Sudden Gains in Cognitive-Behavioral Therapy for Seasonal Affective Disorder

Abstract

Sudden gains are associated with better acute and follow-up outcomes in cognitive-behavioral therapy (CBT) for mood and anxiety disorders. Using the group CBT for seasonal affective disorder (CBT-SAD) condition (n=88) from a parent randomized clinical trial, this study examines whether sudden gains can be detected in CBT-SAD and whether they relate to outcome at post-treatment and at follow-ups one and two winters later. Applying Kelly et al.’s (2005) criteria to week-to-week symptom changes [i.e., scores on the Structured Interview Guide for the Hamilton Rating Scale for Depression—Seasonal Affective Disorder Version (SIGH-SAD)] with SIGH-SAD score drop ≥ 6, 44.3% (39/88) of CBT-SAD participants had a sudden gain. Most sudden gains (25/39, 64.1%) occurred between pre-treatment and Week 3. Sudden gainers had lower depression severity on the Beck Depression Inventory-Second Edition (BDI-II) and a greater likelihood of remission (BDI-II ≤ 8) at both winter follow-ups than non-gainers. Sudden gain groups did not differ on post-treatment depression outcomes. Relative to non-gainers, sudden gainers showed greater improvement in seasonal beliefs, CBT-SAD’s potential mechanism, during CBT-SAD, with more flexible seasonal beliefs than non-gainers at post-treatment. Membership in a particular CBT-SAD group was not related to the effect of sudden gains on treatment outcome.

Winter seasonal affective disorder (SAD) is characterized by recurrent major depressive episodes in the fall or winter that remit in the spring (Rosenthal et al., 1984). Light therapy is the most studied treatment for SAD. However, group cognitive-behavioral therapy for SAD (CBT-SAD; Rohan, 2008) is comparably effective to light therapy in treating an acute SAD episode (Rohan et al., 2007; 2015) and is superior to light therapy on outcomes following treatment (i.e., fewer depression recurrences and less severe symptoms in future winters; Rohan et al., 2009; Rohan et al., 2016a). Clinical trials comparing CBT-SAD and light therapy have used the experimental therapeutics approach to test mediators of treatment effects.

Change towards more flexible seasonal beliefs (i.e., thoughts about the seasons, light availability, and weather) during treatment mediated improvement in depression symptoms during CBT-SAD, but not during light therapy (Rohan et al., 2020). In addition, greater change towards more flexible seasonal beliefs during treatment was associated with lower depression severity one and two winters following treatment and explained lower depression severity following CBT-SAD relative to light therapy (Rohan et al., 2023). These data provide compelling support for the cognitive mediation hypothesis in CBT-SAD, i.e., that CBT-SAD reduces depression symptoms (and has sustained effects on depression symptoms after treatment) via modifying cognitions, specifically seasonal beliefs.

Beyond seasonal beliefs, which are potentially mechanistic of outcome in CBT-SAD, sudden gains represent a potential marker of benefit from CBT-SAD. Sudden gains are large between-session symptom improvements (Tang & DeRubeis, 1999). In the original study on sudden gains, individuals who achieved sudden gains during cognitive-behavioral therapy (CBT) for depression had significantly lower depression scores at post-treatment and at follow-up 18 months later than individuals without sudden gains (Tang & DeRubeis, 1999). In addition, the proportion in “recovery” at post-treatment, defined as Beck Depression Inventory (BDI) score < 10, was significantly larger in the sudden gains vs. no sudden gains group (79% vs. 41%; Tang & DeRubeis, 1999). Since then, sudden gains have been widely studied in CBT and non-CBT psychotherapy treatments for depression, anxiety disorders, and other mental health problems. Meta-analyses suggest that sudden gains have significant, moderate-magnitude effects on primary outcome measures at treatment endpoint (Hedges’ gs = 0.62 and 0.68) and follow-up (Hedges’ gs = 0.56 and 0.61) in Aderka et al. (2012) and Shalom and Aderka (2020), respectively.

Tang and DeRubeis (1999) developed the original criteria for detecting sudden gains, which remain the most commonly adopted (Aderka & Shalom, 2021): (a) the gain was greater than or equal to a drop of 7 BDI points compared to the prior session, (b) the gain was at least 25% of the score before the gain, and (c) the mean BDI score of the three sessions prior to the gain was significantly higher than the mean score of the three sessions following the gain using a two-sample t test. These criteria are meant to capture a between-session drop in symptom severity that is large in both absolute terms (criterion a) and relative to the individual’s pre-gain symptom levels (criterion b) and that is not better accounted for by within-person symptom fluctuations (criterion c). The magnitude of 7 BDI points was determined to be a reliable change and an unusually large drop between sessions as 7 represents about half the total mean improvement in BDI scores across all CBT sessions and corresponds to the standard deviation for BDI scores among depressed patients (Tang et al., 2007).

Over the past 20+ years, Tang and DeRubeis’ (1999) sudden gains criteria have been applied to other measures beyond the BDI, with the cut-off point for a large magnitude drop in score (criterion a) typically based on the scale’s reliable change index (RCI; Jacobson & Truax, 1991). Shalom and Aderka’s (2020) meta-analysis included 50 studies and 57 psychotherapy conditions that adopted Tang and DeRubeis’ (1999) first two criteria, with the majority 37/57 (65%) adopting all three criteria and the rest (20/57; 35%) adopting an alternate to their third criterion. Strict application of Tang and DeRubeis’ (1999) third criterion precludes detecting very early (i.e., occurring between pre-treatment and session one) and very late (i.e., occurring between the last session and post-treatment) sudden gains. Example alternate stability criteria include a gain ≥ 1.5 SD from the individual’s mean score across all treatment sessions or maintenance of at least 50% of the gain for the next two sessions, both of which allow for gains to occur as early as between the first two sessions. The effect of sudden gains on post-treatment outcome was smaller among studies using all three original criteria (Hedges’s g = 0.63) compared to studies using altered stability criteria (Hedges’s g = 0.72; Shalom & Aderka, 2020).

Kelly, Roberts, and Ciesla (2005) were the first to define the third (stability) criterion as an improvement in score that was at least 1.5 SD higher than the individual’s mean score across all sessions, which was intended to capture a clinically meaningful improvement beyond normal within-individual variability in scores and allow for detection of sudden gains across all weeks of treatment. This modification is especially helpful for studies of short-duration treatments or studies with a limited number of observations in their assessment schedule. The original Tang and DeRubeis (1999) stability criterion limits the number of observations that can be used to detect sudden gains to those having at least two sessions before and two sessions after the gain.

The processes underlying the beneficial effects of sudden gains on treatment outcome are not fully understood. In CBT for depression, sudden gains have been examined in the context of cognitive mediation theory (i.e., that cognitive change leads to symptom improvement). This tradition dates to Tang and DeRubeis’ (1999) original study on sudden gains, which found evidence of greater in-session cognitive change during the session preceding the gain relative to control sessions that occurred earlier. Aderka and Shalom’s (2021) review of the broader sudden gains literature concluded that evidence was mixed regarding whether cognitive changes are associated with sudden gains based on studies that compared sessions associated with sudden gains to earlier control sessions. The authors noted that methodological differences between studies may have contributed to the mixed findings, such as self-report or observer ratings of cognitive change, the cognitive measure’s reliability, assessment of cognitive changes before or during the session when a gain occurred, and adopting original or altered sudden gains criteria. Additionally, none of the reviewed studies used cognitive changes as a predictor of later sudden gains (Aderka & Shalom, 2021). Both studies focusing on CBT for depression found greater cognitive change in the pregain session relative to the pre-pregain session (Tang et al., 1999, 2005).

The present study contributes to the literature regarding the impact of sudden gains on cognitive processes and extends the sudden gains literature to a new treatment: CBT-SAD. The study is a secondary analysis of the CBT-SAD condition within a randomized clinical trial comparing CBT-SAD and light therapy during 6-weeks of initial treatment and at follow-ups one and two winters later. First, this study determined whether the phenomenon of sudden gains generalizes to CBT-SAD. Challenges to detecting sudden gains in CBT-SAD include the treatment format (12 sessions delivered at a rate of two sessions/week over 6 weeks) and weekly outcome assessment rather than at each session. As precedent for this, Vittengl, Clark, and Jarrett (2005) examined sudden gains in a dataset where CBT for depression was administered in twice-weekly sessions but outcome was assessed weekly (every-other session) using the BDI. At treatment endpoint, participants with sudden gains had a higher likelihood of treatment response (post-treatment Hamilton Rating Scale for Depression score ≤ 9) but did not differ on depression scores, compared to those without sudden gains (Vittengl et al., 2005). Over 2 years of follow-up, participants with sudden gains during acute treatment had significantly lower depression scores, but did not differ on relapse/recurrence rates, compared to those without sudden gains (Vittengl et al., 2005). Tang et al. (2007) found that participants with sudden gains had lower relapse rates following CBT for depression than those without sudden gains when sudden gains were defined according to the original criteria but not when sudden gains were defined according to Vittengl et al.’s (2005) criteria (i.e., using BDI scores at every-other- session). It is possible that measuring scores weekly, rather than after every session, might miss large changes in scores that would be averaged out into one score. It remains to be seen if this is problematic for studying sudden gains in the context of CBT-SAD.

As another challenge, operational criteria for sudden gains in CBT-SAD require modification of Tang and DeRubeis’ (1999) third (stability) criterion because, with only 6 outcome assessments, it is not possible to compare outcome 3 weeks before vs. 3 weeks after a gain. This study articulated sudden gains criteria that accommodate CBT-SAD’s brief, intensive treatment phase and the weekly outcome measure used in the parent study, the Structured Interview Guide for the Hamilton Rating Scale for Depression—Seasonal Affective Disorder Version (SIGH-SAD). To characterize sudden gains in CBT-SAD, this study reported the proportion of CBT-SAD participants with sudden gains (and with reversals of sudden gains, defined by an increase of the BDI score by at least 50% of the sudden gain drop after the sudden gain), the average magnitude of a sudden gain, and the median treatment week for a sudden gain. This was examined using modified Tang and DeRubeis (1999) criteria, as well as using Kelly et al.’s (2005) criteria defining a gain as ≥ 1.5 SD from the individual’s mean score across all treatment sessions and allowing for gains as early as between the first and second assessment and as late as between the last two assessments. The distributions of sudden gains under each set of criteria were examined to select the sudden gains predictor for outcome analyses.

Second, this study examined whether sudden gains are associated with CBT-SAD outcome at post-treatment and at follow-ups one winter and two winters later using depression scores and dichotomous outcome measures (i.e., remission at post-treatment and recurrence at follow-up). Consistent with the broader literature, it was hypothesized that CBT-SAD participants who achieve sudden gains would have better outcomes than those without sudden gains at treatment endpoint and at follow-up. Third, this study explored a cognitive process in relation to sudden gains in testing whether CBT-SAD participants with sudden gains show greater improvement in seasonal beliefs during treatment than those without gains. As seasonal beliefs mediate outcome in CBT-SAD (Rohan et al., 2020, 2023), it was hypothesized that CBT-SAD participants with sudden gains would show greater improvement in seasonal beliefs over treatment (i.e., more flexible seasonal beliefs at mid- and post-treatment) than those without sudden gains.

Fourth, as CBT-SAD is a group therapy intervention, this study explored whether group membership influences sudden gains. To our knowledge, a few prior studies have examined sudden gains within group therapy interventions (Kelly et al., 2005, Clerkin et al., 2008; Norton et al., 2010; Nogueira-Arjona et al., 2017; Thorisdottir et al., 2018); however, they did not consider within-group dependency in their analyses and treated individual subjects as independent observations. This is potentially problematic due to group member interactions. According to Baldwin et al. (2005), individual group member’s outcomes may be dependent on group characteristics. For example, groups can have cohesion, potentially allowing peer pressure and encouragement to positively affect individual outcomes. Conversely, lack of cohesion can result in individual discouragement and lead to worse outcomes. Treating the individual outcomes as independent from the group might lead to a confounding of treatment effects with secondary effects from the group (Baldwin et al., 2005). A meta-analysis of group interaction effects (Kivlighan et al., 2020) found that group membership effects account for about 6% of the variance in treatment effects, a sizeable effect that should be taken into account in group therapy research. Therefore, the current study examined whether group membership accounted for any participant-level sudden gain effect on treatment outcome.

Methods

This study is a secondary analysis based on a parent randomized clinical trial conducted at the University of Vermont, comparing CBT-SAD to light therapy during 6-weeks of acute treatment and at follow-ups one and two winters later. The parent study was approved by the University of Vermont Committees on Human Subjects Research (protocol 08–116, effective 12/12/2007). Privacy rights of human subjects were observed, and informed consent was obtained. This study explored sudden gains in the CBT-SAD arm of the study (n = 88 of 177 randomized participants). The parent study was approved by the University of Vermont’s Institutional Review Board. The full study protocol is archived as Rohan et al. (2013). Participant baseline characteristics by treatment condition (Rohan et al., 2015) and CONSORT flow diagrams from recruitment to treatment endpoint (Rohan et al., 2015a) and over follow-up (Rohan et al., 2016a) are published.

Participants

Participants were community adults, aged 18 and older, with a primary diagnosis of recurrent major depressive disorder with seasonal pattern on the Structured Clinical Interview for DSM-IV Axis I Disorders—Clinician Version (SCID-CV; First et al., 2002) and in a current winter depressive episode, defined as a total score ≥ 20 and an atypical subscale score ≥ 5 on the Structural Interview Guide for the Hamilton Rating Scale for Depression—Seasonal Affective Disorder Version (SIGH-SAD; Williams et al., 1988). Exclusion criteria included prior treatment with light therapy or CBT for SAD; current psychotherapy for depression; plans for new treatment during the winter of enrollment; history of psychotic or bipolar disorder; current substance use disorder; serious suicidal intent; evidence of current hypothyroidism on thyroid panel at medical screening; or plans to travel outside of the region between December and March of the treatment period. Stable antidepressant medication use (i.e., stable dose for ≥ 4 weeks with no plans to adjust) was allowed. Regarding baseline characteristics for the 88 CBT-SAD participants included in this secondary analysis, 83 (94%) were White and non-Hispanic, 74 (84%) identified as female, 53 (60%) were married or living with a partner, 58 (66%) had an undergraduate degree or higher, 66 (75%) were employed full time, 24 (27%) a comorbid psychological disorder, and 24 (27%) were taking antidepressant medication. The mean age of CBT-SAD participants was 44.4 years (SD = 12.9).

Cognitive-Behavioral Therapy for Seasonal Affective Disorder (CBT-SAD) Treatment

CBT-SAD (Rohan, 2008) was tailored based on traditional CBT for depression. Like traditional CBT, CBT-SAD includes behavioral activation involving scheduling and completing pleasant activities, but the focus is on counteracting the tendency to “hibernate” and identifying activities that are available and enjoyable in the winter. Also, like traditional CBT, CBT-SAD incorporates cognitive restructuring around learning to identify and reframe automatic thoughts and core beliefs that maintain depression, but some of this work in CBT-SAD specifically focuses on negative thoughts about the winter season, light availability, and weather. Participants are encouraged to be proactive in using CBT-SAD skills in future fall/winter seasons, particularly when they notice early onset symptoms or cues that the environment is changing (e.g., photoperiod getting shorter, fall foliage). CBT-SAD is delivered in a small group format at a rate of two 90-minute sessions per week over 6 weeks for a total of 12 sessions. Each CBT-SAD group was led by a licensed doctoral-level clinician and a student co-facilitator (Rohan et al., 2013). There were a total of 16 CBT-SAD groups in this study, with an average of 5.5 participants per group. The parent study reported good compliance with the CBT-SAD manual (Rohan et al., 2015a).

Measures

Structured Interview Guide for the Hamilton Rating Scale for Depression—Seasonal Affective Disorder Version (SIGH-SAD).

The SIGH-SAD (Williams et al., 1988) is the most common measure of SAD symptoms in research. It is a 29-item measure comprised of two main subscales, the 21-item Hamilton Rating Scale for Depression (HAM-D; Williams, 1988) and 8 items that assess atypical SAD symptoms like fatigue and hypersomnia. It was administered at pre-treatment, weekly during CBT-SAD, at post-treatment, and at each winter follow-up by a rater who was blinded to treatment condition. Inter-rater reliability was good when comparing the original rater to a second blinded rater who listened to an audio-recording of the live interview across all study timepoints (see Rohan et al., 2016). SIGH-SAD scores at post-treatment and at each follow-up constituted outcomes. Dichotomous SIGH-SAD outcomes included remission status at post-treatment and at each follow-up (≥ 50% improvement in SIGH-SAD from pre- to post-treatment, with HAM-D score ≤ 7 and atypical score ≤ 7, OR HAM-D score ≤ 2 and atypical score ≤ 10 at post-treatment) and recurrence status at each follow-up (total score ≥ 20 + HAM-D score ≥ 10 + atypical score ≥ 5).

Beck Depression Inventory-Second Edition (BDI-II).

Scores on the BDI-II (Beck et al., 1996), a 21-item self-report measure, at post-treatment and at first and second winter follow-up constituted a secondary measure of depression symptom severity. Questions on the BDI-II are rated on a scale from 0 to 3, with higher scores indicating greater severity over the past two weeks. The measure has good convergent validity and adequate test-retest reliability (Beck et al., 1996). Remission at post-treatment or follow-up is defined here as BDI-II score ≤ 8.

Seasonal Beliefs Questionnaire (SBQ).

The SBQ (Rohan et al., 2019) consists of 26 items that express attitudes and beliefs about the seasons, light availability, and weather conditions. Each item is rated on a 1 (totally disagree) to 7 (totally agree) Likert scale to capture the flexibility of the individual’s seasonal beliefs. Factor analysis supports five subscales of the SBQ: Personal Deficiency Related to Winter (e.g. “I’m ineffective in the winter”), Global Summer Positivism (e.g. “I’m always happier when it’s warmer”), Global Winter Negativity (e.g. “I’m not a winter person”), Effects from Lack of Light (e.g. “Dark, gloomy days are depressing”), and Lack of Perceived Seasonal or Weather Effect (e.g. “The weather is irrelevant to how I feel”). The SBQ has good psychometric properties including internal consistency, convergent and divergent validity, and 2-week test-retest reliability (Rohan et al., 2019). SBQ score has excellent predictive ability for distinguishing between known SAD cases and healthy controls and adequate predictive ability for distinguishing between individuals with SAD and individuals with nonseasonal Major Depression (Rohan et al., 2021). This study used the SBQ as administered at pre-, mid-, and post-treatment in the parent study.

Sudden Gain Criteria

Given that sudden gains have not been examined in CBT-SAD, this study explored the distribution of sudden gains using two different sets of criteria. The first set represented a modification of the original Tang and DeRubeis (1999) criteria. Aside from using a different outcome (SIGH-SAD vs. BDI), the modification from the original criteria was in criterion c. Tang and DeRubeis’ original (1999) criterion (c) tested for a statistically significant difference between the three points before and after the interval being evaluated for sudden gain. In this study’s dataset, there were only two intervals with three observations before and after (i.e., Week 2 to Week 3 and Week 3 to Week 4), a very small number of intervals within which to detect sudden gains. Therefore, we allowed for sudden gains at intervals with at least two observations before and after, extending the number of intervals to four, including Week 1 to Week 2 and Week 4 to Week 5.

Second, this study explored application of Kelly et al.’s (2005) sudden gains criteria, which are the same as Tang and DeRubeis (1999) for criteria (a) and (b) but modify criterion (c) to a gain ≥ 1.5 SD from the individual’s mean score across all timepoints, allowing for early and late sudden gains. Under Kelly et al.’s (2005) criteria, there are six intervals within which to detect sudden gains in CBT-SAD, adding two early gain intervals (i.e., pre-treatment to Week 1 and Week 1 to Week 2) and one late gain interval (Week 5 to post-treatment) beyond those above.

Data Analysis Plan

Analyses were conducted using all available data, with missing SIGH-SAD and BDI-II scores generated through multiple imputation using SPSS 27. At each timepoint, missing scores were minimal; at post-treatment 84/88 participants provided data, and 85/88 and 83/88 provided data at the first and second winter follow-ups, respectively. Imputed data were generated using multivariate normal regression models using baseline participant characteristics, such as age, sex, depression scores, and comorbid diagnostic status. Fifty data sets with imputed SIGH-SAD and BDI-II scores were generated in this manner; these were then combined in order to conduct subsequent analyses.

Based on Wiedemann et al. (2020), the magnitude for a SIGH-SAD score drop (criterion a) was determined as reliable change based on the SD of the pre-treatment SIGH-SAD and Cronbach’s alpha for the 7 scores from pre- to post-treatment in this sample. Using the two sets of sudden gain criteria (based on Tang & DeRubeis, 1999 and Kelly et al., 2005), the frequency of participants with a sudden gain versus those without was computed. A larger proportion with sudden gains was expected under the Kelly et al. (2005) criteria, given they allow six intervals [vs. only four in under the modified Tang & DeRubeis (1999) criteria]. However, variability in sudden gains status is necessary to conduct outcome analysis. In the original studies, sudden gains were detected in 13/31 (42%; Kelly et al., 2005) and 24/61 (39%; Tang and DeRubeis, 1999) of participants in CBT for depression. The sudden gains criteria yielding the largest proportion of participants with sudden gains, not to exceed 50% of the sample with gains, was selected for outcome analysis.

The mean magnitude and SD of a sudden gain in SIGH-SAD points and the frequency of sudden gains among participants with sudden gains were computed. Descriptive statistics were generated for SIGH-SAD scores from pre-treatment through the 6-weeks of CBT-SAD for the sudden gain vs. non-gain groups. The frequencies of sudden gains at each interval provided the median interval for when a sudden gain occurred. We also examined the frequency of sudden gain reversals, defined as an increase in SIGH-SAD points of at least 50% between the week when the gain occurred and the following week (Kelly et al., 2005). Additionally, gainers and nongainers were compared on demographic and baseline characteristics [i.e., age, gender¹, race/ethnicity (White/non-Hispanic vs. all others), antidepressant medication status at baseline, comorbidity status]. Sudden gain groups were compared using a t-test on the continuous variable (age) and using a chi-square test for the remaining (categorical) variables.

Participants with a sudden gain vs. those without were compared on post-treatment depression scores (on the SIGH-SAD and BDI-II) while controlling for pre-treatment depression scores using ANCOVA. This comparison was then repeated on follow-up depression scores, separately, for the first and second winter of follow-up. For dichotomous outcomes (remission status at post-treatment, recurrence status at each follow-up), logistic regressions were used to predict outcome from sudden gain status, controlling for pre-treatment depression severity.

To examine whether sudden gain status is related to change in seasonal beliefs, a 2 (sudden gains status: present vs. absent) × 3 (timepoint: pre-, mid-, and post-treatment) repeated measures ANOVA was conducted on SBQ scores. A Sudden Gains × Time interaction was expected with participants with sudden gains having lower SBQ scores at mid- and post-treatment than those without and sudden gainers showing greater change in SBQ over time than non-gainers. The a priori plan was to graph significant interactions and conduct tests of simple main effects to test for sudden gain group differences at each timepoint and time differences within each group. For any significant ANCOVA comparing participants with vs. without a sudden gain on depression scores, secondary analyses explored whether group membership accounted for the apparent effect of participant-level sudden gains on treatment outcome. Specifically, linear mixed model analyses tested the effect of group membership on the relationship between participants’ sudden gain status and outcome. The model used the groups as the subject, sudden gain status as the fixed effect, and the outcome (depression) scores as the target variable, while controlling for pre-treatment scores.

Results

Detecting Sudden Gains in CBT-SAD: Criterion Selection and Descriptive Statistics

The magnitude for a SIGH-SAD score drop (criterion a) was determined to be a drop of 6 or more points between observations using the method outlined in Wiedemann et al. (2020) to calculate the reliable change index. This value was determined by first calculating the standard error of measurement (SE_M) using the standard deviation of the pre-treatment SIGH-SAD scores (SD = 5.47) and the Cronbach’s alpha for the 7 scores from pre- to post-treatment (α = 0.85), as seen in (1).

$${SE}_M=SD\sqrt{1-\alpha }$$ (1)

This yielded a SE_M value of 2.11. The standard error of the difference (S_DIFF) in scores was then calculated using (2).

$$S_{\mathit{\text{DIFF}}}=\sqrt{{2\left({SE}_M\right)}^2}$$ (2)

This S_DIFF value of 2.99 was multiplied by 1.96 to get a reliable change value of 5.87, rounded to 6 SIGH-SAD points.

Using the SIGH-SAD score drop ≥ 6, the proportion of sudden gainers across CBT-SAD sessions was 27.3% (24/88) using the modified Tang and DeRubeis (1999) criteria and 44.3% (39/88) using the Kelly et al. (2005) criteria. Gains detected using the Kelly et al. (2005) criteria included 23/24 (95.8%) of the gains detected using the Tang and DeRubeis (1999) criteria. Consistent with the a priori decision-making rule, Kelly et al.’s (2005) criteria were selected for the analyses, given that they detected a larger proportion of gains (while not exceeding 50% of the sample).

Frequencies of sudden gains at each interval are presented in Table 1. Examining the pattern of sudden gains across treatment indicated that more than half of all sudden gains (21/39, 53.8%) occurred between pre-treatment and Week 2. Of all sudden gains, one-third (13/39; 33.3%) occurred between pre-treatment and Week 1, and about one-fifth (8/39; 20.5%) occurred between Weeks 1 and 2. The subsequent weeks saw relatively fewer occurrences of sudden gains, ranging from three gains between Weeks 4 and 5 to six gains between Week 5 and post-treatment. The median interval for when a sudden gain occurred was between Weeks 1 and 2. The mean magnitude drop in SIGH-SAD points for a sudden gain was 14.6 with a standard deviation of 5.5. Similarly, Tang and DeRubeis reported a mean magnitude drop of 11.2 BDI points among gainers, well beyond their threshold of 7. This is expected, given that criterion (c) calls for a drop that is at least 1.5 standard deviations larger than the mean score across all time points. Based on criteria for sudden gain reversals (Kelly et al., 2005), only four participants had sudden gain reversals, indicating that only 10.3% (4/39) of sudden gains were reversed. Furthermore, there were no significant differences between gainers versus non-gainers on age, gender, race/ethnicity, comorbid diagnosis status, and baseline antidepressant medication status.

Table 1.

Frequencies of Sudden Gains Across Treatment

Interval	No. of Sudden Gains
PreTx-Week 1	13
Week 1-Week 2	8
Week 2-Week 3	4
Week 3-Week 4	5
Week 4-Week 5	3
Week5-PostTx	6

Note. PreTx = pre-treatment; PostTx = post-treatment

Differences in CBT-SAD Outcome based on Sudden Gain Status

Table 2 presents descriptive statistics by sudden gain status for depression scores on the SIGH-SAD and BDI-II at pre- and post-treatment and at each winter follow-up. Figure 1 presents weekly SIGH-SAD scores over the course of CBT-SAD in the sudden gain vs. no gain groups. In the ANCOVA on post-treatment SIGH-SAD scores, controlling for pre-treatment SIGH-SAD scores, the effect of sudden gain status was ns (F(1,86) = 0.262, p = .610), indicating no significant difference between sudden gainers and non-gainers on depression severity as measured by the SIGH-SAD at treatment endpoint. Similarly, the ANCOVA on post-treatment BDI-II scores, controlling for pre-treatment BDI-II scores, revealed no significant difference between participants with versus without sudden gains on depression severity, as measured by the BDI-II at treatment endpoint, F(1,86) = 0.361, p = .550.

Table 2.

Depression Scores by Sudden Gain Status at Pre- and Post-Treatment and Follow-Ups

	Sudden Gain	No Sudden Gain
	M (SD)	M (SD)
Pre-Treatment
SIGH-SAD	28.76 (5.78)	27.76 (4.85)
BDI-II	24.71 (9.96)	20.94 (8.58)
Post-Treatment
SIGH-SAD	13.63 (7.28)	12.38 (7.48)
BDI-II	8.32 (7.49)	8.16 (6.09)
Winter One Follow-up
SIGH-SAD	14.68 (9.44)	15.55 (8.76)
BDI-II	7.03 (7.74)	9.28 (7.72)
Winter Two Follow-up
SIGH-SAD	13.78 (7.52)	16.39 (9.26)
BDI-II	6.35 (5.65)	9.11 (7.69)

Note. M = mean; SD = standard deviation. SIGH-SAD = Structured Interview Guide for the Hamilton Rating Scale for Depression-Seasonal Affective Disorder Version. BDI-II = Beck Depression Inventory-Second Edition.

Figure 1.

SIGH-SAD Scores Across Treatment by Sudden Gain Status

Note. See Table 2 for measure abbreviations. Error bars shown at +/− 1 SE. Pre-Tx = pre-treatment, Post-Tx = post-treatment.

In the ANCOVAs on follow-up SIGH-SAD scores, controlling for pre-treatment SIGH-SAD scores, there were no significant differences between those who did and did not have a sudden gain in SIGH-SAD scores at either the first (F(1,86) = 1.08, p = .301) or the second (F(1,86) = 3.40, p = .069) winter of follow-up. However, there was a significant difference in BDI-II scores by sudden gain status at winter one (F(1,86) = 6.19, p = .015) and at winter two (F(1,86) = 8.00 p = .006). At winters one and two, sudden gainers had significantly lower mean BDI-II scores than non-gainers (winter one: M = 6.89, SD = 7.68 vs. M = 9.28, SD = 7.72; winter two: M = 6.18, SD = 5.67 vs. M = 9.11, SD = 7.69, respectively).

Table 3 presents the proportions of participants meeting remission criteria at each timepoint and the proportions of participants meeting recurrence criteria at each follow-up by sudden gain status. Table 4 shows the logistic regression results, including odds ratios expressed as integers greater than one when sudden gainers had greater odds than non-gainers of remission, and less than one when non-gainers had greater odds. For recurrence, the odds ratios are expressed as integers less than one when sudden gainers had lower odds than non-gainers of recurrence, and greater than one when non-gainers had lower odds. Sudden gain status was not predictive of BDI-II remission status at post-treatment, χ2(2) = 4.18, p = .124. However, the overall BDI-II remission models were significant at both winters one and two (χ2(2) = 11.12, p = .004 and χ2(2) = 9.45, p = .009, respectively). A significantly larger proportion of sudden gainers achieved remission on the BDI-II at both winter one (p = .029) and winter two (p = .034).

Table 3.

Remission and Recurrence by Sudden Gain Status at Post-Treatment and Follow-Ups

	Sudden Gain	No Sudden Gain
	% (No.)	% (No.)
Post-Treatment
SIGH-SAD Remission	43.6 (17/39)	46.9 (23/49)
BDI-II Remission	53.9 (21/39)	53.1(26/49)
Winter One Follow-up
SIGH-SAD Recurrence	25.6 (10/39)	30.6 (15/49)
SIGH-SAD Remission	38.5 (15/39)	32.7 (16/49)
BDI-II Remission	66.7**(26/39)	51.0** (25/49)
Winter Two Follow-up
SIGH-SAD Recurrence	20.5 (8/39)	30.6 (15/49)
SIGH-SAD Remission	33.3 (13/39)	28.6 (14/49)
BDI-II Remission	66.7** (26/39)	51.0** (25/49)

Note.

^**p < .05 when comparing the sudden gain vs. no sudden gain group.

See Table 2 for measure abbreviations. SIGH-SAD recurrence = total SIGH-SAD score ≥ 20 + HAM-D score ≥ 10 + atypical score ≥ 5. SIGH-SAD remission = ≥ 50% improvement in SIGH-SAD compared to pre-treatment SIGH-SAD + HAM-D ≤ 7 + Atypical ≤ 7 OR HAM-D ≤ 2 + Atypical ≤10. BDI-II remission = BDI ≤ 8.

Table 4.

Binary Logistic Regression Results for Sudden Gain Status and Remission/Recurrence at Post-Treatment and Follow-Ups

	β	SE β	Wald’s χ2	p	Odds ratio
Post-Treatment
SIGH-SAD Remission	−0.09	0.43	0.04	.838	0.915
BDI-II Remission	−0.28	0.45	0.37	.542	0.758
Winter One Follow-up
SIGH-SAD Recurrence	−0.43	0.53	0.67	.412	0.650
SIGH-SAD Remission	0.41	0.47	0.75	.386	1.499
BDI-II Remission	1.09	0.50	4.76	.029*	2.972
Winter Two Follow-up
SIGH-SAD Recurrence	−0.83	0.57	2.15	.143	0.435
SIGH-SAD Remission	0.35	0.48	0.54	.465	1.418
BDI-II Remission	1.04	0.49	4.48	.034*	2.834

Note.

^*Refers to significance at .05 level.

β = logistic regression coefficient; SE = standard error. SIGH-SAD = Structured Interview Guide for the Hamilton Rating Scale for Depression-Seasonal Affective Disorder Version. BDI-II = Beck Depression Inventory-Second Edition.

Sudden gain status was not significantly predictive of SIGH-SAD remission status at post-treatment (χ2(2) = 5.50, p = .064), winter one (χ2(2) = 5.63, p = .060), or winter two (χ2(2) = 3.90, p = .142). For SIGH-SAD recurrence at winters one and two, each overall logistic regression model was significant (χ2(2) = 11.76, p = .003 and χ2(2) = 15.40, p < .001, respectively). However, this was largely due to the contribution of the pre-treatment SIGH-SAD score. Gain status did not contribute significant variance to the model at either winter one (β = −0.43, p = .412) or winter two (β = −0.83, p = .143).

Sudden Gains Status and Seasonal Beliefs

Figure 2 shows SBQ scores in sudden gainers and non-gainers at pre-, mid-, and post-treatment. As seen in Figure 2, SBQ scores generally decreased among sudden gainers over time, whereas, among non-gainers, SBQ scores decreased between pre- and mid-treatment but then stabilized or increased slightly between mid- and post-treatment.

Figure 2.

Seasonal Beliefs Questionnaire Scores at Pre-, Mid-, and Post-Treatment by Sudden Gain Status

Note. SBQ = Seasonal Beliefs Questionnaire. Error bars shown at +/− 1 SE. Pre-TX = pre-treatment, Mid-Tx = mid-treatment (Week 3), Post-Tx = post-treatment.

Mauchly’s test of sphericity revealed that the variance between SBQ scores across timepoints was significantly different, χ2(2) = 29.97, p < .001. Therefore, the degrees of freedom were corrected using the Greenhouse-Geisser method, ε = .74. The ANOVA revealed a significant main effect of time on seasonal beliefs, F(1.49, 107.11) = 21.13, p < .001, η2 = .28. There was also a significant sudden gain status × time interaction on seasonal beliefs, F(1.49, 107.11) = 3.48, p = .048, η2 = .05. Post hoc tests using Bonferroni correction revealed that SBQ scores were significantly lower in the sudden gain group (M = 123.51, SE = 3.55) than in the non-sudden gain group (M = 137.21, SE = 3.36) at post-treatment, mean difference = −13.61, 95% CI [−23.44, −3.94], p = .007. However, sudden gain groups did not significantly differ on SBQ scores at either pre- (mean difference = −3.44, 95% CI [−10.63, 3.75], p = .343) or mid-treatment (mean difference = 1.24, 95% CI [−7.64, 10.12], p = .781).

Additionally, there was a significant main effect of time on SBQ scores within both the non-sudden gain group and the sudden gain group. Within the non-sudden group, SBQ scores were significantly lower at mid- (mean difference = −15.28, 95% CI [−23.56, −7.00], p <.001) and post-treatment (mean difference = −13.18, 95% CI [−20.14, 6.21], p <.001) compared to pre-treatment. For gainers, SBQ scores were significantly lower at post-treatment (mean difference = −23.43, 95% CI [−31.22, −15.64], p <.001) compared to pre-treatment, but the difference between pre- and mid-treatment (mean difference = −10.60, 95% CI [−22.49, 1.29], p =.094) was nonsignificant. Both the gain and non-gain groups showed a significant linear trend in SBQ scores over time (F(1,34) = 57.335, p < .001 and (F(1,38) = 22.493, p < .001, respectively). However, SBQ scores showed a significant quadratic trend for the non-gainers that was not present for the gainers (F(1,38) = 7.542, p = .009 and F(1,34) = .042, p = .838, respectively).

The Influence of CBT-SAD Group Membership on the Relationship between Sudden Gains and Outcome

Linear mixed model analyses did not yield a significant effect of group membership on BDI-II scores at either winter one (F(30,50) = .998, p = .491) or two (F(30,50) = 1.392, p = .151), after controlling for pre-treatment BDI-II. Therefore, it appears that CBT-SAD group membership did not account for the significant effect of participant-level sudden gains on BDI-II scores at winter one and winter two.

Discussion

The original research study examining sudden gains (Tang & DeRubeis, 1999) and the broader literature that followed (Aderka et al., 2012; Shalom & Aderka, 2020) demonstrated that sudden gains during psychotherapy treatment are associated with better outcomes at treatment endpoint as well as at follow-up. The current study examined whether these findings generalize to cognitive-behavioral group treatment for SAD (CBT-SAD) in a secondary analysis of the CBT-SAD arm of a randomized clinical trial. In the parent study, CBT-SAD was compared to light therapy over 6-weeks of treatment and at follow-ups one and two winters later.

Despite CBT-SAD’s brief, intensive format and weekly outcome assessment, sudden gains occurred during CBT-SAD. A larger proportion of CBT-SAD participants experienced a sudden gain using Kelly et al.’s (2005) criteria (44.3%) than using Tang and DeRubeis’ (1999) criteria (27.3%). The proportion of CBT-SAD sudden gainers using the Kelly et al. (2005) criteria was consistent with the proportion of sudden gainers that Kelly et al. (2005) observed (41.9%) in their study of adults receiving CBT for depression. However, the proportion of CBT-SAD sudden gainers using the Tang and DeRubeis (1999) criteria was lower than the proportion of sudden gainers that Tang and DeRubeis observed (37.3%) in their study of CBT for adult depression.

One possibility for this difference is that the parent trial’s design included only seven observations during treatment (at pre-treatment and weekly during treatment), resulting in fewer opportunities to capture sudden gains according to the Tang and DeRubeis (1999) vs. the Kelly et al. (2005) criteria. Modifying the original Tang and DeRubeis criteria to allow for gains with at least two (vs. three) sessions before and after the gain still left only four intervals in which to detect sudden gains during CBT-SAD. In contrast, the Kelly et al. (2005) criteria allowed use of all six intervals from pre- to post-treatment. Indeed, most of the sudden gains missed by the Tang and DeRubeis (1999) criteria and captured by the Kelly et al. (2005) criteria occurred between the pre-treatment and the second week of treatment.

Sudden gains were originally hypothesized to result from cognitive changes (Tang et al., 1999, 2005). As in Kelly et al.’s (2005) study, more than half of all CBT-SAD sudden gains happened in the first third of treatment (i.e., between pre-treatment and Week 2), prior to the sessions devoted to cognitive restructuring. Of note, one-third (13/39; 33.3%) of all sudden gains occurred between pre-treatment and Week 1. Busch et al. (2006) made the distinction between “pre-treatment gains” (occurring between baseline and Session 1) and “first session gains” (occurring between Sessions 1 and 2). Without session-by-session assessments, it remains possible that some, or even all, of the 13 sudden gains detected between pre-treatment and Week 1 were pre-treatment gains, occurring before any CBT-SAD was delivered, which may reflect excitement about beginning an intervention and/or a general expectancy to improve with any intervention. Our data does not suggest such pre-treatment gains, if present here, are due to treatment preference. Prior to randomization, participants had an overwhelming preference for light therapy over CBT-SAD, with 81.9% ranking light therapy as their first choice (Meyerhoff & Rohan, 2016). Given our weekly assessments, it is equally possible that some, or even all, of these very early gains were first session gains. The first session of CBT-SAD focuses on introductions, group goals and purpose, confidentiality, an introduction to CBT, and expected changes. Such information may result in gains by improving participant “buy-in” to CBT-SAD, which affects outcome (Meyerhoff & Rohan, 2016). Sudden gains might have resulted from psychoeducation in Session 2 around SAD symptoms and what triggers them due to greater insight and understanding of one’s experience, instilling hope that symptoms could improve, or the shared experience of being in a group with other people who have SAD, disconfirming beliefs about being “the only one” who feels this way in the winter.

Another nearly one-third (12/39, 30.8%) of all sudden gains occurred between Weeks 1 and 3, during the behavioral activation and introduction to the cognitive mediation model portions of CBT-SAD. Sudden gains may occur directly in response to behavioral activation in Sessions 3 and 4 through greater pleasure and mastery in activities during the wintertime. Alternatively, sudden gains may occur through a cognitive mechanism in response to behavioral activation. Engaging with pleasurable activities and approaching, rather than avoiding, activities in the winter might provide contradictory evidence to challenge negative thoughts about the wintertime and/or one’s ability to function in the winter, prompting cognitive change (e.g., “I can do something to improve my mood, even though I am tired or it’s dark/cold out”). Consistent with the assertation that behavior activation can prompt cognitive change, Jacobson et al. (1996) reported comparable change in negative thinking, as measured with the Automatic Thought Questionnaire (Hollon & Kendall, 1980), over behavioral activation and cognitive therapy interventions for depression. Articulation of the cognitive mediation model in Session 5 and cognitive distortions in Session 6 may prompt additional gains as participants begin keeping thought diaries.

It was expected that CBT-SAD sudden gainers would have better outcomes at post-treatment than non-gainers. However, gainers and non-gainers did not significantly differ in depression severity or remission status at post-treatment, using either the SIGH-SAD or BDI-II cutpoints. Although CBT-SAD sudden gainers, by definition, had a large, rapid drop in depression symptoms and showed relatively few reversals, the non-gainers, in general, showed a gradual reduction in depression severity over the 6-weeks treatment phase. As seen in Figure 1, by treatment endpoint, improvement among the non-gainers had largely made up for the difference in scores attributable to a sudden ≥ 6-point SIGH-SAD score drop among gainers.

The benefit of sudden gains during CBT-SAD on outcome was evident at follow-ups one and two winters later, suggesting that that sudden gains in CBT-SAD improve the durability of treatment effect. Relative to CBT-SAD non-gainers, those who experienced a sudden gain had significantly lower depression scores on the BDI-II and a greater likelihood of BDI-II defined remission at both follow-ups. These results were not replicated when using SIGH-SAD scores or SIGH-SAD remission status as the outcome variable. Sudden gainers did not differ from non-gainers on the primary outcome measure in the parent study, SIGH-SAD recurrence status at follow-up. Significant differences between sudden gain groups at follow-up using the BDI-II but not the SIGH-SAD as outcome may be due to differences in depression symptoms captured by these measures. Factor analysis shows that the BDI-II captures a “negative attitude towards self” as well as a “performance impairment” factor (Seemüller et al., 2023). These cognitive aspects of depression may be missed by the SIGH-SAD, which has a more somatic focus.

Although seasonal beliefs generally became more flexible during CBT-SAD, participants who experienced a sudden gain during CBT-SAD had significantly more flexible seasonal beliefs at post-treatment than those who did not. CBT-SAD participants without sudden gains also improved in seasonal beliefs, but this improvement leveled off at mid-treatment, with a significant quadratic trend in SBQ scores over time. In contrast, CBT-SAD participants with sudden gains showed linear change in seasonal beliefs from pre- to mid- to post-treatment. Experiencing a sudden gain may facilitate modifying seasonal beliefs, which is an established mediator, and possibly a mechanism, of CBT-SAD’s antidepressant effects during and following treatment (Rohan et al., 2020, 2023). Given that CBT-SAD sudden gainers differed from non-gainers in their flexibility of seasonal beliefs at post-treatment but not midway through therapy, cognitive changes around seasonal beliefs are likely not the driving factor in producing sudden gains. Perhaps the gain-related rapid reduction of depression symptoms itself drives later change in seasonal beliefs (e.g., the SBQ item “It’s difficult to feel good on dark, dreary days”) and/or is a marker of internalizing more flexible seasonal beliefs through behavioral activation and cognitive restructuring during CBT-SAD. More research needs to be done to better understand the longer-lasting effects of sudden gains in CBT-SAD and whether they are mediated by change in seasonal beliefs.

To our knowledge, this was the first study to examine whether group membership influences the effect of sudden gains on outcome in a group therapy treatment. Secondary linear mixed model analyses confirmed that the significant differences between sudden gainers and non-gainers on follow-up depression (BDI-II) scores were not attributable to the effects of group membership. This indicates that CBT-SAD group membership was not related to the significant effect of participant-level sudden gains on treatment outcome in this study.

Limitations of the study include the weekly, rather than session by session, outcome assessment. Perhaps this explains why the proportion of gainers detected using the Tang criteria was lower in our sample than was detected in their original study, whereas the proportion detected using the Kelly criteria was in-line with their results. It remains possible that some gains went undetected due to the weekly measurement frequency. Similarly, seasonal beliefs were only assessed at three time points during treatment. More frequent assessment may provide a clearer pattern of seasonal belief change across treatment.

Sudden gains as measured here represent a large amount of change between weekly assessments that is worthy of investigation but likely reflect a different phenomenon than the Session N to Session N+1 drop per Tang and DeRubeis’ (1999) widely adopted definition. In this study, within-person symptom fluctuations at the session-level are unknown. For example, a weekly drop of ≥ 6 SIGH-SAD points could occur with a drop of 3 points between CBT-SAD Sessions N and N+1 and a further drop of 3 points between CBT-SAD Sessions N+1 and N+2, a pattern of steady improvement that combined, but not individually, meets the threshold. Alternatively, a threshold drop may be specifically associated with only one CBT-SAD session in the interval and the other session associated with no change (e.g., a drop of ≥ 6 SIGH-SAD points between CBT-SAD Sessions N and N+1 without further change between CBT-SAD Sessions N+1 and N+2 or vice-versa). A participant could have even deteriorated between CBT-SAD Sessions N and N+1 but then dramatically improved between CBT-SAD Sessions N+1 and N+2 to show a net weekly drop of ≥ 6 SIGH-SAD points at time of assessment. The weekly assessment schedule could also have masked sudden gains if, for example, a participant dropped by 6 points between CBT-SAD Sessions N to N+1 but gained back 1 point between Sessions N+1 and N+2. Therefore, sudden gains, as identified in this study, cannot be directly compared with the session-by-session changes identified in other studies as these constructs may differ.

Another limitation was the non-independence of the measurement used to detect sudden gains (SIGH-SAD) and the outcome measure (SIGH-SAD scores, remission, and recurrence). This concern was somewhat attenuated by including the BDI-II as an additional outcome measure (scores and remission). Finally, this was a single-site study, and the demographics of this sample were quite homogenous. Participants were largely white, female, and educated. Thus, the results of this study may not be generalizable to the larger population experiencing SAD. Future research ought to include a more diverse sample and other geographic regions.

To conclude, this study shows that sudden gains occur in group CBT-SAD, meaningfully relate to outcome one and two winters later, and relate to greater change in seasonal beliefs during CBT-SAD. The latter is exciting, given evidence that seasonal beliefs are a mediator, and potential mechanism, of CBT-SAD’s acute and enduring antidepressant effects. This is also the first study that we are aware of that considers the effects of group interaction on sudden gains and outcome. Future work should consider assessing symptoms after each CBT-SAD session to better capture potential sudden gains.