Introduction
Neurocognitive dysfunction is a common feature of mood disorders. Deficits in a range of cognitive processes have been described during symptomatic episodes in major depressive disorder (MDD) (Zakzanis et al. Reference Zakzanis, Leach and Kaplan1998; Lee et al. Reference Lee, Hermens, Porter and Redoblado-Hodge2012; Rock et al. Reference Rock, Roiser, Riedel and Blackwell2014) and bipolar disorder (BD) (Rubinsztein et al. Reference Rubinsztein, Michael, Underwood, Tempest and Sahakian2006; Kurtz & Gerraty, Reference Kurtz and Gerraty2009; Gallagher et al. Reference Gallagher, Gray, Watson, Young and Ferrier2014, Reference Gallagher, Gray and Kessels2015), including in medication-free patients (Porter et al. Reference Porter, Gallagher, Thompson and Young2003; Taylor Tavares et al. Reference Taylor Tavares, Clark, Cannon, Erickson, Drevets and Sahakian2007). There has long been an emphasis on the extent to which such deficits can be observed in clinical remission (Astrup et al. Reference Astrup, Fossum and Holmboe1959; Bratfos & Haug, Reference Bratfos and Haug1968), with growing consensus that they may be state-independent (Robinson et al. Reference Robinson, Thompson, Gallagher, Goswami, Young, Ferrier and Moore2006; Torres et al. Reference Torres, Boudreau and Yatham2007; Arts et al. Reference Arts, Jabben, Krabbendam and van Os2008; Bora et al. Reference Bora, Harrison, Yücel and Pantelis2013; Bourne et al. Reference Bourne, Aydemir, Balanzá-Martínez, Bora, Brissos, Cavanagh, Clark, Cubukcuoglu, Dias, Dittmann, Ferrier, Fleck, Frangou, Gallagher, Jones, Kieseppä, Martínez-Aran, Melle, Moore, Mur, Pfennig, Raust, Senturk, Simonsen, Smith, Soares, Soeiro-de-Souza, Stoddart, Sundet, Szöke, Thompson, Torrent, Zalla, Craddock, Andreassen, Leboyer, Vieta, Bauer, Worhunsky, Tzagarakis, Rogers, Geddes and Goodwin2013). The further identification – albeit less consistently – of modest dysfunction in the non-affected, first-degree relatives of affected probands (Balanzá-Martínez et al. Reference Balanzá-Martínez, Rubio, Selva-Vera, Martinez-Aran, Sánchez-Moreno, Salazar-Fraile, Vieta and Tabarés-Seisdedos2008; Bora et al. Reference Bora, Yucel and Pantelis2009) has resulted in some aspects of neurocognitive dysfunction being put forward as candidate cognitive endophenotypes for mood disorders. Due to a paucity of studies in some areas, there remains debate over the extent to which specific cognitive deficits can be viewed as true endophenotypes (i.e. heritable, co-segregating, and found in non-affected family members at a higher rate than in the general population; Gottesman & Gould, Reference Gottesman and Gould2003) rather than core illness ‘traits’, emerging consequent to the mood disorder (Glahn et al. Reference Glahn, Bearden, Niendam and Escamilla2004; Christensen et al. Reference Christensen, Kyvik and Kessing2006; Daban et al. Reference Daban, Mathieu, Raust, Cochet, Scott, Etain, Leboyer and Bellivier2012).
Impairments in facets of attentional processing have been described in many studies of neurocognitive function in mood disorders (Cohen et al. Reference Cohen, Lohr, Paul and Boland2001). Deficits have been observed in MDD and BD patients when euthymic (Paelecke-Habermann et al. Reference Paelecke-Habermann, Pohl and Leplow2005; Torrent et al. Reference Torrent, Martinez-Aran, Daban, Sanchez-Moreno, Comes, Goikolea, Salamero and Vieta2006; Preiss et al. Reference Preiss, Kucerova, Lukavsky, Stepankova, Sos and Kawaciukova2009; Robinson et al. Reference Robinson, Thompson, Gray, Young and Ferrier2013) as well as abnormalities in the activation of underlying neurocircuitry when performing attentional tasks (Strakowski et al. Reference Strakowski, Adler, Holland, Mills and DelBello2004; Mullin et al. Reference Mullin, Perlman, Versace, de Almeida, LaBarbara, Klein, Ladouceur and Phillips2012). Following the observation of deficits in first-degree relatives of BD patients, and euthymic recurrent MDD patients, attentional control (cognitive flexibility) has been suggested as a candidate endophenotype for mood disorder in general (but not actual disease phenotypes) (Clark et al. Reference Clark, Sarna and Goodwin2005b ). However, one of the most frequently examined aspects of attention in mood disorders has been vigilance (or sustained attention). Performance decrements, which increase with time-on-task, on the degraded stimulus form of the continuous performance test (CPT) in euthymic BD patients have led to the suggestion that alterations in sustained attention may be an endophenotype for BD (Ancín et al. Reference Ancín, Santos, Teijeira, Sánchez-Morla, Bescós, Argudo, Torrijos, Vázquez-Álvarez, De La Vega, López-Ibor, Barabash and Cabranes-Díaz2010). Numerous other studies have demonstrated CPT deficits in BD and MDD patients in symptomatic states (Koetsier et al. Reference Koetsier, Volkers, Tulen, Passchier, van den Broek and Bruijn2002; Porter et al. Reference Porter, Gallagher, Thompson and Young2003; Fleck et al. Reference Fleck, Eliassen, Durling, Lamy, Adler, DelBello, Shear, Cerullo, Lee and Strakowski2012; Gallagher et al. Reference Gallagher, Gray, Watson, Young and Ferrier2014) and in euthymia (Wilder-Willis et al. Reference Wilder-Willis, Sax, Rosenberg, Fleck, Shear and Strakowski2001; Liu et al. Reference Liu, Chiu, Chang, Hwang, Hwu and Chen2002; Weiland-Fiedler et al. Reference Weiland-Fiedler, Erickson, Waldeck, Luckenbaugh, Pike, Bonne, Charney and Neumeister2004; Doyle et al. Reference Doyle, Wilens, Kwon, Seidman, Faraone, Fried, Swezey, Snyder and Biederman2005; Thompson et al. Reference Thompson, Gallagher, Hughes, Watson, Gray, Ferrier and Young2005; Kolur et al. Reference Kolur, Reddy, John, Kandavel and Jain2006). CPT deficits have also been observed in some (Klimes-Dougan et al. Reference Klimes-Dougan, Ronsaville, Wiggs and Martinez2006; Trivedi et al. Reference Trivedi, Goel, Dhyani, Sharma, Singh, Sinha and Tandon2008) but not all (Clark et al. Reference Clark, Kempton, Scarna, Grasby and Goodwin2005a ; Meyer & Blechert, Reference Meyer and Blechert2005; Jabben et al. Reference Jabben, Arts, Krabbendam and Van Os2009; Walshe et al. Reference Walshe, Schulze, Stahl, Hall, Chaddock, Morris, Marshall, McDonald, Murray, Bramon and Kravariti2012) studies in first-degree relatives. A recent study found both behavioural deficits and functional magnetic resonance imaging differences (increased activation in the insula and parts of the cingulate cortex) during a CPT in euthymic BD-I patients and non-affected relatives compared with controls (Sepede et al. Reference Sepede, De Berardis, Campanella, Perrucci, Ferretti, Serroni, Moschetta, Del Gratta, Salerno, Ferro, Di Giannantonio, Onofrj, Romani and Gambi2012).
One important consideration in the assessment of attentional processes is in the method of performance measurement. In most CPTs, absolute errors, signal detection indices or mean reaction time (RT) over subcomponents or the overall task are typically used. However, increasingly there is recognition of the need to go beyond such measures and take into account inconsistency of responses or intra-individual variability (IIV). This can be achieved most simply by calculation of the standard deviation of item-by-item RT for each individual (or the individual standard deviation; iSD), although as this measure is strongly related to mean RT, the coefficient of variation (CoV) is often preferred (Jackson et al. Reference Jackson, Balota, Duchek and Head2012) which divides the iSD by the corresponding individual's mean RT. Such measures are being increasingly applied in the cognitive ageing literature (Nilsson et al. Reference Nilsson, Thomas, O'Brien and Gallagher2014), where it has been reported that IIV indices are better than mean RT in differentiating early neurodegeneration from healthy ageing (Hultsch et al. Reference Hultsch, MacDonald and Dixon2002), and are strongly related to broader cognitive function (Bielak et al. Reference Bielak, Hultsch, Strauss, MacDonald and Hunter2010) and brain white matter integrity (Fjell et al. Reference Fjell, Westlye, Amlien and Walhovd2011; Jackson et al. Reference Jackson, Balota, Duchek and Head2012). However, empirical RT distributions are fundamentally non-normal and tend to be positively skewed and there is growing interest in the utility of mathematical RT modelling to characterize dissociable components of RT distributions (Balota & Yap, Reference Balota and Yap2011).
The ex-Gaussian distribution, a mathematical convolution of a Gaussian (normal) and exponential distribution, produces a good approximation to empirical RT distributions (Schmiedek et al. Reference Schmiedek, Oberauer, Wilhelm, Su and Wittmann2007). The ex-Gaussian distribution has three parameters: mu and sigma, the mean and standard deviation of the Gaussian (normal) component; and tau, which determines the exponential component and represents the relative strength of the ‘slow-tail’ of the distribution (Ratcliff, Reference Ratcliff1979). As the ex-Gaussian model represents the distribution of RT, it can intuitively be related to ‘standard’ arithmetic properties, for example, the sum of mu and tau equals the overall arithmetic mean of the data (Ratcliff, Reference Ratcliff1979; Heathcote et al. Reference Heathcote, Popiel and Mewhort1991). This methodology has been used to model RT from a number of attentional tasks in older adults, for example, demonstrating a clear increase in the tau component in mild dementia of the Alzheimer's type compared with controls, which correlated with decreased cerebral white matter (Tse et al. Reference Tse, Balota, Yap, Duchek and McCabe2010; Jackson et al. Reference Jackson, Balota, Duchek and Head2012). More generally, RT variability has been linked to white matter integrity across the normal developmental trajectory in healthy children, adolescents and adults: maturation of white matter integrity and connectivity leading to reductions in RT IIV (Fjell et al. Reference Fjell, Westlye, Amlien and Walhovd2011; Tamnes et al. Reference Tamnes, Fjell, Westlye, Østby and Walhovd2012). Given the growing evidence of impaired white matter integrity in MDD and BD patients and those at high-risk (Heng et al. Reference Heng, Song and Sim2010; Macritchie et al. Reference Macritchie, Lloyd, Bastin, Vasudev, Gallagher, Eyre, Marshall, Wardlaw, Ferrier, Moore and Young2010; Sprooten et al. Reference Sprooten, Sussmann, Clugston, Peel, McKirdy, Moorhead, Anderson, Shand, Giles, Bastin, Hall, Johnstone, Lawrie and McIntosh2011; Henderson et al. Reference Henderson, Johnson, Vallejo, Katz, Wong and Gabbay2013; Leow et al. Reference Leow, Ajilore, Zhan, Arienzo, GadElkarim, Zhang, Moody, Van Horn, Feusner, Kumar, Thompson and Altshuler2013; Sarrazin et al. Reference Sarrazin, Poupon, Linke, Wessa, Phillips, Delavest, Versace, Almeida, Guevara, Duclap, Duchesnay, Mangin, Le Dudal, Daban, Hamdani, D'Albis, Leboyer and Houenou2014; Wang et al. Reference Wang, Leonards, Sterzer and Ebinger2014) there is a clear rationale for applying such analyses to attentional RT data in mood disorder.
Despite the potential utility of these approaches, there are very few data on IIV in mood disorders. Increased variability on the Connors CPT in manic and euthymic patients has been reported (Bora et al. Reference Bora, Vahip and Akdeniz2006), although variability was examined between average blocks of trials rather than individual RT. One study found a large effect size in the increase in RT iSD from a CPT in young BD probands and their unaffected first-degree relatives compared with matched controls (Brotman et al. Reference Brotman, Rooney, Skup, Pine and Leibenluft2009). It has been reported that RT iSD from a Go/No-go paradigm was increased in patients with schizophrenia/schizo-affective disorder, but not in those with major depression or borderline personality disorder compared with healthy controls (Kaiser et al. Reference Kaiser, Roth, Rentrop, Friederich, Bender and Weisbrod2008). To date there has been no comprehensive assessment of attentional IIV, with full RT modelling, in mood disorders.
The aim of the present study was therefore to examine RT distributions from an attentional CPT in patients with mood disorders, comparing iSD, CoV and ex-Gaussian components (mu, sigma and tau) in patients with BD (euthymia and depression), medication-free depression and healthy control participants. As the ex-Gaussian is a parametric model of an underlying theoretical distribution, Vincentile analysis was also conducted in order to demonstrate convergence across the two techniques (Tse et al. Reference Tse, Balota, Yap, Duchek and McCabe2010). This non-parametric technique directly assesses raw empirical RT distributions and makes no assumptions about an underlying theoretical distribution (by first ordering and then dividing the empirical distribution into a number of equal-sized ‘bins’ and computing the average RT in each of these bins). It was hypothesized that, overall, the mood disorder groups would show a significantly increased IIV and ex-Gaussian tau component (reflecting increased response variability, especially slowing) compared with matched controls.
Method
Individual RT datasets were collated from multiple studies conducted in the Institute of Neuroscience (Academic Psychiatry), Newcastle University which had used the same attentional task (Porter et al. Reference Porter, Gallagher, Thompson and Young2003; Thompson et al. Reference Thompson, Gallagher, Hughes, Watson, Gray, Ferrier and Young2005; Macritchie et al. Reference Macritchie, Lloyd, Bastin, Vasudev, Gallagher, Eyre, Marshall, Wardlaw, Ferrier, Moore and Young2010; Gallagher et al. Reference Gallagher, Gray, Watson, Young and Ferrier2014).
Participants
Patients aged 18–65 years with a diagnosis of BD, confirmed using the Structured Clinical Interview for DSM-IV (SCID; First et al. Reference First, Spitzer, Williams and Gibbon1995), were recruited from secondary and tertiary care services in the North East of England. All were out-patients and either currently in a depressive episode (SCID-defined) or euthymic, prospectively defined as ≤7 on both the 21-item Hamilton Depression Rating Scale (HAMD21; Hamilton, Reference Hamilton1960) and the Young Mania Rating Scale (Young et al. Reference Young, Biggs, Ziegler and Meyer1978) at initial assessment and after 1 month. Patients were excluded if they met criteria for any other current Axis I disorder (except anxiety) or substance dependence/abuse. All were receiving medication at the time of testing but this had remained stable for ≥4 weeks. For the MDD cohort, patients aged 18–65 years with a DSM-IV diagnosis of MDD, single episode or recurrent, were recruited from general practice clinics. For this latter (MDD) cohort, patients had been entirely psychotropic medication-free for at least 6 weeks before recruitment and were excluded if currently taking other medication active in the central nervous system, including beta-blockers or St John's wort, or if there was a co-morbid medical/psychiatric diagnosis, or recent alcohol/substance misuse. All were tested as soon as possible after recruitment to minimize delay in treatment. For all participants, illness characteristics, clinical ratings and medication history were determined by trained psychiatrists using full history, case-note and medication review and standardized rating scales. All studies were approved by the local National Health Service (NHS) Research Ethics Committee and all participants gave written, informed consent.
Neurocognitive testing
All participants completed the Vigil CPT (Cegalis & Bowlin, Reference Cegalis and Bowlin1991) using the same parameters. In this task, a continuous stream of random letters of the English alphabet is displayed on a computer screen. Each letter appears for 85 ms, followed by a 900 ms inter-stimulus interval (ISI) and is presented as a white letter on a black background in the centre of the screen (see Fig. 1). Participants are instructed to look out for a target sequence (an ‘A’ immediately followed by a ‘K’) and must respond ‘as quickly, but as accurately as possible’ by pressing the spacebar if this target sequence occurs. The letter ‘A’ thereby becomes the signal for the potential occurrence of a target sequence, but responses should only be made once the second letter of the sequence, ‘K’, appears. In total, 480 letters are displayed, in which 100 target sequences occur. These are pseudo-randomized between each quarter of the test, i.e. so there are 25 targets within every 120 trials (The Psychological Corporation, 1998).
Fig. 1. The Vigil continuous performance test. Stimulus timing and example of a reclassification of a late response to a target sequence (a). General response classification rules (b). * If the previous stimulus was a target, the algorithm first checked if this target already had a valid response, in which case the current response was also classified as a commission error. This path is omitted in the figure. ISI, Inter-stimulus interval; RT, reaction time; n/a, not applicable.
Data analysis procedure
Data extraction and cleaning
RT data were re-extracted from the original Vigil CPT output files and any responses were either classified as ‘valid’ or as ‘commission error’ according to their temporal relationship to the target sequenceFootnote 1 Footnote †. Response times were always measured in relation to the onset of the second stimulus of a target sequence (letter ‘K’). In contrast to the standard analysis, we classified responses as ‘valid’ even if they occurred after the onset of the letter that immediately followed a target sequence (see Fig. 1), allowing maximum response times of up to 1970 ms [i.e. (2 × ISI) + (2 × letter duration)]. However, there is one exception to this rule: since it is possible that two (or more) target sequences follow directly after another (i.e. ‘A-K-A-K’), responses to the second ‘A’ would no longer be considered valid for the initial target sequence, as such a response could be a premature response to the new target sequence. Such responses were classified as commission errors. Any other responses that could not be associated with a target stimulus according to the above rules were also classified as commission errors. Target stimuli with no detectable valid response were classified as ‘misses’.
This classification scheme ensured that responses with RT just above the ISI were considered (late) valid responses to the target, instead of resulting, according to the original scheme, in a ‘miss’-classification to the target stimulus and a commission error for the stimulus following the target. While we believe that this classification better reflects the underlying psychological processes, it is important to consider the number of misses when looking at the distribution of response times of an individual. For instance, some individuals may have been better able than others to withhold responses when they detected that those responses would be late (i.e. after the onset of the stimulus following a target), thereby restricting their maximum response times to the ‘standard’ response window. Since such behaviour would reduce the potential range of RTs and therefore RT variability, care must be taken that this reduction does not come at a cost of an increased number of misses.
IIV analysis and ex-Gaussian modelling
From valid responses, basic measures of IIV were derived using the iSD – the SD of all RTs for each individual, and the CoV – the iSD divided by an individual's mean RT. Ex-Gaussian probability density functions were fitted to the distribution of valid response times of each individual using the DISTRIB toolbox (Lacouture & Cousineau, Reference Lacouture and Cousineau2008) in MATLAB® v.R2010b (The MathWorks Inc., 2010). This toolbox uses maximum likelihood principles to estimate the ex-Gaussian distribution parameters mu, sigma and tau. Vincentile plots were also derived as a distribution-free representation of the data. For these data, RTs within each participant were ranked and eight Vincentiles derived (representing the average RT within each sequential 12.5% of valid data, from fastest to slowest). Individual Vincentiles were then averaged across participants.
Healthy control reference data
An SAS algorithm was used (Kosanke & Bergstralh, Reference Kosanke and Bergstralh1995) which sampled from the overall control cohort (n = 138) and matched controls to individual cases according to age, sex and National Adult Reading Test (NART)-estimated intelligence quotient(IQ) (Nelson, Reference Nelson1982). This created very closely matched healthy control groups for each of the three patient groups. Group analyses were made using SPSS v19 (USA).
Results
Subject demographics and clinical details
In total 297 datasets were available for analysis (see Table 1). This included 138 healthy controls (61 males, 77 females) and 159 patients. The three patient samples included: 86 euthymic bipolar patients (41 males, 45 females), 33 depressed bipolar patients (19 males, 14 females) and 39 depressed MDD patients (15 males, 24 females). Data from one further female depressed MDD patient were excluded from the analysis as only 22% valid responses were recorded for this patient. The three patient groups and their respective matched control groups were closely matched for age and NART-estimated IQ (p > 0.69 for all).
Table 1. Demographic and clinical details
Data are given as mean (standard deviation) unless otherwise indicated.
BD, Bipolar disorder; MDD, major depressive disorder; NART, National Adult Reading Test; IQ, intelligence quotient; HAMD21, 21-item Hamilton Depression Rating Scale; SCID, Structured Clinical Interview for DSM-IV.
a Each control comparison was sampled from the overall control group (see Method), so are not independent.
b SCID-diagnosed bipolar type I or II (missing for n = 5 BD depressed).
None of the patients in the MDD group was currently on psychotropic medication; 24 (62%) had never previously taken antidepressant medication; of the remaining 15 (38%), the median time since last treatment was 12 months (range 2–84 months). Of the bipolar patients, five were drug-free at the time of testing. In the euthymic sample, 76 (88%) were taking a mood stabilizer (of which n = 55 lithium), 23 (27%) antidepressant medication, and 23 (27%) antipsychotic medication. In the depressed sample, 27 (84%) were taking a mood stabilizer (of which n = 8 lithium), 26 (81%) antidepressant medication, and 15 (47%) an antipsychotic medication. Medication details of one patient were not recorded.
Response profiles
Within the original raw dataset (n = 296), a total of 29 677 individual trials were recorded, of which 28 482 (96.0%) were responses within the originally defined response window (0–985 ms). The remaining 4.0% were classified as: early (300/29 677; 1.0%), i.e. responses that occurred before the ‘K’ of an ‘AK’ target sequence; or late (201/29 677; 0.7%), i.e. ‘correct’ responses which were slow (985–1970 ms)Footnote 2 ; or misses (694/29 677; 2.3%). Examining these between patients and controls indicated that the greater proportion of early (226/300; 75.3%) and late responses (152/201; 75.6%), and misses (570/694; 82.1%) occurred in the patient sample. Comparing these directly revealed that, on average, significantly more misses occurred in all three patient samples compared with their respective control group, with depressed BD patients also making more early and late responses (see Table 2).
Table 2. Descriptive statistics for RT data and response profile
Data are given as mean (standard deviation).
RT, Reaction time; BD, bipolar disorder; MDD, major depressive disorder; iSD, individual standard deviation; CoV, coefficient of variation.
a Each respective control comparison was resampled from the overall control group (n = 138), so are not independent (see Method).
b For n = 4 datasets (1.3%), sigma was returned as 0 in the ex-Gaussian model.
c Mann–Whitney U test.
* p < 0.05, ** p ≤ 0.01, *** p < 0.0001 compared with respective control comparison data.
Following data cleaning (see above), averages of 94.4 (s.d. = 8.14) responses per participant in patients and 98.3 (s.d. = 3.11) responses per participant in controls were available for RT analysis.
Average RT
The analysis of the standard average RT showed significantly slower RT for the group of euthymic BD patients [F 1,170 = 6.322, p = 0.013; d = 0.38, 95% confidence interval (CI) 0.08–0.68; see Table 2], but not for the group of depressed BD patients (F 1,64 = 1.009, p = 0.319; d = 0.25, 95% CI −0.24 to 0.73) or the group of depressed MDD patients (F 1,76 = 0.048, p = 0.826; d = 0.05, 95% CI −0.39 to 0.49) compared with controls.
IIV indices
The various measures of RT IIV are shown in Table 2. Analysis of the iSD demonstrated significantly greater variability in patients compared with their matched control data, for euthymic BD (F 1,170 = 4.785, p = 0.030; d = 0.33, 95% CI 0.03–0.63), depressed BD (F 1,64 = 32.474, p < 0.00001; d = 1.40, 95% CI 0.85–1.92) and depressed MDD (F 1,76 = 5.662, p = 0.020; d = 0.54, 95% CI 0.08–0.99). Accounting for the overall mean RT, a significantly greater CoV was observed in depressed BD (F 1,64 = 28.824, p < 0.00001; d = 1.32, 95% CI 0.77–1.84). There was also a statistical trend for greater CoV for depressed MDD (F 1,76 = 3.545, p = 0.064; d = 0.43, 95% CI −0.02 to 0.87), but no difference in euthymic BD (F 1,170 = 0.732, p = 0.393; d = 0.13, 95% CI −0.17 to 0.43).
Ex-Gaussian analysis and Vincentile plots
The ex-Gaussian analysis indicated that there were differences across the three distribution parameters (see Table 2). No significant differences between patients and controls were observed in mu (euthymic BD: F 1,170 = 1.943, p = 0.165; d = 0.21, 95% CI −0.09 to 0.51; depressed BD: F 1,64 = 1.864, p = 0.177; d = −0.34, 95% CI −0.82 to 0.15; depressed MDD: F 1,76 = 0.301, p = 0.585; d = −0.12, 95% CI −0.57 to 0.32). No significant differences in the sigma parameter were observed for euthymic BD (F 1,170 = 1.918, p = 0.168; d = 0.21, 95% CI −0.09 to 0.51) or depressed MDD (F 1,76 = 1.901, p = 0.172; d = 0.31, 95% CI −0.14 to 0.76), but sigma was significantly increased in depressed BD (F 1,64 = 5.292, p = 0.025; d = 0.57, 95% CI 0.07–1.05). A significant increase in the exponential part of the RT distribution was observed for both BD patient groups: the tau parameter was greater in euthymic BD patients (F 1,170 = 6.604, p = 0.011; d = 0.39, 95% CI 0.09–0.69) and depressed BD patients (F 1,64 = 21.347, p < 0.0001; d = 1.14, 95% CI 0.60–1.64) compared with controls. There was also a statistical trend for greater tau in depressed MDD patients (F 1,76 = 3.034, p = 0.086; d = 0.39, 95% CI −0.06 to 0.84).
Vincentile plots are shown in Fig. 2, providing convergent support for the ex-Gaussian analyses. For the euthymic BD sample, the plots for patients and controls remain close until the last Vincentile (V8) where they diverge more sharply. This occurs more clearly in the depressed MDD and BD samples, particularly the latter. However, there are also differences evident in the first Vincentile (V1) for the depressed samples, with responses being faster in patients than controls (a difference which is significant in the BD depressed sample; p = 0.024).
Fig. 2. Vincentile plots for all clinical groups compared with matched control data: (a) bipolar disorder (BD) euthymic; (b) BD depressed; (c) major depressive disorder (MDD). V1–V8 denote each Vincentile [sequential 12.5% of reaction time (RT) data] from fastest to slowest RT. Values are means, with standard errors represented by vertical bars.
To facilitate comparison between patient groups, the ex-Gaussian parameters for euthymic BD, depressed BD and MDD groups were expressed as a z-score based on the mean and s.d. of their respective control groups. One-way analysis of variance revealed significant differences for mu (F 2,155 = 4.348, p = 0.015) and tau (F 2,155 = 15.545, p < 0.0001). Post hoc contrasts revealed that the mu parameter was significantly different between euthymic and depressed BD groups (p = 0.006) with a trend between euthymic BD and MDD groups (p = 0.085). For tau, the depressed BD group differed significantly from both euthymic and MDD groups (p < 0.001).
Receiver operating characteristic (ROC) analysis
To demonstrate the degree of differentiation between the clinical groups and controls (i.e. that differences are not consequent to extreme responses from a small number of participants), an ROC plot (Wilcoxon estimate) was used to determine the optimum cut-point to maximize sensitivity and specificity. For MDD patients, a tau value of 56.12 yielded a ROC area under the curve (AUC) = 0.60 (95% CI 0.46–0.73), with sensitivity = 0.74 and specificity = 0.44. For euthymic BD patients, a tau value of 56.35 yielded a ROC AUC = 0.62 (95% CI 0.53–0.70), with sensitivity = 0.77 and specificity = 0.44. For depressed BD patients, a tau value of 85.56 yielded a ROC AUC = 0.82 (95% CI 0.70–0.93), with sensitivity = 0.73 and specificity = 0.88. Comparing between the clinical groups, the tau parameter also differentiated depressed from euthymic BD patients with sensitivity = 0.70 and specificity = 0.71 (ROC AUC = 0.73, 95% CI 0.61–0.84), and depressed BD from depressed MDD patients with sensitivity = 0.73 and specificity = 0.65 (ROC AUC = 0.68, 95% CI 0.55–0.82).
Relationship to severity of depression
Exploratory Spearman's correlations were performed separately for each patient group, between IIV parameters and the HAMD21. No significant correlations between iSD, CoV or ex-Gaussian parameters were observed in euthymic (−0.073 ≤ r s ≤ 0.188, p > 0.080 for all) or depressed BD patients (−0.135 ≤ r s ≤ −0.017, p > 0.450 for all). For MDD patients, a near-significant positive correlation between depression severity and CoV was observed (r = 0.314, p = 0.051).
Discussion
The present study investigated RT IIV during a simple sustained attention task in three groups of patients with mood disorders: euthymic BD, depressed BD and depressed MDD. All three groups showed evidence of increased response variability compared with matched controls. Euthymic BD patients had greater values of iSD and tau, but not in CoV or sigma. Together with the fact that this group also showed greater standard average RT, but not in the fitted mu parameter, these results indicate that the differences between these patients and controls is best characterized as in increase in the exponential part of the RT distribution (i.e. an increased number of ‘disproportionately slow’ responses), as this would cause a shift in mean RT and iSD but not in CoV. Depressed BD patients showed the most consistent evidence of increase in RT variability, as all four indices of variability (iSD, CoV, sigma and tau) were significantly increased in comparison with the healthy control sample. It may at first seem surprising that there was no significant increase in average RT in this group as a result of increased variability. However, as can be seen in the Vincentile plot of this group, the increase in variability was due not only to an increase in the number of slow responses (similar to euthymia), but also the number of fast responses (although not to a sufficient extent to alter mu). Depressed MDD patients showed the weakest evidence for a RT variability increase. While the iSD was significantly higher in this group, both the CoV and the tau parameter showed only statistical trends for larger values. There were no differences in average RT, mu or sigma.
These data are in line with previous reports of increased IIV in attentional performance in BD (Bora et al. Reference Bora, Vahip and Akdeniz2006; Brotman et al. Reference Brotman, Rooney, Skup, Pine and Leibenluft2009). However, to our knowledge this is the first study to comprehensively examine RT distribution parameters and IIV across patients with mood disorders. Previous studies have applied ex-Gaussian RT modelling to tasks in children and adolescents with attention-deficit/hyperactivity disorder (ADHD). The tau parameter has been suggested to produce excellent differentiation between ADHD and controls (Leth-Steensen et al. Reference Leth-Steensen, King Elbaz and Douglas2000). Subsequent findings suggest that there are differences in all three parameters compared with controls, with more variability (sigma) and increases in mu and particularly slow (tau) responses – the latter suggested to reflect attentional lapses in some but not all trials (Hervey et al. Reference Hervey, Epstein, Tonev, Arnold, Conners, Hinshaw, Swanson and Hechtman2006). In the present study, while there was no significant difference in mu between patients and controls, the Vincentile plots did indicate some evidence of faster responses in V1 in the depressed samples (which was significant in BD depression). There was also a significant increase in the number of misses in all patient groups (and early and late responses, in depressed BD), compared with controls. This general inconsistency combined with the frequency of disproportionately slow responses is again consistent with ‘phasic’ attentional task engagement/disengagement. This has been suggested previously during CPT task performance in euthymic BD patients (Robinson et al. Reference Robinson, Thompson, Gray, Young and Ferrier2013). Functional imaging has further revealed that while prefrontal activation occurs early during CPT performance in mania, it cannot be maintained over sustained periods (Fleck et al. Reference Fleck, Eliassen, Durling, Lamy, Adler, DelBello, Shear, Cerullo, Lee and Strakowski2012).
An area of ongoing debate is the extent to which RT distribution characteristics can be linked to specific aspects of neurocognitive function. For example, the utility of ex-Gaussian modelling has been demonstrated across different conditions of the classic Stroop test, revealing attentional shifts which would otherwise be missed with outcomes based on simple central tendency (Heathcote et al. Reference Heathcote, Popiel and Mewhort1991). These authors suggest that no direct attribution can be made between ‘parameter and process’ and while ‘the ex-Gaussian model describes RT data successfully, it does so without the benefit of an underlying theory’ (Heathcote et al. Reference Heathcote, Popiel and Mewhort1991). However, more recently it has been proposed that the tau parameter is strongly related to ‘higher’ cognitive functions (a statistical composite measure of working memory tasks and reasoning) and is therefore a marker of individual differences in attentional/executive control (Schmiedek et al. Reference Schmiedek, Oberauer, Wilhelm, Su and Wittmann2007). As work in this area progresses – and if IIV and ex-Gaussian measures are applied more frequently in clinical studies – it may be possible to derive more precise theoretical accounts, informing our understanding of neurocognition in mood disorders.
A strength of the present study was the assessment of IIV and application of RT modelling to one single attentional CPT which had been used consistently in a series of studies in the same research centre. However, it should be noted that in addition to attention, other cognitive processes such as processing speed have been assessed as putative cognitive endophenotypes in BD (Antila et al. Reference Antila, Kieseppä, Partonen, Lönnqvist and Tuulio-Henriksson2011; Daban et al. Reference Daban, Mathieu, Raust, Cochet, Scott, Etain, Leboyer and Bellivier2012). One caveat is that most studies have used the digit–symbol task as an index of processing speed, but this measure is known to involve multiple interacting lower-level and higher-level cognitive control processes, including executive control and attention (Cepeda et al. Reference Cepeda, Blackwell and Munakata2013). Therefore when utilizing such tasks in the search for candidate endophenotypes, especially if proposing process-specificity, it is necessary to consider more precisely the cognitive processes underpinning performance on any given measure. It is also important to ascertain whether IIV and shifts in the RT distribution in mood disorders are sensitive to the demand characteristics of tasks, such as rate of presentation or cognitive load, and therefore whether they are related more to impairments in attentional control or basic processing efficiency.
Other methodological considerations should be highlighted. The present study utilized a large normative reference sample from which control data were selected by computer algorithm and demographically matched to individual patient cases. This ensured very close group-wise matching of patients and controls which was independent of experimenter selection. The majority of BD patients in the present study were taking psychotropic medication at the time of testing. While several studies have reported minimal effects of medication on performance (Goswami et al. Reference Goswami, Sharma, Varma, Gulrajani, Ferrier, Young, Gallagher, Thompson and Moore2009; Bourne et al. Reference Bourne, Aydemir, Balanzá-Martínez, Bora, Brissos, Cavanagh, Clark, Cubukcuoglu, Dias, Dittmann, Ferrier, Fleck, Frangou, Gallagher, Jones, Kieseppä, Martínez-Aran, Melle, Moore, Mur, Pfennig, Raust, Senturk, Simonsen, Smith, Soares, Soeiro-de-Souza, Stoddart, Sundet, Szöke, Thompson, Torrent, Zalla, Craddock, Andreassen, Leboyer, Vieta, Bauer, Worhunsky, Tzagarakis, Rogers, Geddes and Goodwin2013), the potential impact of medication on performance should be considered and replication in medication-free samples or in cohorts large enough to perform subgroup analysis is needed. The depressed MDD sample in the present study was entirely psychotropic medication-free at the time of testing and some evidence of increased IIV was observed, specifically iSD, but the ex-Gaussian parameters were not significantly different from controls (although tau was increased at a trend level, with a small to medium effect size). Differences in clinical characteristics (see Porter et al. Reference Porter, Gallagher, Thompson and Young2003), such as medication, age (the MDD patients were younger) and number of episodes (the majority of MDD patients being first-episode) mean that comparisons need to be interpreted cautiously. Similarly the inherent difficulty in how to equate stage of illness and other clinical characteristics between MDD and BD in order to reliably compare them should also be noted, along with the issue of statistical power in relation to the sample size characteristics.
The clearest comparison between IIV parameters can be made between the BD groups. It is of note that variability is evident in euthymia (as increased iSD and tau) but increases in depression, reflected in the additional increase in CoV and sigma. It would be of interest for future studies to explore the potential neurobiological mechanisms underlying such effects. For example, it has been demonstrated in animal and human models that corticosteroid (cortisol) levels can exert both positive and negative effects on attention, depending on the relative occupancy of corticosteroid receptors (Lupien & McEwen, Reference Lupien and McEwen1997). Given the evidence of hypothalamic–pituitary–adrenal (HPA) axis dysfunction and hypercortisolaemia in BD (Rybakowski & Twardowska, Reference Rybakowski and Twardowska1999; Gallagher et al. Reference Gallagher, Watson, Smith, Young and Ferrier2007), which is present in euthymia but worse in depression, examining the hypothesized role of systems such as the HPA axis and their potential for causing or exacerbating state-related effects is warranted.
Due to the methodological issues outlined it remains to be established if specific features of cognitive processes, such IIV in sustained attention, could be considered as cognitive endophenotypes. It has previously been suggested that impairment on tasks such as the CPT is more an indicator of general brain dysfunction, underpinning the attentional system, than a disorder-specific marker (Rosvold et al. Reference Rosvold, Mirsky, Sarason, Bransome and Beck1956; Riccio et al. Reference Riccio, Reynolds, Lowe and Moore2002). Given the strong relationship that has been identified between IIV and white matter, it is possible that some measures of IIV or components of the RT distribution such as tau, are sensitive markers of general white matter integrity (Fjell et al. Reference Fjell, Westlye, Amlien and Walhovd2011; Jackson et al. Reference Jackson, Balota, Duchek and Head2012; Tamnes et al. Reference Tamnes, Fjell, Westlye, Østby and Walhovd2012). These links warrant detailed exploration in future studies – especially in combination with focused processing speed and attentional assessment – to ascertain the utility of these measures as markers of structural and functional integrity in a variety of clinical disorders in which white matter impairments are implicated, such as neurodegenerative and mood disorders (Sachdev et al. Reference Sachdev, Wen, Christensen and Jorm2005; Assareh et al. Reference Assareh, Mather, Schofield, Kwok and Sachdev2011; Poletti et al. Reference Poletti, Bollettini, Mazza, Locatelli, Radaelli, Vai, Smeraldi, Colombo and Benedetti2015). Including assessment in individuals with genetic risk, for example for mood disorder, will further inform the extent to which they can be considered endophenotypic markers (Hasler et al. Reference Hasler, Drevets, Gould, Gottesman and Manji2006). Developing understanding of the relationship between specific cognitive processes and their structural and functional underpinnings has clear clinical implications, especially in the potential use of neurocognitive function in the stratification of mood disorders (Insel et al. Reference Insel, Cuthbert, Garvey, Heinssen, Pine, Quinn, Sanislow and Wang2010).
The present study has demonstrated increased RT IIV in sustained attention in mood disorders. Further analysis of RT distribution parameters revealed differences in the parameters affected between MDD and BD, and depression and euthymia in BD. These data highlight the utility of applying measures of IIV to characterize cognitive variability and the potential for future application in studies examining neurocognitive dysfunction and its underlying functional and structural brain connectivity in mood disorder.
Acknowledgements
We are grateful to the participants who contributed to the research and to those clinicians involved in the wider research programme, including recruitment and screening: Niraj Ahuja, Sankalpa Basu, Jane Carlile, Louise Golightly, Thiyyancheri Harikumar, Patrick Keown, Samer Makhoul, Anuradha Menon, Gavin Mercer, Rajesh Nair, Bruce Owen and Nanda Palanichamy. This work was supported by grant funding from the Stanley Medical Research Institute (reference: 03T-429) and the Medical Research Council (reference: GU0401207). P.G., I.N.F., R.H.M.-W. and S.W. received Research Capability Funding from the Northumberland, Tyne and Wear NHS Foundation Trust which also supported this project.
Declaration of Interest
None.
