Deficits in Domains of Social Cognition in Schizophrenia: A Meta-Analysis of the Empirical Evidence

Abstract

Objective: Social cognition is strongly associated with functional outcome in schizophrenia, making it an important target for treatment. Our goal was to examine the average magnitude of differences between schizophrenia patients (SCs) and normal comparison (NCs) patients across multiple domains of social cognition recognized by the recent NIMH consensus statement: theory of mind (ToM), social perception, social knowledge, attributional bias, emotion perception, and emotion processing. Method: We conducted a meta-analysis of peer-reviewed studies of social cognition in schizophrenia, published between 1980 and November, 2011. Results: 112 studies reporting results from 3908 SCs and 3570 NCs met our inclusion criteria. SCs performed worse than NCs across all domains, with large effects for social perception (g = 1.04), ToM (g = 0.96), emotion perception (g = 0.89), and emotion processing (g = 0.88). Regression analyses showed that statistically significant heterogeneity in effects within domains was not explained by age, education, or gender. Greater deficits in social and emotion perception were associated with inpatient status, and greater deficits in emotion processing were associated with longer illness duration. Conclusions: Despite the limitations of existing studies, including lack of standardization or psychometric validation of measures, the evidence for deficits across multiple social cognitive domains in schizophrenia is clear. Future research should examine the role of neurobiological and psychosocial factors in models linking various aspects of deficit in schizophrenia, including social cognition, in order to identify targets for intervention.

Introduction

Schizophrenia is associated with markedly impoverished social dexterity and functioning.¹ The ability to navigate social cues and behaviors is inherently dependent on a knowledge base and set of skills, commonly known as “social cognition,”² defined by schizophrenia researchers as the “(the processes by which) we draw inferences about other people’s beliefs and intentions and how we weigh social situational factors in making these inferences.”³ Allusions to deficits in aspects of social cognition and functioning among people with schizophrenia can be found as early as the writings of Kraepelin (eg, “Loss of sympathy is shown in indifference and want of understanding for the misfortunes of others …” [p. 33]).⁴ Efforts to identify characteristics of schizophrenia that may explain poor functional outcome have been ongoing since the earliest conceptualizations of the condition. The focus on Schneiderian first-rank symptoms as potential predictors of functioning shifted to neurocognitive deficits about three decades ago, with hundreds of studies demonstrating that cognitive functioning is more relevant to real-world functioning than are positive symptoms of schizophrenia.⁵

Social cognition research has been part of the broader research in general social psychology for decades, and applications to schizophrenia can be found in published studies through the 1980s and 1990s. Penn and colleagues,⁶ in 1 of the first reviews of social cognition in schizophrenia, emphasized the importance of studying cognitive processes underlying how people with schizophrenia think about themselves, others, social situations, and social interactions in further understanding the etiology of the disorder. Pivotal publications by Green and colleagues^7,8 further spurred this area of research.

Indeed, the critical role of social cognition in functional disability has now been well established in the current literature.⁹ In a recent meta-analysis, Fett and colleagues¹⁰ investigated the associations between neurocognition, social cognition, and domains of functional outcome in schizophrenia, concluding that social cognition was most strongly related to functioning. Social cognition appears to be moderately related to domains of neurocognition, negative symptoms, and disorganization, as demonstrated in another recent meta-analysis by Ventura and colleagues.¹¹ The study of social cognition in schizophrenia not only has clinical/functional implications but also potentially significant research implications. Socioemotional and higher neurocognitive processes, such as abstract thinking, working memory, and online monitoring, appear to mature late in ontogeny,¹² coinciding with the period of life associated with typical onset of schizophrenia. Therefore, examining these processes in conjunction with each other can further our understanding of the disrupted neurobiological systems in schizophrenia.

Social cognition, like other aspects of cognition, is a multifaceted concept, comprising several sub-domains and processes. The NIMH consensus statement on social cognition in schizophrenia identified five relevant domains: ToM, social perception, social knowledge, attributional bias, and emotion processing.³ Although the consensus statement included emotion perception within the domain of emotion processing, we chose to examine it separately, to investigate whether simply identifying and labeling emotions was less impaired in schizophrenia than understanding one’s own emotions and facilitating or managing them. Table 1 gives a brief description of each of these constructs and their prototypical measures.

Table 1.

Social Cognition Domains and their Descriptions

Domain		Description		Example of a Prototypical Task
Theory of mind (ToM)		Definition: The ability to interpret an individual’s speech and actions in terms of his or her intentions, knowledge, and beliefsTasks involve inferring mental states from facial expressions or perspective-taking“First-order” ToM: the ability infer what another person is thinking“Second-order” ToM: the ability to infer what one person believes another person is thinking (ie, a “belief about a belief”)		Reading the Mind in the Eyes task 32: requires that one infer the mental state of a person only by looking at a photograph of the person’s eyes
Social perception		Definition: The ability to understand and appraise social roles, rules, and contextInvolves using verbal and nonverbal cues in order to make inferences about a social situationMay be central to functioning in a social context, ie, facilitating interactions with people in social settings or establishing relationshipsCan involve making critical appraisals, such as judgments of trustworthiness in other people		Profile of Nonverbal Sensitivity (PONS) 33: video-taped scenes containing facial expressions, voice intonations, and bodily gestures. After watching each scene, participants were asked to select from two situations (eg, saying a prayer or talking to a lost child) that would prompt the social cues observed
Social knowledge		Definition: Refers to representational templates of social situations or awareness of the roles, rules, expectations, and goals that govern social situations34Can be declarative, comprising facts and abstract concepts (eg, social scripts) or procedural (eg, rules, skills, and strategies) processes35		Situational Feature Recognition Test (SFRT) 36: requires subjects to select appropriate actions from a list of actions associated with a particular social situation (eg, going to a movie), followed by a list of goals
Attributional bias		Definition: Attributional bias/style reflects whether one typically makes inferences about the causes of positive and negative events to internal (personal), external (other person), or situational factorsReferred to in terms of externalizing bias (EB) vs personalizing bias (PB)EB is the tendency to overattribute positive rather than negative events to oneself, and PB is the tendency to attribute negative events to others rather than to situational factors		The Internal, Personal, and Situational Attributions Questionnaire (IPSAQ) 18: consists of statements describing an incident; the subject is asked to select one of three causes of the incident. Items reflect internal, external, and situational causes
Emotion perception		Definition: Refers to the ability to accurately identify and name emotions of others, primarily by means of facial expressionsEmotions may also be perceived through vocal prosody		The Face Emotion Identification Test (FEIT) 37: uses the Izard/Ekman emotion photographs and asks subjects to circle one of six “basic” emotions displayed (happiness, sadness, anger, fear, surprise, and shame)
Emotion processing		Definition: Refers to the ability to understand emotions, discriminate between different emotions, and manage emotions and emotional reactions		Mayer-Salovey-Caruso Emotional Intelligence Test (MSCEIT) 38 is part of the MATRICS battery39: is self-administered, performance-based measure, comprising eight tasks. In the widely used Managing Emotions subscale (designed to measure the cognition of emotion regulation) participants must judge actions that are most effective in obtaining a specified emotional outcome for an individual in a story (eg, what a character may do to reduce his or her anger)

There are 2 meta-analyses of ToM in schizophrenia^13,14 and 1 of facial emotion perception,¹⁵ each demonstrating robust deficits in those respective domains. These, at least in part, reflect the disproportionately large numbers of studies of these domains compared with the other four domains, ie, social perception, social knowledge, attributional bias, and emotion processing. Fett and colleagues¹⁰ reported that ToM had the strongest relationship with community functioning (combined correlation = 0.48), followed by social perception and social knowledge (combined) was a close second (combined correlation = 0.41); indeed, deficits in social perception (the ability to understand social roles, rules, and context) and social knowledge (the representational templates of social situations, or awareness of the roles, rules, expectations, and goals that govern social situations) may be critical to functional outcomes among people with the illness, yet have been little studied. Furthermore, finer distinctions between the various community functioning outcomes, such as social and relational functioning and their relationships with the various social cognitive domains are warranted from an interventions perspective. Given the different processes potentially underlying the social cognition domains, it is not possible to make conclusions about social cognition deficits in schizophrenia based on reviews and meta-analyses of a subset of the social cognitive domains. Examining the existing literature in these additional domains, along with up-to-date research on the more widely studied domains of social cognition is, therefore, warranted. The goal of the current meta-analysis was to examine whether social cognitive assessments provide reliable evidence of impairment in schizophrenia. Specifically, we aimed to investigate (1) the magnitude of differences between SC and NC participants across all six domains of social cognition (and consequently, whether the focus on ToM and facial emotion perception has overshadowed the potential relevance of other domains), (2) whether SC participants are equally impaired across domains, and (3) to what extent, if any, could demographic or disease burden variables explain variability within domains.

Methods

The methods of this study met criteria specified by the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) statement.¹⁶

Literature Search

We conducted a literature search of the PsycINFO database using the following keywords: social cognition, theory of mind, emotion perception, social perception, attributional bias, and schizophrenia. In PubMed, we used the following search string: “([social cognition {tiab} OR theory of mind {tesh} OR social perception {mesh} OR attributional style OR emotion perception OR emotion processing] AND schizophrenia [major]) OR (schizophrenia [major] AND [social behavior {major} AND {cognition <major> OR cognition disorders <major>}]).” The abbreviations in the string refer to title or abstract (tiab), medical subject heading (mesh), and major subject heading (major). The 2 searches combined yielded 888 unique articles, all published between 1980 and November 2011. We limited our search to articles written in English and those describing studies with human subjects. We also conducted an additional search in PubMed to capture articles that may have escaped processing by the National Library of Medicine. This was accomplished using the same search in a “keyword” format (ie, no Boolean indicators and limiting the search to “NOT MEDLINE” articles), therefore leaving us with only the most recent, non-indexed references (unique number of studies, ie, “k” = 224).

Study and Data Selection

We included studies that met the following criteria: (1) article written in English; (2) peer-reviewed publication; (3) psychosis sample with at least 90% of participants diagnosed with schizophrenia or schizoaffective disorder; (4) NC sample; (5) results reported as means and standard deviations, or F or t values so that effect sizes could be calculated. In cases where these data were not reported, we contacted the authors and included those studies if we received usable data.

We excluded studies with samples that overlapped with other published reports. In articles reporting results from overlapping samples, we chose the study with either the largest sample or the one with the most relevant/usable data. When multiple tasks were used to assess 1 single social cognition domain, we selected scores on 1 task (that we determined as most psychometrically sound or that was most commonly used across studies) per domain. We did not include results from any studies with overlapping samples within domains, however, we did include the same studies across domains (eg, the study by Addington et al¹⁷ included measures for social perception, social knowledge, and emotion perception, therefore, Addington et al contributed 1 data point to each domain).

A total of 112 studies, yielding 154 effect sizes, met our selection criteria (table 2). We excluded articles from the total number (k = 1112) yielded by our search strategies for the following reasons: reviews or letters to the editor (k = 213), no NC sample (k = 393), no social cognition measures (k = 129), no schizophrenia sample/atypical schizophrenia sample, eg, prodrome, ultra-high-risk, or child sample (k = 198), <90% of sample with schizophrenia or schizoaffective disorder (k = 46), unusable data/no responses from authors to e-mails requesting data (k = 51), overlapping samples (k = 48). (Some articles met multiple exclusion criteria, but we only counted them once.) Among the non-indexed articles we found in PubMed (wherein we could not specify limits), we excluded 26 articles describing animal studies and 10 articles written in a language other than English.

Table 2.

Studies Included in the Meta-analysis: Description of Samples and Domain-wise Social Cognition Tasks

		First Author		Location of Research		N (SCs)		N (NCs)		Years Ill (SCs)		Sample Description (SCs)		Social Cognition Task		Domain
1		Addingtona		Canada		53		55		>3		O		FEIT		EP
														Social Cue Discrimination Test		SP
														Situational Features Recognition Test (action)		SK
2		Baasa		the Netherlands		18		28		9.9		O		Trustworthiness Social Judgment Task		SP
3		Badan Bâ		Switzerland		16		16		9.5		O		ToM: First-order false belief stories		ToM
4		Bediou		France		30		30		8.82		"clinically stable"		Facial affect recognition		EP
5		Benedetti		Italy		24		20		12.7		O		Comic-strip stories (ToM condition; errors)		ToM
														Comic-strip stories (Affective empathy condition; errors)		EP
6		Bigelow		United States		20		14		2.7		B; recent-onset		Movie stills—Unmasked Faces Test (accuracy)		EP
7		Boraa		Australia, Turkey		91		55		10.7		O		Eyes Test		ToM
8		Borod		United States		20		21		13.58		"chronic"		FEIT		EP
9		Bozikas		Greece		36		30		10.92		I		Cartoon stories (developed for Greek population; first-order false belief)		ToM
10		Brünea		Germany		23		18		12.3		Both		Facial affect recognition task (total score)		EP
														Cartoon picture stories (first-order false belief)		ToM
11		Brünea		Germany		38		29		8.8		NR		Cartoon picture stories-Mental states total (a + b)		ToM
12		Brunet		France		25		25		21.6		NR		Attribution of Intention Task		ToM
13		Chambon		France		26		26		9.6		I		Facial emotion recognition task (discriminability for upright faces)		EP
14		Champagne-Lavau		Canada		31		29		16.2		O		Comic-strip stories		ToM
15		Chen		United States		19		30		18.1		NR		NimStim Face Stimulus Test (“happy” condition)		EP
														Eyes task		ToM
16		Chung		United States		35		32		NR		O		Emotional Context Processing Task (valence ratings across conditions)		EPr
17		Corcoran		United Kingdom		59		44		13.9		O		Hinting Task		ToM
18		Corcorana		United Kingdom		55		30		NR		I		Hinting Task		ToM
19		Corrigan		United States		26		14		NR		O		Schema Component Sequencing Task-Revised (Combined Juxtaposition score)		SK
20		Corrigan		United States		24		15		15.2		I		Social Cue Perception Task (sensitivity to social cues)		SP
21		Couturea		United States		44		41		5.5		"mild psychopathology"		Movie stills—with face (total)		EP
														The Abbreviated Trustworthiness task—Untrustworthy faces		SP
														Eyes task		ToM
22		Csukly		Hungary		58		29		10.8		O		The Emotion Hexagon Task (total across conditions)		EP
23		Das		Australia		23		22		9.4		NR		Online implicit mentalizing task (Intentionality condition)		ToM
24		de Achával		Argentina		20		20		NR		O		Faces Test (emotion recognition)		EP
														Reading the Mind in the Eyes Test		ToM
25		Diaz		United States		11		17		13.7		NR		Working memory based on IAPS photographs (accuracy)		EPr
26		Donohoe		Ireland		73		78		18.2		O		IPSAQ EB		EB
														IPSAQ PB		PB
27		Edwards		Australia		29		24		NR		O; first-episode		Facial Affect Computer Tasks (FACT) #4, Emotion labeling task		EP
28		Feinberg		United States		20		20		NR		I		Faces with standardized emotions (Emotions labeling task)		EP
29		Fisher		United States		91		30		NR		O		Facial Affect Recognition		EP
30		Fujiwara		Japan		26		20		10.1		O		Perception of Affect Tasks (PAT)—Matching the social situation with emotional labels		ToM
														PAT—Matching emotional faces with emotional labels		EP
														PAT—Matching emotional faces with nonverbal social situations		EPr
31		Green		United States		81		46		FE		O; first-episode		MSCEIT (Identifying Emotions)		EP
														MSCEIT (Managing Emotions)		EPr
														TASIT (Perception of Social Inference—Enriched)		ToM
														Relationships Across Domains		SP
31a		Green		United States		53		47		>5		O; chronic		MSCEIT (Identifying Emotions)		EP
														MSCEIT (Managing Emotions)		EPr
														TASIT (Perception of Social Inference—Enriched)		ToM
														Relationships Across Domains		SP
32		Green		Australia		20		22		NR		O		Vignette-Face Task (accuracy)		EPr
														Facial emotion processing task (accuracy)		EP
33		Gur		United States		14		14		NR		O		Emotional valence discrimination task		EP
34		Hall		United Kingdom		20		20		NR		NR		Hexagon task (face emotion recognition)		EP
														Social Cognition Test (complex social judgments)		SP
35		Haralanova		Bulgaria		30		30		14.23		I		Task based on IAPS and Munich Affective Picture System (neutral stimuli, subjective emotional arousal)		EPr
36		Harringtona		New Zealand		25		38		10.9		B		Position sequencing position scores (social script and mechanical stories)		SK
														False Belief stories (first-order ToM)		ToM
37		Heimberg		United States		20		20		NR		I; first hospitalization, neuroleptic naïve		Facial emotion discrimination		EP
38		Hirao		Japan		20		20		10.6		NR		Reading the Mind in the Eyes		ToM
39		Hooker		United States		20		27		18.8		O		Facial affect recognition (based on Benton Test of Facial Recognition)		EP
40		Hooker		United States		21		17		24.47		O		The Recognition of Faux Pas test % correct, Faux Pas total score		ToM
41		Horan		United States		45		41		14.6		I		FEIT		EP
42		Ibanez		Argentina		13		13		12.46		NR		Valence categorization task		EP
43		Irani		United States		10		10		NR		O		Reading the Mind in the Eyes Task (accuracy)		ToM
44		Johnston		Australia		18		18		>2		O		Facial emotion recognition (accuracy)		EP
45		Kantrowitz		United States		41		41		NR		I		Voice emotion recognition		EP
46		Kelemen		Hungary		52		30		NR		O		Eyes Test		ToM
47		Kern		United States		50		44		NR		O		TASIT (Perception of Social Inference—Enriched)		ToM
48		Kern		United States		176		300		19.5		O		MSCEIT (Managing Emotions)		EPr
49		Kerr		United States		29		23		NR		I		FEIT		EP
50		Kinderman & Bentall		United Kingdom		20		20		NR		I		IPSAQ EB		EB
														IPSAQ PB		PB
51		Kington		United Kingdom		16		16		11.96		B		Expression Recognition Test—Basic Emotions, Faces		EP
														Expression Recognition Test—Complex Mental States, Eyes		ToM
52		Kline		United States		27		15		NR		O		Facial affect labeling		EP
53		Kohler		United States		35		45		5.6		O		Emotion valence task		EP
54		Kosmidis		Greece		28		26		11.4		NR		TASIT (modified for Greek sample, Perception of Social Inference)		ToM
55		Kosmidis		Greece		37		32		10.9		Both		Emotion discrimination task		EP
56		Kubota		France		25		10		12.5		Both; hearing-impaired		Facial affect labeling task		EP
57		Kucharska-Pietura		Poland		100		50		7.6		I		Facial emotion recognition task		EP
58		Kuperberg		United States		18		18		16.9		O		Two-sentence scenarios to assess perception of emotional valence		EP
59		Langdona		Australia		35		34		12.6		O		IPSAQ EB		EB
														IPSAQ PB		PB
														Picture Sequencing Task—False Belief		ToM
														Picture Sequencing Task—Social Script		SK
60		Lee		United States		12		13		NR		O		Belief Attribution Task: False Belief		ToM
61		Leentjens		Netherlands		26		24		18		O		Facial affect discrimination task		EP
62		Leitman		United States		43		34				NR		FEIT		EP
63		Lepage		Canada		30		28		8.5		O		Facial emotion valence perception		EP
64		Leppanen		South Africa		44		40		13.7		O		Mac-Brain Face Stimulus Set (% hits across "happy" conditions)		EP
65		Lewis		United States		18		10		NR		I		Facial affect recognition (errors)		EP
66		Lincoln		Germany		75		75		10		B		Theory of mind movie task of social situations (“intentions” condition)		ToM
														IPSAQ EB		EB
														IPSAQ PB		PB
67		Marjoram		United Kingdom		15		15		11.3		B		Hinting Task		ToM
68		Martin		France		20		20		12.6		I		Facial affect matching task (accuracy)		EPr
69		Martino		Argentina		21		15		8.57		O		Faux-pas task		ToM
70		Mathewsa		United States		40		40		NR		O		Facial emotion recognition (discrimination)		EP
														TASIT (Perception of Social Inference subtests)		ToM
71, 72		Matsui / Sumiyoshib		Japan		25		32		NR		O		Frequency judgment task		SP
73		Mier		Germany		16		16		NR		O		Affective ToM task		ToM
														Facial recognition of emotion		EP
74		Mo		China		29		22		19.34		I		Sally-Ann stories (first-order false belief)		ToM
75		Montag		Germany		80		80		9.8		B		Movie for Assessment of Social Cognition (mental state decoding)		ToM
76		Mueser		United States		28		15		NR		I		FEIT		EP
77		Novic		United States		17		17		10.6		I		Facial affect recognition		EP
78		Park		United States		20		16		NR		O		Emotional identification		EP
79		Pijnenborg		Netherlands		46		53		7		O		Facial expression identification task		EP
80		Pinkhama		United States		49		44		10.4		O		FEIT		EP
														Schema Component Sequencing Task (accuracy)		SK
														Hinting Task		ToM
81		Pinkhama		United States		24		12		NR		O		Trustworthiness/approachability task (% rated trustworthy)		SP
82		Pousa		Spain		61		51		10.69		O		Picture Sequencing Task (Social Script)		SK
														Picture Sequencing Task (False Belief)		ToM
83		Randalla		United Kingdom		32		18		NR		O		ToM: first-order false belief stories		ToM
														IPSAQ EB		EB
														IPSAQ PB		PB
84		Reske		Germany		12		15		NR		First-episode		Facial Emotion Discrimination Task (% correct)		EP
85		Rubin		United States		22		31		13.23		NR; all women		Penn Emotion Acuity Test		EP
85a		Rubin		United States		26		26		11.58		NR; all men		Penn Emotion Acuity Test		EP
86		Russell		United Kingdom		5		7		13		NR		Reading the Mind in the Eyes Test (errors)		ToM
87		Sachs		Austria		40		43		3.7		I		Emotion differentiation test		EP
														Penn Emotion Acuity Test		EPr
88		Sarfatia		France		25		15		NR		I		Comic-strip stories (ToM)		ToM
89		Sasson		United States		10		10		4.2		O		Movie Stills Task (Face-present condition)		EP
90		Schimansky		Switzerland		40		39		12.5		B		Reading the Mind in the Eyes Test		ToM
91		Schneider		Germany/United States		20		20		NR		I		Facial Emotions for Brain Activation Test-Emotion Discrimination (accuracy)		EP
92		Scholten		The Netherlands		53		42		5.9		Both		Facial affect recognition task		EP
93		Shamay-Tsoorya		Israel		24		28		NR		O		Ekman Faces Test		EP
														Cognitive ToM task		ToM
94		Stewart		United Kingdom		59		38		13.95		B		Request/Response Task (Knowledge Attribution)		ToM
95		Streit		Germany, Japan		15		12		NR		I		Pictures of Facial Affect (first affect recognition run)		EP
96		Suslow		Germany		49		28		NR		Both		Emotion priming/Facial emotion discrimination task (decision latencies)		EPr
97		Tenyi		Hungary		26		26		NR		NR		Violations of the Maxim of Relevance (ToM vignettes)		ToM
98		Tomlinson		United Kingdom		16		24		NR		NR		Point-light images		EP
99		Toomeya		United States		28		28		NR		I; institutionalized		Profile of Nonverbal Sensitivity Test (total)		SP
100		Tse		Hong Kong		40		46		15.15				Facial Affect Perception Test (FAP) (total error)		EP
101		Tso		United States		33		33		17.9		O		Reading the Mind in the Eyes Task		ToM
														MSCEIT (Managing Emotions)		EPr
102		Tsoi		United Kingdom		30		30		17.5		B		Six sets of cartoon picture stories		ToM
103		Turetsky		United States		16		16		9		O		Penn Facial Emotion Stimuli		EP
104		Van't Wout		the Netherlands		37		41		9.62		Both		Facial affect recognition (degraded faces) errors across emotion conditions		EP
105		Vaskinn		United States		72		58		NR		O		The Interpersonal Perception Task-15		SP
106		Vistolia		France		19		21		NR		NR		Comic-strips stories (ToM; errors)		ToM
107		Weniger		Germany		45		30		6.7		NR		Facial affect recognition (errors)		EP
108		Williamsa		Australia		28		72		0.95		O; first-episode		Facial Emotions for Brain Activation Test (Emotion Discrimination accuracy)		EP
109		Wolwer		Germany		32		21		NR		I		Facial affect recognition		EP
110		Wynna		United States		33		42		NR		O		Half-PONS		SP
111		Yamashita		Japan		49		28		14		O		Means-Ends Problem Solving Procedure		SK
112		Zhu		China		40		31		9.2		I		Faux pas Test (Faux pas recognition questions)		ToM
113		Ziv		Israel		30		30		13.2		O		ToM stories (first-order false belief)		ToM
														Emotion Inference Questionnaire		SP

Notes: All references in this table are listed in the supplemental material. Tasks used in multiple studies: FEIT, Face Emotion Identification Test; IAPS, International Affective Picture System; TASIT, The Awareness of Social Inference Test; IPSAQ, Internal, Personal, Situational Attribution Questionnaire; PONS, Profile of Nonverbal Sensitivity; MSCEIT, Mayer–Salovey–Caruso Emotional Intelligence Test (In all cases, we used the names of the task as specified in the report.) Social cognition domains: ToM, Theory of Mind; EP, Emotion Perception; EPr, Emotion Processing; SP, Social Perception; SK, Social Knowledge; EB, Externalizing Bias; PB, Personalizing Bias; SC, schizophrenia sample; NC, normal comparison sample; NR, not reported; O, outpatients; I, inpatients; B, mixed inpatient and outpatient sample; Unless indicated, NOT specified as first-episode.

^aIndicate studies chosen among several with overlapping samples.

^bStudy authors indicated (via e-mail correspondence) that the 2 articles reported data collected at the same time from the SC and NC samples.

Data Extraction and Coding

Prior to conducting our analyses, we coded the following variables from the studies we reviewed: (1) first author, publication year, and country of origin; (2) number of participants, gender distribution, years of education in both groups; (3) duration of illness, inpatient vs outpatient status, first-episode vs not, type of antipsychotic medication, additional sample descriptors for the schizophrenia sample; (4) means and SDs, (or when unavailable), t values, F values, or chi-squared values for one measure per domain for each study. We classified emotion labeling tasks under the emotion perception domain, tasks requiring discrimination between emotion valences, and managing emotions under the emotion-processing domain. For attributional bias, we coded externalizing bias (EB) and personalizing bias (PB) separately; all studies included in our meta-analysis measured attributional bias with the Internal, Personal, and Situational Attributions Questionnaire (IPSAQ)¹⁸ (there are other measures of attributional bias that have been used in people with schizophrenia, but they were not among those used in the studies that met our selection criteria). The IPSAQ yields an EB score (calculated by subtracting the number of internal attributions for negative events from the number of internal attributions for positive events) and a PB score (calculated by dividing the number of personal attributions for negative events by the sum of both personal and situational attributions for negative events).

Statistical Analysis

In all domains except attributional bias, scores reflected social cognitive abilities (with lower scores indicating poorer social cognition). We examined PB and EB separately; the scores in this case reflected the direction of bias, rather than impaired vs intact performance. All analyses were conducted in STATA/IC, version 10 (StataCorp. 2007. Stata Statistical Software: Release 10. College Station, TX: StataCorp LP). We used standardized mean difference effect sizes with Hedges correction for small sample size bias, ie, “g” to compare the schizophrenia and NC samples.¹⁹ We calculated the mean effect sizes for each of the social cognition categories using the meta program. Forest plots and funnel graphs were created using the metagraph and metafunnel programs, respectively. We calculated a fail-safe n using Orwin’s formula,²⁰ using a criterion effect size of 0.2. The fail-safe n is the number of studies with effect sizes of zero that would reduce the mean effect size to 0.2, a small effect.²¹ We examined publication bias with Egger’s regression intercept test, a statistical test of funnel plot asymmetry, using the metabias program.

When the homogeneity analysis (Q-test²²) was significant and the I² index was greater than 50 (ie, more than 50% [a medium level] of total variability is due to true heterogeneity²³), we used a random effects model in the mean effect size analysis and followed up with meta-regressions to examine heterogeneity between studies (metan program).

The independent variables to be examined in the meta-regressions, determined a priori, included the following: age, years of education, percent male, illness duration, first-episode (vs chronic/multiepisode), inpatient status (vs outpatient), and English speaking (vs not). Age, education, and percent male were all entered as 2 variables each (eg, age of schizophrenia sample and age of normal controls) as all studies reported these variables as 2 separate categories. Due to the large amount of missing data from each of the categories, separate univariate regressions were conducted for each pre-determined variable, to ensure that all available data would be utilized. (If all independent variables were included in a single regression, the k would decrease by more than 50% for many of the analyses, thus severely limiting interpretation of results.)

Results

Study Characteristics

Forty-two of the 112 studies included in our final analysis were conducted in the United States. Assessments in 48 of the 112 studies were conducted in a language other than English (we assumed that assessments in the United States and the United Kingdom were conducted in English unless otherwise noted [k = 2]).

Sample Characteristics

The 112 articles (comprising 114 unique samples) reported on 3908 SCs and 3570 NCs. Not all studies reported demographic data, although most reported age of participants (k = 110). The average age of SCs and NCs was 35.1 years (mean SD = 9.1) and 34.5 years (mean SD = 9.4), respectively. SCs had fewer years of education than did NC participants (mean years = 12.3 and 13.9 years, respectively; k = 76). Most of the study participants were men (mean = 67.0% in SCs, k = 109 and mean = 60.5% in NCs, k = 106).

Mean duration of illness of the SCs was 11.9 years (mean SD = 8.1; k = 69), which included four studies of first-episode patients (37 studies did not describe their sample in terms of chronicity). Twenty-four studies were conducted with inpatients with schizophrenia, 52 with community-dwelling outpatients, and 15 with both (21 studies did not report type of treatment setting). Seventy-one-point-three percent of the SCs across the studies were prescribed at least one atypical antipsychotic medication. Studies widely differed in the assessments they used to measure presence and severity of psychopathology among their SC participants; however, from our examination of the symptom scores, it appeared that participants generally had a mild to moderate level of symptoms.

Effect Size Calculations

SCs performed more poorly on all domains of social cognition compared with NCs. Effects sizes are reported in table 3 and are described below. Please see supplementary material for forest plots and funnel graphs for studies under each domain.

Table 3.

Effect Sizes, Heterogeneity across Studies, and Publication Bias

Domain	k	ES (g)	CI	P	Q	df	P	n (NC)	n (SC)	Fail-safe n a	Bias Coefficient	Bias SE	P
Theory of mind	50	0.96	±13	<0.001	146.19	49	<0.001	1536	1760	191	3.2	0.80	<0.001
Social perception	13	1.04	±25	<0.001	37.91	12	<0.001	450	503	54	3.72	1.77	0.06
Social knowledge	7	0.54	±17	<0.001	10.48	6	0.106	263	298	12	3.62	2.69	0.237
Externalizing bias	5	−0.02	±38	0.918	14.56	4	0.006	225	221	−6	−0.71	3.26	0.841
Personalizing bias	5	−0.17	±55	0.532	29.33	4	<0.001	225	221	−9	−6.61	2.52	0.079
Emotion perception	62	0.89	±17	<0.001	324.63	61	<0.001	1715	1935	214	4.17	1.11	<0.001
Emotion processing	12	0.88	±30	<0.001	54.56	11	<0.001	638	574	41	−1.09	1.67	0.528

Note: k, number of studies; ES, effect size (Hedge’s g); CI, confidence interval; Q, homogeneity analysis.

^aNumber of studies with effect size of 0 needed to reduce the mean effect size to a criterion level (ie, d = medium effect size of 0.5 or small effect size of 0.2).

ToM (k = 50, n[NC] = 1536, n[SC] = 1760)

The mean effect for ToM was large (g = 0.96, with 95% confidence interval [CI] = 0.83 – 1.09, P < .001). We examined the potential role of age, education, gender, illness duration, inpatient status, and English-speaking status in the statistically significant heterogeneity across ToM studies (Q = 146.19, P < .001; I² = 66.5). There were no studies of ToM that included only first-episode patients. None of the variables examined significantly explained variability in the effect sizes (P ≥ .063). The fail-safe n was 191. Egger’s bias coefficient was significant (bias = 3.20, se = 0.80, P < .001) indicating possible publication bias.

Social Perception (k = 13, n[NC] = 450, n[SC] = 503)

The mean effect for social perception was large (g = 1.04, 95% CI = 0.79 – 1.29, P < .001). We examined the potential role of age, education, gender, illness duration, first-episode status, inpatient status, and English-speaking status to explain the heterogeneity among the effect sizes (Q = 37.91, P < .001; I² = 68.3).

Only inpatient status significantly accounted for variability in the social perception effect sizes (estimate = 1.07, SE = 0.29, P = .005), with inpatients having greater deficits than outpatients. The remaining variables were not significant (P ≥ .189). The fail-safe n was 54. Egger’s bias coefficient was not significant (P = .060).

Social Knowledge (k = 7, n[NC] = 263, n[SC] = 298)

The mean effect for social knowledge was medium (g = 0.54, 95% CI = 0.37 – 0.72, P < .001), and there was no statistically significant evidence of heterogeneity across these studies (Q = 10.48, P = .106; I² = 42.8). The fail-safe n was 12. Egger’s bias coefficient was not significant (P = .237).

Attributional Bias (k = 5, n[NC] = 225, n[SC] = 221)

The mean effect for both EB and PB was negligible (g = −0.02, 95% CI = −0.40 to 0.36, P = .918 and g = −0.17, 95% CI = −0.72 to 0.37, P = .532, respectively). Because both effect size estimates were nonsignificant, we did not conduct further analyses on these constructs. (We also conducted separate analyses on three of the four studies that included only SCs with persecutory delusions and found no difference in the results.)

Emotion Perception (k = 62, n[NC] = 1715, n[SC] = 1935)

The mean effect for emotion perception (k = 62) was large (g = 0.89, 95% CI = 0.72 − 1.05, P < .001). We examined the potential role of age, years of education, percent male, illness duration, first-episode status, inpatient status, and English-speaking status to explain the statistically significant heterogeneity across studies (Q = 324.63, P < 0.001; I² = 81.2).

Inpatient status was the only variable that significantly accounted for variability in the emotion perception effect sizes (estimate = 0.56, SE = 0.18, P = .002), with inpatients having greater deficits than outpatients. The remaining variables were not significant (P ≥ .140). The fail-safe n was 214. Egger’s bias coefficient was significant (bias = 4.17, se = 1.11, P < .001), indicating possible publication bias.

Emotion Processing (k = 12, n[NC] = 638, n[SC] = 574)

The mean effect for emotion processing (k = 12) was large (g = 0.88, 95% CI = 0.58 − 1.17, P < .001). We examined the potential role of age, education, gender, illness duration, first-episode status, inpatient status, and English-speaking status to explain the statistically significant heterogeneity across studies (Q = 54.56, P < .001; I²=79.8).

Illness duration was the only variable that significantly accounted for variability in the emotion processing effect sizes (estimate = 0.08, SE = 0.03, P = .04); longer illness duration accounted for more pronounced deficits. The remaining variables were not significant (P ≥ .118). The fail-safe n was 41. Egger’s bias coefficient was not significant (P = .528).

Discussion

The goal of this study was to review and examine the quantitative evidence of deficits in domains of social cognition in schizophrenia, including those that have not been extensively studied but deemed important by the NIMH consensus statement on social cognition in schizophrenia.³ SCs performed more poorly than did the NC participants across all domains. Each of the social cognition domains had high levels of heterogeneity among the effect sizes within the domains, except attributional bias. The direction of all effects, however, was the same, and we examined this heterogeneity with regression analyses.

There were no significant differences in the direction of attributional bias between the NCs and SCs, including the subset of patients with persecutory delusions. There was no heterogeneity in the effect sizes for PB or EB, and notably, unlike the other domains, the same measure of attributional bias was used across all the studies we included in our analysis. There was some evidence from the studies included in this meta-analysis that PB was associated with paranoid traits in people across both samples, but not specifically with a schizophrenia diagnosis. There is evidence from other studies comparing SCs with paranoia to those without, suggesting that the former tend to have an EB for negative events.²⁴

The large effect seen for ToM (Hedge’s g = 0.96 in 50 studies) was consistent with previous meta-analyses with fewer studies (Sprong et al¹⁴: k = 29, Cohen’s d = −1.26; Bora et al¹³ k = 36, d [range across individual tasks] = 0.90 − 1.08). The large effect for emotion perception (g = 0.89 in 62 studies) was also as expected, based on reviews and a recent meta-analysis of emotion identification in schizophrenia (Kohler et al¹⁵: k = 86, d = −0.91).²⁵ (Unlike Kohler et al,¹⁵ we selected 1 emotion identification task per study and excluded studies with overlapping samples.) Our results also call attention to the importance of social perception, which had the largest effect size (g = 1.04 in 13 studies), and emotion processing, which had a large effect size (g = 0.88 in 12 studies). These 2 domains have not been studied nearly as extensively as ToM and emotion perception, but the strength of their effects warrants future research on these domains. Social knowledge, too, is a lesser studied domain in schizophrenia; however, like crystallized “cold” cognition abilities, one might expect that declarative social knowledge, such as scripts for common social situations, may be relatively less impaired than other social cognitive abilities in schizophrenia (g = 0.54 in seven studies).

Heterogeneity in effect sizes between studies within the domains was not accounted for by age, gender, education, or language in the schizophrenia samples. Patients with longer duration of illness had greater deficits in emotion processing, and greater deficits in social perception and emotion perception were associated with inpatient status. Studies have demonstrated that fluctuations in social cognition impairments do not seem to be related to symptoms of reality distortion (ie, delusions and hallucinations), but have fairly strong relationships with disorganization and negative symptoms.¹⁶ We could not examine in detail the mechanisms underlying inpatient status and worse social and emotion perception deficits due to the lack of available data. However, given that hospitalization is typically associated with elevated positive symptoms (rather than negative symptoms),²⁶ our results suggest that acute psychosis may disrupt some aspects of social cognition. A recent factor analysis also showed that positive symptoms and agitation are also associated with a “hostile attributional style” among SCs²⁷ (we did not conduct any regression analyses for attributional style because of lack of heterogeneity in the effect sizes).

We were restricted in our ability to conduct multivariate regression analyses by missing data; furthermore, we were unable to use data on symptom severity, type and length of antipsychotic treatment, or similar disease burden variables, given the varied manner in which sample characteristics were assessed. However, the majority of the unexplained variance appears to stem from several factors related to the social cognition measures (table 2 demonstrates the heterogeneity in how a given construct was measured): (1) except for established and widely used tasks such as the Reading the Mind in the Eyes task, few measures were standardized; (2) tasks were modified, culturally adapted, and/or translated, such that equivalence of task versions was unknown. Social cognition tasks also face the problems that underlie neuropsychological tests; when tasks are not matched on relevant psychometric variables (such as task difficulty) and differ in their ability to discriminate between high and low performances, we cannot make inferences about differential deficits in a patient population with certainty.²⁸ Thus, apparent performance differences may be confounded by the differing discriminatory power of the tests.

Despite the limitations of the studies we reviewed, the results are important in that they suggest that social cognition deficits clearly exist in schizophrenia across domains and may reflect some overlapping deficits among people with the illness. Some domains of social cognition have smaller effects than others, and most have substantially smaller effects than certain nonsocial cognitive domains, notably, processing speed as measured by Digit Symbol tasks (g = 1.57).²⁹ However, a large effect size may not necessarily be indicative of a central or core deficit, whereas small effects can have a considerable impact on outcomes, depending on their position within a longer causal sequence. This point is also underscored in reviews and meta-analyses demonstrating that social cognition deficits are more strongly related to functioning than are neurocognitive deficits.^9,10 The meta-analysis by Fett and colleagues¹⁰ separated out the effects of social perception and emotion processing and perception on community functioning, and the results from the current meta-analysis place those findings in a broader context by quantifying the degree of deficit in those domains in a way that has not been done before. There is also evidence from evolutionary biology and cognitive neuroscience research that social and nonsocial higher-order cognitive processes are intricately related constructs and are associated with overlapping neurobiological systems (eg, medial prefrontal cortex) that are implicated in schizophrenia.^30,31 To that end, this meta-analysis does not allow for an estimation of the independence of social cognitive impairments from more general cognitive impairments, or the specific processes underlying social cognitive deficits. Indeed, the clinical study of social cognition may have far outpaced the development of conceptual models and finer measures of social cognition in schizophrenia.

Conclusions

In conclusion, previously published meta-analyses of social cognition in schizophrenia have focused on 2 domains (ToM and emotion perception) without comparing them to other domains. Given the multifaceted nature of social cognition in schizophrenia, comparing results across all six NIMH consensus domains is warranted in order to examine differential deficits across domains and predictors of impairment. Our meta-analysis replicated the results of existing meta-analyses of ToM and emotion perception, but we also examined the other four domains of social cognition in schizophrenia, with an emphasis on the magnitude of and potential variables associated with heterogeneity. Furthermore, our study highlights the importance of further research on social perception and emotion processing, which showed some of the largest effect sizes, yet have been less studied than some of the other domains. Additionally, deficits in both of these domains were related to disease-burden variables (inpatient status and chronicity, respectively) and may have implications for functional intervention. Our study also found no significant differences in attributional style between the schizophrenia and healthy comparison groups, with no heterogeneity in effect sizes within the PB and EB sub-domains. Heterogeneity of results within the other domains was largely unexplained, which underscores the need for development and use of psychometrically sound assessments that can reliably measure social cognition in schizophrenia. Future models of functional impairment in schizophrenia that incorporate social cognition will be needed, along with a better understanding of the biological and environmental underpinnings of such impairments, to inform more effective treatment.

Funding

This work was supported by National Institute of Mental Health grants (T32MH 019934 to G.N.S. and L.V.) and (R01MH 080150 and R34 MH93453 to E.W.T.).

Supplementary Material

Supplementary material is available at http:// schizophreniabulletin.oxfordjournals.org.