Main findings

Using Read codes and prescriptions in the electronic GP record, the estimated prevalence of common mental disorders for the pre-birth period was 9.5% (13.1% in the first postnatal year, 12.8% in the second, and 14.0% in the third). White British women had twice the prevalence and incidence of psychiatric morbidity compared with minority ethnic women. Based on a self-reported screening scale with scores measuring distress at levels likely to indicate caseness our estimate of the overall proportion of missed cases of pre-birth common mental disorders was between 31.3% (95% CI 28.7–33.8) and 46.8% (95% CI 44.7–49.0). Minority ethnic women had twice the rate of potentially missed cases and half the volume of screening records. Across all ethnic groups, lower SES was associated with pre-birth morbidity, whether detected or not. Detection of this morbidity was associated with screening, at least for women using English, although low levels of screening preclude firm conclusions. Increased healthcare utilisation was more likely in women with detected common mental disorders compared with women with low risk. The relationship between healthcare utilisation for the women who were potentially missed compared with detected and low risk varied by ethnic group. Ethnic disparities in detection of significant clinical morbidity continued postnatally.

Strengths and limitations

BiB is a large cohort study thus rates could be estimated with some precision with results robust to a variety of methodological assumptions. Linkage between routine data and BiB avoids traditional attrition problems and we limited missing routine data by excluding women who moved. To minimise bias caused by variation in recording by GP, or time, we broadly defined potential disorders. ^{Reference Walters, Rait, Griffin, Buszewicz and Nazareth18} We consider it unlikely that more minority ethnic women have their mental health managed by health visitors and midwives outside the electronic record, and, regardless, GPs should be notified of suspected cases. ¹⁹ Our study has some limitations. The quantity of missing data from the primary care data-set is unknown. Morbidity or treatment noted by free-text or letter were not available, meaning that results could have been related to differences in free-text recording by GPs caring for either minority ethnic or majority women. An incomplete history of common mental disorders in the medical record and a lack of causal ordering meant we could not be certain which screening records were case-finding for the incident disorder and which were used for monitoring existing psychopathology. We acknowledge the limitations of the GHQ-28 only being administered once and could not find a report of its case-finding properties during pregnancy that would enable us to apply weights for anxiety and mixed episodes. NHS guidance documents advise screening maternal populations for depression, ^19,20 but only for anxiety after our study ended, ²⁰ hence ethnic differences in specific disorder prevalence may have affected our results. We did not aim to identify all women with potentially missed common mental disorders and our predictions contain unknown error quantities, although we presented weighted prevalence and believe we have under- rather than overestimated ethnic differences. The regression analyses may overstate relationships between covariates and these disorders because of multiple testing. It is difficult to quantify the impact of excluded cases because of missing ethnicity along with missing GHQ data. Pooling smaller ethnic groups may have masked important within-group differences in risk. ^{Reference Prady, Pickett, Croudace, Fairley, Bloor and Gilbody12} Our proxy marker for GP contact could be unreliable because of varying quantities of administrative codes. Finally, Bradford is a disadvantaged city with low socioeconomic diversity and it is unclear how findings might translate to more economically diverse areas, or general populations.

Comparison with findings from other studies

We adopted a general approach to morbidity (common mental disorders) in contrast to most research that focuses on either depression or anxiety because mixed episodes are common, ^{Reference Austin, Hadzi-Pavlovic, Priest, Reilly, Wilhelm and Saint21} single episodes are less reliably separated in the GP record using our coding criteria and sequelae for children are equally high. ^{Reference O'Connor, Heron and Glover8} We included descriptive terms and studied an economically disadvantaged community, a risk factor for common mental disorders. Thus, in the absence of underdetection, prevalence and incidence in our study should be comparatively high. However, our period prevalence estimates are low compared with studies in a review of maternal depression (not including anxiety) assessed using diagnostic criteria. ^{Reference Gaynes, Gavin, Meltzer-Brody, Lohr, Swinson and Gartlehner6} They are also low compared with a BiB subsample assessed in a depression diagnostic accuracy study, ^{Reference Mann, Adamson and Gilbody22} implicating common mental disorder under-recognition within the primary care setting. Studies using GP records typically report fewer cases than diagnostic studies because of underdetection and coding problems. ^{Reference Walters, Rait, Griffin, Buszewicz and Nazareth18,Reference Khan, Harrison and Rose23} Prevalence, however, in our study, still appears low compared with analyses of a primary care research data-set of child-bearing age women who had Read-coded anxiety diagnoses or symptoms, ^{Reference Martin-Merino, Ruigomez, Wallander, Johansson and Garcia-Rodriguez9} or depression diagnoses. ^{Reference Martin-Merino, Ruigomez, Johansson, Wallander and Garcia-Rodriguez24} Financial incentives available to GP practices for contributing to primary care research data-sets might be associated with more consistent coding, or possibly better rates of case-finding, than in the practices contributing to our study. Populations represented by practices contributing to primary care research data-sets may also be different to BiB (on average they will serve more advantaged populations), pathology in maternal populations may be different to non-maternal women or there may be pervasive coding differences during pregnancy across GP data. Incidence in our study is high compared with analyses of child-bearing age women in a primary care research data-set using Read codes to identify anxiety diagnoses or symptoms, ^{Reference Martin-Merino, Ruigomez, Wallander, Johansson and Garcia-Rodriguez9,Reference Walters, Rait, Griffin, Buszewicz and Nazareth18} or depression diagnoses. ^{Reference Martin-Merino, Ruigomez, Johansson, Wallander and Garcia-Rodriguez24} Although this could indicate the absence of a problem of absolute underdetection in our study, it could also be caused by underrecording of pregnancy prevalence, for example by midwives outside the GP record, which would falsely inflate future incidence. However, incidence in our study is low or comparable with an analysis of depression diagnoses and depression-related prescriptions in the first postnatal year. ^{Reference Dave, Petersen, Sherr and Nazareth25} This may reflect a genuinely higher risk of incident disorder in maternal populations, similar problems in both studies in accurately accounting for pre-birth prevalence, or the effect of including prescriptions to aid in case-finding. Differences in study populations and analytic and case-finding methodologies make comparisons about our study's findings and others' very challenging. Despite these comparative differences, our estimate of missed diagnoses does agree with other non-maternal and postnatal studies. ^{Reference Cepoiu, McCusker, Cole, Sewitch, Belzile and Ciampi1,Reference Hearn, Iliff, Jones, Kirby, Ormiston and Parr26} Although the evidence base is limited, ^{Reference Tarricone, Stivanello, Poggi, Castorini, Marseglia and Fantini27} there is little to suggest that Pakistani women have a lower common mental disorder burden, ^{Reference Bhui, Bhugra, Goldberg, Sauer and Tylee3,Reference Husain, Cruickshank, Husain, Khan, Tomenson and Rahman28–Reference Gater, Tomenson, Percival, Chaudhry, Waheed and Dunn30} thus, although cultural differences in responses to screening questionnaires may account for some variation, we consider our results to indicate health disparity rather than a genuine difference in prevalence.

Implications

We noted extremely low levels of screening overall, presumably because of non-electronic recording by midwives or health visitors. Our results, however, indicate that case-finding in disadvantaged women has the potential to be successful, and, if systematically and rigorously employed, might reduce detection disparities. The lower observed levels of screening for ethnic minorities could be related to the lack of culturally validated and non-English screening instruments, ^{Reference Downe, Butler and Hinder31} inhibiting case-finding activity during consultations. There may also be cultural sensitivities in asking about psychological symptoms, and patient reporting of symptoms, where there is potential stigma associated with such problems. Screening, case-finding and improved detection in itself does not improve mental health and needs to be accompanied by effective intervention. ^{Reference Gilbody, Sheldon and House32}

White British women who were potentially missed cases had greater healthcare utilisation than those with low common mental disorder risk, with borderline statistical significance for English-speaking Pakistani women. This possibly indicates higher levels of physical health problems and minor psychiatric diagnoses, ^{Reference Naylor, Parsonage, McDaid, Knapp, Fossey and Galea33} or increased visits for the same state of health. Indeed, across ethnic groups the level of healthcare utilisation was similar between potentially missed and detected cases. Ethnic minorities access primary care services at similar rates to majority populations, but may be less likely to consult GPs with a mental health concern, indicating that consultations are less effective for mental healthcare-seeking. ^{Reference Cooper, Bebbington, McManus, Meltzer, Stewart and Farrell2,Reference Gater, Tomenson, Percival, Chaudhry, Waheed and Dunn30,Reference Sashidharan, Commander, Shahid and Atkin34,Reference Nazroo35} We found that non-English language users were least likely to be identified with mental health problems. The effect of language in consultations will vary by practice and GP, but differences could be minimised by improved access to translators. Language could also be a proxy for more recent immigrants, reported to have better mental health, ^{Reference Nazroo36} prima facie observed as decreased prevalence in our study, but a reduced risk for mental disorder should not be assumed in any consultation. Disorder identification for any ethnic group is likely to be influenced by both patient and professional barriers, ^{Reference Saver, Van-Nguyen, Keppel and Doescher37} and GPs working in deprived inner cities need to beware normalisation of distress because of a high volume of pervasive ‘misery’. ^{Reference Chew-Graham, Mullin, May, Hedley and Cole38,Reference Barley, Murray, Walters and Tylee39} Recent and established immigrant populations may experience and express psychopathology somewhat differently, and variation in clinicians' cultural competency may result in health inequality. ^{Reference Weich, Nazroo, Sproston, McManus, Blanchard and Erens29,Reference Bhui, Bhugra, Goldberg, Dunn and Desai40,Reference Kleinman41}

Directions for future research

Our study has highlighted several areas for future research. Prospective longitudinal research is needed to understand how and why some women ‘fall through the gaps’ of continued care. The effects of potentially missed common mental disorders on children's outcomes need to be quantified. Studies into the content of health professional encounters with minority/majority and disadvantaged/advantaged women at risk for psychiatric disorder are needed, with a view to improving the quality of case-finding in routine visits and reducing disparities. Similar investigations into how screening is recorded would help unpack our unanswered question about whether the low levels of screening observed in this study are a result of limited screening, screening undertaken by other health professionals or variation in recording. This would provide evidence to feed into the assessment of robustness and generalisability of effectiveness and cost-effectiveness estimates for screening programmes. Culturally specific and cross-cultural screening instruments need validation, and interventions in their use and effect should be evaluated. More definitive research is required to clarify whether, in vulnerable maternal populations, less healthcare is sought or healthcare-seeking is less effective. Generally, causal studies into factors that predict identification and reduce inequality are required. Finally, routine clinical information can be a rich source of research data but our study highlights the potential for underreporting and masking of substantial health inequalities; greater understanding of potential bias in routine data sources is needed.