Journal of Psychosomatic Research 73 (2012) 10–17

Contents lists available at SciVerse ScienceDirect

Journal of Psychosomatic Research

Phenotype of subsyndromal delirium using pooled multicultural Delirium Rating

Scale—Revised-98 data☆
Paula T. Trzepacz a, b,⁎, Jose G. Franco c, d, David J. Meagher e, Yanghyun Lee f, Jeong-Lan Kim g,
Yasuhiro Kishi h, Leticia M. Furlanetto i, Daniel Negreiros i, Ming-Chyi Huang j, Chun-Hsin Chen k,
Jacob Kean l, Maeve Leonard e
a
Lilly Research Laboratories, Indianapolis, IN, USA
b
Department of Psychiatry, Indiana University School of Medicine, Indianapolis, IN, USA
c
Faculty of Medicine, Universidad Pontificia Bolivariana, Medellín, Colombia
d
Hospital Psiquiatric Universitari Institut Pere Mata, IISPV, Universitat Rovira i Virgili, Reus (Tarragona), Spain
e
Department of Psychiatry, University of Limerick School of Medicine, Limerick, Ireland
f
Department of Psychiatry, Mungyeong Jeil General Hospital, Mungyeong, South Korea
g
Department of Psychiatry, College of Medicine, Chungnam National University, Daejeon, South Korea
h
Department of Psychiatry, Nippon Medical School Musashikosugi Hospital, Kawasaki-city, Kanagawa, Japan
i
Department of Internal Medicine, Federal University of Santa Catarina, Brazil
j
Department of Psychiatry, Taipei City Psychiatric Center, Taipei City Hospital, Taipei, Taiwan
k
Department of Psychiatry, Taipei Medical University–Wan Fang Hospital, Taipei, Taiwan
l
Department of Rehabilitation Medicine, Indiana University School of Medicine, Indianapolis, IN, USA

a r t i c l e i n f o a b s t r a c t

Article history: Objective: There is no consensus definition for the phenotype of subsyndromal delirium (SSD), a subthreshold
Received 8 December 2011 state to full delirium. Without an a priori definition we applied advanced analytic techniques to discern SSD.
Received in revised form 23 April 2012 Method: We pooled Delirium Rating Scale—Revised-98 (DRS-R98) data from 859 DSM-IV diagnosed nonde-
Accepted 24 April 2012 mented delirious adults and nondelirious controls collected by investigators in 7 countries. Discriminant
analyses defined an SSD group that was then compared to Nondelirium and Delirium groups.
Keywords:
Results: SSD (n = 138) had intermediate DRS-R98 item severities between Delirium (n = 497) and Non-
Delirium
Delirium Rating Scale—Revised-98
delirium (n = 224) groups, where groups significantly differed on all DRS-R98 items (ANOVA p b .001) except
Phenotype delusions. Discriminant analysis found SSD phenomenologically closer to Delirium than Nondelirium. Using
Subsyndromal full multinomial logistical regression, SSD was distinguished from Nondelirium by temporal onset, sleep–
wake cycle, perceptual disturbances, motor retardation, delusion, affective lability, and all cognitive items;
SSD was similar to Delirium in thought process, language, motor agitation or retardation, sleep–wake
cycle, all cognitive items, fluctuation and physical disorder. The multivariate model correctly classified
94.2% of Nondelirium, 75.4% of SSD and 97.2% of Delirium subjects. Binary logistic regression of six core do-
main symptoms (sleep–wake cycle, thought process, language, attention, orientation, and visuospatial abili-
ty) together were found as highly differentiating of SSD from Nondelirium, which correctly classified almost
80% of SDD.
Conclusions: SSD is intermediate in severity between nondelirious controls and full syndromal delirium, but
its phenotype is more like delirium. Core domain delirium symptoms present at milder severity in SSD should
be evaluated further for utility in detecting and managing SSD, preventing delirium, and possible inclusion in
DSM-V.
© 2012 Elsevier Inc. All rights reserved.

Introduction

Delirium is an acute neuropsychiatric disorder impairing conscious-

ness and affecting many higher cortical functions. Application of stan-
dardized instruments has increased our understanding of delirium
☆ Research for this study was conducted by Drs. Trzepacz and Franco and data were
phenomenology. Phenomenological research supports three core do-
analyzed at the Spain research site.
⁎ Corresponding author at: Neurosciences Research, Eli Lilly and Company, India-
mains of symptoms comprised of abnormalities of attention and other
napolis, IN 46285, USA. Tel.: + 1 317 433 5391. cognition (memory, visuospatial, orientation), circadian rhythm
E-mail address: PTT@lilly.com (P.T. Trzepacz). (sleep–wake) and higher order thinking (comprehension and thought

0022-3999/$ – see front matter © 2012 Elsevier Inc. All rights reserved.
doi:10.1016/j.jpsychores.2012.04.010
 P.T. Trzepacz et al. / Journal of Psychosomatic Research 73 (2012) 10–17
process) [1]. Delirium symptoms persist at essentially comparable se-
verity throughout an episode while Diagnostic and Statistical Manual
of Mental Disorders, fourth edition (DSM-IV) criteria are met [2].
Subsyndromal delirium (SSD) is not well characterized. It is a sub-
threshold for full delirium by virtue of the number, type and/or severity
of symptoms and its occurrence may be during a prodromal or resolving
delirium phase, or even be chronically persistent. However, there are
only a few studies of SSD, where measurement methods varied. Prodro-
mal symptoms during 1 to 4 days prior to a delirium episode have in-
cluded sleep disturbance, changes in behavior, consciousness, motor
activity, thinking and cognition, and EEG slowing [1,3], supporting the
notion of a transitional state.
Most studies focused on outcomes in elderly and used a priori symp-
tom selection to define SSD. Nondelirious elderly with disorientation,
clouding of consciousness and perceptual disturbances on the Delirium
Symptom Interview had a longer term prognosis intermediate to that
for Delirium or Nondelirium [4]. Compared to controls, nondelirious el-
derly with SSD defined as at least two symptoms of clouding of con-
sciousness, inattention, disorientation or perceptual disturbances had
longer length of stay, reduced cognitive and functional status, and in-
creased institutionalization and mortality rates that correlated with the
number of SSD symptoms [5]. Compared to nondelirious at 6 months fol-
low up, those with SSD defined as the median Memorial Delirium Assess-
ment Scale (MDAS) score in nondelirious elderly hip fracture surgery
admissions had higher mortality and decline in activities of daily living
[6]. Defined as partially fulfilling the Confusion Assessment Method
(CAM) algorithm in elderly acute care geriatric admissions, SSD was an
independent predictor of post-discharge institutionalization [7].
Several reports used mixed age range adult samples and well-
validated severity rating scales without a priori symptom selection for
SSD. Combining Delirium Rating Scale (DRS) and MDAS items for
daily ratings of prodromal and syndromal deliriums in adult bone mar-
row transplantation patients, cognitive deficits, psychomotor behavior
changes, emotional lability and perceptual and thought disturbances
were present during several days prior to delirium diagnosis [8]. Previ-
ous researchers defined SSD as 8 to 13 points on the Delirium Rating
Scale—Revised-98 (DRS-R98) [9] or used a cutoff score of 15 on the
DRS-R98 among DSM-IV delirious cases [10]. These reports found com-
parable severity of SSD to delirium for sleep–wake cycle and perceptual
disturbances, motor activity changes, acute onset, and affective lability,
but intermediate severity between delirious and control groups for
most cognitive items, language, thought process and delusions.
The inadequate research database on SSD phenomenology contrasts
with its clinical relevance and possible consideration by the DSM-V
Neurocognitive Disorders Work Group as to whether SSD criteria
could be added in the proposed version of the manual. Diagnostic
criteria in DSM-IV are less restrictive than International Statistical
Classification of Diseases and Related Health Problems, 10th Revision
(ICD-10) or DSM-III-R [11,12] and, therefore, allow more cases to be di-
agnosed as delirious even if they are subsyndromal using more rigorous
definitions. SSD may represent its own risk factors, outcomes and man-
agement. Identification of those with SSD offers opportunity for pro-
phylactic interventions.
This report analyzed a large pooled research dataset of nonde-
mented delirious and nondelirious adults from investigative sites in
countries in the Americas, Europe and Asia in order to delineate the
phenomenology of SDD and used advanced statistical methods with-
out a priori assumptions as to which DRS-R98 items defined SSD.
Methods
Dataset selection and inclusion criteria
De-identified cross-sectional data for adult patients from 14 studies
(11 of them published) conducted by expert delirium researchers in 7
countries (United States, Brazil, Colombia, Ireland, Taiwan, Korea and
11
Japan) from 4 continents comprised the pooled database (see Supple-
mental Table 1S) [13–23]. Investigators were invited by Dr. Trzepacz,
the developer of the DRS-R98, to participate in this study because they
had previously received permission to use the scale for work related to
its validation. Their work, published or not yet published, needed to
meet the following criteria: 1.) all site investigators were delirium experts
and well-trained in the use of the DRS-R98; 2.) delirium and other psychi-
atric diagnoses in the original projects were made according to DSM-IV/
DSM-IVTR using all sources of clinical information; 3.) DRS-R98 ratings
were independent (blinded) from DSM-IV diagnosis and covered the pre-
vious 24 hour period; and 4.) all subjects (delirium and controls) were
evaluated cross-sectionally using DSM-IV criteria and the DRS-R98.
All studies were approved by the appropriate human ethics com-
mittees at participating sites and informed or proxy consent was
obtained as required.
Patient populations
Patients were consecutively assessed adults (n= 859) from a variety
of inpatient and outpatient clinical settings, where 516 had delirium
and 343 were nondelirious controls according to DSM-IV criteria. Con-
trols were drawn from the same samples as the delirious cases. Demo-
graphic data and comorbid psychiatric diagnoses according to DSM-IV
were collected and demented cases were excluded from the study in
order to avoid contamination from overlapping symptoms. Active med-
ical–surgical diagnoses from the original datasets were categorized
according to the Delirium Etiology Checklist (DEC) that allows classifi-
cation of diverse medical–surgical diagnoses into 13 categories [1].
Procedures
The DRS-R98 is a widely used, well-validated, specific and sensitive
rating scale with phenomenological descriptive anchors for rating the
severity levels for each item (0 is normal to a maximum of 3) with a
maximum scale score of 46 points. It is designed to evaluate the breadth
and severity of delirium symptoms for phenomenological and treat-
ment studies, in addition to diagnosing deliriums [13]. The DRS-R98 is
not derived from any particular diagnostic system and instead was de-
veloped based on known delirium characteristics. Its 16 items include 3
diagnostic items for the Total score where 13 items constitute the Se-
verity scale score. It was originally validated using raters blind to diag-
noses and results were compared against four other diagnostic groups
of inpatients. It has been subsequently translated and revalidated in
countries outside of the U.S. Translations were used, as appropriate,
and raters had been trained as part of individual projects. The DRS-
R98 versions used had very high inter-rater reliability (intraclass corre-
lation coefficient >0.9), and excellent validity as shown by the area
under the curve >0.9 (Receiver–Operator Characteristic analyses)
when comparing DSM Delirium vs. Non-delirium patients for all DRS-
R98 versions utilized in this study [13,14,16,17,19,21,22]. The DRS-
R98 broadly measures delirium phenomenology using anchored item
descriptors and has been validated against other neuropsychiatric dis-
orders making it an ideal instrument to assess phenomenology.
The DEC is a standardized tool that categorizes medical–surgical
etiologies of delirium (for example: drug intoxication, traumatic
brain injury or metabolic/endocrine disturbance). We used the DEC
category table with the only purpose of grouping for description in
a standardized way medical surgical diagnoses across all studies [1].
Statistical analysis
Comparability of DRS-R98 scores across samples
In order to evaluate the consistency of DRS-R98 across studies in-
cluded in this study Cronbach's alpha was performed for its Total, Se-
verity, and Diagnostic item scores, and then alpha was calculated for
the same item scores when removing each study. Finally, Cronbach's
 12
A B C D

P.T. Trzepacz et al. / Journal of Psychosomatic Research 73 (2012) 10–17

Fig. 1. Boxplots of DRS-R98 Total scores to illustrate the effects of the step-wise method used to define the SSD group. Part A shows the two DSM-IV-defined groups for Delirium and Nondelirium. Note the overlapping tail distributions that
may reflect SSD cases. Part B shows DRS-R98 Total scores for DSM-IV Nondelirium cases after being subjected to cluster analysis to delineate the SSD cases. Part C shows the DRS-R98 Total scores for the resultant SSD group after applying the
DRS-R98 full syndromal cutoff [≥ 13 points on Total score] to the Delirium group. Parts C and D show DRS-R98 Total and Severity scores for the three study groups used for analyses of phenomenology in this report. (Median scores are
denoted by a solid line within the boxes, boxes represent the middle 50% of scores, and the tails denote the 25th percentiles.).
 P.T. Trzepacz et al. / Journal of Psychosomatic Research 73 (2012) 10–17
alpha was calculated for the same groups of item scores by comparing
among all published and all unpublished studies.
Delineation of the SSD group
Boxplot distributions of DRS-R98 Total scores for pooled DSM-IV-
defined Delirium cases and nondelirious controls are shown in Fig. 1
( part A). To delineate an SSD group from the DSM-IV defined deliri-
ous and nondelirious groups, we analyzed DRS-R98 data in two steps.
Step 1. In order to obtain an initial group of SSD cases from DSM-IV
Nondelirium defined subjects, we analyzed DRS-R98 items from the
nondelirious cases using hierarchical cluster analysis with Ward's
method to cluster the subjects, and squared Euclidean distance (Σ of
the squares from differences between variable values) as a measure of
dissimilarity between all DRS-R98 item scores from all subjects in the
DSM-IV nondelirious defined group. This is a technique for exploratory
data analysis designed to reveal natural groupings within a collection of
data. In short, data observations close together should fall into the same
cluster while observations far apart should be in different cluster
groups. There are several methods available for performing clustering
and calculating distances of elements [24]. Ward's method was utilized
for clustering in order to minimize loss of information that occurs in the
element grouping process. The DRS-R98 item characteristics (scored al-
most the same interval) made them candidate for squared Euclidean
distance as a measure of distance (first assumption for this distance
measure), so, we accounted for possible colinearity or presence of latent
groups with different numbers of correlated variables or items (unbal-
anced groups) in cluster analysis (second assumption). Even though
we knew by previous factor analysis that DRS-R98 items are distributed
in a balanced way [9], we performed a principal component analysis be-
fore performing the cluster analysis and evaluated the Eigenvalues (i.e.,
variance indices that measure the part of the total variance induced by a
factor) of the components (factors) to determine whether colinearity
was an issue (Eigenvalues close to zero suggest a colinearity problem).
We used the square root of the ratio between the higher Eigenvalue
and the lowest one (following the Belsley criteria: values between 5
and 10 for the square root of the ratio indicate a weak colinearity prob-
lem and values between 30 and 100 indicate a moderate to strong colin-
earity problem). We did not find a colinearity problem because the
higher Eigenvalue was 3.833 and the lower was 0.268 (square root of
the ratio = 3.78). The best 2-cluster solution was selected from the
resulting dendrogram plot to separate Nondelirium and SSD cases.
Cluster analysis obtained groups may depend on method of cluster-
ing and dataset used. So, we divided the DSM-IV defined nondelirious
group according to age in two groups (>65 and all others) and per-
formed cluster analysis with a totally different method (K-means clus-
tering, a non-hierarchical, iterative method for calculation of Euclidean
distances) in both age groups. This second analysis was performed in
order to evaluate in two different subsets with a different clustering
method, the reliability of the hierarchical cluster analysis solution [24].
Percentage of subjects correctly classified in the hierarchical clusters
by the K-means method is reported.
Step 2. Boxplots of DRS-R98 Total scores were made for the 3 groups
(the two cluster-defined groups and the DSM-IV Delirium group)
(Fig. 1, part B). We then determined which cases in the Delirium
group belonged to the SSD group by applying a DRS-R98 Total cutoff
score (b13) that represented cases who were below the 75th percentile
of the SSD group in order to correct for the broad range of milder cases
that DSM-IV criteria captures. This 2-step method allowed us to differ-
entiate three groups from the original two, including a Delirium group
that is more restrictive than the DSM-IV criteria (Fig. 1, part C).
Comparison of groups
Age and DRS-R98 score differences among groups were compared
using one-way ANOVA. Between group comparisons were done using
the post hoc Games–Howell test set at b0.001 significance. Sex and
13
other psychiatric diagnosis differences between groups were com-
pared with chi-square.
To estimate the ability of DRS-R98 items to discriminate among
groups we performed two types of analysis. First, we dichotomized
the items as being absent or present at any severity (score= 0 and
score ≥ 1) and evaluated two subsamples (Nondelirium vs. SSD patients
or SSD vs. Delirium) to obtain sensitivity, specificity and percent cor-
rectly classified subjects as SSD using 2 × 2 tables in a univariate dis-
criminant analysis. Second, we used full factorial multinomial logistic
regression in the whole sample with SSD as the reference category to
discriminate which DRS-R98 dichotomized items (absence or presence)
were the most differentiating of the SSD group from the Nondelirium
and Delirium groups (i.e., which items had an exponentiated value of
the β coefficients, Exp(β), significantly b1), and which items were the
most similar between SSD group and the other two groups (i.e., which
items had an Exp(β) significantly >1) when controlling for all items'
impact. In addition to Exp(β) values for comparisons between SSD
and the other two groups, the percentage of correctly classified subjects
by the multivariate model is reported for all three groups.
Three core domains were evaluated using six DRS-R98 items (sleep–
wake cycle, language, thought process, attention, orientation and visuo-
spatial ability). We performed a binary logistic regression analysis with
those items' scores dichotomized (absent/present) in the subsample of
those with Nondelirium or SSD, with Nondelirium as the reference cat-
egory. Exp(β) is reported where significant values >1 differentiate be-
tween groups when controlling for all items in the model.
Hosmer and Lemesshow goodness of fit statistics and other literature
recommendations were followed for logistic models construction [25].
Results
Comparability of DRS-R98 scores across samples
Cronbach’s alpha for the DRS-R98 Total items in the whole sample was 0.934 (range
from 0.925 to 0.940 when removing each study from the sample), for its Severity items
was 0.917 (range from 0.907 to 0.924 when removing each study), and for its Diagnostic
items was 0.856 (range from 0.843 to 0.865 when removing each study). Cronbach’s alpha
for the DRS-R98 Total items in the 11 published studies was 0.927, for Severity items was
0.910, and for Diagnostic items, 0.847. Cronbach’s alpha for DRS-R98 Total items in the 3
unpublished studies was 0.969, for Severity items was 0.961, and for Diagnostic items,
0.908.
Delineation of the SSD group
Fig. 1 shows boxplots of DRS-R98 Total scores to illustrate the effects of the step-wise
method used to define the SSD group, starting with the DSM-IV-defined groups for Delir-
ium and Nondelirium (part A) then part B showing the effects of applying hierarchical
cluster analysis to the Nondelirium cases (88.2% of those >65 years old and 81.5% of
those ≤65 were classified into the same cluster by K-means method), and then
applying the DRS-R98 syndromal cutoff (≥13 points on DRS-R98 Total) to the Delirium
group (part C). Three groups resulted: Nondelirium (n = 224), SSD (n= 138) and Deliri-
um (n = 497). Part D shows boxplot distributions of the DRS-R98 Severity scores for these
three groups where the SSD group has an overall delirium symptom severity that is inter-
mediate to the other two groups. The lower quartile distribution for the Delirium group
overlaps with the SSD group and the upper quartile for the control group overlaps with
the SSD lower quartile, though medians were different across the three groups (Median
Tests p b .001).
Description of groups
Mean ages of the whole sample (n=859) was 62.90±17.09 and for Nondelirium,
57.40±17.83 (range 18–90), for SSD, 57.83± 19.63 (range 18–98), and Delirium 66.80
±14.81 (range 18–100), where the Delirium group was older than each of the other two
groups (F=32.96, df 2, pb .001). The proportion of males was 47.8% in the Nondelirium
group, 48.6% in SSD, and 65.0% in Delirium. There was a statistically higher proportion of
males in the Delirium as compared to both the Nondelirium (χ2 =19.026, df 1, pb .001)
and SSD groups (χ2 =12.318, df 1, pb .001).
The most frequent medical–surgical diagnoses across all patients as grouped by the
DEC were systemic neoplasm in 108 (12.6%), metabolic/endocrine disturbance in 89
(10.4%), systemic infection in 89 (10.4%), organ insufficiency in 72 (8.4%), cerebrovas-
cular disorder in 47 (5.5%), traumatic brain injury in 38 (4.4%), and intracranial neo-
plasm in 25 (2.9%).
Two hundred sixty-seven (30.1%) subjects had at least one comorbid psychiatric
disorder besides delirium. The three most frequent comorbid disorders were
14 P.T. Trzepacz et al. / Journal of Psychosomatic Research 73 (2012) 10–17

Table 1
Individual DRS-R98 items for Nondelirium, SSD, or Delirium groups.

DRS-R98 Nondelirium SSD Delirium

(n = 224) (n = 138) (n = 497)

Score max Mean SD Range Mean SD Range Mean SD Range

Sleep–wake cycle 3 0.33 0.53 0–2 0.91 0.70 0–3 2.02 0.72 0–3
Perceptual disturbances 3 0.05 0.36 0–3 0.88 1.26 0–3 1.69 1.26 0–3
Delusion 3 0.07 0.33 0–3 0.75 1.19 0–3 0.93 1.06 0–3
Affective lability 3 0.09 0.29 0–1 0.47 0.64 0–3 1.30 0.95 0–3
Language 3 0.05 0.22 0–1 0.23 0.48 0–2 1.20 0.89 0–3
Thought process 3 0.04 0.21 0–1 0.46 0.78 0–3 1.50 0.94 0–3
Motor agitation 3 0.10 0.30 0–1 0.32 0.60 0–3 1.54 1.01 0–3
Motor retardation 3 0.09 0.34 0–2 0.34 0.65 0–2 0.83 0.91 0–3
Orientation 3 0.10 0.30 0–1 0.52 0.61 0–3 1.71 0.86 0–3
Attention 3 0.14 0.38 0–2 0.69 0.69 0–3 2.16 0.79 0–3
Short-term memory 3 0.12 0.47 0–2 0.92 0.98 0–3 2.00 1.04 0–3
Long-term memory 3 0.12 0.41 0–3 0.78 0.87 0–3 1.78 1.00 0–3
Visuospatial ability 3 0.06 0.25 0–2 0.68 0.84 0–3 1.93 1.02 0–3
Temporal onset 3 0.20 0.54 0–3 1.07 1.05 0–3 2.22 0.73 0–3
Fluctuation 2 0.01 0.09 0–1 0.33 0.50 0–2 1.20 0.50 0–2
Physical disorder 2 0.18 0.47 0–2 0.77 0.83 0–2 1.74 0.46 0–2
Severity score 39 1.37 1.67 0–6 7.92 2.51 3–15 20.61 6.26 8–37
Total score 46 1.83 2.09 0–7 10.13 2.82 4–17 25.75 6.54 13–42

All item means between groups were significantly different at p b .001 using a post hoc Games-Howell test comparison for all 3 groups, except for bold and italicized values
(delusion).

Depressive disorders, Schizophrenia or other psychotic disorder, and Alcohol use disor-
der (abuse or dependence): 25 (11.2%) in the Nondelirium group, 24 (17.4%) in the
SSD group and 59 (11.9%) in the Delirium group had Depressive Disorders; 21 (9.4%)
in the Nondelirium group, 39 (28.3%) in the SSD group and 5 (1.0%) in the Delirium
group had Schizophrenia or other Psychotic disorder; 9 (4.0%) in the Nondelirium
group, 7 (5.1%) in the SSD group and 32 (6.4%) in the Delirium group had Alcohol
Use disorder. Only the Schizophrenia or other Psychotic disorder diagnosis frequency
was statistically different among the groups (p b .001 for all chi2).
Sensitivity, specificity and the percent of subjects correctly classified as SSD by indi-
vidual (dichotomized) DRS-R98 items (using univariate discriminant analysis) are listed
in Table 2. The percent correctly classified as SSD as compared to Nondelirium was at
least 65% for every item indicating very good differentiation of SSD from Nondelirium,
whereas the percent correctly classified as SSD as compared to Delirium was much
lower with the lowest at 10.1% and only three items were above 30% though less than
45% (motor retardation, perceptual disturbances and delusion) indicating much similarity
to Delirium.

Comparison for individual delirium symptoms Full multinomial logistic model

Mean scores for the DRS-R98 and its items are shown by group in Table 1. Using
ANOVA, all item scores were significantly different across groups (p b .001) such that
the SSD group's symptoms were intermediate in severity between Nondelirium and
Delirium groups, except for delusions that were different from Nondelirium but similar
for SSD and Delirium groups. DRS-R98 item mean score distributions for the three
groups are visually represented in a radar graph in Fig. 2, where patterns of symptoms
are generally more similar between SSD and Delirium than SSD and Nondelirium, ex-
cept that language appears less impaired in SSD than in Delirium.
The full multinomial logistic regression model using the DRS-R98 items and dichoto-
mized scale scores comparing SSD to the other two groups is shown in Table 3. Given that
comorbid schizophrenia or other psychotic disorders were more frequent in the SSD
group, we conducted the analysis using those as a covariate. We also ran the model with-
out including this covariate. Both models identified essentially the same DRS-R98 items as
being significant in the comparisons except that sleep–wake cycle was not similar be-
tween SSD and Delirium in the model not accounting for psychotic disorder comorbidity
even though its Exp(β) value was only slightly lower (3.669 vs. 4.069).

Fig. 2. Radar chart of DRS-R98 item severity scores for 859 patients by diagnostic group for Delirium, SSD, or Nondelirium.
 P.T. Trzepacz et al. / Journal of Psychosomatic Research 73 (2012) 10–17 15

Table 2
Discriminant analysis of individual DRS-R98 items for differentiating SSD from Nondelirium (n = 362) and from Delirium (n = 635).

DRS-R98 item Item's accuracy of distinguishing SSD from Nondelirium Item's accuracy of distinguishing SDD from Delirium

Sensitivity % Specificity % % Correctly classified as SSD Sensitivity % Specificity % % Correctly classified as SSD

Sleep–wake cycle 72.5 69.2 70.4 72.5 2.6 17.8

Perceptual 39.1 97.8 75.4 39.1 28.6 30.9
Delusion 33.3 95.1 71.5 33.3 46.3 43.5
Affective lability 46.0 90.6 71.5 40.6 23.3 27.1
Language 21.0 95.1 66.8 21.0 23.1 22.7
Thought process 32.6 95.5 71.5 32.6 16.9 20.3
Motor agitation 26.1 90.2 65.7 26.1 19.7 21.1
Motor retardation 25.4 92.9 67.1 25.4 47.3 42.5
Orientation 46.1 89.7 73.5 47.1 8.4 16.8
Attention 57.2 87.5 76.0 57.2 1.0 13.2
Short-term memory 56.5 84.8 74.0 56.6 10.5 20.5
Long-term memory 55.8 90.1 77.1 55.8 12.3 21.7
Visuospatial ability 47.8 95.1 77.1 47.8 11.7 19.5
Temporal onset 63.8 85.3 77.1 63.8 0.6 14.3
Fluctuation 31.2 99.1 73.2 31.2 4.2 10.1
Physical disorder 52.2 85.3 72.6 52.2 1.2 12.3

For the SSD and Nondelirium analysis, sleep–wake cycle, perceptual disturbances, R2 = 0.60). Exp(β) values that were significantly >1 were found for almost all these items
delusion, affective lability, motor retardation, all cognitive items and temporal onset (except for language) and separated the two groups as follows (with 95%CI): sleep–wake
were worse in SSD as compared to Nondelirium controls. The other 5 items were sim- cycle = 5.85 (3.04–11.23), language= 1.65 (0.62–4.37), thought process= 7.06
ilar between SSD group and controls. Exp(β) values for 11 items were significantly b 1 (2.71–18.37), orientation= 3.43 (1.68–6.99), attention = 3.37 (1.74–6.53), and visuospa-
and perceptual disturbances and visuospatial ability had the lowest Exp(β) value tial ability = 8.05 (3.62–17.88). Each item except language individually differentiated
among symptom items and temporal onset had the lowest Exp(β) value among diag- SSD from Nondelirium (Wald statistics significant at b 0.05).
nostic characteristic items.
For the SSD and Delirium analysis sleep–wake cycle, language, thought process,
motor agitation and retardation, all cognitive items, fluctuation and physical disorder
were similar between SSD and Delirium groups and only perceptual disturbances, delu- Discussion
sions and affective lability were different. Attention had the highest Exp(β) value
among symptom items, and fluctuation and physical disorder had the highest Exp(β) We describe an SSD phenotype using pooled DRS-R98 data from 859
values among diagnostic characteristic items. Moreover, sleep–wake cycle, motor retarda- nondemented adults in 14 studies from 7 countries, comprising the
tion, and all cognitive items differentiated SSD from Nondelirium but not from Delirium.
The multivariate model correctly classified 94.2% of the Nondelirium subjects,
largest SSD phenomenology report to date. Rather than making a priori
75.4% of the SSD and 97.2% of the Delirium subjects. assumptions about what constitutes the SSD phenotype, we applied hi-
erarchical cluster analysis to delineate the SSD group. In general, SSD
was found to be intermediate in symptom severity from the two other
Core domains of delirium
groups, though Exp(β) values in regression models suggested more
Binary logistic regression analysis of six DRS-R98 items representing core domains similarity between SSD and Delirium than Nondelirium. Further, five
correctly classified 87.1% of Nondelirium and 79.7% of SSD subjects (n = 362) (Nagelkerke of six core domain symptoms of delirium significantly distinguished
SSD from nondelirious controls. Our data suggest that SSD is a valid clin-
ical state linked to delirium as a unique neuropsychiatric syndrome.
Table 3
Full factorial multinomial logistic regression analysis was performed with the SSD Though our findings are consistent with prior descriptions of SSD,
group as reference category for DRS-R98 items in 859 patients with Delirium, SSD, or they offer more clarity and focus for the SSD phenotype because of
Nondelirium. DRS-R98 items in bold denote those that distinguished the SSD group our analytic methods and use of the DRS-R98 which is specifically
from Nondelirium (Exp(β) significantly b 1) and those that made SSD similar to Delir- designed for detailed phenomenological research.
ium (Exp(β) significantly > 1) (p b .05 for the Wald statistic and with 95% confidence
intervals within the range). The model takes into account a contribution from primary
Symptoms that distinguished SSD from Nondelirium were
psychotic disorder as a covariate. acute onset, psychosis, and alterations in cognition, sleep–wake
cycle, motor retardation and affective lability. Psychosis was less
Multivariate modela
common than other symptoms but when present it is linked to
DRS-R98 item Exp(β) for comparison Exp(β) for comparison SSD. These are consistent with naturalistic literature that found
between SSD and between SSD and
cognitive deficits in SSD or prodromal delirium [8]. Our finding of
Nondelirium groups Delirium groups
affective lability reflects reports of mood and behavioral changes
Sleep–wake cycle 0.147 4.092 in SSD [26,27]. Attentional deficits and circadian abnormalities
Perceptual disturbances 0.002 2.158
Delusion 0.230 1.591
have been described as two of the three core domains of delirium
Affective lability 0.088 1.323 [10,15], and we found these to be present in SSD and differentiating
Language 2.093 4.124 from Nondelirium. Inattention measured by computerized testing
Thought process 0.494 4.688 differentiated SSD from controls in orthopedic surgery inpatients,
Motor agitation 0.885 3.741
supporting the cardinal symptom status of inattention in SSD as
Motor retardation 0.127 4.088
Orientation 0.126 4.558 well as in delirium [28]. We found that core domain symptoms of
Attention 0.124 6.767 delirium were comparable between SSD and Delirium and correctly
Short-term memory 0.076 4.025 classified 79.7% of the SSD subjects compared to the Nondelirium
Long-term memory 0.025 3.946 group, and recommend that future studies of SSD include these
Visuospatial ability 0.019 5.186
Temporal onset 0.076 3.518
three core domains.
Fluctuation 0.160 8.221 Extrapolating from our results and the phenomenological descrip-
Physical disorder 1.375 14.657 tions for DRS-R98 items for milder severities, we propose that clini-
a
χ2 for the likelihood ratio test = 1394,944, df = 34, p b .001, Cox and Snell Pseudo cians interested in detecting SSD should search for an acute change
R2 = 0.803. from baseline for a combination of the following:
16 P.T. Trzepacz et al. / Journal of Psychosomatic Research 73 (2012) 10–17
1. Mild sleep continuity disturbances at night or occasional drowsi-
ness during the day
2. Tangential or circumstantial thought processes
3. Disorientation to time or place
4. Mild distractibility or difficulty sustaining attention, but ability to
refocus with cueing
5. Mild difficulty navigating in one's surroundings.
This can be accomplished through routine cognitive bedside test-
ing that includes an attention task plus observations of acute changes
in thinking ability, emotional control and sleep–wake cycle. Unfortu-
nately these symptoms are not captured by two commonly used
tools, the Mini-Mental State Examination (MMSE) and the CAM [29]
so we cannot recommend their use to detect SSD. A test of simple vi-
sual attention span may be the best single cognitive test for delirium
detection [15,30] and measures two of the five domains in the above
list. The 3-item Delirium Diagnostic Tool-Provisional (DDT-Pro),
which quantitates only the core domains, highly distinguished deliri-
ous from nondelirious post-traumatic brain injury patients [20] and
might be evaluated as a possible tool for SSD.
In summary, we delineated a plausible set of symptoms that de-
scribe SSD and have face validity with what is known about the core do-
mains of delirium. As seen in Fig. 2, the SSD group separated nicely from
the other two groups and its phenomenological pattern mirrors that of
the Delirium group. We also advocate for inclusion of SSD as a condition
in DSM-V and ICD-11 coded within the Delirium category. Inclusion of
an SSD coding will prompt clinicians to become more cognizant of it
and also encourage more research in this important area by standardiz-
ing its definition. Our findings contribute meaningfully to the emerging
literature about the importance of recognizing a subthreshold state of
delirium and we advocate provisional clinical use of the five proposed
symptoms to raise awareness of SSD and possibly improve patient
care through detection, management and prevention of full delirium.
Limitations of our report include that our pooled analysis across con-
tinents did not allow for a single DRS-R98 interrater reliability assess-
ment across all raters, although all were trained on the DRS-R98 as
part of other delirium research. Further, the consistency of items
was high (>84%) according to Cronbach's alpha, with little variation
(≤6.1%) when scores for DRS-R98 Total, Severity, and Diagnostic
items were evaluated for either whole sample, samples when excluding
each study, published or unpublished studies. Dementia was excluded
clinically and not with an instrument, though many of our subjects
were not elderly. Validation studies of DRS-R98 translations report
slightly different full syndromal delirium cutoff values that could
make pooling data difficult, but we used statistical methods to first dis-
cern differences based on phenomenology between SSD and Non-
delirium before applying a cutoff (b13 points) for SSD vs. Delirium.
The latter cutoff may have excluded some cases in countries where
the cutoff value is higher and whose scores were in the SSD range but
attributed to the Delirium group in our report. Although our K-means
cluster analysis in different subsets of patients performed to evaluate
reliability of hierarchic method showed that almost 9/10 of those
>65 years old and that more than 8/10 of those ≤65 were correctly
classified in its hierarchical clusters, the data-dependence of cluster
analysis warrants further studies in diverse populations in order to rep-
licate our findings. Another limitation is the cross-sectional nature of our
data so we cannot comment on relationships in temporal onset among
individual key distinguishing features of SSD. However, our data suggest
that evaluation for an acute change from baseline for the presence of a
cluster of symptoms that define SSD, even though their severity is still
a subthreshold for syndromal delirium, is a reasonable approach and
that whenever these appear together the threshold for reaching an SSD
state is achieved and is relevant. This cluster represents the same core
phenotype as delirium and thereby reduces the likelihood of nonspecific
isolated neuropsychiatric symptoms being erroneously attributed to SSD
as can occur using previously reported a priori definitions.
Supplementary data related to this article can be found online at
http://dx.doi.org/10.1016/j.jpsychores.2012.04.010.
Disclosure statement
Dr. Trzepacz is a full-time salaried employee and minor shareholder
at Eli Lilly and Company. Dr. Trzepacz holds the copyright for the Delir-
ium Rating Scale—Revised-98 but does not charge a fee for a not-for-
profit use.
No other coauthors have conflicts of interest to disclose.
Role of funding source
There was no formal funding for this study.
Competing interest statement
All authors will complete the Unified Competing Interest form at
www.icmje.org/coi_disclosure.pdf upon notification of acceptance.
All relevant author disclosures and/or competing interests have
been declared in the previous section, “Disclosure statement.”
References
[1] Trzepacz PT, Meagher DJ, Leonard M. Delirium. In: Levenson J, editor. American
psychiatric publishing textbook of psychosomatic medicine. 2nd edition. Wash-
ington, DC: American Psychiatric Publishing; 2011.
[2] Meagher DJ. A study of symptom profile and clinical subtypes of delirium. PhD
Thesis, Dublin City University, Dublin, Ireland, 2011.
[3] Osse RJ, Tulen JH, Bogers AJ, Hengeveld MW. Disturbed circadian motor activity
patterns in postcardiotomy delirium. Psychiatry Clin Neurosci 2009;63:56–64.
[4] Levkoff SE, Liptzin B, Cleary PD, Wetle T, Evans DA, Rowe JW, et al. Subsyndromal
delirium. Am J Geriatr Psychiatry 1996;4:320–9.
[5] Cole M, McCusker J, Dendukuri N, Han L. The prognostic significance of sub-
syndromal delirium in elderly medical inpatients. J Am Geriatr Soc 2003;51:754–60.
[6] Marcantonio E, Ta T, Duthie E, Resnick NM. Delirium severity and psychomotor
types: their relationship with outcomes after hip fracture repair. J Am Geriatr
Soc 2002;50:850–7.
[7] Bourdel-Marchasson I, Vincent S, Germain C, Salles N, Jenn J, Rasoamanarivo E,
et al. Delirium symptoms and low dietary intake in older inpatients are indepen-
dent predictors of institutionalization: a 1-year prospective population-based
study. J Gerontol A Biol Sci Med Sci 2004;59:350–4.
[8] Fann JR, Alfano CM, Burington BE, Roth-Roemer S, Katon WJ, Syrjala KL. Clinical
presentation of delirium in patients undergoing hematopoietic stem cell trans-
plantation. Cancer 2005;103:810–20.
[9] Franco JG, Trzepacz PT, Mejia MA, Ochoa SB. Factor analysis of the Colombian trans-
lation of the Delirium Rating Scale (DRS), Revised-98. Psychosomatics 2009;50:
255–62.
[10] Meagher DJ, Trzepacz PT. Delirium. In: Gelder M, Andreasen N, Lopez-Ibor JJ,
Geddes J, editors. Oxford textbook of psychiatry. Oxford, England: Oxford Univer-
sity Press; 2009. p. 325–32.
[11] Laurila JV, Pitkala KH, Strandberg TE, Tilvis RS. The impact of different diagnostic
criteria on prevalence rates for delirium. Dement Geriatr Cogn Disord 2003;16:
156–62.
[12] Cole MG, Dendukuri N, McCusker J, Han L. An empirical study of different diagnos-
tic criteria for delirium among elderly medical inpatients. J Neuropsychiatry Clin
Neurosci 2003;15:200–7.
[13] Trzepacz PT, Mittal D, Torres R, Kanary K, Norton J, Jimerson N. Validation of the
Delirium Rating Scale—Revised-98: comparison with the delirium rating scale
and the cognitive test for delirium. J Neuropsychiatry Clin Neurosci 2001;13:
229–42.
[14] Lim K, Kim S, Lee Y, Lee S, Kim J. A validation study for the Korean version of De-
lirium Rating Scale—Revised-98 (K-DRS-98). J Korean Neuropsychiatr Assoc
2006;45:518–26.
[15] Meagher DJ, Moran M, Raju B, Gibbons D, Donnelly S, Saunders J, et al. Phenome-
nology of delirium: assessment of 100 adult cases using standardized measures.
Br J Psychiatry 2007;190:135–41.
[16] Franco JG, Mejia MA, Ochoa SB, Ramirez LF, Bulbena A, Trzepacz P. Delirium Rat-
ing Scale—Revised-98 (DRS-R-98): Colombian adaptation of the Spanish version.
Actas Esp Psiquiatr 2007;35:170–5.
[17] de Negreiros DP, da Silva Meleiro AM, Furlanetto LM, Trzepacz PT. Portuguese ver-
sion of the Delirium Rating Scale—Revised-98: reliability and validity. Int J Geriatr
Psychiatry 2008;23:472–7.
[18] Lee J, Lee H, Kim S, Kim J. A study for the diagnosis, severity and subtypes of de-
lirium using the Korean version of Delirium Rating Scale—Revised-98 (K-DRS-
98). J Korean Soc Biol Ther Psychiatry 2008;14:101–9.
[19] Huang MC, Lee CH, Lai YC, Kao YF, Lin HY, Chen CH. Chinese version of the Delirium
Rating Scale—Revised-98: reliability and validity. Compr Psychiatry 2009;50:81–5.
 P.T. Trzepacz et al. / Journal of Psychosomatic Research 73 (2012) 10–17
[20] Kean J, Trzepacz PT, Murray LL, Abell M, Trexler L. Initial validation of a brief pro-
visional diagnostic scale for delirium. Brain Inj 2010;24:1222–30.
[21] Kato M, Kishi Y, Okuyama T, Trzepacz PT, Hosaka T. Japanese version of the Delir-
ium Rating Scale, Revised-98 (DRS-R98-J): reliability and validity. Psychosomatics
2010;51:425–31.
[22] Lee Y, Ryu J, Lee J, Kim HJ, Shin IH, Kim JL, et al. Korean version of the Delirium Rating
Scale—Revised-98: reliability and validity. Psychiatry Investig 2011;8:30–8.
[23] Leonard M, Donnelly S, Conroy M, Trzepacz PT, Meagher DJ. Phenomenological
and neuropsychological profile across motor variants of delirium in a palliative
care unit. J Neuropsychiatry Clin Neurosci 2011;23:180–8.
[24] Everitt BS, Landau S, Leese M, Stahl D. Cluster analysis. 4th edition. London, En-
gland: Hodder Arnold; 2001.
[25] Franco JG, Gaviria AM, Torres Y, Cotes JM. Regresión logística en la literatura
psiquiátrica: evaluación de los artículos publicados entre 2002 y 2005 en una
prominente revista. Rev Bras Epidemiol 2007;10:370–9.
17
[26] Morse RM, Litin EM. The anatomy of a delirium. Am J Psychiatry 1971;128:111–6.
[27] Ohki T, Matsushima E, Shibuya M, Sunamori M. An evaluation strategy for the early
detection of postoperative delirium. Psychiatry Clin Neurosci 2006;60:277–82.
[28] Lowery DP, Wesnes K, Brewster N, Ballard C. Subtle deficits of attention after sur-
gery: quantifying indicators of sub syndrome delirium. Int J Geriatr Psychiatry
2010;25:945–52.
[29] Inouye SK, van Dyck CH, Alessi CA, Balkin S, Siegal AP, Horwitz RI. Clarifying con-
fusion: the confusion assessment method. A new method for detection of deliri-
um. Ann Intern Med 1990;113:941–8.
[30] Meagher DJ, Leonard M, Donnelly S, Conroy M, Saunders J, Trzepacz PT. A compar-
ison of neuropsychiatric and cognitive profiles in delirium, dementia, comorbid
delirium-dementia and cognitively intact controls. J Neurol Neurosurg Psychiatry
2010;81:876–81.