Perinatal samples pose challenges to the measurement and conceptualization of depression. First, there is overlap between symptoms of depression and normative perinatal experiences, such as changes in appetite or changes in sleep, which can lead to both over-diagnosis and under recognition of depression in perinatal samples (Matthey and Ross-Hamid, 2011; Pereira et al., 2014; Ross et al., 2005). Second, the maximally sensitive and specific values on depression measures have been found to change across the perinatal period (Ji et al., 2011). And third, there is ongoing debate about whether or not a Major Depressive Episode (MDE) occurring in the postnatal period is a qualitatively different disorder than a non-puerperal MDE (Batt et al., 2020; Di Florio and Meltzer-Brody, 2015). Even with these challenges, accurate measurement and conceptualization of depression during the perinatal period is vital, since perinatal depression is common and associated with long-term adverse outcomes for the affected individual and their families (Gavin et al., 2005; Luca et al., 2019; Stein et al., 2014; Stuart-Parrigon and Stuart, 2014; Woody et al., 2017). With this paper, we built on the literature by considering positive affect, how to identify it from common measures of depression and its correlates, and how it relates to other features of depression.

One approach to addressing these concerns is to consider alternative frameworks for conceptualizing and measuring mental illness. A common feature of such frameworks is that they study narrower constructs than are measured with diagnostic categories found in the Diagnostic and Statistical Manual of Mental Disorders, 5th edition (American Psychiatric Association, 2013). One of the most prominent examples of a framework for psychiatric disorders is the Research Domain Criteria (RDoC). RDoC was developed as a research framework for mental illness that integrates behavioral, psychological, and biological approaches to better the understanding of multiple dimensions of functioning relevant to the range of human behavior. The RDoC framework is expressed as a “matrix” of psychological domains by units of analysis. This structure promotes the synthesis of multiple levels of functioning, from human behavior to molecules, and facilitates a dimensional conceptualization of mental health rather than a categorical one, as is the case in DSM-5 categories or classification based on scores from symptom measures. Another framework uses a bifactor approach and emphasizes a general risk factor for psychopathology, sometimes called a p-factor, which cuts across all diagnoses. The most commonly used bifactor model in psychopathology research is the HiToP framework. HiToP has a hierarchical structure, where narrower domains and symptoms (e.g., substance abuse) are nested within broader domains and symptoms (e.g., externalizing; (Caspi et al., 2014; Caspi and Moffitt, 2018; Conway et al., 2019)). Support for this framework derives from factor analyses that show a bifactor model of psychopathology (which includes a general p-factor in addition to other factors), sometimes improves model fit (Lahey et al., 2012). In a sample of pregnant individuals, Clark et al. (2023) found support for using a HiToP structure to symptoms of psychopathology during pregnancy. However, questions about the utility and interpretation of the p-factor have been raised (Watts et al., 2019).

Given concerns about the identification and measurement of depression symptom severity, particularly in the perinatal period, we sought to characterize symptoms of depression and its correlates across the perinatal period using transdiagnostic factors. In a recently published study (Cochran et al., 2020), we sampled women with a history of mental illness, based on knowledge that such an approach enhances the likelihood of capturing a greater range of symptom presentations relative to sampling the general population. We examined participants' scores on items from seven commonly used, empirically supported measures of depression and common correlates of depression: anxiety, stress, and sleep disturbances. We chose and refined our transdiagnostic constructs iteratively; we began with subdomains from RDoC, and in particular from the Negative Valence System within RDoC, and added factors that represented symptoms frequently measured by depression, stress, anxiety, and sleep scales but not represented in the RDoC matrix. Our process yielded six factors. Four of these factors were derived from the RDoC matrix (Loss, Potential Threat, Frustrative Nonreward, and Sleep-Wakefulness), while the remaining two factors were outside the scope of RDoC (Somatic and Coping). Based on the HiToP literature, we tested these symptom domains with and without a general factor. We found that a model using a general factor and our six transdiagnostic factors characterized symptoms well in our perinatal sample; however, model components needed to vary to capture some factors around delivery. There are normative changes, such as sleep that occur across pregnancy (e.g. Sweet et al., 2020) and the postpartum period (e.g. Rychnovsky and Hunter, 2009) that may account for variability in the accuracy of depression measures at different points of pregnancy and the postpartum period (Ji et al., 2011). Thus, situating symptoms in particular perinatal windows (trimester and postpartum periods) was a strength of our previous study. We found that although factor loadings and intercepts could not remain fixed between certain perinatal periods, our overall factor structure performed well in each period.

Cochran et al. (2020) provided an important step in situating depression and related symptoms experienced by women during the perinatal period in a transdiagnostic framework. However, there were two important limitations to that work. First, we did not test whether creating a factor for items that measure positive affect (PA) improved model fit. Second, the homogeneous sample characteristics, e.g., overwhelmingly white, economically privileged, and married, curtailed the generalizability of our results.

Positive affect (PA) can be defined as any affective state with a positive valence. Individuals who frequently experience positive emotions, particularly in response to stressful events, may exhibit higher trait-level positive emotionality. Low PA is linked to several psychological disorders but is most strongly linked to depression, and is thought to be one of the most robust specifying factors that distinguishes depression from other disorders (Watson and Naragon-Gainey, 2010).

In studies of the general adult population, levels of PA are found to relate to depression and its correlates in important ways. Low PA is a risk factor for psychopathology and maladjustment, and high PA may be a protective factor against psychopathology and maladjustment (Nelis et al., 2015; Watson and Naragon-Gainey, 2010). A series of meta-analyses of the relationship between positive emotionality (operationalized as trait-level constructs defined by positive emotions) and depression and anxiety found that, among samples of adults, positive emotionality was cross-sectionally and longitudinally associated with depression and anxiety symptoms; the longitudinal association remained significant after controlling for baseline depression/anxiety symptoms (Khazanov and Ruscio, 2016). Moreover, in experience sampling methods in twins, Wichers et al. (2007) found that PA buffered against negative affect and the appraisal of events as stressful and reduced lifetime depression risk. PA has also been shown to predict the efficacy of treatment for depression, such that higher levels of PA before treatment are associated with more reduction in depression symptoms (Oren-Yagoda et al., 2018). For example, Riskind et al. (2013) found that elevated negative affect prospectively predicted depression symptom reduction only in individuals with low levels of PA; in individuals with high levels of PA, negative affect did not predict depression symptom change.

Less is known about PA in relation to depression across the perinatal period, though findings from a few published studies support PA playing a similar role as in general population samples i.e., associated with both risk and positive outcomes. In a recent systematic review, Scroggins et al. (2022) reported that nine out of ten studies identifying symptom clusters of postpartum depression using the Edinburgh Postnatal Depression Scale (EPDS) identified anhedonia (markedly reduced ability to experience pleasure) as one of their clusters, despite there being considerable heterogeneity in the gestational timing and sample characteristics of the studies. Buttner et al. (2012) found that positive and negative affect emerged as distinct factors in the early postpartum period in a non-treatment-seeking sample. Evidence suggests that PA during pregnancy is associated with improved offspring outcomes, including better infant social-emotional development (Chaves et al., 2022; Hanley et al., 2013) and healthy birth outcomes (Pesonen et al., 2016) and negatively associated with previous pregnancy loss (Côté-Arsenault, 2007) and postpartum depression symptoms (Bos et al., 2013). Further, tests of two perinatal mindfulness interventions (Lönnberg et al., 2020; Sun et al., 2021) found that their intervention groups concomitantly reduced symptoms of stress or anxiety and depression and increased PA or positive “states of mind.”

In Cochran et al. (2020), items indicating low PA were predominantly categorized in the broader Loss construct. Given that we had aimed to use RDoC constructs whenever possible to characterize our factors, we struggled with how to capture the positive valence system (PVS) as defined by RDoC in our set of measures. The PVS is broadly defined as constructs that “are primarily responsible for response to positive motivational situations or contexts” (National Institute of Mental Health (NIMH)). Most of our scale items did not provide a specific situation or context before describing an affective or emotional state, thus making it difficult to measure reward-driven behavior. However, despite considerable evidence that reward hyposensitivity may at least partially underlie deficits in PA in individuals with MDD (Pizzagalli et al., 2008; Treadway and Zald, 2013), and findings that treatments meant to increase reward-driven behaviors also increase PA (Craske et al., 2019), the operationalization of PA is not synonymous with the definition of PVS constructs. Further, measures of PA are strongly (r's between 0.60 and 0.61) correlated with the Positive Valence System (PVS) Scale, which was developed to measure the RDoC construct (Khazanov et al., 2020). Thus, though our battery precludes measurement of PVS constructs per se, the importance of the construct of PA warrants testing whether we may be able to capture a broader factor encompassing PA. Given the association between PA and depression risk, both longitudinally and cross-sectionally, measuring PA in a sample of perinatal women at risk of depression would enable us to test if PA is associated with the expression of other symptoms of depression. This may inform screening practices for individuals at risk of perinatal depression as well as future prevention efforts and interventions. For example, a combination of high levels of depression symptoms and low levels of PA may imply that an individual is particularly vulnerable to perinatal depression, which would warrant providers escalating a referral and informing treatment options.

In evaluating the generalizability of a study of perinatal depression, it is important to consider demographic characteristics of the sample. Cochran et al. (2020) sampled women who were primarily middle socioeconomic status (SES), married, and white. However, the hardships associated with low SES and single motherhood contribute to these adversities' conceptualization as risk factors for perinatal depression (Dagher et al., 2021; Stuart-Parrigon and Stuart, 2014). Thus, it is unclear to what extent our findings would generalize to perinatal women with less material privilege. Further, the relationship between race/ethnicity and perinatal depression is complex. A systematic review found higher rates of antenatal depression in non-Hispanic black and Hispanic pregnant women compared to white women (Mukherjee et al., 2016); however, few studies included in the review assessed moderators or correlates of racial/ethnic differences in perinatal depression. When included, the majority of the variance between racial/ethnic groups was explained by other factors (e.g., employment, discrimination exposure, history of childhood maltreatment). Critically, all these studies have compared total depression symptom scale scores or diagnoses. We have found no published studies testing the generalizability of factor models of perinatal depression among groups with varying sociodemographic characteristics. However, a small, mixed literature suggests that low SES may be a risk factor for a trajectory of symptoms of perinatal depression that is chronically higher across the perinatal period (Baron et al., 2017).

The aims of this study were as follows: First, to test whether the Cochran et al. (2020) transdiagnostic factor model characterizing women at elevated risk of depression during the perinatal period should be extended to include a PA factor. That is, does the inclusion of a PA factor in each perinatal period yield a better model fit? The ability to measure PA is important for clinicians and researchers, particularly given recent evidence that treatments that attempt to increase PA may be more effective at improving overall clinical outcomes in individuals with depression as compared to treatments that aim to reduce negative affect (Craske et al., 2023). We hypothesized that the addition of a PA factor would improve model fit in each of our perinatal periods. Our second aim was to determine what aspects of the factor model replicated in an independently obtained, more socio-demographically diverse sample. Testing the validity of our factor structure in an additional sample of women at risk for perinatal depression is vital for our understanding of the generalizability of our model. Further, factor analyses can be difficult to replicate; thus, replicating our factor structure in an additional sample would inform interpretations of the appropriateness of our factor structure in conceptualizations of perinatal depression (Osborne and Fitzpatrick, 2012). We hypothesized that we would replicate the general factor structure of the factor model in the second sample.