Methods
This study was embedded as part of the larger M-DEPTH cRCT and followed consolidated health economic evaluation reporting standards. A full overview of trial characteristics can be found in the study protocol20 (see online supplemental file), and preliminary results in short-term intervention efficacy are reported elsewhere.21 22 Briefly, participating hospitals and health centres, located in close proximity to one another outside Kampala, were randomised to either CAU or M-DEPTH. A cluster RCT was used, as opposed to individual randomisation, because risk of contamination was too great if both control and intervention clients were treated by the same providers within a given site. Although we were unable to fully prevent contamination, it was mitigated by the extensive distance between individual sites.
CAU for addressing perinatal depression at ANC clinics in Uganda encompasses referrals among patients who exhibit severe depressive symptoms—either to a district, regional or national referral hospital. WLH attending ANC clinics throughout Uganda are also invited to attend monthly Family Support Groups (FSGs) that provide them with education and guidance to support perinatal and postpartum care, including adherence to medications provided in the context of PMTCT.23 However, FSGs do not provide psychotherapy or pharmacotherapy that might be considered a direct mental health service.
The M-DEPTH intervention represented a stepped care approach to providing behavioural and pharmacological treatments24: women with minimal, subthreshold depressive symptoms (Patient Health Questionnaire (PHQ)-9 <10) were provided with depression psychoeducation and monthly monitoring of depression. In the event depressive symptoms exceeded PHQ-9 >9 at a later assessment period, individuals were offered treatment at this time. Women with mild or moderate depressive symptoms (PHQ-9 between 10 and 19) were recommended individual PST counselling alone, up to seven bi-weekly sessions; and those with severe depressive symptoms (PHQ-9 >19) were recommended ADT alone—for which the first-line medication was fluoxetine.
While the study protocol outlined these cut-off scores as treatment recommendations, women who were eligible for treatment were allowed to select the treatment modality of their choice. Furthermore, in the event that the initial modality selected was not efficacious, women were offered to change to or add the other modality. CAU, the default in the study setting, comprised screening for depression and referral for specialty care, in the event the client was severely depressed.
Women were eligible to participate in the clinical trial if they were receiving ANC services at a participating facility, scored 5 or greater on the PHQ-9 (reflecting at least minimal depressive symptoms), were medically stable (including being on ARVs for at least 4 weeks) and were not identified as high risk for suicide. Pregnancies at enrolment ranged from 4 to 32 weeks of gestation. Women identified at high risk for suicide, including over the course of the trial, were immediately referred for same-day services, following a standardised protocol on which all staff were trained. All women invited to participate provided written informed consent.
The public and clients were included in the planning and implementation of the research study. A community advisory board was assembled for the purpose of providing the community stakeholders with a voice in the planning and implementation of this study. The board consisted of female clients, service providers and local and district health officials. The study team met with the board during key junctures throughout the study to get its input on study design, intervention content and structure and measurement and interpretation of findings. Furthermore, peer mothers served as interventionists.
Measures
The primary outcome measure was remission of perinatal depression, as assessed by the 9-item version of the PHQ-925 and the Mini Diagnostic Neuropsychiatric Interview (MINI).26 The PHQ-9 contains questions pertaining to individuals’ frequency of sadness and other cognitive and somatic symptoms over the previous 2-week period. A threshold of 10 or greater (PHQ-9 ≥10) is commonly used as an estimation of current depressive disorder, as this threshold has been found to be highly predictive of the presence of major depressive disorder. This instrument has also been previously validated in Uganda and was forward-translated and back-translated into Luganda.27 The MINI is a brief diagnostic interview corresponding to DSM criteria for major depressive disorder—quantified as five or more of nine symptoms present and indication of functional impairment. Assessments of depression symptomology were assessed at enrolment, 1 month postpartum as well as 6, 12 and 18 months follow post-partum assessment.
In addition to depression symptoms, we also documented basic demographic and background characteristics of participants. These included patient age in years, receipt of any secondary education (yes/no), relationship status, length of gestation in weeks, and whether the patient was diagnosed with HIV within the last 3 months (yes/no).
Secondarily, we quantified the cost of M-DEPTH and CAU used TDABC. TDABC is a gold-standard framework for cost accounting, in which patients are directly observed as they progress through a health system to quantify all resources they consume.28 Time is used as the primary measurement unit for assigning costs to resources—such as personnel, equipment and space. Finally, as a function of intervention costs and benefits, we measured an ICER. ICERs compare the marginal utility of intervention (M-DEPTH) versus control (CAU) conditions, relative to the marginal cost of the intervention versus control conditions. Further information on this is described below.
Statistical analysis
Health outcomes analysis
Intervention effects were assessed using repeated measures mixed effects logistic regression analysis, from an intention-to-treat perspective. The model contained a random effect to account for repeated observations within each client, and remaining study characteristics were modelled as fixed effects: namely, treatment exposure, time, the interaction between treatment exposure and time as well as a fixed effect for facility assignment. This interaction effect is the primary effect of interest, representing comparative rates of remission (PHQ-9<10) between study arms.
In a secondary analysis, we included four demographic characteristics as covariates: age in years, any secondary education (yes/no), newly diagnosed with HIV (yes/no) and weeks of gestation at enrolment. We found no meaningful difference in the intervention effect estimates between the two analyses: that is, in both models, the interaction between treatment exposure and time was highly significant and ORs were of similar magnitude (less than a 1% difference). Therefore, we report and use the effect estimates from the more parsimonious model.
Cost analysis
We quantified the full economic cost of the intervention, including volunteered and donated resources, and we assumed a societal perspective by incorporating opportunity costs incurred by participants. Unit cost estimates were obtained from various sources, including payroll, invoices and receipts. Staff time was valued at full salary cost, inclusive of benefits and allowances for vacation. We estimated the cost of clinical space based on local rents. All costs were expressed in 2022 US$. Costs measured in 2020 and 2021 were adjusted for inflation.
Cost-effectiveness analysis
Cost-effectiveness of the M-DEPTH intervention was estimated using a Markov chain Monte Carlo approach in TreeAge Pro 2023 V.R2.0.29 The primary outcome, absence from perinatal depressive disorder (PHQ-9<10), was converted to changes in disability weights according to the 2019 Global Burden of Disease study,30 following a standard methodology as outlined by Buttorf and colleagues (2012).31 In short, non-depressed individuals were assigned a health state value equivalent to the level of disability observed among those with a PHQ-9 score <10, while those with depressive disorder (PHQ-9≥10) were assigned a decrement equivalent to the disability weight associated with depressive symptom severity.
Based on WHO-CHOICE cost-effectiveness thresholds,32 we identified $964—the median GDP per capita in Uganda19—as a willingness to pay (WTP) threshold for averting one DALY. We also incorporated two sensitivity analyses to provide upper and lower bound estimates for ICERs. First, we varied the discount rate for future health utility from 0% to 5%,33 with 0% set as the base case.34 Discounting is commonly used to make a balanced comparison between programmes whose costs and outcomes are manifest at different intervals. Second, we modelled cost-effectiveness under three scenarios: (1) the base case, for which both costs and intervention effects were terminated at the end of the observation period and (2) two other scenarios in which intervention effects remain stable over a 3-year and 5-year time period from the point of study enrolment.
Finally, to account for uncertainty in the level of disability associated with depressive disorder, we incorporated a 95% CI of disability estimates as reported by WHO Global Burden of Disease point estimates. Specifically, we conducted a probabilistic sensitivity analysis with 100 000 Monte Carlo simulations and quantified the per cent of simulations that M-DEPTH would be considered cost-effective at WTP thresholds ranging from $0 to $1000 US$ per DALY averted, with the default set to $964 as described above.
