Abstract
Treatments for severe depression have moderate success rates, often take many weeks to yield responses, and are often followed by relapse or recurrence. Neurobehavioral interventions address these limitations by targeting mechanisms of cognitive and emotional dysregulation directly. This study extends data and observations from a pilot study examining effects of 2 weeks (6 sessions) of adjunctive cognitive control training exercises added to medication and psychotherapy in severely depressed patients. We examined acute effects and predictors of change in rumination, and long-term effects on service utilization. Compared with treatment as usual, exercises were associated with decreases in rumination and decreased use of intensive outpatient services in the following year. Responses were strongest among patients who displayed physiological indicators (pupillary oscillations at the task frequency) of task engagement before the intervention. These indices changed following intervention, suggesting that the intervention required capitalization on relevant attentional mechanisms and addressed fundamental emotional processes through their cognitive substrates.
Depression has long been recognized as a heterogeneous disorder, with research foci increasingly shifting from diagnoses writ large to concentration on more specific mechanisms (Insel et al., 2010). Yet, adjunctive treatments targeted at proposed subtypes have not yet been shown to increase outcomes such as early response. Early interventions targeting rumination, a key cognitive and affective process in depression involving perseverative focus on negative thoughts and feelings, have been suggested as a promising approach (Huffziger, Reinhard, & Kuehner, 2009) as rumination predicts delayed response to conventional treatments such as cognitive therapy (Jones, Siegle, & Thase, 2008). This article examines whether targeting a mechanism thought to underlie rumination could affect this and other symptoms early in treatment. Specifically, we examined which severely depressed patients’ rumination responds within 2 weeks to adjunctive executive control exercises added to a typical regime of medication and psychotherapy. Desirable clinical features such as lasting impact on symptoms over a 1-year follow-up period, prediction of clinical outcome, and change in the targeted mechanisms were also examined.
Here, we specifically focus on an intervention designed to increase executive control via repetitive exercises that necessitate prefrontal engagement. Increasing executive control could affect key aspects of depression directly. Disruptions of executive control are hypothesized to be central to depression (Ottowitz, Dougherty, & Savage, 2002). Decreased executive control, associated with decreased dorsolateral-prefrontal cortex (DLPFC) function, particularly in the context of addressing hyperactive limbic reactivity to emotional stimuli, has often been observed in depressed individuals (e.g., Baxter et al., 1989; Bench, Friston, Brown, Frackowiak, & Dolan, 1993; Dannlowski et al., 2009; Davidson, 1994; Goethals et al., 2005; Hooley et al., 2009; Killgore, Gruber, & Yurgelun-Todd, 2007; Mayberg et al., 1999; Siegle, Thompson, Carter, Steinhauer, & Thase, 2007; Wagner et al., 2006). Decreased prefrontal function has specifically been associated with perseveration, and potentially, perseverative, elaborative, or ruminative aspects of depression (Denson, Pedersen, Ronquillo, & Nandy, 2009; Haas, Constable, & Canli, 2008; Israel, Seibert, Black, & Brewer, 2010; Kross, Davidson, Weber, & Ochsner, 2009).
In addition, increasing executive control could potentiate the effects of standard treatments. That is, executive control is necessary for engaging in psychotherapies that require interruption of automatic reactions to emotional information (Berking et al., 2008). It could also augment the effects of common antidepressants, which directly address mechanisms of reactivity (e.g., Sheline et al., 2001). Furthermore, conventional therapies have often been hypothesized to ultimately act by increasing emotion regulation and have been associated with increased prefrontal activity on cognitive tasks (DeRubeis, Siegle, & Hollon, 2008; Frewen, Dozois, & Lanius, 2008); training executive control could help patients to “get a jump” on such processes.
Thus, “neurobehavioral” therapies targeted specifically at aspects of prefrontal function have garnered increasing attention, including direct interventions such as neurofeedback (Baehr & Baehr, 1997; Baehr, Rosenfeld, & Baehr, 1997; Rosenfeld, Baehr, Baehr, Gotlib, & Ranganath, 1996). We have previously suggested that a neurocognitive intervention involving prefrontal exercises was associated with decreased rumination and depression compared with treatment as usual in a small sample (Siegle, Ghinassi, & Thase, 2007). Here we examine a number of theoretically and clinically important generalizations in the full sample involving the robustness of our results for rumination, which has clear emotional as well as cognitive components. We specifically consider theoretically motivated response predictors and generalizations for service use.
The employed intervention is based on the theory that by increasing prefrontal function, particularly in the face of stressful emotional reactions, prefrontal deficits that lead to increased limbic dysregulation could be corrected. The basic approach is to exercise depressed individuals repeatedly on tasks that require prefrontal activity to complete, particularly including a computer-based version of attention training exercises associated with decreasing rumination in a case series (Papageorgiou & Wells, 2000) and an adaptive version of the Paced Auditory Serial Addition Task (PASAT; Gronwall, 1977), following a rehabilitation model. The attention training exercises described by Papageorgiou and Wells are low load, and thus are considered likely to allow for emotional information processing/rumination in those prone to engage in it; returning to the task induces cognitive control (Wells, 2000). In contrast, the PASAT is higher load and frustrating. The high cognitive working memory demand in the task inherently necessitates cognitive control. The frustrating nature of the task in absence of a catchy interface or obvious positive reinforcement requires cognitive control to stay on the task rather than to engage in reactive self-recrimination and rumination (Siegle, Ghinassi, et al., 2007).
Our initial report on this trial was promising, demonstrating increased reductions in symptoms and rumination, as well as increases in behavioral performance, and normalization of brain reactivity, above and beyond treatment as usual (medication and psychotherapy; Siegle, Ghinassi, et al., 2007). But the sample size was small, and results were explicitly preliminary. This article reports the conclusion of that initial trial of adjunctive cognitive control training (CCT) exercises in a severely depressed group compared with a comparably sized group of patients receiving treatment as usual (TAU). It is important that, as the intervention was directed at aspects of brain function subserving rumination, our primary outcome variable is rumination, with depressive symptoms being a secondary outcome variable hypothesized to change potentially less specifically or more slowly.
Given that the examined intervention was mechanistically specific, we did not expect that it would be helpful for all participants. Rather, we hypothesized that the participants who received the greatest benefit in emotion dysregulation from the intervention would be those who had the greatest observable deficits in executive engagement. That said, we also allowed for the potential that some critical expenditure of cognitive resources during the PASAT would be necessary to achieve benefit in the emotional domain. Either of these formulations would have a profound implication for understanding emotion–cognition interactions in recovery—the idea that emotional change is predicated on basic cognitive substrates. Often “basic” cognitive and emotional features of depression are considered to be separate (e.g., attention to nonemotional features is frequently “controlled for” in investigations of emotional information processing). Rather, we are suggesting that a strong understanding of change in emotional functioning must account for basic cognitive capabilities—simply put, whether you ruminate may depend on whether you are attending to the stimuli in your environment or just your ruminations.
Pupillary responses were used to index task-related resource allocation, specifically executive function during a nonadaptive version of the PASAT. Pupillary motility is well known to reflect cognitive load (Beatty, 1982), and using simultaneous pupil and fMRI assessment, we have previously shown pupil dilation to vary with DLPFC activity on a cognitive task (Siegle, Steinhauer, Friedman, Thompson, & Thase, 2011). In particular, in the examined sample, we have shown that compared with healthy participants, depressed participants display decreased pupillary responses at the frequency of task stimuli, and increased pupillary variation off the task frequency, potentially reflecting a lack of task-related resource allocation coupled with non-task-related processing (e.g., rumination, distraction, frustration; Jones, Siegle, Muelly, Haggerty, & Ghinassi, 2010).
The primary questions of interest were therefore geared toward establishing the validity and clinical relevance of the intervention. First, was there clinical change in targeted features? This is important as we claim the intervention is targeted—here we evaluate the target. Specifically we examined whether adjunctive CCT exercises tuned to address prefrontal cortex function were associated, within 2 weeks, with reduction in rumination and depressive symptoms. Our primary focus was whether CCT was associated with decreases in involuntary perseverative thought. The character of the perseverative thought (e.g., very negative, or simply perseverative focus on task-irrelevant information) was not specifically addressed by the intervention. Thus, primary analyses focused on a global measure of rumination (the full-scale rumination score for the Response Styles Questionnaire) rather than subscales for brooding or reflection per se, though these were considered in sensitivity analyses. Second, was the clinical change predicted by mechanistically expected pretreatment patterns of cognitive information processing? This is important because a targeted intervention should work only for individuals with specific mechanistic features to be targeted. Specifically we examined whether rumination was more strongly reduced in participants with observable patterns of physiological cognitive resource allocation on a nonadaptive preintervention version of the PASAT task used in CCT. Third, did the mechanisms change in responders? This question also speaks to whether the nominally mechanistically targeted intervention did target relevant mechanisms. Here we examined whether changes in rumination were associated with changes in physiological indices of cognitive resource allocation. As our primary hypotheses regarded acute change in the CCT group, these changes were examined a priori. To examine the specificity of observed changes with CCT, analogous tests were performed in the TAU group, and the relative size of effects in the CCT versus TAU groups was also examined. To make measurements comparable across individuals, particularly between CCT and TAU participants, we examined reactivity on a nonadaptive version of the PASAT. That said, to understand generalizability to the intervention itself we also examined analogous indices for the adaptive version in CCT participants. Fourth, was the intervention clinically important over a longer period? This question is important for clinical adoption of the intervention as momentary effects on mechanism are uninteresting unless the mechanism is causally connected to longer-term change. Toward this end, we conducted a preliminary chart review analysis of long-term outcome following CCT by comparing service utilization records during the year following CCT in CCT recipients and a matched sample of patients who received the same clinical services but did not enroll in our study. This article thus represents an increased sample size and more nuanced questions than reported in our presentation of initial results on this sample (Siegle, Ghinassi, et al., 2007).
In addition, as a central tenant of the proposed work is that the intervention changes how emotional information is processed, it is important to demonstrate changes in emotional information processing explicitly. Toward that end we conducted an initial analysis of the relationship between examined physiological indices of task-unrelated as well as task-related processing on the working memory task with change in physiological responses on an explicit emotion-processing task—identifying words as positive, negative, or neutral. Here, our expectation was that change on this task would be a function of both task-related resource allocation (i.e., it is a cognitive task) and task-unrelated processing (i.e., salient emotional processing not associated with simple word valence identification).
Method
Participants
CCT and TAU participants
Primary comparisons involved depressed individuals who received CCT or TAU. As shown in Figure 1, for these groups, 99 depressed patients from the Western Psychiatric Institute and Clinic’s intensive outpatient day-treatment program (IOP) were screened for this study, of whom 51 passed inclusion and exclusion criteria and came to at least one assessment; of these participants, 43 returned for a second assessment 2 weeks later (n = 23 CCT, n = 20 TAU) and were thus considered among the final sample. Because of technical errors, one participant in the CCT group had missing PASAT/pupil data at baseline and an additional one participant in the CCT and one participant in the TAU group had missing PASAT/pupil data at posttest, yielding slightly smaller n values for analyses that used PASAT pupil data (Items 2 and 3 in the results section). Inclusion criteria included having a diagnosis of unipolar depression on the Diagnostic and Statistical Manual of Mental Disorders (DSM–IV) via structured clinical interview (First, Spitzer, Gibbon, & Williams, 1996) and being 18 to 55 years old. Exclusion criteria included bipolar, psychotic, or substance use disorders, or being medicated with tricyclic antidepressants or Nefazodone due to likely effects on pupil dilation, which was our primary outcome predictor. The IOP offered a number of group therapies, typically three times each week, and weekly clinical management meetings with attending psychiatrists. Seven participants did not report taking psychotropic medications; the remaining participants were taking psychotropic medications as shown in Table 1, as well as medications for high blood pressure and insomnia and birth control. The duration of medication course varied markedly from a few days to months. Demographic information is presented in Table 1. TAU and CCT participants did not differ significantly on any demographic variable (p > .25 via t test for continuous variables and Fisher’s exact tests for counts), with the exception of SSRIs, for which marginally more CCT participants were prescribed SSRIs than TAU participants (Fisher’s exact p = .094). This difference was not controlled for statistically as (a) it was not significant and (b) it was assumed to work against prediction hypotheses (yielding low rumination regardless of initial cognitive style) though medications might reduce rumination writ large—that said, in absence of differential data on rumination effects for different classes of medication, this was considered hard to deal with statistically. Moreover, there were cohort effects in the type of prescribed SSRIs, as noted later, that may further moderate effects.
CONSORT diagram.
Descriptive Statistics for the Cognitive Control Training, Treatment-as-Usual, and Control Patient Samples
Note: BDI = Beck Depression Inventory; CCT = cognitive control training; IOP = intensive outpatient day-treatment program; RSQ = Response Styles Questionnaire; TAU = treatment as usual. Full-scale IQ equivalent assessed via the North American Adult Reading Test (Nelson & Willison, 1991).
n = 22 participants were used for analyses of PASAT pupil data (Items 2 and 3 in the results section) due to a technical error in one participant’s PASAT data collection at baseline.
Descriptives reported are for sample included in acute outcome analyses. The sample compared with control patients in service utilization analyses differed in two ways: (a) Service utilization records for one subject included in the acute analyses could not be located; (b) one subject who completed a full course of CCT but did not complete posttreatment questionnaires was included in the service utilization analysis, but not in the acute analyses.
For the healthy control sample, 1 participant’s BDI score was missing.
Service-control patients differ from combined CCT and TAU sample, p < .05.
Service-control patients differ from combined CCT and TAU sample, p = .10.
Service-control patients
As described later, following the protocol, TAU participants had the option to undergo CCT procedures, which they all did. As such, to provide comparative preliminary information on long-term outcomes from CCT, we used chart review to assess service utilization at Western Psychiatric Institute and Clinic during 1-year periods before and after CCT, comparing CCT participants to a new matched group of “service-control” patients who did not enroll in the study but received IOP services during the same period. For each CCT and TAU patient completing at least 1 session of CCT (before or after the acute posttest; n = 43), 1 an index month was defined based on the approximate 1-month active study period comprised either of CCT followed by a 2-week wait period (CCT group) or a 2-week wait period followed by CCT (TAU group). Service-control patients enrolled in the IOP during each CCT/TAU patient’s index month and matched to the index CCT patient on age were then identified. Service-control patients were required to have a current diagnosis of major depressive disorder (MDD; n = 39) or depressive disorder NOS (n = 18) and to have no current bipolar or psychotic disorder diagnosis and no record of current substance use. 2 The resulting service-control patient sample (n = 57) was significantly more likely to be male (Fisher’s exact p = .03) and nonsignificantly less likely to be Caucasian (Fisher’s exact p = .10) than the CCT sample (see Table 1).
Healthy controls
In addition, to help put results in context, reaction time and pupil dilation data are compared with 19 healthy never-depressed individuals from Jones et al. (2010) whose demographic data are described in Table 1.
Procedure
Patients were enrolled in cohorts into either the full CCT protocol or a TAU control. The CCT protocol involved a structured diagnostic interview, pretesting on the nonadaptive version of the PASAT as well as other cognitive and emotional tasks and a battery of questionnaires not described here but available from the authors on request, TAU in the IOP plus six 35-min CCT sessions (described in detail later), and posttesting using cognitive and emotional tasks. This 2-week experimental intervention was added to the patients’ regular IOP treatment program. The TAU control condition consisted of the same assessment procedures described earlier, in addition to the standard care provided by the IOP. TAU participants were offered CCT following their posttesting session—nearly all availed themselves of this option. Recruitment proceeded in cohorts. Our original publication on a subset of this sample (19 CCT, 8 TAU; Siegle, Ghinassi, et al., 2007) described randomized assignment; on inspection, due to procedural oversights, we did not end up randomizing at the patient level. Rather, cohorts were roughly randomized yielding alternating recruited subsets as 7 CCT, 8 TAU, 12 CCT, 18 TAU, 8 CCT. There was a 1-year break between the 12 CCT and 18 TAU cohorts. The final 8 CCT participants were recruited in alternation with an active control condition not described here due to extremely low power (n = 4). The recruitment schedule was the product of sporadically available resources. Healthy controls from Jones et al. (2010) were recruited simultaneously with the original 19 CCT participants.
TAU
All participants in both the TAU and CCT groups received TAU at WPIC’s IOP, including medication management, supportive group psychotherapy based on the principles of dialectical behavior therapy (Linehan, Heard, & Armstrong, 1993), and milieu therapy. Primary goals included teaching effective strategies for coping with mental health and safety issues. Patients attended the IOP for 3 hr, 3 days per week and met with the program psychiatrist once each week for clinical management. This treatment venue was chosen for a number of reasons for an initial test of CCT, including the presence of a stably maintained severely depressed population and the standardized intensive environment of psychosocial treatment, irrespective of CCT participation. This program minimized the chances that results associated with CCT would stem from group differences in clinical attention.
Components of the intervention
Participants attended six intervention sessions over 2 weeks in between the pre- and postintervention assessments. Spacing of the intervention sessions was per participants’ convenience and was not regulated. The CCT intervention had two components, each of which was geared toward activating the prefrontal cortex. The first exercises prefrontal function in the context of likely automatic ruminative cognitions. The second requires prefrontal control and use of working memory in the presence of frustration. The full intervention protocol and computer-based stimuli are available from the authors on request.
Tasks
Attention training
The first task exercised prefrontal function in the context of likely negative automatic ruminative cognitions and is based on Wells’s (2000) attention training. This task requires individuals to learn to direct their attention by focusing on one sound at a time, switching attention between sounds, and counting sounds in a naturalistic environment, all the while staying focused on the task rather than naturally occurring depressive ruminative thoughts. Thus, cognitive control is needed to stay with the task rather than more automatic emotional processes. The protocol takes approximately 15 min per session. This protocol has been studied extensively in anxiety disorders; an initial study suggested the intervention was useful in treating four depressed individuals in just a few sessions (Papageorgiou & Wells, 2000). We used a computer-administered version of the protocol. Bird sounds (from Blinkow, 1999), presented in four-speaker surround sound at randomly occurring intervals, provided the environmental stimuli.
Adaptive PASAT
The second task, a variant of the PASAT (Gronwall, 1977), involved continuously adding serially presented digits in working memory. Participants are asked to add each new digit to the digit that preceded it (i.e., sum just the current and one-back digits). Difficulty is manipulated by increasing the speed with which items are presented. Participants are instructed to get as many items right as they can and to resume the task as quickly as possible when they get something wrong. Thus, the task taps both working memory and executive control. One study in which nine healthy individuals completed the PASAT during assessment with fMRI reported that left middle frontal gyrus activity (including the DLPFC) was increased during the PASAT versus a control task (Lazeron, Rombouts, de Sonneville, Barkhof, & Scheltens, 2003). Furthermore, depressed individuals have been shown to score lower than healthy individuals on the task (Landro, Stiles, & Sletvold, 2001), even when performance on other neuropsychological tasks was controlled for. The PASAT is known to be frustrating (Holdwick & Wingenfeld, 1999). Thus, to keep the task tolerable by depressed participants (i.e., control induced frustration on a per-subject basis), likely to engage prefrontal circuitry rather than a “giving up” reaction, and likely to remain a useful exercise even after training, a modified version of the task was used that was adapted to participants’ performance, beginning with a 3,000-ms interstimulus interval (ISI) and speeding up by 100 ms when participants get four consecutive items correct. It slowed down by 100 ms when they missed four consecutive items, to keep participants at a constant level of performance. This technique equates the task for difficulty across participants and sessions. Participants completed three 5-min blocks per session for a total of 15 min. An adaptive variant of the PASAT has previously been used; the speed of presentation on the task was positively correlated with performance on other tests of executive function (Royan, Tombaugh, Rees, & Francis, 2004).
Explicit emotional information processing
To assess resource allocation associated with explicit emotional information processing we employed an alternating word emotion-identification/digit sorting task that we have used in the past (full details provided in Siegle et al., 2011). Briefly, following a fixation mask (1 s) and presentation of briefly presented positive, negative, and neutral words (150 ms), participants push a button to indicated whether they are positive, negative, or neutral (8.5 s), followed by a cognitive trial involving putting digits in numerical order geared to clear participants’ processing of the emotional stimulus (total trial duration 27 s). Decreased sustained pupillary motility on this task is positively predictive of response to cognitive therapy and sustained processing is related to dorsolateral prefrontal activity (Siegle et al., 2011). Sustained pupillary motility on similar alternating word-emotion-identification/cognitive tasks is associated with rumination (Siegle, Steinhauer, Carter, Ramel, & Thase, 2003).
Measuring symptoms and rumination
Change in rumination, a specific mechanism of interest, was the primary outcome measure and was assessed with Nolen-Hoeksema’s Response Style’s Questionnaire (Nolen-Hoeksema, Morrow, & Fredrickson, 1993). This questionnaire is frequently used to index trait rumination associated with depression. As noted previously, the full-scale score was used to index perseverative thought writ large, which we hypothesized would be addressed by increasing cognitive control. In sensitivity analyses, we further examined whether change was general or restricted to subscales of this measure associated with more emotional features of rumination (brooding) and more cognitive nonemotional features (reflection; Treynor, Gonzalez, & Nolen-Hoeksema, 2003). Change in depressive symptomatology was a secondary outcome and was measured with the Beck Depression Inventory (BDI; Beck, Steer, & Brown, 1996). This 21-item self-report questionnaire is frequently used to index depressive severity in clinically depressed individuals.
Measuring long-term outcome
A pre and a post year were defined for both CCT and service-control patients as the 1-year period prior to the start of the relevant index month and the 1-year period following the end of the index month. A tally of the number of IOP sessions, regular (i.e., non-IOP, usually weekly or bi-weekly) outpatient therapy sessions, and outpatient medication management visits was acquired for each patient during these pre and post years and used in ANCOVAs comparing post year tallies for each service type across the two groups, controlling for pre year tallies, age, gender, and ethnicity (Caucasian vs. non-Caucasian). Because of nonnormal (zero-inflated) distributions, SPSS’s bootstrapping function, a procedure robust to normality violations (Delucchi, Ph, & Bostrom, 2004), was used to obtain 95% confidence intervals (CIs) and p values for all service utilization analyses.
Data selection, cleaning, and reduction
Calculation of behavioral indices
Median ISIs on the PASAT were computed for each participant to account for the skewed distribution of ISIs associated with often starting at a nonoptimal starting point. Means of the median ISIs within participants were examined across participants.
Calculation of pupil dilation indices
Data were cleaned using methodology previously described (Siegle, Steinhauer, & Thase, 2004). Briefly, following linear interpolation through blinks, data were smoothed using a 5-point unweighted average filter applied twice. Linear trends in pupil dilation calculated over experimental blocks were then removed from pupil dilation data to eliminate effects of slow drift in pupil diameter.
Wavelet transformation
Indices of task-related and task-unrelated processing were derived using our previously described procedure that differentiated depressed from nondepressed participants (Jones et al., 2010). Cleaned pupil data were subjected to a continuous wavelet transform algorithm to calculate the wavelet power spectrum using the “wavelet” function from Torrence and Compo’s (1998) wavelet toolkit for MATLAB (The MathWorks, 2007). A Morlet wavelet (wavenumber w0 = 6) was chosen as the mother wavelet given its resemblance to the pupillary impulse response. Wavelet spectral power was normalized by dividing each band by its spectral frequency so that equivalent magnitude sine waves at any frequency were considered to have equivalent spectral power. This was done to control for the phenomena that for the same magnitude of time-domain response, wavelet power in high frequencies is actually lower than wavelet power at lower frequencies. Based on our design, mean wavelet spectral power was calculated at 0.417 Hz for ISI 2,400 ms trials; that is, we calculated the mean spectral power for each participant at the frequency at which the task was occurring (henceforth “on-task” power, our index of task-related resource allocation) on the nonadaptive PASAT task. In addition, for each participant, we calculated the mean wavelet spectral power at and below 0.37 Hz, that is, the mean spectral power in all frequency bands occurring slower than the frequency at which the task was occurring (henceforth “off-task” power, our index of off-task resource allocation). Given that the task occurred at 0.417 Hz, we chose to calculate spectral power at and below 0.37 Hz to minimize the overlap between our measures of off-task resource allocation and on-task resource allocation. Power above the stimulus frequency was not examined as it could reflect more transient processes and physiological artifacts.
Outliers
Outliers on all self-report and physiological variables were Windsorized (rescaled to the last valid value within the 25th percentile − 1.5*IQR to 75th percentile + 1.5*IQR) before analysis to ensure normality. This procedure affected very few values in any analysis. Sensitivity analyses confirmed that effect sizes changed very little when this procedure was not used.
Association of indices with explicit emotion processing
Because we were interested primarily in sustained pupillary responses, conventional statistics such as peak response are not appropriate. Rather, contrasts were examined via statistical tests at each sample along pupillary waveforms on the explicit emotion-processing task for groups defined using indices from the nonadaptive PASAT. To control Type I error, 35 consecutive samples (.56 s) of tests in a row significant at p < .1, considered replications, were considered significant at p = .05 (described in Guthrie & Buchwald, 1991, e.g., as used with pupil dilation in Siegle, Ichikawa, & Steinhauer, 2008). Results report tests of the mean pupillary response in significant windows.
Results
Reduction in rumination and depressive symptoms in the first 2 weeks following enrollment
Rumination
As shown in Figure 2, there was a Group (CCT/TAU) × Time (pre/post) interaction in the primary outcome measure, RSQ scores, F(1, 41) = 15.48, p < .001, η p 2 = .27. Reductions in rumination were present in the CCT group, t(22) = −6.79, p < .001, raw difference in RSQ scores (D) = −10.13, standardized difference (Cohen’s d) = −1.42, 2/23 increased, but not in the TAU group, t(19) = −0.20, p = .85, D = −0.40, d = −0.04, 6/20 increased. To understand the differential effects of treatment on rumination at the level of each subject, when initial RSQ score was regressed from final RSQ scores within the full sample, 5/23 participants in the CCT group had residual rumination > 0, whereas 16/20 participants in the TAU group had residual rumination > 0, Fisher’s exact test p < .0005.
Clinical change in treatment as usual (TAU) versus cognitive control training (CCT).
Results were of the same general character for the more restricted Brooding subscale of the RSQ, which measures rumination specifically on negative cognitions, Group × Time F(1, 41) = 9.33, p < .005, η p 2 = .19. Reductions in brooding rumination were present in the CCT group, t(22) = −4.18, p < .001, D = −2.17, d = −0.98, 0/23 increased, but not in the TAU group, t(19) = −0.17, p = .87, D = 0.1, d = −0.04, 9/20 increased. In contrast, there was no differential change across groups in the Reflection subscale, which measures a less negatively focused tendency to reflect on past events, Group × Time F(1, 41) = 1.78, p = .649, η p 2 = .04.
Depressive severity
There was no Group × Time interaction in BDI scores, F(1, 41) = 1.7, p = .64. BDI scores decreased in both the CCT group, t(22) = −5.69, p < .005, D = −10.48(8.83), d = −1.19, 1/23 increased, and the TAU group, t(19) = −2.68, p = .02, D = −6.55, d = −0.60, 5/20 increased. The lack of agreement between depression and rumination depression severity changes was thus due to the fact that nearly all individuals improved in depression severity, but only the only the CCT individuals improved in rumination. Depression and rumination change were uncorrelated in both the CCT group, r = .13, p = .53, and the TAU group, r = −.06, p = .8, yielding a nonsignificant correlation in the full sample, r = .11, p = .46.
Association of improvement in rumination with physiological indices of cognitive resource allocation on the nonadaptive PASAT
Within the CCT group, continuous residual rumination (posttreatment RSQ, after covarying out pretreatment RSQ and Windsorizing outliers) was examined with regard to on- and off-task power, and their interaction. Though the overall model was significant, the interaction was not a significant predictor (p > .4), so a main effects model was examined. Residual rumination was predicted, R2 = .36, F(2, 21) = 5.33, p = .02, by nonadaptive PASAT on-task power, t(21) = −3.41, p = .003, stβ = −2.25, and off-task power t(21) = 3.34, p = .005, stβ = 2.13. Thus, decreased rumination at posttreatment was associated with initially higher on-task power and lower off-task power. The same general character of results, with slightly weaker effects was observed for the Brooding subscale; as predictions regarded rumination in general and not simply brooding, and as results were indeed more robust for the full rumination scale, this scale was used in subsequent analyses.
Formulating a clinically useful “unfocus index.”
A more clinically useful, if less reliable index, is the number of points decreased on the RSQ, which was similarly predicted by on- and off-task power, R2 = .28, F(2, 21) = 3.64, p = .04, via the equation
Because the index is proportional to low on-task power and high off-task power, it is henceforth termed the unfocus index, which also strongly predicted residual rumination, R2 = .36 (see earlier discussion). Thus, when participants were roughly divided into high responders and low responders based on residual final RSQ scores being below or above zero, respectively, the unfocus index correctly predicted 11/12 high-responders and 9/10 low responders, 90% correct, logistic regression R2 = .58, p < .005, χ2proportion = 14.67, p < .005, using a grid-search optimized threshold = −10.72, d′ = 2.66. As shown in Figure 3 (left), for CCT participants, focus was greater in high responders than low responders.
Association of pretreatment unfocus index (difference between on-task and off-task wavelet power) with change in rumination in cognitive control training (CCT) but not treatment as usual (TAU).
Differential prediction
On- and off-task power
To demonstrate differential prediction, when the CCT and TAU groups were combined, residual final RSQ was significantly predicted above and beyond group and task power by Group × On Task and Group × Off Task power terms, R2 = .51, ΔR2 = .14, ΔF(2,38) = 5.17, p = .01, such that power was not significantly predictive of residual RSQ in the TAU group, R2 = .16, F(2, 19) = 1.6, p = .23, and the coefficients were opposite in direction to the CCT group, on-task power stβ = 0.13, t(17) = 0.42, p = .68, off-task power, stβ = −.49, t(17) = −1.5, p = .15.
Unfocus index
More generally, high focus predicted strong response to CCT whereas, as shown in Figure 3 (right), in TAU, low focus was nonsignificantly associated with decreased RSQ scores, grid search optimized threshold = −11.83, 75% correct, d′ = 1.29, p = .27.
Changes in behavioral and physiological indices of executive control
Speed on the adaptive PASAT, a behavioral index of executive control, was compared with the 19 healthy control participants from Jones et al. (2010) and between groups over time subject to outlier Windsorization. At the baseline assessment, depressed participants median ISI (M = 2730.23, SD = 629.64) was slower than controls (M = 2326.31, SD = 552.61), t(60) = 2.41, p = .02, difference (M = 403.92, SD = 607.56), Cohen’s d = 0.66. Following training, the CCT group was faster than controls’ Day 1 performance (M = 1965.22, SD = 512.22), t(40) = −2.19, p = .03, D(ms) = −361.10 (530.78), d = −0.68, whereas the TAU group was nonsignificantly slower than the mean of controls (M = 2700.00, SD = 632.45), t(37) = 1.96, p = .06, D(ms) = 373.68 (594.95), d = 0.63, yielding a significant difference between the two depressed groups, t(41) = 4.21, p < .005, D(s) = 734.78 (571.10), d = 1.29.
On-task power on the nonadaptive version of the PASAT administered pre- and posttreatment was assessed as a measure of task-related neural executive processing. Power indices were highly reliable in controls and moderately reliable in depressed participants despite the interventions (Table 2). For CCT participants, on-task power increased compared with TAU participants, Group (TAU, CCT) × Time (pre, post) F(1, 38) = 3.69, p = .004, η p 2 = .09. Within the CCT group, on-task power increased reliably, t(20) = 3.37, p < .005, D = 0.10 (0.14), d = 0.74. In contrast, on-task power did not increase for the TAU group, t(18) = 0.95, p = .36, D = 0.02 (0.08), d = 0.22. As shown in Figure 4, very few CCT individuals decreased in on-task power (n = 4/21). Unexpectedly, in the CCT group, less of an increase in on-task power at the trial frequency was related to more decreased rumination, r = .44, t(19) = 2.11, p = .05 (Figure 4). On-task power was not related to change in rumination in the TAU group, r = .03, t(17) = 0.11, p = .91, or in the combined depression group, r = .04, t(38) = 0.27, p = .79. Change in on-task power was not related to changes in depressive severity or performance.
Two-Week Test–Retest Reliability of 2,400 ISI Power Indices on the Nonadaptive PASAT Task (Outliers Rescaled)
Note: CCT = cognitive control training; TAU = treatment as usual. CCT and TAU participants were in treatment during this time; controls were not. CCT participants were trained on the adaptive version of the measured task. Correlations reported.
p < .05. **p < .005.
Relationship between change in on-task power on the nonadaptive Paced Auditory Serial Addition Task and change in rumination in the cognitive control training (CCT) group.
Off-task power on the nonadaptive version of the PASAT was assessed as a measure of non-task-related processing. There were no significant group differences in change across time, Group (TAU, CCT) × Time (pre, post) F(1, 38) = 2.04, p = .66, η p 2 = .05. Decreasing off-task power was weakly and nonsignificantly related to changes in rumination, r = .33, t(19) = 1.55, p = .14, in the CCT group, and not related to changes in depressive symptomatology or performance in either group.
In combination, the unfocus index did decrease for nearly all CCT participants (n = 19/21), though as stated earlier, increasing focus was unexpectedly associated with less of a decrease in rumination, R2 = .233, F(1, 20) = 5.780, p = .027.
Exploratory analysis: changes in service utilization up to 1-year post-CCT
As shown in Table 1, chart review revealed that participants (collapsing across CCT and TAU) receiving at least 1 CCT session (n = 43; M = 5.3 CCT sessions, SD = 2.1, range = 1–7) had fewer IOP visits in the post-CCT year than a control sample of matched IOP patients not recruited for our study, controlling for pre year IOP visits, age, gender, and ethnicity; group (CCT vs. control patient) F(1, 94) = 4.9, 95% CI for B = 0.60 to 5.5, p = .031. Identical ANCOVA analyses for regular outpatient therapy and medication management visits were not significant, ps > .2. In the year following treatment, 90.7% of CCT recipients versus 71.9% of control patients had 5 or fewer IOP sessions, two-tailed Fisher’s exact p = .024. Pre versus post year IOP visits decreased significantly in CCT recipients, paired t(42) = 5.6, 95% CI for mean difference = 5.1 to 10.8, p = .001, but not control patients, paired t(56) = 1.6, 95% CI for mean difference = −0.59 to 5.4, p = .12. Pre year IOP visits did not differ by group, 95% CI for mean difference = −0.72 to 5.9; p = .15. Among CCT recipients, the number of CCT sessions received was not correlated with any post year service utilization variable, |r|s < .15, ps > .35.
The principal ANCOVA finding related to post year IOP sessions was upheld at the trend level, p = .055 using a smaller control patient sample, n = 39, composed of only those with an MDD (rather than a depressive disorder NOS) diagnosis.
Associations of on-task and off-task power with change in explicit emotional information processing
A central tenant of this work is that change in the intervention translates to change in how emotional information is processed. Our preliminary report showed pre to post changes in an explicit emotional information processing task associated with the intervention. In the full sample 15 CCT and 15 TAU participants completed the task during assessment of pupil dilation. As shown in Figure 5, within the CCT group an On-Task Power (high/low, on the nonadaptive PASAT) × Day (pre/post) ANOVA at each sample again revealed that pupil dilation decreased following the intervention, day main effect, 4.18 to 9.53 s, F(1, 13) = 11.41, p = .005, 10.9 to 17.42 s, F(1, 13) = 11.45, p = .005. These decreases were qualified by larger decreases in the participants with high PASAT on-task power, during the peak of the valence identification period, Power × Day interaction 4.2 to 5.78 s, F(1, 13) = 6.07, p = .03, and during digit sorting, F(1, 13) = 7.29, p = .02. Although the effect of day was not unique to the CCT group, Group (CCT/TAU) × Day, p > .1, within the TAU group interaction effects were not present during valence identification, Power × Day interaction in the peak/sustained period p > .3, though they were present in the digit sorting period, suggesting they could be associated with task exposure, Power × Day interaction 11.63 to 13.83 s, F(1, 15) = 5.32, p = .04, 16.28 to 17.95 s, F(1, 15) = 5.36, p = .04.
Pupillary motility on the alternating valence identification/digit sorting task separated by high and low on-task power on the nonadaptive Paced Auditory Serial Addition Task (PASAT) given prior to intervention.
Discussion
The current study examined the effects of an adjunctive cognitive training protocol targeting prefrontal function (CCT) on self-reported symptoms of rumination, a form of perseverative negative emotional processing common in depression, and physiological indices of executive function, compared with TAU. Data suggested that CCT was associated with a greater reduction in rumination than TAU from pre- to postintervention, particularly emotional features of rumination (brooding). Reduction in depressive symptomatology occurred for both the CCT and TAU groups and did not significantly differ between groups. Rumination change was positively predicted by a physiological index of task-related processing on the nonadaptive version of the intervention task (PASAT) in the CCT group but not the TAU group. CCT was associated with change in task-related processing, but not non-task-related processing on the nonadaptive PASAT.
Changes in symptoms and rumination were largely consistent with our initial publication on a subset of the presented data (Siegle, Ghinassi, et al., 2007), which did not focus on prediction or pupil dilation during the PASAT. The specific pre- to postintervention decrease in rumination in CCT compared with TAU could suggest that the CCT exercises, which were specifically constructed to address a theoretical mechanism of rumination, served that function. By increasing executive control in the face of otherwise distracting thoughts, participants could better focus on specific tasks. In our original publication on a subset of the current participants, CCT participants (before the last cohorts of TAU and CCT) also displayed greater reductions in symptomatology compared with TAU participants. Greater pre- to postintervention changes in symptomatology were observed in the most recently recruited members of the TAU group, decreasing group differences in this index, potentially due to a change in prescribing patters in our clinic—escitalopram became the first-line prescription of choice for that group and may have facilitated early response. That said, the final dissociation in changes in rumination and depressive severity may suggest that rumination is initially causal for depression (i.e., those individuals who ruminate are more likely to get depressed) but changing depression does not necessarily change its associated vulnerability processes such as rumination.
Prediction of change in rumination by increased task-related physiological responsivity compared with non-task-related responsivity before CCT suggests that participants who strongly engaged in the intervention task on the first day it was administered may have achieved the most benefit from it. This pattern did not hold for TAU participants suggesting that the prediction was specific to the mechanism.
Though our physiological index of task focus reliably increased from pre- to post-CCT, these increases were not positively associated with decreases in rumination. Rather, rumination decreased most for those whose task-related resource allocation increased least. Potentially, those individuals who started high in task focus benefited from the training for other reasons, for example, capitalization on a premorbid ability to recruit prefrontal control, associated with increased emotion regulation, or a motivational difference in willingness to use prefrontal function for this task. In either case, these data could suggest that starting high in task focus allowed participants to achieve a goal different than simply increased focus, for example, increased cognitive control over their emotional reactivity. Those who spent the time simply trying to engage with the task may have been less likely to get to the more beneficial aspects of the intervention. That is, targeting emotional dysregulation with a cognitive task may require first being able to engage with the task.
Initial evidence supports the idea that the training did, indeed, affect emotional information processing, but only for those who strongly engaged with it. That is, neurophysiological engagement with explicitly emotional information, signified by pupil dilation, decreased from initial to postintervention for those with initially high on-task power in just the CCT group. For those with low-on-task-power there were virtually no differences in responses from pre- to postintervention. This was different from effects on a more purely cognitive task. Together, these data could suggest that symptom benefits of cognitive training are through an emotional information processing route, though we did not have the power to test such a model.
More generally, the data suggest that a simple physiological test, administered before a targeted intervention and analyzed in an automated way, could help to predict its utility in improving specific aspects of symptomatology within the first 2 weeks of intervention.
The change in physiological reactivity we observed was a necessary but not sufficient condition for inferring change in the mechanism of interest, cognitive control. Although other potentially confirmatory measures were not administered, this first piece of evidence is consistent with the hypothesis that relevant mechanisms were targeted by PASAT training.
Observations of change in mechanisms with a neurobehavioral treatment targeting prefrontal function is strongly consistent with other similar efforts using different prefrontally intensive tasks in patient groups as diverse as schizophrenia and older adults with cognitive decline (Greig, Zito, Wexler, Fiszdon, & Bell, 2007; Mayer, Bishop, & Murray, 2012; Twamley, Jeste, & Bellack, 2003; Willis & Schaie, 2009). The more novel element of the current work is that such changes can be predicted by pretreatment assessments.
In addition, previous studies of neurocognitive therapies have often been limited by lack of long-term follow-up. Our preliminary analysis of service utilization records suggested that CCT recipients had decreased need for IOP level of care during the 1-year period following study participation in comparison to a matched sample of IOP patients not recruited for the current study. The number of regular outpatient therapy and medication management appointments did not differ across the CCT and control patients during this period, suggesting that the CCT recipients were not more likely than control patients to be lost to follow-up or to discontinue care at our facility. Thus, CCT recipients would likely have been referred for additional IOP services should this level of care have been warranted based on clinicians’ impressions of symptom severity and level of functioning. Decreased IOP sessions during the post-CCT year therefore appear to be a valid, though imprecise, proxy for decreased symptom severity over a 1-year follow-up period. Although we matched control IOP patients to CCT recipients on age, month of active enrollment in the IOP, and diagnosis, and controlled for observed demographic differences (gender, ethnicity) in analyses, unmeasured variables such as motivation and socioeconomic status may have differentiated IOP patients who chose to enroll in our study from those who did not, and accuracy of control patient diagnoses could not be assured. Furthermore, we did not find evidence of a dose-response relationship for the number of CCT sessions received, which would have strengthened our argument for long-term effects of CCT itself. These caveats notwithstanding, our data are suggestive of possible long-term benefits from CCT that warrant further study.
This study had a number of limitations. The lack of an active control condition prevented examination of whether observed changes were a function of increased clinical attention, study rationale, or other factors associated with CCT. The intensive clinical attention received by all participants partially alleviates these concerns. However, of particular note, as participants were informed that change in rumination was a study goal, observed changes in rumination could have been due to demand characteristics. The lack of randomization in recruitment at the patient rather than cohort level was potentially important, particularly as prescribing patterns at the IOP changed over the course of the study. The patient population was severe and heterogeneously medicated, particularly including patients with medications that are typically left out of behavioral treatment trials due to effects on learning (e.g., benzodiazepines). This population was specifically selected because the adjunctive intervention was considered most necessary for treatment in a severe population who might not otherwise respond to conventional interventions. That said, further exploration, particularly regarding the utility of our prediction algorithm in a less severe and more heterogeneous sample, would be useful.
These limitations notwithstanding, these data have theoretical and practical implications. Theoretically, presented data help to support a rarely tested tenant of the original “cognitive models” of depression (Beck, 1967)— that cognition and emotion in depression are fundamentally linked. Whereas most investigations of this relationship are confounded by examining cognition about emotional information, which does not examine basic cognition, the current data demonstrate that cognitive and emotional information processing are strongly linked in depression. Specifically, cognitive information processing styles appears to have measurable effects on the nature and time-course of emotional information processing, and by implication, emotion dysregulation. Addressing cognitive features of information appears to affect emotion dysregulation directly. Moreover, the value of cognitive interventions for emotion regulation may depend on participants’ ability to engage with them in the first place—using these interventions in participants who cannot attend to the basic cognitive substrates of the task may not help to address emotional dysregulation which is predicated on executive dyscontrol. An intervention targeted at a mechanism thought to underlie a specific form of emotion dysregulation in depression, rumination, appears to be (a) associated with reduction in that symptom, (b) most useful for individuals who engage a putative index of the mechanism on the intervention task before the intervention is administered, (c) associated with change in the putative index of the underlying mechanism, and (d) associated with decreased need for intensive outpatient services during a 1-year follow-up. A practical implication of this work could be that patients who score high on a measure of task-related processing on an executive control task may specifically benefit from adjunctive CCT. More broadly, this work could support the growing field of neurobehavioral or “brain training” exercises used to target specific mechanisms of depression and other emotional disorders. Neurobehavioral exercises remain infrequently used in clinical care and, when used, tend to be given without pretreatment mechanistic assessment, based on symptoms. The current work would, instead, suggest the utility of quick, inexpensive pretreatment mechanistic assessment before prescribing such interventions.
Of course, the current work will not change policy alone. It represents a Phase 1 investigation of a potential clinical target. Additional steps are required to recommend such assessments/interventions, validly, in the psychiatry clinic. We suggest that three steps are appropriate for this, and more generally, for those investigating neurobehavioral interventions with promising initial data. These involve establishing relevant psychometrics for the intervention, improving the likely clinical acceptability of the work, and increasing theoretical understanding of the intervention mechanisms. Applied to the current intervention, this practice would involve at least the following important steps. First, further testing should include randomized controlled trials with a compelling placebo (at least something considered to be “brain training” by participants), replicated response prediction, in this case, for those with high task focus, and assessment of not only service use but clear information regarding rumination and brain function at a follow-up assessment. This will also include understanding which populations this intervention is optimally useful for, for example, quite severe individuals as examined here, less severe depressed individuals, or perhaps vulnerable individuals with the examined mechanism, as suggested by DeRaedt and Koster (2010). The second step is making the presented indices useable, that is, development of norms, easy algorithms for clinicians to use, making turnkey, inexpensive hardware/software packages so that clinicians could hope to implement the proposed treatment prediction steps. The third step is more detailed understanding of relevant mechanisms, for example, via neuroimaging in larger samples at preintervention, postintervention, and follow-up. This is particularly important as rumination changed most for those whose task-related power did not increase—finding what mechanisms, for example, more tonic prefrontal control, did increase in association with rumination decreases is thus key. The current data provide compelling rationale for pursuing these directions, and we are happy to note that the proposed steps are in various steps of progress in ongoing studies at a number of institutions.
Footnotes
Acknowledgements
We gratefully acknowledge contributions of the volunteers who participated in this study as well as Lisa Farace and Dimple Sodhi for contributions to data collection, and to the staff of WPIC’s Intensive Outpatient Program for their help in recruiting for this study.
Declaration of Conflicting Interests
The authors declared that they had no conflicts of interest with respect to their authorship or the publication of this article.
Funding
This research was supported by the National Institutes of Health MH082998, MH64159, and NARSAD. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.