Article

Columbia Classification Algorithm of Suicide Assessment

(C-CASA): Classification of Suicidal Events in the FDA’s
Pediatric Suicidal Risk Analysis of Antidepressants

Kelly Posner, Ph.D. Objective: To evaluate the link between suicidal by pharmaceutical companies
antidepressants and suicidal behavior were eliminated, which resulted in a total
and ideation (suicidality) in youth, ad- of 38 discrepant ratings. For the specific
Maria A. Oquendo, M.D.
verse events from pediatric clinical trials label of “suicide attempt,” a relatively low
were classified in order to identify suicidal level of agreement was observed be-
Madelyn Gould, Ph.D., M.P.H. events. The authors describe the Colum- tween the C-CASA and pharmaceutical
bia Classification Algorithm for Suicide As- company ratings, with the C-CASA report-
Barbara Stanley, Ph.D. sessment (C-CASA), a standardized suicidal ing a 50% reduction in ratings. Thus, al-
rating system that provided data for the though the C-CASA resulted in the identifi-
Mark Davies, M.P.H. pediatric suicidal risk analysis of antide- cation of more suicidal events overall,
pressants conducted by the Food and fewer events were classified as suicide at-
Drug Administration (FDA). tempts. Additionally, the C-CASA ratings
were highly reliable (intraclass correlation
Method: Adverse events (N=427) from
coefficient [ICC]=0.89).
25 pediatric antidepressant clinical trials
were systematically identified by pharma-
Conclusions: Utilizing a methodical, an-
ceutical companies. Randomly assigned chored approach to categorizing suicidal-
adverse events were evaluated by three of
ity provides an accurate and comprehen-
nine independent expert suicidologists sive identification of suicidal events. The
using the Columbia classification algo-
FDA’s audit of the C-CASA demonstrated
rithm. Reliability of the C-CASA ratings excellent transportability of this ap-
and agreement with pharmaceutical
proach. The Columbia algorithm was
company classification were estimated. used to classify suicidal adverse events in
Results: Twenty-six new, possibly suicidal the recent FDA adult antidepressant
events (behavior and ideation) that were safety analyses and has also been man-
not originally identified by pharmaceuti- dated to be applied to all anticonvulsant
cal companies were identified in the C- trials and other centrally acting agents
CASA, and 12 events originally labeled as and nonpsychotropic drugs.

(Am J Psychiatry 2007; 164:1035–1043)

A ntidepressant use by children and adolescents dra-

matically increased in recent decades (1, 2), with up to 8
To evaluate the potential association between suicidal-
ity and antidepressants, the Food and Drug Administra-
million prescriptions written annually in the United tion (FDA) decided to undertake a meta-analysis to exam-
States (3). However, the use of antidepressant drug treat- ine suicidal events from 24 randomized placebo-
ment has been fraught with controversy because of ques- controlled pediatric trials of selective serotonin reuptake
tions regarding both efficacy and safety. Efficacy results inhibitors (SSRIs) and other newer generation antidepres-
from pediatric trials are mixed and difficult to interpret, sants. However, inconsistent labeling of potentially sui-
largely because of methodological limitations and regula- cidal events was identified as a significant threat to accu-
tory idiosyncrasies in determining what is an “effective” rate risk-assessment analyses. This concern first arose
study (4–6). Furthermore, regulatory agencies in the during an FDA review of one pediatric SSRI study, in which
United States and the United Kingdom raised concerns in events suggestive of suicidality were labeled “emotional
2003 about the emergence of suicidal thoughts or behav- lability.” Subsequent examination of suicidality data from
iors during antidepressant treatment in pediatric popula- the other eight pediatric antidepressant studies under-
tions, which may have led to a recent decline in prescrip- scored the problem, with a notable example being a sub-
tion rates (7, 8), rendering risk-benefit analyses even ject who slapped herself in the face and was deemed as
more challenging. having made a suicide attempt (Table 1). The FDA deter-

This article is featured in this month’s AJP Audio, is the subject of a CME course, and is discussed in an editorial by Dr. Brent on p. 989.

Am J Psychiatry 164:7, July 2007 ajp.psychiatryonline.org 1035

C-CASA CLASSIFICATION OF SUICIDAL EVENTS

TABLE 1. Examples of Difficulties in Adverse Event Labelinga

Original Label Original Investigator Text From Adverse Event Report
Personality disorder [A] 10-year-old male exhibited symptoms of personality disorder of moderate severity and was dis-
continued. One day later, [the patient] attempted to hang himself with a rope after [a] dispute
with his father. [The] investigator did not consider this a serious adverse event but rather part of
the personality disorder.
Accidental overdose and neurosis The overdose of six capsules of study medication was in fact intentional and in response to an ar-
gument with the subject’s mother.
Medication error Age 14: The patient took 11 tablets impulsively and then went to school...the patient denied that
it was a suicide attempt.
Suicide attempt [The patient] had thoughts of killing self but had no intention of acting on them.
Hostility Age 10: Before his mother’s call to the site and again after arguing with his stepfather, he wrapped
a cord from the miniblinds around his neck, threatening to kill himself.
Emotional lability/suicide attempt Age 14: The patient is reported to have engaged in an episode of “automutilation,” where she
slapped herself in the face.
a These labels were given by the study clinicians in the pharmaceutical company trials. They were given prior to the implementation of C-CASA
and reflect why reclassification was necessary. Some labels are more severe than they should be, and other labels are less severe than war-
ranted.

mined that conclusions based on these data would be un- the links between manifestations of suicidal process de-
reliable and might produce either a false signal that would spite well-documented differences between them (28).
result in unwarranted restriction of useful medications or In the present article, we describe the structure and reli-
an underestimation of risk and subsequent danger to the ability of the Columbia Classification Algorithm of Suicide
general public. Assessment (C-CASA), the classification system of suicidal
The problem of inconsistent nomenclature of suicidal adverse events that produced the data used by the FDA in
ideation and behavior (suicidality) encountered in this their critical assessment of pharmacologic risk.
data set is not unique. Indeed, the ongoing debate con-
cerning nomenclature has perpetuated the use of multiple
Method
terms to refer to the same behavior, frequently with pejo-
rative connotations (e.g., threat, gesture) and descriptors C-CASA
(e.g., “manipulative,” “hostile,” “nonserious”) (9–12). Such The C-CASA is a classification system that utilizes definitions of
variability in terminology has consequences that extend suicidality derived from empirical findings on the phenomenol-
beyond imprecise communication, limiting comparison ogy of suicidality and identified predictive and risk factors. The
of epidemiological prevalence rates and hampering pre- criteria for a suicide attempt include both self-injurious behavior
vention efforts (13). Additionally, it undermines the valid- and suicidal intent (at least some intention to commit suicide).
Intent to die portends a risk for future suicide and repeated at-
ity of risk-benefit analyses. tempts (15, 18, 29, 30) and can be reliably obtained (27). Inclusion
To enhance interpretability of pediatric antidepressant of intent in the definition of suicide allows a distinction between
trial data to be used in their risk analysis, the FDA com- those who self-injure in an attempt to die and those who self-in-
missioned a study by Columbia University/New York jure for purely other nonsuicidal reasons (e.g., to manage affect)
(31). The C-CASA has eight categories that distinguish suicidal
State Psychiatric Institute investigators to classify all
events from nonsuicidal events and indeterminate or potentially
events that could represent suicidality. The investigators suicidal events (Table 2). C-CASA definitions and training exam-
developed a systematic approach to the categorization of ples are presented in Table 2. Figure 1 illustrates the boundaries
potential suicidal adverse events covering the full spec- between categories.
trum of suicidality, rooted in consensus recommenda-
C-CASA Rating Guidelines
tions and empirical findings regarding suicide-related
definitions (10, 12, 14–16). The C-CASA includes operationalized guidelines for inference
of suicidal intent. “Clinically impressive” behavior or circum-
The whole continuum of suicidality was included in the stances are used to infer suicidal intent when the stated intent is
system, given evidence that manifestations along the spec- missing, unclear, or denied. For example, a highly lethal act that is
trum are linked (17, 18). For example, evidence suggests clearly not an accident might mean that no other intent except
that suicide attempts with intent to die are predictive of suicide can be inferred (e.g., a gunshot to the head, jumping from
a high-story building). An illustrative example was a case of self-
completed suicide (16, 18, 19), and individuals who engage
immolation, which was a circumstance allowing inference of in-
in preparatory suicidal behaviors with intent to die are also tent to classify the event a suicide attempt. Alternatively, infer-
at risk for future suicide attempts (20) and completion (21). ence of suicidal intent could also be based on two other pieces of
Epidemiological and clinical studies of adolescents and data, including clinical circumstances such as the method used,
adults have established that severe or pervasive suicidal number of pills ingested, and location of injury on the body. For
ideation is a predictor of both future attempts (17, 22–25) example, cuts on the legs typically represent nonsuicidal self-in-
jurious behavior. According to C-CASA guidelines, other relevant
and completed suicide (26). Moreover, Brown et al. identi- data that could be used included past history of suicide attempt,
fied passive thoughts about wanting to be dead as a risk past history of self-injurious behavior/self-mutilation, and family
factor for completed suicide (27). These studies provide history of suicide/suicide attempts.

1036 ajp.psychiatryonline.org Am J Psychiatry 164:7, July 2007

POSNER, OQUENDO, GOULD, ET AL.

TABLE 2. C-CASA Definitions and Training Examples

Classification/
Category Definition Training Examples
Suicidal events
Completed A self-injurious behavior that resulted in fatality and 1) After a long argument with his girlfriend, which resulted in the
suicide was associated with at least some intent to die as a end of their relationship, the patient collected a rope and rode his
result of the act. bike to an isolated area where he fatally hanged himself. A suicide
note was later found. 2) After four documented attempts at sui-
cide, the patient stole his uncle’s gun and shot himself and was fa-
tally injured.
Suicide attempt A potentially self-injurious behavior, associated with 1) After a fight with her friends at school, in which they discontin-
at least some intent to die, as a result of the act. ued speaking with her, the patient ingested approximately 16 as-
Evidence that the individual intended to kill him/ pirin and eight other pills of different types on the school
herself, at least to some degree, can be explicit or grounds. She said that she deserved to die, which was why she
inferred from the behavior or circumstance. A sui- swallowed the pills. 2) The patient used a razor blade to lacerate
cide attempt may or may not result in actual in- his wrists, his antecubital fossae, and his back bilaterally. He told
jury. his therapist that the “the main objective was to stop feeling like
that,” and he knew that he could die but didn’t care. According to
the patient, he also ingested a bottle of rubbing alcohol because
in his health class he heard “that the medulla will get more sup-
pressed that way,” thereby increasing the chances that he would
be “successful” and die.
Preparatory acts The individual takes steps to injure him- or herself, 1) The patient had run away from home overnight because his fa-
toward but is stopped by self or others from starting the ther had gone to school and retrieved a recent “bad” report card.
imminent self-injurious act before the potential for harm has He was fearful of his father’s reaction. Upon his return home, a 5-
suicidal begun. to 6-hour argument with his parents ensued, and he took a vege-
behavior table (broad, sharp) knife and went to his room. He reported put-
ting the knife to his wrist but never puncturing the skin. 2) The pa-
tient stated that he “couldn’t stand being depressed anymore”
and “wanted to die.” He decided to hang himself. He tied a tele-
phone cord to the door knob and placed the cord loosely around
his neck. Then, he stopped himself and did not follow through
with the attempt.
Suicidal ideation Passive thoughts about wanting to be dead or active 1) Active: The patient reported to the doctor that he was thinking
thoughts about killing oneself, not accompanied about hanging himself in the closet. He was taken to the hospital
by preparatory behavior.a and admitted. 2) Passive: The patient reported ideas about want-
ing to be dead but denied acting on these feelings.
Nonsuicidal
events
Self-injurious Self-injurious behavior associated with no intent to 1) The patient was feeling ignored. She went into the family kitchen
behavior, no die. The behavior is intended purely for other rea- where her mother and sister were talking. She took a knife out of
suicidal intent sons, either to relieve distress (often referred to as the drawer and made a cut on her arm. She denied that she
“self-mutilation,” e.g., superficial cuts or scratches, wanted to die at all (“not even a little”), but she just wanted them
hitting/banging, or burns) or to effect change in to pay attention to her. 2) The patient reported feeling agitated
others or the environment. and anxious after a fight with her parents. She went into her
room, locked the door, and made several superficial cuts on the
inside of her arms. She stated that she felt relieved after cutting
herself and that she did not want to die. She reported that she
had done this before at times of distress and that it usually helped
her feel better. 3) The patient was in class, where a test was about
to begin, and stabbed himself with a pencil in order to be taken
to the nurse’s office. 4) A 14-year-old girl wrote her name on her
arm with a penknife and said that she often does so in order to re-
duce her anxiety. 5) The patient was noted to have multiple super-
ficial burns on his arms. Upon questioning, he denied trying to
kill himself.
Other, no No evidence of any suicidality or deliberate self-inju- 1) The patient had a cut on the neck from shaving. 2) The patient
deliberate rious behavior associated with the event. The was hospitalized for worsening of OCD or depressive symptoms
self-harm event is characterized as an accidental injury, psy- with no suicidal thoughts or actions or 3) aggressive behavior.
chiatric or behavioral symptoms only, or medical 4) Hospitalization was because of an infection, rhinoplasty, or
symptoms or procedure only. pregnancy.
Indeterminate or
potentially
suicidal events
Self-injurious Self-injurious behavior where associated intent to 1) The patient cut her wrists after an argument with her boyfriend.
behavior, die is unknown and cannot be inferred. The injury 2) The patient was angry at her husband. She took 10 to 15 diaz-
suicidal intent or potential for injury is clear, but why the individ- epam tablets and flushed the rest down the toilet. Her husband
unknown ual engaged in that behavior is unclear. called the police for help, and she was taken to the hospital. She
was groggy and stayed overnight in the hospital. 3) A 9-year-old
patient had spoken about suicide frequently. After learning that
his baseball coach was retiring, he began scratching his arm with
a pencil.
(continued)

Am J Psychiatry 164:7, July 2007 ajp.psychiatryonline.org 1037

C-CASA CLASSIFICATION OF SUICIDAL EVENTS

TABLE 2. C-CASA Definitions and Training Examples (continued)

Classification/
Category Definition Training Examples
Not enough Insufficient information to determine whether the 1) A child who “stabbed himself in [the] neck with a pencil.” The
information event involved deliberate suicidal behavior or ide- event may have been deliberate as opposed to accidental, as sug-
ation. There is reason to suspect the possibility of gested by “stabbed,” but not enough information was provided to
suicidality but not enough to be confident that the determine whether the event was deliberate. 2) A cut on the neck.
event was not something other, such as an acci-
dent or psychiatric symptom. An injury sustained
on a place on the body consistent with deliberate
self-harm or suicidal behavior (e.g., wrists), with-
out any information as to how the injury was re-
ceived, would warrant placement in this category.
a If ideation is deemed inherently related to a behavioral act, a separate rating is not given. However, if there is no clear relationship to a be-
havioral event, a separate classification of ideation is warranted.

Data suicidal events and suicide attempts. The string search identified
114 possibly suicidal events; of these, 87 (76.3%) were considered
Adverse event reports from 25 trials of antidepressant medica-
suicide attempts by pharmaceutical companies.
tions with a combined sample of 4,562 pediatric patients were
included. Reports were provided by the FDA. Twenty-four trials Broadening of event search. To insure that all potentially sui-
were sponsored by pharmaceutical companies, and one was cidal events were identified, the scope of the search was broad-
funded by the National Institute of Mental Health (NIMH) (32); ened beyond those events originally identified by pharmaceutical
however, data from that particular trial was subsequently uti- companies to include all accidental injuries, overdoses, and seri-
lized for a pediatric indication by a pharmaceutical company. ous adverse events, such as life-threatening events and hospital-
Twenty-three trials were randomized controlled trials, and two izations. Inclusion of these additional events enabled a blinded
were nonrandomized controlled trials. Participants were pediat- review, since both suicidal and other adverse events were in-
ric patients, ages 6 to 17 years, and clinical trials were conducted cluded. For classification, 427 potentially suicidal adverse events
between 1983 and 2004. The treatment duration, across nine were included. Among these events, 114 were originally rated by
medications, ranged between 4 and 16 weeks. Among SSRI-med- pharmaceutical companies as possibly suicidal.
ication trials, two were on citalopram, three on fluoxetine, one
Adverse event narrative construction. Once adverse events
on fluvoxamine, six on paroxetine, and three on sertraline. Other
were flagged by the string search, pharmaceutical companies
newer generation antidepressants studies were three bupropion
composed narratives for each adverse event using data from case
trials, one mirtazapine study, two nefazodone trials, and four
report forms, recorded by local study investigators during the
venlafaxine trials. Psychiatric diagnoses treated were major de-
course of the trials, and other sources, such as hospital records.
pressive disorder (15 trials), obsessive-compulsive disorder
When available, narratives included age, sex, history of suicidal-
([OCD] five trials), generalized anxiety disorder (two trials), so-
ity, hospitalization status, current psychosocial stressors, and
cial phobia (one trial), and attention deficit hyperactivity disor-
family history of suicide.
der ([ADHD] two trials). Fifteen of the trials were conducted
exclusively in the United States. The two nonrandomized con- Blinding
trolled trials were 1) an open-label trial of bupropion for ADHD
(N=17) and 2) a randomized withdrawal study of paroxetine for Columbia University investigators developed comprehensive
OCD (N=194). The FDA analysis (33) used a subset of events, blinding procedures that removed information from all narratives
classified by the C-CASA, from the 23 randomized controlled tri- that might have biased a classification decision. The FDA then
als described previously and events from an additional federally implemented these procedures, removing all potential drug-
funded trial ( Treatment for Adolescent Depression Study). identifying information, including the drug name, company/
Events from the Treatment for Adolescent Depression Study were sponsor name, patient identification numbers, primary diagno-
classified using the C-CASA but were not included in the present sis, active or placebo arm, and all medication names and types,
reliability study, since a different pool of raters was used and it since treatment with other medications may be associated with a
was sponsored by NIMH. particular antidepressant side-effect profile. Case numbers that
had no link to patient identifying information were randomly as-
Adverse Events signed to narratives by the FDA. Columbia University investiga-
tors further removed all original labels given by the pharmaceuti-
Pharmaceutical company identification of “possibly sui-
cal companies to categorize events (“preferred terms”) as well as
cidal” events. The FDA requested that manufacturers of all nine
adverse event labels given by participating investigators, includ-
antidepressants identify adverse events that could represent
“possibly suicidal” events. Events were identified using an elec- ing “serious” and “nonserious” determinations.
tronic text-string search of trial databases of patient data re-
Expert Raters
corded by local study clinicians. Pharmaceutical companies were
asked to search for any adverse events report that included the Nine internationally recognized experts in suicide and suicide
terms “suic overdos attempt,” “cut,” “gas,” “hang,” “hung,” assessment were recruited as “raters.” Expert review of cases was
“jump,” “mutilate,” “overdos,” “self-damage,” “self-harm,” “self- needed for inference of suicidal intent based on the details of be-
inflict,” “self-injur,” “shoot,” “slash” in the labeling of an event. haviors and related clinical data, since many narratives lacked
The FDA permitted exclusion of obvious false positives (e.g., “gas” stated suicidal intent. Expertise in suicidality was determined by
in “gastrointestinal”). The pharmaceutical companies were also relevant experience and publications. Panel members neither
asked to select a subset of events that were considered suicide at- were involved in these industry trials nor were employed by Co-
tempts. No definitional criteria were given to categorize possibly lumbia University.

1038 ajp.psychiatryonline.org Am J Psychiatry 164:7, July 2007

POSNER, OQUENDO, GOULD, ET AL.

FIGURE 1. Suicidality Classification Schemea

Suicidal Indeterminate Nonsuicidal

Completed Suicidal Self-injurious behavior Not enough

with unknown intent Self-injurious
suicide ideation information Other:
(? suicidal or (? suicidal or behavior
• accidental
nonsuicidal self- “other”) without
Suicide Preparatory action • psychiatric
injurious behavior) suicidal
attempt toward imminent suicidal • medical
intent
behavior (including
interrupted attempt or Non-consensus
aborted attempt)

a Blue boxes=FDA “primary analysis” (includes events deemed suicidal). Blue+green boxes=FDA “sensitivity analysis” (includes any event that
could possibly be suicidal).

Randomization and Expert Review Procedures known>not enough information>self-injurious behavior, no sui-
cidal intent. This approach identified 377 individual subjects, all
Event narratives were randomly distributed among raters using
of whom experienced one or more relevant adverse event. Only
a balanced incomplete block design. Each event was classified by
50 individuals had more than one event, and most of those were
three raters; each triad of raters shared five cases. This random-
accidental injuries.
ization approach reduces rater burden without sacrificing preci-
sion in variance estimates (34). Blinded examination of de-identified case records was consid-
Raters participated in a training teleconference to review clas- ered exempt from review by the institutional review board of the
sification parameters (categories, associated definitions, and case New York State Psychiatric Institute and the Columbia University
examples), followed by training reliability exercises prior to re- Department of Psychiatry.
ceiving narratives. Training exercises of each rater were reviewed
for agreement with C-CASA definitions, and disagreements were Results
discussed with the individual rater.
Each rater classified approximately 125 events. Raters could Frequencies of the 427 events according to C-CASA clas-
consult with a Columbia University trainer regarding the applica- sifications are presented in Table 3. Completed suicides are
tion of classification processes but were restricted from discuss-
not included, since none occurred in the pediatric trials.
ing specific events. Cases with discordant ratings were identified,
and corresponding narratives were resent to raters. If ratings did Reliability of C-CASA
not result in a unanimous agreement, a consensus discussion in-
cluding the three raters assigned to assess the event was held and Excellent overall reliability (median ICC=0.89) was dem-
was led by another rater. The goal was to reach 100% agreement; onstrated among independent ratings of nine experts us-
otherwise, the event was classified as “indeterminate.” Final con- ing the C-CASA. ICCs for the seven categories are pre-
sensus classification determinations were provided to the FDA.
sented in Table 3.
FDA Independent Audit of the C-CASA Of the 427 events, 366 (85.7%) had unanimous agree-
To assess the reproducibility and reliability of the C-CASA ment among the three raters. Fifty-nine events (13.8%)
methodology, four independent, nonsuicidologist FDA clinical had agreement between two of three raters, while two
reviewers were selected, including two pediatricians, one phar- (0.47%) events had no agreement. Consensus discussions
macist, and one psychiatrist. Fifteen percent of the 427 event nar-
ratives were selected for review, with oversampling of “difficult-
were held via teleconference whereby agreement was
to-classify” cases. Raters received the same training and proce- reached for all cases that were not unanimous.
dures as the expert panel. Audit results showed 89% agreement
(kappa=0.84) between audit ratings and expert ratings (35). Comparison With Pharmaceutical Companies
Discrepant cases. Thirty-eight discrepant cases were
Statistical Analysis
identified when comparing C-CASA with pharmaceutical
Reliability coefficients were estimated with a random-effects
linear model using the restricted maximum likelihood algorithm
company ratings (Table 4). Of these, 26 were new, possibly
in SPSS 12.0 for Windows. Random effects modeled event-to- suicidal cases that were originally labeled by pharmaceu-
event, rater-to-rater, and error variation. Intraclass correlation tical companies as something other than suicidal (e.g.,
coefficients (ICC) were estimated by the ratio of the variance be- accidental injury). These cases were as follows: one sui-
cause of the event divided by the total variance (sum of event-to- cide attempt, one suicidal preparatory act, 13 suicidal
event, rater-to-rater, and error variation) (34). ICCs were esti-
mated for each category.
ideation events, four self-injurious behaviors with un-
Cohen’s kappa was used to evaluate the agreement between known intent, and seven cases without enough informa-
pharmaceutical companies and C-CASA classifications. These tion but reason to suspect suicidality. The following is an
analyses were conducted with only one event per subject. For example of a newly identified suicidal event: “The pa-
subjects with multiple events, statistical calculations used the tient, age 11, held a knife to his wrist and threatened to
most severe event, which was chosen according to the severity hi-
erarchy employed by the FDA for their unblinded analyses. This
harm himself. The patient was hospitalized with an acute
severity hierarchy was as follows: suicide attempt>preparatory exacerbation of major depressive disorder.” The original
behavior>suicidal ideation>self-injurious behavior intent un- adverse event label was “exacerbation of major depres-

Am J Psychiatry 164:7, July 2007 ajp.psychiatryonline.org 1039

C-CASA CLASSIFICATION OF SUICIDAL EVENTS

TABLE 3. Frequency and Reliability Results

Frequency Percent Reliability of C-CASA Ratings ICCs
Classification/Category (N=427) (N=427) (Mean=0.89)
Suicide attempt 36 8.4 0.81
Preparatory acts toward imminent suicidal behavior 8 1.9 0.89
Suicidal ideation 62 14.5 0.97
Self-injurious behavior, suicidal intent unknown 35 8.2 0.67
Not enough information 9 2.1 0.47
Self-injurious behavior, no suicidal intent 17 4.0 0.59
Other, no deliberate self-harm 260 60.9 0.93

TABLE 4. Agreement Between C-CASA and Pharmaceutical Company Ratings of Possible Suicidal Events and Suicide Attempts
C-CASA Ratings
Pharmaceutical Company Ratings Yes No Total
Possibly suicidal events
Yes 102 12 114
No 26 237 263
Total 128 249 377
Suicide attempts
Yes 33 45 78
No 1 298 299
Total 34 343 377

sive disorder,” without an indication of suicidality from CASA. This C-CASA grouping was used by the FDA in their
either the site investigator or pharmaceutical company. primary analysis (33).
The new label was preparatory suicidal behavior. This Agreement on possibly suicidal cases. When compar-
event was discovered only because it was within a serious ing the broad nonspecific pooling of all categories that
adverse event report of a hospitalization. could possibly represent suicidality, there was good agree-
Twelve cases that were originally identified as poten- ment between C-CASA (suicide attempts, preparatory be-
tially suicidal by pharmaceutical companies were classi- haviors, suicidal ideation, self-injurious behavior with un-
fied as not potentially suicidal by C-CASA raters. These k nown intent, an d n ot enou gh infor mation) and
events were reclassified as psychiatric, involving no sui- pharmaceutical company identification of possibly sui-
cidality (N=2), accidental injury (N=1), and self-injurious cidal events (kappa=0.77 [SE=0.04]) (Table 4). This C-CASA
behavior without suicidal intent (N=9). grouping was used in the FDA’s “sensitivity analysis” to
Agreement on suicide attempts. Modest agreement conservatively examine results that included anything that
was found between pharmaceutical company and C- could have possibly represented suicidality (i.e., “worst
CASA raters’ classification of suicide attempts (kappa= case”) (33). Thus, the C-CASA identified an increased num-
0.53 [SE=0.06]) (Table 4). Of their 114 possibly suicide-re- ber of possibly suicidal events in the data set overall.
lated events, pharmaceutical companies rated 78 (68.4%)
as attempts, versus the C-CASA raters identifying 34 out of Discussion
128 (26.6%) as attempts. Forty-five of the 78 (57.7%) events
Classification of suicidal adverse events in 25 pediatric
classified as suicide attempts by the pharmaceutical com-
antidepressant trials with the C-CASA resulted in reliable
pany raters were not classified by C-CASA raters as suicide
classification of suicidal events. The C-CASA classification
attempts. One suicide attempt was identified by C-CASA
identified 38 discrepant cases, including events not previ-
raters that had not been identified by pharmaceutical
ously deemed potentially suicidal (N=26) and those
companies. Although the C-CASA identified more poten-
changed from suicidal to nonsuicidal (N=12). Further-
tially suicidal cases overall, the rate of specific suicide at-
more, while C-CASA classification found more suicidal
tempts was lower.
events, estimates of suicide attempts were significantly re-
Agreement on definitely suicidal cases. A g ree m e n t duced. The new potentially suicidal events identified in-
between C-CASA and pharmaceutical company ratings volved both suicidal ideation and behavior, across a range
increased when comparing the broader C-CASA categori- of classifications. Thus, when we expanded the search,
zation of definitely suicidal events (attempts, preparatory many new suicidal events were found that had been
acts, and suicidal ideation) with the pharmaceutical com- missed by the pharmaceutical companies. However, of the
pany rating of possibly suicidal cases (kappa=0.69 [SE= suicidal events that the pharmaceutical companies identi-
0.04]). Thirty-two events identified as possibly suicidal by fied, C-CASA classification resulted in a 50% reduction in
pharmaceutical companies were not classified as defi- the rate of suicide attempts. This reflects a tendency of the
nitely suicidal by the C-CASA. Conversely, 15 newly identi- pharmaceutical companies to label any potentially sui-
fied definitely suicidal cases were identified by the C- cidal event or self-injurious behavior as a suicide attempt

1040 ajp.psychiatryonline.org Am J Psychiatry 164:7, July 2007

POSNER, OQUENDO, GOULD, ET AL.

(e.g., suicidal ideation or a “slap in the face” labeled sui- analysis using C-CASA ratings—by Hammad et al. (33) in
cide attempt). These findings underscore the need for a the FDA’s safety analysis, for example—some discussion of
standardized assessment of suicidality. Additionally, the the limitations of these subsequent analyses is warranted.
need to expand the search for suicidal events as evidenced Suicidal adverse events were not systematically elicited
by the 26 newly found cases suggests that approaches cur- but were revealed spontaneously, allowing the possibility
rently employed in clinical trials lack sensitivity. of ascertainment bias. Subjects receiving active medica-
When comparing the C-CASA ratings with pharmaceuti- tion may be more likely to report suicidal occurrences
cal company ratings, a relatively low level of agreement was than those on placebo because of increased contact with
found with more specific identification of suicidal occur- providers, consequent to other side effects. Such ascer-
rences, namely suicide attempts. Only when identifying a tainment bias is an alternate explanation for differential
“suicidal range” or a broad nonspecific category of “possi- rates among subjects receiving drug treatment versus
bly suicidal” was there better agreement. Pharmaceutical those receiving placebo found in the FDA safety analysis
companies rated 45 events as suicide attempts that C- (33). In addition, improvement from active medication
CASA raters did not. Thus, with respect to suicide attempts, may lead subjects to discuss suicidal thoughts with their
reclassification with C-CASA would yield less of a hazard clinician for the first time, as opposed to such thoughts
from the medication than if the original pharmaceutical being caused by the medication.
ratings were used. Indeed, the FDA safety analysis that Future intervention trials that prospectively and sys-
used these C-CASA ratings (33) found reduced risk esti- tematically monitor occurrence and emergence of suicid-
mates of suicidality in a depressed pediatric sample when ality with consistent methods of ascertainment would be
compared with earlier FDA estimates that relied on the informative. Such investigations would more optimally
pharmaceutical labels (36). Additionally, a more precise delineate the relationship between suicidal adverse events
risk estimate resulted (i.e., tighter confidence interval) us- and antidepressant treatments as well as for any other
ing the C-CASA. These findings support the notion that treatment risk analysis. Improved assessment of suicidal
misclassification may lead to overestimation of true risk events is necessary both to better inform research-derived
(37). Such a change in risk estimation has clinical implica- risk-benefit analyses and to foster improved clinical man-
tions and likely affects risk-benefit analyses. Furthermore, agement and identification. Accordingly, a prospective
the final FDA data set with the C-CASA ratings (33) in- counterpart to this system, the Columbia Suicide Severity
cluded one-third (38/114) of cases that were different com- Rating Scale (39), is being widely used and frequently rec-
pared with the original data set (36), a substantially differ- ommended by the FDA. The Columbia Suicide Severity
ent sample. The use of data sets with imprecisely classified Rating Scale is a tool designed to systematically assess and
suicidal events can result in misleading findings, such as track suicidal adverse events (behavior and ideation)
inaccurate risk and protective factors for suicidality. throughout any clinical trial as well as other settings.
The reliability of this classification approach was con- The strength of this suicide classification system is, per-
firmed by the FDA’s independent audit, which concluded haps, in its ability to comprehensively identify suicidal
that the C-CASA was “robust and reproducible” (35). The events while limiting the overidentification of suicidal be-
reliable use of this classification schema by nonsuicidolo- havior. This classification system is research-based and
gists reflects the transportability of this methodology. No- can be applied in both clinical and research settings. Its use
tably, the FDA has mandated application of C-CASA to might result in more accurate identification of suicidality
classify suicidal adverse events in adult antidepressant tri- and more precise communication among researchers and
als, as well as nonpsychotropic drug classes, and other cen- clinicians, which would ultimately benefit treatment of
trally acting agents, including all anticonvulsants, canna- suicidal individuals. The incorporation of research-sup-
binoid 1 receptor (CB1R) inverse agonists for the treatment ported, standardized suicidality terminology into psych-
of obesity and metabolic disease. C-CASA classified data iatric diagnostic manuals could also promote greater
were used in the recent FDA investigation of an association accuracy in communication between clinicians, allowing
between antidepressants and suicidality in adults (38). dissemination to a broad audience. Such a common lan-
guage of suicide classification could be used in the same
Limitations and Future Directions
way that diagnostic criteria are currently used to provide a
The study findings are limited by the quality of the method for precise, widely understood communication.
available data describing adverse events. Descriptions of
suicidal occurrences were variable and limited, particu- Received June 16, 2006; revision received Dec. 15, 2006; accepted
larly regarding intent. Furthermore, the expanded search Jan. 14, 2007. From the Division of Child and Adolescent Psychiatry
and the Department of Neuroscience, Columbia University/New York
for unidentified occurrences elucidated the inadequate
State Psychiatric Institute, New York; Department of Psychology, John
quality of the elicitation and description of suicidal ad- Jay College of Criminal Justice, City University of New York, New York;
verse events. and the Department of Epidemiology, Mailman School of Public
Health, Columbia University, New York. Address correspondence and
Although neither the C-CASA raters nor Columbia Uni- reprint requests to Dr. Posner, Division of Child Psychiatry, Columbia
versity investigators were responsible for subsequent University/New York State Psychiatric Institute, 1051 Riverside Dr.,

Am J Psychiatry 164:7, July 2007 ajp.psychiatryonline.org 1041

C-CASA CLASSIFICATION OF SUICIDAL EVENTS

Unit 74, New York, NY 10032; [email protected] (e- Schmidtke A, Bille-Brahe U, DeLeo D, Kerkhof A. Gottingen,
mail). Germany, Hogrefe and Huber, 2004, pp 11–14
Supported in part by NIMH grant P20 AA015630-03 to Dr. Stanley. 10. O’Carroll PW, Berman AL, Maris RW, Moscicki EK, Tanney BL, Sil-
The authors thank Thomas Laughren, M.D. (FDA) and Tarek Ham- verman MM: Beyond the Tower of Babel: a nomenclature for
mad, M.D., Ph.D., M.Sc., M.S. (FDA), J. John Mann, M.D. (Columbia Uni-
suicidology. Suicide Life Threat Behav 1996; 26:237–252
versity, New York), David Shaffer, M.D. (Columbia University, New
York), Laurence Greenhill, M.D. (Columbia University, New York), and 11. Goldston DB: Measuring Suicidal Behavior and Risk in Children
the Expert Suicidality Panel: Annette Beautrais, Ph.D.; David Brent, and Adolescents. Washington, DC, American Psychological As-
M.D.; Greg Brown, Ph.D.; Cheryl King, Ph.D.; Peter Marzuk, M.D.; Alec sociation, 2003
Miller, Psy.D.; Patrick O'Carroll, M.D., M.P.H.; David Rudd, Ph.D.; An- 12. Beck AT, Beck R, Kovacs M: Classification of suicidal behaviors:
thony Spirito, Ph.D.; and Kees van Heeringen, M.D., Ph.D. The au- I. quantifying intent and medical lethality. Am J Psychiatry
thors also thank Marjorie Kleinman, M.S., Glenn A. Melvin, Ph.D., and 1975; 132:285–287
S. Sonia Gugga, M.S.
13. Joe S: Self-directed violence definitions: a review of the scien-
tific literature, in the Uniform Definitions of Self-Injury Confer-
CME Disclosure ence, the National Center for Injury Prevention and Control,
Dr. Posner has received funding from the FDA to develop the suicid- Centers for Disease Control and Prevention. Miami, Fla, 2004
ality classification system used in their antidepressant safety analysis. 14. Brown GK, Jeglic E, Henriques GR, Beck AT: Cognitive Therapy,
Subsequently, as part of an effort to help execute the FDA adult sui-
Cognition and Suicide Behavior, in Cognition and Suicide: The-
cidality classification mandates that followed this FDA-sponsored pe-
ory, Research, and Therapy. Edited by Ellis TE. Washington, DC,
diatric study, Dr. Posner has had research support from GlaxoSmith-
Kline, Forest Laboratories, Eisai, Inc., AstraZeneca, Johnson and American Psychological Association, 2006, pp 53–74
Johnson, Abbott Laboratories, Wyeth, Organon USA, Bristol-Myers 15. Goldsmith SK, Pellmar TC, Kleinman AM, Bunney WE (eds): Re-
Squibb, Sanofi-Aventis, Cephalon, Novartis, Shire Pharmaceuticals, ducing Suicide: A National Imperative (Committee on Patho-
and UCB Pharma. Dr. Oquendo has received funding from NIMH, Na- physiology and Prevention of Adolescent and Adult Suicide,
tional Institute on Alcohol Abuse and Alcoholism, Moody’s Founda- Board on Neuroscience and Behavioral Health, Institute of
tion, American Foundation for Suicide Prevention, and Eli Lilly (unre- Medicine of the National Academies). Washington, DC, The Na-
stricted educational grant); she has also served as a consultant to
tional Academies Press, 2002
Pfizer. Dr. Stanley has received fluoxetine/placebo for a randomized
controlled trial from Eli Lilly. Mr. Davies is a shareholder with Merck, 16. Oquendo MA, Galfalvy H, Russo S, Ellis SP, Grunebaum MF,
Pfizer, Wyeth, GlaxoSmithKline, Johnson and Johnson, Amgen, and Burke A, Mann JJ: Prospective study of clinical predictors of sui-
Bard. Dr. Gould reports no competing interests. cidal acts after a major depressive episode in patients with ma-
APA policy requires disclosure by CME authors of unapproved or in- jor depressive disorder or bipolar disorder. Am J Psychiatry
vestigational use of products discussed in CME programs. Off-label 2004; 161:1433–1441
use of medications by individual physicians is permitted and com- 17. Brent DA, Perper JA, Moritz G, Allman C, Friend A, Roth C,
mon. Decisions about off-label use can be guided by scientific litera- Schweers J, Balach L, Baugher M: Psychiatric risk factors for ad-
ture and clinical experience.
olescent suicide: a case-control study. J Am Acad Child Adolesc
Psychiatry 1993; 32:521–529
18. Dorpat TL, Ripley H: The relationship between attempted sui-
References cide and committed suicide. Compr Psychiatry 1967; 8:74–79
1. Olfson M, Marcus SC, Weissman MM, Jensen PS: National trends 19. Nordstrom P, Samuelsson M, Asberg M: Survival analysis of sui-
in the use of psychotropic medications by children. J Am Acad cide risk after attempted suicide. Acta Psychiatr Scand 1995;
Child Adolesc Psychiatry 2002; 41:514–521 91:336–340
2. Vitiello B, Zuvekas SH, Norquist GS: National estimates of anti- 20. Marzuk PM, Tardiff K, Leon AC, Portera L, Weiner C: The preva-
depressant medication use among US children, 1997–2002. J lence of aborted suicide attempts among psychiatric in-pa-
Am Acad Child Adolesc Psychiatry 2006; 43:271–279 tients. Acta Psychiatr Scand 1997; 96:492–496
3. IMS Health America: National Prescription Audit: Information, 21. Steer RA, Beck AT, Garrison B, Lester D: Eventual suicide in in-
Plymouth Meeting. Plymouth, Pa, Information Marketing Sys- terrupted and uninterrupted attempters: a challenge to the
tems cry-for-help hypothesis. Suicide Life Threat Behav 1988; 18:
4. Whittington CJ, Kendall T, Fonagy P, Cottrell D, Cotgrove A, Bod- 119–128
dington E: Selective serotonin reuptake inhibitors in childhood 22. Fergusson D, Doucette S, Glass KC, Shapiro S, Healy D, Hebert P,
depression: systematic review of published versus unpub- Hutton B: Association between suicide attempts and selective
lished data. Lancet 2004; 363:1341–1345 serotonin reuptake inhibitors: systematic review of ran-
5. Cheung AH, Emslie GJ, Mayes, TL: Review of the efficacy and domised controlled trials. Br Med J 2005; 330:396–399
safety of antidepressants in youth depression. J Child Psychol 23. Wichstrom L: Predictors of adolescent suicide attempts: a na-
Psychiatry 2005; 46:735–754 tionally representative longitudinal study of Norwegian ado-
6. Mann JJ, Emslie G, Baldessarini RJ, Beardslee W, Fawcett JA, lescents. J Am Acad Child Adolesc Psychiatry 2000; 39:603–610
Goodwin FK, Leon AC, Meltzer HY, Ryan JD, Shaffer D, Wagner 24. Pfeffer C, Klerman GL, Hurt SW, Kakuma T, Peskin JR, Siefker
KD: ACNP Task Force Report on SSRIs and suicidal behavior in CA: suicidal children grow up: rates and psychosocial risk fac-
youth. Neuropsychopharmacology 2006; 31:473–492 tors for suicide attempts during follow-up. J Am Acad Child Ad-
7. Rosack, J: New data show declines in antidepressant prescrib- olesc Psychiatry 1993; 32:106–113
ing. Psychiatr News 2005; 40:1 25. Lewinsohn PM, Rohde P, Seeley JR: Psychosocial risk factors for
8. Gibbons RD: Efficacy and safety of antidepressants for depres- future adolescent suicide attempts. J Consult Clin Psychol
sion and suicide risk in youth and adults: results of new analy- 1994; 62:297–305
ses, in the American College of Neuropsychopharmacology 26. Brown GK, Beck AT, Steer RA, Grisham JR: Risk factors for sui-
Annual Meeting, Hollywood, Fla, American College of Neuro- cide in psychiatric outpatients: a 20-year prospective study. J
psychopharmacology, 2006 Consult Clin Psychol 2000; 68:371–377
9. Bille-Brahe U, Kerkhof A, DeLeo D, Schmidtke A: Definitions 27. Brown GK, Steer RA, Henriques GR, Beck AT: The internal strug-
and termination used in the World Health Organization/EURO gle between the wish to die and the wish to live: a risk factor
Multicentre Study, in Suicidal Behaviour in Europe. Edited by for suicide. J Consult Clin Psychol 2005; 162:1977–1979

1042 ajp.psychiatryonline.org Am J Psychiatry 164:7, July 2007

POSNER, OQUENDO, GOULD, ET AL.

28. Young EA, Coryell W: Suicide and the hypothalamic-pituitary- the Pediatric Advisory Committee Meeting, Bethesda, Md, FDA,
adrenal axis. Lancet 2005; 366:959–960 2004 (http://www.fda.gov/ohrms/dockets/ac/04/slides/2004–
29. Hawton K, Cole D, O’Grady J, Osborn M: Motivational aspects of 4065S1_07_FDA-Iyasu.htm) (Accessed Aug. 2006)
deliberate self-poisoning in adolescents. Br J Psychiatry 1982; 36. Mosholder A: Comparison between original ODS and current
141:286–291 DNDP analyses of pediatric suicidality data sets, in Psychophar-
30. Beck RW, Morris JB, Beck AT: Cross validation of the Suicidal In- macologic Drugs Advisory Committee and the Pediatric Advi-
tent Scale. Psychol Rep 1974; 34:445–446 sory Committee Meeting. Bethesda, Md, FDA, 2004 (http://
31. Stanley B, Gameroff MJ, Michalsen V, Mann JJ: Are suicide at- w w w. f d a . g o v / o h r m s / d o c k e t s / a c / 0 4 / s l i d e s / 2 0 0 4 –
tempters who self-mutilate a unique population? Am J Psychi- 4065S1_09_FDA-Mosholder.htm) (Accessed Aug. 2006)
atry 2001; 158:427–432 37. Jurek AM, Greenland S, Maldonado G, Church TR: Proper inter-
32. Emslie GJ, Rush AJ, Weinberg WA, Kowatch RA, Hughes CW, Car- pretation of non-differential misclassification effects: expected
mody T, Rintelmann J: A double-blind, randomized, placebo- vs observations. Int J Epidemiol 2005; 34:680–687
controlled trial of fluoxetine in children and adolescents with 38. Laughren TP: Overview for Meeting of Psychopharmacologic
depression. Arch Gen Psychiatry 1997; 54:1031–1037 Drugs Advisory Committee (PDAC) Memorandum Department
33. Hammad TA, Laughren T, Racoosin J: Suicidality in pediatric of Health and Human Services, Public Health Service. Be-
patients treated with antidepressant drugs. Arch Gen Psychia- thesda, Md, Food and Drug Administration, Center for Drug
try 2006; 63:332–339 Evaluation and Research, 2006 (http://www.fda.gov/ohrms/
34. Shrout PE, Fleiss JL: Intraclass correlations: uses in assessing dockets/ac/06/briefing/2006–4272b1–01-FDA.pdf)
rater reliability. Psychol Bull 1979; 86:420–428 39. Posner K, Melvin GA, Stanley B, Oquendo MA, Gould M: Factors
35. Iyasu S: Audit of the Columbia Suicidality Classification Project, in the assessment of suicidality in youth. CNS Spectr 2007; 12:
in FDA Psychopharmacologic Drugs Advisory Committee and 156–162

Am J Psychiatry 164:7, July 2007 ajp.psychiatryonline.org 1043