Journal of Autism and Developmental Disorders, Vol. 35, No.

3, June 2005 (
2005)

DOI: 10.1007/s10803-005-3300-7

Screening Adults for Asperger Syndrome Using the AQ:

A Preliminary Study of its Diagnostic Validity in Clinical
Practice

M. R. Woodbury-Smith,2,3 J. Robinson,1 S. Wheelwright,2 and S. Baron-Cohen2

The Autism Spectrum Quotient (AQ) has been developed to measure the degree to which an
adult with normal intelligence has autistic traits. In this paper it is evaluated for its potential as
a screening questionnaire in clinical practice on one hundred consecutive referrals to a
diagnostic clinic for adults suspected of having Asperger Syndrome or high functioning autism
(AS/HFA). The results indicate that it has good discriminative validity and good screening
properties at a threshold score of 26. The implications of these results are discussed.

KEY WORDS: Asperger Syndrome; Autism Quotient; high-functioning autism; screening; diagnosis;
validity.

Asperger Syndrome (AS) is now widely believed The diagnosis of AS is often a difficult task, such
to lie on the autistic spectrum of conditions (Leek- that it is often delayed until late childhood or even
ham, Libby, Wing, Gould, & Gillberg, 2000; Mayes, early adulthood (Barnard, Harvey, Prior, & Potter,
Calhoun, & Crites, 2001; Ozonoff, South, & Miller, 2001; Howlin and Moore, 1997; Powell, 2002).
2000; Wing, 1997), differing from ‘classical autism’ in Although several different diagnostic instruments
terms of normal language development and intellec- for autistic spectrum conditions, including Asperger
tual ability, and in terms of higher prevalence, with Syndrome, have now been developed, the most
one epidemiological study estimating a population widely used probably being the Autism Diagnostic
prevalence of 0.7% (Ehlers and Gillberg, 1993). In the Interview—Revised (Lord, Rutter, & Le Couteur,
absence of normal early language development, high 1994), the development of screening questionnaires is
functioning autism (HFA) is a possible diagnosis. important. This is primarily because all diagnostic
The exact relationship between AS and HFA is interviews involve a lengthy assessment, in the region
unclear, and, in view of the similarities in clinical of 3 hours. It is therefore advantageous that only
presentation between AS and HFA, Asperger Syn- those who are highly likely to have an AS go through
drome will be used in the remainder of this paper to this diagnostic process. To this end, the Asperger
loosely apply to both. Syndrome Screening Questionnaire (ASSQ, Ehlers,
Gillberg, & Wing, 1999) was developed for possible
1
Cambridge Lifespan Asperger Syndrome Service and Autism
cases of Asperger Syndrome. Unfortunately this
Research Centre, Cambridge, UK. screening questionnaire is designed for use with
2
Departments of Psychiatry and Experimental Psychology, Uni- school-age children.
versity of Cambridge, Cambridge, UK. There are reasons to believe that the develop-
3
Correspondence should be addressed to: Dr Marc Woodbury- ment of a screening questionnaire specifically for use
Smith Child Study Center Yale University School of Medicine
230 South Frontage Road PO Box 207900 New Haven, CT 06520
by adults with suspected AS is also important. First,
USA Email: marc.woodbury-smith@yale.edu AS was not recognised as a clinical entity until

331
0162-3257/05/0600-0331/0
2005 Springer ScienceþBusiness Media, Inc.
332 Woodbury-Smith, Robinson, Wheelwright, and Baron-Cohen

relatively recently, with its inclusion only in the most significantly higher than controls on two of the
recent editions of the Diagnostic and Statistical subscales of the AQ (Bishop et al., 2004).
Manual of Mental Disorders (DSM-IV, American The current study evaluated the usefulness of the
Psychiatric Association, 1994) and International AQ as a screening questionnaire in clinical practice
Classification of Diseases (ICD-10, World Health by exploring its properties in 100 consecutive patients
Organisation, 1992). As a result, many adults will who had been referred to a national diagnostic clinic
have ‘missed’ diagnosis during their childhood. Sec- in the UK for adults suspected of having Asperger
ond, AS is often harder to diagnose than classical Syndrome. The aim was to determine whether the
autism, with much more subtle impairments in the questionnaire was able to usefully distinguish
core clinical phenotypic domains, making it difficult between those individuals who turned out to have
for clinicians to ascertain who should be referred for AS, and those who did not.
further assessment. Third, AS is relatively more
common that classical autism, and therefore represents
a disproportionate number of referrals. And finally, METHOD
adults with possible AS are quite likely to present to
primary care physicians who will need to be able to Participants
recognise and refer those who may have the syndrome. The sample consisted of the first 100 patients
Two adult based screening assessments are evaluated in the Cambridge Lifespan Asperger Syn-
available. One is the Australian Scale for Asperger drome Service (CLASS, median age = 32 years,
Syndrome (ASAS, Garnett and Attwood, 1995), a range 18–69; gender, male: female = 4:1). This is a
clinician rated questionnaire, which can be used for diagnostic clinic for adults, aged 18 years and over,
adult assessments. Its major drawback is its lack of suspected of having Asperger Syndrome or high
clear scoring criteria making it potentially problem- functioning autism. Referrals are accepted from all
atic as a screening questionnaire. The other, the health professions, with most referrals being from
Autism Spectrum Quotient (AQ, Baron-Cohen, general practitioners. All patients complete a short
Wheelwright, Skinner, Martin, & Clubley, 2001), is ten-item checklist following their referral to the clinic.
a brief 50 item self-administered questionnaire for This simply ensures that the patient themselves are
adults that identifies the degree to which any aware of and acknowledge that they have experienced
individual adult of normal intelligence might have the core phenotypic difficulties. It is not used for
features of the core autistic phenotype. screening purposes and does not contribute to the
The 50 questions on the AQ are made up of 10 diagnostic process. It does not include items that are
questions assessing five different areas: social skill, included in the AQ. People with a history of mental
attention switching, attention to detail, communica- retardation (learning disability) are specifically
tion and imagination. Each question allows the excluded from assessment. Furthermore, patients
respondent to answer ‘definitely agree’, ‘slightly are required to be accompanied by a suitable infor-
agree’, ‘slightly disagree’ or ‘definitely disagree’ mant who has known them throughout the develop-
according to the degree to which they believe they mental period.
exhibit the behaviour described in the item. Each item
scores one point if the respondent records the
Design
abnormal behaviour either mildly or strongly. In
order to avoid response bias, approximately half the In order to evaluate the screening properties of
items are worded to produce a ‘disagree’ response, the AQ in clinical practice, each patient referred to
and half an ‘agree’ response. The properties of the the clinic was asked to complete the Autism Quotient
AQ have been reported previously (Baron-Cohen (AQ) questionnaire (questionnaire available from
et al., 2001). In short, 80% of adults with AS or high authors on request). All patients were subsequently
functioning autism scored above a critical minimum seen for further diagnostic assessment irrespective of
of 32, whereas only 2% of control adults did. It was their scores on this. This assessment consisted of a
also shown to have good test–retest and inter-rater detailed interview by two clinicians with the patient
reliability. It is therefore potentially useful as a and their informant. At the end of the clinical
screening questionnaire. It is of interest that parents interview, both clinicians independently rated the
of children with autism (who may have the broader patient according to the DSM-IV diagnostic criteria
phenotype but not at the severity level of AS) score for Asperger Syndrome. In this way the DSM-IV
Screening Adults for Asperger Syndrome 333

Table I. Discriminative Validity of the AQ

N Mean AQ score (s.d.) t-test AUC Std. Err.

AS vs non-AS 73 35.62 (6.63) )5.59*** 0.78 0.06

27 26.22 (9.39)

***p < 0.0001.

diagnosis assigned to each patient was used as the course, other thresholds of cut off may be favoured in
yardstick against which the AQ was evaluated. different circumstances.
Several different aspects of its performance were
measured: the discriminative validity of the AQ
between adults with AS and those with sub-threshold DISCUSSION
difficulties in the core domains (i.e., those who do not
meet the diagnostic criteria for AS) as measured by This study has attempted to determine whether
parametric statistics and receiver operating curves, the AQ has acceptable properties to allow it to be
the sensitivity and specificity of the AQ at different used as a screening instrument in clinical practice. In
threshold scores, and the positive and negative a clinic population of adults referred for assessment
predictive values of the AQ. The data were analysed of possible Asperger Syndrome, all referrals com-
using Stata Version 7 (Stata Corporation, 2001). pleted the AQ and then underwent a more rigorous
Significance was set at the conventional 5% level. assessment. The AQ was shown to have good ROC
characteristics, and, at a cut-off of 26, its sensitivity
and specificity were such that 83% of patients were
RESULTS
correctly classified.
Before discussing these findings further, several
Discriminant Power of the AQ
limitations need to be mentioned. First, the clinicians
The AQ differentiated well between patients who who saw the patients in the clinic were not blind to
received a diagnosis of AS and those who did not AQ scores. This was unavoidable as the AQ was
(Table I). This was demonstrated by significant group being used as part of clinical practice. It is possible
differences as measured by parametric statistics. that this influenced their diagnostic decision, and
Additionally, the area under the ROC curve was therefore the study needs replication to ensure the
0.78 (std. err. 0.06, 95% CI 0.7–0.9), representing validity of our preliminary results. However, for each
accuracy of the AQ in the moderate range. The area case, diagnosis was based upon a detailed clinical
under the ROC is indicative of the overall accuracy of interview that allowed a DSM-IV diagnosis to be
a test, representing the probability that a randomly made independent of AQ score.
selected ‘true-positive’ individual will score higher on Another issue to be considered in the interpre-
the test than a randomly selected ‘true-negative’ tation of the AQ is the possibility that other factors
individual. might have influenced scores. For example, schizo-
phrenia is known to be associated with the pre-
morbid personality traits of impaired social interac-
Threshold Score for Most Effective Use as a Screening
tion and communication. It is possible, therefore,
Questionnaire
that this would also result in higher AQ scores, and
It has previously been suggested that in a general result in ‘false positives’ being referred on for
population study a cut-off of 32 or above should be further assessment. This clearly needs further inves-
employed for correctly identifying individuals with tigation by administering the AQ to adults with
‘autistic traits’. However, examination of the receiver schizophrenia. With regard to the clinic population,
operating characteristics for the total AQ suggested of the one hundred referrals, only two had a
that for this clinic referred sample a threshold score previous diagnosis of schizophrenia: one went on
of 26 resulted in the correct classification of the to be diagnosed with AS (with ICD-10 criteria that
greatest numbers (Table II). At this cut off the allows for such comorbidity) and scored 45 on the
sensitivity is 0.95, specificity 0.52, positive predictive AQ, the other did not have AS and had scored only
value 0.84, and negative predictive value 0.78. Of 20 on the AQ.
334 Woodbury-Smith, Robinson, Wheelwright, and Baron-Cohen

Table II. Detailed report of diagnostic statistics for the Autism Spectrum Quotient (AQ)

Cut-off point Sensitivity (%) Specificity (%) Correctly classified (%)

‡10 100.00 0.00 73.00

‡11 100.00 3.70 74.00
‡13 100.00 7.41 75.00
‡15 100.00 11.11 76.00
‡16 100.00 14.81 77.00
‡19 98.63 18.52 77.00
‡20 98.63 22.22 78.00
‡21 98.63 29.63 80.00
‡23 94.52 29.63 77.00
‡25 94.52 44.44 81.00
‡26 94.52 51.85 83.00
‡27 93.15 51.85 82.00
‡28 87.67 59.26 80.00
‡29 84.93 66.67 80.00
‡30 83.56 70.37 80.00
‡31 79.45 70.37 77.00
‡32 76.71 74.07 76.00
‡33 73.97 74.07 74.00
‡34 65.75 74.07 68.00
‡35 60.27 77.78 65.00
‡36 54.79 77.78 61.00
‡37 47.95 81.48 57.00
‡38 39.73 81.48 51.00
‡39 31.51 85.19 46.00
‡40 27.40 88.89 44.00
‡41 24.66 92.59 43.00
‡42 17.81 92.59 38.00
‡43 13.70 96.30 36.00
‡44 10.96 100.00 35.00
‡45 9.59 100.00 34.00
‡46 6.85 100.00 32.00
‡48 2.74 100.00 29.00
>48 0.00 100.00 27.00

With increasing demands on clinical services to is because in the general population there may be a
assess for the possibility of Asperger Syndrome, as percentage of individuals who have many autistic
demonstrated by the large number of referrals cur- traits but who do not require any clinical support
rently received at our clinic in Cambridge, it is (and are not seeking this) because of a good cognitive
important to be able to identify those people who are match between their cognitive style or personality,
most likely to have AS. We believe our results and their family or occupational or social context
support the AQ as a useful screening instrument in (Baron-Cohen, 2003). In this sense, whether a high
clinical practice. It provides a quick and reliable AQ score becomes disabling may depend on envi-
method of determining the likelihood of any individ- ronmental factors (tolerance by significant others, or
ual falling on the higher functioning end of the being valued for contribution at work, or a place in a
autistic spectrum and warranting further, more social network, protecting against the risks of sec-
detailed, assessment. We suggest that a more conser- ondary depression) rather than solely on factors
vative threshold score of 26 would ensure that false within the individual. This impression warrants
negatives are limited, and equally avoid cases ‘slip- systematic research.
ping through the net’. Of importance is that seventy-five percent of the
However, if the AQ were being used in a general patients seen in the clinic had been referred by their
population screen (and the ethical case for such a use general practitioner. This figure represents all sus-
has yet to be demonstrated) the higher cut off of 32 is pected cases referred by primary care practitioners as
likely to minimise false positives. We suspect that this no one was excluded simply based on their AQ score.
Screening Adults for Asperger Syndrome 335

Therefore our results are also relevant in the primary Baron-Cohen, S. (2003). The essential difference: men, women and
the extreme male brain. London: Penguin.
care setting where, as a result of increasing awareness Baron-Cohen, S., Wheelwright, S., Skinner, R., Martin, J., &
of autistic spectrum conditions, there is likely to be an Clubley, E. (2001). The Autism-Spectrum Quotient (AQ):
increase in the numbers of patients seeking assessment. Evidence from Asperger Syndrome/High Functioning Autism,
Males and Females, Scientists and Mathematicians. Journal of
The GP has the difficult task of deciding who should be Autism and Developmental Disorders, 31, 5–17.
referred on for in-depth assessment. We believe the AQ Bishop, D. V. M., Maybery, M., Maley, A., Wong, D., Hill, W., &
will facilitate this process, and is particularly useful in Hallmayer, J. (2004). Using self-report to identify the broad
phenotype in parents of children with artistic spectrum dis-
this setting as it is a relatively quick and easy to use orders: a study using the Austism-spectrum Quotient. Journal
screening instrument. There is increasing evidence that of child psychology and psychiatry, 45(8), 1431–1436.
by diagnosing even relatively late much can still be Ehlers, F., Gillberg, C., & Wing, L. (1999). A screening question-
naire for Asperger Syndrome and other high functioning
done to effectively manage the social impairments and autistic spectrum disorders in school age children. Journal of
facilitate better social inclusion. Autism and Developmental Disorders, 29, 129–142.
Ehlers, S., & Gillberg, C. (1993). The epidemiology of Asperger
syndrome. A total population study. Journal of Child Psy-
ACKNOWLEDGMENTS chology and Psychiatry, 34, 1327–1350.
Garnett M., Attwood T. (1995). The Australian Scale for Asperger
Syndrome: Paper presented at the 1995 Australian National
The authors would like to thank the Three Autism Conference, Brisbane, Australia.
Guineas Trust for supporting the Cambridge Life- Howlin, P., & Moore, A. (1997). Diagnosis in autism–A survey of
span Asperger Syndrome Service (CLASS). Many over 1200 patients in the UK. Autism, 1, 135–162.
Leekham, S., Libby, S., Wing, L., Gould, J., & Gillberg, C. (2000).
thanks go to Sabia Kayala and Kamrin Abid for Comparison of ICD-10 and Gillberg’s criteria for Asperger
assisting with data collection. Our appreciation also Syndrome. Autism, 4, 11–28.
goes to Jenny Hannah and Paula Naimi for admin- Lord, C., Rutter, M., & Le Couteur, A. (1994). Autism diagnostic
interview – Revised. Journal of Autism and Developmental
istrative support in CLASS. SBC and SW were Disorders, 24, 659–686.
supported by the Medical Research Council (UK) Mayes, S., Calhoun, S., & Crites, D. (2001). Does DSM-IV As-
during the period of this work. This study was perger’s Disorder exist? Journal of Abnormal Child Psychology,
3, 263–271.
completed at University of Cambridge, UK. Marc Ozonoff, S., South, M., & Miller, J. N. (2000). DSM-IV – defined
Woodbury-Smith’s current affiliation is Yale Child Asperger Syndrome: cognitive, behavioural and early history
Study Center, USA. differentiation from high-functioning autism. Autism, 4, 29–
46.
Powell, A. (2002). Taking responsibility. Good practice guidelines for
REFERENCES services: Adults with Asperger Syndrome. London: National
Autistic Society.
American Psychiatric Association. (1994). DSM-IV Diagnostic and Stata Corporation (2001). Stata Version 7, College Station,: TX:
Statistical Manual of Mental Disorders (4th edn.). Washington Stata Corporation.
DC: American Psychiatric Association. Wing, L. (1997). The Autistic Spectrum. London: Pergamon.
Barnard, J., Harvey, V., Prior, A., & Potter, D. (2001). Ignored or World Health Organisation. (1992). The Tenth Revision of the
ineligible? The reality for adults with autistic spectrum disor- International Classification of Diseases and Related Health
ders. London: National Autistic Society. Problems (ICD-10). Geneva: WHO.