Directory UMM :Data Elmu:jurnal:B:Biological Psichatry:Vol48.Issue6.2000:

Methodological Issues in Developing New Acute
Treatments for Patients with Bipolar Illness
A. John Rush, Robert M. Post, Willem A. Nolen, Paul E. Keck, Jr., Trisha Suppes,
Lori Altshuler, and Susan L. McElroy
One important aim of the recent reorganization of the
National Institute of Mental Health (NIMH) is to streamline the development of new treatments for patients with
severe mental illnesses, such as bipolar disorder. Researching new treatments for patients with bipolar disorder presents specific problems not readily addressed by
traditional efficacy trial methodologies that aim to maximize internal validity. This article reexamines several
assumptions that have guided the design of these efficacy
trials but that also create obstacles for studies of bipolar
disorder and suggests potential solutions. This article
draws on literature from neurology and psychiatry and
discussions at a MacArthur Foundation–sponsored Conference on Longitudinal Methodology in 1992 (David J.
Kupfer, M.D., Chair), which brought together investigators to consider alternative designs for patients with
severe and persistent mental illness. In addition, we
benefited from discussions at two NIMH-sponsored conferences, one held in 1989 (Prien and Potter 1990) and the
other in 1994 (Prien and Rush 1996), at which investigators and methodologists discussed issues surrounding the
development and conduct of informative efficacy trials for
patients with bipolar disorder. Based on these discussions
and recent literature reviews, we 1) outline common
problems in the development and evaluation of effective

acute treatments for bipolar disorder and 2) suggest
possible solutions to these impediments. We also discuss
alternative designs by which to build a sequence of acute
treatment studies from which efficacy, safety, and the
comparative value of different treatments can be
established. Biol Psychiatry 2000;48:615– 624 © 2000
Society of Biological Psychiatry

From the Department of Psychiatry, University of Texas Southwestern Medical
Center, Dallas (AJR, TS), Biological Psychiatry Branch, National Insititute of
Mental Health, Bethesda, Maryland, (RMP), HC Ru¨mke Groep en University
Medical Centre Utrecht, Utrecht, The Netherlands (WAN), UCLA Neuropsychiatric Institute and West Los Angeles VA Medical Center, Los Angeles,
California (LA), Biological Psychiatry Program, Department of Psychiatry,
University of Cincinnati College of Medicine, Cincinnati, Ohio (PEK, SLM),
and the Stanley Foundation Bipolar Network, Bethesda, Maryland (AJR, RMP,
WAN, PEK, TS, LA, SLM).
Address reprint requests to A. John Rush, M.D., UT Southwestern Medical Center,
Department of Psychiatry, 5323 Harry Hines Boulevard (9086), Dallas TX
75390-9086.
Received January 4, 2000; revised March 29, 2000; accepted April 4, 2000.

© 2000 Society of Biological Psychiatry

Key Words: Acute phase treatment, bipolar illness, methodological issues

Defining the Impediments
Studies of new acute phase treatments for bipolar disorder
designed to assess efficacy, safety, and tolerability have
largely been modeled after traditional methodologies used
to develop and evaluate acute phase treatments for patients
with nonpsychotic major depressive disorder. This approach usually entails the identification of drug-free,
moderately symptomatic, acutely nonsuicidal outpatients
with moderate levels of disability. These patients provide
informed consent to participate in a randomized, doubleblind, acute phase comparison trial of the new treatment to
a standard control treatment, pill placebo, or both for 6 to
12 weeks. Clinical outcomes typically include symptom
severity (assessed by itemized symptom checklists) and
adverse effects, each obtained at frequent intervals.

Limitations of the Classic Acute Phase

Efficacy Trial
The traditional approach to evaluating the efficacy and
safety of new medications is often extremely difficult to
implement for bipolar disorder. Even if successful, it may
still preclude informative generalizability, depending on
the restrictiveness of inclusion or breadth of exclusion
criteria (Licht et al 1997). For example, a recent study of
the treatment of acute manic episodes, engaged and
randomly assigned drug-free inpatients with bipolar disorder, manic phase, to divalproex sodium, lithium, or pill
placebo (Bowden et al 1994). This study, although highly
informative about the acute phase efficacy of divalproex,
did not answer the clinically important question of where
divalproex alone or combined with other treatment(s) fits
in the management of patients with bipolar disorder during
the acute phase and especially in subsequent longer term
treatment. This carefully designed initial trial had high
internal validity but limited generalizability.
In addition, the Bowden et al (1994) study would
0006-3223/00/$20.00
PII S0006-3223(99)00898-2

616

A.J. Rush et al

BIOL PSYCHIATRY
2000;48:615– 624

probably not be feasible in today’s managed health care
climate where hospitalization is curtailed— even for patients with mania. In this context, patient safety is of
paramount importance. If patients receive placebo and
have a reasonably good initial response while in the
hospital but are subsequently discharged on placebo only,
their relapse would occur as outpatients—which increases
risk.
Unlike major depressive disorder, most patients with
bipolar disorder will likely need a combination of medications to achieve euthymia (Freeman and Stoll 1998; Post
et al 1998a; Solomon et al 1996). Thus, to determine when
a combination of mood stabilizers, with or without atypical antipsychotic agents, is required—and therefore where
a new therapy fits (not only as a monotherapy but also as

a part of combination therapy)—is a key clinical issue
typically not addressed by traditional efficacy trials in
which monotherapy is evaluated against placebo. The
question of where a treatment fits is important because
only about half of manic patients taking either lithium or
divalproex sodium respond ($50% improvement) after 3
weeks (Bowden et al 1994; Keck and McElroy 1996). That
is, 50% improvement from baseline may still leave the
patient symptomatic. Moreover, as emphasized at the 1992
MacArthur Conference on Longitudinal Methodology,
these designs are not clinician friendly because they do not
answer the question in most need of an answer, namely, if
drug A does not work in a given patient, then what are the
chances that drug B will?
The following discussion attempts to address several of
these current impediments by reconceptualizing approaches to the acute phase efficacy– effectiveness studies
of new treatments for bipolar disorder. The discussion is
divided into three sections: 1) a reevaluation of sample
definition and of study participants, 2) a commentary on
the utility of developing consensus-based outcome measures and a time period for outcome sampling, and finally

3) borrowing substantially from the neurology literature,
suggestions for alternative study designs that, when taken
together, provide a matrix for rapidly developing efficacy
data from representative patients while protecting patient
safety. These suggestions are offered with the hope that a
consensus among investigators evaluating new acute
phase treatments for bipolar disorder can be developed to
more rapidly evaluate treatments of potential benefits to
patients.

Sample Composition
Several differences distinguish bipolar disorders from
major depressive disorders and impact the composition of
patient samples and the nature of outcomes to be mea-

sured. Given issues of patient safety, these differences
may well recommend designs other than the traditional
randomized, double-blind, acute phase, placebo-controlled
study of moderately ill outpatients. Among the differences
between the disorders are the presence of concurrent

conditions, which are common in adults with major
depression but are even more frequent in those patients
with bipolar disorder. In addition, two types of comorbidity (lifetime and concurrent) occur in those with bipolar
disorder. They have a 10 to 20 times higher risk of having
three or more lifetime comorbid conditions than any other
psychiatric illness (Keck et al 1997; Kessler et al 1994;
McElroy et al, in press; Strakowski et al 1998). Given this
high rate of concurrent comorbidity, both generalizability
and feasibility are reduced, and these patients would be
excluded if traditional restrictive criteria were applied.
Patients with major depression who are psychotic,
actively suicidal, and judged to be at risk are typically
excluded from placebo-controlled efficacy trials; however,
patients with bipolar disorder often become suicidal more
rapidly and more unpredictably, especially given the high
prevalence of psychotic symptoms in this population.

Outcome Measures
Turning to the issue of meaningful, agreed-on outcome
measures, efficacy studies in major depression generally

use itemized depressive symptom severity rating scales
completed by clinicians, such as the Hamilton Rating
Scale for Depression (HRS-D; Hamilton 1960, 1967), the
Montgomery–Åsberg Depression Rating Scale (MADRS;
Montgomery and Åsberg 1979), or the Inventory of
Depressive Symptomatology—Clinician Rated (IDS-C;
Rush et al 1986, 1996) as primary outcome measures. For
confirmation or elaboration of treatment effects, symptom
self-reports such as the Beck Depression Inventory (BDI;
Beck et al 1961, 1979), the Zung Depression Rating Scale
(ZDRS; Zung 1965), or the IDS-SR (Rush et al 1986,
1996) are often employed.
For bipolar disorder, the clinical instruments used to
assess symptom severity have been borrowed from or
modeled after those used for major depression. For example, the depressed phase of the illness is often assessed by
a depressive symptom clinical rating (e.g., the HRS-D),
whereas the manic phase of illness is often assessed by a
similarly designed rating instrument, such as the Young
Mania Rating Scale (YMRS; Young et al 1978).
A number of difficulties are encountered when using

these simple symptom checklists in bipolar disorder,
however. Reliance on patients’ reports of their symptoms
and assessments for brief time periods (e.g., 7 days before
the rating) carry key limitations, including the following:
1) the different illness phases (depressed, hypomanic,

Developing New Treatments for Bipolar Illness

manic, and mixed) may all occur between measurement
occasions; 2) the co-occurrence of manic and depressive
symptoms may not be fully captured by symptom assessments; 3) the different illness phases often are associated
with psychosis, which can impair ability to identify
symptoms; 4) moreover, many patients, when manic or
hypomanic, underestimate pathology, rendering self-reports insufficient or misleading; and 5) finally, in rapidcycling patients, even monthly cross-sectional symptomatic assessments may miss many periods of severe
depression or mania that can be captured by other, more
frequent, prospective daily assessments (Denicoff et al
1997b).
Another outcome measurement issue, given the waxing
and waning course of bipolar illness, is how to define
response, especially because the morbid state has three

different and sometimes even apparently opposite elements: depressive symptoms, manic symptoms, and mood
lability (i.e., the tendency to shift rapidly from one state to
another with an almost infinite variety of accompanying
patterns). Additionally, patients with bipolar disorder may
experience rapid shifts from euthymic or nearly asymptomatic states into more severe depressive or manic states
(even a psychotic condition). Switches in mood states may
happen so rapidly that scheduled appointments cannot
sufficiently capture acute symptom worsening or
improvement.

Rapid Cycling
A related difference between major depressive and bipolar
disorders is the frequent cycling between illness phases
(i.e., from manic to mixed, from manic to depressed, etc.),
often referred to as “rapid cycling” (i.e., four or more
mood episodes per year). Moreover, rapid cycling may be
so rapid (i.e., ultra rapid and ultradian forms; George et al
1998; Kramlinger and Post 1996; Pazzaglia et al 1993)
that at a particular time, the patient may not qualify for
studies in which a minimum duration of an episode (e.g.,

2 weeks) or no clinical change during an observational
period (e.g., placebo run-in or “washout”) are among the
study inclusion criteria.

Concomitant Medications
An additional, important distinction between major depressive and bipolar disorders is the use of concomitant
medications. Although difficult, it is possible to find
samples of depressed patients receiving no specific antidepressant treatment; however, recruitment of medicationfree patients with bipolar disorder is extremely difficult.
A central problem is what to do with patients receiving
concomitant medications. Is discontinuation of these

BIOL PSYCHIATRY
2000;48:615– 624

617

agents during a washout period appropriate? What ethical
and safety issues are involved in such a procedure? What
if the medication is a mood stabilizer, yet the overall
response has been incomplete? Is the patient excluded? If
patients undergo a washout, many will experience a
worsening of symptoms or develop an episode, thus
becoming an acute clinical and study management problem (i.e., requiring extra care for safety reasons or becoming study ineligible). If, on the other hand, patients
requiring concomitant agents are excluded, both generalizability and feasibility are increasingly compromised.
It has been suggested that a partial response to a
particular medication should not lead to its discontinuation, but rather (as is also common clinical practice) to
continue the treatment and add a second agent in an
attempt to convert a partial response to a complete
response (i.e., augmentation; for reviews, see Post et al
1997a; Woggon 1997). Is there any way to engage such
patients to assist in establishing efficacy and safety,
informing us as to where the new agent might fit in the
treatment plan?

Sample Issues
Concurrent comorbidity, variations in the course of illness,
and different variants (e.g., bipolar I, bipolar II, and
bipolar disorder, not otherwise specified [NOS]) of the
disorder, as well as the high probability that most otherwise study-eligible patients are already taking one or more
“therapeutic” agents, all create substantial impediments to
the feasibility of acquiring a “clean” sample of patients.
Further, the tendency to shift illness phases (depressed,
manic, or mixed) with or without treatment, complicates
both study design and sample definition.
Consensus opinions (Prien and Rush 1996) recommend
broader inclusion and fewer exclusion criteria for patients
with bipolar disorder who enroll in trials of safety,
tolerability, and efficacy of new mood stabilizing, antimanic, or antidepressant medications. The reasons for this
recommendation include the following: 1) generalizability
of findings, whether negative or positive, is nearly immediate because populations under study are basically identical to those in routine practice; 2) by including patients
with greater comorbidity, greater mood lability, or both,
placebo response rates are likely reduced, so that sample
sizes needed to identify an effective treatment should be
smaller than with less complex patient populations; and 3)
feasibility is increased, and cost and time to develop useful
clinical information are reduced.
On the other hand, the inclusion of more complex populations in clinical trials potentially threatens internal validity.
Randomization may fail in that a disproportionate representation of one as opposed to another type of condition enters

618

BIOL PSYCHIATRY
2000;48:615– 624

one treatment cell. For example, patients with a rapid cycling
course—who may be less responsive to drug A, but more
responsive to drug B—may, by chance, be assigned disproportionately to the cell containing drug A. This may in turn
render drug A less effective than it normally would be. The
standard solution to this problem is to stratify based on
parameters that are well known to affect outcome. Unfortunately, with bipolar disorder, parameters that actually affect
outcome are not yet completely known; however, concurrent
substance abuse and rapid cycling have been suggested by
numerous studies (Aagaard and Vestergaard 1990; Calabrese
and Woyshville 1995a; Denicoff et al 1997a; Himmelhoch
and Garfinkel 1986; Kusalic and Engelsmann 1998; Post et al
1996a; Swann et al 1999) to affect the acute phase response
to lithium. Thus, stratification for each of these parameters, if present at baseline, can allow inclusion of patients
with rapid cycling or with active substance abuse. Nonetheless, one must limit the number of stratifying variables
as not to substantially reduce feasibility while retaining
generalizability.
Another limitation of restricted samples that are less
representative is that such patients may preferentially
benefit from the new treatment because their illness may
be more responsive to treatment overall. Specifically, this
strategy leads to a risk that an effective treatment will be
found for only a minority (10 –25%) of patients and that
this treatment will not meaningfully add to what is already
available for more representative patients (i.e., a drug with
only modest “real-world” utility is inappropriately recommended because of inadequate study design; Licht et al
1997).
Thus, the dilemma encountered in sample selection is
whether to study an unrepresentative and difficult-toobtain sample in an attempt to avoid the potentially
confounding affects of variables that may not randomize
equally to each treatment cell or whether to “roll the dice,”
allowing patients with more complex but representative
illnesses into the trial, gambling on the likelihood that
randomization will succeed.
To develop new acute phase treatments with real-world
generalizability, inclusiveness is preferred by allowing in
most concurrent conditions, with the exception of active
substance abuse, which almost always requires independent concurrent treatment. Patients with past histories of
substance abuse not currently abusing are, of course, at
risk for recidivism, even in the course of a new treatment
trial. Using this approach, those patients randomized to
different treatment cells are assumed to be equally likely
to have a recurrence of their substance abuse.
In addition, because the treatment implications (i.e., all
of these patients are likely to require mood stabilizers)
have not been established for the division between types of
bipolar disorder (bipolar I vs. bipolar II), including both

A.J. Rush et al

diagnostic groups in efficacy trials can be justified. Because rapid cycling appears to differentially affect outcome, stratification for this single variable can be planned,
or this group can be completely excluded. The advantage
of a more inclusive, representative sample is especially
relevant to maintenance trials designed to determine
whether acute phase responders continue to have sustained
benefit. Those patients with greater comorbidity, greater
course fluctuations, or both are those most likely to display
early relapse or recurrence. Narrowly defined or restricted
populations that respond to acute phase treatment and then
enter a longer term “maintenance” trial are less likely to
have sufficient symptomatology or disability within a
reasonable period of time to identify a clinically significant relapse or recurrence. This may have been a reason
for the failure to distinguish valproate or lithium from
placebo in the prevention of mania (Bowden et al 2000).
To summarize, the need to generalize results from efficacy
trials to clinical practice argue for broader inclusion and
fewer exclusion criteria. The following lists the agreedupon “rules” at the NIMH-sponsored 1994 conference on
bipolar illness:
1. Broaden inclusion criteria to include rapid cycling
and other variants of bipolar disorder.
2. Inclusion of comorbidities.
3. Inclusion of bipolar I, bipolar II, and bipolar NOS
disorders.
4. Develop a widely acceptable standardized outcome
assessment rating package.
5. Choose an acceptable longitudinal rating tool.
6. Broaden the list of acceptable designs (e.g., crossover) that are clinician friendly.
7. Consider conducting randomized trials in the clinical setting in an open fashion.
8. Adopt more liberal criteria for entry into the
randomization phase, so that the most ill patients
are not excluded.
9. Recognize the unique difficulties of studies in
bipolar patients and fund studies that answer some,
if not all, questions raised.
10. Develop new methods to study comparative complex combination therapies.
Anecdotal reports from multiple applicants, even recently,
suggest that NIMH review committees should be familiarized with these NIMH rules, promulgated over 6 years
ago, to include a major emphasis on inclusion criteria, as
well as other recommendations that could lead to reviews
of grant applications for the treatment studies of bipolar
illness that are consistent with these guidelines.

Developing New Treatments for Bipolar Illness

BIOL PSYCHIATRY
2000;48:615– 624

619

Assessment of Outcome

Alternative Study Designs

It appears to be insufficient to borrow methodologies from
efficacy studies in major depressive disorder to define the
outcome of patients with bipolar disorder because of the
diversity of symptomatology in the latter illness and the
presence of mood instability (especially rapid cycling).
For example, in major depressive disorder, the standard
frequency of measurement (e.g., every week or two) and
the typical time frame assessed (e.g., the past 7 days) may
be inadequate or even misleading for patients with bipolar
disorder. More frequent assessments or observations that
cover more extended time periods, provide a more thorough method by which to evaluate treatment benefits.
Therefore, alternative strategies to measure outcome in
efficacy trials of patients with bipolar disorder are
recommended.
Another impediment is a lack of a consensus regarding
the appropriate outcome measures for bipolar disorder.
Although there are a number of itemized symptom rating
scales available (American Psychiatric Association Task
Force for the Handbook of Psychiatric Measures 2000),
they do not capture the qualitative and quantitative
changes that rapidly occur or the level of disability
associated with bipolar disorder. For example, when patients present in a mixed episode, is it necessary to
complete both mania and depression rating scales? Should
this also occur when patients are in a single phase of the
illness? Is severity the sum of both ratings? Additionally,
how can ratings of both mania and depression be
compared?
Recent innovations using a life charting method (LCM;
Denicoff et al 1997b; Leverich and Post 1996, 1998), and
the recent redesign of the Clinical Global Impression
(CGI; Guy 1976) for bipolar disorder (CGI-BP; Spearing
et al 1997) provide two methods by which to gauge the
type (manic, hypomanic, mixed, or depressed), severity,
and associated functional disability in bipolar disorder
over time periods that range from 1 day to weeks, or even
months to years. These methods also can evaluate acute
phase response, as well as prophylactic efficacy. The LCM
summarizes the severity and functional effect of the illness
without diagnostic bias as to what constitutes an “episode”
of mania or depression.
The LCM captures mood state fluctuations and the
degree of symptom-driven dysfunction either retrospectively (e.g., on a monthly basis) or prospectively (e.g., on
a daily basis). The prospective LCM has been found to be
a relatively reliable and valid method when sampling daily
symptomatology over the previous 2 to 4 weeks (Denicoff
et al 1997b). Specifically, the LCM correlated with the
HRS-D (at 7 days), the YMRS (at 2 days), and the Global
Assessment Scale (Endicott et al 1976, at 1 month).

Another major obstacle encountered when studying
bipolar disorder is treatment trial design. The standard
parallel group, randomized, double-blind, placebo-controlled trial, when used in patients with bipolar disorder
who are often severely ill, can lead to several problems,
including the following: 1) patients or their treating
physicians decline studies because of concerns about
receiving placebo; 2) investigators or institutions limit
eligible patient populations to the mildly to moderately
ill to increase patient safety, and 3) investigators,
patients, and their families worry about untoward outcomes for an increased number of patients as some are
receiving placebo. In the case of bipolar disorder, safety
concerns (for self or others) can present abruptly and,
although uncommon, are not rare.
An analogous set of problems has been encountered in
the research on epilepsy. The neurology literature recommends several alternative study designs for evaluating the
safety, efficacy, and tolerability of new medications, all of
which appear well suited to treatment studies of patients
with bipolar disorder (Pledger and Sahlroot 1993). Specifically, the risk of stopping partially effective mood stabilizing, antimanic, or antipsychotic agents can be as profound as the risk of stopping antiepileptic agents in some
patients, especially given suggestions of “kindling-like”
phenomena apparently present in both disorders. Thus,
“add-on” designs similar to those used in epilepsy research
may have particular utility in acute phase treatment studies
in a large proportion of persons with bipolar disorder.

Add-On Placebo-Controlled Trials
In the treatment of epilepsy, as is case for bipolar disorder,
multiple medications often are used to achieve complete
symptom control, although monotherapy often is initially
attempted. In an add-on placebo-controlled study, patients
who remain symptomatic, while continuing to receive
their current medication(s) at fixed dose(s), are randomized, double-blind, to either placebo or the new treatment
(Handforth et al 1998; Harden et al 1999; Mawer et al
1999). This design has the advantage of being inclusive
with regard to eligible patients, except that patients with
bipolar disorder who respond fully to monotherapy will
not be eligible for this design. This add-on design requires
that some knowledge of drug interactions is available, but
it also allows for the prospective evaluation of drug
interactions in more representative samples.
This add-on design also benefits patient safety, a major
concern, especially with the evaluation of placebo against
new treatments that often have, at the particular time of
study, no demonstrated efficacy. Patients entering an
add-on study are not disadvantaged over their baseline

620

BIOL PSYCHIATRY
2000;48:615– 624

A.J. Rush et al

Figure 1. Modified “add-on” design. *Low-dose
experimental drug may also be used. (Modeled
after Sachdeo et al 1997.)

state because all patients continue to receive the medication(s) that might have been partially helpful. A subset of
them have a potential added advantage (or disadvantage)
of exposure to the new medication.
A variation of this add-on design (Sachdeo et al
1997) incorporates a baseline period to determine symptom severity and consistency and, consequently,
whether patients have sufficient symptomatology to
enter the trial. Once inclusion and exclusion criteria are
met, the experimental medication is given open label for
a short period of time (e.g., 1 week) in addition to the
patient’s prestudy medication(s) at fixed doses. During
the open-label period, a lower than presumably therapeutic dose of the new medication is used to probe for
drug interactions and tolerability (i.e., to identify drug
allergies, drug intolerance, and negative drug interactions). Thus, patients who cannot tolerate modest doses
of the new medication being added to their ongoing
medication(s) are not randomized.
At the end of the open-label period, medications are
masked. What follows is a specified period (e.g., 5
weeks) in which patients are randomized to receive
either a placebo or some experimental medication used
at a low dose during the open-label period, or a higher,
presumably effective, dose of the new medication. As
an alternative, patients may not be assigned to placebo,
but rather continue on the lower dose of the experimental medication originally prescribed. During this randomized phase, prestudy medications are continued in a
stable regimen.
Following the randomization phase, the prestudy medication(s) may be discontinued to compare monotherapy
with the new medication to prestudy medications combined with placebo (or low dose experimental medication;
Figure 1).
Thus, for bipolar disorder, a prospective baseline

evaluation (e.g., 1 to 2 weeks) for patients on an
established regimen is followed by open-label exposure
to the new drug at low doses to determine tolerability,
adverse events, and drug interactions. Thereafter, while
prestudy medications continue, a double-blind randomization to placebo (or to a low dose of the new
medication) or to the new medication at therapeutic
doses follows. This phase evaluates the utility of the
new medication as an adjunct to whatever baseline
treatment(s) patients are receiving. After the randomization phase, a slow discontinuation of the original
regimen could follow, with random assignment while
the new medication is continued at therapeutic doses, to
evaluate efficacy as a monotherapy.
Some have argued that if evidence of efficacy is
established in add-on studies such as those described
above, given that such studies target patients with medication-resistant conditions (i.e., patients already taking one
or more antiepileptic drugs but who continue to have
symptomatology), then separate trials in more amenable
(i.e., less treatment-resistant) patients are unnecessary
because the study medication has already been proven
effective in a more complex, difficult-to-treat population
(Brodie 1996; Brodie and Dichter 1996; Brodie et al
1995). What cannot be deciphered with certainty from this
type of add-on design is whether an effective add-on agent
is an effective monotherapy.
This add-on or “discontinuation” design can also enhance patient safety. A survival analysis can be used in
conjunction with the setting of stringent subject nonresponse or worsening criteria (such as doubling of the
average monthly symptomatology, doubling of the highest
2-day symptomatology, or the development of any psychotic symptoms) to remove subjects. Then, survival
analysis captures the comparative benefits (or lack
thereof) of each masked treatment. Those who exit the

Developing New Treatments for Bipolar Illness

BIOL PSYCHIATRY
2000;48:615– 624

621

Figure 2. A design to define comparative efficacy of two adjunctive medications. *Randomized.

study without meeting those criteria (i.e., administrative or
unexplained withdrawals) provide censored data. They do
not contribute to the survival curve beyond the time they
were in the study. This survival analysis compares the
benefits of the new treatment while protecting patient
safety to an optimal degree. It also allows much easier
initial patient recruitment. It also allows, given randomization, an assessment of the spontaneous waxing and
waning of symptoms and disability over more prolonged
time periods than the traditional 4- to 6-week trials that
require weekly symptom assessments as commonly used
in trials of major depressive disorder.
Alternatively, the add-on of one or more medications in
a randomized trial can be systematically tested for efficacy
in individual patients using an off-on-off-on (B-A-B-A)
design to avoid exposure to placebo altogether for nonresponders (Ballanger and Post 1978; McDermut et al 1995).
The duration of the trial can also be individualized based
on the patient’s presenting baseline cycle frequency,
thereby determining a set threshold for the degree of
acceptable improvement on and off drug (McDermut et al
1995; Post et al 1998b).

uation design noted above, beginning with patients who are
already receiving mood stabilizing agent(s) but who have
sufficient symptomatology to warrant additional pharmacotherapeutic interventions (Figure 2). In this case, the add-on
therapy could be provided initially in a low dose to assay for
tolerability under open label. Then, under double-blind conditions, doses of each of the two candidate monotherapies are
increased during the conversion period. Thereafter, prior
mood stabilizers are discontinued. Thus, one would have a
comparison of two agents as adjunctive therapies and then as
monotherapies in these patients.
Alternatively, one could simply leave patients on their
pretreatment medications and add (double-blind) either an
additional standard medication (A), or the experimental
medication to the mix or one of two new medications (X and
Y) with a crossover to assess individual responsivity. This
design studies the comparative value of agent X versus agent
Y as adjuncts to treatment in those already receiving one or
more insufficiently beneficial mood stabilizers.
Following these observations to test for comparability
of efficacy between the two candidate adjunctive agents,

Defining Where a New Treatment Fits in
Clinical Practice
Brodie (1996) highlighted the need to conduct studies (during
or following regulatory approval) to determine where new
medications fit in our overall therapeutic armamentarium.
Because new medications cost substantially more than generic agents, it is important to determine where these newer
medications best fit in terms of value for the health care
dollar, as well as, whenever feasible, to identify patients who
preferentially respond to either monotherapy or combination
treatment with the new agent.
This issue can be addressed with the add-on or discontin-

Figure 3. A design to define where one or more medications
“fit” in a treatment plan.

622

A.J. Rush et al

BIOL PSYCHIATRY
2000;48:615– 624

baseline medication(s) could be discontinued to assay for
comparative efficacy of the two agents as monotherapies.
The individual added benefit from the combination compared to monotherapy could be further evaluated in the
off-on-off-on design (McDermut et al 1995).
Another way to address the issue of where a medication
fits in the treatment plan is to design medication treatment
algorithms beginning with monotherapy, but adding new
medications in a systematic way to a consistent prestudy
medication regimen (Figure 3). The key decision points at
which each new medication is added are cued by an evaluation of current symptomatology (manic, mixed, depressed,
or mood reactivity). Such algorithms have been proposed
(Bauer et al 1999; Calabrese and Woyshville 1995b; Dennehy and Suppes 1999; Nolen et al 1998; Post et al 1996b;
Sachs 1996; Suppes et al 1998). To our knowledge, no
comparative trials of one algorithm in contrast to another
have been undertaken in bipolar disorder; however, the
simultaneous use of multiple mood stabilizers is becoming
common practice (Freeman and Stoll 1998; Solomon et al
1996), in part because of increasing evidence that full
symptomatic remission, rather than simple symptom reduction, is associated with less disability and a better prognosis
(Fava et al 1998a, 1998b; Judd et al 1997, 1998; Miller et al
1998; Paykel et al 1999).
Finally, there are a plethora of practical issues, not the least
of which are the timely recruitment of large numbers of
representative patients with bipolar disorder, reluctance of
practitioners to “give up” patients and associated income for
research studies, and the overall reduction in length of
hospital stays. One solution to these and other practical
complexities is to engage practitioners to conduct the research and properly reimburse them, similar to what is
currently being planned in the NIMH-sponsored Sequenced
Treatment Enhancement Program-Bipolar Disorder (University of Pittsburgh, Epidemiology Data Center 2000b) and
Sequenced Treatment Alternatives to Relieve Depression
(University of Pittsburgh, Epidemiology Data Center 2000a)
trials. Similarly, subjects in the protocol can be followed in
and out of different care settings (inpatient, day hospital,
intensive outpatient, and outpatient clinic), which in turn is
more representative of actual practice. Nonetheless, post hoc
statistical adjustments may be needed in such situations
because other nonprotocol treatments are often provided to
differing degrees in diverse care settings, and adherence to
protocol procedures may also depend on the intensity and
frequency of staff–patient interactions that vary across these
settings.

Other Issues
This article has focused on studies of acute phase treatment;
however, additional major design, feasibility, patient sample,

ethical, outcome measurement and other issues pertain to
longer term maintenance trials. This topic deserves a detailed
discussion beyond the scope of this article. Designs and
methods developed in longer term chronic disease management (often effectiveness) trials (e.g., in cancer or heart
disease) may be especially useful when approaching these
issues. We plan a subsequent paper on the problems encountered in long-term studies of bipolar disorder.

Conclusions
We have reviewed a number of impediments to developing
and evaluating new treatments for bipolar disorder, and we
have suggested some potential solutions to these problems.
Some of these solutions may also be applicable to studies of
patients with other severe and persistent mental illnesses.
First, broadening the inclusion and narrowing the exclusion
criteria in clinical trials involving patients with bipolar
disorder both increases feasibility and decreases cost, while
protecting sufficient internal validity and increasing external
validity compared with more narrowly defined populations.
Second, using detailed, continuous, longitudinal outcome assessments conducted over longer periods of time
and using broader outcome domains (e.g., both symptoms
and functional status) avoids the difficulty of arbitrarily
brief time periods for outcome evaluation. The LCM is
one readily available method of assessment that can be
used until a consensus is reached regarding standard
outcome measures to be recommended for all efficacy
trials involving patients with bipolar disorder.
Third, by adopting some of the study designs from
efficacy trials of patients with epilepsy, it is possible to
study treatment-resistant (e.g., partial or nonresponders to
standard pharmacotherapy) patients who are in need of
additional medication by using an add-on protocol. Such
designs retain an optimal degree of patient safety while
determining whether the new medication offers additional
therapeutic value beyond current monotherapies or treatment combinations. If these add-on studies establish both
safety and efficacy, they can then be followed by a
subsequent discontinuation of prestudy medications to test
the new agent as a monotherapy. Taken together, these
approaches address the question of whether new medications are effective adjuncts and whether they are effective
monotherapies. If both questions are answered in the
affirmative, then acute phase efficacy studies with a
placebo comparison in treatment-resistant, medicationfree patients may not be needed. Finally, the diligent
introduction of medications, their discontinuation, or both
can help to identify the clinical place of new medications
(either as monotherapy or in combination) in the overall
management of patients with bipolar disorder.

Developing New Treatments for Bipolar Illness

Preparation of this report was supported in part by the Theodore and
Vada Stanley Foundation, National Institute of Mental Health (Grant No.
MH-53899), the Betty Jo Hay Distinguished Chair in Mental Health, the
Rosewood Corporation Chair in Biomedical Science, and the Sarah M.
and Charles E. Seay Center for Basic and Applied Research in Psychiatry. The authors appreciate the secretarial support of David Savage and
Fast Word, Inc., Dallas. We are also grateful to Kenneth Z. Altshuler,
M.D., Stanton Sharp Chair and Professor, Department of Psychiatry, UT
Southwestern Medical Center, for his administrative support.
Aspects of this work were presented at the conference “Bipolar
Disorder: From Preclinical to Clinical, Facing the New Millennium,”
January 19 –21, 2000, Scottsdale, Arizona. The conference was sponsored by the Society of Biological Psychiatry through an unrestricted
educational grant provided by Eli Lilly and Company.

References
Aagaard J, Vestergaard P (1990): Predictors of outcome in
prophylactic lithium treatment. A two-year prospective study.
J Affect Disord 18:259 –266.
American Psychiatric Association Task Force for the Handbook of
Psychiatric Measures (2000): Handbook of Psychiatric Measures. Washington, DC: American Psychiatric Association.
Ballenger JC, Post RM (1978): Therapeutic effects of carbamazepine in affective illness: Preliminary report. Commun
Psychopharmacol 2:159 –175.
Bauer MS, Callahan AM, Jampala C, Petty F, Sajatovic M,
Schaefer V, et al (1999): Clinical practice guidelines for
bipolar disorder from the Department of Veterans Affairs.
J Clin Psychiatry 60:9 –21.
Beck AT, Rush AJ, Shaw BF, Emery G (1979): Cognitive
Therapy of Depression. New York: Guilford.
Beck AT, Ward CH, Mendelson M, Mock JE, Erbaugh JK
(1961): An inventory for measuring depression. Arch Gen
Psychiatry 4:561–571.
Bowden CL, Brugger AM, Swann AC, Calabrese JR, Janicek PG,
Petty F, et al (1994): Efficacy of divalproax sodium vs. lithium
and placebo in the treatment of mania. JAMA 271:918 –924.
Bowden CL, Calabrese JR, McElroy SL, Gyulai L, Wassef A, Petty
F, et al (2000): A randomized, placebo-controlled 12-month trial
of divalproex and lithium treatment of outpatients with bipolar I
disorder. Arch Gen Psychiatry 57:481– 489.
Brodie MJ (1996): Antiepileptic drugs, clinical trials, and the
marketplace. Lancet 347:777–779.
Brodie MJ, Dichter MA (1996): Antiepileptic drugs. N Engl
J Med 334:168 –175.
Brodie MJ, Richens A, Yuen AW, UK Lamotrigine/Carbamazepine Monotherapy Trial Group (1995): Double-blind comparison of lamotrigine and carbamazepine in newly diagnosed
epilepsy. Lancet 345:476 – 479.
Calabrese JR, Woyshville MJ (1995a): Lithium therapy: Limitations and alternatives in the treatment of bipolar disorders.
Ann Clin Psychiatry 7:103–112.
Calabrese JR, Woyshville MJ (1995b): A medication algorithm
for treatment of bipolar rapid cycling? J Clin Psychiatry
56(suppl 3):11–18.
Denicoff KC, Smith-Jackson EE, Disney ER, Ali SO, Leverich
GS, Post RM (1997a): Comparative prophylactic efficacy of

BIOL PSYCHIATRY
2000;48:615– 624

623

lithium, carbamazepine, and the combination in bipolar disorder. J Clin Psychiatry 58:470 – 478.
Denicoff KD, Smith-Jackson EE, Disney ER, Suddath RL,
Leverich GS, Post RM (1997b): Preliminary evidence of the
reliability and validity of the prospective life-chart methodology (LCM-p). J Psychiatr Res 31:593– 603.
Dennehy E, Suppes T (1999): Medication algorithms for bipolar
disorder. J Pract Psychiatry Behav Health 5:142–152.
Endicott J, Spitzer RL, Fleiss JL, Cohen J (1976): The global
assessment scale. A procedure for measuring overall severity
of psychiatric disturbance. Arch Gen Psychiatry 33:766 –771.
Fava GA, Rafanelli C, Grandi S, Canestrari R, Morphy MA (1998a):
Six-year outcome for cognitive behavioral treatment of residual
symptoms in major depression. Am J Psychiatry 155:1443–1445.
Fava GA, Rafanelli C, Grandi S, Conti S, Belluardo P (1998b):
Prevention of recurrent depression with cognitive behavioral therapy: preliminary findings. Arch Gen Psychiatry 55:816–820.
Freeman MP, Stoll AL (1998): Mood stabilizer combinations: A
review of safety and efficacy. Am J Psychiatry 155:12–21.
George MS, Huggins T, McDermut W, Parekh PI, Rubinow D,
Post RM (1998): Abnormal facial emotion recognition in
depression: Serial testing in an ultra-rapid-cycling patient.
Behav Modif 22:192–204.
Guy W (1976): ECDEU Assessment Manual for Psychopharmacology,
U.S. Department of Health, Education and Welfare Publication No.
76-338. Washington DC: U.S. Government Printing Office.
Hamilton M (1960): A rating scale for depression. J Neurol
Neurosurg Psychiatry 23:56 – 62.
Hamilton M (1967): Development of a rating scale for primary
depressive illness. Br J Soc Clin Psychol 6:278 –296.
Handforth A, DeGiorgio CM, Schachter SC, Uthman BM,
Naritoku DK, Tecoma ES, et al (1998): Vagus nerve stimulation therapy for partial-onset seizures: A randomized activecontrol trial. Neurology 51:48 –55.
Harden CL, Lazar LM, Pick LH, Nikolov B, Goldstein MA, Carson
D, et al (1999): A beneficial effect on mood in partial epilepsy
patients treated with gabapentin. Epilepsia 40:1129 –1134.
Himmelhoch JM, Garfinkel ME (1986): Sources of lithium resistance in mixed mania. Psychopharmacol Bull 22:613– 620.
Judd LL, Akiskal HS, Maser JD, Zeller PJ, Endicott J, Coryell
W, et al (1998): A prospective 12-year study of subsyndromal
and syndromal depressive symptoms in unipolar major depressive disorders. Arch Gen Psychiatry 55:694 –700.
Judd LL, Akiskal HS, Paulus MP (1997): The role and clinical
significance of subsyndromal depressive symptoms (SSD) in
unipolar major depressive disorder. J Affect Disord 45:5–17.
Keck PE Jr, McElroy SL (1996): Outcome in the pharmacologic
treatment of bipolar disorder. J Clin Psychopharmacol
16(suppl 1):15S–23S.
Keck PE Jr, McElroy SL, Strakowski SM, Bourne ML, West SA
(1997): Compliance with maintenance treatment in bipolar
disorder. Psychopharmacol Bull 33:87–91.
Kessler RC, McGonagle KA, Zhao S, Nelson CB, Hughes M,
Eschleman S, et al (1994): Lifetime and 12-month prevalence
of DSM-III-R psychiatric disorders in the United States. Arch
Gen Psychiatry 51:8 –19.
Kramlinger KG, Post RM (1996): Ultra-rapid and ultradian cycling
in bipolar affective illness. Br J Psychiatry 168:314 –323.

624

BIOL PSYCHIATRY
2000;48:615– 624

Kusalic M, Engelsmann F (1998): Predictors of lithium treatment
responsiveness in bipolar disorder. A two-year prospective
study. Neuropsychobiology 37:146 –149.
Leverich GS, Post RM (1996): Life charting the course of bipolar
disorder. Curr Rev Mood Disord 1:48 – 61.
Leverich GS, Post RM (1998): Life charting of affective disorders. CNS Spectrums 3:21–37.
Licht RW, Gouliaev G, Vestergaard P, Frydenbert M (1997):
Generalisability of results from randomised drug trials. A trial
on antimanic treatment. Br J Psychiatry 170:264 –267.
Mawer GE, Jamieson V, Lucas SB, Wild JM (1999): Adjustment of
carbamazepine dose to offset the effects of the interaction with
remacemide hydrochloride in a double-blind, multicentre, add-on
drug trial (CR2237) in refractory epilepsy. Epilepsia 40:190–196.
McDermut W, Pazzaglia PJ, Huggins T, Mikalauskas K, Leverich GS, Ketter TA, et al (1995): Use of single case analyses
in off-on-off-on trials in affective illness: A demonstration of
the efficacy of nimodipine. Depression 2:259 –271.
McElroy SL, Altshuler L, Suppes T, Keck PE Jr, Frye MA,
Denicoff KD (in press): Axis I psychiatric comorbidity in a
cohort of patients with bipolar disorder. Am J Psychiatry.
Miller IW, Keitner GI, Schatzberg AF, Klein DN, Thase ME,
Rush AJ, et al (1998): The treatment of chronic depression,
part 3: Psychosocial functioning before and after treatment
with sertraline or imipramine. J Clin Psychiatry 59:608 – 619.
Montgomery SA, Åsberg M (1979): A new depression scale designed
to be sensitive to change. Br J Psychiatry 134:382–389.
Nolen WA, Knoppert-van der Klein EAM, Bouvy PF, Honig A,
Klompenhouwer J-L, Ravelli DP (1998): Richtlijn Farmacotherapie Bipolaire Stoornissen. Amsterdam: Boom.
Paykel ES, Scott J, Teasdale JD, Johnson AL, Garland A, Moore
R, et al (1999): Prevention of relapse in residual depression
by cognitive therapy. Arch Gen Psychiatry 56:829 – 835.
Pazzaglia PJ, Post RM, Ketter TA, George MS, Marangell LB
(1993): Preliminary controlled trial of nimodipine in ultra-rapid
cycling affective dysregulation. Psychiatry Res 49:257–272.
Pledger GW, Sahlroot JT (1993): Alternative analyses for antiepileptic drug trials. In: French JA, Dichter MA, Leppik IE,
editors. New Antiepileptic Drug Development: Preclinical
and Clinical Aspects. Epilepsy Research, Supplement 10.
Amsterdam: Elsevier Science, 167–174.
Post RM, Denicoff KD, Frye MA, Leverich GS (1997a): Algorithms for bipolar mania. In: Rush AJ, editor. Mood Disorders. Systematic Medication Management. Modern Problems
of Pharmacopsychiatry, Vol 25. Basel: Karger, 114 –145.
Post RM, Frye MA, Denicoff KD, Leverich GS, Kimbrell TA,
Dunn RT (1998a): Beyond lithium in the treatment of bipolar
illness. Neuropsychopharmacology 19:206 –219.
Post RM, Ketter TA, Denicoff K, Pazzaglia PJ, Leverich GS,
Marangell LB, et al (1996a): The place of anticonvulsant therapy
in bipolar illness. Psychopharmacology (Berl) 128:115–129.
Post RM, Ketter TA, Pazzaglia PJ, Denicoff K, George MS,
Callahan A, et al (1996b): Rational polypharmacy in the
bipolar affective disorders. Epilepsy Res (suppl 11):153–180.
Post RM, Leverich GS, Denicoff KD, Frye MA, Kimbrell TA,

A.J. Rush et al

Dunn R (1997b): Alternative approaches to refractory depression in bipolar illness. Depression Anxiety 5:175–189.
Post RM, L’Herrou T, Luckenbaugh DA, Frye MA, Leverich GS,
Mikalauskas K (1998b): Statistical approaches to trial durations
in episodic affective illness. Psychiatry Res 78:71– 87.
Prien RF, Potter WZ (1990): NIMH Workshop report on treatment of bipolar disorder. Psychopharmacol Bull 26:409 – 427
Prien RF, Rush AJ (1996): National Institute of Mental Health
Workshop report on the treatment of