LETTERS TO THE EDITOR
J Clin Psychiatry 68:1, January 2007
Predicting Pharmacotherapy Outcome From a
Retrospective Chart Study?
Sir: We read with interest the report by Shetti et al.1 in the
December 2005 issue of the Journal on predicting nonresponse
in obsessive-compulsive disorder (OCD). Though articles on
predicting response to pharmacotherapy for OCD should be
encouraged, this article gives rise to important methodological
First, the authors plead for a more accurate definition of non-
response in OCD, but fail to define nonresponse adequately.
While they describe nonresponse as having had at least 2 inef-
fective serotonin reuptake inhibitor (SRI) trials, the mean num-
ber of trials in the response group was higher than 1, indicating
that some subjects in the response group had had at least 2
ineffective SRI trials as well; they should therefore be qualified
as nonresponders. Apart from this, the criterion used in the
study to define response was a Clinical Global Impressions
scale (CGI) score of 1 or 2, whereas nonresponse was defined as
a CGI score of 3 to 7. The gold standard for assessing response
in OCD is a decrease on the Yale-Brown Obsessive Compulsive
Scale (YBOCS) of 25% to 35%.2 The authors state that they pre-
ferred to use the CGI since the YBOCS lacks sensitivity to mea-
sure changes. We believe that not using any OCD scale is even
less specific and accurate than using the YBOCS with its limits.
Second, the study is seriously flawed by the retrospective na-
ture of the design. Subjects were included after they completed
at least 2 medication trials. At that time, the CGI was adminis-
tered and the treatment history was reviewed with the subjects
with reference to their charts. This retrospective design carries
the risk of attrition. Nonresponse is often a reason for not com-
pleting a medication trial; nonresponders as well as dropouts
due to side effects are not accounted for in this study. The ideal
design for identifying predictors would be prospective, which
seems to be entirely feasible for this topic. Similarly, one of the
inclusion criteria, i.e., after the medication trials, is a YBOCS
score higher than 15, or 8 in the case of predominant obsessions
or compulsions. This method implies that responders having a
YBOCS score less than 15 (or 8) have been excluded before-
Third, crucial aspects from the pharmacotherapy trials have
not been presented clearly. All subjects had had at least 10
weeks of any SRI treatment, but the mean duration of treatment
is not reported. It is not even clear whether all subjects were still
taking the medication at the time of assessment. No information
is available on the mean dosages used in the study. It is possible
that nonresponders experienced more side effects and therefore
received lower mean doses.
Finally, the authors slightly overrate the uniqueness of their
study by claiming that they were the first to systematically char-
acterize SRI nonresponders in OCD. A number of previous stud-
ies have included subjects with a history of SRI treatment (see
Denys et al.3). In our own study,3 81% of subjects had a history
of at least 1 medication trial before entering the study. More-
over, in our study, we presented a prediction model to dis-
criminate response from nonresponse based on a prospective
standardized medication trial in 150 patients. We offered an ex-
haustive review of the literature on response prediction in OCD.
It seems that Shetti et al. failed to screen the existing literature
on predicting response to pharmacotherapy for OCD, since none
of these topics was taken into consideration.
Drs. de Bruijn and Denys report no financial or other affiliation
relevant to the subject of this letter.
1. Shetti CN, Reddy YCJ, Kandavel T, et al. Clinical predictors of drug
nonresponse in obsessive-compulsive disorder. J Clin Psychiatry
2. Goodman WK, Price LH, Rasmussen SA, et al. The Yale-Brown
Obsessive Compulsive Scale, 1: development, use, and reliability.
Arch Gen Psychiatry 1989;46:1006–1011
3. Denys D, Burger H, van Megan H, et al. A score for predicting
response to pharmacotherapy in obsessive-compulsive disorder.
Int Clin Psychopharmacol 2003;18:315–322
Carla de Bruijn, M.D., Ph.D.
De Waag Outpatient Clinic for Forensic Psychiatry
Amsterdam, the Netherlands
Damiaan Denys, M.D., Ph.D.
Department of Psychiatry
University Medical Centre Utrecht
Utrecht, the Netherlands
Drs. Reddy and Kandavel Reply
Sir: We thank Drs. de Bruijn and Denys for their comments
on our recently published article1 on prediction of drug nonre-
sponse in obsessive-compulsive disorder (OCD). Following are
our responses to their comments.
Drs. de Bruijn and Denys claim that our definition of nonre-
sponse is inadequate because some patients in the responder
group may have had at least 2 ineffective serotonin reuptake
inhibitor (SRI) trials. We would like to clarify here that our
sample of responders had 5 subjects who had had more than 2
trials. We classified them as responders because they had re-
sponded to later trials. The argument is that they should have
been classified as nonresponders. However, it is difficult to
classify patients who have responded to treatment as nonre-
sponders simply because, at the time of assessment, they had
improved and the ratings also demonstrated that they had re-
sponded. It would have been possible for us to classify these
5 patients as nonresponders only if we had performed assess-
ments at the time of nonresponse to 2 trials.
We eliminated these 5 patients from the responder group to
see if our results change in any meaningful way. By eliminating
the 5 patients, it can be ensured that the sample had no subjects
who were once nonresponders but who had later become re-
sponders. We performed multiple logistic stepwise forward re-
gression analysis for the same variables to identify predictors of
nonresponse using 55 nonresponders and 62 responders. In per-
forming our original regression analysis, we had chosen only
those variables that were significant in the univariate analysis.
Even with the reduced sample size of responders, variables that
were significant in the univariate analysis were the same. There-
fore, we used the identical predictors.
The final model resulted in 5 variables with 78% overall
correct prediction. The 5 variables that significantly predicted
nonresponse were major depressive disorder (β = 3.534, SE =
1.193, p = .003, OR = 34.277), washing compulsions (β =
1.837, SE = 0.570, p = .001, OR = 6.278), sexual obsessions
(β = 1.709, SE = 0.642, p = .008, OR = 5.525), the baseline
Yale-Brown Obsessive Compulsive Scale (YBOCS) severity
score (β = .069, SE = 0.034, p = .044, OR = 1.072), and age
(β = –0.071, SE = 0.032, p = .027, OR = 0.932). It is clear that
predictors have remained largely similar, with the exclusion
of only miscellaneous compulsions from the model. In the uni-
variate analysis, there was a significant difference between
LETTERS TO THE EDITOR
J Clin Psychiatry 68:1, January 2007
Correction: Supplement 13, 2006
In the article “Trends in the Pharmacologic Management of Insomnia” by Paul P. Doghramji, M.D., F.A.A.F.P. (2006, Supplement
13, pp. 5–8), an incorrect version of Table 1 was printed. The corrected Table 1, along with corrected references, appears below.
The online version of the article has been corrected.
The staff regrets the error.
Table 1. Comparisons of Medications Approved for the Treatment of Insomniaa
Dosage (mg) 5, 104
Restricted to short-term usageYes4
Number of awakenings…20–22
Wake after sleep onset…b
Total sleep time
aData from Sonata [prescribing information],4 Ambien [prescribing information],5 Ambien CR [prescribing information],6 Lunesta [prescribing
information],7 Rozerem [prescribing information],8 Ancoli-Israel et al.,9 Perlis et al.,10 Erman et al.,11 Halas,12 Erman et al.,13 Roth et al.,16
Elie et al.,20 Ancoli-Israel et al.,21 Hedner et al.,22 and Scharf et al.23
bZaleplon is known to have no effect on wake after sleep onset because of its short half-life.
Symbols: ↑ = increased, ↓ = decreased, … = no consistent effect.
Abbreviations: BZ = benzodiazepine, MT = melatonin.
20. Elie R, Ruther E, Farr I, et al., for the Zaleplon Clinical Study Group. Sleep latency is shortened during 4 weeks of treatment with zaleplon, a novel
nonbenzodiazepine hypnotic. J Clin Psychiatry 1999;60:536–544
21. Ancoli-Israel S, Walsh JK, Mangano RM, et al., for the Zaleplon Clinical Study Group. Zaleplon, a novel nonbenzodiazepine hypnotic, effectively treats
insomnia in elderly patients without causing rebound effects. Prim Care Companion J Clin Psychiatry 1999;1:114–120
22. Hedner J, Yaeche R, Emilien G, et al., for the Zaleplon Clinical Study Group. Zaleplon shortens subjective sleep latency and improves subjective sleep quality
in elderly patients with insomnia. Int J Geriatr Psychiatry 2000;15:704–712
23. Scharf MB, Roth T, Vogel GW, et al. A multicenter, placebo-controlled study evaluating zolpidem in the treatment of chronic insomnia. J Clin Psychiatry
BZ1 and BZ27
1, 2, 37
MT1 and MT28
Figure 1. Mean Binge Frequency Over 16 Weeks of Treatment
in Patients With Binge Eating Disorder Randomly Assigned
to Zonisamide or Placebo
Weekly Binge Episodes
02468 1012 1416 Endpoint
Figure 2. Weight Loss in Patients With Binge Eating
Disorder Randomly Assigned to 16 Weeks of Double-Blind
Treatment With Zonisamide or Placebo
Mean Weight Change, kg
Zonisamide (N= 30)
Placebo (N= 30)
02468 10 121416Endpoint
Correction: December 2006
In the article “Zonisamide in the Treatment of Binge Eating Disorder With Obesity: A Randomized Controlled Trial”
by Susan L. McElroy, M.D., et al. (December 2006 issue, pp. 1897–1906), there should be no lines connecting week 16
to endpoint in Figures 1 and 2. The corrected figures are shown below, and the online version of the article has been corrected.
The staff regrets the error.