Home

German Journal of Psychiatry

ISSN 1433-1055

Emotional priming in depressed patients


Matthias Weisbrod1, Jan Trage1, Holger Hill1, Herm-Dieter Sattler1, Sabine Maier1, Markus Kiefer2, Joachim Grothe2, Manfred Spitzer2


 1Department of Psychiatry, University of Heidelberg

2Department of Psychiatry, University of Ulm

 

Corresponding author: Matthias Weisbrod, Psychiatrische Universitätsklinik, Voss-Str. 4, 69115 Heidelberg, Germany, Phone: 06221-562745, Fax: 06221-565477, E-mail: Matthias_Weisbrod@ukl.uni-heidelberg.de

Abstract

Background: Negative automatic thoughts are a common symptom in depression. The most influential model of the intertwining of affective and cognitive processes proposes a specific node for each distinct emotion in an associative network. Activation of such a node gives rise to the experience of the emotion. This assumption implies that in depressed patients negative concepts are overly activated. Therefore, processing of negative outcomes of sentence endings should be facilitated whereas positive outcomes should be inhibited. Methods: To test this model, sentences with negative and positive outcomes were presented to depressed and healthy subjects in the form of a lexical decision task. Neutral sentences were used as primes and emotionally salient words or respectively incongruent words as targets. Results: Analyses of reaction times revealed differences between depressed patients and healthy subjects depending on the emotional salience of the target word. Priming effects were correlated to clinical features in depressed patients. In addition depressed patients had prolonged reaction times for the incongruent condition. Conclusions: When compared to normal control subjects, depressed patients display a bias towards negative associations. This pattern is possibly relevant for therapeutic interventions. Moreover, depressed patients have specific problems in dealing with unusual associations (German J Psychiatry 1999;2:19-47)

 

Key words: depression, lexical decision, semantic priming

Received: Nov 12, 1999

Published: Dec 8, 1999

 

Introduction

Negative automatic thoughts are a common symptom in depression. The thoughts of depressed patients are directed towards negative contents and are difficult to suppress. Affective and cognitive processes interact reciprocally (Haaga et al 1991) and lead to a negative cognitive bias in depressed patients. Within this framework, the effects of cognitive behavioural therapy targeting negative automatic thoughts and feelings in patients with chronic depressive symptoms can be explained (Thase et al 1991). The prophylactic effects of this therapy persist beyond treatment discontinuation (Frank et al 1990). Ellis (Ellis 1962) and Beck (Beck 1967) suggested that cognitive manifestations of depression result from maladaptive cognitive structures that are linked to the disorder. In their view, constantly present and enduring negative assumptions and attitudes are the primary vulnerability factors for depression. Such a negative cognitive bias is thought to exert a significant influence on information processing by selectively screening what information is extracted from both internal and external sources, and by affecting both the encoding and retrieval of information (Alba and Hasher 1983; Kihlstrom and Cantor 1984).

More recent conceptual approaches view cognition and affect as intricately embedded within interactive associative networks (Ingram 1984; Izard 1993). One of the most influential models to explain the interaction of cognition and mood state was proposed by Bower (Bower 1981). He extended the model of spreading activation within associative networks to the domain of emotion. According to network models of semantic memory, semantic (and possibly other) features of words are represented as “nodes” in neuronal networks. In speech production and reception, these units become activated for a short period of time. This model further asserts that activation spreads to related nodes and lowers the thresholds of related concepts to become activated, thereby influencing subsequent thought processes. Bower (Bower 1981) proposed that each distinct emotion has a specific node in an associative network. Activation of such a node gives rise to the experience of the respective emotion and spreads to related nodes in the network. This renders them more likely to become activated themselves. In short, the accessibility of these activated concepts is enhanced. According to this model, in a person who is sad, events and concepts associated with sadness in his or her life are more accessible, i.e. these thoughts will come to mind far more readily than thoughts associated with other emotions.

A considerable body of literature on so-called state-dependent learning and recall supports this model. In depressed mood, healthy individuals tend to remember more negative personal memories, and selectively recall material learned in a depressed mood (Blaney 1986; Clark and Teasdale 1982; Matt et al 1992; Singer and Salovey 1988; Teasdale 1983). In induced mild depressed mood, recovered depressed individuals recall significantly more negative adjectives that had been endorsed as self-descriptive than do individuals who have never been depressed (Teasdale and Dent 1987).

To study the spreading of activation in semantic networks, we measured semantic priming in a lexical decision task, in which words have to be distinguished from non–words. If the word upon which the decision is to be made (target-word) is related to a word (prime-word) which is presented immediately before the target-word, the decision is carried out faster compared to the condition in which the target-word and the prime word are non-related.

In Bower´s model, nodes which are connected to negative emotions easily become activated in the associative networks of depressed patients and thereby facilitate the recall of emotionally negative associations. Therefore, depressed patients who are presented with sentences with emotionally distinct outcomes should differ from healthy controls by recognizing the negative outcome more readily. In the present study we compared healthy controls and depressed patients with respect to the priming effects for the final word of sentences with emotionally distinct (but equally congruent) endings. In the positive condition, the final word closed the sentence in a positive way (i.e. to fly is fun), whereas in the negative condition the sentence was turned negative by the final word (i.e. to fly is dangerous). To contrast the specific emotional manipulation with a merely semantic manipulation, we further included an incongruent condition, in which the final word did not complete the sentence in a meaningful way (i.e. to fly is green). We expected depressed patients to show more priming in the negative condition and less priming in the positive condition compared to healthy controls.

 

Method

Subjects

 Twenty-three healthy controls and 23 depressed patients, all native German speakers and right-handed, were included in the study. Handedness was confirmed using a modified version of the Edinburgh Handedness Inventory (Oldfield 1971). Exclusion criteria for the control group were a personal and family history of psychiatric illness. Exclusion criteria for all subjects were a previous manic episode, alcohol or drug abuse, any neurological diseases, and impaired vision for all subjects. Controls and patients were not different in sex, age, verbal intelligence, or education (see Table 1). Diagnoses were made by the physician on the wards and the chief of services according to ICD 10 (World Health Organization 1991), based upon clinical symptoms and patient history. All patients suffered from unipolar major depression without delusional symptoms (ICD 10: F32.1, F32.2, F33.1, F33.2). At the time of testing, their initial symptomatology had partly remitted. Psychopathology was rated using the Hamilton Rating Scale (Hamilton 1960) and the Beck Depression Inventory (BDI; (Beck 1978). Ratings were performed after testing to avoid mood induction due to the rating procedure (Bargh 1992). Verbal intelligence was measured by the „Mehrfachwahl-Wortschatz-Test“ (Lehrl 1977). All patients were on antidepressant medication during the time of data collection.

 

Table 1: Demographic and psychopathological variables. There are no significant differences between the depressed and the healthy group

 

 

Patients

Controls

t, p

N

23

23

---

Sex (female/male)

16/7

16/7

---

Age (SD)

48.4 (13.3)

42.7 (18.0)

t (44) = 1.23; (n.s.)

MWT-B score (SD)

30.0 (3.4)

31.3 (3.4)

t (44) = -1.40; (n.s.)

Years of Schooling (SD)

10.2 (1.6)

11.0 (1.6)

t (44) = -1.58; (n.s.)

Hamilton score (SD)

15.3 (7.3)

3.4 (3.4)

t (44)= 7.1; p<.001

BDI score

13.3 (6.9)

6.2 (4.4)

t (44)= 4.2; p<.001

 

Stimulus material

240 sentences consisting of 4 to 8 words were carefully constructed such that they ended with one of two semantically congruent final words with distinct emotional valence. In a first step we constructed sentence stems which were completed by final words, either positively (i.e. to fly is fun) or negatively (i.e. to fly is dangerous). To verify expectancy of the target words as final words, the sentences were presented to 54 healthy subjects without the final words. They were instructed to write down as fast as possible the first three words which came into their mind to complete the sentences in a meaningful way. We found 182 sentences which fulfilled the following criteria: 1) At least two final words were produced by most of the subjects in agreement, 2) one of these associated words closed the sentence in a positive way the other in a negative way, 3) neither the positive nor the negative final word was named by most of the subjects as the first association. In a second step, 20 healthy subjects were presented with these 182 sentences, together with both the positive and the negative final word, and were instructed to rate the emotional valence of each of the complete sentences. Only sentences upon which at least 18 subjects (90%) agreed in the rating for both the positive and the negative completion were accepted. Positive and negative words were then checked for length and frequency (according to Ruoff 1990) which left us with 120 sentences. In addition to the positive and the negative condition, an incongruent condition was constructed for each sentence stem. For each of the sentence stems, a word was selected from the negative and positive completion words of the other sentences (i.e., to fly is green). These words closed the sentence in a grammatically correct way, and were not semantically related to any of the words of the sentence.

With this material three different versions of the test were created each consisting of 240 sentences. In each test 120 sentences ended with a non-word, and 120 sentences were completed by a real German word. Of these 120 real word sentences, 40 had an incongruent termination; 40 were closed by a congruent word which turned the content into a negative meaning, and 40 ended positively. Sentences were identical throughout the three versions, except for the final word. Sentences and each target words were presented only once in each version. The 120 sentences with a meaningless string of letters as the final word were identical throughout all versions. The order of sentences within the stimulus list was initially randomised but then remained constant for all presentations.

Design and experimental procedures

All subjects were tested individually using one of the three versions. The experiment was performed in a dimly lit room. Subjects were sitting in an upright position in front of the computer screen on which the stimuli were presented. Stimuli were displayed on an Apple Macintosh computer (Apple, Cupertino, Calif.). Prime and target words were written in lower-case, 18 point Geneva font. Given a viewing distance of 50 cm, each word subtended a visual angle of about 2° to 4° in width and 0.7° in height. Reaction times (RTs) and error rates were automatically recorded by the computer program (Mac Lab 2.0.0.b.2; (Costin 1988). The sequence of one trial is displayed in Figure 1. The trial started with a blank computer screen. As soon as the subjects initiated a trial by pressing the mouse key with their left index finger, a fixation mark was displayed on the screen for 700 ms. Then the screen immediately displayed each word of the sentence, one at a time for 300 ms. A black frame was used to mark the last word of the sentence as the target word. Subjects were instructed to read silently and to decide as fast and as accurately as possible whether the final word in each of the sentences was a real German word (50%) or a meaningless string of characters (50%). Subjects´ responses consisted of pressing one of two keys on the computer´s keyboard with the index (word) or the middle finger (non-word) of their right hand. The target word disappeared as soon as the decision was made. The fixation point and all words were presented at the centre of the screen. A practice experiment, consisting of 24 sentences similar to those of the actual experiment was set up with no data sampled. The 240 sentences were presented in blocks of 60 sentences with a 2-minute break between blocks. Each block consisted of 30 sentences of the non-word condition and 10 sentences of the negative, positive, and incongruent conditions, respectively. The entire session, consisting of the practice experiment, the full length experiment, the handedness questionnaire, and the pychopathological rating required about one hour and 15 minutes to be completed. All subjects were tested between 9 am and 11 am.

 

 

Figure 1: Temporal presentation of the stimuli. As soon as the blank screen appears subjects can initiate the presentation of the stimuli (self paced). Subjects are instructed to decide if the framed word is a real word or a meaningless string of letters as soon as the framed word is presented.

 

 

Results

Error rates and reaction times were calculated for each individual subject. Incorrect responses were excluded from the calculation of RTs. The differential influence of the final words on errors and RTs was evaluated in separate ANOVAs. Post hoc comparisons were conducted using the Newman-Keuls-test.

Error rates were calculated as the percentage of false responses in relation to the correct responses ([number of wrong responses/number of right responses] x 100). A two way ANOVA performed with group (controls and depressed patients) and condition (incongruent, negative, positive, and non-word; repeated measures factor) as independent factors and with error rates as dependent factor revealed a significant main effect for condition (F(3/132) = 21.85 ; p < .001), i.e. the incongruent condition led both groups to make significantly more errors than the other conditions. Whereas depressed patients generally did not make more errors than controls (no significant main effect for group) they performed worse than controls in the incongruent condition (significant interaction between group and condition (F(3/132) = 3.02; p = .03; p < .001 in the post hoc analysis).

 

 

Figure 2: Difference between the RTs for the presentation of the non-word and the positive or the negative word (ms, SE-bars), respectively, for both depressed patients and healthy controls.

Reaction times were shortest in the positive condition, followed by the negative condition, and the non-word condition. The longest RTs were elicited by the incongruent condition. This pattern was present in both groups and was reflected in a significant main effect for condition (F(3/132) = 27.30; p < .001) in the two way ANOVA with RTs as dependent variable and group and condition (repeated measures factor) as independent factors. This ANOVA furthermore revealed a significant main effect for group (F(1/44) = 7.31; p = .001), demonstrating the generally longer RTs in depressed patients. The significant interaction between group and condition (F(3/132) = 2.8; p = .043) was explored by post hoc comparisons. Patients showed no significant RT difference between the positive and the negative condition and presented shorter RTs in both of these conditions compared to the incongruent and the non-word condition (see Fig. 2). In the incongruent condition the RTs were longer compared to the non-word condition (see Fig. 3). Controls´ RTs in the positive and the negative conditions were not different either. In contrast to the patients, controls showed only in the positive, but not in the negative, condition significant RT-reduction compared with the non-word condition (see Fig. 2). Contrary to the depressed group, controls exhibited no significant difference between the incongruent condition and the non-word condition (see Fig. 3). In comparison to controls, patients showed significantly longer RTs in all conditions (p < .05) except the negative condition.

 

 

Figure 3: Reaction times for the presentation of the non-word and the incongruent word (ms, SE-bars) for both depressed patients and healthy controls

 

 

Correlations: In a correlation analysis, we found significant negative correlations between the number of relapses since first depressive episode and priming effects in the positive (r = -.44; p = .028) as well as in the negative (r = -.47, p = .024) condition, i.e. the more relapses depressed patients experienced in the past the smaller were priming effects irrespective of the emotional valence. This result was even more interesting since neither in controls nor in depressed patients were significant correlations between age, verbal intelligence (MWT score), psychopathology (Hamilton and the BDI score) and any priming effects found. There were also no significant correlations between RT data and number of relapses.

In general, healthy subjects responded faster in the positive condition than in the negative condition although this difference was not significant. To evaluate the significance of this "healthy pattern" we calculated the difference between RT in the negative condition and RT in the positive condition (RT-ne minus RT-po) and correlated this difference with clinical variables. In depressed patients this difference was significantly correlated with the time of hospitalisation (r = -.45; p = .034), whereas no such correlations were found with various measures of psychopathological state (see Fig. 4).

 

Figure 4: Correlation between the difference between RT for negative and positive word (labelled emotional priming) in ms and the duration of hospitalisation (days) during the actual depressive episode

Discussion

In this study we set out to measure the effects of emotional valence on semantic information processing in depressed patients and normal controls. In a lexical decision task with neutral sentences as prime and emotionally salient words as targets, depressed patients performed differently from normal controls on final words with negative emotional valence. First, in contrast to controls, depressed patients responded significantly faster in the negative condition than in the non-word condition. Secondly, depressed patients showed longer RTs than the controls in all conditions except the negative valence condition. In sum, our results are indicative of subtle differences in the processing of emotionally salient material between controls and depressed patients, suggesting relatively higher pre-activation of negative concepts in the semantic networks of depressed patients compared to healthy controls. In the following discussion we first want to address possible reasons for the surprisingly small effect size, and secondly, describe and discuss the depressed patients´ relative slowness in responses towards incongruent words.

The general pattern of our results, i.e. the small effect of emotional valence on RTs, suggests that the effect of emotional priming was masked by comparatively larger semantic priming effects. In both groups, semantically congruent sentence endings (regardless of emotional valence) produced markedly faster responses than semantically incongruent words. Possibly, semantic priming produced a floor effect, to which emotional priming could add only little.

Moreover, the positive and negative phrase endings chosen as targets bore an uncertain relation to subjects´ own connotation of these concepts and may have been located more distally in semantic space from subjects´ idiosyncratic depressive content. The design of future studies should aim at an enhancement of emotional priming processes. For example, future designs might incorporate biographically relevant material tailored to the single subject and thereby increase the emotional load of the stimulus material.

Teasdale et al. have shown that, in some contexts, depressed patients show more positive completion than controls when such positive completions are consistent with an underlying depressiogenic model of the self (Teasdale et al 1995). As depressives recover, the frequency of negative completions for this type of material rises. Teasdale and colleagues used a sentence completion task in which sentences operated in opposite directions on the level of schematic models and on the level of specific constructs. This design favoured controlled processes operating at a postlexical access stage. In contrast, our study focused on supposedly automatic cognitive processes since we used a lexical decision task with a comparatively short SOA (stimulus onset asynchrony) of 300 ms. These fundamental differences on task design explain the different findings. Whereas our results suggest that automatic spreading of activation along a semantic network structure is involved in negative thinking of depressed patients, the Teasdale et al. findings show that these basic associations can be overwritten by higher cognitive processes.

 

Furthermore, it is noteworthy that our patients were tested during recovery from the depressed episode. Since studies on emotional memory in recovered depressed patients did not yield significant differences from normal controls (Teasdale and Dent 1987), the subjects in our patient group may already have been returned to normal with respect to their emotionally driven semantic processing. Therefore, further studies should examine depressed patients at least twice, before and after remission of acute depressive symptoms.

Finally, studies on remitted depressed patients have demonstrated that effects of emotional salience on information processing were dependent upon the current mood state of the patient (Williams 1988). The Differential Activation Hypothesis (Teasdale 1988) suggests that the severity and maintenance of depressed states may be determined primarily by differences in the patterns of information processing that become accessible once a person has initially become depressed. Therefore, the thinking of remitted individuals may be quite normal when they are not in a depressive mood state. Mood induction procedures which were not used in the present study should increase emotionally driven cognitive processes and therefore demask dysfunctional thinking in subjects prone to depression (Teasdale and Dent 1987).

Interestingly, the difference between RTs for negative and for positive endwords was negatively correlated to the duration of the present hospitalisation (cf Fig. 4), while it did not correlate to psychopathology as measured by the Hamilton Score and the BDI. In this respect, our analyses suggest that the maintenance of this "healthy pattern" of emotional priming could be predictive for the duration of the depressed episode independently of the actual psychopathology. This finding corresponds to the results from Teasdale et al. who found that differences in the endorsement of negative adjectives during a depressive episode predicted who would remain depressed and who would recover (Teasdale 1988). It is also in line with the suggestions of several researchers that such cognitive structures place people at risk of increased persistence of their disorder (Kuiper et al 1988). In respect to our study, patients with a more "healthy pattern" of emotional priming may show a generally faster remission of depressive episodes. Correspondingly, patients who showed comparatively less semantic priming effects had experienced more depressive episodes in the past. This finding may reflect long lasting effects of depressive episodes on the processing of emotionally salient material as proposed by Lewinsohn et al. (Lewinsohn et al 1981). Differential emotional priming effects may also be explored for use in the definition of subgroups of depressed patients with different prognoses.

The most striking result of this study is the marked RT slowing evoked by the semantically incongruent word. When the incongruent condition was compared to the non-word condition, normal subjects showed no significant difference in response time, whereas depressed patients took markedly longer to identify the incongruent word. This observation translates into a highly significant difference between the groups regarding the RT slowing present in the incongruent condition (see Fig. 3). Similar effects of incongruent stimulus material in depressed patients have been described by Thomas et al. (Thomas et al 1997) and Benoit (Benoit et al 1992), using the Stroop Color-Word Test (SCWT). Trichard et al. (Trichard et al 1995) investigated depressed patients in a longitudinal study with the SCWT and the Verbal Fluency Test (VFT). On admission to the hospital, depressed patients showed impaired performance in both tests, but VFT normalised with successful treatment of depression. These findings suggest that prefrontal areas which were shown to become activated by the VFT (Frith et al 1991) are already functional at discharge, whereas the function of the anterior cingulate which was shown to be involved in the SCWT (Pardo et al 1990) was still deficient. Beyond these findings, little is known about the underlying cognitive mechanisms, although clinically, depression is characterised by a slowing of cognitive as well as psychomotor functions. Further experimental investigation of these mechanisms in depressed patients is clearly warranted.

 

References

Alba JW, Hasher I (1983) Is memory schematic? Physiol Rev 93:207-231.

Bargh JA (1992) The ecology of automaticity: Toward establishing the condition needed to produce automatic processing effects. Am J Psychol 105:181-199.

Beck AT (1967) Depression: Clinical, Experimental, and Theoretical Aspects. Harper & Row, New York

Beck AT (1978) The depression inventory. Center for Cognitive Therapy, Philadelphia

Benoit G, Fortin L, Lemelin S, Laplante L, Thomas J, Everett J (1992) Selective attention in major depression: clinical retardation and cognitive inhibition. Can J Psychol 46:41-52.

Blaney PH (1986) Affect and memory: A review. Psychol Bull 99:229-246.

Bower GH (1981) Mood and memory. American Psychologist 36:129-148.

Clark DM, Teasdale JD (1982) Diurnal variations in clinical depression and accessibility of memories of positive and negative experiences. J Abnorm Psychol 91:87-95.

Costin D (1988) MacLab: A Macintosh system for psychology labs. Behavior Research, Methods, Instruments & Computers 20:197-200.

Ellis A (1962) Reason and Emotion in Psychotherapy. Stuart, New York

Frank E, Kupfer DJ, Perel JM, et al (1990) Three-year outcome for maintenance therapies in recurrent depression. Arch Gen Psychiatry 47:1093-1099.

Frith CD, Friston KJ, Liddle PF, Frackowiak RSJ (1991) A PET study of word finding. Neuropsychologia 29:1137-1148.

Haaga DAF, Dyck MJ, Ernst D (1991) Empirical status of cognitive theory of depression. Psychol Bull 110:215-278.

Hamilton M (1960) A rating scale for depression. J Neurol Neurosurg Psychiatry 23:56-62.

Ingram RE (1984) Toward a information-processing analysis of depression. Cognitive Therapy and Research 8:443-478.

Izard CF (1993) Four systems for emotion activation: Cognitive and noncognitive processes. Psychol Rev 100:68-90.

Kihlstrom JF, Cantor N (1984) Mental representations of the self. In: Berkowitz (ed) Advances in experimental social psychology, Vol 15. Academic Press, New York, pp 1-47.

Kuiper NA, Olinger IJ, MacDonald M (1988) Vulnerability and episodic cognitions in a self worth contingency model of depression. In: Alloy IB (ed) Cognitive processes in depression. Guilford Press, New York, pp 289-309.

Lehrl S (1977) Mehrfach-Wortschatz-Intelligenztest MWT-B. Straube, Erlangen

Lewinsohn PM, Steinmetz JL, Larson DW, Franklin J (1981) Depression-related cognitions: Antecedent or consequence? J Abnorm Psychol 90:213-219.

Matt GE, Vazquez C, Campbell WK (1992) Mood-congruent recall of affectively toned stimuli: A mega-analytic review. Clinical Psychology Review 12:227-255.

Oldfield R (1971) The assessment and analysis of handedness: The Edingburgh Inventory. Neuropsychologia 9:97-113.

Pardo JV, Pardo PJ, Janer KW, Raichle ME (1990) The anterior cingulate cortex mediates processing selection in the Stroop attentional conflict paradigm. Proceedings of the National Academy of Sciences, USA, Neurobiology 87:256-259.

Ruoff A (1990) Häufigkeitswörterbuch gesprochener Sprache. Niemeyer, Tübingen

Singer JA, Salovey P (1988) Mood and memory: evaluating the network theory of affect. Clinical Psychology Review 8:211-251.

Teasdale JD (1983) Negative thinking in depression: cause, effect, or reciprocal relationship? Advances in Behaviour Research and Therapy 5:3-25.

Teasdale JD (1988) Cognitive vulnerability to persistent depression. Cognition and Emotion 2:247-274.

Teasdale JD, Cooper Z, Hayhurst H, Paykel S (1995) Depressive thinking: shifts in construct accessibility or in schematic mental models. J Abnorm Psychol 104:500-507.

Teasdale JD, Dent J (1987) Cognitive vulnerability to depression: An investigation of two hypothesis. Br J Clin Psychol 26:113-126.

Thase ME, Simons AD, Cathalane JF, McGeary J (1991) Cognitive behaviour therapy of endogenous depression: Part 1: An outpatient clinical replication series. Behaviour Therapy 22:457-467.

Thomas J, Raoux N, Everett J, Dantchev N, Widlocher D (1997) Deficit in selective attention and its evolution in depression. Encephale 23:108-112.

Trichard C, Martinot JL, Alagille M, et al (1995) Time course of prefrontal lobe dysfunction in severely depressed in-patients: A longitudinal neuropsychological study. Psychol Med 25:79-85.

Williams RM (1988): Individual differences in the effects of mood on cognition: University of Oxford.

World Health Organization (1991) Tenth Revision of the International Classification of diseases. Clinical Descriptions and Diagnostic Guidelines. World Health Organization,