The Predictive Coding Account Of Psychosis: A Meta-Analysis Of The Relationship Between Mismatch Negativity And Symptom Severity

Background: The predictive coding (PC) theory of schizophrenia has gained traction in recent years as an explanatory model for the presence and severity of psychotic symptoms. PC posits that hallucinations and delusions emerge from a fundamental failure to accurately predict the environment and update prediction models when expectations are violated. Despite the conceptual link between the PC framework and symptom phenomena; however, a clear association between predictive coding abnormalities elicited by experimental paradigms and the severity of positive symptoms has not been consistently observed. Of the paradigms currently used to measure this phenomenon, the mismatch negativity (MMN) is arguably the most established. The MMN is an electrophysiological potential that is elicited when a sequence of tones is unexpectedly interrupted by a tone that deviates in one or more stimulus dimension, and is robustly impaired in people with schizophrenia (Erickson et al., 2015). The MMN is often held up as an exemplar index of the PC phenomenon (e.g., Wacongne et al., 2012), and some studies do suggest a relationship between impaired MMN and greater symptom severity (e.g., Fisher et al., 2011). However, one large study of MMN in people with schizophrenia (N = 877) revealed only a modest relationship between MMN amplitude and positive symptoms (r = .08; Light et al., 2014). Thus, there is a need to clearly establish the relationship between the MMN (a putative PC index) and symptom severity; failure to detect a consistent relationship between these measures may indicate that a need to reconceptualize the relationship between measures of PC and symptom expression. Methods: The present study used a meta-analytic approach to examine the pattern of symptom severity and MMN impairment across studies. Of the 101 studies of MMN impairment in schizophrenia described in a previous meta-analysis (Erickson et al., 2015), 33 studies reported a PANSS positive symptom subscale score and 30 studies reported a PANSS negative symptom subscale score. Separate meta-regressions were used to quantify the relationship between the effect size of MMN impairment in people with schizophrenia and positive and negative symptom severity, respectively. Results: We found that MMN effect size was not significantly associated with either positive (P = .24) or negative symptoms (P = .94). Conclusion: The present results suggest that a simple account of PC as it relates to symptom expression may not be adequate, and that a more nuanced conceptualization of this relationship is needed. For example, it is possible that MMN represents a “low-level” prediction error—one that occurs at a low level of the cortical hierarchy. By contrast, prediction errors that occur at higher levels of the hierarchy may exhibit stronger relationships with symptom measures. Future studies will be needed to test this hypothesis.