Study design

This paper assesses the economic burden of maternal depression from a societal perspective^{Reference Drummond, Sculpher, Claxton, Stoddart and Torrance13} by comparing patient costs, public provider costs and economic outcomes of pregnant and postpartum women according to their depression symptoms. Women were categorised as having depression if they scored >8 on the HRSD.^{Reference Zimmerman, Martinez, Young, Chelminski and Dalrymple12} Costs and economic outcomes were modelled over pregnancy and up to 12 months postpartum. Data for this analysis were collected within an individual-level randomised controlled trial. Trial registration: ClinicalTrials.gov: NCT01977326, registered on 24 October 2013; Pan African Clinical Trials Registry (www.pactr.org): PACTR201403000676264, registered on 11 October 2013.

The trial aimed to test a task-sharing intervention for women with symptoms of maternal depression. Women were randomised to one of two arms: (a) a structured manual-based psychological intervention including six counselling sessions conducted by community health workers trained in counselling techniques; or (b) enhanced usual care consisting of regular monthly telephone calls over a 3-month period, conducted by community health workers trained in conducting structured phone calls but not trained in counselling. Further details of the trial are available in published sources.^{Reference Lund, Schneider, Davies, Nyatsanza, Honikman and Bhana1} Results from the trial indicted that the intervention had a non-significant impact on depression as measured by the HRSD in comparison with the control.^{Reference Lund, Schneider, Garman, Davies, Munodawafa and Honikman14} For the purposes of this paper, which presents an analysis of the economic burden of maternal depression, all women in the trial were included, irrespective of arm.

Participants in this study were recruited at the Site B and Michael Mapongwana Midwife Obstetric Units (MOUs) in Khayelitsha, a peri-urban settlement on the outskirts of Cape Town. Many in Khayelitsha live in shacks without potable water, indoor sewage systems or electricity; the area is characterised by high rates of unemployment, crime and violence.^{Reference Fox and Goemaere15} Additional details of the study protocol are available in published sources.^{Reference Lund, Schneider, Davies, Nyatsanza, Honikman and Bhana1} Recruitment was undertaken by trained field workers in collaboration with nurses providing antenatal services. Potential participants were pregnant women, aged 18 or older, at their first antenatal visit – typically in the second trimester of gestation.

Women who gave informed consent were recruited using a three-stage screening process: (a) verbal confirmation of the eligibility criteria for the study; (b) completion of a set of demographic questions; and (c) completion of the Edinburgh Postnatal Depression Scale (EPDS).

Women who scored 13 and above on the EPDS were recruited into the study, and were asked to complete a full baseline interview with a field worker who entered the responses into a mobile device linked to Mobenzi Researcher (a system that facilitates electronic data collection). If responses in this interview revealed signs of suicide ideation (scoring 17 or higher on the Mini International Neuropsychiatric Interview 6.0.0^{Reference Sheehan, Lecrubier, Sheehan, Amorim, Janavs and Weiller16}), the woman was referred to the nursing sisters at the clinics for further assessment and possible onward referral to Khayelitsha District Hospital. In total, 2192 women were screened, of whom 425 screened positive on the EPDS and were enrolled. Of these, 6 were excluded when it was found that they did not meet inclusion criteria, and 419 remained in the study. In addition to the baseline assessment at first antenatal visit, women were interviewed again at 1 month before their due date, and at 3 and 12 months postpartum (four interviews in total).

Across these interviews, we collected a range of demographic, socioeconomic, health service utilisation and patient cost data, as further described below.

Demographic, socioeconomic and health service utilisation data

In addition to baseline information such as median gestation, number of pregnancies, number of live births and educational status, at each interview we collected data on HRSD, employment status, partnership status, monthly household income and ownership or access to a range of assets and services.

In order to estimate the utilisation of health services, during each of the four interviews, women were asked about their utilisation of public ambulatory services in the preceding 3 months (including clinic and hospital out-patient department visits), and, if admitted to hospital, the number of in-patient days in the preceding 6 months. In addition, at their second interview (1 month before their due date), women were asked to estimate their utilisation of antenatal visits over their pregnancy; at their third interview (3 months postpartum) women reported the number of days spent at a health facility during labour and birth as well as their place of birth. In addition, women were asked about their utilisation of ambulatory and in-patient care for their child(ren) across the four interviews; at 12 months postpartum, women estimated the number of well-baby visits during the preceding year.

In order to estimate the opportunity cost of using services and out-of-pocket payments (i.e. the patient perspective), women were asked about travel, waiting and consultation times and about travelling expenses for each of their ambulatory visits in the preceding 3 months, for antenatal services over their pregnancy and for well-baby services in the first year after birth. As before, data were collected for services utilised by the women and by her child(ren). In the South African public health system, ambulatory services are free at the point of use, and fees for in-patient care differ depending on economic status. On the other hand, user fees are incurred for most private health services (for example private general practitioners). For the patient perspective, women were therefore asked about any user fees they had incurred across the range of public and private healthcare services utilised.

Valuation of economic outcomes and costs

Household income was collected as a categorical variable (given that such a strategy has been shown to improve response rates^{Reference Cleary, Birch, Chimbindi, Silal and Mcintyre17}) and transformed into a quantitative variable using a regression approach, where household income was predicted as a function of baseline employment, education and socioeconomic status. Per capita household income was computed as total household income divided by the total number of household members. Following the approach of similar studies,^{Reference Cleary, Birch, Chimbindi, Silal and Mcintyre17,Reference Foster, Vassall, Cleary, Cunnama, Churchyard and Sinanovic18} socioeconomic status was estimated at each time point using multiple correspondence analysis across a range of variables relating to access to services, main source of income, whether income is fixed or variable, employment and education status, ownership and type of dwelling, type of shop where food is bought and access to banking services.

All costs were expressed in 2014/15 prices and converted to US dollars (US$1 = 11.69 South African Rand) using an average exchange rate over the same period.¹⁹ Provider and patient average unit costs were calculated using a number of approaches. For public health services, a gross costing approach was used to compute an average cost per clinic visit, per hospital out-patient department visit, per in-patient day, per delivery day at an MOU and per delivery day at a hospital, using audited expenditure and service utilisation data from all health facilities across the Western Cape Province.²⁰ For clinics and community health centres, the overall expenditure was divided by the patient headcount over the same period; for hospitals, total expenditure was split into in-patient and out-patient unit costs using the patient day equivalent method, where the cost of an out-patient department visit is assumed to be one-third of the cost of an in-patient day.^{Reference Docrat, Besada, Cleary, Daviaud and Lund10} Once these average unit costs were computed, the costs of public health services for the women and their child(ren) were modelled by multiplying estimates of service utilisation (from the four interviews) against these unit costs. This modelling approach is described in more detail below.

As mentioned above, patient costs included travel, waiting and consultation times for the ambulatory services utilised by the women and their child(ren). Following the literature, the opportunity cost of time can be valued using wages/salary earnings foregone^{Reference Palmer and Raftery21} and is therefore based on assumed lost working hours. In this study, in order to value these costs equitably, the mean per capita income reported by the women at the baseline interview was used as a proxy of this opportunity cost. In other words, the opportunity cost of the time women spent accessing services for themselves and for their child(ren) was valued at the same rate. This opportunity cost was based on an assumed 228 working days per annum and 8 working hours per day. We did not include an opportunity cost of children's time (given that they do not work). In addition, we did not include an opportunity cost of time for delivery or in-patient care given that these times would likely include non-working hours. All other patient costs (travelling costs and user fees) were based on the values reported by respondents. In the analysis, estimates of time, travel and user fee costs were compared with the mean per capita income of the respondent's own household in order to assess the share of per capita household income spent on these costs.

Modelling approach

Using Markov modelling with a 3-month cycle length, we have estimated the economic burden of maternal depression over pregnancy and up to 12 months postpartum by comparing women with depression symptoms to those without. In other words, responses from women at each of the four interviews were used to populate either a ‘depressed Markov model state’ or ‘not depressed Markov model state’ using a cut-off of 8 on the HRSD.^{Reference Zimmerman, Martinez, Young, Chelminski and Dalrymple12}Figure 1 illustrates the Markov modelling approach. For the purposes of this model, Markov states are developed to capture variability in costs and economic burden as reported by the women across the four interviews. As is shown, modelled participants enter the model at 0–6 months antenatal, remaining in this Markov state for two cycle lengths (i.e. 6 months). Thereafter, they move to 6–9 months antenatal, and 0–3 months postnatal, spending one cycle in each of these Markov states. Finally, they move to 3–12 months postnatal, and spend three cycles in that state. This is therefore not a longitudinal analysis of economic burden based on participants’ baseline depressive symptoms; such an approach was precluded by the low numbers of women without depression at baseline (n = 16, see Fig. 1). Instead, building on a Markov modelling approach, women are allocated to a ‘depressed Markov state’ or ‘non-depressed Markov state’ depending on their HRSD score at each interview in order to generate a modelled estimate of the patient and provider costs of women with and without depression and their child(ren) across pregnancy and up to 12 months postpartum. To do so, we have assumed that the estimates of service utilisation and patient costs reported in the baseline interview are applicable to the first and second trimesters of pregnancy; services and costs reported at 1 month before the due date are assumed to occur across the third trimester; services and costs reported at 3 months postnatal are assumed to apply to that period, including the costs of delivery; while services and costs reported at 12 months postnatal apply to the 3–12 months postnatal period.

Fig. 1 Relationship between timing of interview and assumptions made to model costs during pregnancy and up to 12 months postpartum.

As mentioned above, the utilisation of antenatal and well-baby care and the associated time, travel and user fee costs were collected at 1 month before the due date and at 12 months postpartum, respectively. The antenatal costs were spread across the pregnancy whereas the well-baby visit costs were spread across the postpartum period. This approach to utilising the interview data to estimate costs by depression status over the full period (pregnancy and 12 months postpartum) is also illustrated in Fig. 1.

For both the public provider and patient perspective, estimates of service utilisation and patient time, travel and user fee costs and opportunity costs were generated within StataSE 16, and then were entered into a Markov model created in Treeage Pro 2013. Separate models were created for estimating the costs for mothers and their child(ren) within both the provider and patient perspectives and for participants with and without depression, based on their HRSD scores at each time point. In order to assess data variability, the estimated means and s.d. were entered as gamma distributions in the models. Each model was run 10 000 times in order to capture data uncertainty using probabilistic sensitivity analysis. Where P-values are presented, these have been derived using the Kruskal–Wallis comparison of means test for quantitative data and the Pearson's χ²-test for categorical data.

Ethical approval

The study obtained ethical approval from the University of Cape Town Human Research Ethics Committee (Ref no 226/2011 and 469/2015) and the National Institute of Mental Health Data Safety and Monitoring Board. All participants provided written informed consent.