Clusters of Event Reactions Among Recipients of Project Liberty Mental Health Counseling

Project Liberty services were provided by more than 175 community-based organizations in a wide variety of settings, as described by Felton ( 10 ). Contracted service providers were paid to provide individual crisis counseling, educational services to groups, and group crisis counseling. Under FEMA funding conditions, all services provided under Project Liberty were anonymous; no participant names were reported. As an administrative requirement for reimbursement, providers were required to complete service encounter log forms for each contact, indicating date and location of service (such as a school or place of employment) and borough or county. For persons receiving individual counseling, service providers indicated whether the counseling contact was a first visit or a follow-up visit and provided demographic information (including gender, preferred language, and race or ethnicity, as observed by the counselor).

Logs included a record of the presence of up to 31 event reactions. Event reactions were categorized as behavioral (extreme change in activity level; excessive use of drugs, alcohol, or prescription drugs; isolation or withdrawal; hypervigilance; reluctance to leave home; and violent behavior), emotional (sadness or tearfulness, irritability or anger, feeling anxious or fearful, despair or hopelessness, feeling guilty or shameful, and feeling emotionally numb or disconnected), physical (headaches, stomach problems, difficulty falling or staying asleep, difficulty eating, worsening of chronic health conditions, fatigue or exhaustion, and chronic agitation), and cognitive (inability to accept or cope with the death of a loved one, distressing dreams, intrusive thoughts or images, difficulty concentrating, difficulty remembering things, difficulty making decisions, preoccupation with death, and suicidal thoughts or feelings). Each category also had items labeled other, unknown, and none. Counselors checked event reactions that were present. The Mount Sinai School of Medicine Institutional Review Board found that this project involved analyzing anonymous, secondary data and was therefore exempt from the board's review.

Hierarchical cluster analysis was performed on the 31 event reactions recorded for 465,428 logs of initial counseling sessions that occurred between September 2001 and December 2003. Because the event reaction data were coded as either present or absent, absolute binary phi 4-point correlations were used (the equivalent of a Pearson correlation between two dichotomous variables) to create clusters. Hierarchical cluster analyses first link the two variables that are most strongly correlated with one another and then join subsequent variables to a new variable or to an existing cluster on the basis of the strength of their correlations. This linking is done in a hierarchical fashion by entering larger correlations first and the smallest correlation last. Because each successive cluster is based on weaker associations, later clusters are less likely to comprise meaningful groupings. Moreover, event reactions that occur rarely (such as suicidal thoughts or feelings) have attenuated correlations and are thus less likely to be included in the clusters. SPSS 12.0 for Windows was used to conduct these analyses ( 11 ). We conducted a hierarchical cluster analysis specifying between-groups linkage using an absolute phi 4-point correlation measure.

Individual crisis counseling services were accessed by a broad spectrum of the population. Of the total sample, 91.0 percent (N=423,315) were adults, 97.5 percent (N=453,580) lived in New York City, 31.2 percent (N=145,198) were Caucasian, 30.1 percent (N=140,209) were African American, 25.3 percent (N=117,834) were Hispanic, and 52.9 percent (N=246,070) were women ( 12 ). Rates of endorsement of event reactions ranged from 42.6 percent (N=198,382) for sadness and tearfulness to .6 percent (N=2,777) for suicidal thoughts or feelings. Table 1 displays the numbers and proportions of service recipients who exhibited those event reactions. [An expanded version of this table is available at ps.psy chiatryonline.org and includes a 31×31 correlation matrix.]

Seven distinct clusters were identified from the 31 items. The weakest included a wide array of responses coded under "other" reactions in the behavioral, emotional, physical, and cognitive domains. This cluster was dropped from further discussion.

The two most statistically robust and coherent clusters included event reactions that resembled criteria for DSM-IV diagnoses of major depressive disorder and PTSD ( 1 ). The cluster that most closely resembled major depressive disorder (hereinafter referred to as "depressive episode cluster") contained ten items, seven of which matched major depressive disorder criteria listed in DSM-IV: despair and hopelessness, feeling guilty or shameful, difficulty eating, fatigue or exhaustion, difficulty concentrating, difficulty remembering things, and difficulty making decisions. The three remaining items were isolation and withdrawal, feeling emotionally numb or disconnected, and stomach problems.

The cluster that most closely resembled PTSD (hereinafter referred to as "traumatic stress cluster") contained eight items, seven of which corresponded to DSM-IV criteria for the disorder: hypervigilance, reluctance to leave home, feeling anxious or fearful, difficulty falling or staying asleep, distressing dreams, intrusive thoughts or images, and preoccupation with death. The final item of the cluster was extreme change in activity level.

Two additional clusters were designated as aggression and grief clusters. The aggression cluster included violent behavior, irritability and anger, and chronic agitation, whereas the grief cluster included inability to cope with the death of a loved one, sadness and tearfulness, and headaches. The three remaining items—excessive use of drugs, alcohol, or prescription drugs; worsening of a chronic health condition; and suicidal thoughts or feelings—did not cluster with any of the previous four groups.

Next, hierarchical cluster analyses were conducted separately for different subgroups (for example, males only, females only, and African Americans only) to determine whether similar clusters would emerge. The depressive episode cluster consistently included nine of ten reactions found in the overall group (all but feeling guilty or shameful, which typically clustered with grief in the subgroup analyses). The traumatic stress cluster consistently included seven of eight reactions found in the overall group (all but preoccupation with death, which also typically clustered with grief). The clusters used in subsequent analyses included only the depressive episode cluster (nine items) and traumatic stress cluster (seven items).

Next, we calculated depressive symptom and traumatic stress scores for each service recipient on the basis of his or her score on these clusters. The rarely observed item on suicidal thoughts and feelings was added to the depressive episode cluster in order to include individuals who reported this rare but serious reaction. Thus depressive symptom scores could range from 0 to 10, whereas traumatic stress scores could range from 0 to 7. Resulting cluster scores were highly skewed, with a mean±SD of 1.18±1.46 (range=0-10; median=1.0) for depression and a mean of 1.64±1.50 (range=0-7; median=1.0) for traumatic stress.

Service recipients were also divided into two subgroups based on whether they had three or more event reactions within either cluster to identify individuals who likely fit Marshall and colleagues' ( 8 ) description for manifesting subthreshold conditions. The Marshall study reported significant linear increases in impairment as the number of endorsed symptoms increased from one to four. The presence of three symptoms was chosen to match the criterion that Marshall and colleagues set for defining subthreshold PTSD; 16 percent of recipients met these criteria for subthreshold or greater major depressive disorder, 26 percent for PTSD, and 8 percent for both. Although displaying three or more event reactions is not the equivalent of meeting full criteria for either of these diagnoses, the scores were used to show the pattern of morbidity for these syndromes over the first 27 months since the attacks. As illustrated in Figure 1, individuals in all three of these groups sought first-time Project Liberty counseling on a relatively consistent basis for more than two years after the attacks on the World Trade Center.

Longitudinally, 55.8 percent (275 of 493 persons) of individuals receiving first-time crisis counseling in the first month after the attacks met our high-risk criteria for either one or both clusters. Percentages dropped across the next five months to a low of 27.1 percent (2,505 of 9,237 persons), then rose to a second peak of 40.1 percent (10,465 of 26,120 persons) at ten months. Percentages then dropped again in a linear fashion to 24.4 percent (5,681 of 23,307 persons) at 15 months, and peaked again at 40.2 percent (10,290 of 25,599 persons) at 22 months. The ten- and 22-month peaks correspond to anticipation of the first-year and second-year anniversaries of the attacks on the World Trade Center.