The role of diet and nutritional supplementation in perinatal depression: a systematic review

Abstract

This article presents a systematic literature review on whether dietary intake influences the risk for perinatal depression, i.e. depression during pregnancy or post‐partum. Such a link has been hypothesized given that certain nutrients are important in the neurotransmission system and pregnancy depletes essential nutrients. PubMed, EMBASE and CINAHL databases were searched for relevant articles until 30 May 2015. We included peer‐reviewed studies of any design that evaluated whether perinatal depression is related to dietary intake, which was defined as adherence to certain diets, food‐derived intake of essential nutrients or supplements. We identified 4808 studies, of which 35 fulfilled inclusion criteria: six randomized controlled trials, 12 cohort, one case‐control and 16 cross‐sectional studies, representing 88 051 distinct subjects. Studies were grouped into four main categories based on the analysis of dietary intake: adherence to dietary patterns (nine studies); full panel of essential nutrients (six studies); specific nutrients (including B vitamins, Vitamin D, calcium and zinc; eight studies); and intake of fish or polyunsaturated fatty acids (PUFAs; 12 studies). While 13 studies, including three PUFA supplementation trials, found no evidence of an association, 22 studies showed protective effects from healthy dietary patterns, multivitamin supplementation, fish and PUFA intake, calcium, Vitamin D, zinc and possibly selenium. Given the methodological limitations of existing studies and inconsistencies in findings across studies, the evidence on whether nutritional factors influence the risk of perinatal depression is still inconclusive. Further longitudinal studies are needed, with robust and consistent measurement of dietary intake and depressive symptoms, ideally starting before pregnancy.

Introduction

Perinatal depression, also referred to as maternal depression, is defined as depression during pregnancy and up to 1 year post‐partum. It is a common morbidity during pregnancy and lactation and can have severe and long‐term consequences for women and their children. A systematic review on the prevalence of perinatal depression reported that the pooled point prevalence for minor depression alone was 11% during pregnancy and 13% post‐partum (Gavin et al. 2005). In low‐income and middle‐income countries (LMICs), perinatal depression is thought to be more common because of concomitant stress, anxiety, familial and political instability and poverty (Shidhaye et al. 2013). Fisher et al. (2012) reported a prevalence of 15.6% of women having depressive symptoms during the antenatal period (from 13 studies) and 19.8% during the post‐partum period (from 34 studies) in LMICs.

There is growing concern about the consequences of perinatal depression (Howard et al. 2014b), with evidence accumulating of its negative effects on women's health and child development (Walker et al. 2007; Stein et al. 2014; Patel et al. 2004). Studies show that maternal depression is associated with poor self‐care and caregiving, poor child growth and cognitive development, child malnutrition and increased incidence of child illness (Stewart 2007; Bennett et al. 2004; Rahman et al. 2004; Stein et al. 2008). Although the detrimental effects of maternal depression are increasingly recognized and studied, the focus has been on treatment and intervention rather than on risk factors and prevention (Rahman et al. 2013).

It has been hypothesized that nutrition could play a biological role in depression (Kaplan et al. 2007). Various nutrients are needed for synthesis and modulation in the neurotransmission system and may therefore be involved in mood regulation (Rechenberg & Humphries 2013; Bodnar & Wisner 2005). The biochemistry of each nutrient and its role in mood regulation are discussed elsewhere (Kaplan et al. 2007; Rechenberg & Humphries 2013; Leung & Kaplan 2009). During pregnancy and lactation, nutritional demands increase, and deficiencies thus arise more easily. It is also possible that hormonal and life changes post‐partum increase the risk for depression (Burt & Stein 2002). Nonetheless, women may experience heightened risks from both nutritional deficiencies and hormonal shifts during the perinatal period.

Previous reviews on certain nutrients and their link to depression concentrate on one nutrient or class of nutrients, such as polyunsaturated fatty acids (PUFAs) (Freeman et al. 2006). Most considered depression in general and few had a focus on perinatal depression (Jans et al. 2010; Leung & Kaplan 2009; Makrides 2008). Reviews also often mix studies of prevention of depression (supplementation in non‐depressed subjects) and treatment of already depressed subjects with certain nutrients.

This review aims to synthesize the evidence on whether dietary intake influences the risk of depression in the perinatal period. Dietary intake is defined as adherence to certain diets, food‐derived intake of essential nutrients or supplementation, which encompasses complex interactions between different ingested nutrients and compounds. Given that intake of foods and supplements is not always indicative of bioavailability, blood levels of certain nutrients will be the focus of a separate review.

Key messages.

• Depression in pregnancy and post‐partum is common and can have adverse consequences for mother and child.

• During pregnancy and lactation, nutritional demand increases. Lack of key nutrients may negatively impact neurotransmission and increase the risk of perinatal depression.

• We conducted a systematic literature review and present a synthesis of the evidence on whether dietary intake influences perinatal depression.

• The evidence is still inconclusive, because of both mixed results of studies and a lack of high‐quality investigations.

• Further research is needed using robust, longitudinal measurement of nutritional exposures and depression outcomes, including low‐income settings.

Methods

Search strategy

We ran a sensitive and comprehensive search strategy using both MeSH terms and free text keywords in two separate searches that was designed to identify all relevant studies. The first search combined nutrition terms with maternal depression terms, while the second combined nutrition terms with depression terms and maternity terms separately (Fig. 1). PubMed, EMBASE and CINAHL databases were searched for English‐language publications from 1966 to 30 May 2015. Two independent reviewers screened all titles and abstracts identified by the search and then reviewed the full text of potentially eligible articles for inclusion.

Figure 1.

Search strategy.

Eligibility

Articles were eligible for inclusion if they were peer‐reviewed and presented measures of association between dietary intake before, during or after pregnancy and depression during pregnancy or up to 1 year post‐partum. Analytical studies of any design were eligible for inclusion.

We included studies that evaluated distinct dietary patterns and adherence to those diets or intake of specific nutrients through food or supplementation if they used validated food frequency questionnaires (FFQs) or supplement history questionnaires. We excluded studies using non‐specific intake questions to measure health behaviours, such as whether or not a woman took supplements but without a measure of specific amount or type. Studies using other measures such as food insecurity, body composition (e.g. body mass index), hormone levels or one‐item food questions to assess health behaviours were also excluded from this review.

Studies in women already diagnosed with depressive symptoms or women with underlying health conditions (HIV positive, for example) were excluded. Perinatal depression was required to be assessed through a validated depression screening tool or through clinical diagnosis from a trained interviewer. Studies using prescription of antidepressants as a proxy for a clinical diagnosis were also included. The Center for Epidemiological Studies Depression Scale (CES‐D) and Beck's Depression Inventory (BDI) are frequently used for depression screening, also in pregnancy and post‐partum, as they do not emphasize somatic symptoms such as appetite changes and sleep disturbance and have been found to be highly accurate in screening for both minor and major depression (Tandon et al. 2012). However, the only screening tools that totally exclude these symptoms and are technically validated for the antenatal (pregnancy) period and post‐partum period are the Edinburgh Post‐partum Depression Scale (EPDS) and the Post‐partum Depression Screening Scale (PDSS), which have been shown to have higher sensitivity than other tools (Gaynes et al. 2005; Beck & Gable 2001). Thus, we assigned a lower risk of bias to studies using these tools.

Risk of bias assessment

The methodological quality of the included studies was assessed using the Quality in Prognostic Studies tool (Hayden et al. 2013) for observational studies and the Cochrane Collaboration tool for assessing risk of bias in randomized control trials (RCTs) (Higgins et al. 2011). Although the Quality in Prognostic Studies tool was designed for prognostic studies, ‘prognoses’ are similar to ‘risks’ in epidemiology. The tool evaluates risk of bias in studies and comprises six domains: study participation, study attrition (which we replaced by response in cross‐sectional studies and by selection of controls in the one case–control study), prognostic factor (in this case exposure) measurement, outcome measurement, confounding, and statistical analysis and reporting.

The confounding domain was assessed based on potential confounders being controlled for and the method by which they were controlled. Several socioeconomic and psychosocial variables have shown relationships to both depression and self‐care or eating habits (and thus dietary intake), specifically, a history of depression, social support, poor general health, income and employment (Bennett et al. 2004). Studies that did not control for any of these confounders were given a high risk of bias, studies including some confounders, a medium risk of bias and studies that accounted for several of these confounders, including history of depression, a low risk of bias. For Vitamin D studies, measuring a proxy of sunlight exposure, a main source of Vitamin D, was required to be ranked as having low risk of bias. In most studies, confounding was considered in stepwise models, where a univariable analysis was conducted first to identify potential confounders. If important potential confounders were included in a crude model but not associated with the outcome, and the authors therefore did not include them in the final adjusted model, this was still considered sufficient control for confounding in the risk of bias analysis.

For statistical analysis and reporting, if studies used stepwise models including important potential confounders and found no association between the exposure(s) of interest and the outcome and therefore did not proceed to a fully adjusted multivariable model, then they were still given a low risk of bias in this category. However, if an adjusted model was not attempted despite significant associations in a crude model or the analysis strategy was not clear, then studies were ranked as having moderate or high risk of bias for statistical analysis and reporting.

The Cochrane tool for risk of bias in RCTs uses the following domains: random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias), incomplete outcome data (attrition bias) and selective reporting (reporting bias).

Two reviewers independently ranked each study as having high, medium or low risk of bias in each domain. The reviewers compared scores, disagreements were discussed and a consensus reached.

Data extraction

Study characteristics and overall results were extracted from all included studies using a standard data form and are presented in Table 1. The type of analysis performed and detailed results were also extracted and are shown in Appendices 1–4. As studies varied widely in their choice of nutritional exposures and timing of dietary intake assessment and depression assessment, performing a meta‐analysis of associations between dietary exposure and perinatal depression was not possible. Therefore, a descriptive summary of all included studies is presented.

Table 1.

Summary of studies evaluating associations between dietary intake and perinatal depression

Study characteristics					Exposure			Outcome		Results
Author (year)	Country	n	Study design	Depression prevalence	Nutrients/diet	Tool; recall/intervention period	Time assessed	Tool; cut‐point	Time assessed	Adj. model	Protective associations (unless noted; final models reported)
DIETARY PATTERNS
Chatzi et al. (2011)	Greece॑	519	Cohort	14%	PCA grouped into ‘Health conscious’ and ‘Western’	FFQ; pregnancy	Mid‐pregnancy	EPDS; 13	PP: 8–10 weeks	Y	High (vs. low) adherence to health conscious diet
Murakami et al. (2008)	Japan	865	Cohort	14%	GI and GL index	FFQ; 1 month recall	Pregnancy	EPDS; 9	PP: 2–9 months	Y	Third quartile (vs. first) of GI score
Okubo et al. (2011)	Japan	865	Cohort	14%	Factor analysis grouped into ‘Healthy’, ‘Western’ and ‘Japanese’	FFQ; 1 month recall	Pregnancy	EPDS; 9	PP: 2–9 months	Y	Second quartile (vs. first) of adherence to Western pattern
Pina‐Camacho et al. (2015)	UK†	7814	Cohort	Not reported	Factor analysis to determine level of unhealthy diet	FFQ: 1 week recall	Third trimester	EPDS; linear	Second, Third trimester PP: 8 weeks, 8 months	Y	None
Vilela et al. (2014)	Brazil	246	Cohort	Not reported	PCA grouped into ‘Common Brazilian’, ‘Healthy’ and ‘Processed’	FFQ; 6 months prior to pregnancy	First trimester	EPDS; linear	Each trimester	Y	High (vs. low) adherence to healthy dietary pattern before and during pregnancy
Fowles et al. (2011)	United States	50	Cross‐section	Not reported	Frequency of eating fast food	DQI‐P; 3 × 24 h recall	First trimester	EPDS; 10 or mean difference	First trimester	N	Low (vs. high) frequency of fast food consumption
Fowles et al. (2012)	United States	71	Cross‐section	Not reported	Dietary Quality Index during pregnancy score, high vs. low	DQI‐P; 3 × 24 h recall	First trimester	EPDS; mean difference	First trimester	N	High (vs. low) DQI‐P score
George et al. (2005)	United States	146	Cross‐section	Not reported	Compliance with US dietary guidelines	FFQ: 6 months recall	1 year PP	CES‐D; mean difference	PP: 1 year	N	Highest (vs. lowest) tertile of dietary compliance
Miyake et al. (2015b)	Japanℓ	1745	Cross‐section	19%	Intake of seaweed	FFQ; 1 month recall	Pregnancy	CES‐D; 16	Pregnancy	Y	High (vs. low) seaweed consumption
TOTAL NUTRIENT INTAKES
Paoletti et al. (2013)	Italy	552	RCT	14%	Group A: Vit A, B1, B2, B6, B12, B7, B9, C, D3, E, Ca, Fe, Mg, Mn, Cu, P and Zn Group B: calcium and Vit D3	Randomized; pregnancy	N/A	EPDS; 12	PP: 3, 15 and 30 days	N	Full multivitamin supplementation (group A) vs. calcium and Vit D3 (group B)
Leung et al. (2013)	Canada°	475	Cohort	12%	Intake of Vit B1, B3, B6, B9, B12, D, iodine, Fe, Mg, Se, Zn and n‐3 PUFA	Supplement Questionnaire; pregnancy	Each trimester	EPDS; 10	Each trimester PP: 12 weeks	Y	High (vs. low) intake of selenium
Bae et al. (2010)	Korea	114	Cross‐section	43%	Intake of kcal, protein, fat, carbohydrate, fibre, Ca, P, Fe, Na, K, Vit A, B1, B2, B3, B6, B9, C, E, cholesterol, PUFA, MUFA, SFA, P/M/S ratio	FFQ; 24 h recall	Pregnancy	BDI; 10	Pregnancy	N	High (vs. low) intake of total calcium, plant calcium, plant iron, potassium, total folate and dietary folate
Fowles et al. (2011)	United States	18	Cross‐section	94%	Intake of grains, fruit, vegetables, kcal, % kcal from fat, Ca, Vit B9, Fe	FFQ+DQI‐P; 3 × 24 h recall	First trimester	CES‐D; 16	First trimester	N	Low (vs. high) intake of calcium‐rich foods
Lukose et al. (2014)	India	365	Cross‐section	33%	Intake of kcal, protein, fat, Vit C, B9, B12, Ca, P, Fe, Cu, Mn, Zn	FFQ; 3 months recall	First trimester	K‐10; 6	First trimester	Y	None
Watanabe et al. (2010)	Japan	86	Cross‐section	62%	Intake of total energy, % energy from protein, fat, carbohydrate, Ca, Fe, Zn, Vit B9, B6, B12, C	BDHQ; 1 month recall	First trimester	CES‐D; 16	First trimester	Y	None
B VITAMINS
Blunden et al. (2012)	UK∞	2856	Cohort	32%	Intake of folate (B9), Vit B6 and B12	FFQ; 3 months recall	First and third trimester	EPDS; 13	PP: 6 months, 1 year	N	None
Lewis et al. (2012)	UK†	6809	Cohort	Preg.: 14% PP: 9%	Folic acid supplementation during pregnancy	Supplement Questionnaire; pregnancy	Second and third trimester	EPDS; linear/12	Second, Third trimester PP: 8 and 21 months	Y	None
Miyake et al. (2006a)	Japan	865	Cohort	14%	Intake of folate (B9), Vit B2, B6 and B12	FFQ; 1 month recall	Pregnancy	EPDS; 9	PP: 2–9 months	Y	Third quartile (vs. first) of Vit B2 intake
Cho et al. (2008)	Korea	1277	Cross‐section	8%	Folic acid supplementation during pregnancy	Folic acid supplementation questionnaire; pregnancy	Pregnancy	Goldberg's depression scale; 22	Pregnancy	N	None
CALCIUM
Harrison‐Hohner et al. (2001)	United States	293	RCT	14% (OR) 22% (NM)	Group A: calcium supplement Group B: placebo	Randomized; pregnancy	Second trimester	EPDS; 14	PP: 6 and 12 weeks	N/A	Calcium supplementation
Miyake et al. (2015a)	Japanℓ	1745	Cross‐section	19%	Intake of dairy and calcium (milk products and calcium‐rich foods)	FFQ; 1 month recall	Pregnancy	CES‐D; 16	Pregnancy	Y	High (vs. low) intake of yoghurt and calcium
VITAMIN D
Miyake et al. (2015b)	Japanℓ	1745	Cross‐section	19%	Dietary intake of Vitamin D	FFQ; 1 month recall	Pregnancy	CES‐D; 16	Pregnancy	Y	High (vs. low) Vit D intake
ZINC
Roy et al. (2010)	Canada	2030	Cross‐section	Not reported	Intake of zinc	FFQ, supplements; 1 month recall	First or second trimester	CES‐D; 16	First or second trimester	Y	High (vs. low) quintiles of zinc intake
FISH FAT AND PUFAs
Doornbos et al. (2009)	Netherlands	119	RCT	6%	Group A: DHA+AA Group B: DHA Group C: placebo	Randomized; First trimester – 3 months PP	N/A	EPDS; 12	First and second trimester PP: 6 weeks	N/A	None
Judge et al. (2014)	United States	42	RCT	10% (PDSS ≥80)	Group A: DHA Group B: placebo	Randomized; Second to third trimester	Third trimester	PDSS: mean difference	PP: 2–6 weeks	N/A	DHA supplementation (vs. placebo)
Llorente et al. (2003)	United States	138	RCT	7% (BDI >20)	Group A: DHA Group B: placebo	Randomized; 0–4 months PP	N/A	BDI; linear EPDS, SCID‐CV	Third trimester PP: 3 weeks and 2 and 4 months	N/A	None
Makrides et al. (2010)	Australia	2399	RCT	10% vs. 11%	Group A: DHA+EPA Group B: placebo	Randomized; pregnancy	N/A	EPDS; 12	PP: 6 months	N/A	None
Da Rocha and Kac (2012)	Brazil	106	Cohort	26%	Intake of PUFA ratio: n‐6/n‐3 ratio >9:1	FFQ; First trimester recall	First trimester	EPDS; 11	PP: ≥30 days	Y	n‐6/n‐3 PUFA ratio <9:1 (vs. >9:1)
Miyake et al. (2006b)	Japan	865	Cohort	14%	Intake of fish, meat, eggs, dairy, total fat, SFA, MUFA, n‐3 PUFA, n‐6 PUFA, LA, ALA, AA, EPA, DHA, ratio of n‐3 to n‐6 PUFA	FFQ; 1 month recall	Pregnancy	EPDS; 9	PP: 2–9 months	Y	None
Strom et al. (2009)	Denmark	54 202	Cohort	PPD admit: 0.3% PPD Rx: 1.6%	Intake of fish and n‐3 PUFA	FFQ; 1 month recall	Mid‐pregnancy	PPD admit PPD Rx (%)	PP: within 1 year	Y	High (vs. low) fish intake group for PPD prescription
Browne et al. (2006)	New Zealand	80	Case–control	BDI: 17% SCID‐CV: 9%	Intake of PUFA	FFQ; recall NR	Pregnancy	BDI; SCID‐CV; EPDS	PP: 6 months	N	None
Cosatto et al. (2010)	Australia	94	Cross‐section	Not reported	Intake of n‐3 PUFA	FFQ; 3 months recall	First trimester	EPDS; 10	First trimester	N	None
Golding et al. (2009)	UK†	9960	Cross‐section	14%	Intake of n‐3 PUFA	FFQ; pregnancy recall	Third trimester	EPDS; 13	Third trimester	Y	High (vs. low) intake of omega‐3 from seafood
Miyake et al. (2013)	Japan॑	1745	Cross‐section	19%	Intake of fish, meat, total fat, SFA, MUFA, n‐3 PUFA, n‐6 PUFA, LA, ALA, AA, EPA, DHA, ratio of n‐3 to n‐6 PUFA	FFQ; 1 month recall	Pregnancy	CES‐D; 16	Pregnancy	Y	High (vs. low) intake of fish, EPA and DHA
Sontrop et al. (2008)	CanadaØ	2394	Cross‐section	19%	Intake of fish and EPA+DHA	FFQ; 1 month recall	First and second trimester	CES‐D; 16	Second or third trimester	Y	None

Adj. model, statistical analysis adjusted for potential confounding factors; RCT, randomized control trial; PCA, principle component analysis; EPDS, Edinburgh Post‐partum Depression Scale; CES‐D, Center for Epidemiological Studies Depression Scale; BDI, Beck's Depression Inventory, SCID‐CV, Structured Clinical Interview; K‐10, Kessler Depression Scale; PDSS, Post‐partum Depression Screening Scale; Rx, prescription for antidepressants; admit, hospital admission for depression; NR, not reported; PPD, post‐partum depression; Preg, pregnancy; PP, post‐partum; FFQ, food frequency questionnaire; DQI‐P, Dietary Quality Index in Pregnancy; BDHQ, Brief Diet History Questionnaire; RDA, Recommended Daily Allowance; Vit, Vitamin/s; Ca, calcium; Fe, iron; Mg, magnesium; Mn, manganese; Cu, copper; P, phosphorus; Se, selenium; Zn, zinc; B1, thiamin; B2, riboflavin; B3, niacin; B6, pyridoxine; B7, biotin; B9, folic acid; B12, cobalamin; PUFA, polyunsaturated fatty acid; MUFA, monounsaturated fatty acid; SFA, saturated fatty acid; n‐3 PUFA, omega‐3 polyunsaturated fatty acids; n‐6 PUFA, omega‐6 polyunsaturated fatty acids; DHA, docosahexaenoic acid; EPA, eicosapentaenoic acid; ALA, alpha‐linolenic acid; LA, linolenic acid; AA, arachidonic acid; OR, Oregon study site; NM, New Mexico study site; GI, glycemic index; GL, glycemic load.

^॑ Rhea cohort;

^ℓ KOMCHS, Kyushu Okinawa Maternal and Child Health Study;

OMCHS, Osaka Maternal and Child Health Study;

^° APrON, Alberta Pregnancy Outcomes and Nutrition study;

^∞ Southampton Women's Survey;

^† ALSPAC, Avon Longitudinal Study of Parents and Children;

^Ø Prenatal Health Project.

Results

Study selection

The process of study selection is shown in Fig. 2. After removing duplicates, 4808 articles were identified from the three databases. Following the screening of all titles and abstracts, 165 full‐text articles were retrieved and checked against eligibility criteria. Twenty‐three articles that used biomarkers instead of diet or intake measures were excluded and will be separately reviewed. We finally included 35 articles in this review, representing 88 051 distinct subjects (seven study populations were not counted because the analysis came from the same cohorts as other included studies).

Figure 2.

Flowchart of study selection.

Studies were grouped according to four categories of dietary intake: (1) dietary patterns, such as ‘healthy’ or ‘processed’ (nine studies); (2) food intake separated into a full panel of essential nutrients (six studies); (3) food intake used to identify a certain class of nutrients, including intake of B vitamins (four studies), calcium (two studies), Vitamin D (one study) and zinc (one study); and (4) fish and PUFAs (12 studies). The study populations were population‐based prospective cohorts (n = 14), recruited from hospitals or clinics (n = 15), a baseline assessment of an RCT (Lukose et al. 2014), RCTs recruiting from health facilities (n = 5) and one RCT that did not report the source population (Doornbos et al. 2009).

The studies measured depression using a structured clinical interview, such as the Structured Clinical Interview for DSM‐IV Diagnoses, or a screening tool for depressive symptoms, such as the EPDS or CES‐D. The prevalence of perinatal depression reported in the general population was between 10% and 15% (Blunden et al. 2012; Chatzi et al. 2011; Da Rocha & Kac 2012; Leung et al. 2013). Several of the cross‐sectional studies reported a very high prevalence of depressive symptoms, ranging from over 40% to 94% (Bae et al. 2010; Fowles et al. 2011a; Watanabe et al. 2010). In the cohort studies, prevalence of perinatal depression ranged from 12% (Leung et al. 2013) to 31.7% (Blunden et al. 2012).

Risk of bias assessment

There were several common sources of bias of the included studies, presented in Figures  3, 4, 5. Because risk of bias in certain domains (such as study attrition) was different for cross‐sectional and cohort studies, these studies are presented separately. One case–control study is presented with the cohort studies (Browne et al. 2006). The RCTs were assessed by a different tool and are also presented separately.

Figure 3.

Risk of bias assessment for cross‐sectional studies.

Figure 4.

Risk of bias assessment for cohort and case–control studies.

Figure 5.

Risk of bias assessment for randomized control trials.

All of the cross‐sectional studies had some risk of bias in one or more domains, often because study characteristics were not adequately reported. In the study participation domain, only four studies achieved low risk of bias, while 75% (12/16) had medium or high risk of bias. Only 6/16 (38%) studies had low risk of bias concerning participation rates and potential differences between participants and non‐participants. Overall, cross‐sectional studies described appropriate exposure measurement, with only three out of 16 studies having high or medium risk of bias. Nine studies had high or medium risk of bias in the outcome measurement domain, which often resulted from using a depression screening tool not validated for pregnancy (i.e. not the EPDS or the PDSS) and from inconsistent timing of outcome measurement. Another weak domain was the presentation of statistical analysis, in which only 44% (7/16) had low risk of bias.

Out of 12 cohort studies and one case–control study, only two studies had low risk of bias across all domains. Studies performed best on measurement of exposure and outcome and statistical analysis with nearly all studies ranked as low risk of bias on these domains. Half of the studies suffered from high or unreported study attrition. Not accounting sufficiently for potential confounding was a weakness in many observational studies. Only 8/16 (50%) cross‐sectional studies and 8/12 (67%) cohort studies had a low risk of bias in this domain.

Only one RCT, Makrides et al. (2010) in Australia, had low risk of bias across all domains. Five out of six RCTs had incomplete outcome data causing possible attrition bias. Doornbos et al. (2009), Harrison‐Hohner et al. (2001) and Paoletti et al. (2013) failed to report on at least some of the methods for allocation concealment or blinding of participants and outcome assessors.

Results from studies

Eleven studies showed clear protective effects from dietary intake: four cohort studies, three RCTs and four cross‐sectional studies (Chatzi et al. 2011; Vilela et al. 2014; Paoletti et al. 2013; Miyake et al. 2015a; Da Rocha & Kac 2012; Miyake et al. 2013; Strom et al. 2009; Miyake et al. 2014; Miyake et al. 2015b; Harrison‐Hohner et al. 2001; Judge et al. 2014). Eleven studies (seven cross‐sectional and four cohort studies) detected associations that could have been spurious, as they were only observed within one or two strata (i.e. quartiles or quintiles) but not as a trend. Some of these studies observed only one or two associations from many tests of different nutrients or dietary components but not in an overall pattern or group. Thirteen of 35 studies reported weak or no associations at a threshold significance level of 0.05 (five cross‐sectional studies, five cohort studies and three RCTs). All studies are described by group of exposure.

(1) Dietary patterns (Appendix 1)

Four cross‐sectional studies and five cohort studies examined the relationship between dietary patterns and perinatal depression. Chatzi et al. (2011) used principle component analysis (PCA) and factor loading to identify two different dietary patterns, termed ‘healthy’ and ‘Western’, in pregnant women in Greece. They were ranked as having low risk of bias in all domains except study attrition and found that the ‘healthy’ pattern was protective against depressive symptoms 8–10 weeks post‐partum in both the second and third tertiles [second tertile: risk ratio (RR) = 0.52, 95% confidence interval (CI) 0.30–0.92, P = 0.03; third tertile: RR = 0.51, 95% CI 0.25–1.05, P = 0.07; P‐value for trend = 0.04] compared with the least ‘healthy’ tertile, after adjusting for potential confounders. Similar results were observed when analysing the depression score as a continuous variable. There was no evidence that the degree of adherence to the Western dietary pattern was associated with depression.

Vilela et al. (2014) conducted a cohort study in Brazil and also found a healthy dietary pattern to be protective against antenatal depression. Using PCA, they identified a ‘healthy’ pattern, a ‘common Brazilian’ pattern and a ‘processed’ pattern. They were ranked as having a low risk of bias in 5/6 domains and showed evidence for an association between adherence to a healthy dietary pattern and a lower EPDS score (β‐coefficient = −0.72, 95% CI −1.28 to −0.17, P = 0.01), while there was no evidence for any effect of the other dietary patterns tested.

Two cohort studies from Osaka, Japan, analysed dietary patterns identified with factor analysis (‘Healthy’, ‘Western’ and ‘Japanese’) as well as glycemic index and glycemic load, both of which suffered from risk of bias in several domains. One found evidence that a ‘Western pattern’ was associated with less perinatal depression but only in the second quartile of dietary compliance compared with the first [adjusted odds ratio (OR) = 0.52, 95% CI 0.30–0.93] (Okubo et al. 2011). The other study found that glycemic load was not associated with perinatal depression but that the third highest quartile (vs. lowest) of glycemic index score was associated with less depressive symptoms (adjusted OR = 0.56, 95% CI 0.32–0.995) (Murakami et al. 2008). Pina‐Camacho et al. (2015) used a large birth cohort from England and found no association between unhealthy diet, determined using PCA, and post‐partum depression, although their primary outcome of interest was not perinatal depression.

Three cross‐sectional studies from the United States found some protective associations with depression, for instance, with low (vs. high) frequency of fast food consumption, high (vs. low) Dietary Quality Index in Pregnancy and high (vs. low) adherence to US national dietary guidelines (Fowles et al. 2011b, Fowles et al. 2012; George et al. 2005), despite very small sample sizes (51, 70 and 146, respectively). The authors did not present an analysis adjusted for confounding factors. Miyake et al. (2014) examined cross‐sectional associations between seaweed intake and antenatal depression and found in an adjusted model that those in the highest quartile of seaweed consumption were less likely to be depressed than those in the lowest, observing a linear trend between quartiles (highest quartile OR 0.68 (CI 0.47–0.96), P for trend = 0.03), but this study was ranked as having unclear or medium risk of bias in 3/6 domains.

(2) Full panel of nutrients (Appendix 2)

Four cross‐sectional studies, one cohort study and one RCT used FFQs to quantify a whole range of essential nutrient intake. Paoletti et al. (2013) conducted an RCT in Italy, comparing Vitamin D and calcium supplementation with a full multivitamin supplement, measuring EPDS scores at 3 and 30 days post‐partum. They found that at 1 month post‐partum, the mean EPDS score was significantly (P < 0.01) lower in the full multivitamin supplementation group (3.2, SD 2.8, n = 274) than in the Vitamin D and calcium group (4.5, SD 3.8, n = 278). They reported that the difference was greater in the subgroup of women who screened positive for depression at entry into the study (EPDS score ≥12) but did not provide an interaction test. Their study had high or medium risk of bias in 4/6 domains. Leung et al. (2013) in their cohort study in Canada quantified the mean, range and percent of Recommended Daily Allowance in intake of various B vitamins, Vitamin D, omega‐3 PUFAs, iodine, iron, zinc and selenium. Of these nutrients, only selenium supplementation seemed protective against depression at 12 weeks post‐partum (adjusted OR = 0.76, 95% CI 0.74–0.78, P = 0.02). This study had low risk of bias across all domains.

Four cross‐sectional studies examined a range of nutrients and their possible associations with depression in pregnancy, three of which were carried out in the first trimester. These studies had medium and high risk of bias in several domains. Bae et al. (2010) in Korea found that intake of total calcium, plant calcium, plant iron, potassium, dietary folate and total folate (including supplements) was associated with depression but did not present an adjusted model. Fowles et al. (2011a) in the United States included 18 women in their sample, 17 of whom screened positive for depression, and found that higher calcium intake was associated with more depressive symptoms. Lukose et al. (2014) in India and Watanabe et al. (2010) in Japan found no evidence for protective associations from any of numerous vitamins and minerals tested.

(3) Specific nutrients (Appendix 3)

B vitamins

None of the studies evaluating B vitamins showed clear evidence of protective effects against perinatal depression. Cho et al. (2008) in a cross‐sectional study in Korea, ranked as having high and medium risk of bias in all domains, found no evidence that supplementation with folic acid (Vitamin B9) was protective against depressive symptoms during pregnancy. Blunden et al. (2012) measured intake of B vitamins 6, 9 (folate) and 12, finding no relationship with depressive symptoms at either 6 months or 1 year post‐partum in a study with low risk of bias in 5/6 domains. Lewis et al. (2012) did not find any evidence for an effect of folic acid supplements during pregnancy on risk of depression at 8 months post‐partum, although there was a protective effect at 21 months post‐partum in those with the MTHFR C677T genotype. Only one of four studies evaluating B vitamins found weak evidence of a protective association, but only in the third highest, and not in the highest quartile of riboflavin intake (Miyake et al. 2006a).

Calcium, Vitamin D and zinc

Harrison‐Hohner et al. (2001) supplemented pregnant women with calcium or a placebo and then measured post‐partum depression in two different study sites. Results reported for one study site showed 6% of the supplementation group and 15% of the placebo group screened positive for major depressive disorder. This study had high or medium risk of bias in blinding of participants and assessors, incomplete outcome data and selective reporting.

Miyake et al. (2015a) in a cross‐sectional study in Japan found a protective linear association between overall calcium and yoghurt intake (in quartiles) and depressive symptoms during pregnancy (CES‐D score; ≥16 vs. <16): highest vs. lowest quartile of calcium intake: OR 0.59, 95% CI 0.40–0.88, P = 0.01; highest vs. lowest quartile of yoghurt intake: OR 0.69, 95% CI 0.48–0.99, P = 0.03. The authors adjusted for many different potential confounders and found little difference between adjusted and unadjusted results.

Miyake et al. (2015b) used the same cross‐sectional data to examine Vitamin D intake and antenatal depression. The study had a high or medium risk of bias in several domains and did not incorporate proxy measures of sunlight exposure. They detected a protective effect of intake when comparing the highest with lowest quartile (adjusted OR 0.52 CI 0.30–0.89, P for trend = 0.02).

Roy et al. (2010) measured zinc intake in over 2000 women in Canada and found that women in the lowest quintile of zinc consumption had a higher risk of antenatal depression between 10 and 22 weeks of gestation (lowest quintile β = 1.0, standard error 0.5, P = 0.03).

(4) Fish and PUFA intake (Appendix 4)

Only one of the four RCTs on supplementation with PUFAs, comprising docosahexaenoic acid (DHA), arachidonic acid and/or eicosapentaenoic acid (EPA), found an effect on perinatal depression. Judge et al. (2014) randomized 42 participants to receive either DHA in fish oil or a placebo during pregnancy and found that post‐partum depression was lower in the DHA supplementation group (repeated measures least square mean: 46.03 (2.17) vs. 52.11 (2.4), P = 0.02). Doornbos et al. (2009) in the Netherlands supplemented 119 women with DHA and arachidonic acid, only DHA or placebo from the first trimester until 3 months post‐partum and found no evidence of an effect on depressive symptoms using the EPDS at 6 weeks post‐partum. This study suffered from high or medium risk of bias across all domains. Llorente et al. (2003) in a trial with 138 women in the United States also found no evidence that women supplemented with DHA for the first 4 months post‐partum scored differently than those on placebo on the BDI. Makrides et al. (2010) ran a large trial (n = 2399) in Australia, supplementing women for the entire pregnancy with DHA and EPA or placebo and measuring their depressive symptoms at 6 months post‐partum using the EPDS. They also found no strong evidence of a protective effect against depressive symptoms (adjusted RR 0.85, CI 0.70–1.02, P = 0.09) and also had a low risk of bias across all domains.

Two of four cohort studies on fish and essential fatty acid intake showed a clear protective effect. Da Rocha & Kac (2012) found that pregnant women in Brazil who ate a high ratio of n‐6 to n‐3 PUFAs had a 2.5 times higher prevalence of depressive symptoms (95% CI 1.21–5.14, P = 0.01) compared with women with a lower ratio, but they had high risk of bias from high study attrition. Strom et al. (2009) used an innovative approach linking fish and n‐3 PUFA intake data from the Danish National Birth Cohort to registries containing information on hospital admission for depression and prescriptions for antidepressants within 1 year of giving birth. They found no associations with hospital admission for post‐partum depression, but women who were in the lowest group of fish intake (vs. the highest) had 1.46 times the odds of having filled a prescription for anti‐depressants (95% CI 1.12–1.90, P = 0.04). Browne et al. (2006) in a case–control study and Miyake et al. (2006b) in a cohort study found no evidence of associations between PUFA intake and post‐partum depression.

Two out of four cross‐sectional studies on fish and PUFA intake found evidence of protective effects on depressive symptoms during pregnancy. Golding et al. (2009) in the UK found a protective effect of eating fish (none vs. any fish intake (three groups combined): OR 1.54, 95% CI 1.25–1.89, test for trend over four groups: P < 0.01). Miyake et al. (2013) in Japan found evidence of protective effects from overall fish intake, DHA and EPA intake (highest vs. lowest quartiles of fish intake: OR 0.61, 95% CI 0.42–0.87, P = 0.01; EPA: OR 0.66, 95% CI 0.46–0.95, P = 0.02; and DHA: OR 0.64, 95% CI 0.44–0.93, P = 0.007). They also found evidence that higher intakes of total fat and saturated fat were risk factors for depressive symptoms (highest vs. lowest quartiles of total fat: OR 1.42, 95% CI 1.00–2.03, P = 0.06; saturated fat: OR 1.74, 95% CI 1.22–2.49, P = 0.001). Cosatto et al. (2010) and Sontrop et al. (2008) found no evidence for protective effects against depressive symptoms of eating fish or consuming PUFAs during pregnancy.

Discussion

Overall, there is limited evidence that dietary intake influences the risk of perinatal depression. While 13 studies, including three PUFA supplementation trials, found no evidence of an association, 22 studies showed some protective effects from healthy dietary patterns, multivitamin supplementation, fish and PUFA intake, calcium, zinc and possibly selenium. However, given the methodological limitations of studies on this topic, the lack of evidence does not necessarily imply the absence of true associations. In order to clearly answer the question whether certain diets and nutrients influence the risk of perinatal depression, the issues discussed below warrant consideration in future study designs.

Exposure measurement

Most dietary intake assessments rely on FFQs, which are the best available option for ascertaining dietary and nutrient intake but are still imprecise measures and can suffer from reporting bias (Ioannidis 2013). In the case of Vitamin D, measuring dietary intake alone is insufficient because a main source of bioavailable Vitamin D is sunlight exposure (Millen & Bodnar 2008). Blood analysis can verify what is processed and made available by the body, and four of the reviewed studies evaluated both dietary intake and nutrient blood levels (Bae et al. 2010; Fowles et al. 2011a; Lukose et al. 2014; Watanabe et al. 2010). However, blood analysis is costly, often limited to a few nutrients and does not capture unknown food compounds and food synergies. For these reasons, it is difficult to detect robust associations between nutrition and health outcomes.

Often, nutrition is only measured at one point during pregnancy, while long‐term dietary habits may be more relevant, and pregravid nutritional status and intake is an important baseline for nutrient changes (primarily losses) during pregnancy and lactation. Dietary intake prior to pregnancy was not measured in any included study, although some studies had a recall period covering pre‐pregnancy, and four of the Japanese studies included a question about whether there were significant changes in the diet in the previous month (Miyake et al. 2006a, 2006b; Okubo et al. 2011; Murakami et al. 2008). Da Rocha & Kac (2012), Strom et al. (2009) and 2014, 2015a, 2015b included pre‐pregnancy BMI. Only 6/35 studies (17%) measured dietary intake more than once (Blunden et al. 2012; Lewis et al. 2012; Roy et al. 2010; Sontrop et al. 2008; Vilela et al. 2014; Leung et al. 2013), and the timing of measurements was highly variable between subjects enrolled. Because dietary intake and nutritional needs can fluctuate heavily during the perinatal period, the timing of exposure measurement should be considered in any multivariable model of nutrition and depression but was included in only 10/35 studies (29%). Ideally, dietary intake should be measured at baseline or prior to pregnancy and again at several follow‐up time points to offset the imprecision of a FFQ with a long recall period. Such successive intake measurements were performed by only five studies included in this review (Leung et al. 2013; Blunden et al. 2012; Lewis et al. 2012; Roy et al. 2010; Sontrop et al. 2008).

Outcome measurement

Clinical diagnosis of perinatal depression is the gold standard but requires a trained interviewer and an evaluation that often takes over an hour and thus is usually not feasible for epidemiological purposes (Bennett et al. 2004). Depression screening tools are commonly used instead, and the EPDS and PDSS are particularly designed and validated for the perinatal period. Because symptoms can change rapidly and dramatically, especially during the perinatal period, a standard and well‐timed measurement is particularly important (Noble 2005). While screening tools for perinatal depression are not a substitute for a clinical diagnosis of depression, they can have good sensitivity and specificity, depending on the cut‐off point and population (Gibson et al. 2009; Hewitt et al. 2009). The EPDS has been validated for many cultural contexts (Gibson et al. 2009) and for use during pregnancy. A cut‐off point of ≥13 for the EPDS (used by Chatzi et al., Blunden et al. and Golding et al.) has been shown to have a sensitivity of 0.91 (95% CI 0.84–0.99) and a specificity of 0.91 (95% CI 0.88–0.94) for post‐partum depression (Hewitt et al. 2009). Nineteen of the 35 studies used the EPDS or PDSS or a clinical diagnosis of depression, while the others used screening tools that may not produce valid results in the perinatal period.

Depression prior to pregnancy is a known risk factor for perinatal depression, and antenatal depression may predict post‐partum depression; therefore, it is critical to include pre‐pregnancy or antenatal screening measurements. However, only eight studies took repeated measurements of depression (Vilela et al. 2014; Leung et al. 2013; Paoletti et al. 2013; Blunden et al. 2012; Lewis et al. 2012; Judge et al. 2014; Harrison‐Hohner et al. 2001; Pina‐Camacho et al. 2015), and only 10 studies included a history of depression in their analysis (Chatzi et al. 2011; Leung et al. 2013; Blunden et al. 2012; Miyake et al. 2013, 2015a; Strom et al. 2009; Miyake et al. 2014, 2015b; Harrison‐Hohner et al. 2001; Judge et al. 2014).

Other sources of bias

Many of the included studies suffered from design issues. Ten studies had sample sizes less than 200, limiting the power to detect differences between depressed and non‐depressed women. Sixteen of the 35 studies were cross‐sectional, and associations in these studies could be due to reverse causality: depression may cause changes in self‐care and eating habits, which may change dietary intake. In order to assess whether nutritional factors cause depression, intake has to be measured prior to ascertaining depression with an appropriate lag time. However, of the 11 reviewed studies that reported clear protective effects, eight were cohort studies or RCTs, and therefore, only four were limited by potential reverse causation (Miyake et al. 2013, 2015a, 2014, 2015b). Thirteen of the studies were nested in large cohorts and therefore benefitted from larger sample sizes, lag time between exposure and outcome measurement, and longer follow‐up periods. In some studies, there was high variability in the timing of both exposure and outcome measurements, which limited comparability between subjects. The enrollment and inclusion criteria of study participants were often not described in detail, with the exception of those within research cohorts. Many studies did not adjust for potential confounders, meaning that associations observed could have been due to other factors related to both dietary intake and perinatal depression.

Prevalence of nutritional deficiency

Most studies included in this review were undertaken in high‐income settings where dietary intake is generally sufficient and the prevalence of nutrient deficiencies is often very low. In order to create an evidence base that is generalizable and relevant for vulnerable populations, studies must also be carried out in LMICs where prevalence rates of perinatal depression, as well as insufficient nutrition, are higher (Fisher et al. 2012; Patel et al. 2004). Only one of the 35 studies was from a lower‐middle‐income country, India, and two from a higher‐middle‐income country, Brazil. If we assume that there is a nutritional ‘threshold’, i.e. a critical point below which deficiencies will lead to symptoms of depression, and the studies included few or no women below the threshold, this could explain the lack of associations. In other words, there may have been insufficient power to detect effects because of the low number of participants in the included studies eating a diet that places them in the most deficient range. Conversely, a Canadian study in a socio‐demographically homogenous clinic population in London, Ontario, found that women in the lowest quintile of zinc intake (around or below the recommended daily intake) had a one point higher depression score on average than women in the highest quintile, a difference that the authors did not consider ‘clinically meaningful’ but which reached statistical significance because of a large sample size of 2030 women (Roy et al. 2010). Another explanation for the absence of associations in many studies could be a lack of variation in dietary intake in the sample. For example, Blunden et al. (2012) report null results for folate (Vitamin B9), Vitamin B6 and Vitamin B12, but 96% of the women in that study reported taking folic acid supplements during pregnancy, and the prevalence of folate deficiency in the study sample was below 5%.

Practical implications

The study of single nutrients or foods may not capture how ingested foods interact in complex biochemical systems to influence mood regulation and outcomes such as depression. Considering diets and dietary patterns, however, better reflects the synergies that result from cumulative food consumption, nutrient interactions, food quality, unknown substances or unknown effects of substances in food. More holistic approaches to exposure measurement may therefore be a promising approach in epidemiological studies of the effects of nutrition on disease outcomes (Jacobs & Steffen 2003). Two studies included in this review suggest that there may be a protective effect against perinatal depression from eating a healthy diet (Chatzi et al. 2011; Vilela et al. 2014), a finding that may encourage further investigation. Other studies have examined the role of specific dietary patterns on depression outside of the perinatal period (Akbaraly et al. 2009; Jacka et al. 2010; Sanchez‐Villegas et al. 2006), and one found that a ‘healthy’ diet was protective against depression in Australian women (Jacka et al. 2010).

The current interest in the relationship between dietary intake and perinatal depression focuses mostly on treatment of depression with non‐pharmacological interventions, such as PUFAs (Ellsworth‐Bowers & Corwin 2012). Existing interventions for perinatal depression are often inaccessible or ineffective for women who are exposed to numerous risk factors, so a shift of focus to prevention is warranted.

A series of articles published in The Lancet in 2014 highlighted risk factors for perinatal depression such as low socioeconomic status, trauma, domestic violence, lack of support, migration status, history of psychopathology, and chronic illness and medical problems (Howard et al. 2014a). Several reviews highlight non‐nutritional treatment options for perinatal depression in LMICs that focus on social support and complex interventions (Rahman et al. 2013; Howard et al. 2014a). Some of the complex interventions suggested comprise health and economic components that are also designed to improve nutrition. Better understanding the root causes of perinatal depression may allow a shift from treatment to prevention, and targeting these root causes might simultaneously alleviate other adverse health outcomes.

Strengths and limitations

We opted for broad inclusion criteria and a sensitive search strategy, which meant that the studies reviewed were heterogeneous and covered a wide range of dietary intake measurements, including whole diets. We included both antenatal and post‐partum depression in our review, which have previously often been separated (O'Hara & Wisner 2014; Howard et al. 2014b). Recent epidemiological studies suggest that antenatal depression is both common unto itself and a risk factor for post‐partum depression (Patel et al. 2002; Liabsuetrakul et al. 2007; Husain et al. 2006), and therefore, we decided that these should be considered together. The heterogeneity of studies excluded the possibility of a meta‐analysis, but this review is the most current synthesis of the evidence available on this topic.

Only English‐language papers were included, which limits our ability to incorporate evidence presented in other languages that might be relevant. Additionally, this review only included published study results, which may have introduced bias as studies that found no associations between dietary intake and perinatal depression are less likely to be published.

A main limitation of this review is that few studies were from settings where severe or widespread nutrient deficiencies exist, and nutrient intake levels may therefore have been too high and their range too narrow to detect any association with depression. It should also be noted that several studies came from the same groups of researchers, one group from Japan contributed eight studies and another group from the United States contributed three. Because they provided distinct analyses of different nutrients, all of these were included, but the populations and study designs in these groups were the same or very similar, and thus, they do not represent independent pieces of evidence.

Conclusions

This review shows that there is inconclusive evidence that individual nutrients or dietary patterns contribute to the development of perinatal depression. Further studies in populations that have a wider variation in nutrient intake and compromised nutrition are needed to better elucidate this relationship. Also, future studies should incorporate baseline levels of nutrition and depression, as well as measuring dietary intake and depression at several time points throughout the perinatal period. High‐quality longitudinal studies could also overcome the problem of reverse causality between dietary intake and perinatal depression.

Source of funding

TS and SG are funded by a grant (Award number 01ER1201) of the German Ministry of Education and Research (BMBF). The content of this publication is solely the responsibility of the authors.

Conflicts of interest

The authors declare that they have no conflicts of interest.

Contributions

TS proposed the concept and compiled the search. SG and NH advised on conceptual issues. NH, RN and TS screened titles, abstracts and full text articles for inclusion, and completed the risk of bias assessment. TS led the review process and the writing. SG led the editing process, and all authors read, edited and approved the final manuscript.

Appendix Table 1. Results of studies evaluating associations between dietary patterns and perinatal depression

Author (year)	n	Nutrients/diet (components of analysis)	Analysis	Unit of exposure	Unit of outcome	Unadjusted associations significant at p < 0.10 (effect size, 95% CI, P‐value)	Adjusted associations significant at p < 0.10 (effect size, 95% CI, P‐value)	Confounding considered	Limitations
Chatzi et al. (2011)	519	FFQ into principle component analysis, factor loading. • ‘Healthy pattern’ = high vegetables, fruit, pulses, olive oil, fish/seafood, milk and nuts. • ‘Western pattern’ = high meat/meat products, potatoes/starches, sugar/sweets, cereals, fats except olive oil, eggs, alcoholic and non‐alcoholic drinks, snacks and sauces	Multivariable log‐binomial regression	Tertiles of dietary compliance (T1: lowest/reference, T2: middle, T3: highest)	EPDS score; ≥13 vs. <10	Healthy: T1: ref. • T2:e RR = 0.54 (CI 0.31–0.95); P = 0.03 • T3: RR = 0.56 (CI 0.28–1.12); P = 0.10 • P for trend = 0.07 • Western: tertiles NS, P for trend: NS	Healthy: T1: ref. • T2: RR = 0.52 (CI 0.30–0.92); P = 0.03 • T3: RR = 0.51 (CI 0.25–1.05); P = 0.07 • P for trend = 0.04 • Western: tertiles NS, P for trend: NS	Age, education, parity, house tenure, depression in previous pregnancies and total energy intake during pregnancy.	• Low participation rate (57%), but analysis shows no significant differences between participants and non‐participants
			Multivariable linear regression		EPDS score; linear	Healthy: T1: ref. • T2: β = −0.80 (CI −1.86–0.25); P = 0.14 • T3: β = −0.63 (CI −1.69 to −0.43); P = 0.04 • P for trend = 0.24 Western: T1: ref. • T2: β = −0.91 (CI −0.14−1.96); P = 0.09 • T3: β = 1.27 (CI 0.21–2.32); P = 0.02 • P for trend = 0.02	Healthy: T1: ref. • T2: β = −1.13 (CI 2.25–0.0); P = 0.05 • T3: β = −1.75 (CI −3.22 to −0.28); P = 0.02 • P for trend = 0.02 Western: T1: ref. • T2 β = 0.96 (CI −0.17–2.00); P = 0.10 • T3 β = 1.32 (CI −0.19–2.76); P = 0.07 • P for trend = 0.07
Murakami et al. (2008)	865	FFQ to quantify high GI and GL in diet. • GI = consumed foods * weighted GI value. • GL = dietary GI * total daily available carbohydrate intake (divided by 100), based on a total of 72 major carbohydrate‐containing foods	Multivariable logistic regression	Quartiles of dietary indices (Q1: lowest/reference, Q2, Q3, Q4: highest)	EPDS score; ≥9 vs. <9	GI compliance: Q1: ref. • Q2: OR 0.68 (CI 0.40–1.15) • Q3: OR 0.56 (CI 0.33–0.98) • Q4: OR 0.73 (CI 0.44–1.23) • P for trend = 0.17 • GL quartiles and trends NS	GI diet compliance: Q1: ref. • Q2: OR 0.68 (CI 0.39–1.17) • Q3: OR 0.56 (CI 0.32–0.995) • Q4: OR 0.72 (CI 0.41–1.26) • P for trend = 0.18 • GL quartiles and trends NS	Age, gestation, parity, smoking, family structure, occupation, family income, education, changes in diet in previous month, season of baseline data, BMI, timing of delivery, problems during pregnancy, baby's sex and baby's birth weight, as well as intakes of n‐3 PUFA and Vit B2.	• Assessment timing not standard for all participants • No history of depression considered
Okubo et al. (2011)	865	FFQ into factor analysis grouped into ‘Healthy’: high loadings of green/yellow/white vegetables, seaweeds, fish/seafood and fruits; ‘Western’ high loadings of vegetable oil, beef/pork/chicken, salty seasonings and processed meat and eggs; ‘Japanese’: high loadings of rice, miso soup, fish/seafood and pickled vegetables.	Multivariable logistic regression	Quartiles of dietary patterns (Q1: lowest/reference, Q2, Q3, Q4: highest)	EPDS score; ≥9 vs. <9	Western pattern: Q1: ref. • Q2: OR 0.50 (CI 0.29–0.87); P < 0.05 • Q3: OR 0.68 (CI 0.40–1.14) • Q4: OR 0.71 (CI 0.42–1.19) • P for trend = 0.31	Western pattern: Q1: ref. • Q2: OR 0.52 (CI 0.30–0.93); P < 0.05 • Q3: OR 0.71 (CI 0.41–1.20) • Q4: OR 0.73 (CI 0.42–1.24) • P for trend = 0.36	Age, gestation, parity, smoking, family structure, occupation, family income, education, changes in diet in previous month, season of baseline data, BMI, time of delivery, problems during pregnancy, baby's sex and baby's birth weight.	• Assessment timing not standard for all participants • Potential confounders not included in adjusted model • No history of depression considered
Pina‐Camacho et al. (2015)	7814	FFQ into factor analysis into unhealthy diet score (higher = worse): defined by two first‐order latent factors: processed food (i.e. fried food, meat pies or pasties and chips) and confectionery (i.e. crisps, chocolate bars, cakes or buns and biscuits)	Adjusted path analysis	Correlations between pathway variables	EPDS; linear	Correlation coefficients: •Unhealthy prenatal diet and antenatal depression: 0.024, P ≤ 0.05 • Unhealthy prenatal diet and PP depression: 0.012, P > 0.05	Correlations between factors of path analysis: • Unhealthy prenatal diet and antenatal depression: NR • Unhealthy prenatal diet and PP depression: −0.03, P > 0.05	Age, ethnicity, education, marital status, social class, sex of offspring, birth weight, parity, birth complications, mother involved with police, substance use, partner cruelty, inadequate basic living conditions, inadequate housing, housing defects, poverty, being a single caregiver, early parenthood and low educational attainment	• Primary study outcome not depression • Correlations presented, no causation able to be concluded • Study variables were associated with exclusion and loss to follow‐up • No history of depression considered
Vilela et al. (2014)	246	FFQ into principle component analysis grouped into • ‘Common Brazilian’: high loadings of rice, beans, vegetables, spices, meats and eggs. • ‘Healthy’: high loadings of dairy, fruits, juice, green vegetables, legumes, candies, fish, cakes, cookies/crackers, noodles, pasta, roots and tea. • ‘Processed’: high loadings of bread, fat, fast food, snacks, sugar, sausages, soft drinks and coffee.	Multivariable longitudinal linear regression	Score for dietary pattern compliance	EPDS score; linear	EPDS score: • Common Brazilian: 8.7 (CI 7.7–9.6) • Healthy: 8.3 (CI 7.2–9.3) • Processed: 10.0 (CI 8.9–11.2) • P = 0.06	• Healthy pattern: β = −0.723 (CI −1.277 to −0.169); P = 0.01 • Other dietary patterns NS	Age, education, parity*, early pregnancy BMI, gestational age*, total energy intake, marital status* and unplanned pregnancy*	• Diet measured in first weeks of pregnancy with a recall of 6 months, (analysis was for diet before and in very early pregnancy). • Potential confounders not included in model
Fowles et al. (2011b)	50	3 × 24 h quantitative food recall, dichotomized into high vs. low frequency of eating fast food	t‐test	Frequency of eating fast food	EPDS score; ≥10 vs. <10	Low frequency mean: 6.8 (SD 4.1) High frequency mean: 10.4 (SD 6.0); P < 0.05	Not reported	No	• No adjusted model of the association of interest • No history of depression considered
Fowles et al. (2012)	71	3 × 24 h quantitative food recall, eight‐component Dietary Quality Index–Pregnancy (quantified by intake of grains, fruit, vegetables, kcal, % kcal from fat, calcium, folate, iron and meals/snacks); DQI‐P score high vs. low	Chi‐squared and t‐test	DQI‐P score	EPDS score; linear	High vs. low DQI‐P score: • EPDS score: 6.7 vs. 9.6; P = 0.02	Not reported	Tested for association with the outcome but not included in final model: race/ethnicity, US citizen, marital status, education, insurance type, physical activity, smoking prior to pregnancy, current smoking and pregnancy intendedness	• No adjusted model of the association of interest • No history of depression considered
George et al. (2005)	146	FFQ into DIETSYS statistical software, into compliance with US dietary guidelines index score	ANOVA and chi‐squared test and Pearson correlation	Tertiles of dietary compliance (T1: lowest, T2, T3: Highest)	CES‐D score; linear	• T1: 21.2 (SD 1.9) • T2: 15.1 (SD 1.6) • T3: 15.0 (SD 1.5) • P = 0.02	Not reported	No	• No adjusted model of the association of interest • No history of depression considered • Primary study outcome not depression
Miyake et al. (2014)	1745	FFQ to quantify seaweed consumption (g/d)	Multivariable logistic regression	Quartiles of intake (Q1: reference, Q2, Q3, Q4: highest)	CESD score; ≥16 vs. <16	Seaweed intake: Q1: ref. • Q2: OR 0.70 (CI 0.51–0.97) • Q3: OR 0.62 (CI 0.45–0.86) • Q4: OR 0.60 (CI 0.43–0.83) • P for trend = 0.002	Seaweed intake: Q1: ref. • Q2: OR 0.72 (CI 0.51–1.004) • Q3: OR 0.71 (CI 0.50–1.01) • Q4: OR 0.68 (CI 0.47–0.96) • P for trend = 0.03	Age, gestation, residence, parity, family structure, history of depression, family history of depression, smoking, secondhand smoke exposure, job type, household income, education, BMI and intake of fish and yoghurt	• Assessment timing not standard for all participants • Participation and rate and characteristics not available

BMI, body mass index; CES‐D, Center for Epidemiological Studies Depression Scale; CI, confidence interval; DQI‐P, Dietary Quality Index in Pregnancy; EPDS, Edinburgh Post‐partum Depression Scale; GI, glycemic index; GL, glycemic load; NS, not significant; OR, odds ratio; g, grams; kcal, kilocalories.

^*Significant (P < 0.05) in final model.

Appendix Table 2. Results of studies evaluating associations between nutrient intake and perinatal depression

Author (year)	n	Nutrients/diet (components of analysis)	Analysis	Unit of exposure	Unit of outcome	Unadjusted associations significant at P < 0.10 (effect size, 95% CI, P‐value)	Adjusted associations significant at P < 0.10 (effect size, 95% CI, P‐value)	Confounding considered	Limitations
Paoletti et al. (2013)	552	Group A: Vit A, 3600 IU; Vit B1, 1.55 mg; Vit B2, 1.8 mg; Vit B6, 2.6 mg; B7, 0.2 mg; Vit B12, 4 mcg; Vit C, 100 mg; Vit D3, 500 IU; Vit E, 15 IU; Vit B9, 800 mcg; Ca, 125 mg; Fe, 60 mg; Mg, 100 mg; Mn; 1 mg; Cu, 1 mg; P, 125 mg; Zn, 7.5 mg Group B: Ca, 500 mg; Vit D3, 400 IU	Mean difference between groups	Group A vs. group B	Mean EPDS score	Mean differences in EPDS score with P < 0.05: • All participants: group A 3.2 (SD 2.8) vs. group B 4.5 (SD 3.8) • Basal EPDS <12: group A 2.7 (SD 2.5) vs. group B 3.5 (SD 3.2) • Basal EPDS ≥12: group A 6.8 (SD 3.8) vs. group B 9.9 (SD 3.8 )	Not reported	No baseline imbalance for age, BMI, baseline haematological parameters and EPDS score.	• Food intake not measured • Loss to follow‐up not analysed for differences in completers and non‐completers • No history of depression considered
Leung et al. (2013)	475	Supplement questionnaire to quantify mean, range and percent of RDA through prenatal supplement intake of Vit B1 (mg), B3 (mg), B6 (mg), B9 (mcg), B12 (mcg), D (IU), iodine (mcg), Fe (mg), Mg (mg), Se (mcg), Zn (mg) and n‐3 PUFA (mg)	Multivariable logistic regression	Mean nutrient intake	EPDS score <10 vs. ≥10	Univariable analysis: • n‐3 PUFA supplement intake (for EPDS <10 vs. ≥10) 180 vs. 90 mg/d, P = 0.01 • Selenium supplement (for EPDS <10 vs. ≥10) 25 vs. 19 mcg/d P = 0.002 Full model: • Selenium supplement intake OR = 0.97 (CI 0.95–0.99), P = 0.03 • Mean nutrient intakes from other supplements higher in women with low EPDS scores but NS.	Final model: • Selenium supplement intake OR = 0.76 (CI 0.74–0.78); P = 0.02	Age, parity, marital status, education, income, born outside Canada, ethnicity, BMI, stressors during pregnancy, stressors prior to pregnancy, stressors prior to age 17, social support*, prenatal EPDS scores* and prenatal selenium supplement*	• 90% above RDA, very few women deficient • Multi‐colinearity in sets of nutrients commonly combined in supplements • Food intake not measured • No history of depression included
Fowles et al. (2011a)	18	3 × 24 h quantitative food recall including intake of grains (oz/d), fruit (cups/d), vegetables (cups/d), energy (kcal/d), % kcal from fat, calcium (mg/d), folate (mg/d) and iron (mg/d)	Spearman's rho	Mean nutrient intake	CES‐D score; linear	Higher depression score associated with higher calcium intake: • Spearman's rho 0.60, P = 0.02	Not reported	Not included in univariable or adjusted model	• No adjusted model of the association of interest • No potential confounders included in model
Bae et al. (2010)	114	Energy (kcal/d), protein (g/d), fat (g/d), carbohydrate (g/d), fiber (g/d), total Ca(mg/d), plant Ca (mg/d), animal Ca (mg/d), P (mg/d), total Fe (mg/d), plant Fe (mg/d), animal Fe (mg/d), supplement Fe (mg/d), Na (mg/d), K(mg/d), Vit A (mcg/d), B1 (mg/d), B2 (mg/d), B3 (mg/d), B6 (mg/d), Total B9 (mg/d), dietary B9 (mg/d), supplement B9 (mg/d), C (mg/d), E (mg/d), Cholesterol (mg/d), PUFA (g/d), MUFA (g/d), SFA (g/d), PUFA/MUFA/SFA ratio	T‐test and chi‐squared test	Mean nutrient intake	BDI score ≥10 vs. <10 BDI ≥10 vs. BDI <10:	• Higher intake of energy, protein, fat, in non‐depressed group, but NS Significant associations: • Total Ca: 648.33 (SD 252.11) vs. 533.08 (SD 251.52); p=0.02 • Plant Ca: 308.50 (SD 118.78) vs. 252.80 (SD 117.39); p=0.01 • Plant iron: 11.43 (SD 8.56) vs. 8.33 (SD 3.02); p=0.01 • K: 3,073.10 (SD 1192.30) vs. 2,430.30 (SD 881.11); p=0.001 • Total folate: 281.15 (SD 113.99) vs. 230.47 (SD 89.16) p=0.04 • Dietary folate: 288.07 (SD 128.93) vs. 229.26 (SD 91.94); p=0.006	Not reported	Tested for association with the outcome, but not included in final model: age, pre‐pregnancy BMI, pregnancy BMI, delivery birth BMI, parity*, education, income, occupation, morning sickness, nutritional supplements	• Cut‐point for depression based on mean, not validated cut‐point for screening depressed vs. non‐depressed • Assessment timing not standard for all participants • No adjusted model of the association of interest
Lukose et al. (2014)	365	FFQ to quantify energy (kcal/d), protein (g/d), fat (g/d), Vit C (mg/d), Vit B9 (mcg/d), Vit B12 (mcg/d), Ca (mg/d), P (mg/d), Fe (mg/d), Cu (mg/d), Mn (mg/d) and Zn (mg/d)	Chi‐squared test; multivariable log binomial regression	Mean nutrient intake	K‐10; ≥6 vs. <6	Nutrient intakes NS	Not reported	Age, education, occupation, parity, BMI, anaemia*, nausea and vomiting*	• No adjusted model of the association of interest • No history of depression considered • Only univariable tests of nutrients, no effect sizes or p‐values reported if p > 0.05 and not included in adjusted model.
Watanabe et al. (2010)	86	Brief self‐administered diet history questionnaire (BDHQ) to quantify intake of total energy, % energy from protein, fat, carbohydrate, Ca, Fe, Zn, Vit B9, B6, B12 and C	t‐test, chi‐squared test and logistic regression	Mean nutrient intake	CES‐D <16 vs. ≥ 16	Iron (mg/1000 kcal) • Non‐depressed group: 4.3 (SD 0.9) • Depressed group: 3.9 x(SD 0.8); P = 0.06 Other nutrient intakes NS but higher intake in non‐depressed group	Folate, Vitamins B6, B9 and B12 intake included; NS	Age, height, pre‐pregnancy weight, pre‐pregnancy BMI, gestational age*, parity* and vomits per day*.	• Unrealistically high depression prevalence • BDHQ not a robust FFQ • Depression measured in first trimester • No history of depression considered

Admit, hospital admission for depression; B1, thiamin; B2, riboflavin; B3, niacin; B5, pantothenic acid; B6, pyridoxine; B7, biotin; B12, cobalamin; B9, folic acid; BDI, Beck's Depression Inventory; BMI, body mass index; Ca, calcium; CES‐D, Center for Epidemiological Studies Depression Scale; CI, confidence interval; Cu, copper; EPDS, Edinburgh Post‐partum Depression Scale; Fe, iron; K‐10, Kessler Depression Scale; Mg, magnesium; Mn, manganese; n‐3 PUFA, omega‐3 polyunsaturated fatty acids; n‐6 PUFA, omega‐6 polyunsaturated fatty acids; NS, not significant; OR, odds ratio; P, phosphorus; PUFA, polyunsaturated fatty acid; RCT, randomized control trial; Rx, prescription for antidepressants; Se, selenium; Vit, Vitamin; Zn, zinc; g, grams; mg, milligrams; mcg, micrograms; oz, ounces; kcal, kilocalorie.

^*Significant (P < 0.05) in final model.

Appendix Table 3. Results of studies evaluating associations between intake of specific nutrient groups and perinatal depression

Author (year)	n	Nutrients/diet (components of analysis)	Analysis	Unit of exposure	Unit of outcome	Unadjusted associations significant at P < 0.10 (effect size, 95% CI, P‐value)	Adjusted associations significant at P < 0.10 (effect size, 95% CI, P‐value)	Confounding considered	Limitations
B VITAMINS
Blunden et al. (2012)	2856	FFQ to quantify intake of folate (B9), Vit B6 and B12	Multivariable Poisson regression	Strata of intake, per 10 units	EPDS score ≥13 vs. <13	Nutrient intakes NS at any time point	Not reported	History of mental illness*, education, scoring positive for depression on the GHQ*, smoking, alcohol, social benefits, financial strain, breastfeeding*, folic acid supplements and pregravid and early pregnancy red‐cell folate.	• Less than 5% of study population with low/marginal folate • PPD screening done on average at 1 year PP
Lewis et al. (2012)	6809	Self‐reported folic acid supplementation during pregnancy	Multivariable linear regression, stratification by genotype	Supplementation vs. no supplementation	EPDS score; linear	Supplementation NS within 1 year post‐partum.	Not reported	Age, pre‐pregnancy BMI, education and parity, and other confounders tested by MTHFR C677T genotype stratification	• Self‐reported supplementation history, no amount or frequency • Women taking multivitamins were included in unexposed group • Folate deficiency in population not known • No history of depression considered
Miyake, et al. (2006a)	865	FFQ to quantify intake of folate (B9), Vit B2, B6 and B12	Multivariable logistic regression	Quartiles of intake (Q1: lowest/reference, Q2, Q3, Q4: highest)	EPDS score; ≥9 vs. <9	Vitamin B2 intake: Q1: ref. • Q2: ref. OR 0.58 (CI 0.33–0.98) • Q3: OR 0.50 (CI 0.28–0.87) • Q4: OR 0.82 (CI 0.49–1.35) • P for trend = 0.35 Other B vitamins NS in any quartile	Vitamin B2 intake: Q1: ref. • Q2: OR 0.61 (CI 0.34–1.06) • Q3: OR 0.53 (CI 0.29–0.95) • Q4: 0.88 (CI 0.51–1.50) • P for trend = 0.55 Other B vitamins NS in any quartile	Age, gestation, parity, smoking, family structure, occupation, family income, education, changes in diet in previous month, season of baseline data, BMI, time of delivery, problems during pregnancy, baby's sex and baby's birth weight.	• Assessment timing not standard for all participants • Potential confounders not included in model • No history of depression considered
Cho et al. (2008)	1277	Multivitamin and folic acid supplementation (MV/FA) during pregnancy	Chi‐squared test	Supplementation vs. no supplementation	Goldberg's Depression Scale ≥22 vs. <22	MV/FA intake in preconception period NS: • 10.7% vs. 7.9%, OR 1.4 (CI 0.77–2.54) MV/FA intake at any time in pregnancy NS: • 9.4% vs. 6.9%, OR 0.72 (CI 0.48–1.08)	Not reported	Tested for association with the outcome but not included in final model: age, gestational age, pregnancy intended, education, income, family history of depression* and history of premenstrual syndrome*	• Nutrition measure not robust (yes/no for MV/FA supplementation) • No adjusted model of the association of interest
CALCIUM
Harrison‐Hohner et al. (2001)	293	Group A: 2000 mg elemental calcium per day Group B: placebo	Comparison of means	Calcium vs. placebo group	EPDS score; ≥14 vs. <14; mean difference	% of calcium vs. placebo group ≥14 on EPDS groups: • 6 weeks: Oregon: 11% vs. 18%, P = 0.07 New Mexico: 24% vs. 21%, P > 0.05 • 12 weeks, Oregon only: 6% vs. 15%, P = 0.014 Mean EPDS scores (12 weeks, Oregon only): • 6.4 (SEM 0.38) vs. 7.3 (SEM 0.45), P >0.05	Not reported	Baseline differences in age, education, history of depression, gestation, birth weight of baby, breastfeeding, life stress (OR only), serum Vit D, calcitonin and parathyroid hormone tested between groups, NS	• High loss to follow‐up (29% completed in NM, 60% completed in OR) • Study not designed/powered for post hoc analysis of depression • Blinding and selective reporting information not available • Study sites not comparable
Miyake, et al. (2014a)	1745	FFQ to quantify intake of total energy, total dairy, milk, yoghurt, cheese, calcium, fish, SFAs, EPA+DHA, Vitamin D	Multivariable logistic regression	Quartiles of intake (Q1: reference, Q2, Q3, Q4: highest)	CESD score; ≥16 vs. <16	• Yoghurt: Q1: ref. • Q2: OR 0.81 (CI 0.58–1.11) • Q3: OR 0.66 (CI 0.47–0.92) • Q4: OR 0.68 (CI 0.49–0.95) • P for trend = 0.01 Calcium: Q1: ref. • Q2: OR 0.79 (CI 0.57–1.10) • Q3: OR 0.67 (CI 0.48–0.93) • Q4: OR 0.68 (CI 0.49–0.95) • P for trend = 0.01 Other nutrient variables NS	• Yoghurt: Q1: ref. • Q2: OR 0.80 (CI 0.57–1.13) • Q3: OR 0.70 (CI 0.49–0.99) • Q4: OR 0.69 (CI 0.48–0.99) • P for trend = 0.03 Calcium: Q1: ref. • Q2: OR 0.75 (CI 0.53–1.06) • Q3: OR 0.63 (CI 0.44–0.92) • Q4: OR 0.59 (CI 0.40–0.88) • P for trend = 0.006 Other nutrient variables NS	Age*, gestational age, residential area, parity*, family structure, personal history of depression, family history of depression, ever smoked*, smoke exposure, job type, household income*, education*, BMI, intake of fish, SFAs, EPA, DHA and Vitamin D	• Assessment timing not standard for all participants • Participation and rate and characteristics not available
VITAMIN D
Miyake et al. (2015b)	1745	FFQ to quantify intake of Vit D (mcg/d)	Multivariable logistic regression	Quartiles of intake (Q1: reference, Q2, Q3, Q4: highest)	CESD score; ≥16 vs. <16	Vit D intake: Q1: ref. • Q2: OR 0.80 (CI 0.58–1.10) • Q3: OR 0.73 (CI 0.53–1.01) • Q4: OR 0.51 (CI 0.36–0.72) • P for trend = 0.001	Vit D intake: Q1: ref. • Q2: OR 0.79 (CI 0.55–1.11) • Q3: OR 0.73 (CI 0.49–1.07) • Q4: OR 0.52 (CI 0.30–0.89) • P for trend = 0.02		• Assessment timing not standard for all participants • Participation and rate and characteristics not available • Seasonal variance and sunlight exposure not measured
ZINC
Roy et al. (2010)	2030	FFQ and supplement questionnaire to quantify zinc intake (mg/d)	Multivariable linear regression	Quintiles of intake (mg/d): Q1 (lowest): <11.52 Q2: 11.52–14.82 Q3: 14.83–17.28 Q4: 17.29–19.80 Q5 (highest): reference ref.	EPDS score; linear	Zinc intake: Q5: ref. • Q4: β 0.03 (SE 0.5), P = 0.95 • Q3: β −0.02 (SE 0.5), P = 0.97 • Q2: β 1.3 (SE 0.5), P = 0.02 • Q1: β 1.1 (SE 0.5), P = 0.04	Zinc intake: Q5: ref. • Q4: β 0.03 (SE 0.5), P = 0.94 • Q3: β 0.4 (SE 0.5), P = 0.38 • Q2: β 0.8 (SE 0.5), P = 0.07 • Q2: β 1.0 (SE 0.5), P = 0.03	Education*, marital status*, parity*, employment*, income, preexisting medical conditions, age, shift work and stress	• 75% took supplements during pregnancy • Effect size very small • History of depression not included in model

B1, thiamin; B2, riboflavin; B3, niacin; B6, pyridoxine; B7, biotin; B9, folic acid; B12, cobalamin; BMI, body mass index; CESD, Center for Epidemiological Studies Depression Scale; CI, confidence interval; EPDS, Edinburgh Post‐partum Depression Scale; MV/FA, multivitamin and folic acid supplementation; NS, not significant; OR, odds ratio; SEM, standard error of the mean; g, grams; mg, milligrams; mcg, micrograms.

^*Significant (P < 0.05) in stratum or final model.

Appendix Table 4. Results of studies evaluating associations between intake of fish and fats and perinatal depression

Author (year)	n	Nutrients/diet (components of analysis or intervention)	Analysis	Unit of exposure	Unit of outcome	Unadjusted associations significant at p < 0.10 (effect size, 95% CI, P‐value)	Adjusted associations significant at p < 0.10 (effect size, 95% CI, P‐value)	Confounding considered	Limitations
Doornbos et al. (2009)	119	Group A: DHA+AA (220 mg) Group B: DHA (220 mg) Group C: placebo	ANOVA and chi‐squared tests	DHA+AA vs. DHA vs. placebo group	EPDS; ≥12 vs. <12	Mean differences between average EPDS scores NS at any time point	Not reported	Baseline differences in fish intake, social and obstetric characteristics tested between groups, NS	• High loss to follow‐up (31%) • Only powered for large differences • Low PPD prevalence found • Only 60 women completed PP blues evaluation; 16/60 had PP blues • Low dose of DHA compared with other studies • No history of depression considered
Judge et al. (2014)	42	Group A: DHA (300 mg DHA, fish oil) 5 days/week Group B: placebo	t‐test and chi‐squared test, repeated measures analysis	DHA group vs. placebo group	CES‐D: mean PDSS: mean	DHA group vs. placebo group means and standard error of means (SEM): • 2 weeks: 47.65 (12.96) vs. 53.86 (15.25) • 6 weeks: 47.61 (14.31) vs. 47.40 (12.42) • 3 months: 45.28 (12.25) vs. 42.63 (9.52) • 6 months: 45.55 (13.50) vs. 48.42 (17.18) • Repeated measures LS mean: 46.03 (2.17) vs. 52.11 (2.4), P = 0.016	Not reported	Baseline differences in age, pre‐pregnancy weight, number of live births, income, education, ethnicity, household size, history of depression or previous treatment for depression tested between groups, NS	• Low response rate (58%) and high loss to follow‐up (42% lost at 6= months)
Llorente et al. (2003)	138	Group A: DHA (200 mg/d) Group B: placebo	t‐test and chi‐squared test, regression analysis	DHA group vs. placebo group	BDI; ≥10 vs. <10 EPDS and clinical interview in subsample	NS at any time point, for any measure (BDI, EPDS, SCID‐CV)	Not reported	Baseline differences in age, education, parity, gestational age, delivery weight, pre‐pregnancy weight, ethnicity, caesarean delivery, baby's sex, Apgar score tested between groups, NS	• High loss to follow‐up (27%) • Powered to detect 30% difference in groups • Low PPD prevalence found • Low dose and short duration of intervention • No history of depression considered
Makrides et al. (2010)	2399	Group A: 800 mg/d of DHA, 100 mg/d of EPA Group B: vegetable oil capsules without DHA	Log binomial regression	DHA group vs. placebo group	EPDS score; ≥12 vs. <12	• RR 0.86 (CI 0.71–1.05), P = 0.15 (9.67% in supplement group vs. 11.19% in placebo group)	• RR 0.85 (CI 0.70–1.02), P = 0.09	Baseline differences in maternity centre, parity, Maternal Social Support Index score, age, history of depression and smoking status tested between groups, NS
Da Rocha and Kac (2012)	106	FFQ to quantify mean n‐6/n‐3 PUFA intake ratio (mg/d), also tested energy (kcal/d), carbohydrate (g/d), lipids (g/d), protein (g/d), n‐3 PUFA intake (g/d) and n‐6 PUFA intake (g/d)	Multivariable Poisson regression	n‐6/n‐3 PUFA ratio >9:1	EPDS score; ≥11 vs. <11	n‐6/n‐3 PUFA ratio in EPDS score ≥11 vs. <11: • Prevalence ratio: 2.73 (CI 1.44–5.18), P = 0.002 • All other nutrient variables NS	n‐6/n‐3 PUFA ratio in EPDS score ≥11 vs. <11: • Prevalence ratio: 2.50 (CI 1.21–5.14), P = 0.01	Income, age, schooling*, skin colour, marital status, smoking, alcohol, age at menarche, delivery type, LBW, prematurity, marital problems, stress, parity and pre‐pregnancy BMI*	• High loss to follow‐up • Small sample size in second and third trimesters • No history of depression considered
Miyake, Y., et al. (2006b)	865	FFQ to quantify fish, meat, eggs, dairy, total fat (g/d), SFA, MUFA, n‐3 PUFA, n‐6 PUFA, LA, ALA, AA, EPA, DHA and ratio of n‐3 to n‐6 PUFA (mg/d)	Multivariable logistic regression	Quartiles of intake (Q1: lowest/reference, Q2, Q3, Q4: highest)	EPDS score; ≥9 vs. <9	• Increased intake of several nutrient variables protective but NS • Trends NS	• Increased intake of several nutrient variables protective but NS • Trends NS • Similar ORs for unadjusted and adjusted	Age, gestation, parity, smoking, family structure, occupation, family income, education, changes in diet in previous month, season of baseline data, BMI, time of delivery, problems during pregnancy, baby's sex and baby's birth weight	• Assessment timing not standard for all participants • No history of depression considered
Strom et al. (2009)	54 202	FFQ to quantify fish intake (g/d, quintiles), n‐3 PUFA intake (mg/d, deciles)	Multivariable logistic regression	Intake groups (g/d): • G1: 0–3 • G2: >3–10 • G3: >10–20 • G4: >20–30 • G5: >30 Deciles of n‐3 PUFA intake (D1, lowest‐D10, highest)	PPD admit	• NS for fish or n‐3 PUFA	• NS for fish or n‐3 PUFA	Age, parity, marital status, smoking, alcohol, pre‐pregnant BMI, occupation, education, home ownership, social support and previous depression	• Depression measured in severe clinical manifestation, very specific, not very sensitive
					PPD Rx	Fish intake: • G1: OR 1.61 (CI 1.26–2.06) • G2: OR 1.11 (CI 0.90–1.38) • G3: group OR 1.14 (CI 0.94–1.40) • G4: group OR 1.01 (CI 0.80–1.28) • G5: reference • P for trend = 0.001 n‐3 PUFA intake: • D1: OR 1.41 (CI 1.10–1.81) • D2: OR 1.24 (CI 0.96–1.60) • D3: OR 1.05 (CI 0.80–1.37) • D4: OR 1.31 (CI 1.02–1.68) • D5: OR 1.06 (CI 0.81–1.38) • D6: OR 1.09 (CI 0.83–1.42) • D7: OR 1.02 (CI 0.78–1.34) • D8: OR 0.89 (CI 0.67–1.18) • D9 and D10: reference • P for trend = 0.04	Fish intake: • G1: OR 1.46 (CI 1.12–1.90) • G3: OR 1.10 (CI 0.87–1.38) • G2: OR 1.18 (CI 0.95–1.45) • G4: OR 1.03 (CI 0.81–1.32) • G5: reference • P for trend = 0.04 n‐3 PUFA intake: • D1: OR 1.24 (CI 0.96–1.61) • D2: OR 1.17 (CI 0.90–1.53) • D3: OR 0.99 (CI 0.75–1.31) • D4: OR 1.29 (CI 0.99–1.68) • D5: OR 1.09 (CI 0.83–1.44) • D6: OR 1.11 (CI 0.84–1.46) • D7: OR 1.04 (CI 0.79–1.38) • D8: OR 0.89 (CI 0.67–1.20) • D9 and D10: reference • P for trend = 0.33
Browne et al. (2006)	80	FFQ to quantify mean PUFA intake (mg/d)	Logistic regression, F‐test and t‐test	Some fish (any freq.) vs. none	EPDS score; ≥9 vs. <9	F‐test for nutrients in logistic regression NS (P = 0.29)	Not reported	Alcohol, breastfeeding*, dietary supplement use, education, ethnicity, fish consumption, household income* and smoking	• No adjusted model of the association of interest • No participants consumed oily fish regularly
Cosatto et al. (2010)	94	FFQ to quantify intake of EPA, DPA, DHA and total n‐3 PUFA (mg/d)	Wilcoxon test	Mean nutrient intake	EPDS ≥10 vs. <10	NS for any nutrient	Not reported	None	• No adjustment for confounding factors • No adjusted model of the association of interest
Golding et al. (2009)	9960	FFQ to quantify n‐3 PUFA intake from fish (g/week)	Multivariable logistic regression	Groups: g/week • G1: None • G2: 0.1–0.4 • G3: 0.4–1.5 • G5: >1.5	EPDS; ≥13 vs. <13	Groups: • G1: OR 1.97 (CI 1.63–2.38) • G2: OR 1.64 (CI 1.37–1.96) • G3: OR 1.31 (CI 1.13–1.52) • G4: reference • P for trend ≤ 0.01	Groups: • G1: OR 1.54 (CI 1.25–1.89) • G2: OR 1.37 (CI 1.13–1.66) • G3: OR 1.20 (CI 1.03–1.41) • G4: reference • P for trend ≤ 0.01	Age, parity, outcome of immediately preceding pregnancy, education, housing, crowding, events in childhood, recent events, chronic stress, smoking, alcohol, ethnicity and energy intake	• High levels of depressive symptoms were associated with not returning FFQ questionnaire
Miyake et al. (2013)	1745	FFQ to quantify mean total meat, total fish, total fat, SFAs, MUFAs, n‐3 PUFA, n‐6 PUFA, LA, ALA, AA, EPA, DHA, ratio of n‐3 to n‐6 PUFA and cholesterol (mg/d)	Multivariable logistic regression	Quartiles of intake (Q1: lowest/reference, Q2, Q3, Q4: highest)	CES‐D score; ≥16 vs. <16	Fish intake: • Q1: ref. • Q2: OR 0.77 (CI 0.56–1.07) • Q3: OR 0.79 (CI 0.57–1.10) • Q4: OR 0.58 (CI 0.41–0.82) • P for trend = 0.003 EPA: • Q1: ref. • Q2: OR 0.91 (CI 0.66–1.26) • Q3: OR 0.77 (CI 0.55–1.06) • Q4: OR 0.60 (CI 0.42–0.84) • P for trend = 0.002	Fish intake: • Q1: ref. • Q2: OR 0.72 (CI 0.51–1.02) • Q3: OR 0.79 (CI 0.56–1.10) • Q4: OR 0.61 (CI 0.42–0.87) • P for trend = 0.01 EPA: • Q1: ref. • Q2: OR 0.93 (CI 0.66–1.30) • Q3: OR 0.79 (CI 0.56–1.12) • Q4: OR 0.66 (CI 0.46–0.95) • P for trend = 0.02	Age, gestational age, residential area, parity, family structure, personal history of depression, family history of depression, ever smoked, second‐hand smoke exposure, job type, household income, education and BMI	• Assessment timing not standard for all participants • Participation and rate and characteristics not available
						DHA: • Q1: reference • Q2: OR 1.06 (CI 0.77–1.45) • Q3: OR 0.84 (CI 0.60–1.17) • Q4: OR 0.60 (CI 0.42–0.85) • P for trend = 0.001 Total fat: • Q1: reference • Q2: OR 1.17 (CI 0.83–1.66) • Q3: OR 1.14 (CI 0.80–1.61) • Q4: OR 1.46 (CI 1.04–2.04) • P for trend = 0.03 SFA:0 • Q1: reference • Q2: OR 1.16 (CI 0.81–1.66) • Q3: OR 1.29 (CI 0.91–1.83) • Q4: OR 1.66 (CI 1.19–2.34) • P for trend = 0.002	DHA: • Q1: reference • Q2: OR 1.06 (CI 0.75–1.48) • Q3: OR 0.87 (CI 0.61–1.23) • Q4: OR 0.64 (CI 0.44–0.93) • P for trend = 0.007 Total fat: • Q1: reference • Q2: OR 1.18 (CI 0.82–1.69) • Q3: OR 1.12 (CI 0.78–1.62) • Q4: OR 1.42 (CI 1.00–2.03) • P for trend = 0.06 SFA: • Q1: reference • Q2: OR 1.13 (CI 0.78–1.64) • Q3: OR 1.28 (CI 0.89–1.84) • Q4: OR 1.74 (CI 1.22–2.49) • P for trend = 0.001
						MFA: • Q1: reference • Q2: OR 1.08 (CI 0.77–1.53) • Q3: OR 0.94 (CI 0.66–1.33) • Q4: OR 1.46 (CI 1.05–2.03) • P for trend = 0.03 ALA: • Q1: reference • Q2: OR 0.90 (CI 0.63–1.27) • Q3: OR 0.87 (CI 0.61–1.23) • Q4: OR 1.38 (CI 1.00–1.92) • P for trend = 0.04 n‐3/n‐6 PUFA ratio: • Q1: reference • Q2: OR 1.07 (CI 0.78–1.48) • Q3: OR 0.78 (CI 0.55–1.09) • Q4: OR 0.74 (CI 0.52–1.04) • P for trend = 0.03 All other nutrient variables NS	MFA: • Q1: reference • Q2: OR 1.02 (CI 0.71–1.45) • Q3: OR 0.91 (CI 0.63–1.31) • Q4: OR 1.38 (CI 0.98–1.96) • P for trend = 0.08 ALA: • Q1: reference • Q2: OR 0.90 (CI 0.63–1.29) • Q3: OR 0.83 (CI 0.57–1.18) • Q4: OR 1.30 (CI 0.92–1.82) • P for trend = 0.12 n‐3/n‐6 PUFA ratio: • Q1: reference • Q2: OR 1.12 (CI 0.80–1.57) • Q3: OR 0.83 (CI 0.58–1.18) • Q4: OR 0.83 (CI 0.58–1.19) • P for trend = 0.15 All other nutrient variables NS
Sontrop et al. (2008)	2394	FFQ to quantify intake of fish	Multivariable linear regression	Intake of fish: <1 vs. ≥1 serving per week; EPA+DHA: < 85 vs. ≥85 mg/d	Mean CES‐D score	Difference in CES‐D score between intake categories: • Fish intake: −0.8 (CI −1.5 to −0.1); P < 0.05 • EPA+DHA intake: −0.8 (−1.4 to −0.1); P < 0.05 Subgroup analysis for EPA+DHA: • Smokers: −3.0 (−5.1 to −0.9); P < 0.05; • Single status: −3.5 (−5.9 to −1.1); P < 0.05	Difference in CES‐D score between intake categories: • Fish intake: −0.2 (CI −0.9 to 0.4); NS • EPA+DHA intake: not reported; NS Subgroup analysis for EPA+DHA: • Smokers: −2.4 (−4.4 to −0.4); P < 0.05 • Single:−2.8 (−5.1 to −0.4); P < 0.05	Age, marital status*, education*, income*, occupation*, smoking*, physical activity* and meeting Canada Food Guide to Healthy Living guidelines	• Not all potential confounders included in model • No history of depression

AA, arachidonic acid; Admit, hospital admission for depression; ALA, alpha‐linolenic acid; BDI, Beck's Depression Inventory; BDHQ, Brief Diet History Questionnaire; BMI, body mass index; CES‐D, Center for Epidemiological Studies Depression Scale; CI, confidence interval; DHA, docosahexaenoic acid; DPA, docosapentaenoic acid; EPDS, Edinburgh Post‐partum Depression Scale; EPA, eicosapentaenoic acid; FFQ, food frequency questionnaire; LA, linolenic acid; LS, least square; n‐3 PUFA, omega‐3 polyunsaturated fatty acids; n‐6 PUFA, omega‐6 polyunsaturated fatty acids; NS, not significant; OR, odds ratio; PDSS, Post‐partum Depression Screening Scale; PPD, post‐partum depression; PP, post‐partum; Preg, pregnancy; PUFA, polyunsaturated fatty acid; RDA, Recommended Daily Allowance; RR, risk ratio; Rx, filled prescription for antidepressants; SCID‐CV, Structured Clinical Interview; SEM, standard error of the mean; g, grams; mg, milligrams; mcg: micrograms.

^*Significant (P < 0.05) in stratum or in final model.

Sparling T. M., Henschke N., Nesbitt R. C., and Gabrysch S. (2017) The role of diet and nutritional supplementation in perinatal depression: a systematic review, Maternal & Child Nutrition, 13, e12235. doi: 10.1111/mcn.12235.