Skip to main content
NCBI home page
Schizophrenia Bulletin logoLink to Schizophrenia Bulletin
. 2025 Oct 29;52(5):sbaf188. doi: 10.1093/schbul/sbaf188

Relational Memory in Psychosis: Learning Trajectories and Associative Inference

Aslıhan İmamoğlu 1,✉, Suzanne Avery 2, Kristan Armstrong 3, Stephan Heckers 4
PMCID: PMC13645818  PMID: 41159629

Abstract

Background and Hypothesis

Schizophrenia is characterized by relational memory impairments that disrupt the ability to form novel inferences based on prior knowledge. The current study examined three aspects of relational memory impairments in psychosis. First, do relational memory deficits reflect specific difficulties in learning across repeated exposures? Second, can associative inference deficits be explained by a more general deficit in paired associate learning? Third, are relational memory deficits specific to schizophrenia or do they extend to affective psychosis?

Study Design

We studied persons with schizophrenia spectrum disorders (n = 244), persons with affective psychosis (n = 77), and healthy control participants (n = 186) with the Associative Inference Paradigm. Participants learned overlapping (A-B, B-C) and non-overlapping (D-E) pairs across multiple training blocks and were later tested on both directly studied (e.g., D-E) and inferred (i.e., A-C) associations.

Study Results

Persons with psychosis exhibited slower learning, a lower learning ceiling, and increased forgetting when compared to healthy control participants. Critically, associative inference impairments were disproportionally greater in schizophrenia and persisted even after accounting for paired-associate learning deficits.

Conclusions

Our results reveal that relational memory impairments in psychosis reflect difficulties in learning and retaining associations and that their integration, necessary to form novel inferences, is especially impaired.

Keywords: schizophrenia, cognitive impairments, episodic memory, forgetting

Introduction

Schizophrenia is associated with widespread episodic memory impairments,1 resulting in prominent functional deficits.2,3 Impairments are particularly pronounced in the ability to encode and retrieve relationships among multiple elements of an experience, namely relational memory, as opposed to memory for individual items.4,5 Relational memory supports the integration of information across multiple experiences, thereby facilitating novel inferences based on prior knowledge.6 Relational memory is compromised at various stages of schizophrenia,7–11 including first episode.12 These deficits are also present in first-degree relatives of individuals with schizophrenia,13–15 suggesting a heritable cognitive vulnerability associated with psychosis.5

The Cognitive Neuroscience Treatment Research to Improve Cognition in Schizophrenia Initiative (CNTRICS)16 recommended the Associative Inference Paradigm (AIP)17 for the study of relational memory. Since then, we have employed the AIP to assess relational memory in first episode18,19 and chronic schizophrenia.20,21 The paradigm involves an extensive training phase, during which participants learn overlapping (A-B, B-C) and non-overlapping (D-E) associations between item pairs across multiple blocks. A subsequent testing phase then assesses the ability to retrieve relationships between directly studied items and infer relationships between indirectly linked items (i.e., A-C). The ability to draw an associative inference reflects a specialized form of relational memory, which allows individuals to flexibly integrate related memories across separate episodes to derive novel associations.6 This process is thought to depend on hippocampal-prefrontal mechanisms that support both the encoding of individual episodes and the integration of overlapping events.22,23

In prior studies, we have consistently shown that individuals with schizophrenia exhibit greater impairments when responding to inferential (A-C) relative to non-inferential (D-E) pairs, suggesting that associative inference deficits are particularly pronounced in this population.18–21 However, given that successful inference relies on learning and maintaining directly studied associations, a process disrupted in psychosis,24 associative inference deficits need to be explained in the context of simpler deficits of paired associate learning.

Here, we report on 3 aspects of the AIP that we had not previously explored. First, we study the learning trajectories underlying associative memory. We examine whether group differences in relational memory are evident as early as the first testing block, persist with additional training, and reflect slower rates of learning or increased forgetting. Second, we evaluate the robustness of previously reported associative inference deficits in the largest psychosis cohort to date, and test if these deficits remain after accounting for impairments in learning directly studied associations. Finally, consistent with dimensional frameworks (e.g., HiTOP25) positing that cognitive deficits extend across diagnostic boundaries, we compare individuals with schizophrenia and affective psychosis to determine if relational memory deficits represent a transdiagnostic impairment across psychotic disorders.

Consistent with prior literature, we hypothesized that individuals with psychosis would show impaired relational memory compared to healthy control participants. We expected group differences in learning trajectories, with individuals with psychosis demonstrating slower acquisition across training blocks and greater forgetting between training and testing. We also expected to replicate prior findings of disproportionate impairments on inferential relative to non-inferential pairs and expand on them by showing that these deficits persist even when analyses are limited to trials in which participants successfully recalled the corresponding direct associations. Finally, consistent with prior literature,26,27 we hypothesized that relational memory impairments would be more severe in schizophrenia than in affective psychosis.

Methods

All study procedures were approved by the Institutional Review Board at Vanderbilt University.

Participants

We studied 507 individuals between April 2009 and August 2024. Individuals with affective psychosis (n = 77) and schizophrenia spectrum disorders (n = 244) were recruited from the Vanderbilt Psychiatric Hospital inpatient unit and outpatient clinics in the community, whereas healthy control participants (n = 186) were recruited from the community through flyers and advertisements. Diagnoses were determined based on the Structured Clinical Interview for DSM disorders (SCID), with 374 participants assessed under the DSM-IV-TR criteria and 133 under the DSM-5 criteria.28 Exclusionary criteria for all 3 groups included major physical and neurological illnesses, active substance abuse or dependence, and significant head injury. Additionally, healthy control participants were only recruited if they did not endorse any history of Axis I psychiatric disorders. All participants were assessed for premorbid IQ using the Wechsler Test of Adult Reading (WTAR).29 Individuals with psychosis were further assessed for the severity of psychotic symptoms using the Positive and Negative Syndrome Scale (PANSS).30 See Table 1 for demographic and clinical characteristics.

Table 1.

Demographic and Clinical Characteristics of Participants

Healthy control (n = 186) Affective psychosis (n = 77) Schizophrenia (n = 244) P-values a P-values b P-values c
Age, years 25.92 (9.02) 30.79 (14.05) 26.39 (10.04) <.01 .89 <.01
Age range 17-60 16-63 15-60 – – –
Female % (N) 31.18% (58) 45.45% (35) 29.10% (71) – – –
Race (White/Black/Other) 132/38/16 55/15/7 137/95/12 – – –
IQ, WTAR 113.96 (9.53) 104.74 (14.82) 100.78 (16.01) <.001 <.001 .08
Education, years 15.51 (2.11) 14.06 (2.03) 13.48 (2.21) <.001 <.001 .09
Parental education, years 14.87 (2.22) 14.82 (2.37) 14.50 (2.83) .99 .30 .61
PANSS—total – 52.58 (18.96) 68.53 (18.36) – – <.001
PANSS—positive – 13.82 (8.74) 18.73 (7.43) – – <.001
PANSS—negative – 11.71 (5.04) 16.61 (7.30) – – <.001
PANSS—general – 27.05 (8.96) 33.19 (8.78) – – <.001
Open in a new tab

Abbreviations: WTAR = Wechsler Test of Adult Reading; PANSS = Positive and Negative Syndrome Scale. aValues reflect differences between healthy control participants and the affective psychosis group. bValues reflect differences between controls and the schizophrenia group. cValues reflect differences between the two clinical groups. Bolded values denote significance.

The schizophrenia spectrum disorders group included individuals diagnosed with schizophreniform disorder (n = 127), schizophrenia (n = 88), and schizoaffective disorder (n = 29), whereas the affective psychosis group included individuals diagnosed with bipolar disorder with psychotic features (n = 74) and major depressive disorder with psychotic features (n = 3). Ninety-three (73.23%) participants with schizophreniform disorder were followed for 2 years during which time 57 (61.29%) converted to schizophrenia (n = 47) or schizoaffective disorder (n = 10). Of the 507 participants included in the current study, 85 participants (59 schizophrenia and 26 healthy control) were previously reported in Armstrong and colleagues,20 135 additional participants (79 schizophrenia and 56 healthy control participants) were reported in Armstrong and colleagues,19 and 39 additional participants (12 schizophrenia, 4 affective psychosis, and 23 healthy control) were reported in Avery and colleagues.18 The current study includes data from 248 participants not previously reported.

Materials

Participants completed the AIP17 as a measure of relational memory (Figure 1). Stimuli consisted of 120 colored images of faces (60 male, 60 female) and 30 colored images of houses. These images were drawn from online databases and used to create three distinct sets of paired associates that were first encountered during the training phase. The first two sets (i.e., A-B, B-C) consisted of face-house pairs, while the third one (i.e., D-E) consisted of face–face pairs. The A-B set included 30 faces (Stimuli A), which were each paired with a unique house (Stimuli B). These houses (Stimuli B) were then paired with a new set of 30 faces (Stimuli C) during the B-C set. This design facilitated an association between each of the house stimuli and two faces. Finally, the D-E set included 30 novel faces (Stimuli D) that were paired with 30 novel faces (Stimuli E).

Figure 1.

Figure 1

Open in a new tab

Visual depiction of the Associative Inference Paradigm. The paradigm consisted of two phases: training and testing. During the training phase, participants completed four study-test cycles for A-B and B-C pairs, as well as a one study-test cycle for D-E pairs. During the testing phase, they viewed directly studied (i.e., A-B, B-C, D-E) or indirectly learned (i.e., A-C) pairs

Procedure

Experimental data were collected on a lab computer using E-Prime software.31 The experiment consisted of a training and testing phase. During the training phase, participants completed four study-test cycles for both A-B and B-C sets. During the study phase of each study-test cycle, participants were instructed to memorize the associations between pairs of items and were presented with each of the pairs for 4 seconds. Immediately following each of the four study phases, participants completed a two-alternative forced-choice (2AFC) recognition test with no feedback. The test required participants to choose between two stimuli of the same category (e.g., between two houses for A-B pairs, or between two faces for B-C and D-E pairs) to identify the target item that had been previously paired with the cue. For all trials, foil items were drawn from the same stimulus set but had been paired with a different cue and were therefore equally familiar. Participants first completed four study-test cycles for the A-B sets, followed by four study-test cycles for the B-C sets. Finally, participants completed an identically formatted study-test cycle of the D-E set. Notably, participants were presented with the D-E study-test cycle only once to obtain a measure of paired-associate performance that is not affected by training effects.

Next, participants completed the testing phase consisting of 120 trials. Specifically, they were presented with 90 trials of previously studied pairs (30 from each set) and 30 trials of inferential A-C pairs. These A-C pairs consisted of two faces that were previously associated with the same house stimuli (Stimuli B) and therefore linked through their association with the same house. Since participants had not previously encountered these A-C pairs during the training phase, successful performance for these pairs relied on participants’ ability to integrate the learned face-house associations to infer the association between the two faces. Participants were presented with six blocks of each pair type, with each block consisting of five trials. These blocks were presented in a pseudo-randomized order, such that a given set of A-C pairs always appeared before the corresponding sets of A-B and B-C pairs. For each pair presented, participants engaged in a 2AFC recognition test.

Data Analysis

All analyses were conducted using R version 4.3.1.32 For analyses involving nested data, we used mixed-effects models via the lme4 package that included participants as a random effect.33 This approach is robust to unequal sample sizes and violations of the assumption of homogeneity of variances.34 Hypothesis tests were performed using Type II Wald chi-square tests via the Anova function from the car package.35 Estimated marginal means and pairwise comparisons were calculated using the emmeans package,36 with Tukey adjustments for multiple comparisons. To examine learning trajectories, we fit logistic growth models using the nlme package,37 which allowed us to estimate two key parameters (growth rate and asymptote) for each diagnostic group. The significance level was set at α = .05.

Whereas prior studies included participants achieving at least 80% accuracy in the final training block,18–21 we included all participants irrespective of their training performance, as one of our primary focuses was on learning trajectories. The analysis scripts are available on the Open Science Framework: https://osf.io/s3h2e/.

Results

Training Accuracy Patterns

A-B Training Accuracy Patterns

We first examined how accurately each group learned A-B associations across training blocks (Figure 2). All three groups improved over time (main effect of Block, χ2(3) = 1978.33, p < .001). The largest improvements occurred between Blocks 1 and 2, followed by smaller improvements between Blocks 2 and 3, with no significant improvement between Blocks 3 and 4, suggesting that all groups reached a learning plateau by Block 3. Healthy control participants exhibited the highest accuracy, followed by individuals with affective psychosis, while individuals with schizophrenia demonstrated the lowest accuracy (main effect of Diagnosis Type, χ2(2) = 122.41, p < .001). While all groups reached a plateau by Block 3, individuals with schizophrenia plateaued at a significantly lower level of accuracy compared to the other groups (Block × Diagnosis Type interaction, χ2(6) = 27.98, p < .001).

Figure 2.

Figure 2

Open in a new tab

Accuracy throughout the training phase. Each point (dot) represents the mean accuracy for a specific group, block, and pair type. Errors bars around each point represent 95% confidence intervals. Solid lines represent A-B pairs, while dashed lines represent B-C pairs

We characterized learning trajectories by using a growth curve analysis modeling accuracy across training for each diagnostic group (see Figure S1). We estimated the highest level of accuracy (i.e., asymptote) and the rate of learning (i.e., k) for each group. Healthy control participants showed the highest asymptotic accuracy (Asymptote = 0.99) and fastest learning rate (k = 1.83), followed by individuals with affective psychosis (Asymptote = 0.96; k = 1.02). Individuals with schizophrenia showed the lowest asymptotic accuracy (Asymptote = 0.90) and a slower learning rate (k = 1.16) compared to healthy control participants, indicating both slower acquisition and a lower modeled performance ceiling.

B-C Training Accuracy Patterns

All groups improved across training blocks for B-C associations (main effect of Block, χ2(3) = 1412.10, p < .001) (Figure 2). As with A-B associations, the largest improvements occurred between Blocks 1 and 2, followed by smaller improvements between Blocks 2 and 3, and minimal improvement between Blocks 3 and 4. Healthy control participants and individuals with affective psychosis approached a learning plateau by Block 3, while individuals with schizophrenia continued to show small but significant improvements into Block 4. Healthy control participants achieved higher accuracy than individuals with affective psychosis, who in turn performed better than individuals with schizophrenia (main effect of Diagnosis Type, χ2(2) = 152.84, p < .001). The schizophrenia group once again plateaued at a lower level of accuracy (Block × Diagnosis Type interaction, χ2(6) = 57.28, p < .001).

Growth curve analyses further elucidated these patterns by showing that healthy control participants exhibited the highest asymptotic accuracy (Asymptote = 0.99) and fastest learning rate (k = 1.80), followed by individuals with affective psychosis (Asymptote = 0.95; k = 1.02) (Figure 3). Individuals with schizophrenia showed the lowest asymptotic accuracy (Asymptote = 0.91) and the slowest rate of learning (k = 0.75), indicating both slower acquisition and a lower modeled performance ceiling.

Figure 3.

Figure 3

Open in a new tab

Rate of forgetting between training and testing blocks. Error bars represent standard errors. Forgetting scores were calculated as (A) the difference between the final training block and the subsequent testing block and (B) the proportion of previously learned pairs (i.e., those correctly recalled in the final training block) that were not retained at test

Rate of Forgetting

Next, we examined whether forgetting rates differed across diagnostic groups and pair types by comparing performance between the final training block and the subsequent testing block. Given that D-E pairs were learned immediately before testing, our analyses focused on A-B and B-C pairs. Participants showed greater forgetting for A-B than B-C pairs (main effect of Pair Type, χ2(1) = 100.80, p < .001), consistent with the fact that A-B pairs were learned first. Forgetting varied by diagnostic group, with individuals with schizophrenia and affective psychosis showing greater forgetting than healthy control participants (main effect of Diagnosis, χ2(2) = 75.20, p < .001) (Figure 3). The magnitude of this difference depended on the pair type (Diagnosis × Pair Type interaction, χ2(2) = 15.66, p < .001). For A-B pairs, both clinical groups showed greater forgetting than healthy control participants (smallest t(995) = 5.67, p < .001), but did not differ from each other (t(995) = 0.66, p = .79). For B-C pairs, individuals with schizophrenia showed more forgetting than healthy control participants (t(995) = 3.92, p < .001), but neither group differed from the affective psychosis group (largest t(995) = 1.83, p = .16).

To address the possibility that group differences in forgetting were driven by weaker initial learning in the psychosis groups, we conducted a secondary analysis focusing exclusively on pairs correctly recalled during the final training block. Consistent with our primary analysis, all groups exhibited more forgetting for A-B than B-C pairs (main effect of Pair Type, χ2(1) = 80.45, p < .001), and forgetting varied by diagnostic group (main effect of Diagnosis, χ2(2) = 142.30, p < .001). For A-B pairs, individuals with schizophrenia forgot more than those with affective psychosis (t(714) = 3.53, p < .01), who in turn forgot more than healthy control participants (t(714) = 5.12, p < .001). A comparable pattern emerged for B-C pairs, where individuals with schizophrenia again showed the greatest forgetting, surpassing the affective psychosis group (t(714) = 4.08, p < .001), who also forgot more than healthy control participants (t(714) = 3.05, p < .01).

Testing Accuracy Patterns

Finally, we tested the hypothesis that all three groups would show an inferential memory effect, such that they would have better accuracy for inferential A-C pairs (presented four times during training) than non-inferential D-E pairs (presented one time during training). Consistent with our prior reports,18–20 accuracy was significantly higher for A-C than for D-E pairs for all groups (main effect of Pair Type, χ2(1) = 164.32, p < .001) (Figure 4). Accuracy varied by diagnosis (main effect of Diagnosis Type, χ2(2) = 188.13, p < .001), and the effect of diagnosis on accuracy differed by pair type (Pair Type × Diagnosis Type interaction, χ2(2) = 29.20, p < .001). For A-C pairs, individuals with schizophrenia performed worse than those with affective psychosis (t(803) = 4.74, p <.001), who in turn performed worse than healthy control participants (t(803) = 5.78, p < .001). For D-E pairs, while both clinical groups performed worse than healthy control participants (smallest t(803) = 5.32, p < .001), they did not differ from each other (t(803) = 1.44, p = .32).

Figure 4.

Figure 4

Open in a new tab

Accuracy for inferential (AC) and non-inferential (DE) Pairs during testing. The heights of the white diamonds represent the group means for each diagnostic group and pair type. Error bars extending from the diamonds indicate 95% confidence intervals around the group means. The midlines within boxplots represent the median, while the edges of the boxes indicate the interquartile range (IQR) for each diagnostic group. Each dot represents an individual participant’s data point

To examine whether associative inference deficits extend beyond deficits in learning directly studied associations, we calculated an alternative measure of A-C accuracy (ACalt). This measure was calculated by only including A-C trials where participants had correctly recalled the corresponding A-B and B-C pairs during the final training block. We observed that ACalt differed by diagnosis (main effect of Diagnosis Type, χ2(2) = 74.42, p < .001), such that individuals with schizophrenia performed worse than those with affective psychosis (t(504) = 4.13, p < .001), who in turn performed worse than healthy control participants (t(504) = 4.78, p < .001) (Figure 4). Together, these results indicate that individuals with schizophrenia show disproportionate impairments in making inferences even when controlled for learning the underlying direct associations.

Associations with Symptom Severity

We also examined whether psychotic symptom severity related to performance across pair types in individuals with schizophrenia and affective psychosis. Pearson correlations indicated that greater negative symptom severity was significantly associated with lower accuracy for A-B, B-C, D-E, and A-C pairs (ps < .05). In addition, higher positive symptoms were associated with lower accuracy for B-C and A-C pairs (ps < .05). To address the possibility that symptom severity may account for group differences between individuals with schizophrenia and affective psychosis, we conducted follow-up analyses that included positive and negative symptoms as covariates. These analyses showed that the pattern of group differences remained unchanged after adjusting for symptom severity.

Discussion

The current study used the well-established AIP to further characterize relational memory impairments in psychosis. Much of our prior work with the AIP in psychosis had focused on the ability to make inferences between indirectly linked items. Here, we shifted our focus to the learning of directly studied associations and whether deficits in associative inferences (a special form of relational memory) persist even after accounting for impairments in paired associate learning (a simpler form of relational memory).24

Consistent with prior research in individuals with first episode18,19 and chronic schizophrenia,20,21 individuals with psychosis showed impaired relational memory compared to healthy control participants. Deficits in learning both A-B and B-C associations were evident in both individuals with schizophrenia and affective psychosis. These deficits were present from the first training block and persisted across all training blocks, suggesting that relational memory impairments in schizophrenia involve persistent learning difficulties rather than deficits restricted to initial encoding. This pattern aligns with the findings from the episodic memory literature showing that individuals with schizophrenia are impaired even on their first learning trial,38 indicating that memory dysfunction emerges early and persists despite repeated learning trials.

Notably, individuals with schizophrenia showed more severe impairments in both relational memory and associative inference compared to those with affective psychosis, suggesting that while such deficits may be transdiagnostic, they are more pronounced in schizophrenia. This pattern is consistent with prior studies that found relational memory26,27 and more broadly, episodic memory39 to be more impaired in schizophrenia when compared to individuals with bipolar disorder with psychotic features. Since individuals with schizophrenia also endorsed significantly more psychotic symptoms than the affective psychosis group, we examined whether diagnostic differences in symptom severity were related to differences in memory accuracy. Consistent with meta-analyses linking memory impairments to negative symptoms in psychosis,40,41 higher severity of negative symptoms was correlated with lower accuracy across all pair types. In comparison, higher positive symptoms were correlated with lower accuracy for only B-C and A-C pairs. These findings are consistent with dimensional frameworks (e.g., HiTOP) positing that cognitive deficits exist on a continuum across psychotic disorders but are most severe in schizophrenia.42 Furthermore, follow-up analyses showed that differences between individuals with schizophrenia and affective psychosis persisted even when accounting for symptom severity.

We further quantified group differences by modeling growth curves, which allowed us to estimate how accuracy changed across training for each group. We confirmed our hypothesis that individuals with psychosis demonstrate a slower rate of learning. We also observed that the psychosis groups reached lower asymptotes than healthy control participants, indicating that even with repeated training, their accuracy plateaued at a lower level. Together, these findings suggest that schizophrenia-related relational memory impairments involve performance differences that are evident from the beginning of training and persist despite repeated training.

Memory depends not only on effective encoding but also on the persistence of encoded information over time.43 Previous research using recall tasks has consistently shown that individuals with schizophrenia exhibit greater forgetting than healthy control participants,44 likely contributing to the pronounced episodic memory dysfunction observed in this population. We found that individuals with psychosis, particularly those with schizophrenia, forgot significantly more than healthy control participants. Notably, while both clinical groups showed comparable forgetting for A-B pairs, individuals with schizophrenia forgot significantly more than those with affective psychosis on B-C pairs, suggesting that memory decay may be more pronounced in schizophrenia. It is also possible that proactive interference contributed to this pattern,45 such that earlier A-B learning interfered with the encoding and retention of later B-C associations. However, a body of research suggests that individuals with schizophrenia are not more susceptible to proactive interference.46–48 Nevertheless, we were not positioned to test this possibility directly due to the fixed order of pair learning.

While our findings support the notion of greater forgetting in psychosis, there remains a debate as to whether such storage deficits may be explained by initial poor performance.44,49 For example, Gold and colleagues50 found that when individuals with schizophrenia and healthy control participants are matched on initial recall, they displayed comparable forgetting scores, suggesting that baseline encoding differences could artificially inflate the extent of forgetting seen in individuals with schizophrenia. To address this possibility, we computed an alternative forgetting index, which focused only on pairs successfully learned by the final training block. This more stringent measure of forgetting, which accounts for baseline performance, revealed that individuals with psychosis continue to forget more than healthy control participants. The affective psychosis group showed intermediate forgetting, suggesting that while elevated forgetting may be common across psychiatric disorders, its severity may differ by diagnosis. Nevertheless, it remains a possibility that the rates of forgetting observed here could be explained by retroactive interference,47,51 wherein newly learned associations disrupt the retention of earlier ones.45 Thus, further research is needed to clarify the relative contributions of interference.

Finally, we examined associative inference performance. Prior research using the AIP has consistently shown that individuals with schizophrenia show disproportionately greater impairments on inferential relative to non-inferential pairs,18–21 a pattern that has been interpreted to reflect a specific associative inference deficit. However, interpretation of this pattern can be complicated by overall group differences in task performance. Since accuracy is not linearly related to memory strength, baseline differences between groups can make effects appear larger or smaller than they are, even when the underlying memory changes are comparable.52 To address this issue, we used two complementary approaches: First, we replicated prior studies by comparing performance on inferential and non-inferential pairs. Second, we calculated an alternative measure of associative inference that included only trials where participants had successfully recalled the corresponding A-B and B-C pairs during the final training block, following the approach used by Spalding and colleagues.53 Both approaches revealed that associative inference deficits are significantly more pronounced in schizophrenia. Critically, the alternative measure of associative inference confirmed that, even when controlling for paired-associate learning, individuals with schizophrenia continued to show significant impairments in making inferences.

This approach of isolating associative inference from paired associate learning deficits opens an avenue for identifying distinct neural mechanisms underlying relational memory impairments in schizophrenia. Prior research suggests that the ability to form and retrieve direct associations between items is heavily dependent on the hippocampus,54,55 whereas associative inference is thought to rely on a broader network including the ventromedial prefrontal cortex (vmPFC). Evidence from lesion studies suggests that damage to the vmPFC selectively disrupts associative inference while sparing memory for directly studied associations,53 indicating a potential dissociation between these two forms of memory. Neuroimaging studies support this notion by showing that hippocampal-vmPFC interactions predict successful associative inference, even when controlling memory for directly studied associations.56 While prior research has linked relational memory deficits in schizophrenia to dysfunction in these regions,57,58 it remains unclear whether the two types of relational memory impairments observed here stem from distinct neural mechanisms. Given established hippocampal and prefrontal volume reductions in schizophrenia,59 future research should examine whether structural abnormalities in these regions differentially predict performance on these two types of memory.

There are several caveats to consider when interpreting the current study. Although our analyses did not evaluate the impact of affective symptoms, prior work suggests that depressive and manic mood states can influence cognition,60 raising the possibility that some of the impairments observed here may have been impacted by mood at the time of participation. We also did not examine how other cognitive domains, such as working memory and processing speed, may have interacted with the pattern of results reported here. While some studies report that memory deficits persist even after accounting for domain-general impairments,61 we cannot rule out the possibility that broader cognitive dysfunction may have contributed to some of the effects observed here. Finally, our design did not allow us to disentangle potential effects of proactive and retroactive interference, as all participants studied A-B and B-C pairs in a fixed order.

In conclusion, our results provide a comprehensive characterization of relational memory impairments in psychosis. Individuals with psychosis demonstrated altered acquisition and increased forgetting of directly studied associations, impairments that were evident from the earliest stages of training and persisted across repeated exposure. Notably, these disruptions extended to associative inference, where individuals with psychosis showed disproportionate difficulty inferring novel relationships, even when analyses were limited to trials in which the underlying associations were successfully recalled. This pattern suggests a specific deficit in integrating related experiences to support flexible memory-based reasoning. Given the central role of relational memory in navigating novel situations and guiding future behavior, such deficits likely contribute to broader difficulties in adaptive functioning. Future research should therefore explore whether interventions targeting relational memory can help mitigate these cognitive challenges and improve functional outcomes in individuals with psychosis.

Supplementary Material

AI_Supplement_SB_Final_sbaf188
ai_supplement_sb_final_sbaf188.docx (106.8KB, docx)

Contributor Information

Aslıhan İmamoğlu, Department of Psychiatry and Behavioral Sciences, Vanderbilt University Medical Center, Nashville, TN 37232, United States.

Suzanne Avery, Department of Psychiatry and Behavioral Sciences, Vanderbilt University Medical Center, Nashville, TN 37232, United States.

Kristan Armstrong, Department of Psychiatry and Behavioral Sciences, Vanderbilt University Medical Center, Nashville, TN 37232, United States.

Stephan Heckers, Department of Psychiatry and Behavioral Sciences, Vanderbilt University Medical Center, Nashville, TN 37232, United States.

Author Contributions

A.İ.: Conceptualization, Data Curation, Methodology, Formal analysis, Writing—original draft, Writing—review & editing, Visualization. S.A.: Conceptualization, Methodology, Data Curation, Software, Writing—review & editing. K.A.: Conceptualization, Methodology, Data Curation, Software, Project Administration, Writing—review & editing. S.H.: Conceptualization, Methodology, Resources, Writing—Review & Editing, Supervision, Project Administration, Funding Acquisition

Funding

This project was supported by the National Institute of Mental Health NIMH (MH070560 to SH).

References

  • 1. Leavitt  VM, Goldberg  TE. Episodic memory in schizophrenia. Neuropsychol Rev. 2009;19:312–323. 10.1007/s11065-009-9107-0 [DOI] [PubMed] [Google Scholar]
  • 2. Fett  AK, Viechtbauer  W, Dominguez  MD, Penn  DL, van Os  J, Krabbendam  L. The relationship between neurocognition and social cognition with functional outcomes in schizophrenia: a meta-analysis. Neurosci Biobehav Rev. 2011;35:573–588. 10.1016/j.neubiorev.2010.07.001 [DOI] [PubMed] [Google Scholar]
  • 3. Halverson  TF, Orleans-Pobee  M, Merritt  C, Sheeran  P, Fett  AK, Penn  DL. Pathways to functional outcomes in schizophrenia spectrum disorders: Meta-analysis of social cognitive and neurocognitive predictors. Neurosci Biobehav Rev. 2019; 105:212–219. 10.1016/j.neubiorev.2019.07.020 [DOI] [PubMed] [Google Scholar]
  • 4. Achim  AM, Lepage  M. Is associative recognition more impaired than item recognition memory in schizophrenia?  A meta-analysis Brain Cogn. 2003;53:121–124. 10.1016/s0278-2626(03)00092-7 [DOI] [PubMed] [Google Scholar]
  • 5. Lepage  M, Hawco  C, Bodnar  M. Relational memory as a possible neurocognitive marker of schizophrenia. JAMA Psychiatry. 2015;72:946–947. 10.1001/jamapsychiatry.2015.0488 [DOI] [PubMed] [Google Scholar]
  • 6. Zeithamova  D, Schlichting  ML, Preston  AR. The hippocampus and inferential reasoning: building memories to navigate future decisions. Front Hum Neurosci. 2012;6:70. 10.3389/fnhum.2012.00070 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7. Elvevåg  B, Egan  MF, Goldberg  TE. Paired-associate learning and memory interference in schizophrenia. Neuropsychologia.  2000;38:1565–1575. 10.1016/s0028-3932(00)00074-9 [DOI] [PubMed] [Google Scholar]
  • 8. Lepage  M, Montoya  A, Pelletier  M, Achim  AM, Menear  M, Lal  S. Associative memory encoding and recognition in schizophrenia: an event-related fMRI study. Biol Psychiatry. 2006;60:1215–1223. 10.1016/j.biopsych.2006.03.043 [DOI] [PubMed] [Google Scholar]
  • 9. Luck  D, Montoya  A, Menear  M, Achim  AM, Lal  S, Lepage  M. Selective pair recognition memory impairment with no response bias in schizophrenia. Psychiatry Res. 2009;169:39–42. 10.1016/j.psychres.2008.06.037 [DOI] [PubMed] [Google Scholar]
  • 10. Ragland  JD, Censits  DM, Gur  RC, Glahn  DC, Gallacher  F, Gur  RE. Assessing declarative memory in schizophrenia using Wisconsin card sorting test stimuli: the paired associate recognition test. Psychiatry Res. 1996;60:135–145. 10.1016/0165-1781(96)02811-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Rushe  TM, Woodruff  PW, Murray  RM, Morris  RG. Episodic memory and learning in patients with chronic schizophrenia. Schizophr Res. 1999;35:85–96. 10.1016/s0920-9964(98)00117-0 [DOI] [PubMed] [Google Scholar]
  • 12. Hutton  SB, Puri  BK, Duncan  LJ, Robbins  TW, Barnes  TR, Joyce  EM. Executive function in first-episode schizophrenia. Psychol Med. 1998;28:463–473. 10.1017/s0033291797006041 [DOI] [PubMed] [Google Scholar]
  • 13. Goldberg  TE, Torrey  EF, Gold  JM, et al.  Genetic risk of neuropsychological impairment in schizophrenia: a study of monozygotic twins discordant and concordant for the disorder. Schizophr Res. 1995;17:77–84. 10.1016/0920-9964(95)00032-h [DOI] [PubMed] [Google Scholar]
  • 14. Laurent  A, Moreaud  O, Bosson  JL, et al.  Neuropsychological functioning among non-psychotic siblings and parents of schizophrenic patients. Psychiatry Res. 1999;87:147–157. 10.1016/s0165-1781(99)00064-5 [DOI] [PubMed] [Google Scholar]
  • 15. Toulopoulou  T, Rabe-Hesketh  S, King  H, Murray  RM, Morris  RG. Episodic memory in schizophrenic patients and their relatives. Schizophr Res. 2003;63:261–271. 10.1016/s0920-9964(02)00324-9 [DOI] [PubMed] [Google Scholar]
  • 16. Ragland  JD, Cools  R, Frank  M, et al.  CNTRICS final task selection: long-term memory. Schizophr Bull. 2009;35:197–212. 10.1093/schbul/sbn134 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Preston  AR, Shrager  Y, Dudukovic  NM, Gabrieli  JD. Hippocampal contribution to the novel use of relational information in declarative memory. Hippocampus.  2004;14:148–152. 10.1002/hipo.20009 [DOI] [PubMed] [Google Scholar]
  • 18. Avery  SN, Armstrong  K, McHugo  M, et al.  Relational memory in the early stage of psychosis: a 2-year follow-up study. Schizophr Bull. 2021;47:75–86. 10.1093/schbul/sbaa081 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Armstrong  K, Avery  S, Blackford  JU, Woodward  N, Heckers  S. Impaired associative inference in the early stage of psychosis. Schizophr Res. 2018;202:86–90. 10.1016/j.schres.2018.06.049 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Armstrong  K, Kose  S, Williams  L, Woolard  A, Heckers  S. Impaired associative inference in patients with schizophrenia. Schizophr Bull. 2012;38:622–629. 10.1093/schbul/sbq145 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Armstrong  K, Williams  LE, Heckers  S. Revised associative inference paradigm confirms relational memory impairment in schizophrenia. Neuropsychology.  2012;26:451–458. 10.1037/a0028667 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Schlichting  ML, Preston  AR. Memory integration: neural mechanisms and implications for behavior. Curr Opin Behav Sci. 2015;1:1–8. 10.1016/j.cobeha.2014.07.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Zeithamova  D, Preston  AR. Flexible memories: differential roles for medial temporal lobe and prefrontal cortex in cross-episode binding. J Neurosci. 2010;30:14676–14684. 10.1523/JNEUROSCI.3250-10.2010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24. Raucher-Chéné  D, Lavigne  KM, Lepage  M. Episodic memory and schizophrenia: from characterization of relational memory impairments to neuroimaging biomarkers. Curr Top Behav Neurosci. 2023;63:115–136. 10.1007/7854_2022_379 [DOI] [PubMed] [Google Scholar]
  • 25. Kotov  R, Krueger  RF, Watson  D, et al.  The hierarchical taxonomy of psychopathology (HiTOP): a dimensional alternative to traditional nosologies. J Abnorm Psychol. 2017;126:454–477. 10.1037/abn0000258 [DOI] [PubMed] [Google Scholar]
  • 26. McKinney  RA, Avery  SN, Armstrong  K, Blackford  JU, Woodward  ND, Heckers  S. Relational memory in the early stage of psychotic bipolar disorder. Psychiatry Res. 2020;294:113508. 10.1016/j.psychres.2020.113508 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27. Sheffield  JM, Williams  LE, Cohen  N, Heckers  S. Relational memory in psychotic bipolar disorder. Bipolar Disord. 2012; 14:537–546. 10.1111/j.1399-5618.2012.01036.x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28. First  MB, Spitzer  RL, Gibbon  M, Williams  JB. The structured clinical interview for DSM-III-R personality disorders (SCID-II). Part I: Description J Pers Disord. 1995;9:83–91. [Google Scholar]
  • 29. Wechsler  D. Wechsler Test of Adult Reading: WTAR. Psychological Corporation, 2001. [Google Scholar]
  • 30. Kay  SR, Fiszbein  A, Opler  LA. The positive and negative syndrome scale (PANSS) for schizophrenia. Schizophr Bull. 1987;13:261–276. 10.1093/schbul/13.2.261 [DOI] [PubMed] [Google Scholar]
  • 31. Psychology Software Tools Inc . E-Prime 2.0. [Computer Program]. Psychology Software Tools, 2016. [Google Scholar]
  • 32. R Core Team . R: A Language and Environment for Statistical Computing. [Computer Program]. R Foundation for Statistical Computing, 2023. [Google Scholar]
  • 33. Bates  D, Mächler  M, Bolker  B, Walker  S. Fitting linear mixed-effects models using lme4. J Stat Softw. 2015;67:1–48. [Google Scholar]
  • 34. Schielzeth  H, Dingemanse  NJ, Nakagawa  S, et al.  Robustness of linear mixed-effects models to violations of distributional assumptions. Methods Ecol Evol. 2020;11:1141–1152. 10.1111/2041-210X.13434 [DOI] [Google Scholar]
  • 35. Fox  J, Weisberg  S. An R Companion to Applied Regression 3rd edn. Sage, 2018. [Google Scholar]
  • 36. Lenth  RV, Singmann  H, Love  J, et al.  emmeans: Estimated marginal means. R package version 1.4.4. 2020. https://rvlenth.github.io/emmeans/ Accessed November 11, 2024.
  • 37. Pinheiro  J, Bates  D, DebRoy  S, Sarkar  D, R Core Team. nlme: Nonlinear Mixed-Effects Models. R package version 3.1-131; 2013. https://svn.r-project.org/R-packages/trunk/nlme/ Accessed November 11, 2024.
  • 38. Paulsen  JS, Heaton  RK, Sadek  JR, et al.  The nature of learning and memory impairments in schizophrenia. J Int Neuropsychol Soc. 1995;1:88–99. 10.1017/s135561770000014x [DOI] [PubMed] [Google Scholar]
  • 39. Hill  SK, Harris  MS, Herbener  ES, Pavuluri  M, Sweeney  JA. Neurocognitive allied phenotypes for schizophrenia and bipolar disorder. Schizophr Bull. 2008;34:743–759. 10.1093/schbul/sbn027 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40. Aleman  A, Hijman  R, de Haan  EHF, Kahn  RS. Memory impairment in schizophrenia: a meta-analysis. Am J Psychiatry. 1999;156:1358–1366. 10.1176/ajp.156.9.1358 [DOI] [PubMed] [Google Scholar]
  • 41. Pillny  M, Krkovic  K, Buck  L, Lincoln  TM. From memories of past experiences to present motivation? A meta-analysis on the association between episodic memory and negative symptoms in people with psychosis. Schizophr Bull. 2022;48:307–324. 10.1093/schbul/sbab120 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42. Ringwald  WR, Abramovitch  A, Agelink van Rentergem  JA, Kotov  R. Do cognitive functions belong in the hierarchical taxonomy of psychopathology model? A meta-analysis. Perspect Psychol Sci. 2025. 10.1177/17456916251347926 [DOI] [PubMed] [Google Scholar]
  • 43. Squire  LR, Genzel  L, Wixted  JT, Morris  RG. Memory Consolidation. Cold Spring Harbor (NY): Cold Spring Harbor Laboratory Press. 2015;7:a021766. 10.1101/cshperspect.a021766 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44. Cirillo  MA, Seidman  LJ. Verbal declarative memory dysfunction in schizophrenia: from clinical assessment to genetics and brain mechanisms. Neuropsychol Rev. 2003;13:43–77. 10.1023/a:1023870821631 [DOI] [PubMed] [Google Scholar]
  • 45. Wixted  JT. The psychology and neuroscience of forgetting. Annu Rev Psychol. 2004;55:235–269. 10.1146/annurev.psych.55.090902.141555 [DOI] [PubMed] [Google Scholar]
  • 46. Kaller  CP, Loosli  SV, Rahm  B, et al.  Working memory in schizophrenia: behavioral and neural evidence for reduced susceptibility to item-specific proactive interference. Biol Psychiatry. 2014;76:486–494. 10.1016/j.biopsych.2014.03.012 [DOI] [PubMed] [Google Scholar]
  • 47. O'Carroll  RE, Murray  C, Austin  MP, Ebmeier  KP, Goodwin  GM, Dunan  J. Proactive interference and the neuropsychology of schizophrenia. Br J Clin Psychol. 1993;32:353–356. 10.1111/j.2044-8260.1993.tb01067.x [DOI] [PubMed] [Google Scholar]
  • 48. Torres  IJ, Flashman  LA, O'Leary  DS, Andreasen  NC. Effects of retroactive and proactive interference on word list recall in schizophrenia. J Int Neuropsychol Soc. 2001;7:481–490. 10.1017/s1355617701744049 [DOI] [PubMed] [Google Scholar]
  • 49. Danion  JM, Huron  C, Vidailhet  P, Berna  F. Functional mechanisms of episodic memory impairment in schizophrenia. Can J Psychiatr. 2007;52:693–701. 10.1177/070674370705201103 [DOI] [PubMed] [Google Scholar]
  • 50. Gold  JM, Rehkemper  G, Binks  SW  3rd, et al.  Learning and forgetting in schizophrenia. J Abnorm Psychol. 2000;109:534–538. 10.1037/0021-843X.109.3.534 [DOI] [PubMed] [Google Scholar]
  • 51. Ayesa-Arriola  R, Tordesillas-Gutierrez  D, Setién-Suero  E, et al.  Verbal memory and voxel based morphometry in first episode non-affective psychosis: a process oriented approach. Neuropsychology.  2019;33:568–580. 10.1037/neu0000540 [DOI] [PubMed] [Google Scholar]
  • 52. Luck  SJ, Gold  JM. The translation of cognitive paradigms for patient research. Schizophr Bull. 2008;34:629–644. 10.1093/schbul/sbn036 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53. Spalding  KN, Schlichting  ML, Zeithamova  D, et al.  Ventromedial prefrontal cortex is necessary for normal associative inference and memory integration. J Neurosci. 2018;38:3767–3775. 10.1523/JNEUROSCI.2501-17.2018 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54. Konkel  A, Warren  DE, Duff  MC, Tranel  DN, Cohen  NJ. Hippocampal amnesia impairs all manner of relational memory. Front Hum Neurosci. 2008;2:15.Published 2008 Oct 29. 10.3389/neuro.09.015.2008 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55. Konkel  A, Cohen  NJ. Relational memory and the hippocampus: representations and methods. Front Neurosci. 2009;3:166–174. 10.3389/neuro.01.023.2009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56. Zeithamova  D, Dominick  AL, Preston  AR. Hippocampal and ventral medial prefrontal activation during retrieval-mediated learning supports novel inference. Neuron.  2012;75:168–179. 10.1016/j.neuron.2012.05.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57. Öngür  D, Cullen  TJ, Wolf  DH, et al.  The neural basis of relational memory deficits in schizophrenia. Arch Gen Psychiatry. 2006;63:356–365. 10.1001/archpsyc.63.4.356 [DOI] [PubMed] [Google Scholar]
  • 58. Ragland  JD, Ranganath  C, Harms  MP, et al.  Functional and neuroanatomic specificity of episodic memory dysfunction in schizophrenia: a functional magnetic resonance imaging study of the relational and item-specific encoding task. JAMA Psychiatry. 2015;72:909–916. 10.1001/jamapsychiatry.2015.0276 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59. Shenton  ME, Dickey  CC, Frumin  M, McCarley  RW. A review of MRI findings in schizophrenia. Schizophr Res. 2001;49:1–52. 10.1016/s0920-9964(01)00163-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60. Tavares  JVT, Drevets  WC, Sahakian  BJ. Cognition in mania and depression. Psychol Med. 2003;33:959–967. 10.1017/S0033291703008432 [DOI] [PubMed] [Google Scholar]
  • 61. Sheffield  JM, Gold  JM, Strauss  ME, et al.  Common and specific cognitive deficits in schizophrenia: relationships to function. Cogn Affect Behav Neurosci. 2014;14:161–174. 10.3758/s13415-013-0211-5 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

AI_Supplement_SB_Final_sbaf188
ai_supplement_sb_final_sbaf188.docx (106.8KB, docx)

Articles from Schizophrenia Bulletin are provided here courtesy of Oxford University Press

RESOURCES