Theory of mind in schizophrenia: a comparison of subgroups with low and high IQ

Abstract

Background

Social cognitive impairment is common in schizophrenia, but it is unclear if it is present in individuals with high IQ. This study compared theory of mind (ToM) in schizophrenia participants with low or high IQ to healthy controls.

Methods

One hundred and nineteen participants (71 healthy controls, 17 high IQ (IQ ≥115), and 31 low IQ (IQ ≤95) schizophrenia participants) were assessed with the Movie for the Assessment of Social Cognition, providing scores for total, cognitive, and affective ToM, along with overmentalizing, undermentalizing, and no-mentalizing errors. IQ was measured with Wechsler Abbreviated Scale of Intelligence; clinical symptoms with the Positive and Negative Syndrome Scale.

Results

Healthy controls performed better than the low IQ schizophrenia group for all ToM scores, and better than the high IQ schizophrenia group for the total score and under- and no-mentalizing errors. The high IQ group made fewer overmentalizing errors and had better total and cognitive ToM than the low IQ group. Their number of overmentalizing errors was indistinguishable from healthy controls.

Conclusion

Global ToM impairment was present in the low IQ schizophrenia group. Overmentalizing was not present in the high IQ group and appears related to lower IQ. Intact higher-level reasoning may prevent the high IQ group from making overmentalizing errors, through self-monitoring or inhibition. We propose that high IQ patients are chiefly impaired in lower-level ToM, whereas low IQ patients also have impaired higher-level ToM. Conceivably, this specific impairment could help explain the lower functioning reported in persons with intact IQ.

Keywords:

• Social cognition
• mentalizing
• cognitive heterogeneity
• intelligence

1. Introduction

Cognitive impairments are considered core features of schizophrenia [1–4], and both social and nonsocial cognition show inter-individual heterogeneity [4–6]. The two types of cognitive impairments are related [5], exemplified by social cognition serving as a mediator of the association between nonsocial cognition and functional outcome [7]. Furthermore, whereas social cognitive impairments can be seen in patients with normal-range nonsocial cognition, normal-range social cognition and impaired nonsocial cognition are rarely seen together [8]. This ‘single dissociation’ could suggest that normal-range nonsocial cognition is necessary, but not sufficient, for intact social cognition.

Among the social cognitive domains, theory of mind (ToM) has the greatest impact on functional outcomes [9]. ToM is the ability to infer the mental states of others, considered to be a complex, higher-level cognitive process which is dependent on lower-level processes, such as emotion perception [10,11] and basic nonsocial cognition [11,12]. It can be divided into two components: cognitive ToM, attributing thoughts, knowledge and plans, and affective ToM, attributing emotional states [13,14]. In clinical populations, researchers often distinguish between three error-types that people can make when solving ToM tasks. These are overmentalizing, i.e. a propensity to excessively attribute mental states and intentions, undermentalizing, i.e. reduced ability to understand and attribute mental states, and lack of mentalizing (or ‘no-mentalizing’), i.e. entirely failing to attribute mental states [15,16].

A recent meta-analysis concluded that there are moderate associations between various domains of nonsocial cognition and ToM in schizophrenia without significant differences between domains [17]. Vaskinn et al. [11] found associations between ToM and a nonsocial composite score, accounting for between 8 and 18% of the variance in different ToM subscores. In a follow-up study of the specific neuropsychological tests included in that composite score, Sjølie et al. [12] reported that no single neuropsychological test, including a measure of IQ, uniquely contributed to ToM.

To better account for the cognitive heterogeneity observed in schizophrenia, there have been attempts at subgrouping based on nonsocial cognitive test scores. Studies indicate that cognitive heterogeneity in schizophrenia can be described by three distinct and reliably identified nonsocial cognitive subgroups [18]: a relatively intact group, with good cognitive performance, an intermediate group, with moderate levels of overall impairment, and a globally impaired group, with severe cognitive deficits. These subgroups have been identified both empirically and clinically from intelligence trajectories [19–21]. Whereas some authors [22,23] have found that the subgroups differ only in their level of overall impairment, others [20,21] have concluded that they also exhibit different cognitive profiles. They find that the relatively intact group has a profile more similar to healthy controls (HC) than to other schizophrenia subgroups. In this view, it is characterized by more isolated deficits than the other groups, for instance in executive functioning and attention [20], however, see also, [21].

Similarly, there are also reports of social cognitive heterogeneity in schizophrenia. Studies using cluster analysis have identified three subgroups, with relatively intact, impaired or very impaired social cognition [24–26]. The relatively intact groups appear to have only subtle social cognitive deficits when compared to HCs [25–27], and have better results than the low-performing groups on several measures of nonsocial cognition [24,25].

In conclusion, subgrouping based on IQ differences has been shown to be a useful tool in examining cognitive heterogeneity [18,28]. However, it is not known whether such subgroups have distinct social cognitive profiles. Developing a better understanding of social cognitive heterogeneity in schizophrenia is important. This could enable us to better tailor treatments to different groups of patients. ToM is a particularly interesting target in this respect due to its strong associations with functional outcomes.

The aim of the current study is to investigate subgroup differences in ToM, including ToM components and error-types, by comparing low and high IQ schizophrenia groups with a HC group. We hypothesize group differences for all ToM variables, with HC outperforming the high IQ schizophrenia group, which in turn will perform better than the low IQ schizophrenia group. We make no predictions for group differences in specific patterns of ToM performance, due to the mixed results described in the social cognition literature above.

2. Materials and methods

2.1. Participants

Forty-eight participants (31 male, 17 female) with a DSM-IV diagnosis of schizophrenia (n = 34) or schizoaffective disorder (n = 14), selected from a larger sample [11], were included, together with 71 HC (42 male, 29 female). We chose to include patients with schizoaffective disorder as their cognitive impairments are largely similar to those seen in schizophrenia [29]. All were participants in the Thematically Organized Psychosis (TOP) study at the Norwegian Centre for Mental Disorders Research (NORMENT). Schizophrenia participants were recruited from in- and outpatient units at Oslo and Akershus University Hospitals.

Inclusion criteria were Norwegian as first language or all compulsory schooling in Norway and age 18–55 years. Exclusion criteria were neurological disease or having been hospitalized following a head trauma. For the current study, schizophrenia participants with an IQ of 115 or higher, as assessed with the subtests Similarities and Matrix Reasoning from Wechsler Abbreviated Scale of Intelligence (WASI) [30], were selected for the high IQ schizophrenia group, whereas participants with an IQ between 71 and 95 were selected for the low IQ schizophrenia group. The upper cutoff of 95 was chosen to achieve statistical power and to adhere to our previous studies [23]. HCs from the same geographical area were randomly selected from national statistical records and invited by letter. They were screened with the Primary Care Evaluation of Mental Disorders (PRIME-MD) [31] interview and excluded if mental, neurological, or somatic disorder was present. Participants gave informed consent to participate in the study, which was approved by the Regional Ethical Committee. Clinical assessments were based on semi-structured interviews and chart reviews. Diagnostic assessments were conducted with the Structured Clinical Interview for DSM-IV (SCID-I) [32] by medical doctors and psychologists who had completed a clinical training program from University of California, Los Angeles (UCLA) [33]. Cognitive assessments were conducted by psychologists or master level psychology students, trained by an experienced clinical psychologist (AV).

2.2. ToM measure

ToM was measured with the Norwegian version of the MASC test [34]. It consists of a 15-minute video of two men and two women at a dinner party. Participants are asked to answer 45 multiple choice questions about the characters’ thoughts, feelings, and intentions. The test yields six scores, including a total score, MASCtot (range 0–45). The individual items were further divided into two ToM component categories, cognitive ToM (MASCcog, 0–26) and affective ToM (MASCaff, 0–18), as described in [11]. In addition, the MASC test distinguishes between three different error types (range 0–45 for each score): overmentalizing (MASCexc), undermentalizing (MASCless), and no-mentalizing (MASCno).

2.3. Symptom measure

Symptoms were measured using the Positive and Negative Syndrome Scale (PANSS) [35]. The Wallwork/Fortgang five-factor model [36] was chosen, as it has been shown to provide the best fit in several studies [37–39]. It includes the factors positive (PANSSpos), negative (PANSSneg), disorganized (PANSSdis), excited (PANSSexc) and depressed (PANSSdep). Ranges are listed in Table 1.

Table 1. Demographics and clinical features.

	HC (n = 71)	High IQ (n = 17)	Low IQ (n = 31)	Statistic		Post hoc
	Mean (SD)	Mean (SD)	Mean (SD)	Value	Sig.	Results
Demographics:
Age	29.3 (7.7)	30.7 (8.1)	27.9 (7.6)	F = 0.743	p = 0.478	–
Sex (males/females)	42/29	9/8	22/9	x²=1.877	p = 0.399	–
Education (years)	14.2 (2.1)	13.1 (1.7)	11.0 (2.2)	F = 25.617	p < 0.001	HC, High > Low
Living situation^a	–	11/0/6	12/10/9	x²=7.174	p = 0.013	–
WASI IQ	110.7 (10.8)	118.0 (2.7)	85.3 (6.6)	F = 104.367	p < 0.001	High > HC > Low
Clinical features:
GAF-symptoms	–	46.2 (10.4)	41.1 (8.1)^b	t = 1.847	p = 0.071	–
GAF-function	–	45.4 (6.9)	40.7 (8.0)^b	t = 2.017	p = 0.050	–
PANSS positive (range 1–28)	–	9.0 (3.3)	10.7 (4.0)	t = 1.462	p = 0.151	–
PANSS negative (range 1–42)	–	12.2 (4.8)	14.1 (6.0)	t = 1.084	p = 0.284	–
PANSS disorganized (range 1–21)	–	4.4 (1.5)	6.7 (2.3)	t = 3.592	p = 0.001	–
PANSS excited (range 1–28)	–	4.8 (1.2)	5.9 (1.8)	t = 2.167	p = 0.035	–
PANSS depressive (range 1–21)	–	9.0 (2.6)	6.9 (3.0)	t = 2.380	p = 0.021	–
Illness duration (yrs)	–	7.8 (6.2)	5.5 (4.7)^c	t = 1.427	p = 0.161	–
Medication (DDD)	–	1.38 (1.00)^d	1.45 (0.86)^c	t = 0.245	p = 0.808	–

Note. The significance tests were one-way ANOVA with Tukey’s HSD (age, education, IQ), Fisher–Freeman–Halton exact test (sex), Fisher’s exact test (living situation), and independent samples t-tests.

^aOutpatient/inpatient/missing

^bn = 30 due to missing data

^cn = 29 due to missing data

^dn = 16 due to missing data.

2.4. Statistical analyses

Analyses were performed using IBM SPSS Statistics 27th version (2020). The Kolmogorov–Smirnov–Lilliefors significance correction revealed non-normal scores in several cells. We therefore chose nonparametric statistics in all analyses. We applied six Kruskal Wallis H-tests to determine whether the low IQ schizophrenia, high IQ schizophrenia, and HC groups differed for any of the MASC scores, followed by a post hoc Dunn-Bonferroni comparison to identify differences between each pairing of the three groups. Effect sizes were calculated as eta squared (η2) for the Kruskal Wallis H-tests and as Hedge’s g for the post hoc tests. We adjusted the p-values for the two ToM components (p = 0.05/2 = 0.025) and the three error types (p = 0.05/3 = 0.017). The Hettmansperger and McKean method [40] – a nonparametric alternative to the ANCOVA – was used to investigate whether any significant group difference in ToM between the two schizophrenia groups remained after controlling for differences in symptom load. Due to its high rate of type II errors, nonsignificant results should be interpreted with some caution.

3. Results

Table 1 presents the preliminary analyses with demographics and clinical features. The low IQ schizophrenia group had shorter education than the other two groups. It also consisted of more inpatients, with more disorganized, excited and depressive symptoms (PANSS) than the high IQ schizophrenia group.

Table 2 presents the group comparisons of MASC performance.

Table 2. Statistical analyses of MASC scores.

	HC (n = 71)	High (n = 17)	Low (n = 31)	Statistic			Post hoc	Pairwise ES (g)^a
	Mean (SD)	Mean (SD)	Mean (SD)	H	Sig.	ES (η^2)	Results	HC + high	HC + low	High + low
MASCtot	35.1 (4.1)	31.4 (6.1)	25.9 (7.5)	36.53	<0.001	0.298	HC > High > Low	0.83	1.74	0.78
Components:
MASCcog	21.3 (2.7)	19.2 (4.0)	15.2 (4.8)	38.60	<0.001	0.315	HC, High > Low	0.71	1.76	0.87
MASCaff	13.6 (1.9)	11.9 (2.6)	10.5 (3.2)	22.49	<0.001	0.177	HC > Low	0.78	1.27	0.47
Error types:
MASCexc	4.4 (2.8)	3.4 (2.5)	6.6 (3.8)	12.38	0.002	0.089	HC, High < Low	0.38	−0.71	−0.96
MASCless	3.8 (2.0)	6.8 (3.6)	8.0 (4.0)	29.95	<0.001	0.241	HC < High, Low	−1.23	−1.49	−0.31
MASCno	1.6 (1.4)	3.5 (2.4)	4.6 (3.0)	28.40	<0.001	0.228	HC < High, Low	−1.15	−1.48	−0.39

Note. High: high IQ schizophrenia group; low: low IQ schizophrenia group; ES: effect size; g: Hedge’s g; H: Kruskal–Wallis test statistic; η²: eta squared.

^aAs error type scores measure the number of errors made, effect sizes with negative values are in the hypothesized direction.

All six Kruskal–Wallis H-tests were statistically significant. The HC group had significantly better scores than the low IQ schizophrenia group for all MASC variables. Further, they significantly outperformed the high IQ schizophrenia group for MASCtot, MASCless and MASCno. The high IQ schizophrenia group performed significantly better than the low IQ schizophrenia group for three MASC variables: MASCtot, MASCcog and MASCexc.

A graphical depiction of group differences on the components is shown in Figure 1. Mean differences in error types are presented in Figure 2.

Figure 1. Cognitive (MASCcog) and affective (MASCaff) ToM in healthy controls (HC, red), high IQ schizophrenia (SZ high IQ, green) and low IQ schizophrenia (SZ low IQ, blue). The lines connecting individual bars indicate significant (*) and nonsignificant (ns) comparisons. Scores are presented as a percentage rate of correct answers.

Figure 2. Overmentalizing (MASCexc), undermentalizing (MASCless) and no-mentalizing (MASCno) errors in healthy controls (HC, red), high IQ schizophrenia (SZ high IQ, green) and low IQ schizophrenia (SZ low IQ, blue). The lines connecting individual bars indicate significant (*) and nonsignificant (ns) comparisons.

The Hettmannsperger and McKean follow-up analyses yielded significant group differences between the two schizophrenia groups after controlling for PANSSdep and PANSSexc but not after controlling for PANSSdis.

4. Discussion

We investigated ToM performance using the MASC test in high and low IQ groups of patients with schizophrenia compared to a HC group. As expected, the high IQ schizophrenia group performed at an intermediate level between the HCs and the low IQ schizophrenia group for most of the measured ToM indices. However, there were some important nuances to this picture which extend previous research and may be of help in tailoring interventions to specific patient groups.

First, the reduced performance of the high IQ schizophrenia group on the overall score (large effect size) and for affective and cognitive ToM (medium-large effect sizes) indicate a generalized ToM impairment. Group differences for the two ToM components, however, did not reach statistical significance, and impairments did not appear to be due to overmentalizing. The high IQ schizophrenia group made numerically fewer such errors than HCs, with a small-moderate effect size. In contrast, the low IQ schizophrenia group had reduced scores for all ToM variables compared to HCs, indicating global ToM deficits. These were especially pronounced for the overall ToM score and for cognitive ToM.

Second, although the high IQ schizophrenia group had better overall ToM than the low IQ schizophrenia group, this was reflected primarily in better cognitive ToM and less overmentalizing. One reason for the larger difference for cognitive than affective ToM may be that MASCcog is more cognitively demanding, perhaps requiring explicit mentalizing to a larger degree than MASCaff. Complex forms of mentalizing rely more on a cognitively demanding, explicit mentalizing system [41], and are indeed correlated with IQ [42]. Future studies should attempt to settle if there is in fact a larger difference in cognitive than affective ToM in IQ-based subgroups.

The similar level of overmentalizing in the high IQ schizophrenia and HC groups is consistent with previous studies showing less overmentalizing than undermentalizing and lack of mentalizing in schizophrenia [13,34,43]. However, these studies did not subgroup patients by IQ. The current findings suggest that overmentalizing in schizophrenia may be chiefly associated with lower IQ, whereas other errors may be determined by other factors. One could speculate that a superior intellect enables the high IQ schizophrenia group to evaluate the fallacious arguments involved in overmentalizing more critically. The excessive or irrational attribution of mental states seen in schizophrenia could require a ‘driver’, such as the ‘jumping to conclusions’ bias (which is also correlated with IQ; e.g. Tripoli et al. [44],) or the related construct of hypersalience [45]. Counteracting this tendency to overmentalize, a better self-monitoring and reasoning ability may enable the high IQ schizophrenia group to avoid making explicit overmentalizing errors. Thus, patients with lower IQ, but comparable levels of psychosis, may be as likely to generate overmentalizing responses, but less likely to inhibit them.

Undermentalizing or lack of mentalizing, on the other hand, may reflect an inability to initiate or maintain mentalizing, such as by failing to perceive or properly consider relevant social information. For example, undermentalizing and lack of mentalizing have been associated with emotion perception, while overmentalizing has not [11,43]. The two schizophrenia groups did not differ significantly with regard to the two former types of errors. Consequently, people with schizophrenia and high IQ could have substantial basic social cognitive impairments despite relatively intact self-monitoring (which is related to IQ in clinical populations, e.g. [46,47]) and higher-level reasoning ability.

Our main analyses did not take into account the IQ-level of our HCs, which may be of importance given that our high IQ schizophrenia group had a WASI IQ that was 8 points higher. With only six individuals with IQ ≤95 in the HC sample, significant power limitations make it problematic to conduct subgroup analyses. However, a follow-up analysis with four groups (high and low IQ HCs, high and low IQ schizophrenia) confirmed our findings (see Supplementary Table 1). Both schizophrenia groups had impairments compared to their IQ-matched HCs (no significant interaction effects). Thus, our findings point to the importance of social cognitive impairments in schizophrenia regardless of level of intellectual functioning. In other words, intact nonsocial cognition is not sufficient to prevent social cognitive impairment [8]. Our results align well with findings from studies on nonsocial cognition that identify similar residual deficits after accounting for IQ scores [23,48,49]. Furthermore, these specific social cognitive deficits could help to account for the marked discrepancy between nonsocial cognitive ability and functioning in patients with preserved IQ [21], explaining why they may struggle in everyday life despite having an intact IQ.

When controlling for differences in levels of disorganized symptoms (PANSSdis), the two schizophrenia groups no longer differed significantly for MASCtot, MASCcog or MASCexc. Disorganized symptoms are conceptually related to nonsocial cognition, and higher scores correlate with lower scores on measures of nonsocial cognition, while the other symptom factors correlate less [36,50]. PANSSdis has also been shown to mediate the relationship between nonsocial cognition and functional outcomes [51,52]. Hence, a significant portion of what is measured by PANSSdis likely overlaps with nonsocial cognition. While we cannot completely rule out the presence of a non-cognitive component in PANSSdis, we find it likely that PANSSdis does not represent a real confounding influence on our results. Further, there were more inpatients in the low IQ schizophrenia group. This is not surprising given the strong association between IQ and functioning in schizophrenia [21]. In fact, there were no inpatients in the high IQ schizophrenia group. When repeating the analyses with outpatients only, the differences in ToM between high and low IQ schizophrenia were no longer statistically significant, probably due to loss of power, although effect sizes remained largely similar (see Supplementary Table 2). For generalization purposes we argue that all participants should be included, across treatment levels.

The present study indicates that overmentalizing may be more associated with IQ than other types of mentalizing errors. Further, patients with schizophrenia and low IQ appear to have a larger impairment in cognitive than affective ToM. Such findings suggest that subgrouping methods may provide additional information about patients’ ToM functioning. This is in accordance with the conclusions drawn in IQ subgrouping studies on nonsocial cognition [19,21,53]. Our study adds to this research by demonstrating that people with schizophrenia and high IQ have deficits in social cognition, as they do in nonsocial cognition, and that these impairments may in fact be larger than the nonsocial cognitive impairments [19,21].

4.1. Limitations

There are a few limitations to our study. First, several assumptions of the parametric ANCOVA were not satisfied, in particular for the fairly small high IQ schizophrenia group. All nonparametric alternatives to the ANCOVA have issues with loss of power [54,55]. This could have contributed to our nonsignificant results when controlling for PANSSdis. A similar study using a larger sample might have the power to ascertain some of the differences we expected. This is also particularly relevant for our follow-up analyses with four groups and when excluding inpatient participants. Second, we chose to define our low IQ group as having an IQ score ≤95, which is a higher cutoff than what is usually seen in the literature. Despite this, we still found significant differences between the patient groups which in fact strengthen our results. A final limitation of our study is the use of a 2-subtest IQ measure, which prevented us from examining verbal and nonverbal contributions to social cognition separately.

4.2. Clinical implications

The functional implications of social cognitive impairments are well documented. Thus, it is important that clinicians be aware that these are present also in high IQ schizophrenia. Although these patients do not overmentalize, they have a propensity to undermentalize or not mentalize at all. Low IQ patients have an additional susceptibility to making overmentalizing errors. To overcome the negative consequences related to these impairments, specific strategies should be considered. These may include psychoeducation directed at those afflicted as well as their caregivers. Both low and high IQ groups may also benefit from cognitive remediation strategies targeting social cognition but perhaps with differentiated foci and content.

5. Conclusions

This study identified reduced ToM in intellectually high-functioning patients with schizophrenia. This group had similar levels of undermentalizing and no-mentalizing errors as the low IQ schizophrenia group, while their performance on overmentalizing errors was on par with the HC group. We suggest that this might be explained by a dissociation between impaired basic social cognitive functions and relatively well-preserved higher-level ToM abilities in the high IQ group. These qualitative differences in ToM performance between high and low IQ groups highlight the potential of IQ subgrouping methods for studying cognitive heterogeneity in schizophrenia.

Acknowledgements

We are grateful for the effort and time provided by our research participants. We would also like to extend our thanks to our colleagues and involved research personnel at the NORMENT center.

Disclosure statement

Anja Vaskinn has received honorarium from VeraSci Inc. Ole A. Andreassen is a consultant to HealthLytics and has received speaker’s honorarium from Lundbeck and Sunovion. No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the South-Eastern Norway Regional Health Authority (grants # 2010007 and # 2017069 to AV) and the Research Council of Norway (grant #271560 student research in psychology to ACS and HFR, and grant # 223273). The funding sources were not involved in study design, data collection, data analyses, drafting the manuscript or the decision to publish.

Notes on contributors

André C. Sahl

André Sahl, PsyD from the Department of Psychology, University of Oslo. He currently works as a clinical psychologist at CRUX in Stavanger, Norway.

Henning F. Rognlien

Henning Fleischer Rognlien, PsyD from the Department of Psychology, University of Oslo. He is currently working as a clinical psychologist at Drammen District Psychiatric Centre, Vestre Viken Hospital Trust, Norway.

Ole A. Andreassen

Ole Andreassen, MD PhD, Professor in psychiatry at University of Oslo. Director of the Norwegian Centre for Mental Disorders Research (NORMENT). His research activities has involved clinical, neurocognitive, brain imaging and molecular genetics tools to identify causes and underlying pathophysiology of psychotic disorder, and develop precision medicine tools in psychiatry.

Ingrid Melle

Ingrid Melle, MD PhD, Professor of adult psychiatry, Institute of Clinical Medicine, University of Oslo. She is co-director of the Norwegian Centre for Mental Disorders Research (NORMENT). Her research focuses on the course and outcome of psychotic disorders including schizophrenia and bipolar disorders.

Torill Ueland

Torill Ueland, PhD & specialist in clinical neuropsychology. Senior Scientist at the Norwegian Centre for Mental Disorders Research, Oslo University Hospital, Norway and Associate Professor in neuropsychology at the Department of Psychology, University of Oslo, Norway. Her research focuses on cognition in schizophrenia and bipolar disorder.

Anja Vaskinn

Anja Vaskinn, PhD & specialist in clinical adult psychology. She is currently a senior researcher at the Centre for Research and Education in Forensic Psychiatry at Oslo University Hospital, Gaustad, Norway, and has been affiliated with the TOP study since its beginning in 2002 (now called the Norwegian Centre for Mental Disorders Research). Her research has primarily focused on social and nonsocial cognition in different clinical populations, notably schizophrenia and bipolar disorder.