe combined subtypes are associated with marked impairment in academic and social settings. As judged by Child Behavior Checklist (CBCL) criteria ¹², these subtypes are characterized by increased attention problems and social problems subscales.¹³ The severe inattentive and combined subtypes are more likely to have poor grades and to have had an Individual Education Plan (school program to address specific special education needs) compared to DSM-IV and other population-defined subtypes.¹⁴ These ADHD subtypes also have high levels of autistic symptoms and motor problems, especially the severe combined type.^{15, 16}

In the present study, we provide a direct comparison of DSM-IV and LCA subtypes, with a view toward addressing a shortcoming to the adoption of this alternative classification approach in clinical practice: translating the many possible symptom endorsement profiles of the 18 DSM-IV ADHD symptoms to specific ADHD subtype assignments. For LCA-derived subtypes to be useful clinically, the main motivation for such research, the novel subtypes must be easy to identify. We thus describe the utility of a simple ADHD symptom counting approach to assigning youth to severe ADHD subtypes for the DSM-IV and latent class classification systems. The findings suggest that revisions in nosology, such as anticipated in DSM-V, can incorporate results from a variety of approaches that produce simple assignment algorithms practical for clinical applications.

Method

Results

Discussion

These results show that it is possible to identify clinically relevant latent class ADHD subtypes using symptom counts only (Tables 4 and 5). Symptom utility estimates from the DSM-IV field trials indicated similar predictive ability for all 18 ADHD symptoms. Inattentive and hyperactive-impulsive items were found to better identify subjects with those named characteristics (ex: inattentive symptom 1 with primarily inattentive type) than cross-dimension (ex: inattentive symptom 1 with primarily hyperactive/impulsive type).²⁸ This would argue for the retention of all 18 symptoms in future revisions of diagnostic schedules, such as DSM-V or ICD-11. The present analysis confirms recommendation, as the factor analysis found that all symptoms have similar factor loadings. This allows the derivation of a simple symptom count algorithm to identify ADHD subtypes derived from more complex statistical approaches and their translation into criteria that are easily applicable to real life clinical situations. Briefly, a total of 6 or more symptoms, of which fewer than three are hyperactive-impulsive, identify the severe inattentive subtype. A total of eleven or more, with 4 or more hyperactive-impulsive symptoms identify the severe combined subtype. Fewer than 6 total symptoms identify the absence of clinically significant ADHD. Consideration of these findings may help increase the utility of future diagnostic schema for ADHD.

The results presented here expand on earlier efforts to identify, in population samples, homogeneous ADHD subtypes with significant clinical impairments. The correspondence between the latent class defined ADHD subtypes and the corresponding DSM-IV subtypes is imperfect (Table 2). On the one hand, severe inattentive latent class subjects are almost perfectly classified as predominantly inattentive subtype by DSM-IV criteria (N=377 out of 379). On the other, of the 755 DSM-IV predominantly inattentive subtype subjects in our sample, a large number fall into the severe combined, mild combined and mild inattentive latent classes (N = 102, 210 and 66, respectively). The DSM-IV predominantly inattentive type allows endorsement of up to 5 hyperactive-impulsive symptoms without meeting criterion, while the latent class severe inattentive definition allows only up to two. The ratio of inattentive to hyperactive-impulsive symptoms appears to be important as the “pure” inattentive subjects endorsed fewer hyperactive-impulsive symptoms than the DSM-IV predominately inattentive subtype. Many of the subjects classified as mild combined by LCA and members of a clinically relevant subtype would have been diagnosed as predominantly inattentive by DSM-IV criteria, resulting in a less homogenous group. This may explain in part why twin studies suggest that the severe inattentive latent class is a genetically distinct subtype of ADHD, while results for the DSM-IV predominately inattentive subtype are equivocal.^{8, 24} Using population-defined criteria for the severe inattentive subtype may be especially valuable in assessing genetic risk factors for this disease. The alternative cut-points presented here capitalize on this difference and provide a method to identify this under-diagnosed, yet impaired, subtype.¹³

Principal Component Analysis of this very large sample suggests two factors (symptom number and symptom type) in the DSM-IV ADHD symptom structure (Table 3). This is fully consistent with previous factor analyses of DSM-IV ADHD symptom data that suggested two correlated factors, one inattentive and one hyperactive-impulsive. ⁴ A simple rotation of the uncorrelated symptom number factor and the type factor recreates the correlated inattention and hyperactive- impulsive factors described previously. The use of the uncorrelated factor structure rather than the previously identified correlated inattentive and hyperactive factors allows easy visual identification of ADHD subtypes by either classification. Subjects belonging to the clinically relevant and markedly impaired severe inattentive and severe combined latent classes cluster in distinct areas on a plot of these two underlying factors.

This is not the first empirical study of ADHD nosology. Faraone and colleagues used ROC analysis to examine the efficiency of diagnostic criteria creating limited symptom clusters for DSM-III-R ADD subjects versus subjects with ADD and comorbidity.²⁹ Similar methods have been employed to predict impairment in subjects with DSM-IV ADHD.³⁰

Limitations of this study include using only symptoms from the DSM-IV definition of ADHD. There may be additional symptoms that would further discriminate between syndrome appearance and severity. For example, items describing sluggish cognitive tempo (ex: daydreams) were included as potential ADHD symptoms during the DSM-IV field trials. The inclusion of sluggish cognitive tempo symptoms does not significantly alter the structure of the latent class derived ADHD subtypes.³¹ Additionally, we did not include symptoms of disorders commonly comorbid with ADHD in this analysis since previous work did not indicate differential clustering by ADHD latent class subtype.⁹

Another limitation of the study is that, due to its low prevalence, we could not meaningfully analyze the predominately hyperactive-impulsive ADHD subtype. No age or sex specific criteria were incorporated in the development of ADHD subtypes for both nosologies, but our results indicate that the classification rule remains accurate across gender, country and method of data collection.

Further, the data come from three twin and family samples. Inclusion of non-twin siblings and random allocation of one number of each family to a training set and the remainder to validation sets in the current analyses helps avoid biases in results due to correlations between twins and siblings. While the prevalence of DSM-IV and latent class subtypes in twins may vary from the general population, no comparable studies have been conducted in non-twin samples. As noted in Neuman et al¹⁹ the prevalences of ADHD subtypes in Australian non-twin siblings of twins are similar to those reported in Missouri twins.^{7, 19, 24} Similarly, no differences in the prevalence or familiality of these ADHD subtypes were found between non-twin sibling pairs and dizygotic twins in the Australian sample.²⁴ Additionally, the prevalences of ADHD and individual ADHD subtypes in the ATAP sample were similar to those in an Australian national epidemiologic sample. Finally, the MOTWIN study is slightly enriched with DSM-IV primarily inattentive subtype subjects. The effects on the present results are limited since the screening instrument used for selection demonstrated low PPV ¹⁹ and the key step of assigning subjects to a particular latent class was done using parameter estimates derived from population representative samples.

Despite these limitations, our study has several strengths. These analyses were conducted on three large, international population-based samples of subjects from Missouri and Australia. Overall, this sample is representative of the twin populations and the data were collected using highly reliable diagnostic interviews (MOAFT and MOTWIN) and questionnaires (ATAP). Because the standard 18 ADHD symptoms from the DSM-IV ADHD diagnosis were used to create alternative symptom cutoffs, our analysis can thus easily be replicated by others. As demonstrated in the present study, clinically significant ADHD latent class subtypes can be assigned by both researchers and clinicians using parent report of the current presence or absence of the 18 DSM-IV symptoms that are available on a variety of widely used screening questionnaires and diagnostic interviews.