Clinical Hub,UW Health Clinical Tool Search,UW Health Clinical Tool Search,Questionnaires,Related

The WHO Adult ADHD Self-Report Scale (ASRS) - A Short Screening Scale for General Population

The WHO Adult ADHD Self-Report Scale (ASRS) - A Short Screening Scale for General Population - Clinical Hub, UW Health Clinical Tool Search, UW Health Clinical Tool Search, Questionnaires, Related

The World Health Organization adult ADHD
self-report scale (ASRS): a short screening scale
for use in the general population
Department of Health Care Policy, Harvard Medical School ; Departments of Psychiatry and Neurology,
New York University School of Medicine ; Department of Psychiatry, Massachusetts General Hospital ;
Eli Lilly and Company, Global Health Outcomes ; Global Burden of Disease Unit, World Health Organization
Background. A self-report screening scale of adult attention-deficit/hyperactivity disorder
(ADHD), the World Health Organization (WHO) Adult ADHD Self-Report Scale (ASRS) was
developed in conjunction with revision of the WHO Composite International Diagnostic Interview
(CIDI). The current report presents data on concordance of the ASRS and of a short-form ASRS
screener with blind clinical diagnoses in a community sample.
Method. The ASRS includes 18 questions about frequency of recent DSM-IV Criterion A symp-
toms of adult ADHD. The ASRS screener consists of six out of these 18 questions that were selected
based on stepwise logistic regression to optimize concordance with the clinical classification. ASRS
responses were compared to blind clinical ratings of DSM-IV adult ADHD in a sample of 154
respondents who previously participated in the US National Comorbidity Survey Replication
(NCS-R), oversampling those who reported childhood ADHD and adult persistence.
Results. Each ASRS symptom measure was significantly related to the comparable clinical symp-
tom rating, but varied substantially in concordance (Cohen’s k in the range 0
81). Optimal
scoring to predict clinical syndrome classifications was to sum unweighted dichotomous responses
across all 18 ASRS questions. However, because of the wide variation in symptom-level concord-
ance, the unweighted six-question ASRS screener outperformed the unweighted 18-question ASRS
in sensitivity (68
7% v.56
3%), specificity (99
5% v.98
3%), total classification accuracy (97
2%), and k (0
76 v.0
Conclusions. Clinical calibration in larger samples might show that a weighted version of the
18-question ASRS outperforms the six-question ASRS screener. Until that time, however, the un-
weighted screener should be preferred to the full ASRS, both in community surveys and in clinical
outreach and case-finding initiatives.
Although it has long been known that attention-
deficit/hyperactivity disorder (ADHD) is one of
the most common psychiatric disorders among
children (Shekim et al. 1985; Bird et al. 1988)
and that ADHD often persists into adulthood
(Menkes et al. 1967; Mannuzza et al. 1993), the
fact that adult ADHD is a commonly occurring
and seriously impairing disorder has only re-
cently become the focus of attention (Wender
et al. 2001; Pary et al. 2002; Wilens et al. 2002).
One commentator has gone so far as to suggest
* Address for correspondence: Dr R. C. Kessler, Department of
Health Care Policy, Harvard Medical School, 180 Longwood
Avenue, Boston, MA 02115, USA.
(Email : kessler@hcp.med.harvard.edu)
Psychological Medicine, 2005, 35, 245–256. f 2004 Cambridge University Press
DOI: 10.1017/S0033291704002892 Printed in the United Kingdom

that ADHD is probably ‘the most common
chronic undiagnosed psychiatric disorder in
adults ’ (Wender, 1998). However, no large-scale
epidemiological data exist to evaluate this claim,
as none of the many adult community psychi-
atric epidemiological surveys carried out over
the past two decades with either the Diagnostic
Interview Schedule (DIS; Robins et al. 1981) or
the Composite International Diagnostic Inter-
view (CIDI; Robins et al. 1988) included an
assessment of adult ADHD.
Several attempts have been made to estimate
the general population prevalence of adult
ADHD by extrapolation from childhood preva-
lence estimates in conjunction with adult per-
sistence estimates (Weiss et al. 1985; Mannuzza
et al. 1998; Biederman et al. 2000; Barkley et al.
2002) or by direct estimation from small samples
of adults (Murphy & Barkley, 1996) or college
students (Heiligenstein et al. 1998), yielding
prevalence estimates in the range 1–6%. How-
ever, these estimates are all based on convenience
samples. In an effort to obtain more represen-
tative estimates, adult ADHD is included in
the new World Health Organization (WHO)
World Mental Health (WMH) Initiative surveys
(Kessler & Ustun, 2000), a series of general
population psychiatric epidemiological surveys
currently underway in 28 different countries in
all regions of the world with a combined sample
size of more than 200 000 respondents.
Both retrospective assessments of childhood
ADHD and a screen for adult ADHD were
developed for use in the expanded version of
CIDI that forms the core of the WMH surveys.
The retrospective assessment was part of a series
of four sections devoted to adult recall of child-
hood disorders (oppositional-defiant disorder,
conduct disorder, and separation anxiety dis-
order in addition to ADHD) that were based
on the comparable sections in the DIS (Robins
et al. 1995). There was no precedent, though, for
assessing adult ADHD in previous DIS-CIDI
surveys. Although other self-report measures
of adult ADHD exist (Barkley, 1995; Brown,
1996; Conners et al. 1998; Mehringer et al.
2002; West et al. 2003), a review of these
measures showed that they either fail to include
all 18 DSM-IV Criterion A symptoms or assess
some of these symptoms with questions that
were judged to be suboptimal by an Advisory
Group of clinical experts in adult ADHD
assembled by the WHO to consult on this aspect
of the WMH survey assessment. Based on this
evaluation, a decision was made to develop a
new self-report measure of adult ADHD for the
WMH surveys. The present report describes this
measure: the WHO Adult ADHD Self-Report
Scale (ASRS) Version 1.1. A short ASRS
screener, which turned out to outperform the
full ASRS in the clinical calibration study re-
ported here, was also developed based on step-
wise logistic regression.
The clinical calibration of the ASRS was carried
out by re-interviewing a quota subsample of
154 respondents from the US National Co-
morbidity Survey Replication (NCS-R; Kessler
et al. 2003), a nationally representative face-
to-face household survey of 9083 respondents
aged 18 and older in the co-terminous United
States who were interviewed by trained lay inter-
viewers between February 2001 and December
2002. NCS-R respondents were selected from a
stratified, multi-stage, clustered area probability
sample of the USA non-institutionalized civilian
population, with over 1000 segments (i.e. block-
equivalents) in 170 counties in 34 states. The
response rate among primary respondents was
9%. All respondents were administered the
WMH version of the CIDI in Part I of the in-
terview, while a probability subsample of 5692
respondents also received a Part II interview
that included assessments of risk factors and
other disorders. The Part II sample consisted of
all respondents who screened positive for any
Part I WMH-CIDI disorder plus a probability
subsample of 25% other Part I respondents.
Less than 1% of the Part I respondents who
were selected into the Part II sample failed to
complete Part II. This conditional response rate
was unrelated to the Part II sampling stratum.
A more detailed description of the NCS-R
sample design and field procedures is presented
elsewhere (Kessler et al. in press).
Based on a concern that older adults would
have difficulty responding to retrospective
questions about childhood, the assessment of
ADHD was limited to respondents in the age
range 18–44 years. Respondents who reported
symptoms that were classified as having met
246 R. C. Kessler et al.

criteria for ADHD in childhood were asked a
single follow-up question about whether they
continued to have any current problems with
attention or hyperactivity-impulsivity. The re-
spondents who received the ADHD assessment
were divided into four sampling strata for the
adult ADHD clinical calibration sample: those
who denied any childhood symptoms of ADHD,
those who reported at least some symptoms
but were classified as not meeting full criteria,
those who were classified as meeting criteria
but denied having current adult symptoms, and
those who were classified as meeting criteria
who reported having current adult symptoms.
An attempt was made to contact by telephone
and re-interview 30 respondents in each of the
first three strata and 60 in the fourth stratum.
The final quota sample of 154 was slightly larger
than the sum of these targets because a higher
than expected proportion of pre-designated
respondents kept their appointments to be
interviewed. Before beginning the interviews,
respondents were told that the purpose of the
interviews was to test the accuracy and expand
the measures in the original survey, that par-
ticipation was completely voluntary and confi-
dential, and that we would send respondents a
check for $50 as a token of our appreciation for
their participation in this phase of the research.
Respondents were then allowed to ask any
additional questions before obtaining verbal
informed consent and beginning the interview.
Interviews were tape-recorded with the per-
mission of respondents. The Harvard Medical
School Human Subjects Committee approved
these recruitment and consent procedures.
Two levels of weighting were required to
make the 154 respondents representative of
the total NCS-R sample. First, the sample was
weighted using the NCS-R Part II composite
weight to adjust for differential probabilities
of selection into the overall sample within
households; differences in intensity of recruit-
ment effort among hard-to-recruit cases ; dif-
ferential non-response across sample segments
based on aggregated Census Block Group data;
discrepancies between the sample and the
Census population distribution on the cross-
classification of various sociodemographic vari-
ables ; and differential probabilities of selection
into the Part II sample. The development of
these weights is discussed in detail elsewhere
(Kessler et al. 2003). Second, the sample was
weighted to adjust for the oversampling of
Part II respondents who reported childhood
ADHD and adult persistence of ADHD symp-
toms. It has previously been shown that the
weighted NCS-R Part II sample distribution
closely matches the Census population on a
variety of geographic and demographic variables
(Kessler et al. 2003). The weighted ADHD clini-
cal calibration sample distribution was found
roughly to approximate the Part II NCS-R dis-
tribution on these same sociodemographic vari-
ables. (Appendix tables that include details of
these comparisons along with other results that
are too detailed to be reported in this paper are
available at www.hcp.med.harvard.edu/ncs).
The screening scale item pool
Two board-certified psychiatrists (L.A., T.S.)
and the WHO advisory group of clinical experts
in adult ADHD (see Acknowledgments) gener-
ated an initial pool of fully structured questions
about the symptoms of ADHD as they are
typically expressed among patients with adult
ADHD and mapped these onto each of the 18
DSM-IV Criterion A symptoms. The survey
methodology collaborators (R.C.K., T.B.U.)
then modified these questions to remove double-
barreled descriptions, reduce ambiguities in
meaning, and create a consistent temporal focus.
The clinical collaborators then made modifi-
cations to improve face validity. One question
was selected from this final item pool for each
of the 18 DSM-IV Criterion A symptoms of
ADHD based on face validity. Eleven more
questions were selected to span the range of
symptoms not in DSM-IV that were thought by
the clinical experts to be common expressions
of adult ADHD. Each question asked how often
a symptom occurred over the past 6 months on a
0–4 scale with responses of never (0), rarely (1),
sometimes (2), often (3), and very often (4). The
focus of this report is on the 18 questions de-
signed to operationalize the DSM Criterion A
symptoms (Table 1).
The clinical interview
The clinical interview used in the calibration
study had three parts. The first part was the
semi-structured clinical ADHD Rating Scale
(ADHD-RS; DuPaul et al. 1998), a state-of-
the-art retrospective assessment of childhood
A screening scale for adult ADHD 247

ADHD designed for administration to adults.
The second part was the semi-structured clinical
interview for recent (past 6 months) DSM-IV
adult ADHD that is used in most clinical trials
of this disorder (Spencer et al. 1995, 1998, 2001;
Michelson et al. 2003). The third part was the
self-report battery, which was administered at
the end in order not to bias interviewers when
they made their clinical symptom ratings. Four
experienced clinical interviewers (Ph.D. clinical
psychologists with between 5 and 20 years of
clinical experience) carried out these interviews.
Each interviewer received 40 hours of training
from two board certified psychiatrists who
specialize in research on adult ADHD (L.A.,
T.S.) and successfully completed five practice
interviews in which their symptom ratings
matched those of the trainers before they began
production interviewing.
Clinical supervisor (M.H.) reviewed tape
recordings of clinical interviews. Discrepancies
between interviewer and supervisor ratings
were referred to the board certified psychiatrists
(L.A., T.S.) for discussion and resolution and,
as needed, additional contacts with respondents.
The supervisor and clinical collaborators also
held weekly group interviewer calibration meet-
ings, while the supervisor held weekly one-on-
one feedback meetings with each interviewer
separately. Consistent with DSM-IV criteria, a
clinical diagnosis of adult ADHD required that
a respondent have at least six symptoms of
either inattention or hyperactivity-impulsivity
during the 6 months before the interview (DSM-
IV Criterion A), at least two Criterion A symp-
toms of ADHD before age seven (Criterion B),
some impairment in at least two areas of living
during the past 6 months (Criterion C), and
clinically significant impairment in at least one
area of living over the same time period (Cri-
terion D). No attempt was made to operation-
alize the DSM-IV diagnostic hierarchy rules for
ADHD (Criterion E).
Statistical methods
Symptom-level concordance between ASRS
self-reports and blind clinician ratings of
6-month prevalence was evaluated by dichot-
omizing the 0–4 ASRS response scale separately
for each question to maximize overall classi-
fication accuracy with dichotomous clinical
symptom ratings. A wide range of possible
simple scoring methods that used all 18 ques-
tions was then investigated. Diagnostic ef-
ficiency statistics were calculated for each of
these to select the best method, including sen-
sitivity (the percent of respondents with the
clinician-rated syndrome classified as having
Table 1. The WMH-CIDI Adult ADHD Self-Report Scale (ASRS) Questions
I. Inattention
1. How often do you make careless mistakes when you have to work on a boring or difficult project?
2. How often do you have difficulty keeping your attention when you are doing boring or repetitive work?
3.* How often do you have difficulty concentrating on what people say to you, even when they are speaking to you directly?
4.*# How often do you have trouble wrapping up the fine details of a project, once the challenging parts have been done?
5.*# How often do you have difficulty getting things in order when you have to do a task that requires organization?
6.# When you have a task that requires a lot of thought, how often do you avoid or delay getting started?
7. How often do you misplace or have difficulty finding things at home or at work?
8. How often are you distracted by activity or noise around you?
9.*# How often do you have problems remembering appointments or obligations?
II. Hyperactivity-Impulsivity
1.# How often do you fidget or squirm with your hands or your feet when you have to sit down for a long time?
2.* How often do you leave your seat in meetings or other situations in which you are expected to remain seated?
3. How often do you feel restless or fidgety?
4. How often do you have difficulty unwinding and relaxing when you have time to yourself?
5.# How often do you feel overly active and compelled to do things, like you were driven by a motor?
6. How often do you find yourself talking too much when you are in a social situation?
7.* When you’re in a conversation, how often do you find yourself finishing the sentences of the people that you are talking to,
before they can finish them themselves?
8. How often do you have difficulty waiting your turn in situations when turn-taking is required?
9.* How often do you interrupt others when they are busy?
Response options are: never, rarely, sometimes, often, and very often. Patients were asked to answer the questions using a 6-month recall
* Clinically significant symptom levels were defined for these seven questions as responses of sometimes, often, and very often. For all
remaining 11 questions, often and very often were the clinically significant symptom levels. See text for the rationale for this difference.
# The six-question ASRS screener.
248 R. C. Kessler et al.

the syndrome by the screening scale) ; specificity
(the percent of respondents without the clin-
ician-rated syndrome classified as not having
the syndrome by the screening scale) ; total
classification accuracy (the percent of all respon-
dents consistently classified by the ASRS and
clinician ratings) ; the odds ratio (OR) of the
2r2 table between yes–no syndrome classifi-
cations based on the ASRS and the clinical
ratings; Cohen’s k (a measure of concordance
that adjusts for chance agreement) ; and the area
under the receiver operator characteristic curve
(AUC; the probability that a randomly selected
clinical case would score higher on the ASRS
than a randomly selected non-case).
In order to investigate whether most of the
precision of the full 18-question ASRS could
be captured with fewer items, stepwise logistic
regression analysis was used to select the best
subset of ASRS questions to create a short-form
screener. Only six questions were found to enter
this stepwise analysis significantly. However,
all-possible subsets logistic regression showed
that a number of different six-question subsets
of the 18 questions were roughly equivalent
in reproducing clinical diagnoses. As a result,
the psychometric analyses described in the last
paragraph were repeated for all these alternative
six-question short-form scales. Inspection of test
statistics was used to select a final optimal short-
form scale.
The conventional screening approach creates
a dichotomy to differentiate predicted cases
and non-cases. However, this dichotomization
often discards potentially useful information
that would be retained in a polychotomous
screening scale, such as the distinction between
a nearly definite case and a probable case. As a
result, the use of polychotomous screening scales
is becoming increasingly popular in evidence-
basedmedicine (Peirce&Cornell, 1993). In order
to investigate whether this might be useful for
the ASRS, polychotomous versions of both the
optimal 18-question and six-question scoring
methods were created.
As the sample design features clustering and
weighting of cases, design-based methods were
used to calculate standard errors and confidence
intervals. The jackknife pseudo-replication
method (Wolter, 1985) implemented in an SAS
8 macro (SAS Institute, 1999) was used to make
these calculations.
Symptom-level concordance
In order to assess symptom-level concordance,
the 0–4 ASRS response scale was collapsed into
a dichotomy that mapped onto the dichotomous
clinical ratings. The conventional way to do this
is to select the same dichotomy for all questions,
with responses of often or very often usually
considered above the clinical threshold and
other responses (never, rarely, and sometimes)
below the threshold (e.g. O’Donnell et al. 2001).
However, this approach can lead to error either
when the self-report questions vary in severity
or when respondents are more reluctant to
admit the frequent occurrence of some symp-
toms than others, in which case it is preferable
to allow between-question variation in thresh-
olds (Kessler et al. 2002). Based on this thinking,
the 2r5 cross-classification of each 0–4 ASRS
response with the dichotomous clinical symp-
tom rating was inspected and the ASRS re-
sponse scale was dichotomized in such a way
that the numbers of false positives (ASRS posi-
tives who were rated as asymptomatic in the
clinical interview) and false negatives (ASRS
negatives who were rated as having the symp-
tom in the clinical interview) in the weighted
data were as equal as possible. This rule resulted
in clinically significant symptom levels being
defined as those that occurred often or very
often for 11 screening questions and as some-
times, often, or very often for the other seven
screening questions.
Cohen’s k was used to assess concordance
between dichotomized ASRS symptom re-
sponses and clinical symptom ratings. Applying
commonly used standards for assessing strength
of k coefficients (Landis & Koch, 1977), con-
cordance was slight (less than 0
2) for two
questions, fair (0
4) for seven, moderate
6) for six, and substantial (0
8) for
the remaining three. Eleven of the 18 questions
were found to be unbiased in the sense that the
number of false positives did not differ signifi-
cantly from the number of false negatives at
the 0
05 level of significance using two-sided
tests. The proportion of biased questions was
comparable for inattention and hyperactivity-
impulsivity. Three of the seven biased ques-
tions were biased downward and the other four
were biased upward. Detailed results of these
A screening scale for adult ADHD 249

comparisons are included in the appendix ma-
terials posted at the website mentioned earlier in
the paper.
Optimal scoring of the ASRS
The most direct mapping of ASRS responses
onto DSM-IV Criterion A is to classify respon-
dents as cases if they have positive responses
either to six or more inattention questions or to
six or more hyperactivity-impulsivity questions.
However, as concordance between the ASRS
questions and clinical symptom ratings is far
from perfect, application of the six-of-nine rule
might not be optimal in predicting clinical syn-
drome classifications. A total of six dimensional
scoring methods were consequently applied to
the data and concordance with the clinical syn-
drome classifications assessed for all dichot-
omizations of each of these six. The first three of
the six scoring methods were based on separate
scores for the inattention and the hyperactivity-
impulsivity domains, with each respondent re-
ceiving the higher of these two scores.
The first method counted the number of
positive symptom screens in each domain (one
dichotomization of which is the DSM-IV six-of-
nine categorization).
The second method assigned greater weight
to responses of ‘very often’ than to other posi-
tive responses and then counted the weighted
number of positive symptom screens in each
domain. A range of relative weights was con-
sidered here up to a maximum of 2 : 1.
The third method included the full 0–4 range
of responses to the screening questions and
summed these responses in each domain.
The fourth through sixth scoring methods
were similar to the first three except that they
were based on a summation across all 18 ques-
tions rather than on the higher score in the sep-
arate inattention and hyperactivity-impulsivity
Test statistics are presented in Table 2 for
the optimal dichotomous cut-point for each
of these six scoring methods, where optimality
is defined as minimization of the difference
between false positive (FP) and false negative
(FN). Method 4 is clearly the best one in this
set, yielding the highest total classification
accuracy (96
2%), OR (73
4), k (0
58), and
close to the highest AUC (0
77). While com-
pared to Method 4, Method 3 has a marginally
higher sensitivity (60
2% v.56
3%) and AUC
(78 v. 77). This is achieved at the expense of
a lower specificity (96
3% v.98
3%), resulting
in lower total classification accuracy (94
2%) and a dramatically lower OR (39
5 v.
4) for Method 3 than Method 4. Based on
these considerations and the previously specified
optimality criterion, Method 4 was selected
as the optimal simple scoring method of the
Table 2. Concordance of optimally dichotomized
versions of the 18-question ASRS with blind
ADHD-RS clinical syndrome classifications using a variety of scoring methods
Optimal cut-point
Cohen’s kappa
Odds ratio
%(S.E.) % (S.E.) % (S.E.) x
k (S.E.) OR (95% CI)
(1) 0–5 v. 6–9 51
0) 97
0) 94
1) 0
48 (0
16) 37
1* (7
4) 0
(2) 0–6 v. 7–18 34
4) 96
1) 93
2) x0
32 (0
17) 15
9* (3
8) 0
(3) 0–20 v. 21–36 60
9) 96
1) 94
1) 1
50 (0
15) 39
5* (9
0) 0
(4) 0–8 v. 9–18 56
8) 98
7) 96
9) x0
58 (0
16) 73
4* (25
6) 0
(5) 0–7 v. 8–36 55
1) 96
1) 94
1) 1
48 (0
15) 35
4* (8
0) 0
(6) 0–36 v. 37–72 57
1) 96
1) 94
1) 1
48 (0
15) 36
5* (8
3) 0
Optimality was defined as minimizing the difference between the weighted number of false positive (FP) and false negative (FN) responses.
The first three scoring methods calculated separate scores for the inattention and the hyperactivity-impulsivity domains and assigned the
higher of these two scores. The first method counted the number of positive symptom screens (defined in Table 3) in each domain. The second
method assigned greater weight to responses of ‘very often’ than to other positive responses and then counted the weighted number of positive
symptom screens in each domain, with a range of relative weights up to a maximum of 2 : 1. Only results of the 2 : 1 version are reported here.
The third method included the full 0–4 range of responses to the ASRS questions and summed these responses separately for the inattention
and hyperactivity-impulsivity domains. The fourth through sixth scoring methods were similar to the first three except that they were based on
a summation across all 18 questions rather than on the higher score in the separate inattention and hyperactivity-impulsivity domains.
See the text for definitions of the test statistics.
The McNemar test evaluates the significance of the difference between FP and FN. None of these tests was significant at the 0
05 level.
TCA, total classification accuracy; AUC, area under the receiver operator characteristic curve.
* Significant at the 0
05 level.
250 R. C. Kessler et al.

In order to explore the usefulness of a poly-
chotomous classification, the optimally-scored
ASRS was collapsed into strata so that the
probability of being classified as a clinical case
did not differ meaningfully across cells within
each stratum, but did differ meaningfully across
strata. Three strata were found to meet these
criteria, corresponding to scale scores in the
range 0–3, 4–8, and 9–18. Part I of Table 3
shows the sensitivity and specificity for each
stratum. Part II presents estimates of Positive
predictive values (PPV) for plausible values of
population prevalence. As shown in the first
row of Part I, 56
3% of adults in the general
population who meet clinical criteria for ADHD
are in the highest stratum, with 30
3% in the
middle stratum and the remaining 13
4% in the
lowest stratum. Nearly three-quarters (70
of clinical non-cases are in the lowest ASRS
stratum, 27
8% in the middle stratum, and only
7% in the highest stratum. Given that best
estimates from available studies put the com-
munity prevalence of adult ADHD in the range
1–6% (Wender et al. 2001) and prevalence in
general medical populations as much as twice
as high, the PPV of the three strata in popu-
lation studies would be in the range 0
the lowest stratum, 1
0% in the middle
stratum, and 24
7% in the highest stratum,
if sensitivity and specificity were constant.
The ASRS screener
As an 18-question screening scale is too long
for many purposes, we investigated whether
we could develop a useful short-form screener.
Stepwise logistic regression was used to make
this evaluation, beginning with the selection of
the dichotomously coded screening questions
that most accurately predicted the clinical syn-
drome classifications. An inspection of success-
ive changes in AUC as new questions were
added to the prediction equation was used to
select an optimal number of questions, leading
to the conclusion that six questions were the
optimal number. The equation with this number
of predictors had AUC=0
95 and x
031). We then attempted to select ad-
ditional predictors that distinguished between
responses of very often and other positive re-
sponses. No significant predictors of this sort
could be found (x
4, p=0
702 for the three
best predictors in this set). We then attempted
to select additional predictors that added infor-
mation on the full 0–4 range of responses, but
again found no significant predictors of this sort
2, p=0
978 for the three best predictors
in this set).
Based on these results, further analysis
focused on six-question screeners that used
dichotomous symptom scoring. All-possible-
subsets logistic regression analysis showed that
several six-question subsets had very similar
values of AUC and x
in predicting clinical
syndrome classifications. The concordance of
these scales with clinical syndrome classifications
was assessed for all logically possible dichot-
omizations. Test statistics showed one of these
to be marginally superior to the rest in that
it had close to the highest sensitivity (68
the highest specificity (99
5%), and the highest
Table 3. Polychotomous stratification of the optimally scored 18-question ASRS
Stratum categorizations
0–3 4–8 9–18
I. Test statistics*
Sensitivity (S.E.) 13
4% (6
9) 30
3% (8
1) 56
3% (8
1 – Specificity (S.E.) 70
5% (8
0) 27
8% (8
0) 1
7% (0
II. Positive predictive values and standard errors at plausible values of P
01 0
1) 1
5) 24
03 0
3) 3
4) 50
3 (11
06 1
9) 6
1) 67
09 1
9) 9
1) 76
12 2
4) 13
9) 81
* See the text for definitions of the test statistics.
# P
is the population prevalence of ADHD in a hypothetical population.
A screening scale for adult ADHD 251

overall concordance as indicated by total classi-
fication accuracy (97
9%), OR (414
1), k (0
and AUC (0
84). Detailed results of these com-
parisons are included in the appendix materials
posted at the website mentioned earlier in the
As with the optimal 18-question version of
the ASRS, the optimal six-question ASRS
screener was collapsed into three strata (0–1,
2–3, and 4–6). Sensitivities and specificities for
the strata are presented in Part I of Table 4.
Approximately two-thirds (68
7%) of clinical
cases scored in the highest stratum, with only
3% in the lowest stratum. About three-
quarters of clinical non-cases (74
8%), in com-
parison, scored in the lowest stratum and only
5% in the highest stratum. Part II of the table
shows that PPVs of the three strata are in the
range 0
8 for the lowest stratum, 1
for the middle stratum, and 56
7 for the
highest stratum for plausible values of popu-
lation prevalence. Comparing these PPVs with
the distribution of sensitivities in Part I of the
table shows that about one-third of clinical
cases (i.e. those in the lowest and middle strata)
would be missed with this screen, while two-
thirds in the highest stratum would be classified
as having a very high probability of being cases.
ASRS discrimination among screener positives
Comparison of the results in Tables 3 and 4
shows that the six-question screener out-
performs the full 18-question ASRS. However,
two further results suggest that the full ASRS
might nonetheless be useful among people
who are positive on the screener. First, the
percent of clinical cases screening positive on
a dichotomous version of the full ASRS in
which positives are defined as those scoring
11–18 (49
1%, with an S.E.of11
0%) is signifi-
cantly higher than the percent of clinical non-
cases screening positive on the same dichotomy
2%, with a S.E.of12
4%, z=2
0, p=0
documenting that administration of the full
ASRS can significantly improve classification
of true cases among people who are positive on
the six-question screener. Second, a substantial
Pearson correlation (r=0
43, p<0
001) exists
between a version of the ASRS that sums re-
sponses to the 18 ASRS questions using the
full 0–4 response scale (generating a scale with a
0–72 range) and the scale of current clinical
symptom severity, documenting that repeat
administration of the full ASRS might be useful
in charting clinical improvement among cases in
It is important to note that no data have ever
been published on the validity of the clinical
interview used as the gold standard in this study
even though it has become the standard in
clinical studies of adult ADHD (Spencer et al.
1995, 1998, 2001; Michelson et al. 2003). To
the extent that it imperfectly operationalizes
the DSM-IV criteria, the validity of the ASRS
might be underestimated. Another important
boundary condition is that we developed the
scale in an explicit attempt to equalize the
Table 4. Polychotomous stratification of the optimally scored six-question ASRS screener
Stratum categorizations
0–1 2–3 4–6
I. Test statistics*
Sensitivity (S.E.) 4
3% (2
9) 27
0% (8
0) 68
7% (8
1 – Specificity (S.E.) 74
8% (6
6) 24
7% (6
5) 0
5% (0
II. Positive predictive values and standard errors at plausible values of P
01 0
0) 1
4) 56
8 (14
03 0
1) 3
3) 80
06 0
2) 6
5) 89
09 0
4) 9
7) 92
12 0
5) 13
7) 94
* See the text for definitions of the test statistics.
# P
is the population prevalence of ADHD in a hypothetical population.
252 R. C. Kessler et al.

number of false positives and false negatives.
Optimization rules that put different weights on
false positives and false negatives might have led
to different scoring rules or different questions
being selected (Kramer, 1992).
A potential limitation of the study design is
that the self-report questions were administered
after the completion of the clinical assessment,
possibly resulting in some respondents becom-
ing more sensitized to their symptoms and re-
sponding to the ASRS differently than they
would have otherwise. This problem could be
resolved in future face-to-face clinical re-
appraisal studies by having respondents self-
administer a paper and pencil version of the
ASRS before blind administration of the clinical
interview. Another limitation regarding validity
is that all data are obtained from respondents
rather than also from informants.
Methodological studies comparing adult
self-reports versus informant reports of ADHD
symptoms generally show the same pattern of
disagreement as in studies of child self-reports
versus informant reports (Jensen et al. 1999) ;
namely, that informants report higher symptoms
than respondents (Gittelman & Mannuzza,
1985; Zucker et al. 2002). This suggests that
both the clinical interviews and the ASRS
results might be conservative. It is important
to note, however, that the one adult self-versus-
informant ADHD symptom comparison study
that was carried out in a non-clinical sample
found fairly strong associations between the two
reports and no self-informant difference in re-
ported symptom severity (Murphy & Schachar,
An additional limitation is that a single set
of scoring rules was presented even though the
optimal thresholds and appropriate values of
PPV might differ as a function of gender, edu-
cational status, marital status, or other known
correlates of adult ADHD. No attempt was
made to generate subsample scoring rules,
however, based on the clinical reappraisal sample
being too small for powerful subsample analy-
sis. This small sample size also raises concerns
about the generalizability of the results. This
is especially true for the ASRS screener, which
was developed based on the use of stepwise
regression analysis and might have capital-
ized on chance in selecting items. It is worth
noting, in this regard, that the variation in the
symptom-level concordance between the ASRS
and the clinical ratings could also have been
taken into consideration in scoring by including
question-level weights, such as those generated
in a logistic regression analysis that regressed
the dichotomous clinical syndrome classifica-
tions on the 18 separate ASRS questions. We
decided against this, however, based on concern
about over-fitting the data.
Within the context of these limitations,
the results suggest that the six-question ASRS
screener is a very good tool for reproducing the
overall clinical evaluations made by carefully
trained and closely monitored clinical inter-
viewers. The three-stratum version of the scale,
in particular, has excellent concordance with
blind clinical diagnoses. This means that the
transformation of the scale’s stratum classifi-
cations into individual-level predicted prob-
abilities of clinical diagnoses can be used in
general population epidemiological surveys to
generate an outcome variable that will be a good
surrogate for clinical syndrome classifications.
In light of the evidence that the six-question
ASRS screener out-performs the full ASRS, a
question can be asked whether the latter has any
value. As noted in the discussion of limitations,
our scoring of the full ASRS was based on an
unweighted summation of responses. Aweighted
version of the full scale might prove to have
much greater concordance with clinical syn-
drome classifications than the screener, although
it would be necessary to have a cross-validation
sample to investigate this possibility rigorously.
Furthermore, we showed that the full ASRS
both refines prediction of the clinical classifi-
cation among respondents who are positive on
the six-question screener and correlates signifi-
cantly with clinician-rated overall symptom
severity in this same subsample.
The six-question ASRS screener, in compari-
son, seems to hold more promise than the full
ASRS for clinical screening purposes. As
shown in Table 4, over two-thirds of clinical
cases screen positive on the six-question screener
compared to an extremely low proportion of
non-cases (0
5%), resulting in a high proportion
of screened positives being true cases under all
plausible assumptions about the population
prevalence of the disorder. The situation with
the roughly one-third of clinical cases who are
negative on the six-question ASRS screener is
A screening scale for adult ADHD 253

also important to consider. None of the scoring
rules we considered was able to generate a
stratum that could reliably distinguish these
cases from non-cases. Not surprisingly, these
screened negative clinical cases had an average
clinical symptom severity score lower than
clinical cases that screened positive (1
5 v.1
7, p=0
091). More detailed analyses ex-
plored whether we could capture these false
negatives by using information in the remaining
12 ASRS questions or in the additional 11
questions that were thought by the clinical
experts to be other common expressions of
adult ADHD. No evidence was found that the
screening scale could be improved by using this
additional information.
The probability of a person with a given score
on a screening scale meeting criteria for a dis-
order (i.e. the PPV at that point on the scale)
will have the same expected value in a given
population as in a calibration sample only if the
calibration sample is representative of that
population. That is why estimates of PPV were
reported for a range of plausible prevalence
values. Importantly, we found that the PPV
of the highest stratum in the ASRS screener is
quite high even under the assumption of an im-
plausibly low prevalence. We also found that the
PPV of the middle stratum in the ASRS screener
is quite low even under the assumption of a high
prevalence. These results tell us that patients
who screen into the highest stratum of the
ASRS screener in primary care samples should
routinely be considered likely cases who warrant
further evaluation, while primary care patients
who screen into the middle stratum should
only be evaluated further when there is other
evidence to suggest that they might be cases.
Uncertainty about PPV is of considerably
more importance, in comparison, in epidemi-
ological surveys that focus on segments of the
population that cannot be considered represen-
tative of the total USA population. Although it
is conventional to use standardized cut-points
in such surveys, these can lead to substantial
error in estimating prevalence and correlates.
A less biased approach is to generate expected
stratum distributions for all logically possible
values of population prevalence (noting that
the expected sample proportion in a given stra-
tum is the sum of the products of prevalence
times sensitivity and the additive inverse of
prevalence times the additive inverse of speci-
ficity) based on the sensitivities and specificities
reported in Table 4 and to use maximum-likeli-
hood comparisons of these theoretical distri-
butions with observed stratum distributions in
the sample to select the most likely prevalence
in the population from which the sample was
selected. Once this maximum-likelihood preva-
lence estimate is obtained, individual-level pre-
dicted probabilities can be calculated easily
(Guyatt & Rennie, 2001).
Besides using the six-question ASRS screener
in epidemiological surveys and in primary care
screening, the good results about the precision
of the screener and the fact that it can be
self-administered easily and quickly (less than
2 minutes) might make it a useful secondary
measure to include in clinical studies. This could
be a useful complement to the dimensional clini-
cal assessments of ADHD symptom severity
typically used in such studies to define a lower-
bound severity threshold that distinguishes
community cases from non-cases (i.e. the high-
est versus middle strata in the ASRS screener).
The use of the ASRS screener in clinical studies
would also provide a useful crosswalk between
clinical research and community epidemiologi-
cal research by allowing a comparison of the
severity distribution between community and
clinical cases. The absence of such comparative
data has restricted our ability to interpret the
clinical significance of categorical prevalence
estimates of most mental disorders in com-
munity epidemiological studies up to now
(Kessler et al. in press). The inclusion of ident-
ical short dimensional assessments of adult
ADHD in both clinical and community studies
would be a useful step in the direction of
addressing this important problem for this
heretofore understudied disorder.
The National Comorbidity Survey Replication
(NCS-R) is supported by the US National In-
stitute of Mental Health (U01-MH60220) with
supplemental support from the US National
Institute of Drug Abuse, the Substance Abuse
and Mental Health Services Administration,
and the Robert Wood Johnson Foundation
(Grant no. 044780). Collaborating investi-
gators include Ronald C. Kessler (Principal
254 R. C. Kessler et al.

Investigator, Harvard Medical School),
Kathleen Merikangas (Co-Principal Investi-
gator, NIMH), Doreen Koretz (Co-Principal
Investigator, Harvard University), William
Eaton (The Johns Hopkins University), Jane
McLeod (Indiana University), Mark Olfson
(Columbia University College of Physicians and
Surgeons), Harold Pincus (University of Pitts-
burgh), PhillipWang (HarvardMedical School),
KennethWells (UCLA), and ElaineWethington
(Cornell University).
Additional support for the ADHD screening
scale validation re-interviews was provided
by an unrestricted educational grant from the
Eli Lilly Company. The WMH-CIDI Advisory
Group for adult ADHD includes Lenard Adler
(New York University Medical School), Russell
Barkley (Medical College of South Carolina),
Joseph Biederman (Massachusetts General
Hospital and Harvard Medical School), Keith
Conners (Duke University Medical School),
Stephen Faraone (Massachusetts General Hos-
pital and Harvard Medical School), Laurence
Greenhill (New York State Psychiatric Insti-
tute), Molly Howes (Harvard Medical School),
Ronald Kessler (Harvard Medical School),
Thomas Spencer (Massachusetts General Hos-
pital) and T. Bedirhan Ustun (World Health
The authors thank the other members of
the advisory group for helpful comments on
this paper. All NCS-R instruments are posted at
L. Adler, S. Faraone, R. C. Kessler and T.
Spencer have all served as paid consultants of Eli
Lilly and Company. K. Secnik is an employee of
Eli Lilly and Company.
Barkley, R. A. (1995). ADHD behavior checklist for adults. ADHD
Report 3,16.
Barkley, R. A., Fischer, M., Smallish, L. & Fletcher, K. (2002). The
persistence of attention-deficit/hyperactivity disorder into young
adulthood as a function of reporting source and definition dis-
order. Journal of Abnormal Psychology 111, 279–289.
Biederman, J., Mick, E. & Faraone, S. V. (2000). Age-dependent
decline of symptoms of attention deficit hyperactivity disorder :
impact of remission definition and symptom type. American
Journal of Psychiatry 157, 816–818.
Bird, H. R., Canino, G. & Rubio-Stipec, M. (1988). Estimates of
the prevalence of childhood maladjustments in a community
survey in Puerto Rico. Archives of General Psychiatry 45,
Brown, T. E. (1996). Brown Attention Deficit Disorder Scales.
Psychological Corporation: San Antonio, CA.
Conners, C. K., Erhardt, D. & Sparrow, E. P. (1998). Conners
CAARS-Self-report: Long Version (CAARS-S:L). Multi-Health
Systems: North Totawanda, NY.
DuPaul, G. J., Power, T. J., Anastopoulos, A. D. & Reid, R. (1998).
ADHD Rating Scale-IV: Checklists, Norms, and Clinical Interpret-
ation. Guilford: New York.
Gittelman, R. & Mannuzza, S. (1985). Diagnosing ADD-H in ado-
lescents. Psychopharmacology Bulletin 21, 237–242.
Guyatt, G. & Rennie, D. (2001). User’s Guide to the Medical Litera-
ture: a Manual for Evidence-based Clinical Practice. AMA Press :
Chicago, IL.
Heiligenstein, E., Conyers, L. M., Berns, A. R., Miller, M. A.
& Smith, M. A. (1998). Preliminary normative data on
DSM-IV attention deficit hyperactivity disorder in college
students. Journal of the American College Health Association 46,
Jensen, P. S., Rubio-Stipec, M., Canino, G., Bird, H. R., Dulcan,
M. K., Schwab-Stone, M. E. & Lahey, B. B. (1999). Parent and
child contributions to diagnosis of mental disorder : are both
informants always necessary? Journal of the American Academy of
Child and Adolescent Psychiatry 38, 1569–1579.
Kessler, R. C., Andrews, G., Colpe, L. J., Hiripi, E., Mroczek, D. K.,
Normand, S. L., Walters, E. E. & Zaslavsky, A. M. (2002). Short
screening scales to monitor population prevalences and trends
in non-specific psychological distress. Psychological Medicine 32,
Kessler, R. C., Berglund, P., Chiu, W. T., Demler, O., Heeringa, S.,
Hiripi, E., Jin, R., Pennell, B. E., Walters, E. E., Zaslavsky, A.
& Zheng, H. (in press). The US National Comorbidity Survey
Replication (NCS-R): an overview of design and field procedures.
International Journal of Methods in Psychiatric Research.
Kessler, R. C., Berglund, P., Demler, O., Jin, R., Koretz,
D., Merikangas, K. R., Rush, A. J., Walters, E. E. & Wang,
P. S. (2003). The epidemiology of major depressive disorder:
results from the National Comorbidity Survey Replication
(NCS-R). Journal of the American Medical Association 289,
Kessler, R. C. & Ustun, T. B. (2000). The World Health Organization
World Mental Health 2000 Initiative. Hospital Management
International, 195–196.
Kramer, H. C. (1992). Evaluating Medical Tests : Objective and
Quantitative Guidelines. Sage Publications: Newbury Park, CA.
Landis, J. R. & Koch, G. G. (1977). The measurement of observer
agreement for categorical data. Biometrics 33, 159–174.
Mannuzza, S., Klein, R. G., Bessler, A., Malloy, P. & LaPadula, M.
(1993). Adult outcome of hyperactive boys. Educational achieve-
ment, occupational rank, and psychiatric status. Archives of
General Psychiatry 50, 565–576.
Mannuzza, S., Klein, R. G., Bessler, A., Malloy, P. & LaPadula, M.
(1998). Adult psychiatric status of hyperactive boys grown up.
American Journal of Psychiatry 155, 493–498.
Mehringer, A. M., Downey, K. K., Schuh, L. M., Pomerleau, C. S.,
Snedecor, S. M. & Schbiner, H. (2002). The assessment of hyper-
activity and attention (AHA): development and preliminary vali-
dation of a brief self-assessment of adult ADHD. Journal of
Attention Disorders 5, 223–231.
Menkes, M. M., Rowe, J. S. & Menkes, J. H. (1967). A twenty-five-
year follow-up study on the hyperkinetic child with minimal brain
dysfunction. Pediatrics 39, 393–399.
Michelson, D., Adler, L., Spencer, T., Reimherr, F. W., West,
S. A., Allen, A. J., Kelsey, D., Wernicke, J., Dietrich, A. &
Milton, D. (2003). Atomoxetine in adults with ADHD: two
randomized, placebo-controlled studies. Biological Psychiatry 53,
Murphy, K. & Barkley, R. A. (1996). Attention deficit hyperactivity
disorder in adults : comorbidities and adaptive impairments.
Comprehensive Psychiatry 37, 393–401.
A screening scale for adult ADHD 255

Murphy, P. & Schachar, R. (2000). Use of self-ratings in the assess-
ment of symptoms of attention deficit hyperactivity disorder in
adults. American Journal of Psychiatry 157, 1156–1159.
O’Donnell, J. P., McCann, K. K. & Pluth, S. (2001). Assessing
adult ADHD using a self-report symptom checklist. Psychological
Reports 88, 871–881.
Pary, R., Lewis, S., Matuschka, P. R., Rudzinskiy, P., Safi, M. &
Lippmann, S. (2002). Attention deficit disorder in adults. Annals
of Clinical Psychiatry 14, 105–111.
Peirce, J. C. & Cornell, R. G. (1993). Integrating stratum-specific
likelihood ratios with the analysis of ROC curves. Medical
Decision Making 13, 141–151.
Robins, L. N., Cottler, L., Bucholz, K. & Compton, W. (1995). Diag-
nostic Interview Schedule for DSM-IV. Washington University :
St. Louis, MO.
Robins, L. N., Helzer, J. E., Croughan, J. L. & Ratcliff, K. S. (1981).
National Institute of Mental Health Diagnostic Interview
Schedule: its history, characteristics and validity. Archives of
General Psychiatry 38, 381–389.
Robins, L. N., Wing, J., Wittchen, H.-U., Helzer, J. E., Babor,
T. F., Burke, J. D., Farmer, A., Jablenski, A., Pickens, R.,
Regier, D. A., Sartorius, N. & Towle, L. H. (1988). The Composite
International Diagnostic Interview: an epidemiologic instrument
suitable for use in conjunction with different diagnostic systems
and in different cultures. Archives of General Psychiatry 45,
SAS Institute (1999). SAS User’s Guide, release 8. SAS Institute Inc. :
Cary, NC.
Shekim, W. O., Kashani, J. & Beck, N. (1985). The prevalence of
attention deficit disorders in a rural Midwestern community
sample of nine-year-old children. Journal of the American Academy
of Child Psychiatry 24, 765–770.
Spencer, T., Biederman, J., Wilens, T., Faraone, S., Prince, J.,
Gerard, K., Doyle, R., Parekh, A., Kagan, J. & Bearman, S. K.
(2001). Efficacy of a mixed amphetamine salts compound in adults
with attention-deficit/hyperactivity disorder. Archives of General
Psychiatry 58, 775–782.
Spencer, T., Biederman, J., Wilens, T., Prince, J., Hatch, M., Jones,
J., Harding, M., Faraone, S. V. & Seidman, L. (1998). Effectiveness
and tolerability of tomoxetine in adults with attention deficit hy-
peractivity disorder. American Journal of Psychiatry 155, 693–695.
Spencer, T., Wilens, T., Biederman, J., Faraone, S. V., Ablon, J. S.
& Lapey, K. (1995). A double-blind, crossover comparison of
methylphenidate and placebo in adults with childhood-onset
attention-deficit hyperactivity disorder. Archives of General Psy-
chiatry 52, 434–443.
Weiss, G., Hechtman, L., Milroy, T. & Perlman, T. (1985). Psychi-
atric status of hyperactives as adults : a controlled prospective
15-year follow-up of 63 hyperactive children. Journal of the
American Academy of Child Psychiatry 24, 211–220.
Wender, P. H. (1998). Attention-deficit hyperactivity disorder in
adults. Psychiatric Clinics of North America 21, 761–774.
Wender, P. H., Wolf, L. E. & Wasserstein, J. (2001). Adults with
ADHD. An overview. Annals of the New York Academy of
Sciences 931, 1–16.
West, S. L., Muslow, M. & Arredondo, R. (2003). Factor analysis
of the Attention Deficit Scales for Adults (ADSA) with a clinical
sample of outpatient substance abusers. American Journal of
Addiction 12, 159–165.
Wilens, T. E., Biederman, J. & Spencer, T. J. (2002). Attention
deficit/hyperactivity disorder across the lifespan. Annual Review
of Medicine 52, 113–131.
Wolter, K. M. (1985). Introduction to Variance Estimation. Springer-
Verlag: New York.
Zucker, M., Morris, M. K., Ingram, S. M., Morris, R. D. & Bakeman,
R. (2002). Concordance of self- and informant ratings of adults’
current and childhood attention-deficit/hyperactivity disorder
symptoms. Psychological Assessment 14, 379–389.
256 R. C. Kessler et al.