lable at ScienceDirect

Behaviour Research and Therapy 51 (2013) 872e881

Contents lists avai

Behaviour Research and Therapy

journal homepage: www.elsevier .com/locate/brat

Cognitive predictors and moderators of winter depression treatmentoutcomes in cognitive-behavioral therapy vs. light therapy

Lilya Sitnikov, Kelly J. Rohan*, Maggie Evans, Jennifer N. Mahon, Yael I. NillniUniversity of Vermont, Department of Psychology, John Dewey Hall, 2 Colchester Avenue, Burlington, VT 05405-0134, United States

a r t i c l e i n f o

Article history:Received 4 October 2012Received in revised form21 September 2013Accepted 30 September 2013

Keywords:Seasonal affective disorderCognitive-behavioral therapyLight therapyCognitive vulnerabilityPrescriptive factorsPrognostic factors

* Corresponding author. Tel.: þ1 802 656 0915; faxE-mail addresses: lsitniko@uvm.edu, lilsit06@gma

0005-7967/$ e see front matter � 2013 Elsevier Ltd.http://dx.doi.org/10.1016/j.brat.2013.09.010

a b s t r a c t

There is no empirical basis for determining which seasonal affective disorder (SAD) patients are bestsuited for what type of treatment. Using data from a parent clinical trial comparing light therapy (LT),cognitive-behavioral therapy (CBT), and their combination (CBT þ LT) for SAD, we constructed hierar-chical linear regression models to explore baseline cognitive vulnerability constructs (i.e., dysfunctionalattitudes, negative automatic thoughts, response styles) as prognostic and prescriptive factors of acuteand next winter depression outcomes. Cognitive constructs did not predict or moderate acute treatmentoutcomes. Baseline dysfunctional attitudes and negative automatic thoughts were prescriptive of nextwinter treatment outcomes. Participants with higher baseline levels of dysfunctional attitudes andnegative automatic thoughts had less severe depression the next winter if treated with CBT than iftreated with LT. In addition, participants randomized to solo LT who scored at or above the sample meanon these cognitive measures at baseline had more severe depressive symptoms the next winter relativeto those who scored below the mean. Baseline dysfunctional attitudes and negative automatic thoughtsdid not predict treatment outcomes in participants assigned to solo CBT or CBT þ LT. Therefore, SADpatients with extremely rigid cognitions did not fare as well in the subsequent winter if treated initiallywith solo LT. Such patients may be better suited for initial treatment with CBT, which directly targetscognitive vulnerability processes.

� 2013 Elsevier Ltd. All rights reserved.

Winter seasonal affective disorder (SAD) is characterized byrecurrent Major Depressive Episodes that begin in the fall or winterand remit in the spring (APA, 2000). Untreated and annuallyrecurring SAD episodes lead to impairments in activities of living,emotional well-being, and overall health in the winter (Schlager,Froom, & Jaffe, 1995). Given that 10%e20% of all cases of recurrentdepression follow a seasonal pattern (Blazer, Kessler, & Schwartz,1998; Magnusson, 2000), it is a public health priority to developinterventions that prevent recurrence of depressive episodes oversubsequent winter seasons.

Bright light therapy (LT) and SAD-tailored cognitive-behavioraltherapy (CBT) have been shown to be efficacious in the acutetreatment of SAD (Golden et al., 2005; Rohan et al., 2007). In anuncontrolled feasibility study comparing LT, CBT, and their combi-nation in the treatment of adult SAD patients, CBT (alone or com-bined with LT) was comparably efficacious to LT alone in reducingacute SAD symptoms (Rohan, Tierney Lindsey, Roecklein, & Lacy,2004). A subsequent controlled, randomized clinical trial found

: þ1 802 656 8783.il.com (L. Sitnikov).

All rights reserved.

that participants randomized to CBT, LT, or combination treatmentevidenced significant and comparable reductions in depressivesymptoms at post-treatment relative to a concurrent wait-listcontrol group (Rohan et al., 2007). Although there were no statis-tically significant differences between treatments in full remissionstatus at post-treatment, the combined CBT þ LT condition had thelargest proportion of participants classified as remitted (73e79%).In a naturalistic follow-up during the next winter season, the CBT(7.0%) and CBT þ LT treatments (5.5%) had significantly smallerproportions of winter depression recurrences than the solo LTtreatment (36.7%; Rohan, Roecklein, Lacy, & Vacek, 2009). In addi-tion, solo CBT, but not combination treatment, had less severe blindinterviewer- and self-rated depressive symptoms the next winterthan solo LT. Therefore, although daily LT use has significant anti-depressant effects during the initial winter of treatment (Goldenet al., 2005), results of recent clinical trials suggest that CBT maybe an effective alternative treatment to LT in treating acute SAD(Rohan et al., 2004; 2007) and may have more enduring effectsthan LT in preventing SAD episode recurrence and reducingsymptom severity during the subsequent winter (Rohan, Roecklein,& Haaga, 2009; Rohan, Roecklein, Lacy, et al., 2009).

L. Sitnikov et al. / Behaviour Research and Therapy 51 (2013) 872e881 873

Despite empirical support for LT and CBT as efficacious acuteSAD treatments, approximately 43e60% of patients randomized toeither solo CBT or LT do not meet remission criteria at post-treatment (Rohan et al., 2007; Terman, Terman, & Quitkin, 1989).Moreover, many formerly treated SAD patients continue to expe-rience clinically significant depressive symptoms in the wintersubsequent to study treatment (proportions of participants inremission the next winter: 30% LT, 37% CBT þ LT, 58% CBT; Rohanet al., 2009; Rohan, Roecklein, Lacy, et al., 2009). Yet, little isknown about factors that predict favorable or unfavorable acute ornext winter treatment outcomes in SAD. More work is needed toinform an empirical basis for determining which SAD patients arebest suited for what type of treatment. To our knowledge, only onestudy has examined a cognitive predictor of LT response (Levitan,Rector, & Bagby, 1998) and no studies have evaluated cognitivemoderators of CBT outcomes among SAD patients. Levitan et al.(1998) reported that baseline negative attributional style (i.e., thetendency to make stable and global attributions about life events)did not predict response to LT among SAD patients. However, thisone study does not informwhether this or other putative cognitivevulnerability constructs might be prognostic of treatment out-comes across treatment modalities or prescriptively impact therelative efficacy of CBT, LT, or their combination.

Studies examining predictors of nonseasonal depression treat-ment outcomes have distinguished between prognostic (i.e.,predict efficacy irrespective of the treatment modality) and pre-scriptive (i.e., account for a different pattern of outcome betweentreatments e moderators; Kraemer, Wilson, Fairburn, & Agras,2002) patient baseline characteristics and traits (e.g., Fournieret al., 2009). Identifying prognostic factors has both theoreticaland clinicalutility. Pre-randomization factors that are associatedwith superior or refractory treatment response in SAD may help toidentify novel etiological risk and protective processes, which couldthen be incorporated into theory and become novel targets oftreatment and prevention. SAD patients who do not respondfavorably to available treatments, irrespective of treatment mo-dality, may benefit from a novel, extended, or supplementarytreatment. That is, novel interventions could be developed orexisting empirically-supported interventions could be revised (e.g.,increased intensity, extended length, new components added) tomitigate or explicitly target prognostic factors. Identification ofprescriptive factors may enhance clinical practice by providingguidelines for selecting an appropriate and evidence-supportedintervention for each presenting SAD patient (i.e., personalizedmedicine). Comparisons of prescriptive factors for LT, CBT, and theircombination would be particularly informative because CBT andLT are designed to target different etiological processes, havedifferent putative mechanisms of action, and are vastly differentinterventions in terms of what is required of the patient. Baselinecognitive constructs are candidate prescriptive factors, giventheir explicit focus in CBT for SAD (Rohan, Sigmon, & Dorhofer,2003) and their accessibility to measurement.

Recent studies on differential response to depression treatmentshave distinguished between compensation and capitalization ap-proaches for adapting treatments to patients’ needs. Under thecompensation model (Cheavens, Strunk, Lazarus, & Goldstein,2012), an intervention targets disorder-specific etiological pro-cesses and vulnerabilities with the aim of addressing or modifyingthe vulnerability. According to this model, patients who display acertain characteristics should, theoretically, benefit more from anintervention aimed at modifying this disorder-specific vulnerabilitythan patients who receive a treatment that does not directly targetsaid vulnerability. In contrast, under the capitalization model, themost effective intervention option would target an individual’sstrengths to bolster them further (Cheavens et al., 2012; Simon &

Perlis, 2010). The rationale underlying most, if not all, SAD treat-ments is consistent with a compensation conceptualization (i.e., thetreatment is presumed to remedy vulnerabilities associated withwinter depressive symptoms). Therefore, if compensation appliesto SAD, SAD patients with a high cognitive vulnerability shouldbenefit from CBT more than those with a lower cognitive vulner-ability who receive CBT and more than those with a high cognitivevulnerability treated with LT because only CBT is designed to targetmaladaptive cognitions. Conversely, according to compensation,those with a high physiological vulnernability (e.g., a circadianphase shift in the winter) should benefit more from LT than thosewith a lower physiological contribution to their SAD and thosewitha high physiological vulnerability treated with CBT. It is lessstraightforward to conceptualize SAD treatments from a capitali-zation perspective. To match depressed patients to cognitive,behavioral, mindfulness, or interpersonal intervention modulesbased on capitalization; Cheavens et al. (2012) administered thetwo treatments that were most consistent with the types of mood-regulation strategies patients were already using at baseline.Applying this type of personalization based on capitalization toSAD, it is possible that patients who cope with their symptomsusing behaviors such as adopting a new perspective and pushingthemselves to stay active in the winter benefit the most from CBTwhereas patients who cope by taking a vacation to a warm, sunnyplace benefit the most from LT. In keeping with capitalization, it isalso possible that a more general construct such as locus of controlmoderates SAD treatment outcome. For example, perhaps thosewith a more internal locus of control benefit from CBT whereasthose with a more external locus of control benefit from LT.

In nonseasonal depression, higher pre-treatment levels ofdysfunctional attitudes (i.e., stable and global beliefs concerningperfectionism, need for approval, inadequacy, and perceived re-quirements for happiness) “appear to be a prognostic indicator ofpoor response to CBT” (Hamilton & Dobson, 2002, p. 887). Contraryto the compensation model, across studies, after controlling forpre-treatment depression severity; depressed patients with higherpre-treatment levels of dysfunctional attitudes had poorer out-comes with CBT in terms of post-treatment scores on self-reportand interviewer-administered measures of depressive symptoms(Jarrett, Eaves, Brannemann, & Rush, 1991; Keller, 1983; Simons,Gordon, Monroe, & Thase, 1995; Sotsky et al., 1991) and a greaterlikelihood of relapse over 1-year follow-up (Thase et al., 1992).Therefore, a theoretical match between a patient’s depressogenicvulnerability at baseline (i.e., highly dysfunctional attitudes) andthe domain targeted by the applied intervention (i.e., CBT) may nottranslate into superior therapeutic outcomes. It is possible thatextremely dysfunctional attitudes at baseline limits the efficacy ofCBT by interfering with a patient’s capacity to identify and effec-tively challenge unhelpful patterns of thinking as well as with apatient’s ability to plan and engage in distracting and pleasant ac-tivities. Preliminary evidence suggests that higher baselinedysfunctional attitudes may be prognostic of worse depressiontreatment outcomes not only in CBT, but across different treatmentmodalities, including interpersonal psychotherapy (Carter et al.,2007) and antidepressant medications (Jacobs et al., 2009;Peselow, Robins, Block, Barouch, & Fieve, 1990; Sotsky et al.,1991). Thus, highly rigid depressogenic attitudes may workagainst beneficial treatment effects, regardless of depressiontreatment modality.

Studies have also examined the prognostic and prescriptiveutility of a ruminative response style (i.e., the tendency to respondto dysphoric mood and symptoms of depression by repetitivelythinking about their causes and consequences; Nolen-Hoeksema,1987) in predicting treatment outcomes in nonseasonal depres-sion. In several studies, higher baseline rumination was prognostic

L. Sitnikov et al. / Behaviour Research and Therapy 51 (2013) 872e881874

of poor outcome in CBT (Jones, Siegle, & Thase, 2008; Kuehner &Weber, 1999), as well as in other types of treatment, includingparoxetine, problem-solving therapy, and pill placebo (Schmaling,Dimidjian, Katon, & Sullivan, 2002). However, other studies havefound that pre-treatment rumination scores were not associatedwith change in depression severity or remission status at post-treatment in other types of depression treatment (outside of CBTbased on Beck, Rush, Shaw, & Emery, 1979), including nefazodone,the Cognitive Behavioral Analysis System of Psychotherapy, or theircombination (Arnow, Spangler, Klein, & Burns, 2004) and phar-macotherapy (Bagby & Parker, 2001). Therefore, the literature todate is mixed on whether rumination is prognostic of poor treat-ment outcomes, regardless of modality, or is prescriptive of pooroutcome in CBT. Similar to highly dysfunctional attitudes, a strongpredisposition to ruminate may interferewith the patient’s abilitiesto decenter and logically examine negative thoughts and/or to usepleasurable activities to distract from negative mood states,thereby diminishing the effectiveness of CBT. Thus, consistent withfindings reported by Cheavens et al. (2012), seasonally depressedpatients who ruminate a lot may benefit less from a treatment thattargets that relative deficit (i.e., compensation) than from a treat-ment that seeks to bolster a relative strength (i.e., capitalization).

The current study used data from our parent randomized clin-ical trial comparing CBT, LT, and their combination for SAD (Rohanet al., 2004, 2007; Rohan et al., 2009; Rohan, Roecklein, Lacy, et al.,2009) to evaluate whether theoretically-relevant baseline cognitivevulnerabilities predict acute and next winter treatment outcomesfor SAD, in general (i.e., are prognostic factors), or differentially as afunction of treatment modality (i.e., are prescriptive factors thatmoderate treatment response). These exploratory secondary ana-lyses represent a first step in establishing an empirical basis fordetermining which patients are more suited for CBT, LT, or a com-bination treatment. We examined a broad range of baselinecognitive constructs as predictors: dysfunctional attitudes, rumi-nation, and negative automatic thoughts. The predictable seasonalpattern of SAD facilitates this study’s dual focus on cognitive pre-dictors of acute (treatment endpoint) and long-term (the nextwinter when there is a high risk for depressive symptoms in SAD)treatment outcomes. We are aware of only one nonseasonaldepression study (Thase et al., 1992) that examined a baselinecognitive vulnerability (i.e., dysfunctional attitudes) in relation tolong-term treatment (i.e., CBT) outcomes. Given that no studies ofthis nature have been conducted in SAD, this exploratory workmade no a priori predictions about whether pre-treatment cogni-tive variables would be prognostic or prescriptive of acute or long-term SAD treatment outcomes.

Method

Participants

The current study pooled data from two pilot studies testing theefficacy of CBT for SAD, our initial feasibility study (Rohan et al.,2004) and our subsequent controlled, randomized clinical trial(Rohan et al., 2007), and used next winter follow-up data from theparent trial (Rohan et al., 2009; Rohan, Roecklein, Lacy, et al., 2009).Screening and enrollment procedures were consistent across thetwo pilot studies and have been reported elsewhere (Rohan et al.,2004; 2007; Rohan et al., 2009; Rohan, Roecklein, Lacy, et al.,2009). The main difference between the two studies is that thesecond pilot’s study design included a concurrent wait-list controlgroup over the acute 6-weeks of treatment (Rohan et al., 2007),which was not included in these analyses because that cell is lessrelevant to questions about the predictive value of baseline cogni-tions for active treatment outcomes and because thewait-list group

was not assessed the next winter. For both studies, participantswere randomly assigned to conditions using a pregeneratedrandomization list using random permuted blocks, with conditionsstratified based on sex and race. The two studies were also com-parable in terms of demographic characteristics of the participants,treatment outcomes, and attrition rates at each time point. Thetreatment protocols employed in the two studies were also iden-tical, with one exception. In the second pilot study, an expert lighttherapy consultant individually-tailored the time of the day forlight administration to maximize response while reducing sideeffects. However, post-treatment outcomes for the LT group werealmost identical across the two studies. In brief, participants in bothstudies were adults, aged 18 and older, recruited from the metro-politan Washington, DC area via community advertisements. Thestudy’s methodology was approved by the Uniformed ServicesUniversity of the Health Sciences’ Institutional Review Board.

Participants in both studies met DSM-IV criteria for majordepression, recurrent, with a seasonal pattern on the StructuredClinical Interview for DSM-IV Axis I Disorders e Clinical Version(SCID; First, Spitzer, Gibbon, & Williams, 1996) with no comorbidAxis I disorder and met the threshold for a current SAD episode asassessed by the Structured Interview Guide for the Hamilton RatingScale for DepressiondSeasonal Affective Disorder Version (SIGH-SAD; Williams, Link, Rosenthal, Amira, & Terman, 1992). Potentialparticipants were excluded from both studies if they: (a) werereceiving current psychological or psychiatric treatment (i.e., psy-chotropic medications, psychotherapy, light therapy) or had im-mediate plans to initiate such treatment, (b) met diagnostic criteriaof any other current Axis I disorder, (c) had plans for major vaca-tions or absences through March, or (d) had bipolar-type SAD.

See Rohan et al. (2009) and Rohan, Roecklein, Lacy, et al. (2009)for a CONSORT flow diagram and for complete baseline de-mographic characteristics, overall and within each treatment con-dition. Across four waves of annual fall/winter recruitmentbeginning in 2000, 72 participants (24 CBT, 25 LT, 23 CBTþ LT) wereinitially randomized to active treatment. However, this includedthree feasibility study participants (one in each treatment group)who were excluded from these analyses due to antidepressantmedication at baseline. Of the 69 participants (23 CBT, 24 LT, 22CBT þ LT) included in the current analyses, 61 provided data atpost-treatment (19 CBT, 22 LT, 20 CBT þ LT), and 52 participantsprovided next winter follow-up data (17 CBT, 19 LT, 16 CBT þ LT).This indicates a 12% attrition rate over the initial 6-weeks oftreatment and an additional 13% attrition at the next winter follow-up. The majority of randomized participants were Caucasian (80%),college educated, and female (92%), with a mean age of 46.4 years.

Procedures

A detailed description of the study design and methods (e.g.,treatment protocols, treatment integrity, randomization procedure,interviewer reliability) was provided in previous reports (Rohanet al., 2007; Rohan et al., 2009; Rohan, Roecklein, Lacy, et al.,2009). The current study used data collected at the baselineassessment (immediately before randomization to treatment), at 6-weeks treatment endpoint (post-treatment), and at the next winterfollow-up in this secondary analysis to explore cognitive prognosticand prescriptive factors. In both studies, participants who provideddata at post-treatment were invited to return for a naturalisticfollow-up in January or February of the next wholly new winter(i.e., approximately 1 year after completing study treatment in theinitial winter, range ¼ 10e14 months). We selected January andFebruary as the timeframe for conducting follow-up assessmentsbecause these months are associated with the largest proportion ofindividuals with SAD in a current depressive episode (78%) relative

L. Sitnikov et al. / Behaviour Research and Therapy 51 (2013) 872e881 875

to other fall/winter months (Rosenthal, Sack, & Gillin, 1984). At pre-and post-treatment, participants completed self-report question-naires, described below, to assess potential prognostic and pre-scriptive cognitive variables relevant to the current study: theDysfunctional Attitudes Scale, Response Styles Questionnaire, andAutomatic Thoughts Questionnaire. At pre- and post-treatmentand at next winter follow-up, participants completed measuresassessing depressive symptoms (i.e., the outcome variables),described below: the Beck Depression Inventory-Second Editionand Structured Interview Guide for the Hamilton Rating Scale forDepression-Seasonal Affective Disorder Version, which wasadministered live by a trained rater, blind to treatment condition.Subsequently, a second trained, blind rater re-rated an audiotape ofeach SIGH-SAD.

In both studies, participants were randomly assigned to one ofthe following 6-week treatments: CBT (12 1.5-h group therapysessions at a frequency of twice per week), LT [two 45-min dailydoses of LT, one in the morning between 6:00 and 9:00 am and onein the evening between 6:00 and 9:00 pm, administered using a10,000-lux SunRay� light box (SunBox Company, Gaithersburg,MD)], or the combination treatment, which involved both fullprotocols concurrently. In the CBT for SAD protocol (Rohan, 2008),traditional CBTcomponents, including psychoeducation, behavioralactivation, cognitive restructuring, and relapse prevention, arepresented as strategies for improved coping with seasonal changesin weather and light availability. Some cognitive restructuringcenters on maladaptive thoughts about the seasons, light avail-ability, and weather conditions (e.g., “I’m stuck in a rut in thewinter,” “All is well if the sun is shining,” and “I can’t be productiveon dark dreary days”). CBT sessions are more frequent and longer induration than in cognitive therapy for depression (Beck et al., 1979),reflecting the need to complete treatment before spontaneousremission of symptoms in the spring.

Predictor variables: measures of potential pre-treatment cognitivepredictors and moderators of treatment effects

Dysfunctional attitudes scale (DAS)-form AThe Dysfunctional Attitudes Scale-Form A (DAS; Weissman &

Beck, 1978) was used to assess participants’ conviction in stableand global beliefs that are commonly endorsed by depressed in-dividuals. Using a 7-point scale (1 ¼ “totally agree” to 7 ¼ “totallydisagree”), respondents indicate the extent to which they agreewith 40 statements such as “If I do not do well all the time, peoplewill not respect me” and “If a person asks for help, it is a sign ofweakness.” DAS total scores range from 40 to 280 with higherscores reflecting more maladaptive beliefs and attitudes. The scalepossesses high internal consistency and testeretest reliability(Cane, Olinger, Gotlib, & Kuiper, 1986; Marton, Churchard, &Kutcher, 1993; Weissman & Beck, 1978). In the current study,Cronbach’s alpha for the DAS was .92 at pre-treatment.

Response styles questionnaire (RSQ)Participants’ tendency to use rumination and distraction stra-

tegies in response to a depressed mood was assessed via theResponse Styles Questionnaire (RSQ; Nolen-Hoeksema, personalcommunication, 1995). This 32-item self-report measure asks re-spondents to indicate the frequency with which they engage inbehaviors consistent with rumination (e.g., “Think ‘Why do I alwaysreact this way?’”) or distraction (e.g., “Do something that has madeyou feel better in the past”) when feeling depressed on a 4-pointscale, ranging from 0 (“almost never”) to 3 (“almost always”). TheRumination and Distraction subscales of the RSQ have high internalconsistency and correlate significantly with respondents’ actual useof rumination and distraction responses (Nolen-Hoeksema &

Morrow, 1991). The Cronbach’s alphas were .95 for the RSQ Rumi-nation subscale and .77 for the RSQ Distraction subscale at pre-treatment.

Automatic Thoughts Questionnaire (ATQ)The 30-item Automatic Thoughts Questionnaire (ATQ; Hollon &

Kendall, 1980) was used to assess frequency of automatic negativethoughts characteristic of depression. Participants indicate thefrequency of thoughts such as “I am so disappointed with myself”and “My future is bleak” over the past week on a 5-point scale,ranging from 1 (“not at all”) to 5 (“all the time”). The ATQ has goodreliability (i.e., split-half reliability and coefficient alpha) whenused with depressed patients (Harrell & Ryon, 1983) and possesseshigh convergent validity with other depressionmeasures (Harrell &Ryon, 1983; Hollon & Kendall, 1980; Hollon, Kendall, & Lumry,1986). Cronbach’s alpha in this study was .98 for the ATQ at pre-treatment.

Outcome variables: measures of depressive symptom severity

Structured interview guide for the Hamilton rating scale fordepression-seasonal affective disorder version (SIGH-SAD)

The Structured Interview Guide for the Hamilton Rating Scalefor Depression-Seasonal Affective Disorder Version (SIGH-SAD;Williams et al., 1992) is the most commonly used outcomemeasurein SAD treatment research. It is a semi-structured interview andconsists of two parts that are summed for a total score: the 21-itemStructured Interview Guide for the Hamilton Rating Scale forDepression (HAM-D; Williams, 1998) and a supplementary 8-itemsubscale used to assess atypical symptoms associated with SAD,such as changes in eating preferences and hypersomnia. For bothstudies, a trained rater, blind to the treatment condition, adminis-tered the SIGH-SAD at pre-treatment, post-treatment, and at nextwinter follow-up. As described in Rohan et al. (2007), the PI metwith all raters to discuss each of the 29 SIGH-SAD items and thenuances of scoring them. The raters then practiced rating audio-tapes of SIGHeSADs from past studies and reviewed their ratings ina group session lead by the PI. To become a “trained” rater on thisstudy, a trainee had to perform a mock SIGHeSAD interview on thePI (role-playing a patient) with good flow and proficiency andobtain item ratings that correspondedwell to the PI’s judgments. Toestablish inter-rater reliability, a second trained rater also blind tothe treatment condition rated an audiotape of the live SIGH-SADinterview at each assessment point. As detailed elsewhere, forboth studies, high levels of inter-rater reliability were observed forall three assessment points (Rohan et al., 2009; Rohan, Roecklein,Lacy, et al., 2009).

Beck depression inventory-second edition (BDI-II)The Beck Depression Inventory-Second Edition (BDI-II; Beck,

Steer, & Brown, 1996) is a 21-item self-report measure of depres-sive symptom severity over the past 2 weeks. Each item is rated ona 0 to 3 scale, indicting increasing severity. The BDI-II possessesgood test-retest reliability and convergent validity (Beck et al.,1996).

Data analytic strategy

All analyses were conducted using SPSS version 20 for Win-dows, except for the multiple imputation analyses (MI) which wereconducted using SAS Proc MIANALYZE. An Analysis of Variance(ANOVA) framework was used to test for baseline differences intreatment groups on each of the proposed predictors. Hierarchicalmultiple regression analyses were used to examine the associationsbetween pre-treatment cognitive characteristics and continuous

Table 1Bivariate correlations between pre-treatment cognitive measures and depressionoutcome measures.

1 2 3 4 5 6 7 8

1. RSQ Distraction e �.18 �.24 �.45b �.17 �.07 �.06 .002. RSQ Rumination �.19 e .48b .64b �.17 .20 �.17 .113. DAS �.24 .48b e .66b �.02 .36b �.11 .244. ATQ �.45b .64b .66b e �.15 .31a �.18 .165. Post-tx BDI-II �.17 �.17 �.02 �.15 e .23 .59b .216. Next winter BDI-II �.07 .20 .36b .31a .23 e �.03 .77b

7. Post-tx SIGH-SAD �.06 �.17 �.11 �.18 .59b �.03 e .138. Next winter SIGH-SAD .00 .11 .24 .16 .21 .77b .13 e

Notes. BDI-II ¼ Beck Depression Inventory-Second Edition. SIGH-SAD ¼ StructuredInterview Guide for the Hamilton Rating Scale for Depression-Seasonal AffectiveDisorder Version. DAS ¼ Dysfunctional Attitudes Scale total score. ATQ ¼ AutomaticThoughts Questionnaire total score. RSQ-Rumination and RSQ-Distraction ¼ theRumination and Distraction subscales of the Response Styles Questionnaire,respectively. Post-tx ¼ post-treatment (treatment endpoint).

a Correlation is significant at the .05 level (2-tailed).b Correlation is significant at the .01 level (2-tailed).

L. Sitnikov et al. / Behaviour Research and Therapy 51 (2013) 872e881876

depression scores (on the BDI-II and SIGH-SAD) at post-treatmentand at next winter follow-up. Separate regression models wereestimated to evaluate the effect of baseline cognitive characteristicson acute and next winter treatment outcome. Consistent with anintent-to-treat (ITT) approach, regression models were firstcomputed using all available data with multiple imputations toestimate missing values. We conducted separate ITT analyses basedon MI to derive post-treatment and winter follow-up BDI-II andSIGH-SAD scores. Imputed values for missing depression scoreswere generated from multivariate normal regression modelsderived from participants with available data. Five data sets withdifferent imputed values were generated for both the SIGH-SADand BDI-II scores and combined for the purpose of regression an-alyses via SAS PROC MIANALYZE. Analyses were then repeated in acompleter analysis including only participants who provided dataat all time points (N ¼ 52). Completer analyses were conducted toexamine consistency between completers-only and the ITT ana-lyses, however, the ITT analyses are considered primary.

Consistent with procedures outlined by Kraemer et al. (2002),each model tested for predictors and moderators of treatmentoutcome. A non-specific predictor (prognostic factor) is identifiedby a significant main effect of a predictor, whereas amoderator (prescriptive factor) is indicated by a significanttreatment � predictor interaction. Because we used an ITTapproach, separate models were used to identify significant pre-dictors and moderators of treatment outcome. For all models, themain effect of treatment was entered as two dummy-coded vari-ables. For analyses aimed at identifying predictors of treatmentoutcome, pre-treatment cognitive variable was then entered.Models aimed at identifying significant moderators of treatmentincluded lower-order terms and interaction terms comparing theeffects of the pre-treatment cognitive variable on treatmentoutcome in CBT vs. LT and in CBT þ LT vs. LT. As LT is considered thegold-standard treatment for SAD, wemade an a-priori decision thatit would serve as the comparison group for all analyses. Accord-ingly, parameter estimates for the treatment terms reflect meandifferences for LT-vs.-CBT and for LT-vs.-CBT þ LT, whereasparameter estimates for the interaction terms reflect treatmentgroup differences in the effect of the pre-treatment cognitive var-iable on outcome.

We did not include pre-treatment depression scores as a co-variate for two reasons. First, depression severity was controlled forby design (i.e., only currently depressed patients were enrolled inthe study). That is, SAD episode onset was defined as a total SIGH-SAD score of equal or greater than 20 and a HAM þ D score greateror equal to 10 þ an atypical score greater or equal to 5. Only par-ticipants meeting this criterion were randomized to treatment.Accordingly, distribution of depression scores at pre-treatment waslimited by design, such that variability in depression scores at pre-treatment was restricted, whereas the full range of depressionscores could be observed at post-treatment and next winter follow-up. Thus, including pre-treatment depression scores would haveviolated assumptions of statistical methods employed in currentstudy (e.g., homogeneity of variance, etc.). Second, the relationshipbetween pre-treatment and next winter SIGH-SAD and BDI-IIscores was different in participants receiving CBT compared tothose receiving LT wherebymore severe depression at baseline wasprescriptive of treatment outcome and was associated with moresevere depression the next winter in LT, but was associated withless severe depression the next winter in CBT (Rohan et al., 2009;Rohan, Roecklein, Lacy, et al., 2009). Accordingly, multipleimputation models were computed separately for each treatmentgroup. In addition, although pre-treatment depression severitymay interact with cognitive vulnerabilities, we did not examine athree-way interaction between treatment group � pre-treatment

depression severity � cognitive vulnerability due to insufficientpower.

All continuous predictors were centered around the grand-mean prior to calculation of higher-order interaction terms(Cohen, Cohen, West, & Aiken, 2003). Predictors/moderatorsincluded the following pre-treatment cognitive characteristics:response styles (RSQ-Rumination and RSQ-Distraction), dysfunc-tional attitudes (DAS total), and negative automatic thoughts (ATQtotal). Significant interactions were plotted and probed through theanalyses of simple slopes (Aiken & West, 1991). Due to theexploratory, hypothesis-generating, nature of our study, a 2-sidedp < .05 was used in all analyses to indicate statistical significance.

Results

Bivariate correlations between predictor and outcome variablesare presented in Table 1. Means and standard deviations for eachtreatment condition on baseline cognitive variables and depressionoutcomes are presented in Table 2. No significant treatment groupdifferences were observed at pre-treatment (Table 2), whichreplicated our previously published finding of no baseline differ-ences in depression severity between the treatment conditions(Rohan et al., 2007). Bivariate correlations were used to examineassociations between the dependent measures and cognitive pre-dictors of treatment outcome. None of the pre-treatment cognitivevariables were significantly associated with BDI-II scores at post-treatment (p > .05). At next winter follow-up, there were signifi-cant and moderate correlations between next winter BDI-II andpre-treatment DAS (r¼ .36, p< .01) and between next winter BDI-IIand pre-treatment ATQ (r ¼ .31, p < .05), such that next winterdepression severity was positively associated with endorsement ofdysfunctional attitudes and with frequency of negative automaticthoughts at baseline. In contrast, none of the cognitive predictors/moderators examined in the current study were significantlycorrelated with post-treatment or follow-up SIGH-SAD scores.

Cognitive predictors and moderators of post-treatment and nextwinter treatment outcomes

Intent-to-treat analysesResults from regression analyses are shown in Tables 3 and 4.

For models examining moderators of treatment outcome, the un-standardized b coefficient for the cognitive variable reflects theestimated effect of that variable on continuous depression scoresamong participants randomized to LT; the b estimates for the

Table 2Pre-treatment scores on cognitive measures and depression outcomes by treatment group.

Treatment group ANOVA

CBT (N ¼ 24)M (SD)

LT (N ¼ 25)M (SD)

CBT þ LT (N ¼ 23)M (SD)

F (df) p

BDI-II 27.92 (10.08) 25.64 (7.01) 23.61 (6.52) 1.69 (2, 69) .19SIGH-SAD 29.04 (6.36) 28.12 (5.49) 26.74 (5.33) .95 (2, 69) .39DAS 127.96 (27.69) 121.04 (29.85) 129.32 (28.08) .57 (2, 69) .58ATQ 72.13 (27.37) 60.00 (19.58) 66.96 (23.31) 1.50 (2, 69) .23RSQ-Rumination 27.83 (10.02) 27.35 (12.23) 28.08 (9.92) .28 (2, 69) .97RSQ-Distraction 14.35 (8.60) 14.09 (3.91) 14.24 (6.33) .01 (2, 69) .99

Notes. CBT ¼ Cognitive-behavioral therapy, LT ¼ light therapy, and CBT þ LT ¼ combination treatment of cognitive-behavioral therapy plus light therapy. See Table 1 formeasure abbreviations.

L. Sitnikov et al. / Behaviour Research and Therapy 51 (2013) 872e881 877

interaction terms represent the difference in the effect of thecognitive variable between the two treatments (i.e., LT vs. CBT aswell as LT vs. CBT þ LT). To aid interpretation, significant in-teractions were probed and graphed at mean and high and lowvalues (i.e., �1 SD above the grand mean) of the moderator of in-terest (Aiken & West, 1991).

Consistent with previously published results (Rohan et al.,2007), no treatment group differences in depression outcomeswere observed at treatment endpoint. There was not a significanteffect of any of the cognitive variables on BDI-II or SIGH-SAD scoresat the end of treatment. Neither RSQ-Rumination nor RSQ-Distraction was a predictor or moderator of acute or next winterfollow-up depression outcomes. As detailed below, both DAS andATQ scores were moderators of next winter follow-up SIGH-SADand BDI-II scores. Significant interactionswere explored graphicallypost-hoc (see Fig. 1).

With treatment type, DAS score, and the interactions of treat-ment group and DAS score in the model predicting next winterfollow-up BDI-II scores, the CBT vs. LT � DAS interaction was sig-nificant (t [31.90] ¼ �2.73, b ¼ �.18, p ¼ .01), such that participantswith higher DAS scores had lower BDI-II scores at next winterfollow-up in CBT vs. LT. Follow-up within-group simple slope an-alyses revealed that DAS total scores were predictive of the esti-mated trajectory of mean BDI-II scores at next winter follow-uponly for participants randomized to solo LT. As can be seen in Fig. 1,

Table 3ITT Analyses of Next-Winter Depression Outcomes: Predicting next winter follow-up BDI-II scores.

b S.E. t p

Intercept 11.43 1.42 9.21 <.01CBT vs. LT �6.52 1.89 �3.44 <.01CBT þ LT vs. LT �3.13 2.09 �1.50 .15Predictors of outcomeDAS .05 .03 1.52 .15ATQ .04 .04 1.02 .32RSQ-Rumination �.03 .08 �.50 .62Moderators of OutcomeDAS .15 .06 2.64 .02CBT vs. LT * DAS �.18 .07 �2.73 .01CBT þ LT vs. LT * DAS �.12 .07 �1.79 .08ATQ .14 .08 1.76 .11CBT vs. LT * ATQ �.20 .1 �1.94 .08CBT þ LT vs. LT * ATQ �.08 .1 �.8 .43RSQ-Rumination �.01 .16 �.70 .50CBT vs. LT * RSQ-Rumination �.13 .19 �.09 .93CBT þ LT vs. LT * RSQ-Rumination .02 .19 �.65 .52

Note. Multiple imputation analyses are displayed. Separate models were computedfor each predictor of interest. For predictors of outcome analyses lower order termsvalues represent effects of putative cognitive predictor averaged across the treat-ment. In the full (i.e., moderators of outcome analyses) parameter estimates forlower-order terms are from the full model and thus reflect the effect of continuouspredictors for participants randomized to LT. See Table 1 for measure abbreviations.

average and above sample mean (by > þ1 SD) levels of dysfunc-tional attitudes were associated with worse prognosis (higher BDI-II scores) at next winter follow-up for participants randomized tosolo LT. When dysfunctional attitudes were 1 SD below the samplemean, next winter BDI-II scores were estimated to be comparableacross the three treatments. That is, the effect of dysfunctional at-titudes on treatment group differences in next winter depressionoutcomes was only evident for LT participants demonstratingaverage or above average levels of dysfunctional attitudes atbaseline. A similar pattern emerged when winter follow-up SIGH-SAD scores were examined. With main effects and interactionterms in the model, the CBT vs. LT � DAS scores interaction (t[49.10] ¼ �2.06, b ¼ �.18, p ¼ .04) was statistically significant inpredicting next winter SIGH-SAD scores.

In the prediction of winter follow-up SIGH-SAD scores based onbaseline frequency of negative automatic thoughts, the modelconsisting of treatment type, pre-treatment ATQ score, the inter-action of treatment type and ATQ score, and the interaction of CBTvs. LT � ATQ score approached statistical significance (t[64.98]¼�2.04, b¼�.17, p¼ .05), indicating the same trend as wasobserved for the DAS, i.e., higher ATQ scores at baseline wereassociated with lower next winter SIGH-SAD scores in CBT than inLT. Parameter estimates in a model predicting BDI-II depressionscores at next winter follow-up based on ATQ scores were notstatistically significant.

Table 4ITT Analyses of Next-Winter Depression Outcomes: Predicting next winter follow-up SIGH-SAD scores.

b S.E. t p

Intercept 17.21 1.69 10.17 <.01CBT vs. LT �8.00 2.47 �3.24 <.01CBT þ LT vs. LT �5.30 2.67 �1.99 .05Predictors of outcomeDAS .03 .04 .71 .49ATQ �.06 .04 �1.29 .21RSQ-Rumination �.02 .11 �.20 .84Moderators of outcomeDAS .15 .06 2.24 .03CBT vs. LT * DAS �.18 .09 �2.06 .04CBT þ LT vs. LT * DAS �.17 .08 �2.04 .05ATQ .17 .11 1.57 .16CBT vs. LT * ATQ �.25 .12 �2.03 .06CBT þ LT vs. LT * ATQ �.16 .13 �1.28 .22RSQ-Rumination .20 .26 .78 .45CBT vs. LT * RSQ-Rumination �.39 .30 �1.29 .21CBT þ LT vs. LT * RSQ-Rumination �.27 .28 �.95 .36

Note. Multiple imputation analyses are displayed. Separate models were computedfor each predictor of interest. For predictors of outcome analyses lower order termsvalues represent effects of putative cognitive predictor averaged across the treat-ment. In the full (i.e., moderators of outcome analyses) parameter estimates forlower-order terms are from the full model and thus reflect the effect of continuouspredictors for participants randomized to LT. See Table 1 for measure abbreviations.

Fig. 1. Estimated next winter mean scores in each treatment for mean, high, and low levels of Cognitive Predictors at baseline.

L. Sitnikov et al. / Behaviour Research and Therapy 51 (2013) 872e881878

Completer analyses. The overall pattern of results fromcompleter analyses was similar to those from the ITT analyses.Consistent with ITT analyses, none of the cognitive variablesexamined were prognostic or prescriptive of acute treatment out-comes as indexed by BDI-II and SIGH-SAD scores. Likewise, nosignificant results were obtained examining the predictive andprognostic utility of RSQ-Rumination and RSQ-Distraction. In bothITT analyses and completer analyses, the CBT vs. LT � DAS inter-action (t [48]¼ �2.84, b ¼ �.20, p< .05) was a significant predictorof next winter BDI-II scores. In completer analyses only, DAS scores(t [46] ¼ 2.10, b ¼ .07, p < .05) and the CBT þ LT vs. LT � DASinteraction also approached statistical significance (t [48] ¼ �1.79,b ¼ �.13, p ¼ .08). Unlike in ITT analyses, the CBT vs. LT � DASinteraction was not a significant predictor of next winter SIGH-SADscores (t [45] ¼ �1.2, b ¼ �.12, p ¼ .24).

Consistent with ITT analyses, CBT vs. LT � ATQ interaction pre-dicted next winter SIGH-SAD scores. In completer analyses only, theCBT vs. LT � ATQ interaction was also a significant predictor of nextwinter BDI-II scores. That is, in completer analyses only, pre-treatment ATQ scores had differential effects on both next winterBDI-II (t [45] ¼ �2.51 b ¼ �.22, p < .05) and SIGH-SAD scores (t[45] ¼ �2.33 b ¼ �.27, p < .05) for participants randomized to CBTvs. those randomized to solo LT. Again, treatment group differencesin depression outcomeswere only observed atmean or abovemeanfrequency of negative automatic thoughts. Specifically, CBT wassuperior to LT in terms of next winter depression scores only amongparticipants demonstrating average or above average frequency ofnegative automatic thoughts at baseline, with CBT þ LT fallingbetween the two groups.

Discussion

The current study aimed to evaluate whether cognitive vulner-abilities implicated in SAD etiology are prognostic and/or pre-scriptive of acute and long-term treatment outcomes. To ourknowledge, this is the first study to evaluate potential cognitivepredictors and moderators of acute and long-term treatment out-comes in seasonal depression. We explored these questions usingdata from our parent clinical trial comparing light therapy (LT),SAD-tailored cognitive-behavioral therapy (CBT), and combinedCBT þ LT treatment in the acute treatment of SAD and at a follow-up the next winter. We examined whether baseline dysfunctionalattitudes, negative automatic thoughts, rumination, and distraction(a) were generally predictive of depression outcomes, regardless oftreatment modality, or (b) differentially influenced depressionoutcomes depending on treatment type.

We found that pre-treatment cognitive characteristics weredifferentially associated with depression outcomes the followingwinter depending on initial treatment modality. Specifically, dif-ferential treatment effects at next winter assessment were presentonly for participants demonstrating higher frequency of negativeautomatic thoughts and more entrenched dysfunctional attitudes.Participants with higher baseline cognitive vulnerability on thesemeasures had less severe depression the next winter if they wererandomized to solo CBT than if randomized to solo LT. Only par-ticipants randomized to solo LT who scored at or above the mean(by þ1 SD or more) on the cognitive measures (the DysfunctionalAttitudes Scale and the Automatic Thoughts Questionnaire) atbaseline fared worse than participants randomized to solo LT who

L. Sitnikov et al. / Behaviour Research and Therapy 51 (2013) 872e881 879

scored below the sample mean. In contrast, baseline dysfunctionalattitudes and negative automatic thoughts were not predictive ofpost-treatment or next winter depression scores in participantsrandomized to solo CBTor CBTþ LT, such that depressive symptomswere comparable at next winter follow-up for participants high andlow on these cognitive vulnerabilities who were randomized to aninitial treatment including CBT. Consistent with some previouslyreported findings in the nonseasonal depression literature (Arnowet al., 2004; Bagby & Parker, 2001), a ruminative response style wasnot associated with next winter depression outcomes.

This pattern of findings is consistent with the rationale that LTtargets physiological mechanisms of SAD, specifically a proposeddysfunctional circadian phase shift associated with seasonalchanges in photoperiod (Lewy, Sack, & Singer, 1988). Consequently,maladaptive cognitive content and processes are not directly tar-geted in LT, although cognitive measures may improve in a state-like fashion as winter depression improves over acute LT treat-ment (Evans et al., 2013). These findings suggest that if a SAD pa-tient high in cognitive vulnerability receives initial treatment withsolo LT, his or her putative cognitive vulnerability is left uncheckedto resurface with the annual recurrence of fall/winter depressivesymptoms. Although CBT was associated with better next winteroutcomes than LT in the primary efficacy analyses (Rohan et al.,2009; Rohan, Roecklein, Lacy, et al., 2009), it may be particularlyimportant to treat patients characterized by such cognitive styleswith CBT. A poor prognosis the next winter seems particularlylikely for a high cognitive vulnerability SAD patient treated initiallywith solo LT who does not resume LT or initiate another effectivetreatment in the subsequent winter. Our previously published pa-per on the next winter outcome data (Rohan et al., 2009; Rohan,Roecklein, Lacy, et al., 2009) reported that only four participants(2 solo LT, 2 CBT þ LT) reported any use of LT the next winter, andonly two of them (both in CBT þ LT) described re-initiating LT at afrequency and duration that would be expected to confer anytherapeutic benefits per the clinical practice guidelines (Lam &Levitt, 1999). Non-compliance with LT in subsequent wintersfollowing a supervised LT trial appears to be the norm rather thanthe exceptionwith only 41% of formerly treated patients continuingregular LT use (Schwartz, Brown,Wehr, & Rosenthal, 1996), perhapsdue to the considerable and consistent daily time commitment LTrequires from the patient 3e6 months of each year (Lam & Levitt,1999). Given the poor long-term compliance with LT, it seemsparticularly risky to treat SAD patients who are high in cognitivevulnerability with solo LT in terms of long-term depressionoutcomes.

The current results are also consistent with the integrativecognitive-behavioral theory of SAD, the basis for SAD-tailored CBT,which proposes maladaptive cognitions contribute to SAD onsetand maintenance (Rohan et al., 2009; Rohan, Roecklein, Lacy, et al.,2009). Components of the CBT for SAD protocol explicitly target thecognitive vulnerabilities assessed in the current study in that CBTrequires active practice of skills taught, such as cognitive restruc-turing to identify and challenge negative thought content (i.e.,automatic thoughts and dysfunctional attitudes) and plannedengagement in pleasurable activities. Patients who are high incognitive vulnerability at baseline may benefit most in the long runif they receive a treatment that directly targets these cognitivevulnerability processes such as CBT or CBT þ LT. Indeed, we re-ported that a greater degree of pre- to post-treatment improve-ment in dysfunctional attitudes and negative automatic thoughts isuniquely associated with next winter BDI-II scores in solo CBT, butnot in solo LT or combined CBT þ LT, suggesting that these con-structs might represent unique cognitive mechanisms underlyingsolo CBT that explain its enduring effects the next winter (Evanset al., 2013). Accordingly, a SAD patient who scores high on this

cognitive profile and is randomized to solo LT would not be ex-pected to reap the potential benefit of offsetting these cognitiveprocesses in future winters though ongoing use of skills learned inCBT.

In contrast to findings for automatic thoughts and dysfunctionalattitudes, ruminative response style was not predictive or pre-scriptive of treatment outcome the next winter. As conceptualizedin Beck’s cognitive model of depression (1967, 1979), automaticthoughts and dysfunctional attitudes are constructs that reflectmood-state dependent cognitive contents (i.e., surface-level andintermediate cognitions, respectively). In contrast, a ruminativeresponse style reflects a trait-like disposition towards a cognitiveprocess (Nolen-Hoeksema, 1991) that is relatively stable acrossmood states (Just & Alloy, 1997; Nolen-Hoeksema, Parker, & Larson,1994). In linewith these theories, baseline cognitive profile score asa prescriptive factor of inferior next winter outcomes in LT relativeto CBT should involve dysfunctional attitudes and negative auto-matic thoughts more so than a general tendency to ruminate. TheCBT for SAD intervention explicitly targets cognitive content (e.g.,keeping thought diaries, challenging and restructuring negativethoughts, identifying and challenging underlying core beliefs) anddoes not directly target rumination. However, it is possible thatindividuals with a ruminative response style who are randomizedto CBT may learn how to respond to negative thoughts that arisewhile ruminating (i.e., to engage in Socratic questioning) and/orlearn to distract from rumination via behavioral activation. Even ifso, individuals who have a high tendency to ruminate in responseto dysphoric mood may not lose that dispositional tendency as aresult of CBT. This interpretation and pattern of findings is consis-tent with our finding that pre- to post- change in dysfunctionalattitudes and automatic thoughts, but not in rumination, wasassociated with better next winter depression outcomes uniquelyin CBT (Evans et al., 2013).

The current study has several methodological limitations thatshould be addressed in future studies. First, the sample size wasrelatively small (N ¼ 69 randomized patients included in the ITTanalyses, with only 52 patients included in completer analyses), butattrition was relatively low and our results were comparable whenall available data with imputations vs. data for completers onlywere used. Second, as previously noted, patients were assessed atthree time points, which limited our ability to estimate morecomplex trajectories of change, conduct doseeresponse analyses,and assess impact of cognitive characteristics on change during thecourse of treatment. Third, this study focused on baseline cognitiveconstructs as predictors of treatment outcome and did not measureconstructs that are a better match to the LT rationale as predictors.Fourth, given the small sample size, the limited number ofassessment points, and the exploratory nature of these analyses;the current study did not examine more complex interactions inthe current model, such as simultaneously examining all cognitivepredictors, interactions between predictors, or interactions be-tween initial depression severity and cognitive predictors. There-fore, future studies explicitly powered to replicate and extendcurrent findings are needed. In addition, some studies of pre-scriptive factors in treatment outcome have found that stressful lifeevents interact with cognitive variables to predict differentialtreatment effects (Jones et al., 2008; Pedrelli, Feldman, Vorono,Fava, & Petersen, 2008; Simons et al., 1995). Although stressfullife events presumably play a lesser role in triggering recurrence inseasonal than in nonseasonal depression, future studies shouldmeasure stressful life events so they can be examined in interactionwith baseline cognitive variables as prescriptive predictors of SADtreatment outcomes. As another limitation, although the primaryinvestigator and a graduate student facilitated all CBT groups,which ensured consistency in the delivery of CBT, analyses did not

L. Sitnikov et al. / Behaviour Research and Therapy 51 (2013) 872e881880

account for the nested nature of our datawithin the CBT conditions.Given that both studies were pilot studies, they were notadequately powered to support analyses that account for nesteddata for participants randomized to CBT or CBT þ LT.

In conclusion, results of this exploratory investigation suggestthat SAD patients who demonstrate more problematic cognitiveprofiles at baseline appear to have a better prognosis the nextwinter following initial treatment with solo CBT than solo LT, atreatment that is palliative by design and does not address cogni-tive mechanisms central to depression onset and maintenance. Inaddition, uniquely among those treated with solo LT, highercognitive vulnerability SAD patients fared worse the next winterrelative to those with lower cognitive vulnerability. Baselinecognitive vulnerability was not associated with next winterdepression severity in CBT or CBT þ LT. If replicated, this observedpattern of prescriptive indicators of treatment outcome may beused in clinical practice to customize a first line SAD treatment to aparticular patient. That is, individuals with SAD who demonstratemore problematic cognitive styles may be particularly ill-suited forsolo LTand appear to benefit more in the long-run if initially treatedwith CBT, a modality that explicitly targets maladaptive cognitivecontent and processes. It may be particularly important for SADpatients demonstrating higher presenting levels of dysfunctionalattitudes and negative automatic thoughts to receive an initialtreatment that matches their vulnerability such as CBT.

Acknowledgment

This study was supported by grants R03 MH0659 from the Na-tional Institute of Mental Health and C072DV and C072EJ from theUniformed Services University of the Health Sciences (USUHS) toKelly J. Rohan. The funding sources had no involvement in studydesign; in the collection, analysis, and interpretation of data; in thewriting of the report; or in the decision to submit the article forpublication. This work was presented in part at the annualmeeting of the Association for Behavioral and Cognitive Therapies,November 2010, San Francisco, CA.

References

Aiken, L. S., & West, S. G. (1991). Multiple Regression: Testing and interpreting in-teractions. Newbury Park, CA: Sage.

American Psychiatric Association. (2000). Diagnostic and statistical manual of mentaldisorders (4th ed., text rev.). Washington, DC: Author.

Arnow, B. A., Spangler, D., Klein, D. N., & Burns, D. D. (2004). Rumination anddistraction among chronic depressives in treatment: a structural equationanalysis. Cognitive Therapy and Research, 28(1), 67e83.

Bagby, R. M., & Parker, J. D. A. (2001). Relation of rumination and distraction withneuroticism and extraversion in a sample of patients with major depression.Cognitive Therapy and Research, 25(1), 91e102.

Beck, A. T., Rush, A. J., Shaw, B. F., & Emery, G. (1979). Cognitive therapy of depression.New York, NY: Guilford.

Beck, A. T., Steer, R. A., & Brown, G. K. (1996). Manual for the Beck depression in-ventory-II. San Antonio, TX: Psychological Corporation.

Blazer, D. G., Kessler, R. C., & Schwartz, M. S. (1998). Epidemiology of recurrentmajor and minor depression with a seasonal pattern. The national comorbiditysurvey. British Journal of Psychiatry, 172, 164e167.

Cane, D. B., Olinger, J., Gotlib, I. H., & Kuiper, K. A. (1986). Factor structure of thedysfunctional attitude scale in a student population. Journal of Clinical Psy-chology, 42(2), 307e309.

Carter, J. D., Luty, S. E., McKenzie, J. M., Mulder, R. T., Framptom, C. M., & Joyce, P. R.(2007). Patient predictors of response to cognitive behaviour therapy andinterpersonal psychotherapy in a randomised clinical trial for depression.Journal of Affective Disorders, 128, 252e261.

Cheavens, J. S., Strunk, D. R., Lazarus, S. A., & Goldstein, L. A. (2012). The compen-sation and capitalization models: a test of two approaches to individualizingthe treatment of depression. Behaviour Research and Therapy, 50, 699e706.

Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences (3rd ed.). Mahwah, NJ: LawrenceErlbaum Associates.

Evans, M., Rohan, K. J., Sitnikov, L., Mahon, J. N., Nillni, Y. I., Tierney Lindsey, K., et al.(2013). Cognitive change across cognitive-behavioral and light therapy

treatments for seasonal affective disorder: what accounts for clinical status thenext winter? Cognitive Therapy and Research. http://dx.doi.org/10.1007/s10608-013-9561-0.

First, M. B., Spitzer, R. L., Gibbon, M., & Williams, J. B. W. (1996). Structured clinicalinterview for DSM-IV axis I disorders, clinician version (SCID-CV). Washington, DC:American Psychiatric Press, Inc.

Fournier, J. C., DeRubeis, R. J., Shelton, R. C., Hollon, S. D., Amsterdam, J. D., &Gallop, R. (2009). Prediction of response to medication and cognitive therapy inthe treatment of moderate to severe depression. Journal of Consulting andClinical Psychology, 77(4), 775e787. http://dx.doi.org/10.1037/a0015401.

Golden, R. N., Gaynes, B. N., Ekstrom, R. D., Hamer, R. M., Jacobsen, F. M., Suppes, T.,et al. (2005). The efficacy of light therapy in the treatment of mood disorders: areview and meta-analysis of the evidence. American Journal of Psychiatry, 162(4),656e662.

Hamilton, K. E., & Dobson, K. S. (2002). Cognitive therapy of depression: pretreat-ment patient predictors of outcome. Clinical Psychology Review, 22, 875e893.

Harrell, T. H., & Ryon, N. B. (1983). Cognitive-behavioral assessment of depression:clinical validation of the automatic thoughts questionnaire. Journal of Consultingand Clinical Psychology, 51, 721e725.

Hollon, S. D., & Kendall, P. C. (1980). Cognitive self-statements in depression:development of an automatic thoughts questionnaire. Cognitive Therapy andResearch, 4, 383e395.

Hollon, S. D., Kendall, P. C., & Lumry, A. (1986). Specificity of depressotypic cogni-tions in clinical depression. Journal of Abnormal Psychology, 95, 52e59.

Jacobs, R. H., Silva, S. G., Reinecke, M. A., Curry, J. F., Ginsburg, G. S., & Kratochvil, C. J.(2009). Dysfunctional attitudes scale perfectionism: a predictor and partialmediator of acute treatment outcome among clinically depressed adolescents.Journal of Clinical Child & Adolescent Psychology, 38(6), 803e813.

Jarrett, R. B., Eaves, G. G., Brannemann, B. D., & Rush, A. J. (1991). Clinical, cognitive,and demographic predictors of response to cognitive therapy for depression: apreliminary report. Psychiatry Research, 37, 245e260.

Jones, N. P., Siegle, G. J., & Thase, M. E. (2008). Effects of rumination and initialseverity on remission to cognitive therapy for depression. Cognitive TherapyResearch, 32, 591e604. http://dx.doi.org/10.1007/s10608-008-9191-0.

Just, N., & Alloy, L. B. (1997). The response styles theory of depression: tests and anextension of the theory. Journal of Abnormal Psychology, 106(2), 221e229. http://dx.doi.org/10.1037//0021-843X.106.2.221.

Keller, K. E. (1983). Dysfunctional attitudes and the cognitive therapy for depres-sion. Cognitive Therapy Research, 7, 437e444.

Kraemer, H. C., Wilson, T., Fairburn, C. G., & Agras, S. (2002). Mediators and mod-erators of treatment effects in randomized clinical trials. Archives of GeneralPsychiatry, 59, 877e883.

Kuehner, C., & Weber, I. (1999). Responses to depression in unipolar depressedpatients: an investigation of Nolen-Hoeksema’s response styles theory. Psy-chological Medicine, 29, 1323e1333.

Lam, R.W., & Levitt, A. J. (Eds.). (1999). Canadian consensus guidelines for the treatmentof seasonal affective disorder. Vancouver: Clinical & Academic Publishing.

Levitan, R. D., Rector, N. A., & Bagby, R. M. (1998). Negative attributional style inseasonal and nonseasonal depression. American Journal of Psychiatry, 155(3),428e430.

Lewy, A. J., Sack, R. L., & Singer, C. M. (1988). Winter depression and the phase shifthypothesis for bright light’s therapeutic effects: history, theory and experi-mental evidence. Journal of Biological Rhythms, 3(2), 121e134.

Magnusson, A. (2000). An overview of epidemiological studies on seasonal affectivedisorder. Acta Psychiatry Scandinavia, 101, 176e184.

Marton, P., Churchard, M., & Kutcher, S. (1993). Cognitive distortion in depressedadolescents. Journal of Psychiatry Neuroscience, 18(3), 103e107.

Nolen-Hoeksema, S. (1987). Sex differences in unipolar depression: evidence andtheory. Psychological Bulletin, 101(2), 259e282.

Nolen-Hoeksema, S., & Morrow, J. (1991). A prospective study of depression andposttraumatic stress symptoms after a natural disaster: the 1989 Loma Prietaearthquake. Journal of Personality and Social Psychology, 61(1), 115e121.

Nolen-Hoeksema, S., Parker, L., & Larson, J. (1994). Ruminative coping withdepressed mood following loss. Journal of Personality and Social Psychology, 67,92e104. http://dx.doi.org/10.1037//0022-3514.67.1.92.

Pedrelli, P., Feldman, G. C., Vorono, S., Fava, M., & Petersen, T. (2008). Dysfunctionalattitudes and perceived stress predict depressive symptoms severity followingantidepressant treatment in patients with chronic depression. PsychiatryResearch, 161, 302e308.

Peselow, E. D., Robins, C., Block, P., Barouch, F., & Fieve, R. R. (1990). Dysfunctionalattitudes in depressed patients before and after clinical treatment and innormal control subjects. American Journal of Psychiatry, 147, 439e444.

Rohan, K. J. (2008). Coping with the seasons: A cognitive-behavioral approach toseasonal affective disorder, therapist guide. New York, NY: Oxford UniversityPress.

Rohan, K. J., Roecklein, K. A., & Haaga, D. A. F. (2009). Biological and psychologicalmechanisms of seasonal affective disorder: a review and integration. CurrentPsychiatry Reviews, 5, 37e47.

Rohan, K. J., Roecklein, K. A., Lacy, T. J., & Vacek, P. M. (2009). Winter depressionrecurrence one year after cognitive-behavioral therapy, light therapy, andcombination treatment. Behavior Therapy, 40, 225e238.

Rohan, K. J., Roecklein, K. A., Tierney Lindsey, K., Johnson, L. G., Lippy, R. D., Lacy, T. J.,et al. (2007). A randomized controlled trial of cognitive-behavioral therapy,light therapy, and their combination for seasonal affective disorder. Journal ofConsulting and Clinical Psychology, 75(3), 489e500.

L. Sitnikov et al. / Behaviour Research and Therapy 51 (2013) 872e881 881

Rohan, K. J., Sigmon, S. T., & Dorhofer, D. M. (2003). Cognitive-behavioral factorsin seasonal affective disorder. Journal of Consulting and Clinical Psychology, 71,22e30.

Rohan, K. J., Tierney Lindsey, K., Roecklein, K. A., & Lacy, T. A. (2004). Cognitivebehavioral therapy, light therapy, and their combination in treating seasonalaffective disorder. Journal of Affective Disorders, 80, 273e283.

Rosenthal, N. E., Sack, D. A., & Gillin, J. C. (1984). Seasonal affective disorder: adescription of the syndrome and preliminary findings with light therapy. Ar-chives of General Psychiatry, 41, 72e80.

Schlager, D., Froom, J., & Jaffe, A. (1995). Winter depression and functionalimpairment among ambulatory primary care patients. Comprehensive Psychia-try, 36(1), 18e24.

Schmaling, K. B., Dimidjian, S., Katon, W., & Sullivan, M. (2002). Response stylesamong patients with minor depression and dysthymia in primary care. Journalof Abnormal Psychiatry, 111(2), 350e356.

Schwartz, P. J., Brown, C., Wehr, T. A., & Rosenthal, N. E. (1996). Winter seasonalaffective disorder: a follow-up study of the first 59 patients of the nationalinstitute of mental health seasonal studies program. American Journal of Psy-chiatry, 153(8), 1028e1036.

Simon, G. E., & Perlis, R. H. (2010). Personalized medicine for depression: can wematch patients with treatments? American Journal of Psychiatry, 167, 1445e1455.

Simons, A. D., Gordon, J. S., Monroe, S. M., & Thase, M. E. (1995). Toward an inte-gration of psychological, social, and biologic factors in depression: effects onoutcome and course of cognitive therapy. Journal of Consulting and ClinicalPsychology, 63, 369e377.

Sotsky, S. M., Glass, D. R., Shea, M. T., Pilkonis, P. A., Collins, J. F., Elkins, I., et al.(1991). Patient predictors of response to psychotherapy and pharmacotherapy:findings in the NIMH treatment of depression collaborative research program.American Journal of Psychiatry, 148, 997e1008.

Terman, M., Terman, J. S., & Quitkin, F. M. (1989). Light therapy for seasonal affectivedisorder: a review of the efficacy. Neuropsychopharmacology, 2, 1e22.

Thase, M. E., Simons, A., McGeary, J., Cahalane, J. F., Hughes, C., Harden, T., et al.(1992). Relapse after cognitive behavior therapy of depression: potential im-plications for longer courses of treatment. American Journal of Psychiatry, 149,1046e1052.

Weissman, A. N., & Beck, A. T. (1978, November). Development and validation of thedysfunctional attitudes scale: A preliminary investigation. Toronto, Ontario, Can-ada: American Educational Research Association.

Williams, J. B. W. (1998). A structured clinical interview guide for the Hamiltondepression rating scale. Archives of General Psychiatry, 45, 742e747.

Williams, J. B., Link, M. J., Rosenthal, N. E., Amira, L., & Terman, M. (1992). Structuredinterview guide for the Hamilton depression rating scale e Seasonal affectivedisorder version (SIGH-SAD). New York, NY: New York State Psychiatric Institute.