The aim of our study was to investigate cross-sectional associations among different neurocognitive domains and social withdrawal in a large and heterogeneous sample of SCZ patients. Furthermore, we investigated the variance explained of social withdrawal indicator by cognitive domains and additional socio-demographic and clinical factors.
Firstly, we found a negative association between global cognitive functioning and the degree of social withdrawal (Table 2). Our result is consistent with previously published results, although the association appears slightly weaker (r = 0.12) in comparison to the result of a recent review reporting a medium-sized average association (r = 0.14–0.26) between overall neurocognition and functional outcome in SCZ . Considering the high percentage of variation (up to 70%) of effect sizes of this association due to heterogeneity among studies, our findings can be partly explained by the choice of a social functioning measure that focused on a particular aspect of social behavior (i.e. social withdrawal) in comparison with more general measures of functional outcome considered in the review’s studies (such as community functioning, social behavior in the milieu, etc.) .
Secondly, our results are consistent with previous literature in confirming the associations among neurocognitive performance and different measures of social functioning in SCZ patients [29, 33, 34]. Indeed, previous studies showed that discrete neuropsychological domains are associated in different and specific ways with various indicators of social functioning. In particular, our findings suggested the association of three specific cognitive domains (i.e. verbal memory, processing speed, and working memory) with social withdrawal behaviour. Among them, “verbal memory” scores were most strongly associated with social withdrawal indicator (r = 0.14). In SCZ, verbal memory deficits appear to be independent of clinical state or medication effects [49, 50] and are considered by some authors as the greatest cognitive predictive factors of social outcome [26, 51]. Consistently, in previous studies verbal memory deficits have been associated with impairments in recreational activities , interpersonal relationships , and social problem solving .
On the other hand, “processing speed” and “working memory” scores were associated with social withdrawal indicator as well (r = 0.09 and r = 0.11, respectively). Within these two domains, the tasks associated with social withdrawal assessed respectively verbal fluency skills (assessed by COWAT and Category Instances test) and auditory working memory skills (evaluated by LNS test). Processing speed and working memory deficits are thought to contribute to impairments in other cognitive domains in SCZ [54, 55] and their association with general functioning is well-established in SCZ, as well as their effects on specific functional domains (e.g. social behavior, community functioning, competitive employment) [24, 56].
In a previous large study focused on the specific relationships between neurocognition and general functioning in SCZ , verbal memory was found to predict only functional competence (an indicator of everyday functioning skills), whereas processing speed and working memory were associated with both functional and social competence and performance domains (with interpersonal behavior and work performance, respectively) . In another study with a large sample of SCZ patients, Perlick and colleagues  showed an association among different cognitive domains and functional status assessed through QOLS . Only motor and memory skills were associated with social behavior (evaluated by the “Interpersonal relationship” domain of QOLS). Finally, Lipkovich and colleagues  found in a sample of 414 SCZ outpatients that multiple functional domains (assessed with QOLS ) were cross-sectionally associated with processing speed, working memory and verbal memory, although only processing speed skills were associated with the “Interpersonal relationship” domain of QOLS. These results consistently suggested specific associations among verbal memory, processing speed and working memory with different aspects of general functioning and social behavior, as showed by our results. Nonetheless, the paucity of available studies prevents us from drawing clear conclusions about the specificity and the reproducibility of these associations.
Exploratory analyses found in “Caucasian” and “Employed” subgroups associations between the neurocognitive composite score and social withdrawal score, but the same association was not found in “Non-Caucasian” and “Unemployed” subgroups. Moreover, in the “Employed” subgroup, we found an association between social withdrawal and “Reasoning” domain. Therefore, it seems that the relationship between social withdrawal and neurocognition tends to be mainly expressed within the “Caucasian” and “Employed” subgroups. Interestingly, although the “Unemployed” subgroup results to be strongly associated with social withdrawal, we found only one correlation between social withdrawal and cognitive domains (specifically with “Verbal Memory” domain) within this subgroup. We assume that “Verbal Memory” might have a critical impact on social withdrawal and this is somehow confirmed by the persistence of a significant association between this cognitive domain and social withdrawal in the multiple regression analysis. Although potentially interesting, these results should be interpreted with care because of the intrinsic increased risk of false positive findings in exploratory analyses and of the limited characterization of the different ethnicities within the sample.
Regarding socio-demographic features, social withdrawal was associated with lower education and “Employment” status (with “Unemployed” patients showing higher degree of social withdrawal). These results are consistent with the previous literature. Indeed, several studies showed that low IQ and scholastic underperformance represent predictive factors of further worse functional outcome [57, 58]. Conversely, the presence of a lower IQ and cognitive impairment after SCZ onset was found to be predicted by poor premorbid social functioning . The higher degree of social withdrawal shown by unemployed patients can be explained in a bidirectional way as well. It could depend either on the negative effect of pre-existing social deficits on the attainment of employment by SCZ patients , or, on financial constraints due to unemployment that would impede the access to recreational activities .
Concerning psychopathological symptoms, social withdrawal was significantly associated with each of the PANSS subscale scores (positive, negative and general psychopathology). Our finding is consistent with previous studies, which reported stronger associations with social dysfunction for negative and general symptoms [16, 30, 32, 61]. On the other hand, the observed relationship among negative and general psychopathology scale scores with social withdrawal was quite predictable because of the contiguity and the overlap of several items of the two rating scales with social withdrawal indicator (e.g. N4 = “Passive/apathetic social withdrawal”; G16= “Active social avoidance”). In addition, numerous other items concerning depressive or anxious states and cognitive alterations (e.g. G10 = “Disorientation”, G11= “Poor attention”) may contributed to strengthening this correlation .
Finally, the multiple regression analysis reported a total explained variance (calculated considering cognitive, clinical and socio-demographic factors) of the social withdrawal indicator of 8%. Our findings showed a smaller explained variance of social functioning by cognitive, clinical and socio-demographic factors in comparison to previously reported findings [61, 62]. Nonetheless, this could be due to the exclusion in our analyses of PANSS negative and general psychopathological scores (due to the presence of items that overlap with the social withdrawal indicator, as previously stated) normally considered in past studies and by the choice of a specific measure of social dysfunction (represented by social withdrawal) that could have further reduced the strength of the association.
4.1 Limitations of the study
There are several limitations in this study that need to be acknowledged. Firstly, the study was based on cross-sectional data not allowing causal inferences that require longitudinal observations. Secondly, the use of a post-hoc and derived measure of social withdrawal (item 8 of QOLS) can be arguable because it lacks specificity and direct validation for the purpose. Moreover, single item indices generally have more measurement error that multi-item indices reducing the magnitude of potential correlations with other measurements. Furthermore, the evaluation of social withdrawal was based on clinical judgement (QOLS ) that could be affected by bias due to its subjective nature. At this moment in time, efforts are ongoing to develop and implement new longitudinal, quantitative and objective measures of social functioning using digital phenotyping methods , and future studies are needed to understand the relationship between subjective and objective measures for this behavioural domain. Finally, since we decided not to apply any statistical correction because our analyses were all hypothesis-driven, we set the statistical significance at the standard level of p < 0.05 with the increased risk of false positive findings, particularly for the exploratory analyses.