The psychological impact of early pregnancy loss

Correspondence address. Professor Tom Bourne, Tommy’s National Early Miscarriage Research Centre, Imperial College, Queen Charlottes and Chelsea Hospital, Du Cane Road, London W12 0HS, UK. E-mail: tbourne@ic.ac.uk

Human Reproduction Update, Volume 24, Issue 6, November-December 2018, Pages 731–749, https://doi.org/10.1093/humupd/dmy025

Cite

Jessica Farren, Nicola Mitchell-Jones, Jan Y Verbakel, Dirk Timmerman, Maria Jalmbrant, Tom Bourne, The psychological impact of early pregnancy loss, Human Reproduction Update, Volume 24, Issue 6, November-December 2018, Pages 731–749, https://doi.org/10.1093/humupd/dmy025

Abstract

Early pregnancy loss (EPL) is a common event, with scope for long-term personal and societal impact. There are three decades worth of published evidence of profound psychological sequelae in a significant proportion of women. However, the wide variety of outcomes, screening instruments, assessment timings and geographical locations makes it challenging to form a coherent picture of the morbidity within the whole group and its subgroups.

This review aims to investigate three questions. (1) What is the evidence for depression, anxiety and post-traumatic stress disorder (PTSD) following a miscarriage or an ectopic pregnancy in women and/or their partners? (2) What is the intensity and duration of these conditions, and how do they compare to those without losses? (3) Which patients have been found to be at highest risk of psychopathology? Answers to these questions are salient not only in day-to-day clinical interactions with those experiencing EPL, whose psychological needs may not be prioritized, but should also form the basis for tailoring healthcare policy in terms of screening for and treating the associated psychological morbidity.

The following databases were searched, from the start of each database up to July 2017: MEDLINE (Ovid interface, 1948 onwards), Embase classic + Embase (Ovid interface, 1947 onwards), and PsychINFO (Ovid interface, 1806 onwards). Search strategies were developed using medical subject headings (MeSH). The concepts of psychological morbidity (anxiety, depression or PTSD) and pregnancy loss (miscarriage or ectopic pregnancy) were first expanded with the Boolean operator ‘or’, then linked together using ‘and’.

Included studies were of prospective cohort design, including women or men following EPL (with the majority to have experienced losses before 24 weeks gestation), and reporting standardized psychometric measures for anxiety, depression and post-traumatic stress disorder. The timing of follow-up had to be specified and standardized across participants. Manuscript quality and risk of bias was assessed using the Newcastle–Ottawa Scale.

We found evidence of significant depression and anxiety in the first month following EPL in women. Partners were also shown to display depression and anxiety, albeit to a generally lower level. There is also evidence of post-traumatic stress symptoms relating to the EPL in three studies.

In view of their high frequency, EPLs can significantly contribute to the overall burden of psychopathology within a population. Recognition of this impact is important, so that severely affected individuals may be screened and treated appropriately. Further research to establish risk factors to promptly identify and treat these patients, and to optimize their management, is crucial.

Introduction

Pregnancy loss is a common problem, with miscarriage estimated to affect 25% of women who have been pregnant by 39 years of age ( Blohm et al., 2008), and ectopic pregnancy occurring in approximately 1% of pregnancies ( Varma and Gupta, 2012). There is scope for long-term emotional impact ( Beutel et al., 1995). However, historically, this has been overlooked. Reflecting this, it was little considered in the literature until the 1990s. Early studies focused on grief, which is not a discrete psychiatric diagnosis and is variably defined ( Janssen et al., 1997). They were also hampered by variable timings of assessment, retrospective recruitment, unrepresentative study populations and qualitative data ( Seibel and Graves, 1980; Cecil and Leslie, 1993; Harker, 1993).

However, since then, research in the field has proliferated, with a number of sizeable studies screening for psychiatric diagnoses (anxiety, depression and PTSD) following pregnancy loss ( Neugebauer et al., 1992a, 1992b; Engelhard et al., 2001; Cumming et al., 2007). These studies have shown that women appear to be at very significant risk of psychological morbidity following miscarriage, with up to 41% self-reporting clinically significant levels of anxiety ( Prettyman et al., 1993) and 36% reporting depression ( Neugebauer et al., 1992b) within 1 month. In the author’s own study, 39% of women self-reported meeting criteria for post-traumatic stress disorder (PTSD) at 3 months ( Farren et al., 2016). As an example, an estimated 250 000 miscarriages take place in the UK each year; these figures point to a significant public health issue. It is clearly essential to understand the natural history of these reactions and when treatment may be indicated.

However, hindering overall interpretation of the morbidity reported in these studies is the diverse nature of recruitment methodologies, assessment tools, timing of assessments and comparison groups. Studies have been published across a variety of journals (general, gynaecological or psychiatric) and over a period of three decades. There is also significant clinical heterogeneity, with some studies choosing to exclude those with a past psychiatric history, for example.

We aimed to conduct a narrative review of the existing literature on the psychological impact of EPLs, focusing on the prevalence, intensity and duration of symptoms of anxiety, depression and PTSD. We explore the variation in the literature and try to summarize evidence of predisposing factors, both background and encounter related, for psychological morbidity. A pertinent question, which we explore in the available literature, is whether psychopathology is related to delays to conception and involuntary childlessness or to the loss itself.

Methods

The methods set out below, in accordance with the PRISMA checklist, were specified in advance and documented in the study protocol. This was registered with PROSPERO (http://www.crd.york.ac.uk/PROSPERO, registration number CRD42017082754).

Eligibility criteria

Studies were considered eligible if they were of prospective cohort design, involving women and/or men following a miscarriage or an ectopic pregnancy. Studies including participants with late losses (>24 weeks) were eligible if the overwhelming majority (≥90%) had experienced loss within the first 24 weeks of pregnancy.

There is strong evidence that psychopathology declines over time ( Prettyman et al., 1993; Cordle and Prettyman, 1994; Beutel et al., 1995; Janssen et al., 1996; Cumming et al., 2007). Therefore, only studies with set specified timings of assessment, standardized across all participants, were included.

In order to comprehensively collate results, only empirical studies using a validated psychometric measure or interview to define the proportion of the study group meeting defined criteria for anxiety, depression or PTSD, or to compare to another group (pregnant women, women after live birth, women after termination, non-pregnant women, partners, or women who conceived with IVF) were included. Studies reporting on grief alone were excluded.

Studies with abstracts in languages other than English were excluded.

Information sources

The following databases were used: MEDLINE (Ovid interface, 1948 onwards), Embase classic + Embase (Ovid interface, 1947 onwards) and PsychINFO (Ovid interface, 1806 onwards).

This search was limited to the English language and was last updated on 27 July 2017.

Search strategy

Literature search strategies were developed using medical subject headings (MeSH). The concepts of psychological morbidity (anxiety, depression or PTSD) and pregnancy loss (miscarriage or ectopic pregnancy) were first expanded with the Boolean operator ‘or’, then linked together using ‘and’.

Details of the search strategies are included in Supplementary Table SI . The reference lists of included studies were also scanned to ensure complete literature saturation.

Study selection

Initially, titles and abstracts were screened against the eligibility criteria above. This was performed independently by two researchers (JF and NMJ). If the article was clearly not eligible based on the title and/or abstract alone, it was discarded. Otherwise, the full-text was reviewed by both researchers independently and the eligibility criteria were again applied. Disagreements were resolved by consensus.

Data collection, data items and summary measures

A spreadsheet was prospectively designed to accommodate the data, piloted on five papers by two reviewers, and refined. The primary researcher (JF) extracted data into this, which was then checked by a second reviewer (NMJ). Disagreements were resolved during research meetings involving two senior researchers (MJ & TB).

Data were extracted on the study population, participation and response rate, and assessment tools and timing. The proportion of participants reaching pre-defined, reported thresholds (‘caseness’) for anxiety, depression, PTSD or Acute Stress Disorder (ASD) was recorded, along with any comparison of measures in the EPL group versus the comparison group (by absolute scores or caseness) and change over time. Statistically significant (P < 0.05) risk factors for morbidity were extracted into a further table.

Quality assessment

Studies were assessed according to the Newcastle–Ottawa Quality Assessment Scale (NOAS) for Cohort Studies ( Supplementary Table SII ). This was performed by JF and checked by NMJ, with disagreements resolved through discussion. Where a comparison cohort was not used, this was documented under the ‘comparability’ section. Where the comparison group was exposed partners, an assessment was made according to NOAS comparability criteria, although it was originally designed for a non-exposed cohort. Stars for comparability were awarded if the cohorts were controlled for (a) parity and (b) any additional factor.

The NOAS requirement for assessment of outcome to be blind was omitted given the dominance of self-report measures and non-controlled studies. Follow-up duration of at least 1 month was required for a star to be awarded. Follow-up was considered adequate if above 75% (based on clinical consensus among the research team), or if there was comparison of the preceding psychological pathology in those who responded and those who did not (only possible in studies performing psychological assessment at two time points).

Risk of bias across studies

To avoid multiple publication bias, where different papers reported on different time points or types of assessment for the same study group, these were reported in a single tabulated row entry of the review. For example, Neugebauer’s group published four eligible articles on the same group of participants ( Neugebauer et al., 1992a, 1992b, 1997; Klier et al., 2000). Data from these were reported within one row, as one entry.

Results

Study selection

As shown in the PRISMA diagram (Fig. 1), 109 articles underwent full-text review, and 27 were included in the final review (comprising 21 separate entries in total, with 20 quantifying psychological response and one with conclusions relating to risk factors only). No relevant unpublished studies were obtained.

Study characteristics

Baseline characteristics are presented in Table I.

Baseline characteristics of included studies. (Studies reporting on the same cohort (with multiple publications for different time points) are combined as one entry.).

Abbreviations: A, anxiety; AR, assisted reproduction; BDI, Beck Depression Inventory; CES-D, Centre for Epidemiological Study Depression Scale; d, days; D, depression; D-S, Depression Scale (Depressivitats–Skala); DIS, Diagnostic Interview Schedule; DSM, Diagnostic and Statistical Manual of Mental Disorders; EMM, expectant management of miscarriage; EPU, early pregnancy unit; GHQ-12/28–12 or 28 item General Health Questionnaire; HADS, Hospital Anxiety and Depression Scale; IES, Impact of Events Scale (-R, research version); IVF, in vitro fetilisation; MMM, medical management of miscarriage; m, months; N, not available; NC, natural conception; P - partners; PDS, Post-traumatic Diagnostic Scale; PHQ-9, Patient Health Questionnaire; PSE, Present State Examination; SCID, Structured Clinical Assessment for DSM; SMM, surgical management of miscarriage; SASRQ, Stanford Acute Stress Reaction; SCL-90, Symptom Checklist 90; SCID, Structured Clinical Assessment for DSM; STAI, State Trait Anxiety Inventory; T, traumatic impact; ToP, termination of pregnancy; W, women; w, weeks; y, year.

a Using data provided by the author, three cases of stillbirth (≥24 w) were excluded. b Mean age, prior loss and children reflect the whole group.

Baseline characteristics of included studies. (Studies reporting on the same cohort (with multiple publications for different time points) are combined as one entry.).

Abbreviations: A, anxiety; AR, assisted reproduction; BDI, Beck Depression Inventory; CES-D, Centre for Epidemiological Study Depression Scale; d, days; D, depression; D-S, Depression Scale (Depressivitats–Skala); DIS, Diagnostic Interview Schedule; DSM, Diagnostic and Statistical Manual of Mental Disorders; EMM, expectant management of miscarriage; EPU, early pregnancy unit; GHQ-12/28–12 or 28 item General Health Questionnaire; HADS, Hospital Anxiety and Depression Scale; IES, Impact of Events Scale (-R, research version); IVF, in vitro fetilisation; MMM, medical management of miscarriage; m, months; N, not available; NC, natural conception; P - partners; PDS, Post-traumatic Diagnostic Scale; PHQ-9, Patient Health Questionnaire; PSE, Present State Examination; SCID, Structured Clinical Assessment for DSM; SMM, surgical management of miscarriage; SASRQ, Stanford Acute Stress Reaction; SCL-90, Symptom Checklist 90; SCID, Structured Clinical Assessment for DSM; STAI, State Trait Anxiety Inventory; T, traumatic impact; ToP, termination of pregnancy; W, women; w, weeks; y, year.

a Using data provided by the author, three cases of stillbirth (≥24 w) were excluded. b Mean age, prior loss and children reflect the whole group.

Setting

The 27 included articles were published between December 1989 and December 2016. Seven study groups originated from the UK, four were from Hong Kong, two each from Germany, the USA and the Netherlands, and one each from Sri Lanka, Norway, France and Ireland. One study was published in French (with an English abstract) and the rest were in English.

Participants

Details of participants in each study are provided in Fig. 2.

Numbers and types of participants in each paper included within qualitative synthesis, in reverse order of year of publication.

Eight out of 21 study groups had results from fewer than 100 participants with pregnancy loss. The smallest study had 22 participants ( Daly et al., 1996) and the largest 626 ( Cumming et al., 2007).

Comparison groups, where present, were variable: pregnant women, women or partners after live birth, women after termination, or non-pregnant women (contacted via random-digit telephoning or when looking for contraceptive advice).

In studies of women alone, all but one study ( Bowles et al., 2006) specified the proportion of women approached who agreed to participate; this ranged from 83% ( Friedman and Gath, 1989) to 97% ( Prettyman et al., 1993) in all but Nordal Broen’s study, in which the figure was 50%. This was proposed to be related to staff motivation ( Nordal Broen et al., 2005). In Cumming’s study, which required potential participants to return a consent form by post, an overall participation of 48% in women and partners was described ( Cumming et al., 2007). The participation of partners was lower in one study: 75% of women versus 30% of partners ( Kong et al., 2010).

Where participation was not defined by first response, response rates at first assessment in women varied from 61% (at 1 month ( Farren et al., 2016)) to 94% (at 1 week ( Prettyman et al., 1993)).

Outcomes

Depression was the most studied outcome, reported by all but two groups ( Bowles et al., 2006; Cheung et al., 2013).

Additional data (not included in the original article) was obtained from the study authors for two publications and presented ( Engelhard et al., 2001; Cumming et al., 2007).

Methods of assessment

Supplementary Table SIII summarizes the methods of assessment used.

The majority of groups used self-report measures. Five groups used a structured assessment by interview in addition to self-report measures; four used both in all participants ( Friedman and Gath, 1989; Lee et al., 1997; Neugebauer et al., 1997; Klier et al., 2000; Walker and Davidson, 2001), and the other performed interviews in those who scored ≥4 on the General Health Questionnaire (GHQ) ( Sham et al., 2010). Garel used only an interview ( Garel et al., 1996). Cumming compared outcome from the Hospital Anxiety and Depression Scale (HADS) with the Composite International Diagnostic Interview (CIDI) in 37 randomly selected participants, in order to confirm its validity in this context ( Cumming et al., 2007).

Cut-offs to define morbidity were variably defined by study groups, as detailed within Supplementary Table SIII .

Follow-up

The follow-up period ranged from 1 month ( Friedman and Gath, 1989; Bowles et al., 2006) to 5 years ( Nordal Broen et al., 2005). Nine of the study groups reported outcomes at 1 year or more, with only two studies reporting beyond 18 months.

Of those with follow-up of 3 months or more, one assessed for pregnancy at follow-up ( Cordle and Prettyman, 1994), one excluded women who were pregnant ( Garel et al., 1996), and two assessed for the impact of future pregnancies on morbidity ( Nordal Broen et al., 2005; Farren et al., 2016). The majority of studies did not assess for further pregnancies.

Suitability for meta-analysis

We considered the feasibility and value of meta-analysis.

It was not possible to obtain individual participant data for the vast majority of studies. Consideration was given to using aggregate data from those studies with a comparison group. However, there is significant clinical diversity, as can be seen in Table I, for example, in psychiatric history (two studies excluded women with any psychiatric history), parity (between 0 and 71% of women had children) and exposure (one study included late losses). The majority of data was derived from cohorts in Western Europe and the USA, with approximately one-fifth from Hong Kong. There has been discussion of the possibility of significant cultural differences in approach to EPL, which calls into question the appropriateness of any summary measures ( Garel et al., 1996; Sham et al., 2010). The data were also derived over a period of three decades.

There was also significant methodological diversity in terms of study design (timing of assessment, measure, comparison group used).

With such differences in both the subjects and the outcomes reported, it was deemed inappropriate to attempt to provide a meaningful summary estimate through meta-analysis.

Risk of bias within studies

Supplementary Table SIV details the quality assessment of each study using the Newcastle–Ottawa Scale. Quality issues relating to selection were noted as follows: three studies used a self-selected population ( Janssen et al., 1996; Engelhard et al., 2001; Johnson and Baker, 2004) and three studies did not specify the derivation of their cohorts ( Daly et al., 1996; Cheung et al., 2013; Kulathilaka et al., 2016). Participant groups were generally said to be representative of the EPL population attending the units, but in two studies, at least 50% declined participation, without discussion of the representativeness of those who agreed ( Nordal Broen et al., 2005, 2006; Cumming et al., 2007). One study group able to collect data on those who declined showed that these women were significantly older, and both women and partners had been married for longer, had a lower level of formal education and a greater number of children, than those who had participated ( Kong et al., 2010; Lok et al., 2010). Partners were also more likely to take part if they reported planning/wanting the pregnancy. Beutel’s study of couples found that partners who participated were more frequently married, and their female partners were less likely to believe that they had desired the pregnancy less than they had ( Beutel et al., 1996).

Ascertainment of exposure to EPL was by hospital records in all but two studies ( Janssen et al., 1996; Engelhard et al., 2001).

Few studies recruited prior to the diagnosis of EPL, thus precluding proof that psychopathology was not present prior to exposure to EPL. Only one study, in partners ( Johnson and Baker, 2004), showed similar early pregnancy assessments in those whose partners went on to experience healthy pregnancies.

No studies matched exposed and non-exposed individuals in the design of the study (although in two studies, couples were recruited together, and thus matched in terms of exposure ( Beutel et al., 1996; Kong et al., 2010)). Studies frequently found the loss group was older or had more previous losses ( Beutel et al., 1995; Janssen et al., 1996). One study found a significantly higher rate of history of prior depression (42.4% vs. 18.8%) in the miscarriage group ( Beutel et al., 1995). Five study groups controlled for potential confounders ( Neugebauer et al., 1992a, 1992b; Janssen et al., 1996; Nordal Broen et al., 2005, 2006; Lok et al., 2010; Kulathilaka et al., 2016).

Outcome assessment was always by assessment for psychopathology performed exclusively for the purposes of the study (usually by self-assessment using standardized self-report measures). Follow-up duration was adequate (≥1 month) in all but two studies ( Friedman and Gath, 1989; Daly et al., 1996). Drop-out ranged from 6% over 18 months (involving volunteers responding to an advert ( Janssen et al., 1996)) to 61% over 3 months (in a study with consecutive recruitment from an early pregnancy unit ( Farren et al., 2016)). Most studies assessed for a relationship between drop-out from first assessment and baseline psychometric scores. Some found no evidence of selective drop-out on this basis ( Thapar and Thapar, 1992; Prettyman et al., 1993; Beutel et al., 1995; Lee et al., 1997; Lok et al., 2010). Other studies suggested that those most affected may drop-out ( Janssen et al., 1996; Engelhard et al., 2001; Farren et al., 2016). Contrastingly, one French study found that those with more distress initially were more likely to take part in the 3-month assessment ( Garel et al., 1996).

Psychological morbidity: depression, anxiety and PTSD

Tables II and III detail psychological morbidity according to the time point of assessment. Table II summarises studies without comparison groups and Table III studies with comparison groups.

Cohort studies that have applied standardized psychometric questionnaires or diagnostic interviews at set time points after a miscarriage to assess the proportion of participants reaching a threshold for psychological morbidity (depression (D), anxiety (A) or measures of acute stress disorder (ASD), or post-traumatic stress disorder (PTSD)).

Abbreviations: ASD, Acute Stress Disorder; A, anxiety; BDI, Beck Depression Inventory; D, Depression; PSE, Present State Examination; PTSD, post-traumatic stress disorder; SCID, Structured Clinical Assessment for DSM.

a Data allowing exclusion of three losses ≥24 weeks gestation provided by author and presented here.

b Unable to extract all relevant data; unable to contact authors, or authors unable to provide raw data.

Cohort studies that have applied standardized psychometric questionnaires or diagnostic interviews at set time points after a miscarriage to assess the proportion of participants reaching a threshold for psychological morbidity (depression (D), anxiety (A) or measures of acute stress disorder (ASD), or post-traumatic stress disorder (PTSD)).

Abbreviations: ASD, Acute Stress Disorder; A, anxiety; BDI, Beck Depression Inventory; D, Depression; PSE, Present State Examination; PTSD, post-traumatic stress disorder; SCID, Structured Clinical Assessment for DSM.

a Data allowing exclusion of three losses ≥24 weeks gestation provided by author and presented here.

b Unable to extract all relevant data; unable to contact authors, or authors unable to provide raw data.

Cohort studies that have applied standardized psychometric questionnaires or diagnostic interviews at set time points after a miscarriage or ectopic pregnancy to assess psychological morbidity (Depression (D), Anxiety (A) or measures of acute stress, post-traumatic stress or traumatic impact (T)): caseness of disorders (given to 0 decimal places), and difference in caseness (C) or scores (S) compared to comparison group (with P-value, if available: results in bold).

Abbreviations: A, anxiety; ASD, acute stress disorder; BDI, Beck Depression Inventory; C, caseness; CoC, community control; D, depression; DSM, Diagnostic and Statistical Manual of Mental Disorders; DIS, Diagnostic Interview Schedule; gest., gestation; GHQ-28–28 item general health questionnaire (GHQ–D, depression score; GHQ-A, anxiety and insomnia score); HADS, Hospital Anxiety and Depression Scale; LbC, live birth controls; n/a, not available; NOS, not otherwise specified; OR, odds ratio; P, partners; PrC, pregnant control; PDS, Post-traumatic Diagnostic Scale; P, partners; PSE, Present State Examination; RR, relative risk; S, scores; SASRQ, Stanford Acute Stress Reaction; SCID, Structured Clinical Assessment for DSM; T, traumatic impact (using the Impact of Events Scale); ToP, termination of pregnancy; vs., versus; W, women.

a Unable to extract all relevant data; unable to contact authors, or authors unable to provide raw data.

b Results controlled for potential confounders.

c Raw data requested and presented here.

d Lower cut-off used to define caseness (using BDI) than elsewhere – see Supplementary Table S3.

Cohort studies that have applied standardized psychometric questionnaires or diagnostic interviews at set time points after a miscarriage or ectopic pregnancy to assess psychological morbidity (Depression (D), Anxiety (A) or measures of acute stress, post-traumatic stress or traumatic impact (T)): caseness of disorders (given to 0 decimal places), and difference in caseness (C) or scores (S) compared to comparison group (with P-value, if available: results in bold).

Abbreviations: A, anxiety; ASD, acute stress disorder; BDI, Beck Depression Inventory; C, caseness; CoC, community control; D, depression; DSM, Diagnostic and Statistical Manual of Mental Disorders; DIS, Diagnostic Interview Schedule; gest., gestation; GHQ-28–28 item general health questionnaire (GHQ–D, depression score; GHQ-A, anxiety and insomnia score); HADS, Hospital Anxiety and Depression Scale; LbC, live birth controls; n/a, not available; NOS, not otherwise specified; OR, odds ratio; P, partners; PrC, pregnant control; PDS, Post-traumatic Diagnostic Scale; P, partners; PSE, Present State Examination; RR, relative risk; S, scores; SASRQ, Stanford Acute Stress Reaction; SCID, Structured Clinical Assessment for DSM; T, traumatic impact (using the Impact of Events Scale); ToP, termination of pregnancy; vs., versus; W, women.

a Unable to extract all relevant data; unable to contact authors, or authors unable to provide raw data.

b Results controlled for potential confounders.

c Raw data requested and presented here.

d Lower cut-off used to define caseness (using BDI) than elsewhere – see Supplementary Table S3.

Depression

Within the first 2 weeks after a miscarriage, studies have shown 22% ( Prettyman et al., 1993), 28% ( Nordal Broen et al., 2005) and 36% ( Neugebauer et al., 1992a, 1992b) of women reaching a threshold for depression. Studies have shown significantly higher depression scores ( Thapar and Thapar, 1992; Beutel et al., 1995) and a higher proportion of cases, with relative risks of meeting criteria between 3.1 ( Lok et al., 2010) and 4.3 ( Neugebauer et al., 1992a, 1992b) times that in a non-pregnant comparison group. No significant difference was found between women following miscarriage and termination ( Nordal Broen et al., 2005). In partners, one study in Hong Kong showed a 17% prevalence of probable depression ( Kong et al., 2010), with significantly lower depression scores, but not the proportion meeting criteria, than in the women ( Beutel et al., 1996; Kong et al., 2010), and no difference in scores compared to partners following live birth ( Johnson and Baker, 2004).

Between one and two and a half months after loss, eight study groups published rates between 8% and 20% of women with EPL meeting a defined cut-off for moderate depressive symptoms ( Friedman and Gath, 1989; Neugebauer et al., 1992a, 1992b; Prettyman et al., 1993; Lee et al., 1997; Engelhard et al., 2001; Cumming et al., 2007; Farren et al., 2016; Kulathilaka et al., 2016). A higher proportion of 35% was found in a study utilizing volunteers and in which 10% were >20 weeks gestation ( Janssen et al., 1996). In studies utilizing a comparison group, GHQ depression subscale (but not HADS-depression) and SCL-90 depression subscale scores were significantly higher than in the comparison group ( Thapar and Thapar, 1992; Janssen et al., 1996). A subgroup of women interviewed for the first time at 6 weeks (excluding those also interviewed at 2 weeks) were 2.6 (95% CI 1.5–4.4) times more likely than non-pregnant controls to be highly symptomatic for depression ( Neugebauer et al., 1992a, 1992b). In a Sri Lankan study, after controlling for gestation and age, the higher proportion of cases in women between 6 and 10 weeks after losses versus pregnant controls was no longer significant ( Kulathilaka et al., 2016). Of partners, 4% met a threshold for moderate depression at 1 month, with lower scores compared to women ( Cumming et al., 2007).

Assessed at 3 months post-miscarriage, one French study found that 51% women had experienced a major depressive episode, although they noted selective drop-out of those less affected ( Garel et al., 1996).

Overall, depression caseness has been found to decline between consecutive assessments in women (for example, from 10% at 1 month to 3% at 6 months ( Cumming et al., 2007), and 27% immediately to 19% at 3 months ( Lok et al., 2010)). Three studies found results comparable to those without EPL by 1 year ( Janssen et al., 1996; Nordal Broen et al., 2005; Cumming et al., 2007). One study suggesting increased pathology at 1 year reported a significantly higher prevalence of psychiatric history in the miscarriage group as compared to the non-pregnant comparison group ( Beutel et al., 1995). Depression scores in partners also declined over time ( Beutel et al., 1996; Cumming et al., 2007; Lok et al., 2010). One study comparing partners a year after miscarriage and live birth showed higher depression scores in the former ( Johnson and Baker, 2004).

Anxiety

Studies looking at both anxiety and depression using HADS show anxiety to be the more frequent and sustained morbidity ( Thapar and Thapar, 1992; Prettyman et al., 1993; Cordle and Prettyman, 1994; Nordal Broen et al., 2006; Cumming et al., 2007; Farren et al., 2016). Immediately after loss, 41% participants met criteria for anxiety ( Prettyman et al., 1993), with scores significantly higher than in an antenatal comparison group ( Thapar and Thapar, 1992). The proportion of women meeting criteria for anxiety was significantly higher than in the control group at 1 month in the author’s own study ( Farren et al., 2016). Scores declined over time in all except one study, which found a further increase in anxiety scores at 12 weeks, which they hypothesized related to resumed efforts to conceive ( Prettyman et al., 1993). Scores were found to normalize compared to non-pregnant community participants at 6 months in two studies ( Beutel et al., 1995; Nordal Broen et al., 2005, 2006), but not until 12 months in Janssen’s study when compared to women at an equivalent time point after delivery ( Janssen et al., 1996).

Sham’s study, in which Chinese women were assessed in person at 3 months post-miscarriage, showed a lower prevalence of only three patients (1.8%) with anxiety disorders (including one with PTSD) ( Sham et al., 2010).

Partners showed significantly ower anxiety scores than women at 1-, 6- and 13-month assessments ( Cumming et al., 2007). Partners of women with miscarriage also showed no difference in anxiety scores to those of women with live births, either within 3 weeks or at 1 year post-loss or birth ( Johnson and Baker, 2004).

PTSD and traumatic impact

The largest study of PTSD following miscarriage showed a 25% incidence at 1 month, and 6% incidence at 4 months ( Engelhard et al., 2001). The author’s own study group, in a pilot study using more lenient criteria for diagnosis, found 38% of women with EPL met criteria for PTSD at 3 months (45% in miscarriage, 18% in ectopic pregnancy) ( Farren et al., 2016). A small study (25 women) by Bowles found a 39% incidence at 1 month after miscarriage ( Bowles et al., 2006). They also reported a 28% incidence of Acute Stress Disorder (ASD) (often seen as a precursor to PTSD) at 7 days. Another small study reported an incidence of ASD of 15% at three weeks ( Walker and Davidson, 2001). A study of 161 women 3 months after miscarriage in Hong Kong using structured clinical assessment found only one patient with PTSD (0.6%) and three with adjustment disorder (1.8%) on interview ( Sham et al., 2010).

Two other studies reported on Impact of Event Scale (IES) scores. The IES is a scale of subjective distress related to a specific event, with questions reflecting PTSD criteria, although not diagnostic of the condition. Nordal Broen compared IES scores following miscarriage to termination, and found significantly higher intrusion scores in the miscarriage group initially (with 47.5% women meeting criteria for intrusion at 10 days) and lower avoidance scores at later time points (2 and 5 years) ( Nordal Broen et al., 2005). Another study looking only at partners, comparing those following miscarriage to those after live birth, showed higher overall IES scores in the miscarriage group at 3 weeks and 1 year ( Johnson and Baker, 2004). The clinical significance of these differences in scores was not discussed.

Correlation between outcomes

In studies reporting on more than one outcome, a minority reported assessing for correlation between outcomes. Prettyman reported a statistically significant positive correlation between higher depression and anxiety scores ( Prettyman et al., 1993). Engelhard found that those meeting criteria for PTSD were significantly more likely to meet criteria for depression ( Engelhard et al., 2001). Johnson and Baker reported correlation between scores on the Beck Depression Inventory, State Trait Anxiety Inventory and IES in partners ( Johnson and Baker, 2004).

Factors associated with psychological morbidity

Although a minority of studies was sufficiently powered to assess risk factors for psychological morbidity, we have summarized the findings showing a significant association in Table IV. Generally, in studies with more than one assessment, factors were correlated to levels of distress at the earliest assessment.

Factors identified to have a statistically significant (at an alpha of 0.05) association with higher outcome measures of anxiety, depression or traumatic impact (scores (S), or caseness (C)) in women following miscarriage.

• Sham et al. (2010) SCID D C at 3 m (P = 0.01) a
• Cumming et al. (2007)HADS-D S at 1, 3 and 13 m

• Beutel et al. (1996) STAI S at 1w (P ≤ 0.05)
• Cordle and Prettyman (1994) HADS-A C at 2 yr (P< 0.001)

• Beutel et al. (1996) D-S S at 1w (P ≤ 0.05)
• Lee et al. (1997) SCID C at 6w (P = 0.02)
• Garel et al. (1996) major depressive episode at 3 m (P< 0.05)

• Engelhard et al. (2001) BDI C at 1 m (P< 0.01)
• Friedman and Gath (1989)PSE C 1 m (P< 0.005)

• Janssen et al. (1996) SCL-90 D S at 2.5 m (P = 0.004) & 1 yr (P = 0.002) b
• Engelhard et al. (2001)BDI S at 1 m (P< 0.01) b

• Sham et al. (2010) SCID D C at 3 m (P = 0.01) a
• Cumming et al. (2007)HADS-D S at 1, 3 and 13 m

• Beutel et al. (1996) STAI S at 1w (P ≤ 0.05)
• Cordle and Prettyman (1994) HADS-A C at 2 yr (P< 0.001)

• Beutel et al. (1996) D-S S at 1w (P ≤ 0.05)
• Lee et al. (1997) SCID C at 6w (P = 0.02)
• Garel et al. (1996) major depressive episode at 3 m (P< 0.05)

• Engelhard et al. (2001) BDI C at 1 m (P< 0.01)
• Friedman and Gath (1989)PSE C 1 m (P< 0.005)

• Janssen et al. (1996) SCL-90 D S at 2.5 m (P = 0.004) & 1 yr (P = 0.002) b
• Engelhard et al. (2001)BDI S at 1 m (P< 0.01) b

Abbreviations: A, anxiety; ASD, Acute Stress Disorder; BDI, Beck Depression Inventory; C, caseness; CES-D, Centre for Epidemiologic Studies Studies Depression scale; D, depression; HADS, Hospital Anxiety and Depression Scale (–A, anxiety score, -D, depression score); h, hours; IES, Impact of Events Scale (-R, revised); m, months; PSE, Present State Examination; PDS, Post-traumatic Diagnostic Scale; PTSD, post-traumatic stress disorder; S, scores; SASRQ, Stanford Acute Stress Reaction Questionnaire; SCID, Structured Clinical Interview for DSM; STAI, State Trait Anxiety Inventory; SCL-90, Symptom Checklist 90; w, weeks; yr, years.

a Multivariate analysis.

b Studies including late losses (stillbirth).

Factors identified to have a statistically significant (at an alpha of 0.05) association with higher outcome measures of anxiety, depression or traumatic impact (scores (S), or caseness (C)) in women following miscarriage.

• Sham et al. (2010) SCID D C at 3 m (P = 0.01) a
• Cumming et al. (2007)HADS-D S at 1, 3 and 13 m

• Beutel et al. (1996) STAI S at 1w (P ≤ 0.05)
• Cordle and Prettyman (1994) HADS-A C at 2 yr (P< 0.001)

• Beutel et al. (1996) D-S S at 1w (P ≤ 0.05)
• Lee et al. (1997) SCID C at 6w (P = 0.02)
• Garel et al. (1996) major depressive episode at 3 m (P< 0.05)

• Engelhard et al. (2001) BDI C at 1 m (P< 0.01)
• Friedman and Gath (1989)PSE C 1 m (P< 0.005)

• Janssen et al. (1996) SCL-90 D S at 2.5 m (P = 0.004) & 1 yr (P = 0.002) b
• Engelhard et al. (2001)BDI S at 1 m (P< 0.01) b

• Sham et al. (2010) SCID D C at 3 m (P = 0.01) a
• Cumming et al. (2007)HADS-D S at 1, 3 and 13 m

• Beutel et al. (1996) STAI S at 1w (P ≤ 0.05)
• Cordle and Prettyman (1994) HADS-A C at 2 yr (P< 0.001)

• Beutel et al. (1996) D-S S at 1w (P ≤ 0.05)
• Lee et al. (1997) SCID C at 6w (P = 0.02)
• Garel et al. (1996) major depressive episode at 3 m (P< 0.05)

• Engelhard et al. (2001) BDI C at 1 m (P< 0.01)
• Friedman and Gath (1989)PSE C 1 m (P< 0.005)

• Janssen et al. (1996) SCL-90 D S at 2.5 m (P = 0.004) & 1 yr (P = 0.002) b
• Engelhard et al. (2001)BDI S at 1 m (P< 0.01) b

Abbreviations: A, anxiety; ASD, Acute Stress Disorder; BDI, Beck Depression Inventory; C, caseness; CES-D, Centre for Epidemiologic Studies Studies Depression scale; D, depression; HADS, Hospital Anxiety and Depression Scale (–A, anxiety score, -D, depression score); h, hours; IES, Impact of Events Scale (-R, revised); m, months; PSE, Present State Examination; PDS, Post-traumatic Diagnostic Scale; PTSD, post-traumatic stress disorder; S, scores; SASRQ, Stanford Acute Stress Reaction Questionnaire; SCID, Structured Clinical Interview for DSM; STAI, State Trait Anxiety Inventory; SCL-90, Symptom Checklist 90; w, weeks; yr, years.

a Multivariate analysis.

b Studies including late losses (stillbirth).

One uncontrolled study found a significant protective impact of older age on depression caseness ( Sham et al., 2010). Another study demonstrated that women with no children or previous miscarriage have significantly higher anxiety scores immediately after loss ( Thapar and Thapar, 1992). Women have been found to be more at risk of depression if they have fewer existing children, if they have a history of infertility, or if they have had previous miscarriage ( Thapar and Thapar, 1992; Neugebauer et al., 1992b; Engelhard et al., 2001; Sham et al., 2010).

A longer gestation has been related to higher anxiety, depression and PDS scores (for depression scores, this relationship was found to persist at 1 year) ( Janssen et al., 1996; Engelhard et al., 2001). A pregnancy achieved through assisted reproduction was associated with higher IES avoidance and hyperarousal scores ( Cheung et al., 2013). Having previously seen a viable foetus on ultrasound increased the likelihood of depression caseness in women ( Kong et al., 2010).

In partners, a planned pregnancy was associated with a high prevalence of depression caseness in one study ( Kong et al., 2010). By contrast, in women, an unplanned pregnancy has been associated with more cases of anxiety at 12 weeks, and higher depression and IES scores immediately ( Thapar and Thapar, 1992; Prettyman et al., 1993; Walker and Davidson, 2001).

Being single at the time of the loss has been related to higher levels of depression shortly afterwards, and both anxiety and depression at 5 years ( Friedman and Gath, 1989; Nordal Broen et al., 2006). Measures of lower marital satisfaction or higher marital discord have been associated with increased caseness of depression (measured simultaneously) ( Lee et al., 1997; Kong et al., 2010). Lower perceived spousal support at the time of EPL has been associated with both anxiety and depression scores (assessed simultaneously, both soon after loss ( Beutel et al., 1996; Lee et al., 1997) and retrospectively at 5 years ( Cordle and Prettyman, 1994)).

Past psychiatric history (or self-reported poor psychiatric health prior to the EPL) has been found to be significantly associated with both anxiety and depression after EPL ( Friedman and Gath, 1989; Walker and Davidson, 2001; Nordal Broen et al., 2006; Cumming et al., 2007; Sham et al., 2010).

A history of abuse, as well as reporting feeling personally responsible for the loss or a lack of control over one’s life, has been associated with increased acute stress symptoms at 7 days ( Bowles et al., 2006).

Only two studies assessed for the impact of further pregnancies in the follow-up period. Neither reported any impact, although were underpowered to do so ( Nordal Broen et al., 2006; Farren et al., 2016).

Risk of bias across studies

Across studies, a large proportion declined participation (up to 52.5% ( Cumming et al., 2007)) or dropped out (up to 39% at first assessment ( Farren et al., 2016)). Only one study group collected background sociodemographic and clinical data on those who declined participation, and it did find some differences as detailed above ( Lok et al., 2010). In the author’s own study, collection of data on those who declined participation was not ethically permitted: it is likely that other studies were similarly constrained. The unassessed group may have different psychological outcomes: either more or less affected. Participation was lower in partners: 22% versus 75% ( Kong et al., 2010) and 35% versus 50% ( Cumming et al., 2007), and thus the potential for bias higher.

There may also be concern of unintended therapeutic effect of study involvement across studies. This was specifically assessed by Neugebauer’s team ( Neugebauer et al., 1992b), who found evidence in support of this, but it not considered in other studies. Care should thus be taken in the interpretation of repeated assessments, in which resolution of symptoms may rely on the impact of the assessment rather than time itself.

A minority relied on interviews as opposed to written assessments, generally excluding non-native language speakers or individuals with low literacy. However, some authors discussed potential ‘conscious non-disclosure of genuine symptoms’ due to a fear of stigmatization during interviews ( Sham et al., 2010).

Discussion

Summary of evidence

Studies performed over three decades, in a variety of locations and involving over 2500 women with EPL, have consistently demonstrated an association between pregnancy loss and psychological morbidity. We found that four to six weeks after loss, studies reported between 8% and 20% of women experiencing symptoms above a threshold for moderate depression. Anxiety appeared to have a prevalence of between 18% and 32%. For both conditions, resolution to background rates seems complete within 1 year. Partners show a lower intensity of depressive and anxiety reactions initially: for anxiety, this difference seems to be sustained for at least a year. Three studies reported incidences of between 25% and 39% of women reaching a threshold for the diagnosis of PTSD at 1 month. The duration and degree of resolution of post-traumatic stress is not clear.

An unselected EPL group appears to have higher pre-loss psychological morbidity (which may relate to previous loss or subfertility), but there is evidence from controlled studies that post-loss morbidity cannot be explained by baseline characteristics alone.

The most consistently identified associated factors relate to past psychiatric history and low marital satisfaction or support. An unplanned pregnancy might be associated with a higher risk of anxiety, depression and PTSD in women. Other identified factors associated with anxiety and depression include being single, having no children or having had a previous miscarriage. Longer gestation appeared to be a risk factor in studies including later losses.

Limitations

There is variability in rates of disorders across studies. This is most marked for PTSD at 3 months: one study using a clinician administered scale found only one case (a rate of 0.6%) ( Sham et al., 2010); another, using the PDS, found a rate of 7% ( Engelhard et al., 2001); and, in the author’s own pilot study, a rate of 39% was found ( Farren et al., 2016). The difference between the latter two studies may in part relate to different recruitment and scoring methods used, but variation also brings into question whether research conducted in different geographical locations, with different demographics and cultural norms, can be appropriately and usefully interpreted in other populations.

There is also limited concordance across studies with regard to risk factors identified. This is likely to reflect the small size of the majority of the studies, thus limiting power to detect statistically significant factors. The absence of consistent relationships of morbidity to background factors could also suggest that better elucidation and stratification of groups at risk is necessary, for which a study assessing all three outcomes (anxiety, depression and PTSD) would be helpful. Different groups may be at risk of different disorders and would merit different management.

There is evidence for selective participation and retention of participants across studies, both under-representation of those most affected (who may avoid the questionnaires as upsetting reminders of the event) or least affected (who may be less motivated to take part). In other literature, there is evidence for selective non-participation by those likely to be most affected by industrial disasters and termination of pregnancy ( Weisaeth, 1989; Soderberg et al., 1998). While unavoidable, this introduces an important selection bias throughout this body of research.

Factors associated with psychological morbidity at first assessment (soon after the EPL) have generally been identified. However, arguably, more important are risk factors for longer term morbidity, which cannot be assumed to be the same. Furthermore, there has been limited investigation with regard to risk factors for PTSD. An understanding of this will be especially important in targeting appropriate treatment.

A pertinent question to healthcare professionals and researchers working in early pregnancy is whether management strategy may affect emotional response. Only one reported assessing for the impact of management of the miscarriage ( Sham et al., 2010). In this study, a very small minority had medical management, and there was no difference in the proportions of women having conservative or surgical management in the groups with or without depressive disorders. The impact of management is more appropriately assessed in interventional studies, where women are randomized to conservative, medical and surgical approaches. However, there is 2-fold difficulty in performing these trials. First, many women will have a clear preference, precluding randomization in a substantial (and potentially characteristically different) group of women. Indeed, Wieringa-de Ward’s analysis of their substantial non-randomized group showed those who opted for expectant management had better mental health scores at the outset ( Wieringa-De Waard et al., 2002). Second, many of those randomized will deviate from their allocated management strategy, for clinical reasons or newly demonstrated preferences. At present, there is no evidence to guide management except to support a woman’s preference.

Fundamental to research of this nature is a reliance on self-report measures, which use arbitrary cut-offs to define pathology that have not been validated in EPL populations. Furthermore, the diagnoses of depression, anxiety and PTSD, while helpful in defining an under-recognized and unmet need in the EPL population, may be of limited usefulness in terms of understanding the exact nature of the distress. For example, anxiety may be generalized, specifically associated with concerns about involuntary childlessness (in which case having further children may be expected to resolve it), or focused on the trauma of the event (i.e. adjustment disorder or PTSD).

Studies without a comparison group enable quantification of the overall psychological pathology in the population with EPL, but cannot prove association with the loss itself. The risk of miscarriage is related to age, previous miscarriage and previous subfertility ( Maconochie et al., 2007). As such, an EPL population is likely to be older and to have had more miscarriages or higher rates of antecedent subfertility than a comparison group outside or within healthy pregnancy, all of which may themselves be related to psychological morbidity. Comparability between the cohorts was a consistent issue in the included studies, most studies commented that comparison groups were broadly similar in terms of age, past EPL, and number of children, but did not control from them. Two study groups retained statistically significant results after controlling for possible confounders ( Neugebauer et al., 1992b, Lok et al., 2010).

It is also important to consider that the psychological conditions seen may relate to the uncertain period of involuntary subfertility or childlessness while waiting to conceive again. In support of this is evidence (though not demonstrated in all studies) that existing children are protective from depression after miscarriage ( Thapar and Thapar, 1992a; Neugebauer et al., 1997). One would also expect improvement with future successful procreation, although there is yet no evidence of this ( Nordal Broen et al., 2006). There is, however, sizeable evidence included in this review that those who conceived without delay or planning experience notable distress after a miscarriage, supporting a notion that the loss itself is at least partially causative.

The significant decline in anxiety and depression scores over time may also be pertinent in ascribing morbidity to the EPL itself, rather than solely to any predisposing factors ( Neugebauer et al., 1992a; Prettyman et al., 1993; Beutel et al., 1995; Janssen et al., 1996; Cumming et al., 2007; Lok et al., 2010). However, hampering interpretation of this is the fact that few studies with prolonged follow-up have assessed for further pregnancies (healthy or otherwise). In order to fully understand the natural trajectory of psychological sequelae after a loss, it is necessary to separate out women who are newly pregnant or have experienced a further loss.

Although this review supports a strong association between EPL and psychopathology, categorical proof of the direction of causation is lacking. This would require pre-loss psychological assessment. Only one study in women was methodologically capable of this ( Janssen et al., 1996), and this study found higher depression and anxiety scores in the group who subsequently lost the pregnancy. This was incompletely explained by the higher rate of past loss in this group. It is possible that symptoms such as bleeding or pain led this group to expect a poor outcome prior to diagnosis. However, the possibility that psychopathology itself could (directly or indirectly) increase risk of EPL cannot be excluded (available literature both supports ( Sugiura-Ogasawara et al., 2002) and refutes ( Nelson et al., 2003; Kolte et al., 2015) this). Overall, however, even if causation cannot be proven, the body of research presented identifies a group at high risk of psychological morbidity.

Conclusions

Women, and to a lesser extent, their partners, are at significant risk of symptoms of anxiety, depression and PTSD after an EPL. Clinicians should be especially alert to this risk in women who have a history of psychiatric illness, women who have less support, women who have had multiple miscarriages, and women who have experienced subfertility and/or are left childless by the loss.

It is important to remember that reactions are diverse. Amongst the distributions of responses to EPL will be those who are unaffected and those who are completely debilitated. Different groups of women may be at risk of different disorders and beneath the heading of one ‘disorder’ may be a variety of reactions with different predispositions, patterns of progression or resolution, and response to treatments. A priority for future research is to establish what types of emotional support are appropriate and efficacious ( Prior et al., 2017). Evidence on the benefits of general counselling in an unselected population has so far been disappointing ( Murphy et al., 2012). Larger studies, and studies assessing all outcomes, will help separate and define groups at risk, as well as the impact of our clinical management, and may be the first step in individualizing treatment and seeing benefit.

While awaiting such research, clinicians should be sensitive to the risk of psychopathology in this population. We suggest enquiring about emotional well-being at any routine follow-up and encouraging those who have experienced EPL to contact their general practitioner if they experience severe or prolonged psychological symptoms. A policy for systematic screening of these women should be considered and evaluated.

Acknowledgements

The research team would like to thank the researchers who responded to queries and provided additional data for this review. We would also like to thank Jacqueline Cousins and the staff at Imperial College library for their support in performing a systematic search procedure and obtaining original texts.

Authors’ roles

JF originally proposed the review. All authors contributed to the protocol design during research meetings, during which consensus was reached for any discrepancies found during screening or data extraction. JF performed the initial literature searches, and JF and NMJ assessed the results of these searches against inclusion criteria. JF extracted data and made the initial assessment of study quality, with advice from JV, which was checked by NMJ. JF and TB wrote the first draft of the manuscript, which was then critically reviewed and revised by JV, MJ, DT and NMJ. All authors approved the final version of the manuscript for submission.

Funding

JF was supported by Imperial Health Charity grant number 141517. TB is supported by the National Institute for Health Research (NIHR) Biomedical Research Centre based at Imperial College Healthcare NHS Trust and Imperial College London. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR or the Department of Health. The Tommy’s National Centre for Miscarriage Research at Imperial College NHS Trust is supported by the Tommy’s charity.

Conflict of interest

The authors do not have any conflicts of interest.