Effectiveness of the Systems Analysis and Improvement Approach to Optimize Outpatient Mental, Neurological, and Substance-use Treatment Cascades in Mozambique: a Cluster Randomized Trial

Abstract

Background.

To our knowledge, there are no evidence-based implementation strategies for care cascade optimization of mental, neurological, and substance-use disorder (MNS) treatment in low- and middle-income countries. This trial evaluated the effectiveness of the Systems Analysis and Improvement Approach for Mental Health (SAIA-MH) implementation strategy to improve MNS care cascade outcomes in Mozambique. The SAIA-MH strategy combines external facilitation, clinical consultation, and provider team meetings with system-engineering tools in a continuous quality improvement framework.

Methods.

We conducted a 3-year cluster randomized trial comparing an 8-month baseline with a 2-year implementation period. All patients diagnosed with MNS conditions across 16 government facilities in Mozambique (8 intervention; 8 attentional placebo control) were eligible. The primary outcome was low functional impairment or functional improvement measured using the WHODAS 2.0. Secondary outcomes were medication adherence and appointment attendance. We involved people with related lived experience in all elements of the research and writing process.

Outcomes.

Between February 4, 2022 and October 14, 2024, 3,837 MNS patients attended 33,055 outpatient visits. The mean age was 26.0 (range 0–103), 53.1% (n=2,037) were male and 46.9% were female (n=1,800), and 67.3% were diagnosed with epilepsy (n=2,581). Ethnicity data were unavailable. The SAIA-MH arm showed 46.0 percentage points higher functional improvement or low functional impairment (95% CI: 34.0, 58.0, p < 0.001), 18.1 percentage points higher medication adherence (95% CI: 15.4, 20.7, p < 0.001), and 18.4 percentage points higher appointment attendance (95% CI: 15.1, 21.7, p < 0.001) compared with control. Among non-adherent patient visits, the intervention arm had 11.9 fewer non-adherent days than controls (95% CI: −17.6, −6.2; p < 0.001). WHODAS 2.0 scores decreased by 5.9 points more in the intervention group (95% CI: −6.5, −5.2; p < 0.001).

Interpretation.

The SAIA-MH implementation strategy shows evidence of effectiveness in increasing appointment attendance, medication adherence, and patient functioning for MNS patients treated in outpatient primary care. Our findings warrant further research across implementation contexts to determine how differences in diagnostic and comorbidity case mix may influence the ability of SAIA-MH to reduce gaps in task-shared MNS care.

Trial Registration.

Clinicaltrials.gov identifier: NCT05103033

Introduction

Mental neurological and substance use disorders (MNS) are leading causes of disability worldwide resulting in 27% of years lived with disability (YLD) and approximately 5 trillion USD in economic losses ⁹. In Mozambique, MNS conditions are estimated to be the number one cause of years lived with disability, collectively accounting for 25% of national YLD ⁹. However, average government health spending on mental health in LMICs is only 0.5% of health budgets despite accounting for almost a quarter of YLDs ¹⁰. As a result, many LMICs struggle with under resourced mental health systems and limited human resources to meet the substantial burden of MNS. Studies have found the mental health (MH) treatment gap in LMICs to often exceed 90% ^11,12.

To address the MH treatment gap, academic and policy leaders have advocated for investments in rapid training programs for lower-level MH providers ^13–15. Mozambique has successfully scaled-up task-shared MH care through a 2-year rapid training program for psychiatric technicians who can diagnose and prescribe medication for most common and severe mental health problems in primary care. As of 2024 there were 328 psychiatric technicians covering all provinces of Mozambique ¹⁶. However, with most attention in LMICs focused on scaling-up rapid task-sharing training programs to improve access to MH care, few studies have focused on implementation strategies to evaluate and improve the quality of care delivered by providers with limited training, resources, and supervision. Task-shared outpatient management of common mental disorders in Southern Africa has demonstrated significant quality concerns: diagnoses were undocumented in >40% of cases; psychiatric histories were inadequately recorded in 89%; inappropriate polypharmacy was prescribed in 88% of cases; >40% of patients were lost to follow-up; and >60% of patients had suboptimal medication adherence ^2,3. Similar patterns of high rates of LTFU, poor medication adherence, and patient outcomes have been reported in other LMICs (5,6). Previous work evaluating the quality of care of outpatient management of MNS by psychiatric technicians in Mozambique has shown LTFU rates >50%, sub-optimal medication adherence >80%, and <30% of patients achieving functional improvement ¹⁷.

There is an urgent need for low-cost implementation strategies focused on improving the performance of the MH care cascade in LMICs. Until quality concerns are addressed, the population health effects of scaling-up task-shared MH care in LMICs will be limited or potentially harmful. The MH care cascade details linked steps along the MH treatment pathway that a person would likely pass through from detection to remission of illness. Quality problems in one step of a treatment cascade can have non-linear and compounding impacts across the larger complex care system. Implementation strategies focused on only one step in a cascade can potentially contribute to unintended system bottlenecks and quality of care issues.

The Systems Analysis and Improvement Approach (SAIA) is a multicomponent implementation strategy focused on optimizing treatment cascades in their entirety. The SAIA strategy has been tested and shown to be effective for optimizing cascades of care for HIV/AIDS and hypertension management within the HIV context ¹⁸. Preliminary data has shown that applying SAIA to MNS treatment cascade optimization (SAIA-MH) is feasible, acceptable, and can result in clinically significant treatment cascade improvements in Mozambique. Six months of pilot SAIA-MH implementation resulted in 1.5-times higher odds of patient medication adherence (aOR 1.5; CI 1.2, 1.9) and 3.7-times higher odds of patient functional improvement (aOR 3.7; CI 2.5, 5.4) ¹⁷.

We describe findings from a large cluster RCT of the SAIA-MH implementation strategy including a census of all patients diagnosed and treated for MNS conditions for over two years across 16 government health facilities in Mozambique. To our knowledge, this is the first large scale RCT of an implementation strategy focused on improving the quality and performance of the task-shared primary care MNS care cascade in an LMIC context.

Methods

Study design and participants

We utilized a parallel cluster randomized difference-in-difference framework to examine clinical changes comparing an eight-month baseline period to a two-year implementation phase for a census of all MNS patients across 8 intervention and 8 attentional placebo control primary care facilities. Inclusion criteria were all patients diagnosed with a MNS condition in outpatient primary care, prescribed any medication, and given a follow-up date for clinical evaluation and management in target facilities. Exclusion criteria included: patients who after evaluation were diagnosed with a non-MNS condition and referred and patients transferred for treatment outside of target trial clinics. Participants included a census of 3,837 MNS patients who were prescribed medication and who attended 33,055 outpatient visits between February 4^th, 2022, to October 13^th, 2022 (baseline) and October 14^th, 2022, to October 14^th, 2024 (implementation). Local and national institutional review boards approved the study, including the University of Washington IRB and the National Health Bioethics Committee of Mozambique (CNBS). As the intervention was delivered at the provider level and used routinely collected clinical data, individual patients were neither recruited nor consented, and patient-level adverse events were not assessed. Adverse events were therefore monitored and reported at the level of participating health providers. A waiver of informed consent was obtained for access to de-identified patient medical records. This study followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guidelines ¹⁹. The study protocol is available at ClinicalTrials.gov (identifier: NCT05103033) and has been previously published ²⁰.

Per reviewer request, we have included a brief discussion of adoption, fidelity, and implementation of SAIA-MH micro interventions in the present manuscript. Additional analyses of sustainability outcomes, cost-effectiveness, and post-hoc evaluation of SAIA-MH barriers and facilitators will be published separately.

Setting

This study was conducted in 16 primary care facilities currently providing MNS service across Sofala and Manica Provinces in Mozambique (see supplementary appendix pg. 1 for map of study sites). These provinces were selected due to the absence of existing structural interventions for MNS systems improvement. Since 1996, the Mozambican Ministry of Health has been a leader in sub-Saharan Africa in scaling-up task-shared outpatient MNS care led by mid-level specialist providers called psychiatric technicians ²¹. These providers can diagnose and treat all major categories of MNS conditions using both pharmacological and psychosocial evidence-based interventions. Despite investment in training task-shared providers in Mozambique, the quality of mental health services remains unknown or poor, largely due to constraints in financial and human resources, limited access to ongoing supervision and training, periodic shortages of essential psychiatric medications, and pervasive stigma surrounding mental illness. In Mozambique, the Ministry of Health delivers more than 90% of formal healthcare services, improving the potential for population-level effects for clinical supply-side interventions.

Randomization and Blinding

Eligible health facilities were: (1) naïve to the SAIA-MH implementation strategy; (2) currently providing MNS services including prescribing medication; (3) within a three hour one-way drive from Chimoio City, Manica or Beira City, Sofala; (4) those with at least one psychiatric technician and one psychologist currently practicing; and (5) those with ≥100 annual outpatient MNS visits during 2020–2021. Quaternary or provincial hospitals were excluded due to the complexity of MNS workflows across specialty services, As of August 31^st 2022, there were 8 eligible facilities in Sofala and 8 in Manica. Facilities were allocated 1:1 to intervention or control using constrained randomization to maximally balance province, level of health facility, MNS human resources, number of annual outpatient consultations, and baseline cascade performance (see supplementary appendix pg. 3). Randomization was conducted by BHW using the cvcrand command in Stata 16 on August 31^st 2022. There was no blinding of facility assignment at the level of providers, health facilities managers, or investigators since it is not feasible given SAIA-MH implementation procedures. The only study blinding occurred at the level of the statistical analyst during initial primary outcome assessment.

Procedures

The 8-month baseline period aimed to represent routine MNS care in the 16 eligible health facilities. However, due to a lack of ability to track patients over time and across follow-up visits, all health facilities received enhanced MNS patient registries and patient tracking tools as used in the pilot study during this baseline period. These enhanced registries included the new workflow of applying the WHODAS 2.0 for all MNS patients ≥15 years of age. Facility teams attended a two-day in-person training in the first week of February 2022 and received monthly in-facility visits by study staff to educate them on the new tools. During these monthly visits, data from enhanced registries were entered into a CommCare database by study staff.

After the 8-month baseline period, facility-level teams receiving SAIA-MH attended a five day in person training October 10^th – 14^th, 2022. These teams were determined by facility staff but were required to include a minimum of one psychiatric technician, one psychologist, and the clinic manager or director who oversees MNS delivery. We conducted a practical training in which facility teams were introduced to the SAIA-MH model and strategy processes and subsequently utilized data on their own health facility performance during the baseline period to select, prioritize, plan, and then implement their first system change. The goal was for each facility to complete their first action plan during the training so they could implement their micro intervention at their health facility the following week.

This study utilized an attentional placebo controlled design whereby control health facilities also attended a five day in person training October 17^th – 21^st, 2022 including the same minimum staff as defined in the SAIA-MH arm. The goal of the attentional placebo controlled design is that the control facilities will mimic the activities in the intervention group but without the “active ingredient” of the SAIA-MH implementation strategy. This placebo training included review of the baseline implementation period enhanced patient registries and patient tracking tools, as well as discussion of research ethics, MNS stigma and discrimination, occupational stress in MNS care, and introduction to research methods.

Following the completion of training, both arms of the study received continuous monthly in-clinic mentorship and facilitation visits. For the SAIA-MH arm these visits focused on supporting fidelity to the SAIA-MH implementation strategy protocols as well as continued use of the enhanced registries, patient tracking tools, and the WHODAS 2.0. For the control arm these visits focused exclusively on the registries and the WHODAS 2.0. During these visits, study staff collected data into the CommCare database and resolved data quality issues in both arms. We involved individuals with related lived experience of MNS conditions in all elements of the research, implementation, and writing process.

Intervention

The SAIA-MH implementation strategy is a multicomponent package of implementation strategies each targeting a diverse range of individual, organizational, and contextual factors that determine implementation success of MNS care delivery. SAIA-MH was developed to assist task-shared workers to identify and prioritize problems; assess potential improvement options using optimization; and develop, execute, evaluate, and integrate micro interventions for continuous systems improvement considering effects on the full MNS treatment cascade.

Details on the intervention have been published previously ^{20 8,22}. Briefly, SAIA-MH is organized around monthly improvement cycles in which facility improvement teams meet in person for 2–3 hours supported by an external SAIA-MH facilitator. These meetings start by conducting cascade analysis. Utilizing the mental health cascade analysis tool (MHCAT), facility teams visualize the flow of patients in their own clinic through the steps of the mental health care cascade. The MHCAT helps teams identify where they are losing the most patients and includes an optimization function to help identify which step in the care cascade should be prioritized for systems improvement for maximum performance gain. Second, teams conduct process mapping whereby they jointly draw a physical map of all steps/processes a patient passes through in the process of receiving MNS care in a specific health facility. The purpose of this is to gain consensus on the current system architecture, highlight system inefficiencies, bottlenecks, redundancies, barriers, or otherwise areas for improvement that do not add value to the patient. Third, utilizing information from the MHCAT and process map, teams propose and prioritize potential micro interventions (low-cost, small changes to workflow or service organization that are within the purview of the psychiatric technician and psychologists and done in a timescale of days or weeks). These micro interventions are targeted at addressing an identified barrier at a specific step in the MNS care cascade. Teams create an implementation blueprint to operationalize micro intervention design, roles, and anticipated changes for the coming month. Fourth, teams implement their micro intervention throughout the next month while tracking their progress via selected process indicators. Fifth, teams assess the impact of their micro intervention by reviewing process indicators, changes in MHCAT patient performance and changes in the system structure. Teams adopt, adapt, or abandon the tested micro interventions based on the results, thus modifying and improving routine care over the long term. Meanwhile, facility teams iteratively repeat the above steps during each monthly cycle to identify and prioritize new areas of the MNS care cascade and select new micro interventions to address them.

Outcomes

Our primary study outcome was patient-visit level functional improvement or low functional impairment defined by visits in which patients ≥15 years of age achieved a WHODAS 2.0 score of <10 or a ≥50% reduction in baseline WHODAS 2.0 score. Secondary outcomes were visit-level medication adherence and patient appointment attendance. Medication adherence was defined as patient visits where patients returned for medication refills without having missed a dose of medication based on pill counts combined with patient reports of adherence. If pill counts and patient self-report were discordant, the visit was classified as nonadherent. On time appointment attendance was defined as patient visits occurring ≤5 days before or after their scheduled appointment.

Choice of Primary Measure

The WHODAS 2.0 was developed by the World Health Organization (WHO) specifically to allow cross-cultural disability measurement across different cultures, settings, and patient populations²³. The measurement tool is open source and freely available. It has good reliability, item-response characteristics, and a robust factor structure that has been replicated across Sub-Saharan Africa (SSA) ^24,25. In the original WHO validation study across 19 countries (9 LMICs) the test-retest reliability had an intra-class coefficient of 0.98, the total Cronbach’s alpha (n=1,565 across 19 countries) was 0.98 and also 0.98 for MH patients^23,25. The WHODAS 2.0 has been used extensively in LMICs^24–26, has been applied successfully in Mozambique in previous studies²⁶ (and our pilot study), and has been extensively applied in LMICs (including SSA) to evaluate function improvement in high-impact pragmatic MH trials^27–31. There is no single universally accepted threshold for clinical improvement on the WHODAS 2.0. For this trial, we utilized a combination of the common metric of ≥50% reduction in baseline score to represent treatment response, as well as patients achieving an absolute score of <10 to represent relatively low functional impairment³².

Statistical analysis

Statistical analysis was performed by BHW from November 28, 2024 to August 8^th, 2025 blinded to study arm. A priori we estimated power (1-β) ≥ 90% to detect an increase as small as 2.2% in functional improvement or low functional impairment, 3.6% in adherence, and 4.9% for appointment attendance assuming target clinics saw a mean of 300 follow-up patients per 6 months. Using an alpha value of 0.05 and Stata 16 for all analyses, we fit mixed-effects Poisson regression models with a log link to estimate risk ratios for each binary primary and secondary outcome in a difference-in-difference (DID) framework to predict differential changes for intervention versus control facilities comparing pre- to post-intervention implementation. All models utilized robust standard errors using the Huber-White sandwich estimator to provide valid inference under possible variance misspecification. These models included a random intercept at the patient level, health facility fixed effects, and fixed effects indexing the intervention group, pre- versus post-implementation time period, and an interaction term between intervention and implementation time period representing our primary DID intervention effect. Clustering at the health facility level was modeled through a fixed effect due to some facilities having zero events in our pre-intervention period for functional improvement leading to model instability when estimating health facility random effects.

Predicted absolute probabilities were estimated from nonlinear combinations of the model-fitted coefficients for each group and time point. Pre-specified sub-group analyses examined whether intervention effects differed by patient sex by including a three-way interaction term between sex, intervention group, and time in our mixed-effects Poisson models. Post-hoc sub-group analyses were conducted as requested by peer reviewers and examined whether intervention effects differed by: (1) MNS diagnosis as defined by International Classification of Diseases – 10^th edition code (ICD-10); and (2) HIV or tuberculosis (TB) status defined as positive, unknown, or negative status at intake. Additional analyses utilized gaussian mixed-effects DID models with random intercepts at the patient and health facility level to examine: (1) absolute differences in patient WHODAS 2.0 scores attributable to the intervention; (2) absolute differences in number of days non-adherent to medication among nonadherent return patient visits; and (3) absolute differences in number of days late for scheduled appointments among patients who returned >5 days after scheduled follow-up appointment date. Intracluster correlation coefficient (ICC) estimates were generated using an intercept-only three-level mixed effects logistic model including random intercepts at the health facility and patient level. There was no data monitoring committee for the present study.

Results

Participant characteristics

In total, 2,394 eligible patients were treated during the study period in the 8 intervention MNS facilities and 2,099 in the 8 control facilities. Of these, 2,153 (89.9%) were included in analyses of secondary outcomes for the intervention arm and 1,684 (80.2%) for the control arm (see Figure 1 for full PRISMA patient flow chart). The primary reason for ineligibility for analyses was not having return visits. Patients attended 21,904 MNS consultations in the intervention arm and 11,151 in the control arm. A subset of 966 patients in the intervention arm (with 7,697 visits) and 785 in the control arm (with 3,804 visits) who had their first visit during the study period, who completed the WHODAS 2.0 measurement at that visit, and who were ≥15 years old were included in the primary analysis of patient-visit level functional improvement or low functional impairment to allow for examination of change from baseline. Patterns of missingness for this primary outcome did not differ by study arm (see supplementary appendix pg. 5).

Figure 1.

PRISMA flow chart for cluster randomized controlled trial of the Systems Analysis and Improvement Approach for Mental Health (SAIA-MH) in Mozambique.

Overall, 2,037 included patients were male (53.1%) and 46.9% were female (n=1,800). The mean age was 26.0 (SD: 15.4, range 0–103), with 55.9% (n=2,144) being ≤25 years old (Table 1). Ethnicity data were unavailable as it is not routinely collected in the Mozambican health system. However, we anticipate health system visits to reflect the national population demographics of 99% African ethnicity³³. Most patients were single (n=2,662; 69.4%), 10.8% (n=413) reported living with HIV, 471 (12.3%) reported current or past drug use, and 157 (4.1%) reported thoughts of suicide at their intake visit. Most patients (n=2,581; 67.3%) had a primary diagnosis of Epilepsy (ICD10 Codes: G40-G41), followed by Schizophrenia, Schizotypal, and Delusional Disorders (n=519; 13.5%); (ICD10 Codes: F20-F29), Affective Disorders (n=173; 4.5%); (ICD10 Codes: F39-F39), and Mental and Behavioral Disorders due to Psychoactive Substance Use (154; 4.0%) (ICD10 Codes: F10-F19). Only 3.1% (n=120) of participants had a secondary diagnosis, with the most common being Intellectual Disabilities (21.7%, n=26, F70–79), Schizophrenia, Schizotypal, and Delusional Disorders (20.8%, n=25, F20-F29), and Epilepsy (18.3%, n=22, G40).

Table 1.

Demographic characteristics of patients seeking care before and after implementation of the SAIA-MH trial across 16 public health facilities in Sofala and Manica Provinces, Mozambique.

	Combined	Intervention	Control
	N (%)	N (%)	N (%)
Total Patients	3,837 (100)	2,153 (56.1%)	1,684 (43.8%)
Sex
Male	2,037 (53.1%)	1,124 (52.2%)	913 (54.2%)
Female	1,800 (46.9%)	1,029 (47.8%)	771 (45.8%)
Age at first visit (Mean: 26.0; SD: 15.4)
0–14	849 (22.1%)	490 (22.8%)	359 (21.3%)
15–18	404 (10.5%)	200 (9.3%)	204 (12.1%)
19–25	891 (23.2%)	493 (22.9%)	398 (23.6%)
26–35	851 (22.2%)	479 (22.3%)	372 (22.1%)
36–50	556 (14.5%)	322 (15.0%)	234 (13.9%)
51–64	173 (4.5%)	105 (4.9%)	68 (4.0%)
65+	109 (2.8%)	60 (2.8%)	49 (2.9%)
Missing	4 (0.1%)	4 (0.19%)	0 (0.0%)
Relationship status at first visit
Married	293 (7.6%)	162 (7.5%)	131 (7.8%)
Single	2,662 (69.4%)	1,459 (67.8%)	1,203 (71.4%)
Domestic union	590 (15.4%)	359 (16.7%)	231 (13.7%)
Divorced/widowed/separated	278 (7.3%)	163 (7.6%)	115 (6.8%)
Missing	14 (0.4%)	10 (0.46%)	4 (0.24%)
HIV diagnosis at first visit
Positive	413 (10.8%)	200 (9.3%)	213 (12.7%)
Negative	2,313 (60.3%)	1,310 (60.9%)	1,003 (59.6%)
Unknown	778 (20.3%)	322 (15.0%)	456 (27.1%)
Missing	333 (8.9%)	321 (14.9%)	12 (0.7%)
TB diagnosis at first visit
Positive	152 (4.0%)	78 (3.6%)	74 (4.4%)
Negative	2,062 (53.7%)	1,327 (61.6%)	735 (43.7%)
Unknown	1,550 (40.4%)	689 (32.0%)	861 (51.1%)
Missing	73 (1.9%)	59 (2.7%)	14 (0.83%)
Alcohol use at first visit
Current	416 (10.8%)	246 (11.4%)	170 (10.1%)
Past	543 (14.2%)	315 (14.6%)	228 (13.5%)
Never	2,794 (72.8%)	1,530 (71.1%)	1,264 (75.1%)
Missing	84 (2.2%)	62 (2.9%)	22 (1.3%)
Drug use at first visit
Current	191 (5.0%)	112 (5.2%)	79 (4.7%)
Past	280 (7.3%)	169 (7.9%)	111 (6.6%)
Never	3,278 (85.4%)	1,807 (83.9%)	1,471 (87.4%)
Missing	89 (2.3%)	65 (3.0%)	23 (1.4%)
Positive	157 (4.1%)	128 (5.6%)	29 (1.7%)
Negative	3,654 (95.2%)	2,012 (93.5%)	1,642 (97.5%)
Missing	26 (0.68%)	13 (0.60%)	13 (0.77%)
Primary diagnosis at first visit
F00-F09 Organic, including symptomatic, mental disorders	77 (2.0%)	45 (2.1%)	32 (1.9%)
F10-F19 Mental and behavioral disorders due to psychoactive substance use	154 (4.0%)	81 (3.8%)	73 (4.3%)
F20-F29 Schizophrenia, schizotypal and delusional disorders	519 (13.5%)	274 (12.7%)	245 (14.6%)
F30-F39 Affective Disorders	173 (4.5%)	82 (3.8%)	91 (5.4%)
F40-F48 Neurotic, stress-related and somatoform disorders	101 (2.6%)	67 (3.1%)	34 (2.0%)
F50-F59 Behavioral syndromes associated with physiological disturbances and physical factors	114 (3.0%)	58 (2.7%)	56 (3.3%)
F60-F69 Disorders of adult personality and behavior	14 (0.36%)	12 (0.56%)	2 (0.12)
F70-F79 Intellectual Disabilities	29 (0.76%)	11 (0.51%)	18 (1.1%)
F80-F89 Disorders of psychological development	7 (0.18%)	6 (0.28%)	1 (0.06%)
F90-F98 Behavioral and emotional disorders with onset usually occurring in childhood/adolescence	45 (1.2%)	27 (1.3%)	18 (1.1%)
G40 Epilepsy	2,581 (67.3%)	1,476 (68.6%)	1,105 (65.6%)
Missing	23 (0.60%)	14 (0.65%)	9 (0.53%)

Clinical outcomes

ICC estimates for primary and secondary outcomes are provided in the supplementary appendix (pg. 4). Patients in intervention facilities showed improved functional impairment, medication adherence, and on time appointment attendance relative to patients in control facilities. The SAIA-MH arm showed 46.0 percentage points higher functional improvement or low functional impairment (95% CI: 34.0, 58.0, p < 0.001), 18.1 percentage points higher medication adherence (95% CI:15.4, 20.7, p < 0.001), and 18.4 percentage points higher appointment attendance (95% CI: 15.1, 21.7, p < 0.001) compared with attentional placebo control. Pre-intervention, at baseline, there were no significant differences in predicted probability of functional improvement or low functional impairment comparing intervention to control (RR: 0.77, p=0.14). Post-intervention, 86.1% (81.4, 90.9) of patient visits for those treated in SAIA-MH facilities showed functional improvement or low functional impairment compared to 49.7% (44.1, 55.2) in control facilities (supplementary appendix pg. 2). In terms of relative changes, patient MNS visits in the SAIA-MH arm had 2.2-fold greater likelihood of functional improvement or low functional impairment (95% CI: 1.6–3.1, p < 0.001), 1.5-fold greater medication adherence (95% CI: 1.4–1.7, p < 0.001), and 1.4-fold greater on time appointment attendance (95% CI: 1.4–1.5, p < 0.001) compared to the attentional placebo control (Table 2).

Table 2.

Primary and secondary outcomes from mixed-effects Poisson regression model for the difference-in-difference effect of the SAIA-MH implementation strategy in Sofala and Manica, Mozambique.

Outcome	Model predicted probability* at pre-intervention (95% CI)	Model predicted probability* post-intervention (95% CI)	Risk Ratio† for DID effect (95% CI)	Absolute Percentage Point Difference* DID effect (95% CI)	N patients included
Secondary: Appointment Attendance
Control	54.2% (50.1, 57.6)	40.2% (38.1, 42.5)	1.43 (1.36, 1.52)‡	18.4% (15.1, 21.7) ‡	3,837
Intervention	68.3% (65.0, 71.4)	72.7% (70.0, 75.5)
Secondary: Medication Adherence
Control	24.3% (21.8, 26.8)	21.1% (19.2, 23.0)	1.53 (1.42, 1.66) ‡	18.1% (15.4, 20.7) ‡	3,837
Intervention	45.0% (42.3, 47.8)	60.0% (56.8, 63.0)
Primary: Functional Improvement
Control	42.1% (32.1, 52.2)	49.7% (44.1, 55.2)	2.24 (1.63, 3.08) ‡	46.0% (34.0, 58.0) ‡	1,751
Intervention	32.6% (24.7, 40.4)	86.1% (81.4, 90.9)

^*Predicted probabilities were estimated from linear combinations of the model-fitted coefficients for each group and time point

^†Risk ratio estimated from a mixed-effects Poisson regression model with robust standard errors

^‡p<0.001

In terms of secondary absolute clinical outcomes, WHODAS 2.0 scores were an average of 5.9 points lower for patient visits in the SAIA-MH arm compared to the control arm (95% CI: −6.5 to −5.2; p < 0.001); (Table 3). Among non-adherent patient visits, the SAIA-MH arm had 11.9 fewer non-adherent days between visits compared to controls (95% CI: −17.6 to −6.2; p < 0.001). Among patient visits not “on time”, patients in the SAIA-MH arm returned to their follow-up visits 12.7 days earlier compared to controls (95% CI: −4.3 to −21.1, p = 0.003).

Table 3.

Secondary outcome analysis for absolute mean difference-in-difference changes in function improvement for the SAIA-MH implementation strategy in Sofala and Manica, Mozambique.

Outcome	Pre-Intervention Predicted Mean (95% CI)	Post-Intervention Model Predicted Mean (95% CI)	Absolute Beta for DID effect (95% CI)
Among Visits Not “On Time”, Mean Days Late ‡
Control	29.5 (10.3, 48.7)	72.0 (53.5, 90.5)	−12.7 (−4.3, −21.1)*
Intervention	34.1 (15.1, 53.1)	63.9 (45.5, 82.3)
Among Non-Adherent Visits, Mean Days without Medication §
Control	18.9 (3.7, 34.0)	55.7 (40.9, 70.5)	−11.9 (−6.2, −17.6)†
Intervention	18.5 (3.4, 33.5)	43.5 (28.8, 58.2)
Mean WHODAS 2.0 Score||
Control	10.3 (8.1, 12.6)	9.2 (7.0, 11.4)	−5.9 (−5.2, −6.5) †
Intervention	13.1 (10.9, 15.3)	6.1 (3.9, 8.3)

^*p<0.05

^†p<0.001

^‡This model is restricted to patient visits that were not “on time” (defined as ≤5 days before or after scheduled appointment).

^§This model is restricted to patients that were not adherent (defined as returning in more days than pills dispensed or attesting they were nonadherent during the follow-up time).

Intervention effects showed no differential effects by patient sex (supplementary appendix pg. 8). Per reviewer request, we conducted further post-hoc unpowered sub-group analyses (see supplementary appendix pg. 7) that showed mixed findings and should be interpreted with caution.

One adverse event was reported during study implementation. This event involved the death of a health facility provider (psychologist) in the attentional placebo control arm and was determined to be unrelated to study participation (Table 4).

Table 4.

Adverse events, withdrawal, or lost to follow-up at the level of the SAIA-MH provider disaggregated by sex.

Adverse event, withdrawal, or lost to follow-up	Intervention			Control
	Total N=33	Male N=17	Female N=16	Total N=37	Male N=17	Female N=20
Any adverse event	0 (0.0%)	0 (0.0%)	0 (0.0%)	1 (2.7%)	0 (0.0%)	1 (2.7%)
Serious adverse event	0 (0.0%)	0 (0.0%)	0 (0.0%)	1 (2.7%)	0 (0.0%)	1 (2.7%)
Death determined to be related to study participation	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)
Death determined to be unrelated to study participation	0 (0.0%)	0 (0.0%)	0 (0.0%)	1 (2.7%)	0 (0.0%)	1 (2.7%)
Injury	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)
Hospitalization	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)
Withdrawal	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)	0 (0.0%)
Lost to follow-up	4 (12.1%)	2 (11.8%)	2 (11.8%)	14 (37.8%)	6 (35.3%)	8 (40.0%)

^*Deaths are included in any adverse events and serious adverse events

Implementation Outcomes

Thirty-three staff were enrolled in the SAIA-MH strategy arm across the eight intervention health facilities. All eight facilities (100%) completed 24 cycles of the five-step SAIA-MH protocols, including cascade analysis, process mapping, creating an implementation blueprint, implementing micro interventions, and assessing the impact of micro interventions implemented and planning next steps. Of 33 staff engaged in SAIA-MH implementation (22 psychologists, 10 psychiatric technicians, and 1 physician), 4 psychologists (12.1%) transferred out of their facility during the follow-up intervention period. In the attentional placebo control arm,\of 37 staff enrolled (20 psychologists, 16 psychiatric technicians, 1 psychiatrist), 14 (7 psychologists, 7 psychiatric technicians, 37.8%) transferred out during the intervention period. Intervention health facility teams tested between 11 (minimum) – 25 (maximum) distinct micro interventions over the 24-month implementation phase. In 56.2% (n=108/192) of monthly assessment meetings the tested micro intervention was adopted by the team for integration into routine practice, 41.6% (n=80/192) were adapted or modified for further testing in the coming month, and 2.1% (4/192) were abandoned. More detailed descriptive information on problems identified and micro interventions tested is provided in supplementary appendix pg. 10.

Discussion

This study demonstrates that the SAIA-MH multicomponent implementation strategy improved care engagement and functional outcomes for outpatient MNS patients in diverse, real-world Ministry of Health facilities. Compared to the attentional placebo control, SAIA-MH was associated with greater functional improvement, better medication adherence, and improved timeliness of clinical visits. These findings suggest that SAIA-MH is an effective systems-level strategy for strengthening MNS service delivery in routine care settings in contexts like Mozambique.

This study advances the science of MNS care in LMICS in several important ways. To our knowledge, this study is the first to evaluate the effectiveness of a multicomponent implementation strategy aimed at optimization of MNS outpatient treatment cascades in a LMIC. To date, most global investment in clinical and implementation science research regarding MNS care globally has focused on testing novel clinical interventions focused on task-shifting MNS care to lower-level providers ¹. However, as shown through the control arm in the present trial, the majority of MNS patients currently receiving care in Mozambique are not returning for scheduled follow-up appointments and only approximately one quarter of patients who do return demonstrate adherence to their prescribed psychiatric medications. Of those patients who do return, they show limited improvement in function. Simply focusing on improving access to MNS care and increased training of providers may not result in population-level improvement in mental well-being. The positive outcomes in our trial suggest that systems analysis and improvement implementation strategies may hold promise for low-cost and rapid improvements in MNS treatment outcomes in LMICs. The highly adaptable SAIA-MH implementation strategy aims to support facility teams to utilize their own routine health facility data for barrier identification and prioritization of areas in the MNS care cascade that need improvement. However, the large effects on clinical outcomes observed in this study are likely influenced by the relatively low baseline MNS system performance (between 33–42% of visits with functional improvement at baseline) and lack of previously implemented MNS quality improvement strategies in this context. Further studies testing the SAIA-MH implementation strategy in diverse contexts are warranted.

Unlike most MNS trials that test a single clinical intervention or implementation strategy, the SAIA-MH strategy enables facility teams to design, implement, and evaluate their own low-cost micro interventions targeting contextually specific barriers to MNS care. Encouraging quality improvement teams to customize micro interventions to meet their unique circumstances promotes perceived ownership of the intervention and subsequent sustainability if the micro intervention is adopted for ongoing use ^36,37. An adopted micro intervention can be further adapted in later iterative cycles, consistent with the Dynamic Sustainability Framework proposed by Chambers et al. which views the interplay between interventions and their contexts as dynamic ³⁷. The importance of ensuring that quality improvement interventions are contextually appropriate has been widely discussed ^38,39. This dynamic contextual flexibility may be especially important for interventions to be effective in the quickly evolving implementation landscape of MNS care in LMICs. Future work on SAIA-MH could examine the transferability of the implementation strategy across mental healthcare providers as well as other MNS care settings such as school-based settings, humanitarian or post-conflict settings, correctional or detention settings, general medical settings, as well as digital or tele-mental health platforms.

Limited evidence exists on what components of systems improvement interventions may be most effective in LMIC contexts. A previous Cochrane systematic review of audit and feedback interventions mostly delivered in high-resource contexts found that strategies are most effective when baseline system performance is low, feedback is delivered by a supervisor or senior colleague, feedback is provided often, and when feedback includes specific measurable targets and a formal implementation plan – all key design elements of the SAIA-MH implementation strategy tested here ⁴⁰. Given the positive effects observed in this trial, more implementation research is warranted to understand the optimal design of systems and quality improvement approaches to rapidly and effectively improve the quality of MNS care as it is scaled-up in LMIC contexts, often led by task-shared providers with relatively limited training, supervision, or continuing educational opportunities.

Per reviewer request, supplementary post-hoc subgroup analyses by MNS diagnosis were conducted and showed mixed findings. These analyses were exploratory and not powered to detect subgroup differences, and findings were mixed. While no inferences regarding differential effectiveness can be made, the results may inform hypotheses for future adequately powered studies examining heterogeneity of effects and generalizability of SAIA-MH across settings with differing diagnostic case mixes.

Additional unpowered post-hoc subgroup analyses by HIV/TB comorbidity status were conducted and yielded mixed findings. The complexity of HIV/TB care in LMIC settings may introduce barriers not fully addressed by the SAIA-MH systems-level strategy as implemented in this study. Further research is needed to determine whether pairing SAIA-MH with complementary HIV/TB interventions or adapting it to better address comorbidity-related barriers, could improve outcomes for MNS patients with complex comorbidities.

Our study had some key limitations. To measure primary and secondary outcomes in this trial we implemented enhanced patient tracking tools and registries within both intervention and control health facilities during the baseline pre-intervention period. Thus, it is likely that the control arm also had improvement in patient outcomes simply as a measure of improved patient tracking across follow-up visits. This was a single-blind study, with only the outcomes assessor masked to condition. Health facility teams were aware of assignment and could have disclosed information to patients or had biased expectations from assignment. Outcome data used for primary and secondary trial outcomes was sourced from routine clinical records completed during each patient visit by psychiatric technicians. Thus, data quality and reliability could have been lower than researcher-implemented data collection protocols. While SAIA-MH is designed for system-level implementation regardless of patient diagnostic mix, the high proportion of epilepsy diagnoses in the study population may limit generalizability to systems serving different patient populations. Last, diverse MNS conditions may differentially affect domains of functioning captured by the WHODAS 2.0, which could introduce bias when comparing overall function improvement across diagnoses. Further research is needed to assess measurement stability, domain-level performance, and cross-diagnostic comparability of the WHODAS 2.0 for MNS patients in primary care LMIC settings.

In conclusion, this trial demonstrates that the SAIA-MH strategy can improve appointment attendance, medication adherence, and patient functioning among MNS patients receiving routine outpatient primary care in an LMIC context. Future studies should examine whether and how differences in diagnostic and comorbidity case mix modify the effectiveness of SAIA-MH in reducing gaps in task-shared MNS treatment.

Research in context.

Evidence before this study

A 2020 systematic review of implementation science for depression interventions globally by our team showed no evidence-based implementation strategies focused on optimizing the mental, neurological, and substance-use disorder (MNS) treatment cascade in primary care ¹. This was supplemented by an ad hoc search of PubMed and Google Scholar for English and Portuguese reports published through 2022 of studies regarding care cascade outcomes for primary mental healthcare in Sub-Saharan Africa. Search terms used included: “mental health services” [mesh], “mental disorders”, “mental health”, “MNS disorders”, “psychiatric”, “depression” [mesh], “epilepsy” [mesh], “schizophrenia” [mesh] and “primary health care” [mesh], “community health services” [mesh], “primary care”, “district mental health” and “developing countries” [mesh], “low- income countries”, “middle-income countries”, “LMIC”, “Sub-Saharan Africa”, “Mozambique”, “Southern Africa”, and “health services administration” [mesh], “quality improvement” [mesh], “implementation science”, “implementation”, “quality improvement”, “task sharing”, and “patient compliance” [mesh], “treatment adherence and compliance” [mesh], “adherence”, “care cascade”, “treatment cascade”, “loss to follow-up”, “treatment coverage”. Reference lists of identified articles of interest were abstracted to supplement articles found in the initial search. We repeated this search when completing this article in 2025 and added new supplementary articles. We found that mental health system performance is documented as poor, with inappropriate polypharmacy exceeding 80% of cases, loss-to-follow-up (LTFU) exceeding 40% and sub-optimal adherence exceeding 60% in many settings^2–5. Similar patterns of suboptimal LTFU, adherence, and treatment outcomes have been reported across many other LMICs^6,7. A pilot study conducted by our team in Mozambique (September 2018 to August 2019) adapted and tested the Systems Analysis and Improvement Approach for Mental Health (SAIA-MH) which is a low-cost multicomponent implementation strategy aimed at improving appointment attendance, medication adherence, and patient functioning for outpatient primary mental healthcare in Mozambique. Our pilot study showed similar quality issues to other contexts, with rates of LTFU >50%, sub-optimal medication adherence >80%, and <30% of patients achieving functional improvement. Six months of SAIA-MH implementation showed the strategy to be feasible to implement and associated with rapid improvements in MNS care cascade performance ⁸.

Added value of this study

In contrast to most existing implementation strategies which may focus on one small element of the mental health care cascade like supervision or provider training, SAIA-MH is a multicomponent implementation strategy focused on optimizing an entire treatment cascade. SAIA blends facilitation, enhanced local clinical consultation, and the creation of facility-level learning collaboratives with systems-engineering tools in a 5-step approach specifically developed for task-shared providers, which include: (1) cascade analysis to visualize treatment cascade drop-offs and prioritize areas for system improvements; (2) process mapping to identify modifiable facility-level bottlenecks; (3) identification and implementation of modifications to improve system performance; (4) assessment of modification effects on the cascade; and (5) repeated analysis and improvement cycles. We found that SAIA-MH leads to substantially improved patient outcomes, including 46.0 percentage points higher functional improvement or low functional impairment, 18.1 percentage points higher medication adherence, and 18.4 percentage points higher appointment attendance. These findings did not differ by patient gender.

Implications of all the available evidence

Systems Analysis and Improvement implementation strategies focused on cascade analysis, process mapping, and guided structured implementation of repeated testing and evaluation of micro interventions for process improvement can lead to rapid gains in patient outcomes in LMIC MNS primary care settings. The SAIA-MH implementation strategy shows evidence of effectiveness in increasing appointment attendance, medication adherence, and functioning for MNS patients treated in outpatient primary care. Further research across implementation contexts is needed to evaluate how variation in diagnostic and comorbidity case mix may influence SAIA-MH effectiveness.

Data sharing:

Data emanating from enrolled and consented SAIA-MH implementers will be submitted to the National Institute of Mental Health Data Archive (NDA) at https://nda.nih.gov/edit_collection.html?id=3898.