Prediction of postpartum depression with an online neurocognitive risk screening tool for pregnant women

Postpartum depression (PPD) is a severe mental illness affecting 10–15% of mothers. Emerging evidence indicates that negative neurocognitive bias in response to infant distress during pregnancy marks an increased risk of PPD. This proof-of-concept study aimed to investigate the association between negatively biased neurocognitive processing of infant distress during pregnancy and subsequent PPD and to explore the feasibility of an online risk screening tool. In the second or third trimester of pregnancy, 87 participants underwent two online tests of reactivity to and evaluation of infant distress and completed questionnaires regarding psychosocial risk factors. After birth, participants rated their depressive symptoms online and underwent a diagnostic telephone interview concerning PPD. Irrespective of depressive symptoms during pregnancy, negative reactivity to and evaluation of infant distress predicted PPD (reactivity: Exp(B) = 1.33, p = 0.04) and depressive symptoms after birth (reactivity: B = 0.04, p = 0.048; evaluation: B = 0.10, p = 0.04). The negative reactivity toward infant distress showed high sensitivity and moderate speciﬁcity (89% and 77%, respectively), while the evaluation of infant distressed cries showed lower


Introduction
The prevalence of postpartum depression (PPD) has been estimated to 30 and 50% for women with a history of depression or bipolar disorder, respectively, and to 8% in women without previous mental illness ( Viguera et al., 2011 ). According to the ICD-11 , diagnostic criteria for PPD are identical to depression outside the peripartum period but with an onset within six weeks of delivery ( WHO, 2019 ). While some symptoms are identical to those for Major Depressive Disorder (MDD) outside the peripartum period, including depressed mood, loss of interest and motivation, feelings of guilt and worthlessness and suicidal thoughts, research indicates that anxiety may be more prevalent in PPD than MDD ( Batt et al., 2020 ). Moreover, mothers are at increased depression risk for at least six months after delivery, and researchers and clinical practitioners consider depression within this period as PPD ( Musters et al., 2008 ;Wisner et al., 2002 ). One of the strongest risk factors for PPD is a history of depression ( Liu et al., 2022 ). In Denmark, most pregnant women who previously have had a depression or other mental illness are offered extra care from family clinics or maternity wards ( Sundhedsstyrrelsen, 2021 ). However, since the extra care are mostly offered to pregnant women with previous mental illness, the 8% of pregnant women without previous mental illness who will have their first-time onset of depression in the postpartum period ( Viguera et al., 2011 ) are not offered any extra preventive intervention. This might partly be due to lack of precise prediction of PPD. In Denmark, the Edinburgh Postnatal Depression Scale (EPDS) is routinely administered by health professionals during home visits 2 months after birth to detect PPD symptoms in mothers ( Sundhedsstyrrelsen, 2021 ). Nevertheless, there are no national guidelines on how and when to screen for PPD risk in pregnancy before depressive symptoms develop ( Sundhedsstyrrelsen, 2021 ).
The aetiology of PPD is, like MDD, multifactorial with contribution from genetic, physiological, neural, and psychosocial factors and their interaction ( Batt et al., 2020 ;Nguyen et al., 2019 ;Zhao and Zhang, 2020 ). Genetic factors, previous depressive episodes, family history of MDD, acute stress, decreased neural processing of reward, lack of emotional support from a partner, and low socioeconomic status are shared risk factors for PPD ( Beck, 2001 ;Hansotte et al., 2017 ;Tambelli et al., 2019 ;Viguera et al., 2011 ) and MDD ( Batt et al., 2020 ). Risk factors unique to PPD include low attachment feelings to the unborn child ( Austin et al., 2013 ;Petri et al., 2018 ), parenting stress ( Batt et al., 2020 ) and the major fluctuation in sex steroid hormones during pregnancy and after birth ( Frokjaer, 2020 ). The massive increase in hormones throughout pregnancy is also associated with neural plasticity and neurocognitive changes that are thought to reflect preparation for motherhood ( Roos et al., 2012( Roos et al., , 2011. For example, healthy firsttime mothers, scanned a few days after birth, showed reduced cortical thickness in the dorsolateral prefrontal cortex, superior parietal gyrus, right inferior parietal sulcus and supramarginal gyrus compared to nulliparous women ( Chechko et al., 2021 ). Moreover, pregnant women display greater and faster attentional processing of infant stimuli compared with non-pregnant nulliparous women, which has been associated with more maternal sensitivity in interactions with own infant after birth ( Dudek et al., 2020 ;Pearson et al., 2011 ). In contrast, pregnant women with depressive symptoms show attentional avoidance of infant stimuli ( Macrae et al., 2015 ;Rutherford et al., 2016 ). In keeping with this, mothers with PPD show hypo-activity in response to infant cries in paralimbic, cingulate, and prefrontal regions ( Laurent and Ablow, 2012 ). This likely reflects disturbance of neural networks involved in emotional response and regulation and contribute to the parenting difficulties associated with PPD. We found that such negative neurocognitive bias exists even in remitted pregnant women with a history of depression, as evidenced by more negative evaluation of infant cry sounds ( Bjertrup et al., 2021a ). Moreover, more negative evaluations of infant cries during pregnancy were associated with a markedly enhanced risk of PPD across all participants with or without depression history also after adjusting the association for current subsyndromal depressive symptoms. Specifically, more negative evaluations of infant cry directly predicted PPD in a model adjusted for subsyndromal depressive symptoms and antenatal attachment quality ( Bjertrup et al., 2021a ). This is consistent with research on neurocognitive trait-markers for depression in general, which has shown that negatively biased emotion processing predicts depression onset and recurrence of depressive episodes ( Bhagwagar et al., 2004 ;Milders et al., 2010 ;Miskowiak and Carvalho, 2014 ). It has therefore been proposed that negatively biased emotion processing constitutes a trait biomarker for depression ( Miskowiak and Carvalho, 2014 ). Trait biomarkers are defined as properties that can be objectively measured, indicate a pathological process or treatment response, and considered to be persistent as they exist before onset, during and after remission of a disorder. Trait biomarkers can therefore be useful in determining risk for developing a disorder such as PPD ( Hacimusalar and E ş el, 2018 ).
Notably, a recent study found that pregnant women's neurocognitive processing of infant emotional cues was associated with their behaviours toward their own infants after birth ( Leerkes et al., 2022 ). Further, PPD affects the mother's ability to stay appropriately engaged and attuned in the relationship with her infant ( Field et al., 2006 ;Goodman and Liu, 2014 ), which can have negative conse-quences for mother-infant interaction and child development . Therefore, negatively biased processing of infant emotions may not only constitute a risk factor for PPD but also -both directly and indirectly -be a key early environmental risk factor for the mental health of the infant after birth.
There is a need for a comprehensive screening tool comprising questions regarding well-known risk factors as well as the evaluation of and reactions to infant distress for accurate prediction of PPD risk. Accordingly, better early identification of risk factors is critical to develop new intervention methods individually tailored to prevent PPD and related negative consequences for the infant ( van den Heuvel, 2022 ). Since psychological preparation for motherhood, and in some cases depressive symptoms, begin during pregnancy ( Gollan et al., 2014 ), this offers the first important window of opportunity to prevent PPD and improve future mother-infant relations. Several studies have consistently associated psychosocial risk factors with PPD, and some screening tools assessing these factors during pregnancy do exist. For example, The Antenatal Risk Questionnaire (ANRQ) measures multiple psychosocial risk factors and have a validated cut-off for which to consider a pregnant woman at risk of PPD ( Austin et al., 2013 ). However, no study has investigated whether a screening tool assessing both psychosocial and affective neurocognitive risk can correctly identify those at high and low risk of PPD.
The overall aim of this study was to investigate whether negative neurocognitive bias can serve as a biomarker for predicting PPD, and whether the assessment of this bias with an online PPD risk screening tool is feasible. Specifically, we aimed to firstly investigate whether we could replicate our previously observed association between negatively biased neurocognitive processing of infant stimuli during pregnancy and PPD. Secondly, we aimed to investigate the sensitivity and specificity of negatively biased neurocognitive processing of infant distress during pregnancy for predicting onset of PPD. Finally, we explored the feasibility of conducting the PPD risk screening with an online instrument. We hypothesised that PPD and self-rated depressive symptoms within six months would be predicted by (I) negative reactivity in response to intense infant distress and (II) negatively biased evaluation of intense infant distress in two separate regression models adjusted for psychosocial risk factors. For exploratory purposes, we also examined reactivity to and evaluation of neutral, happy, and medium distressed infant stimuli.

Recruitment and participants
We obtained names and personal identification (CPR) numbers for women in second and third trimesters of pregnancy from the national register, The Danish Civil Registration System (CPR), through 'Forskerservice', Danish Health Data Authority (Sundhedsdatastyrelsen). Potential participants received an invitation sent via the online digital mailbox "E-boks" that is linked to the CPR numbers of all adult citizens in Denmark ( E-boks, 2019 ). The invitation e-mail included short information about the project and a link that redirected to REDCap (Research Electronic Data Capture) which is the secure and GDPR compliant application used for building and managing the online screening tool for this study ( Harris et al., 2019 ). Inclusion criteria were pregnancy in second or third trimester ability to read Danish, age ≥18 years and residence in the Capital Region of Denmark. Exclusion criteria were severe neurological disorder, schizophrenia, and schizoaffective disorder. Approvals for the study were obtained from the local ethics committee in the Capital Region of Denmark (journal number: H-20056223) and from the Danish Data Protection Agency (approval number: P-2020-801). The study was conducted in compliance with existing law on data protection and in accordance with the Declaration of Helsinki (version 2013). Participants gave informed consent online through REDCap.

Procedures and design
Pregnant participants were assessed with the online screening tool via the REDCap webpage. They were instructed to make sure they had time to complete the entire screening without interruption ( ∼45 minutes). The webpage provided contact information on the researchers and participants could call or write at any time if they had questions or technical issues. Participants rated infant sounds and then their emotional reactions to infant emotional videos were assessed. After this, they filled out questionnaires regarding psychosocial factors, including, demographic information, social relations, pregnancy, clinical history, current depressive symptoms, attachment to unborn infant, and bonding to own parents. Participants also typed in their due dates, and from this field, REDCap automatically sent out questionnaires after birth on a monthly basis for the first six months after birth assessing potential complications regarding the birth or infant health and depressive symptoms. In these online questionnaires, participants also indicated whether they had sought support from their general practitioner, a psychologist, psychiatrist, or other clinician due to mood changes or if they received any psychotropic medication. Participants received reminders in case they failed to complete the online questionnaires. Six months after birth, participants were interviewed by AJB over the telephone which enabled assessment of the incidence of PPD.

Materials
During pregnancy, participants first rated 50 short (4 s) infant sounds of five emotional categories: most distressed, moderately distressed, neutral, moderately happy and most happy. Infant sounds were obtained from the validated Oxford Vocal (OxVoc) sounds database, consisting of a validated set of non-acted OxVoc infant sounds recorded in their own homes in the UK during a play and feeding session with their primary caregiver ( Parsons et al., 2014 ). The computer screen was blank, while the short infant sound clips were played. Following each sound, a horizontal rating bar appeared on the screen and participants were instructed to rate the infants' feelings on a scale from very distressed (far left) over neutral (middle) to very happy (far right) by clicking directly on the bar. After this, another similar rating bar appeared, and they were asked to rate how they themselves felt when listing to the infant sound on a scale from very unpleasant to very pleasant. They were instructed to respond as fast as possible. Participants were also presented with two infant videos, a negative and positive videoclip, each with a duration of 28 s. The negative clip involved with a highly distressed four-month-old infant crying heavily, while being unattended by a caregiver. In contrast, the positive clip showed a mother and her quadruplets, all laughing continuously. Mothers were instructed to passively watch the videos. After each video, they rated the infants' and their own feelings on a rating bar.
For both infant emotion rating tasks (infant sounds and videos), the rating bar represented a visual analogue scale of 0-100. As the two infant videos displayed distinctly distressed and happy infants, respectively, the rating scale for these videos ranged from 0-100, with neutral (0) to most distressed (100) for baby distress, and neutral (0) to most happy (100) for laughing babies. However, the infant sounds were ambiguous and challenging to interpret. To prevent participants' responses from being influenced by rating bars that only ranged from "not distressed" to "very distressed" or "not happy" to "very happy," as seen in the videos, participants were given a uniform rating bar for all 50 infant sounds. High scores signified happiness (100 being the happiest), low scores indicated distress (0 being the most distressed), and a mid-point score (50) represented neutral ratings.
After the infant rating tasks, participants completed questionnaires assessing age, due date, level of education, pregnancy week, medication status, job status, and income level. Participants provided a "yes" or "no" answer to having financial worries, social support, family history of psychiatric disorders (and which), own personal history of mental illness (and which), loss of a parent, and to having experienced a traumatic event in their childhood. Participants who responded affirmatively were then prompted to describe the specific traumatic event. Based on these descriptions, the presence of childhood trauma was verified based on whether they involved any of the following: sexual, physical, and/or emotional abuse, emotional or physical neglect, the sudden death of a close relative or a divorce of parents. Intensity and quality of attachment to unborn child was assessed with the Maternal Antenatal Attachment Scale (MAAS) ( Maas et al., 2016 ). The Parental Bonding Inventory (PBI) was used to assess the two dimensions of parental care and parental overprotection that the participants' received from own mother and father in their childhood ( Parker et al., 1979 ). Depressive symptoms were assessed with the Edinburgh Postnatal Depression Scale (EPDS) ( Cox et al., 1987 ;Smith-Nielsen et al., 2018 ). Finally, in the last questionnaire, participants gave feedback on using the online screening tool. The order of the instruments in the screening tool was fixed. The EPDS was used for the monthly online ratings of depressive symptoms. For the six month telephone interviews after birth, the incidence of PPD was assessed with the short structured diagnostic interview Mini-International Neuropsychiatric Interview (M.I.N.I.) ( Sheehan et al., 1998 ).

Statistical analyses
We had two outcome measures in this study: 1) the occurrence of PPD in the first six months after birth (PPD or no PPD) and 2) average depressive symptoms after birth (mean EPDS score). Predictor variables comprised the affective neurocognitive processing of infant stimuli: emotional reaction to the infant distress video and evaluation of most distressed cries (two rating variables). For exploratory purposes, we also investigated the association between the outcome measures and 14 psychosocial predictors: age, depressive symptoms during pregnancy, level of education, pregnancy week, parental bonding, attachment to unborn child, job status, financial worries, income level, social support, family history of psychiatric disorders, own history of psychological illness, loss of a parent, and experience traumatic event during childhood.
As a first step, we explored the bi-variate associations between the 16 predictor variables and the two outcome variables with Pearson correlation between two continuous variables, Spearman's Rho between mixed continuous and categorical variables and Chi-square tests for two categorical variables. We then conducted logistic regression and multiple linear regression with PPD and depressive symptoms after birth, respectively, as outcome variables. In the regression models, we included those predictor variables, which correlated strongest with the outcome, and that did not have high collinearity with each other. We investigated hypothesis (I) with negative reactivity in response to the infant distress video as a predictor, since we previously found this stimulus to be highly aversive across both healthy mothers and mothers with a mood disorder ( Bjertrup et al., 2021b ). Further, we investigated hypothesis (II) with negatively biased spontaneous evaluation of the most distressed infant sounds as a predictor, since we previously found negatively biased perception of this specific infant emotion to be the strongest predictor of PPD ( Bjertrup et al., 2021a ). Predictive validity of the neurocognitive tests of emotional reaction to the infant distress video and evaluation of most distressed cries was investigated with analyses of receiver operating characteristic (ROC) curves. Sensitivity, specificity and positive and negative likelihood ratios (LR) were computed ( Jaeschke et al., 1994 ). Feasibility was assessed as response rate during pregnancy and follow-up responses after birth and from the feedback received through the online tool. Data were analysed in the Statistical Package for Social Science (SPSS) version 25.

Response rate and feasibility
In total, 1400 pregnant women were invited via their E-boks account. Of these, 200 (14%) accepted the invitation, gave informed consent, and started the test. During the infant rating task, 113 discontinued the online screening, while 87 (44%) completed the full online assessment. After birth, 31 responded to the questionnaires within the first month, 45 responded from the first to the second month, and eight, 41, 34, 19 and 14 responded in the third, fourth, fifth, sixth and seventh month, respectively. Responses on the EPDS were most complete the first two months after birth, and we therefore chose the average EPDS scores within these first two months as the outcome measure of depressive symptoms. Averaging EPDS scores within the first two months after birth rendered 60 (69%) responses, i.e., responses from 27 participants were missing. Finally, 83 (95%) participants were interviewed six months after birth. Of the four participants lost to follow-up, one experienced a still birth, one wished not to be interviewed, and two did not reply to any calls.

Participant characteristics
For the pregnant sample included in the study ( N = 87), the mean age was 32.6 (SD = 4.5), average pregnancy week was 32.4 (SD = 3.0), and 65.5% were primiparous. They were generally well-educated, with a masters' degree or higher education for 61%, a bachelors' degree for 24%, high school for 6% and a vocational qualification for 9%. Further, 94% of participants were currently employed. In terms of risk status, 24% of participants had a first-degree family history of severe mental illness, and another 24% reported a personal history of depression (with 6% having both). Additionally, 37 (43%) reported experiencing other psychological problems, including three participants (3%) who had been diagnosed with bipolar disorder, 19 (22%) who had experienced anxiety, seven (8%) who sought professional help due to trauma or prolonged grief related to abuse, severe illness (personal or in near relatives), or a breakup, and seven who reported work-related stress. Notably, of the 37 partic- Fig. 1 More negative reactivity toward the infant distress video during pregnancy was associated with a greater level of depressive symptoms, rated with the Edinburgh Postnatal depression scale, on average within the first two months after birth. ipants who reported experiencing other psychological problems, 13 (35%) also had a personal history of depression. On average depressive symptoms during pregnancy were low (EPDS mean = 3.2, SD = 3.7) and 5 participants (6%) had EPDS-scores of ≥11, which is the cut-off for PPD risk ( Smith-Nielsen et al., 2018 ). Of the 83 participants who were interviewed six months after birth, 9 (11%) of them developed PPD. See Table 1 for details on psychosocial variables and affective neurocognitive scores.

Correlations
In line with hypothesis (I), more negative reactivity (i.e., higher scores on a 0-100 rating scale) in response to the infant distress video during pregnancy correlated with higher EPDS scores in the two first months after birth ( r = 0.26, p = 0.04) and with PPD (r s = 0.36, p ≤ 0.001) ( Figs. 1 and 2 ). Consistent with hypothesis (II), there was a significant correlation between more negative evaluation (i.e., lower scores on a 0-100 rating scale) of the most distressed infant cries and higher EPDS scores in the two first months after birth ( r = −0.27, p = 0.04) ( Fig. 3 ). In contrast with hypothesis (II), there was no statistically significant association between more negative evaluation of the most distressed infant cry sounds and PPD (p-values ≥0.06).

Regression analyses
In two multiple regression analyses with EPDS score as outcome, we included five predictors: lower income, having financial worries during pregnancy, personal history of depression, depressive symptoms during pregnancy and 1) reactivity to infant distress in one model and 2) evaluation of infant cries in the other model. In two logistic regression models with PPD as outcome, we included the two predictors: experience of trauma in childhood and ratings of own emotional reactivity in response to 1) the infant distress video in one model and 2) reactivity to most distressed infant sounds in the second model.  Table 1 Psychosocial factors and affective neurocognition in pregnancy for all participants and groups of participants who were followed up or not at two and six months after birth.    Table 2 and Fig. 4 display the results of the ROC-analyses. For the test of negative emotional reactivity toward infant distress, the ROC curve yielded a large area under the curve (AUC) = 0.83. An optimal cut-off score of ≤96.0 had a sensitivity of 89% and a specificity of 77%. The positive and negative LRs were 3.90 and 0.14, respectively, demonstrating that the chance of obtaining a true positive result was 3.90 times higher than the risk of getting a false positive and the risk of a false negative was 0.14 times the likelihood of obtaining a true negative result.

Predictive validity of the affective neurocognitive tests
The test of negative evaluation of most distressed infant cries had medium diagnostic accuracy with AUC = 0.70. The most optimal cut-off score of ≤17.4 had a sensitivity of 67% and a specificity of 66%. The positive and negative LRs were 1.98 and 0.50, respectively. Thus, that the likelihood of obtaining a true positive result was 1.98 times greater than the risk of getting a false positive and the likelihood of a false negative was 0.5 times the likelihood of obtaining a true negative result.

Exploratory additional analyses
More positive ratings of own emotions to the most happy infant sounds (i.e., higher scores on a 0-100 rating scale) correlated with PPD (r s = 0.23, p = 0.03). There was a significant correlation between more negative reactivity in response to the most distressed infant sounds (i.e., lower scores on a 0-100 rating scale during pregnancy and PPD (r s = −0.23, p = 0.03). Reactivity to or evaluation of medium distressed, medium happy and neutral infant emotions did not correlate with PPD (p-values ≥0.12) or depressive symptoms after birth (p-values ≥0.16).

Discussion
This proof-of-concept study investigated negative neurocognitive bias in pregnancy as a biomarker for predicting PPD and the feasibility of assessing this bias in an online risk screening tool. The key finding was that (i) more negative emotional reactivity while watching a short infant distress video and (ii) more negative interpretations of infant cry sounds during pregnancy predicted depression symptom severity within two months after birth and PPD within six months after birth, in regression models adjusted for subsyndromal depressive symptoms during pregnancy, history of previous depression, income, financial worries, and childhood trauma. From a feasibility perspective, the response rate was relatively low, suggesting a need for a shorter more optimised online risk screening tool.
In accordance with the hypothesis, negatively biased neurocognitive processing of infant distress during pregnancy independently predicted depression symptom severity within two months after birth and PPD within six months after birth. This is a replication of our previous finding that negative neurocognitive bias during pregnancy was associated with PPD risk ( Bjertrup et al., 2021a ), indicating that negative neurocognitive bias might represent a trait biomarker for PPD risk, rather than merely representing the depressive symptoms during pregnancy. Moreover, we demonstrated good predictive validity of the test of negatively biased emotional reactivity toward the infant distress video at the cut-off score of ≥96.0 with high sensitivity (89%) and moderate specificity (77%) for classifying PPD. We argue that a test screening for PPD risk should prioritise sensitivity over specificity as the potential consequence of being at heightened risk but not receiving a preventive intervention is worse than receiving a preventive intervention without being at heightened risk of PPD. At least if the intervention is non-pharmacological. The test of negatively biased evaluation of infant distressed cries had poorer predictive validity with sensitivity and specificity of 67% and 66%, respectively at the cut-off≤17.4. Emotional reactivity toward the infant distress video was most strongly associated with PPD and had better predictive validity in terms of sensitivity and specificity in predicting PPD than the evaluation of distressed infant cries test. One explanation for this could be that rating own emotional reactivity vs. the infants ' emotions is more closely related to motherhood experiences after birth when participants' own infants are crying. A lower threshold for tolerating intense infant distress is likely associated with more negative emotional states and parental stress, which could increase the risk of developing PPD. However, the threshold for tolerating infant distress is also contingent on the emotionally neurocognitive biased interpretation of infant distress, as affective neurocognitive processes are interrelated so that more negative interpretation of emotional stimuli is associated with greater negative emotional reactivity and vice versa ( LeMoult, 2020 ;LeMoult and Gotlib, 2019 ). In line with the present results, previous research found that negative interpretation bias predicted depressive symptoms four to six weeks later in healthy undergraduate students ( Rude et al., 2002 ). Further, reduction in negative neurocognitive bias in the processing emotional faces has been associated with response to antidepressant treatment before symptom reduction and is thus a putative early marker of treatment effect ( Godlewska and Harmer, 2021 ). Interventions targeting negative neurocognitive bias in response to infant distress during pregnancy could therefore potentially reduce the risk of PPD, although the causal relationship between negative neurocognitive bias and depression is unknown.
Our finding that lower income, having financial worries, personal history of depression, depressive symptoms during pregnancy and having experienced a trauma in childhood were associated with depressive symptoms after birth and/or PPD concur with previous studies ( Beck, 2001 ;Hansotte et al., 2017 ;Tambelli et al., 2019 ;Viguera et al., 2011 ). Assessment of these psychosocial risk factors should thus be included in a future screening tool. In contrast to our expectation, age, level of education, pregnancy week, parental bonding, attachment to unborn child, job status, social support, family history of psychiatric disorders, and loss of a parent were not associated with depressive symptoms or PPD. This lack of association is likely due to the relatively low sample size, and these risk factors should therefore not be disregarded just yet as since they have previously been associated with PPD ( Smorti et al., 2019 ). For example, a previous study of > 3000 pregnant women reported an association between perceived low social support during pregnancy and more depressive symptoms after birth ( Gan et al., 2019 ). However, in our study only four pregnant participants reported lack of social support, which limits the statistical power. Further, only one participant scored above 11 on the EPDS (the validated cut-off for depression risk) during the first two months after birth and nine (11%) went on to develop PPD within six months after birth, which limits statistical power to predict these outcomes. This also suggests that the included participants might represent a slightly healthier sample compared with the general population of pregnant women, although, the PPD rate was similar to previous studies ( Viguera et al., 2011 ).
From a feasibility perspective, the online screening tool was suboptimal. The response rate of 14% was low compared a previous study using E-boks to invite pregnant women to a study of worries during the COVID-19 pandemic, which reported a response rate of 60% ( Schrøder et al., 2021 ). A meta-analysis reported an average response rate to online survey invitations to 40% for studies on depression and anxiety ( Burgard et al., 2020 ). The main reason for this discrepancy in response rates is that our screening tool is time con-suming, and many participants quitted the online screening during the infant sound rating task due to time or because of problems with playing the infant sounds. Although the infant sound rating task was feasible and reliable in previous in-person studies ( Bjertrup et al., 2021a ;Kringelbach et al., 2016 ;Parsons et al., 2014Parsons et al., , 2017, it is less feasible for a non-supervised online screening and may, also in the interest of brevity, not be included in a future screening tool. In contrast, the infant distress video task was shorter, simpler, and thus more suitable for an online screening. Moreover, the infant distress video had high sensitivity and specificity and should therefore be included in a future screening tool. Participants may feel less committed to online studies with no in-person contact to the researchers compared to studies with personal presence ( Khadjesari et al., 2011 ), and study invitation through e-mail yields the lowest response rates as compared with in-person, telephone, letter, or advertisements ( Wu et al., 2022 ). The use of incentives such as monetary reward or lotteries have been shown to improve response-rates ( Burgard et al., 2020 ) and could therefore be implemented for future online screening. Response and completion rates could possibly be increased if the screening was recommended by the healthcare professionals (e.g., midwifes) that the pregnant women receive care from. Yet, in the current form, the tool may not be a feasible method for research or clinical purposes. Studies suggest that the duration of web-based surveys and tests should have a maximum duration of 20 min to balance attrition and completion rates ( Menon and Muraleedharan, 2020 ;Yan et al., 2010 ).
The findings should be considered in light of several limitations. Firstly, the low sample size confined the number of predictors we could include in the regression models, and we could not test the predictive validity of a tool integrating all risk factors. Attrition after birth might be due to PPD, as the four participants that were lost to followup in the current study displayed significantly more PPD risk, including, financial worries, less social support, and more traumatic events in childhood. A second limitation is that the study did not account for hormonal risk factors, or factors after birth such as obstetrical complications and birth trauma, infant illness, or parenting stress, which are known risk factors for PPD ( Liu et al., 2022 ;Smorti et al., 2019 ). Therefore, participants with a low risk-score on our screening, who went on to develop PPD, could for example potentially have experienced a traumatic birth. Nevertheless, this study aimed to identify risk of PPD already during pregnancy, regardless of postpartum events. Moreover, the included factors likely pose risk for women who also have hormonal or obstetric risk factors, although interaction and/or dose-response effects of risk factors are unknown. Thirdly, assessing PPD within six months after birth could obscure the distinction between distinct PPD and a relapse of a previous depression, as it has been suggested that these are two aetiologically different subtypes of depression ( Batt et al., 2020 ). In future larger studies it would therefore be informative to differentiate those with previous depression from those with first onset during the postpartum. Fourth, our we only included pregnant women in the Capital Region of Denmark and the sample may therefore not fully represent the general pregnant population. Indeed, the sample consisted of a majority of well-educated, employed individuals with strong social support networks; factors that are considered protective and could have influenced our findings. Finally, the participants were pregnant during the COVID-19 pandemic, which may have increased their levels of stress, anxiety, and worry. However, it is also possible that this was a sample of women who were particularly resilient to stress and physically healthy enough to consider pregnancy despite the pandemic. Nevertheless, our findings do replicate those of a smaller pre-pandemic study ( n = 37) ( Bjertrup et al., 2021 ), which suggests that our sample was not biased by the pandemic context.
Based on the present findings our planned next step is to conduct a larger-scale national study. The perspective is an online tool with high sensitivity and specificity in the identification of pregnant women at risk for PPD, which can provide a basis for, and lead to greater precision in, the allocation of preventive strategies. If the findings are replicated in a larger sample more representative of the general population of pregnant women in Denmark, this can have important implications for future digital screening of PPD risk.
In conclusion, we were able to replicate our previously demonstrated association between negative neurocognitive bias in response to infant distress during pregnancy and risk of PPD onset, indicating that negative bias is a robust risk marker. Further, a cut-off of ≥96 for negative emotional reactivity to the infant distress video had high sensitivity and specificity for predicting PPD, suggesting that this is a valid biomarker of PPD risk during pregnancy. In contrast, only some previously proposed psychosocial risk factors were associated with PPD, possibly due to the relatively small sample size. The response rate was low, suggesting that the online screening tool should be shortened and investigated in a larger-scale study. The potential is an integrative PPD-risk profile screening tool to optimize identification of PPD risk that enables ultra-early PPD prophylactic strategies.