Recurrent involuntary autobiographical memories (IAMs), unintentional and repetitive recollections of personal past events, are a common experience, yet their role in mental health remains debated. Some researchers propose that recurrent IAMs contribute to various clinical disorders, while others view them as benign occurrences in everyday life. This discrepancy highlights the need to understand how the content of recurrent IAMs might differentiate between maladaptive (linked to poorer mental health) and benign memories. Previous research primarily focused on self-reported valence (emotional tone) of memories, neglecting the actual content. While negative valence has been associated with elevated mental health symptoms, this approach is limited as valence is intertwined with the content of the memory itself. This study aimed to determine whether content analysis, specifically using computational methods to identify recurring themes (topics) within recurrent IAM descriptions, could provide further insights into the relationship between recurrent IAMs and mental health, distinguishing between maladaptive and benign memories. Prior research using manual content analysis of autobiographical memories has yielded inconsistent results, potentially due to small sample sizes, limitations of single-topic categorization methods, and the confounding of valence and content. This study addresses these limitations by using a large nonclinical sample and computational text analysis to overcome challenges associated with manual coding.

Existing literature on autobiographical memory (AM) and mental health reveals mixed findings regarding the role of AM content. Some studies have found associations between specific AM content (e.g., illness, abuse) and increased psychopathology, while others show no significant differences in AM content based on mental health status. These inconsistencies might stem from limitations such as small sample sizes, use of single-membership content models (categorizing memories into only one topic), and the entanglement of valence and content. The existing studies failed to adequately disentangle content and valence, leading to inconclusive results regarding the independent contributions of content and valence to the relationship between AMs and mental health symptoms. For instance, studies found that negative appraisals of IAMs, rather than the event itself, better predicted depression symptoms, which highlights the importance of considering both content and valence in the analysis. This study sought to address these gaps in the literature by using a larger, nonclinical sample and a computational approach to comprehensively analyze both the content and valence of recurrent IAMs in relation to psychopathology.

This study utilized data from a previous research project involving 6187 undergraduate students at the University of Waterloo. Participants completed online surveys between September 2018 and February 2020, including the Recurrent Memory Scale, which assessed the presence and characteristics of recurrent IAMs within the past year. Participants who reported experiencing at least one recurrent IAM were asked to describe their most frequently recurring memory and rate its valence on a 5-point Likert scale (-2 = very negative, 0 = neutral, 2 = very positive). Mental health symptoms were measured using the Depression Anxiety Stress Scales-21 (DASS-21), Posttraumatic Stress Disorder Checklist for DSM-5 (PCL-5), Social Phobia Inventory (SPIN), and State-Trait Inventory of Cognitive and Somatic Anxiety-Trait Version (STICSA-T). Data preprocessing involved supervised machine learning to identify and remove invalid text responses, followed by standard text cleaning procedures (tokenization, stop word removal, lemmatization). Structural topic modeling (STM) was then applied to the cleaned text data to identify coherent topics within the recurrent IAM descriptions. STM is an unsupervised machine learning technique that identifies underlying themes or topics within a corpus of text. In this analysis, the researchers incorporated self-reported valence along with participants’ scores on mental health indices to predict the prevalence of different topics. This allowed for examining the unique contributions of content (topics) and valence to the relationship between recurrent IAMs and mental health symptoms.

The study replicated previous findings showing a significant negative correlation between negative valence in recurrent IAMs and symptoms of depression, PTSD, social anxiety, and general anxiety. Importantly, the analysis revealed that specific topics within the recurrent IAMs were uniquely associated with symptoms of particular disorders, even after controlling for valence. Depression symptoms were significantly predicted by the prevalence of the "Abuse and trauma" topic. PTSD symptoms were associated with higher prevalence of the "Negative past relationships" topic and lower prevalence of topics such as "Embarrassing events," "Conversations," and "Interactions with friends." Social anxiety symptoms were linked to a higher prevalence of the "Communication and miscommunication" and "Reflections on decisions" topics and a lower prevalence of "Negative past relationships" and "Abuse and trauma." Finally, general anxiety symptoms were positively associated with the "Conversations" topic. The study found that many negative topics were not significantly related to symptoms of any disorder, suggesting that the mere negativity of a memory is not sufficient to explain its relationship with psychopathology. The combination of valence and specific content seems to uniquely predict mental health symptoms.

The findings highlight the importance of considering both valence and content when examining the relationship between recurrent IAMs and mental health. While negative valence consistently correlated with increased psychopathology across disorders, the specific content of these memories provided further insights into the nature of this relationship and varied across disorders. The results provide support for models proposing that recurrent IAMs and their associated emotions are involved in the development and maintenance of psychopathology. Specifically, the disorder-specific content suggests that not all negative recurrent memories are equally relevant to mental health difficulties. The findings contribute to the understanding of how specific memory content may reflect the unique cognitive and emotional processes involved in different mental health disorders. Furthermore, the utilization of computational text analysis enabled a comprehensive examination of both valence and content in a large sample, advancing the methodology used in this research area.

This study demonstrates that both valence and content of recurrent IAMs are significantly linked to mental health symptoms. While negative valence is a transdiagnostic factor, the specific content of memories differs across disorders. This underscores the importance of considering both aspects in understanding the role of recurrent IAMs in psychopathology. Future research could extend this work by comparing recurrent IAMs to other thought processes (e.g., future thinking, rumination) and investigating larger clinical samples to further refine our understanding of this complex relationship.

The study used a large nonclinical sample of undergraduates, which may limit the generalizability of findings to other populations. The cross-sectional design prevents conclusions about causality. The focus on the most frequently recurring IAM might not fully capture the diversity of memories experienced by individuals. Future research could address these limitations by including larger clinical samples, longitudinal studies, and the analysis of multiple recurrent IAMs per participant.