A systematic review of screening instruments for depression for use in antenatal services in low resource settings

A limited search of the Cumulative Index of Nursing and Allied Health Literature (CINAHL) and Medline was undertaken to identify relevant keywords contained in the title, abstract, and subject descriptors. Search terms and synonyms were then identified for use in searching different databases for screening studies conducted in antenatal clinics in low resource settings. Low resource settings refer to settings where health care systems do not meet the minimum standards set by the World Health Organisation (WHO) or any other quasi-governmental organisation [11]. In this review, low resource settings were defined as health care settings synonymous with those found in low income and lower middle income countries as defined by World Bank [12] and some health care settings in upper middle income countries (UMICs), such as South Africa, where disparities in the public health infrastructure or supplies or human resources [13] are found. Some articles from low resource settings are not indexed to indicate that they are reporting about health outcomes or disparities for under-served populations in low resource settings [14] and the term, ‘low resource settings’, was not included in the search terms but applied manually at the article review stage. Date limits were set from 2000 to 2015 in anticipation that a wider period to be searched will yield many relevant studies with recent evidence. Detailed search terms are supplied in Table 1.

The following databases were searched: ScienceDirect, CINAHL, MEDLINE, PubMed, SABINET and PsychARTICLES and results were imported into Endnote. Reference lists of key articles identified were hand searched to identify further relevant articles. Manual searches of indexes and “grey” literature databases were not carried out. The preliminary searches were conducted between August and September 2015 and the final search was done on 4th September 2015.

After the initial search, duplicates and irrelevant articles (conferences, congresses, editorials, commentaries, reviews, news, old) in the Endnote database were removed and the search data were exported to Excel. Articles for review were then selected in three phases.

The Quality Assessment of Diagnostic Accuracy Studies (QUADAS) [17] was used by both reviewers to assess the psychometric quality of the final selected articles. The QUADAS has 14 items with three possible responses ‘Yes’, ‘No’ and ‘Unclear’. In the QUADAS, the target condition was depression during pregnancy, the index test was a screening instrument used to screen for depression, and the reference standard was the gold standard against which the index test was validated. The QUADAS items measure the variability of study samples (items 1–2), methodological rigor and bias (items 3–7, 10–12 and 14), and the quality of reporting methodology (items 8, 9 and 13). The scoring of QUADAS is not standardised [18] but studies were categorised as ‘excellent’ (11 to 14 items), ‘good’ (9 to 10 items), ‘adequate’ (6 to 8 items), ‘poor’ (4 to 5 items) or ‘unacceptable’ (0 to 3 items) based on the number of items that were answered ‘Yes’ [17].

Descriptive data extraction and presentation was done to compare screening instruments’ psychometrics data in a between-study literature analysis [19]. A meta-analysis was conducted using REVMAN by pooling individual and all instruments sensitivity and specificity data to show the pooled ability of the screening instruments to identify depression. Upper and lower confidence intervals (95%) for sensitivity and specificity of screening instruments were calculated.

The electronic search yielded 3666 published articles (Fig. 1). Eleven (11) additional articles were sourced from authors on ResearchGate and reference lists of full text articles resulting in a total number of 3677 published articles. A total of 1676 duplicates were removed leaving 2001 articles. Irrelevant articles consisting of conferences, congresses, editorials, commentaries, reviews, news and old articles (≤ 1999) were removed (n = 1750), leaving 251 articles. The 251 articles which remained were then screened for relevancy by the reviewers using the PICOS criteria, excluding a further 210 articles [Participants (n = 133), Outcome (n = 21) and HICs articles (n = 28)], leaving 41 articles (38 primary research studies and 3 systematic reviews). The reviewers’ ratings were in agreement with a Kappa = .97.

The systematic reviews (n = 3) were excluded after being screened for relevancy for inclusion in this review. One systematic review [20] focused on the efficacy of antenatal group interventions aimed at reducing postnatal depression in at risk women. This systematic review did not report any validity data of the depression screening instruments and thus was excluded. The second systematic review by Akena and colleagues [21] examined the accuracy of depression screening instruments validated in general health settings in low and middle income countries (LMICs). This systematic review included three studies conducted in antenatal settings [4, 22, 23] which also had been identified as part of the 38 articles for primary studies in our review. The third systematic review focused on the reliability and validity of instruments for screening perinatal depression in African settings [24]. This systematic review included eight articles for studies which were conducted in antenatal settings of which four [3, 25, 26] were included in the 38 primary articles in our review. The other four articles [27‐30] were published before 2000 and were excluded due to the time limits of the search terms. Further review of the full texts of the 38 articles showed that two pairs of articles [25, 31] and [3, 26]] reported the same data from two different studies and one article from each pair was retained resulting in 36 articles included for further review.

The study characteristics of the 36 selected studies for further review are provided in Table 2. The majority of the studies were published between 2010 and 2015 and only one study was published in a nursing journal. Most of the articles (n = 18) were cross sectional prevalence studies and five (n = 5) were psychometric validation studies measuring reliability and validity of screening instruments. In reviewing these studies for reported psychometrics of sensitivity, specificity, Area under the curve and the relevant gold standards, two studies [32, 33] were excluded (no gold standard as defined by this study) and a further 23 studies were excluded due to inadequate reporting of psychometrics. One third of the articles (n = 11) reported psychometrics and a gold standard and met the final selection criteria for inclusion in the review (Table 2).

All 11 articles were published in medical journals, mostly from 2010 onwards (n = 8). A number of articles were validation studies (n = 5) that reported psychometrics (reliability and validity). There were also 4 cross sectional prevalence studies (n = 4), one prospective study and one randomised trial. These last-mentioned 6 studies generally reported on prevalence of prenatal depression and risk factors but included psychometric properties of the screening instruments. All the screening instruments reported in the selected articles were adapted by translating them to local languages in each setting.

All 11 articles were rated for quality by both reviewers. Overall the quality was satisfactory with six articles [1, 23, 25, 34‐36] rated as excellent, three [37‐39] good and two [3, 4] adequate. All the articles clearly described the selection criteria for the sample and reported the index test as independent of the gold standard. All articles, except one [39], regardless of overall quality, used random samples. The two articles rated as ‘adequate’ [3, 4] did not sufficiently report the execution of a gold standard and it was difficult to ascertain whether individuals who administered index tests or gold standards were blinded to each other’s results. Articles with ‘excellent’ quality were the psychometric validation studies and the randomised controlled trial.

The articles included seven (n = 7) screening tools, namely the Beck Depression Index (BDI), Centre for Epidemiologic Studies Depression Scale (CES-D)-20, Edinburgh Postnatal Depression Scale (EPDS), Hamilton Rating Scale for Depression (HAM-D), Hopkins Symptoms Checklist (HSCL)-25, Kessler Psychological Distress Scale (K-10) and Self-Reporting Questionnaire (SRQ) that were used for screening antenatal depression in low resource settings (Table 3). The BDI and HAM-D are not normally used for diagnostic purposes or screening purposes but to estimate the severity of depression for the past 3 or 7 days. EPDS was designed for use in postnatal period and it has been investigated for antenatal use as well.

Seven studies (n = 7) used a single screening instrument while four (n = 4) used a combination of two or three instruments. The EPDS was the most widely used instrument (8 studies), followed by the BDI and K-10 (2 studies each). The MINI was the most widely used gold standard being used in five of the 11 studies. In assessing the accuracy of screening instruments in detecting depression among pregnant women, an AUC score range is classified as low (.500 to .700), moderate (>.700 to .900) and high (>.900) [40]. The EPDS had the highest level of accuracy (AUC = 0.965) while K-10 had the lowest level of accuracy (AUC = .660). The BDI, CES-D, HAM-D, HSCL-25 and SRQ had moderate accuracy with AUC ranges from .820 to .900. A forest plot showed that the included studies were heterogeneous because error bars for sensitivity and specificity plots did not include the summary values-sensitivity of .82 and specificity of .79 (Fig. 2). As such 5 distinct subgroups based on participants or type of instrument were formulated and graphical test using forest plots showed that one EPDS studies subgroup of all pregnant women was heterogeneous while other four were homogeneous (Figs. 3, 4 and 5). Schriger and colleagues recommended that a forest plot should consist of a minimum of two studies and discourages conducting heterogeneity tests when there are less than five studies [41].

One of the key strengths of the review is the specific evidence on screening tools used in antenatal services in low resource settings. It may serve as an efficient, reliable and comprehensive source of evidence for decision-makers in low resource settings [14] since most evidence, generated from HICs, may not be applicable in low resource settings. A limitation of this review is that restrictions on language and date limits may have resulted in missing out some relevant articles.