Reduced discrimination between signals of danger and safety but not overgeneralization is linked to exposure to childhood adversity in healthy adults

Joint Public Review:

Here, the authors compare how different operationalizations of adverse childhood experience exposure related to patterns of skin conductance response during a fear conditioning task. They use a large dataset to definitively understand a phenomenon that, to date, has been addressed using a range of different definitions and methods, typically with insufficient statistical power. Specifically, the authors compared the following operationalizations: dichotomization of the sample into "exposed" and "non-exposed" categories, cumulative adversity exposure, specificity of adversity exposure, and dimensional (threat versus deprivation) adversity exposure. The paper is thoughtfully framed and provides clear descriptions and rationale for procedures, as well as package version information and code. The authors' overall aim of translating theoretical models of adversity into statistical models, and comparing the explanatory power of each model, respectively, is an important and helpful addition to the literature. However, the analysis would be strengthened by employing more sophisticated modelling techniques that account for between-subjects covariates and the presentation of the data needs to be streamlined to make it clearer for the broad audience for which it is intended.

Strengths
Several outstanding strengths of this paper are the large sample size and its primary aim of statistically comparing leading theoretical models of adversity exposure in the context of skin conductance response. This paper also helpfully reports Cohen's d effect sizes, which aid in interpreting the magnitude of the findings. The methods and results are generally thorough.

Weaknesses
The largest concern is that the paper primarily relies on ANOVAs and pairwise testing for its analyses and does not include between-subjects covariates. Employing mixed-effects models instead of ANOVAs would allow more sophisticated control over sources of random variance in the sample (especially important for samples from multi-site studies such as the present study), and further allow the inclusion of potentially relevant between-subjects covariates such as age (e.g. Eisenstein et al., 1990) and gender identity or sex assigned at birth (e.g. Kopacz II & Smith, 1971) (perhaps especially relevant due to possible to gender or sex-related differences in ACE exposure; e.g. Kendler et al., 2001). Also, proxies for socioeconomic status (e.g. income, education) can be linked with ACE exposure (e.g. Maholmes & King, 2012) and warrant consideration as covariates, especially if they differ across adversity-exposed and unexposed groups. On a related methodological note, the authors mention that scores representing threat and deprivation were not problematically collinear due to VIFs being <10; however, some sources indicate that VIFs should be <5 (e.g. Akinwande et al., 2015).

Additionally, the paper reports that higher trait anxiety and depression symptoms were observed in individuals exposed to ACEs, but it would be helpful to report whether patterns of SCR were in turn associated with these symptom measures and whether the different operationalizations of ACE exposure displayed differential associations with symptoms. Given the paper's framing of SCR as a potential mechanistic link between adversity and mental health problems, reporting these associations would be a helpful addition. These results could also have implications for the resilience interpretation in the discussion (lines 481-485), which is a particularly important and interesting interpretation.

Given that the manuscript criticizes the different operationalizations of childhood adversity, there should be greater justification of the rationale for choosing the model for the main analyses. Why not the 'cumulative risk' or 'specificity' model? Related to this, there should also be a stronger justification for selecting the 'moderate' approach for the main analysis. Why choose to cut off at moderate? Why not severe, or low? Related to this, why did they choose to cut off at all? Surely one could address this with the continuous variable, as they criticize cut-offs in Table 2.

In the Introduction, the authors predict less discrimination between signals of danger (CS+) and safety (CS-) in trauma-exposed individuals driven by reduced responses to the CS+. Given the potential impact of their findings for a larger audience, it is important to give greater theoretical context as to why CS discrimination is relevant here, and especially what a reduction in response specifically to danger cues would mean (e.g. in comparison to anxiety, where safety learning is impacted).