74 Journal of Neurology, Neurosurgery, and Psychiatry 1997;63:74–82 J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.63.1.74 on 1 July 1997. Downloaded from Abnormal response to negative feedback in unipolar depression: evidence for a diagnosis specific impairment

R Elliott, B J Sahakian, J J Herrod, T W Robbins, E S Paykel

Abstract scribed this mechanism as a “catastrophic Objectives—To assess in further detail the response to perceived failure” (after Beats et al3 ) specific form of motivational impairment and suggested that this eVect represents an influencing neuropsychological perform- important link between cognitive and emo- ance in depression—oversensitivity to tional processes aVecting the performance of perceived failure. The present study con- depressed patients. siders two questions: firstly whether this is In depression, various psychological mecha- specific to depression and secondly how nisms closely related to clinical symptoms may the eVect relates to clinical features. influence cognitive function. An often cited Methods—Unipolar depressed patients example is that many neuropsychological and matched controls were assessed on impairments may be due to a more fundamen- two neuropsychological tests giving ex- tal deficit of motivation. Psychotic patients in plicit performance feedback. The data general, and depressed patients in particular, were analysed in two separate studies to do show motivational deficits influencing consider the questions above. The first cognition.8 These authors attempted to distin- study considered the specificity of the guish between the “computational” and “ener- eVect to depressed patients, using data on getical” (including motivational) aspects of the same tests collected from other patient information processing, described by Hockey groups. The second study was a longitudi- et al9 and argued that the energetical aspects nal assessment of the depressed patients were impaired in depression. However, Rich- on clinical recovery to determine whether ards and RuV10 have shown that manipulations the eVect is specific to the depressed state. of motivation do not necessarily aVect cogni- Results—The eVect was not seen in non- tive performance. depressed patient groups, either Motivation should not be considered as a neurological or psychiatric groups. The unitary concept, rather as one made up of vari- longitudinal study showed a residual ab- ous interrelated components. One of these, and normal response to negative feedback on the one which will be considered in the present clinical recovery. paper, is the role of reinforcement in control- http://jnnp.bmj.com/ Conclusions—Abnormal response to ling behaviour. Many neuropsychological tests Department of negative feedback is specific to a primary give subjects feedback for their responses and Experimental diagnosis of depression and may be a trait there is some evidence to suggest that the Psychology, University rather than a state factor of the disorder. response of depressed patients to this feedback of Cambridge, These results are discussed in relation to Downing Street, is diVerent from that of control subjects. the putative neuropathology of depression 11 Hughes et al showed that depressed patients Cambridge CB2 3EB, and also to cognitive and behavioural UK were less responsive to reward than controls. on September 29, 2021 by guest. Protected copyright. R Elliott accounts of the disorder. The findings They used a progressive reinforcement sched- T W Robbins presented here have important implica- ule for financial reward and found that tions for establishing a link between mood depressed subjects worked less for reward and Department of and cognition in unipolar depression. Psychiatry, University earned less. There is also evidence that subjects of Cambridge School judge their performance in a more negative way of Clinical Medicine, (J Neurol Neurosurg Psychiatry 1997;63:74–82) than controls. Wener and Rehm12 gave de- Addenbrookes pressed and non-depressed students either Keywords: negative feedback; depression; CANTAB Hospital, Cambridge 20% or 80% success feedback on a word CB2 2QQ, UK tests; mood and cognition B J Sahakian association task then asked for an estimate of J J Herrod their success rate. Depressed patients reported E S Paykel Numerous studies have shown the presence of a lower self assessed success rate than controls. wide ranging neuropsychological deficits asso- Thus it seems that depressed patients show a Correspondence to: 1–7 Dr B J Sahakian, ciated with unipolar depression, but a coher- bias in their response to reinforcement. It is a Department of Psychiatry, ent account of these deficits has proved elusive. reasonable prediction that depressed patients One reason for this is that it is not clear to what will respond diVerently to negative feedback School of Clinical Medicine, than controls, resulting in altered cognitive Addenbrookes Hospital, extent these neuropsychological impairments Cambridge CB2 2QQ, UK. may be mediated by abnormal psychological performance. mechanisms. In our recent study of unipolar In a recent study of depression in elderly Received 10 June 1996 and depression,1 we identified a specific psychologi- people, Beats et al3 found that these patients in final revised form 5 March 1997 cal mechanism which can have a general eVect showed what was interpreted as a “catastrophic Accepted 6 March 1997 on neuropsychological performance. We de- response to executive failure”. On a test of Negative feedback and depression 75 J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.63.1.74 on 1 July 1997. Downloaded from

planning, the Tower of London test from the Table 1 Characteristics of groups whose performance was CANTAB battery,13 14 the patients solved as analysed on the delayed matching to sample task many problems correctly as controls but once Patients they had made a mistake on a problem, their (n) Mean age Mean NART-IQ performance deteriorated rapidly. Thus al- Depressed 28 48.1 (1.2) 111.8 (1.7) though they solved as many problems within Control 22 50.0 (1.7) 113.7 (1.3) the minimum number of moves as the controls, Parkinson’s 17 61.1 (1.8) 108.2 (2.2) they solved fewer within the maximum number Temporal lobe 23 34.2 (1.6) 107.9 (1.7) Schizophrenic 11 43.4 (3.3) 110.4 (3.0) of moves. This eVect suggests a link between psychological and neuropsychological impair- Values in parentheses are SEM. ments and our recent study of middle aged patients with unipolar depression1 examined it to negative feedback, perhaps because they more systematically. experience more of it during the tests than We focus here on two tests; the delayed controls. The finding may be an artefact due to matching to sample test (DMTS) from the raised overall failure rates, and therefore may CANTAB battery15 16 and a novel one touch not have any import for the understanding of Tower of London test17 based on the planning the neuropsychology of depression. To con- task from the CANTAB battery used by Beats sider this question, the performance of non- et al.3 As discussed in Elliott et al,1 the depressed groups with impairments on these depressed patients showed pronounced deficits tests was considered. on both tasks. On the delayed matching to sample test, patients were significantly im- METHOD paired under all delay conditions, including the Subjects simultaneous condition. On the new Tower of Depressed patients—The depressed patients London test, they were significantly impaired were the 28 inpatients and outpatients de- at all levels of diYculty. scribed by Elliott et al,1 of whom 19 were These tasks have very diVerent neural women and eight were men. All patients met substrates including temporal lobe structures DSM-III-R20 criteria for major unipolar de- in the first case18 but parietal and frontal pressive disorder, and patients with concurrent regions in the second (based on PET diagnoses of psychoactive substance misuse or evidence19). However, they have certain proce- neurological or general medical disorders likely dural similarities; both require subjects to to aVect cognition (for example, dementia, process complex stimuli and then make a single stroke, Parkinson’s disease, head injury, un- choice response from an array of alternatives; treated thyroid disease) were excluded. Four- both present easier problems interspersed with teen of the patients were in their first episode of more diYcult ones and, crucially, both give depression and 14 had recurrent depressive subjects immediate and explicit feedback. disorder. Phenomenology was recorded using Thus subjects are given feedback to one prob- the present state examination. Patients were lem and then are immediately required to not tested within three months of receiving process the subsequent problem. We found electroconvulsive therapy. Patients taking anti-

that on both tests, failure on one problem in the depressant medication were included, but not http://jnnp.bmj.com/ depressed subjects significantly raised the those taking neuroleptics or steroids, or those probability of failure on the subsequent having taken benzodiazepines in the 24 hours problem relative to controls.1 before testing. All the patients were taking The present study aimed to develop these medication; five clomipramine, four amitriptyl- findings more fully by introducing data from ine, four dothiepin, one lofepramine, three ser- other patient groups and from a longitudinal traline, two fluoxetine, two paroxetine, two study of depression. This study had two moclobemide, two lithium with dothiepin, one

specific aims. The first was to determine lithium with amitriptyline, and one lithium on September 29, 2021 by guest. Protected copyright. whether the abnormal response to perceived with phenelzine. failure is relatively specific to a diagnosis of Control subjects—The 22 control subjects (as depression. It is possible that the eVect seen described by Elliott et al1) matched the patients was due to the raised overall failure rate in for age and premorbid IQ (as measured by the depressed patients compared with controls national adult reading test, NART21) (tables 1 simply because exposure to higher rates of and 2). Fifteen (68.2%) of the control subjects negative feedback alters response to it. Sec- were women and seven were men. Control ondly, this paper assesses the relation between subjects were excluded if they showed any evi- abnormal response to perceived failure and dence of psychiatric or neurological history, clinical variables in depression, considering substance misuse, or were taking any concur- whether it represents a trait or a state factor. rent medication which could influence cogni- These questions are considered in two separate tion. The patients and controls did not diVer in studies. terms of male:female ratio, or in terms of age or NART-IQ, as shown by unpaired t tests Study 1 (t=0.63, P>0.25; and t=1.06, P>0.5 respec- COMPARISON WITH OTHER PATIENT GROUPS tively). It is important to consider whether the abnor- Groups of neurological patients and intellec- mal response to negative feedback is specific to tually intact patients with (all depression. It may be, for example, that a non- scoring above 85 on the WAIS-R and with cur- depressed patient group who showed impair- rent IQs within 10 points of premorbid IQs) ments on the tests would respond abnormally who had been tested on the DMTS and Tower 76 Elliott, Sahakian, Herrod, Robbins, Paykel J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.63.1.74 on 1 July 1997. Downloaded from

Table 2 Characteristics of groups whose performance was delays were presented in a randomised order, the analysed on the new Tower of London task same order for each subject. New Tower of London task—(see Owen et al17 n Mean age (y) Mean NART-IQ for a full description) Subjects were presented Depressed 28 48.1 (1.2) 111.8 (1.7) with two arrays of three coloured balls and Control 22 50.0 (1.7) 113.7 (1.3) Parkinson’s 15 59.8 (1.7) 109.6 (2.2) were asked to work out, using specified rules, 12 50.9 (3.6) 114.7 (2.8) the minimum number of moves it would take Schizophrenic 12 43.4 (3.3) 110.4 (3.0) to achieve the goal arrangement. This diVers Values in parentheses are SEM. from earlier Tower of London tasks in that it requires subjects to mentally manipulate the of London tasks were selected to match the arrays without physically moving the balls. depressed group for overall performance on the There were a total of 20 problems of one to five specific tests rather than for age or NART-IQ. moves, presented in a random order, the same For this purpose there was a strong rationale order for each subject. for using respectively patients with frontal lobe excision for making comparisons on the Tower Data analysis of London test and patients with temporal lobe As discussed in Elliott et al1 various post hoc excision for making comparisons on the analysis methods could be used to assess DMTS test as these two groups show selective whether subsequent performance is influenced impairments on the two tests.17 18 We also by feedback. The method we chose was one of selected two relevant neuropsychiatric groups the simplest, which considered the “condi- for comparison with the depressed patients tional probability”of a subject failing a problem who exhibit deficits on both tasks—patients in the DMTS or new Tower test given that they with Parkinson’s disease17 22 or schizophrenia.23 had failed the directly preceding problem. Condi- The comparison with Parkinson’s disease is tional probabilities are usually calculated by important because it represents a form of Bayes’ theorem, which states that the probabil- “subcortical (or frontosubcortical) dementia”24 ity of event B occurring given that event A has and depression has also been considered in the already occurred (denoted p(A I B)) is the same category on the grounds of overlap in probability of both events occurring (p(AB)) clinical features and possible neural substrates, divided by the probability of A occurring. If including basal ganglia pathology.25 26 The events A and B are independent, this will be the comparison with schizophrenia is informative same as the probability of B occurring. because it represents a distinct form of psycho- In the present analysis, event A is failing a sis, with likely pathology in the temporal lobe, problem, say problem x, and event B is failing as well as in frontostriatal circuitry.27 28 Full the subsequent problem, problem x+1. There- details of the neurological groups, including fore the conditional probability is: basic neuropsychological performance, have p (x and x + 1 wrong) been published recently.17 18 22 23 Tables 1 and 2 p(x+1wronglxwrong) = show the demographic characteristics and p (x wrong) overall performance of these patients. This is equivalent to dividing the number of

errors made immediately after another error by http://jnnp.bmj.com/ Procedure the total number of errors made by that Clinical evaluation—Severity of depression was subject. As the calculation involves dividing by assessed with the Hamilton depression scale the overall probability of failure, the diVerence (Ham-D29) and the Montgomery Asberg de- in overall failure rate is controlled for. If, for pression rating scale (MADRS30). Mean scores example, a patient failed twice as many of the patient group were as follows: Ham-D problems as a control (say 2x as opposed to x), 22.4 (SD 0.8), range 15 to 35; MADRS 34.1 and if there was no diVerence in the eVect of

(SD 1.1), range 25 to 46. failure on subsequent performance they would on September 29, 2021 by guest. Protected copyright. Neuropsychological evaluation—Patients were be expected to fail twice as many problems assessed on a battery of computerised neu- after other failed problems as controls (say 2y ropsychological tests including core tests from as opposed to y). Thus the conditional the CANTAB battery, which has been dis- probability for patients would be 2y/2x which is cussed in detail elsewhere (for example, exactly the same as that for controls; y/x. This Robbins et al16). Results from the full battery method of analysing the data would be were discussed in our recent paper.1 For the inappropriate if the performance of the patient purposes of the present discussion, two tests group was approaching floor level; however, all are of particular relevance. the patients were performing well above this. Simultaneous and delayed matching to Having calculated this probability for all sample—This test has been described in detail subjects, univariate F tests were used to elsewhere.16 Subjects were presented with a compare the values for depressed patients and complex abstract pattern and then asked to the other patient groups. These were calculated pick this pattern from a choice of four similar using the statistical package for social sciences stimuli. In simultaneous matching to sample (SPSS26). (MTS) the sample stimulus remained on the screen while choice stimuli appeared. In the RESULTS delayed matching to sample (DMTS) task, the EVects of negative feedback sample stimulus disappeared before the choice As described in detail in Elliott et al,1 depressed stimuli were presented after delays of 0, 4, or 12 patients solved fewer problems correctly than seconds. The MTS and diVerent DMTS controls in both the tests (fig 1A). Conditional Negative feedback and depression 77 J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.63.1.74 on 1 July 1997. Downloaded from

100 A

90

80

70 Correct (%)

60

50 DMTS NTOL

0.3 B

0.2

0.1 Conditional probability

0.0 DMTS NTOL Figure 1 The percentage of problems solved correctly (A) and conditional probabilities of failing problem x+1 having just

failed problem x (B) on the DMTS and one touch Tower tests for depressed patients, controls, patients with Parkinson’s http://jnnp.bmj.com/ disease, neurosurgical patients, and patients with schizophrenia. Bars represent SEM. probabilities of failing a problem having failed problem given that they had failed the preced- the preceding problem are shown in the ing problem. One way ANOVA showed a relevant columns of fig 1B. When these significant main eVect of group (F(4,97) = 8.5, probabilities were compared, there was a P<0.001). Post hoc Newman-Keuls tests significant main eVect of group, both on showed that this eVect was due to the DMTS (F(1,48) = 25.0, P<0.001) and on new depressed patients diVering from the other four Towe r (F(1,48) = 19.5, P<0.001). For the groups at the 0.01 significance level. The other on September 29, 2021 by guest. Protected copyright. patient group these conditional probabilities patient groups did not diVer significantly from for the two tests correlated significantly with the controls. each other (r= 0.53, P<0.01). For the overall percentage correct score on the new Tower of London task, a one way, Comparison with non-depressed groups between subjects ANOVA performed on the For the overall percentage correct score on the DMTS task, one way, between subjects analy- scores of the depressed patients, controls, sis of variance (ANOVA) was performed on the patients with Parkinson’s disease, patients with scores of the depressed patients, controls, frontal lobe lesions, and patients with schizo- patients with Parkinson’s disease, patients with phrenia. This analysis showed that there was a temporal lesions, and schizophrenia. The significant main eVect of group (F(4,84) = 4.7, analysis showed that there was a significant P<0.01). Post hoc Newman-Keuls tests main eVect of group (F(4,97) = 9.7, P<0.001). showed that this eVect was due to all four Post hoc Newman-Keuls tests showed that this patient groups diVering from controls at the eVect was due to all four patient groups diVer- P<0.05 level of significance but not diVering ing from controls at the P<0.01 level of signifi- from each other. Thus the patient groups were cance but not diVering from each other. Thus roughly matched for overall level of impair- the patient groups matched for overall level of ment (fig 1A). impairment relative to controls (fig 1A). Figure 1B shows the conditional probabili- Figure 1B shows the conditional probabili- ties of failing a new Tower problem given that ties for these five groups of failing a DMTS they had failed the preceding problem for these 78 Elliott, Sahakian, Herrod, Robbins, Paykel J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.63.1.74 on 1 July 1997. Downloaded from

four groups. One way ANOVA showed a that patients and controls did not diVer in significant main eVect of group (F(4,84) = 3.7, terms of age or IQ. P<0.005). Post hoc Newman-Keuls tests showed that this eVect was due to the Procedure and data analysis depressed patients diVering from the other four Clinical evaluation—The recovered patients groups at the 0.05 significance level. The other were assessed with the Hamilton depression patient groups did not diVer from the controls. scale29 and the Montgomery Asberg depression These analyses show that the abnormal rating scale.30 Mean scores were 3.0 (SEM response to negative feedback is relatively spe- 0.75) on the Ham-D and 5.1 (SEM 2.3) on the cific to depression. They clearly show that the MADRS. eVect is not simply an artefact due to patients Neuropsychological evaluation and data failing more problems overall, as other groups analysis—Subjects completed the same test of patients matched for overall level of impair- battery as at initial assessment, including the ment do not show the same eVect. It also seems DMTS and Tower of London tasks. Response that the abnormal response to negative feed- to perceived failure was assessed by analysing back is not a general trait of psychiatric conditional probabilities as described above. patients, as the eVect was not seen in the For simplicity the conditional probabilities on patients with schizophrenia, but is specific to the two tests were summed to give an overall depression. “response to negative feedback score”. These scores, together with percentage correct scores, for patients and controls were compared using Study 2 repeated measures ANOVA. RELATIONS BETWEEN ABNORMAL RESPONSE TO NEGATIVE FEEDBACK AND CLINICAL VARIABLES RESULTS Given that the abnormal response to negative Follow up of patients on clinical recovery feedback is an influence on performance which Figure 2A shows the overall percentages seems to be specific to a diagnosis of correct on the DMTS and Tower of London depression, it is important to determine how it tests. Repeated measures ANOVA of the may relate to clinical symptoms. In particular, DMTS scores showed significant eVects of the eVect may be either a state or a trait factor group (F(1,26) = 10.8, P<0.01) and test of the disorder. session (F(1,26) = 5.7, P<0.05). Depressed patients performed worse overall than controls METHODS and both groups performed better on the Subjects second session than the first. There was also a Depressed patients on clinical recovery—In the near significant group x test session interaction follow up stage of this longitudinal study, the (F(1,26) = 4.1, P=0.054) due to the improve- patients in the group described above were ment in patients’ performance on clinical monitored after the initial test session until recovery, which would probably reach signifi- they were judged to be clinically recovered by a cance in a larger sample of patients. Repeated consultant psychiatrist. Additionally, clinical measures ANOVA of the Tower of London

recovery was defined as a score of 8 or less on scores showed a significant main eVect of http://jnnp.bmj.com/ the Hamilton depression scale. Of the 28 group (F(1,26) = 9.1, P<0.01), with patients patients originally studied, one committed sui- solving fewer problems than controls. There cide, two were lost to follow up, two withdrew was also a significant group x test session inter- from the study, and nine did not recover within action (F(1,26) = 5.1, P<0.05) which simple the available time scale (most showed improve- main eVects analysis showed was due to ment but did not fall within the defined range patients performing much less accurately than for clinical recovery). This left 14 patients who controls at initial assessment (F(1,26) = 10.9,

did recover, of whom 10 were female and four P<0.01) and to patients performing signifi- on September 29, 2021 by guest. Protected copyright. were male. These patients were retested cantly better on clinical recovery than when between five and 18 months (mean 12.6 depressed (F(1,26) = 9.3, P<0.01). months) after the initial test session. The mean Figure 2B shows the conditional probabili- age of these patients was 47.9 (SEM 1.5) and ties of failing problem x+1 having failed prob- their mean NART-IQ was 110.7 (2.5). As cur- lem x on the DMTS and Tower of London rently accepted practice is for patients to tests at initial assessment and follow up. These continue to take medication for at least six probabilities were summed to give a total months after recovery and then to withdraw “response to negative feedback score” as gradually, these patients were still taking described above. Two way repeated measures antidepressant medication at the time of follow ANOVA showed only a significant main eVect up. of group (F(1,26) = 13.3, P<0.01). Figure 2B Control subjects—Fourteen of the original shows a trend towards improvement on clinical controls described in Elliott et al ,1 10 women recovery and in a larger group of patients this and four men to match the patients, were may have reached significance; however, the retested between six months and two years crucial finding is that clinically recovered (mean 14.1 months) after the initial test patients are still impaired relative to controls. It session, such that the mean time elapsed should also be noted that recovered patients between test sessions was the same for both show abnormal response to negative feedback groups. The mean age of these controls was on the new Tower of London test even though 50.2 (SEM 2.3) and their mean NART-IQ was their performance level was not significantly 114.5 (SEM 1.6). Univariate F tests showed diVerent from that of controls. This is further Negative feedback and depression 79 J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.63.1.74 on 1 July 1997. Downloaded from

100 A

90

80

70 Correct (%)

60

50 DMTS NTOL

Depressed 0.3 Recovered B Controls 1 Controls 2

0.2

0.1 Conditional probability

0.0 DMTS NTOL Figure 2 (A)The mean total percentages correct for the depressed patients at initial assessment and on recovery with

control patients tested across two sessions on the DMTS and one touch Tower tasks. (B) The corresponding mean http://jnnp.bmj.com/ conditional probabilities of failing problem x+1 having failed problem x. Bars represent SEM.

evidence that raised conditional probability of groups (F(1,21) = 4.72, P<0.05), with the subsequent failure is not simply a consequence group who subsequently recovered showing a of a higher overall failure rate. less abnormal response than the group who did The Pearson product moment correlation not recover, with total scores of 0.39 (SEM coeYcients were calculated between the re- 0.07) and 0.66 (SEM 0.1) respectively. sponse to failure score and the recovered scores on September 29, 2021 by guest. Protected copyright. on the Ham-D and MADRS clinical rating Comparison between first episode and recurrent scales to assess whether these residual impair- depression ments may be related to residual depression. There was no significant diVerence between These correlations were not significant at the the total response to failure scores of patients P<0.05 level. with first episode depression and those with recurrent depression (0.59 (SEM 0.1) and Comparison of recovered with non-recovered 0.49 (SEM 0.08 respectively, F(1,26)=1.7, patients P>0.05). It is possible that clinical recovery is predicted byadiVerent pattern of neuropsychological Discussion performance while depressed. To study this, The results of this study strongly suggest that the performance (at initial assessment, not on one of the general processes influencing cogni- recovery) of the 14 patients who did recover in tive performance in depressed patients is a the time available was compared with that of highly specific form of deficit involving an the nine who did not. A one way ANOVA was abnormal response to negative feedback. Our carried out with group (recovered v non- previous study1 explicitly showed, we think for recovered) as the between subject variable and the first time, that failure on problem (x) delay or diYculty level as the within subject dramatically increases the chance of failure on variable where appropriate. There was a problem (x+1) in depressed patients relative to significant diVerence between the total re- controls. The present study shows, crucially, sponse to negative feedback score of the two that this eVect is specific to depression within 80 Elliott, Sahakian, Herrod, Robbins, Paykel J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.63.1.74 on 1 July 1997. Downloaded from

the context of the comparison groups used probability of failure, 0.1). By contrast, these here. Other pertinent patient groups with the two measures are much more similar in the same overall failure rate did not show the same depressed group (conditional probability of eVect. This shows firstly, that the eVect cannot failure 0.26, average probability of failure be an artefact of the raised failure rates and 0.23). Therefore these results could be inter- secondly, that the eVect is relatively specific to preted as reflecting an unaltered response to depression. Further evidence that the eVect is negative feedback in the depressed group rela- not an artefact due to raised failure rate is that tive to controls and to non-depressed patient it remained significant in clinically recovered groups. In this study, control subjects appar- patients whose overall performance level had ently responded to negative feedback by reduc- improved. Although there was a strong trend ing errors on the subsequent trials. This eVect for response to negative feedback to improve could therefore represent either the enhanced on clinical recovery, it remained significantly use of knowledge of results to monitor and diVerent from controls. This residual abnor- improve performance, an increased motivation mality suggests that the eVect may be a trait to respond for positive reinforcement, or a rather than a state factor for depression. How- combination of these cognitive and emotive ever, it might simply reflect the greater factors. There is some evidence to support this sensitivity of this state measure over others. It account from response latencies to problems would be interesting, therefore, to pursue the after a failed problem. In control subjects, these trait hypothesis in a prospective study. response latencies showed a strong trend to be It should be noted that the present study has significantly longer than overall response laten- not assessed whether the altered response to cies (P<0.08) to problems of the same level of negative feedback is accompanied by an abnor- diYculty. In depressed patients, by contrast, mal response to positive feedback. An analo- these latencies were not significantly diVerent gous method of analysis to that described in from overall response latencies. Whereas it is Elliott et al,1 and used here, to assess the eVect therefore clear that patients were not “giving of negative feedback would not be appropriate up” and making rapid guesses, they were not for analysing the eVects of positive feedback as taking longer over problems to improve their all the subjects studied were scoring more than performance. 50% correct. The abnormal response to negative feedback Whereas the present study has shown in depressed patients may thus reflect ineY- unequivocally an abnormal response to nega- ciency in using knowledge of results to monitor tive feedback relatively specific to depression, performance. Alternatively it may represent a the precise psychological mechanisms underly- motivational impairment such that patients are ing this eVect have not been clearly character- less motivated by potential reward than con- ised. Feedback potentially has both cognitive trols and so do not experience the same deter- and emotional sequelae, the second of which mination to improve performance in the light could either improve or disrupt performance. of negative feedback. It is also possible that Subjects can use negative feedback to monitor both these factors are involved in the abnor- and improve their performance but this cogni- mality; indeed, they may be interdependent.

tive component to their response is accompa- Using the cognitive information provided by http://jnnp.bmj.com/ nied by emotional reactions. This may take the negative feedback to monitor and improve per- form of an increased determination to succeed, formance may depend on subjects being moti- motivated by a desire to achieve positive vated to achieve positive rather than negative reinforcement or it may take the form of a more feedback. negative reaction such that subsequent per- There are thus two diVerent explanations of formance is disrupted. The overall response of the findings presented in this paper; an impair- subjects to negative feedback may depend on a ment of performance in response to negative

subtle interplay of these positive and negative feedback in depressed patients representing an on September 29, 2021 by guest. Protected copyright. factors. Previously, we suggested that the adverse emotive reaction, or an enhancement abnormal response in the depressed group may of performance in controls representing a pos- represent an “oversensitivity to negative feed- sible interaction of cognitive and emotional back”. The resultant disruptive eVect on eVects. Of course, these explanations are not subsequent cognitive performance has been necessarily mutually exclusive and it is entirely described as a “catastrophic response to plausible that both are contributing to the failure”. This account suggests that the nega- overall eVect. Further studies are necessary to tive emotive response to feedback on incorrect consider these hypotheses explicitly; however, trials impairs performance in depressed pa- what can be concluded unequivocally from the tients. present data is that the response to negative However, closer scrutiny of the data suggests feedback is abnormal in patients with depres- that an alternative explanation of the eVect may sion. also be viable. Thus when the conditional prob- The abnormal response to negative feedback ability of failure measure relative to the overall can be seen as an important link between mood probability of failure is considered, the evi- and performance in depression. Many previous dence suggests that it may be the control sub- attempts to link these dimensions of dysfunc- jects who show a greater response to negative tion have relied on the use of emotionally toned feedback. Specifically, their mean conditional information in neuropsychological tasks. The probability of failure (0.03, fig 1) is numerically eVect described here is a psychological eVect lower than their average probability of failure related to depressed mood which can influence (mean percentage correct 90%, fig 1, therefore the processing of neutral material. Suggesting a Negative feedback and depression 81 J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.63.1.74 on 1 July 1997. Downloaded from

critical role for reinforcement in the perform- However, no such deficit was seen in this high ance of depressed patients has much in functioning patient group. common with behavioural theories of depres- In neural terms, the possible existence of a sion. Lewinsohn et al32 33 proposed an account general factor influencing performance of cog- of depression based on reinforcement, stressing nitive tasks that engage probably very diVerent that depressed patients experience and expect neocortical networks may implicate neuro- low rates of positive reinforcement. The modulatory function of those ascending neuro- importance of response to feedback in depres- transmitter pathways innervating wide areas of sion can also be linked to cognitive accounts of the neocortex such as noradrenergic and sero- the disorder. An influential cognitive theory of tonergic systems, which are, of course, already depression has been developed by Beck34–36 and associated with depression.44 45 Another possi- 37–39 modified by Teasdale and colleagues. In ble neural substrate for the eVect is in the fron- brief, this theory argues that people vulnerable tal cortex. Damasio46 described in detail a to depression are characterised by enduring patient who had extensive damage to the dysfunctional schemata made up of negative prefrontal cortices, particularly the ventrome- attitudes and assumptions. A role for oversen- dial regions. This patient showed relatively sitivity to negative feedback is consistent with preserved neuropsychological function but was these cognitive accounts. Patients may be impaired at decision making, due to an inabil- distracted by internal ruminations or by exter- ity to attach emotional salience to situations. nal stimuli with a detrimental eVect on subse- This impairment has clear parallels with the quent performance. Their bias towards suggested abnormal emotional response to processing the negative feedback may detract feedback in depressed patients and an involve- from their processing of the subsequent ment of orbitofrontal pathology in the second problem. Alternatively, or additionally, the eVect could therefore be hypothesised. negative attitudes and assumptions of de- Regardless of what the neural substrate of pressed patients may compromise their ability the eVect might be, it is clear that response to or motivation to improve subsequent perform- negative feedback has a significant influence on ance in the light of negative feedback. neuropsychological performance. Obviously, Abnormal responses to negative feedback this e ect is likely to be most striking in tests are therefore consistent with cognitive and V behavioural accounts of depression and may where immediate and explicit feedback is given prove an important component underpinning after each problem. However, even in tests in both neuropsychological and clinical aspects of which this is not the case, subjects can usually the disorder. The comparisons with other form an impression of how well they are patient groups described here suggest that the performing. Whereas the eVect cannot be eVect may be relatively specific to depression explicitly measured in such tests, it may still be rather than being a general consequence of occurring. If oversensitivity to negative feed- raised failure rates or an artefact of the chosen back plays a part in the abnormality in method of statistical analysis. When groups of depressed patients, this would be further exac- non-psychiatric patients with a similar overall erbated if, as predicted by cognitive and behav- ioural accounts, they are more self critical than level of impairment were compared with the http://jnnp.bmj.com/ depressed patients, they did not show the same controls and are thus more likely to view their eVect of oversensitivity to negative feedback. performance as inadequate. The conditional probabilities of failing a prob- The finding that abnormal response to failure lem having just failed the preceding problem may be more closely related to whether did not diVer significantly from controls for the depressed patients are admitted to hospital patients with severe Parkinson’s disease or even than severity of depression as measured neurosurgical patients with frontal or temporal on clinical scales,1 provides evidence for the clinical validity of the measure which may also lobe lesions. Although these groups failed more on September 29, 2021 by guest. Protected copyright. problems overall than the controls, the same have implications for the behavioural manage- proportion of incorrect responses followed other ment and treatment of depression. The finding incorrect responses. This is an important result of study 2, that clinically recovered depressive in the light of the debate surrounding the patients still show an abnormal response to extent to which impairments associated with negative feedback is also consistent with the Parkinson’s disease can be attributed to findings of Vaughn and LeV47 that high levels of co-occurring depression.40–42 The present find- criticism increased relapse rate in depression. ings suggest that there are qualitative diVer- In conclusion, the abnormal response to ences in the performance of the two groups negative feedback described in this paper which means that several of the deficits found represents an advance in our understanding of in patients with Parkinson’s disease cannot be depression and potentially provides a crucial attributed to depression. The abnormal re- link between mood and performance. The sponse to negative feedback was also absent in finding is consistent with cognitive and behav- schizophrenic patients, suggesting that it is not ioural accounts of the disorder and may have a phenomenon which is common to psychiatric implications for treatment. Further studies are disorders but one which is relatively specific to needed to replicate this eVect independently, depression. Interestingly, one of our explana- perhaps in comparison with other forms of tions of the eVect as a monitoring deficit might aVective disorder—for example, generalised predict a similar impairment in patients with anxiety, obsessive-compulsive disorder, and schizophrenia, a disorder hypothesised to be bipolar depression—and characterise it more associated with monitoring impairments.43 fully in cognitive terms. 82 Elliott, Sahakian, Herrod, Robbins, Paykel J Neurol Neurosurg Psychiatry: first published as 10.1136/jnnp.63.1.74 on 1 July 1997. Downloaded from

This work was funded by a Programme Grant 22 Owen AM, Roberts AC, Hodges JR, Summers BA, Polkey to Drs TW Robbins, BJ Everitt, AC Roberts, and BJ Sahakian. CE, Robbins TW. Contrasting mechanisms of attentional RE was funded by a Wellcome Trust prize studentship. We set-shifting in patients with frontal lobe damage or Parkin- thank Drs Adrian Owen, Peter McKenna, and Paula McKay for son’s disease. Brain 1993;116:1159–75. their contribution to studies from which these data were 23 Elliott R. Neuropsychological deficits in a small group of schizo- 117182223 obtained. phrenic patients with preserved intellectual function. [PhD the- sis.] Cambridge; University of Cambridge. 1 Elliott R, Sahakian BJ, McKay AP, Herrod JJ, Robbins TW, 24 Cummings JL. Subcortical dementia: neuropsychology, Paykel ES. Neuropsychological impairments in unipolar neuropsychiatry, and pathophysiology. Br J Psychiatry depression: the role of perceived failure on subsequent per- 1986;149:682–97. formance. Psychol Med 1996;26:975–89. 25 Baxter LR, Phelps ME, Mazziotta J, Schwartz J, Gerner R, 2 Austin M-P, Ross M, Murray C,O’Carroll RE, Ebmeier KP, Selin C, Sumida R. Cerebral metabolic rates for glucose in Goodwin GM. Cognitive function in major depression. J mood disorders. Arch Gen Psychiatry 1985;42:41–7. AVect Disord 1992;25:21–30. 26 Drevets WC, Videen TO, Price JL, Preskorn SH, Car- 3 Beats BC, Sahakian BJ, Levy R. Cognitive performance in michael ST, Raichle ME. A functional anatomical study of tests sensitive to frontal lobe dysfunction in the elderly unipolar depression. J Neurosci 1992;12:3628–41. depressed. Psychol Med 1996;26:591–603. 27 Robbins TW. The case for frontostriatal dysfunction in 4 Brown RG, Scott LC, Bench CJ, Dolan RJ. Cognitive func- schizophrenia. Schiz Bull 1990;16:391–402. tion in depression: its relationship to the presence and 28 Crow TJ. Temporal lobe asymmetries as the key to the aeti- severity of intellectual decline. Psychol Med1994;24:829– ology of schizophrenia. Schizophr Bull 1990;16:433–43. 47. 29 Hamilton M. A rating scale for depression. J Neurol Neuro- 5 GolinkoV M, Sweeney JA. Cognitive impairment in depres- surg Psychiatry 1960;23:56–62. sion.JAVect Disord 1989; 17:105–12. 30 Montgomery SA, Asberg M. A new depression scale 6 Miller WR. Psychological deficit in depression. Psychol Bull designed to be sensitive to change. Br J Psychiatry 1975;82:238–60. 1979;134:382–9. 7 Weingartner H, Cohen RM, Martello JD, Gerdt C. 31 Nie NH, Hadlai Hull C, Jenkins JG, Jenkins JG, Steinbren- Cognitive processes in depression. Arch Gen Psychiatry ner K, Bent DH. Statistical package for the social sciences. 1981;38:42–7. New York: McGraw-Hill, 1970. 8 Schmand B, Kuipers T, Van der Gaag M, M, Bosveld J, 32 Lewinsohn PM, Youngren MA, Grosscup SJ. Reinforce- Bulthuis F, Jellema M. Cognitive disorders and negative ment and depression. In: RA Depue, ed. The psychobiology symptoms as correlates of motivational deficits in psychotic of depressive disorders. New York: Academic Press, 1979: patients. Psychol Med 1994;24:869–84. 291–316. 9 Hockey GRJ, Giallard A, Coles MGH. Energetics and human 33 Lewinsohn PM. A behavioural approach to depression. In: information processing. Dordrecht: NijhoV, 1986. 10 Richards PM, RuV RM. Motivational eVects on neuropsy- RJ Friedman, MM Katz, eds. The psychology of depression: chological functioning: comparison of depressed v non contemporary theory and research. New York: Winston and Wiley, 1974:157–185. depressed individuals. J Consult Clin Psychol 1989;57:396– 402. 34 Beck AT. Depression: clinical, experimental and theoretical 11 Hughes JR, Pleasants CN, Pickens RW. Measurement of aspects. New York: Harper and Row, 1967 reinforcement in depression: a pilot study. J Behav Ther Exp 35 Beck AT. Cognitive therapy and the emotional disorders.New Psychiatry 1985;16:231–6. York: International Universities Press, 1976. 12 Wener AE, Rehm L. Depressive aVect: a test of behavioral 36 Beck AT. Cognitive therapy of depression. New York: J Wiley, hypotheses.J Abnorm Psychol 1975;84:221–7. 1979. 13 Owen AM, Downes JD, Sahakian BJ, Polkey CE, Robbins 37 Teasdale JD. Cognitive vulnerability to persistent depres- TW. Planning and spatial following fron- sion. Cognition and Emotion 1988;2:247–74. tal lobe lesions in man. Neuropsychologia 1990;28:1021–34. 38 Teasdale JD, Barnard PJ. AVect, cognition and change. Hove: 14 Robbins TW, James M, Lange KW, Owen AM, Quinn NP, Lawrence Erlbaum Associates, 1993. Marsden CD. Cognitive performance in multiple system 39 Teadale JD, Dent J. Cognitive vulnerability to depression: an atrophy. Brain 1992;115:271–91. investigation of two hypotheses. Br J Clin Psychol 1987;26: 15 Sahakian BJ, Morris RG, Evenden JL, Heald A, Levy R, 113–26. Philpot M, Robbins TW. A comparative study of visuospa- 40 Rogers D, Lees AJ, Smith E, Trimble M, Stern GM. Brady- tial memory and learning in Alzheimer-type dementia and phrenia in Parkinson’s disease and psychomotor retarda- Parkinson’s disease. Brain 1988;111:695–718. tion in depressive illness. An experimental study. Brain 16 Robbins TW, James M, Owen AM, Sahakian BJ, McInnes 1987;110:761–76. L, Rabbitt P. Cambridge neuropsychological battery 41 Starkstein SE, Prezios TJ, Berthier ML, et al. Depression (CANTAB): a factor analytic study of a large sample of and cognitive impairment in Parkinson’s disease. Brain normal elderly volunteers. Dementia 1994;5:266–81. 1989;112:1141–53. 17 Owen AM, Sahakian BJ, Hodges JR, Summers BA, Polkey 42 Taylor AE, Saint-Cyr JA, Lang AE, Parkinson’s disease CE, Robbins TW. Dopamine-dependent frontostriatal et al. and depression: a critical evaluation. Brain 1986;109:279– planning deficits in early Parkinson’s disease. Neuropsychol-

92. http://jnnp.bmj.com/ ogy 1995;9:126–40. 18 Owen AM, Sahakian BJ, Semple J, Polkey CE, Robbins TW. 43 Frith CD. The cognitive neuropsychology of schizophrenia. Visuo-spatial short term recognition memory and learning Hove: Lawrence Erlbaum Associates, 1992. after temporal lobe excisions, frontal lobe excisions or 44 Zubenko GS, Moossy J, Koppu M. Neurochemical amygdalo-hippocampectomy in man. Neuropsychologia correlates of major depression in primary dementia. Arch 1995;33:1–24. Neurol 1990;47:209–14. 19 Baker SC, Rogers RD, Owen AM, Frith CD, Dolan RJ, 45 Deakin JFW. Amine mechanisms in anxiety and depression. Frackowiak RSJ, Robbins TW. Neural systems engaged by In: JM Elliott, DJ Heal, CA Marsden, eds. Experimental planning: a PET study of the Tower of London task. Neu- approaches to anxiety and depression. New York: J Wiley, ropsychologia 1996;6:515–26. 1992:233–43. 20 American Psychiatric Association. Diagnostic and statistical 46 Damasio AR. Descartes’ error. New York: G Putnam, 1994. manual of mental disorders, DSM III-R. APA, 1987. 47 Vaughn C, LeV J. The influence of family life and social fac- 21 Nelson HE. National adult reading test (NART): test manual. tors on the course of psychiatric illness. Br J Psychiatry on September 29, 2021 by guest. Protected copyright. Windsor: NFER-Nelson, 1982. 1976;129:125–37.