The Impact of a Family-Based Economic Intervention on the Mental Health of HIV-Infected Adolescents in Uganda: Results From Suubi + Adherence

Abstract

Purpose:

This study examines the extent to which three mental health measures (hopelessness, depression, and poor self-concept) are improved through a family-based economic intervention implemented among adolescents living with HIV in Uganda.

Methods:

We used repeated measures from Suubi + Adherence, a large-scale 6-year (2012–2018) longitudinal randomized controlled trial. Bivariate analyses were conducted to test for observable group differences between the intervention and control conditions. Multilevel piecewise repeated measure mixed models were then conducted to assess hypothesized time × intervention interaction in changes in hopelessness, depression, and self-concept using participant-specific follow-up intervals.

Results:

At 24-month postintervention initiation, adolescents in the intervention condition reported a statistically significant lower hopelessness score than adolescents in the control condition (4.79 vs. 5.56; p = .018; N = 358). At 36-month follow-up, the intervention condition reported a statistically significant lower score on depression in the depression subgroup (N = 344) than the control condition (4.94 vs. 5.81; p = .029).

Conclusions:

The results indicate that family-based economic interventions such as Suubi + Adherence can effectively improve the mental health of adolescents living with HIV who evidenced mental health challenges at baseline. Given the promising positive effects of these interventions, at least in the short term, future studies should investigate strategies to promote the sustainability of these mental health benefits.

As of 2017, there are an estimated 36.9 million people living with HIV (PLHIV), of whom 1.8 million are children aged <15 years [1]. Studies in Sub-Saharan Africa (SSA) demonstrate that young people, including adolescents, residing in low-resource settings experience more HIV-related mortality and comorbidities [2–4]. Relatedly, such populations tend to face mental health challenges that are often because of the stigma associated with an HIV diagnosis, including feelings of hopelessness, depression, and poor self-concept [5–7].

Depression is the most common psychiatric disorder among PLHIV [8]. In SSA, the prevalence of depression among PLHIV is approximately 8% [9]. Notably, the prevalence of depression among Ugandan adolescents living with HIV (ALHIV) is more than 40% [10,11]. In addition, hopelessness and negative self-views (i.e., self-concept, self-esteem, etc.) are risk factors that increase the odds of developing a psychiatric disorder [12], including depression [7,13,14].

For vulnerable subpopulations such as adolescents who reside in low-resource settings, it is imperative to develop effective, sustainable, and implementable interventions that have a positive psychological impact [15]. To this end, family-based economic (FE) interventions have been shown to positively improve the psychosocial well-being of underserved ALHIV [16,17]. This may be due, in part, to the assets effect as hypothesized by asset theory, which posits a positive psychological and social impact on adolescents as a result of parental/family assets ownership [18,19]. Moreover, there are benefits to the use of peer mentoring programs to assist in the development of essential financial skills among individuals with mental health complications [20]. Likewise, financial literacy workshops have been shown to decrease depressive symptomology and improve mental health functioning among Sub-Saharan adolescents affected by HIV [21]. For example, having a financial saving account can be empowering and improve the psychological well-being among young people who reside in Ghana, Kenya, and Uganda [22,23].

Although FE interventions have been shown to increase savings and improve HIV outcomes in SSA populations [24,25], robust and longitudinal analyses are necessary to delineate their short- and long-term impacts on psychological outcomes. In response, the purpose of the present study is to determine the extent to which mental health conditions (i.e., hopelessness, depression, and poor self-concept) are improved through an FE intervention that has been implemented in a low-resource setting among ALHIV. Specifically, we leverage a large-scale, longitudinal randomized control trial, Suubi + Adherence, to delineate the relationship mechanisms between economic factors and mental health challenges among ALHIV in Uganda.

Methods

We used data from the Suubi + Adherence study, a 6-year (2012–2018) National Institutes of Health–funded longitudinal cluster randomized controlled trial (Clinical trials identifier #NCT01790373). The study recruited 702 ALHIV, aged 10–16 years, across 39 health clinics in southwestern Uganda. Repeated measures data were collected from adolescents at five time points (baseline, 12 months, 24 months, 36 months, and 48 months), using evidence-based clinical measures and standardized, culturally adapted interviewer-administered assessments. Research assistants were trained in good clinical practices, Collaborative Institutional Training Initiative certificate, and National Institutes of Health certificate for the protection of research participants.

Economic intervention

The Suubi + Adherence intervention (treatment condition) included four sessions on financial management and training in income-generating activities and combined a matched savings account—opened in the child and caregiver names. The account was matched on 1:1 ratio. The intervention was intended to meet the needs associated with managing HIV as a chronic illness, including support for adherence to antiretroviral therapy (ART), microenterprises to generate family income, and/or to pay for schooling. Participants were also paired with a mentor and could attend 12 educational sessions, including, but not limited to, financial planning, business development, saving, setting short- and long-term goals, and avoiding risk-taking behaviors. Intervention was provided for the first 24 months.

Bolstered standard of care

All participants (control and treatment conditions) received medical and psychosocial support as part of bolstered standard of care (SOC). In the Ugandan context, lay counselors trained in standardized ART adherence counseling provide psychosocial support to patients at HIV care clinics; patients also receive about 2–4 sessions of adherence counseling at initiation of care. Realizing that such adherence counseling varies significantly, the control arm usual care was enhanced (bolstered) with adherence sessions to provide standardized and sufficient adherence counseling including up to six sessions to review HIV, ART, as well as ART resistance and adherence.

Randomization

Stratified random sampling was used to assign clinic/health centers to four strata based on two characteristics: (1) geographical location (rural vs. urban); and (2) health facility (hospital vs. health centers). The restricted randomization technique of Hayes and Moulton was implemented within the four strata to assure overall health facility balance across the study arms [26]. Each of the original 40 health facilities was randomly assigned to one of the two study conditions, such that all selected ALHIV in the same health facility received the same intervention, to reduce contamination. In sum, of the 40 health facilities, 20 facilities were randomly assigned to receive bolstered SOC, and the other 20 received the FE intervention plus the bolstered SOC. However, one health facility was disqualified because of lack of proper registration, resulting in 19 health facilities in the treatment arm.

To be included in this study, adolescents had to meet the following inclusion criteria: (1) aged 10–16 years; (2) HIV positive and having this status already disclosed to them; (3) receiving ART and care from one of the participating government of Uganda’s Ministry of Health accredited health facilities in the study area; and (4) living with a family (not necessarily biological parents). In the study health facilities, most ALHIV in this age group were perinatally infected. Although poverty was not a criterion of inclusion, participants were recruited from health centers located in poor communities, attracting children from low resourced rural households.

Ethics and consent

This study’s protocols were reviewed and approved by Columbia University (Protocol AAAK3852), the Makerere University School of Public Health (Protocol 210), and the Uganda National Council for Science and Technology (Protocol SS 2969) Ethics and Institutional Review Boards. All participants provided written assent, and caregivers gave informed consent for their adolescent’s participation.

Measures

Three domains of mental health were measured: hopelessness, depression, and self-concept. The instruments used have been previously tested and adapted to be culturally appropriate for use with adolescents in Uganda [17,21,27].

Hopelessness.

The Beck Hopelessness scale includes 20 items and assesses the extent of an individual’s pessimism and negative expectations about the future [28]. Each item is answered as true or false. Items in the inverse direction were reverse coded to create summated scores, with higher scores representing more hopelessness. The theoretical range was 0–20, with an acceptable Cronbach’s alpha of .65 at baseline.

Depressive symptoms.

Children’s depressive symptoms were assessed using the Children’s Depression Inventory (CDI), which is one of the most widely used standardized self-report instruments for assessing depressive symptoms among adolescents and has proven successful in several different cultural contexts [29,30]. The 14 items of the CDI used in the present study were adapted from the original 28-item long version scale, which measures both emotional and functional problems that correspond with depression in adolescents. Each item on the CDI has three response options that correspond to varying levels of symptomology for clinical depression [31]. Items are summed and higher scores indicate greater depressive symptoms. The theoretical range was 0–28, with an acceptable Cronbach’s alpha of .63 at baseline.

Self-concept.

We used 17 items adapted from the original 100-item of Tennessee Self-Concept Scale to measure adolescents’ self-concept [32]. Responses ranged from 1 = always false to 5 = always true. We summed the responses to create a total score, with higher scores indicating higher self-concept. The theoretical range was 17–85, with an acceptable Cronbach’s alpha of .75 at baseline.

Covariates.

We adjusted for baseline sociodemographic covariates, including age, gender (female vs. male), primary caregiver (parents vs. grandparents vs. other relatives), and school enrollment (yes vs. no). Although no significant differences were observed at baseline, these are critical factors in determining overall well-being of ALHIV [3].

Statistical analyses

Analyses were conducted among the entire sample and three mental health subgroups in which the adolescents showed some challenges with mental health at baseline: the hopelessness, depression, and low self-concept subgroups. The hopelessness and the depression subgroups included adolescents who scored above the mean (M = 5.66) on the Beck Hopelessness Scale and the CDI (M = 5.18) at baseline [31]. The low self-concept subgroup included those who scored below the mean (M = 67.36) on the Tennessee Self-Concept Scale at baseline [32].

We present sample characteristics among the entire sample and across subgroups in Table 1 and Supplementary Tables 1.1–1.3, respectively. Margin plots were used to show the impact of time × intervention interaction for the five time points. p values were set a priori and considered statistically significant at <.05. Multilevel piecewise repeated measure mixed models were conducted to assess hypothesized time × intervention interaction in changes in hopelessness, depression, and self-concept using participant-specific follow-up intervals. A knot was placed at 24 months to examine the intervention effect separately during and after the intervention provided. Mixed effect models allow for estimates of changes in repeated measures in the presence of missing data, assuming those data were missing at random. We used unstructured covariance pattern matrices for the random effects [33]. The fixed effects included baseline covariates, baseline intervention condition assignment, time, and the time × intervention interaction. We included participant-specific and cluster-specific random intercepts to account for within-cluster and within-participant correlation. The components of the random effect were determined using the likelihood ratio tests and minimum values for the Akaike information criterion and the Bayesian information criterion. Although the likelihood ratio test, Akaike information criterion, and Bayesian information criterion values suggested a better model fit with random slopes at the participant level, we did not include them because they were extremely small and standard errors could not be estimated. The resulting model for each mental health outcome was a three-level multilevel model with measurement of each mental health outcome at each time follow-up (Level 1) being nested within participants (Level 2) who were in turn nested with clinics (Level 3). For follow-up time i, adolescents j, and clinic k, the model equation can be expressed as the following:

$$\begin{gathered} E\left(Y_{\mathit{\text{ijk}}}/b_{\mathit{\text{jk}}}\right){={\beta }_1+{\beta }_2{t}_{\mathit{\text{ijk}}}+{\beta }_3(\mathit{\text{tijk}}-24)}_{+}+\beta 4\mathit{\text{Grou}}{p}_{\mathit{\text{jk}}}\times t_{\mathit{\text{ijk}}}+{\beta }_5\mathit{\text{Grou}}{p}_{\mathit{\text{jk}}}\times {\left(t_{\mathit{\text{ijk}}}-24\right)}_{+}+{\beta }_6\mathit{\text{Grou}}{p}_{jk} \\ +{\beta }_7\mathit{\text{Gende}}{r}_{\mathit{\text{jk}}}+{\beta }_8\mathit{\text{Agejk}}+{\beta }_9\mathit{\text{School}}{ }_{\mathit{\text{jk}}}+{\beta }_{10}\mathit{\text{Caregive}}{r}_{\mathit{\text{jk}}}+b_{\mathit{\text{jk}}}+\mathit{\text{bk}} \end{gathered}$$

Table 1.

Baseline sample characteristics among the entire sample (N = 702)

	Total	Intervention (n = 358)	Control (n = 344)	Design-based F	p
Gender, n (%)				.02	.889
Male	306 (43.59)	155 (43.30)	151 (43.90)
Female	396 (56.41)	203 (56.70)	193 (56.10)
Age, mean (95% CI)	12.42 (12.01–12.74)	12.46 (11.96–12.96)	12.38 (12.02–12.73)	.08	.773
School enrolment, n (%)				.016	.901
Yes	613 (87.32)	312 (87.15)	301 (87.50)
No	89 (12.68)	46 (12.85)	43 (12.50)
Primary caregiver, n (%)				.981	.364
Parents	330 (47.01)	179 (50.00)	151 (43.90)
Grandparents	206 (29.34)	104 (29.05)	102 (29.65)
Other relatives	166 (23.65)	75 (20.95)	91 (26.45)
Beck Hopelessness Scale, mean (95% CI)	5.66 (5.18, 6.15)	5.61 (4.78, 6.43)	5.72 (5.28, 6.16)	.06	.814
Child Depression Inventory Scale, mean (95% CI)	5.18 (4.83, 5.53)	5.17 (4.63, 5.71)	5.19 (4.76, 5.62)	<.01	.957
Tennessee Self-Concept Scale, mean (95% CI)	67.36 (66.51, 68.22)	67.38 (66.13, 68.62)	67.35 (66.18, 68.52)	<.01	.973

All results were adjusted for within-clinic clustering using Taylor-linearized variance estimation.

CI = confidence interval.

The null hypotheses of no time × intervention interaction in changes in hopelessness, depression, and self-concept can be expressed as H₀: β₄ = β₅ =0 for the baseline to month 24 and H₀: β₄ + β₅ =0 from Month 24 to Month 48. For participants in the intervention condition from baseline to 24 months, the total effect included both time effect (β2 for baseline to 24 months; β3 for 24 months to 48 months) and intervention effect (β4 for baseline to 24 months; β5 for 24 months to 48 months; i.e., total effect = β2 + β4 for baseline to 24 months; β3 + β5 for 24 months to 48 months). For participants in the control condition who did not receive the intervention, there is only a time effect (β2 for baseline to 24 months; β3 for 24 months to 48 months). The null hypotheses were tested using the Wald test [34]. Alpha levels were set a priori at <.05. Standard errors and test statistics were adjusted for heteroscedasticity using robust Huber-Whiter sandwich variance estimation [35]. Unstandardized regression coefficients (β) and the 95% confidence interval (CI) are presented in Tables 2–4. All analyses were conducted using Stata SE. Version 15.

Table 2.

Multilevel piecewise repeated measure mixed model results of hopelessness among the entire sample and hopelessness subgroup^a

	Total			Hopeless subgroup
	β	95% CI		β	95% CI
Fixed effects
Gender
Female	.46	.13	.79	.56	.16	.96
Male	Ref.			Ref.
Age	−.31	−.38	−.25	−.16	−.24	−.09
School enrollment
No	1.10	.72	1.48	.50	−.07	1.08
Yes	Ref.			Ref.
Primary caregiver
Others	.69	.43	.96	.59	.19	.99
Parents	Ref.			Ref.
Intervention effect at baseline	.19	−.41	.80	.21	−.17	.60
Time effect
Before 24 months	−.03	−.05	−.02	−.12	−.14	−.10
After 24 months	.04	.00	.07	.13	.09	.17
Time × intervention interaction
Before 24 months: time × intervention	−.02	−.04	.01	−.04	−.07	−.01
After 24 months: time × intervention	.02	−.03	.07	.07	.01	.13
Intercept	8.90	8.14	9.66	9.87	9.02	10.72
Random effects
Clinic intercept	.14	.04	.55	.07	.01	.65
Individual intercept	2.93	2.39	3.58	1.77	1.26	2.47
Residual variances	7.37	6.86	7.90	7.01	6.50	7.56
Time × intervention interaction null hypothesis	χ2 (df)	p		χ2 (df)	p
Before 24 months (β4 = β5 = 0)	2.04 (2)		.361	7.67 (2)		.022
After 24 months (β4 + β5 = 0)	.03 (1)		.857	2.18 (1)		.140

All results were adjusted for within-clinic clustering using robust Huber-Whiter sandwich variance estimation.

CI = confidence interval.

^aHopelessness subgroup: children who had a total score above the average of the Beck Hopelessness Scale at baseline.

Table 4.

Multilevel piecewise repeated measure mixed model results of self-concept among the entire sample and low self-concept subgroup^a

	Total			Hopeless subgroup
	β	95% CI		β	95% CI
Fixed effects
Gender
Female	−1.58	−2.66	−.50	−1.51	−2.53	−49
Male	Ref.			Ref.
Age	.34	.11	.57	.13	−.23	.50
School enrollment
No	−2.59	−4.09	−1.10	−1.88	−3.57	−.20
Yes	Ref.			Ref.
Primary caregiver
Others	−2.01	−3.11	−.91	−1.44	−3.11	.23
Parents	Ref.			Ref.
Intervention effect at baseline Time effect	−.49	−2.37	1.39	.08	−1.40	1.55
Before 24 months	.02	−.02	.06	.36	.30	.41
After 24 months	−.02	−.10	.06	−.44	−.56	−.32
Time × intervention interaction
Before 24 months: time × intervention	.04	−.07	.15	.02	−.08	.11
After 24 months: time × intervention	−.03	−.21	.14	.03	−.15	.21
Intercept	65.54	62.28	68.80	59.05	54.62	63.48
Random effects
Clinic intercept	1.04	.20	5.34	1.23	.38	3.79
Individual intercept	28.19	24.25	23.77	18.03	14.34	22.68
Residual variances	71.22	67.03	75.68	66.51	61.93	71.45
Time × intervention interaction null hypothesis	χ2 (df)	p		χ2 (df)	p
Before 24 months (β4 = β5 = 0)	.80 (2)		.670	.97 (2)		.673
After 24 months (β4 + β5 = 0)	.03 (1)		.859	.50 (1)		.478

All results were adjusted for within-clinic clustering using robust Huber-Whiter sandwich variance estimation.

CI = confidence interval.

^aLow self-concept subgroup: children who had a total score below Tennessee Self-Concept the Scale at baseline.

Results

Baseline characteristics

Of the 702 adolescents, 358 were randomly assigned to the intervention condition, and 344 were assigned to the control condition (Table 1). In addition, 56% of the sample was female, and the mean age was 12 years (range 10–16). Approximately half (47%) of the participants’ primary caregivers were biological parents, and most participants were enrolled in school (87%). The mean baseline hopelessness score among the entire sample was 5.66 (range 0–20), the mean depression score was 5.16 (range 0–28), and the mean self-concept score was 67.36 (range 0–85). The groups were not statistically significantly different on baseline characteristics or primary outcomes.

Changes in mean scores of primary outcomes

Our subgroup analyses of the hopelessness-related measure at 24-month follow-up identified significant differences between the control and treatment conditions; the intervention condition reported a statistically significantly lower score than the control condition (4.79 vs. 5.56; p = .018). In addition, the intervention condition had a statistically significant lower score on depression than the control condition in the depression subgroup at 36-month follow up (4.94 vs. 5.81; p = .029).

The impact of time × intervention interaction on primary outcomes among the entire sample and subgroups by group assignment are presented using the margins plots in Figure 1. Similarly, changes in mean scores of depression, hopelessness, and self-concept over time are presented in Supplementary Table 2.1 (entire sample) and Supplementary Table 2.2 (subgroups).

Figure 1.

The impact of time × intervention interaction on hopelessness, depression, and self-concept among the entire sample and low self-concept subgroup by group assignment.

Mixed model findings

Table 2 shows the results of the multilevel piecewise mixed effect models assessing hypothesized time × intervention interaction in changes in hopelessness. There was no statistically significant time × intervention interaction on hopelessness among the entire sample during the entirety of follow-up. In the subgroup analyses, there was a statistically significant time × intervention interaction on hopelessness in the hopelessness subgroup from baseline to month 24 (β₄ = −.04, 95% CI [−.07, −.01] vs. β₅ = .07, 95% CI [.01, .13], β₄ ≠ β₅; χ² (2) = 7.67; p = .022), which indicates that the decrease of the intervention condition in the mean of hopelessness score from baseline to month 24 is statistically significantly larger than the decrease in the control condition. Specifically, we found that for each month participants received the intervention, the mean hopelessness score decreased by .16 (β₂ + β₄ = −.16, 95% CI [−.18, −.14]) points for each month. This suggests that after receiving the intervention for 24 months, the mean hopelessness score decreased by 3.84 points. For participants in the control condition, the mean hopelessness score decreased by .12 for each month and 2.88 in total from the baseline to Month 24. From Months 24 to 48, however, the time × interaction did not remain statistically significant (β₄ + β₅ = 0; χ² (1) = 2.18; p = .140).

Table 3 shows the results of the multilevel piecewise mixed effect models assessing hypothesized time × intervention interaction in changes in depression. Similar to the findings for hopelessness, there was no statistically significant time × intervention interaction on depression among the entire sample during the entirety of follow-up. In the subgroup analyses, there was a statistically significant time × intervention interaction on depression in the depression subgroup from baseline to Month 24 (β₄ = −.06, 95% CI [−.10, −.02] vs. β₅ = .10, 95% CI [.03, .17], β₄ ≠ β₅; χ² (2) = 7.86; p = .020), which indicates that the decrease of the intervention group in the mean of depression score from baseline to Month 24 is statistically significantly larger than the decrease in the control group. Specifically, we found that for each month participants received the intervention, the mean depression score decreased by .16 points (β₂ + β₄ = −.16, 95% CI [−.18, −.14]). This indicates that after receiving the intervention for 24 months, the mean depression score decreased by 3.84 points. For the participants in the control condition, the mean depression score decreased by .10 for each month and 2.4 in total from the baseline to Month 24. From Months 24 to 48, however, the time × intervention interaction did not remain statistically significant (β₄ + β₅ = 0; χ² (1) = 3.80; p = .051).

Table 3.

Multilevel piecewise repeated measure mixed model results of depression among the entire sample and depression subgroup^a

	Total			Hopeless subgroup
	β	95% CI		β	95% CI
Fixed effects
Gender
Female	.36	.04	.68	.75	.27	1.24
Male	Ref.			Ref.
Age	−.19	−.28	−.10	−.19	−.32	−.05
School enrollment
No	1.01	.52	1.49	.96	.25	1.67
Yes	Ref.			Ref.
Primary caregiver
Others	.45	.06	.83	.01	−.67	.69
Parents	Ref.			Ref.
Intervention effect at baseline	.18	−.36	.72	.29	−.30	.88
Time effect
Before 24 months	−.01	−.02	.01	−.10	−.14	−.07
After 24 months	−.02	−.06	.02	.08	.02	.14
Time × intervention interaction
Before 24 months: time × intervention	−.03	−.05	.00	−.06	−.10	−.02
After 24 months: time × intervention	.04	.00	.09	.10	.03	.17
Intercept	6.93	5.77	8.09	10.04	8.41	11.67
Random effects
Clinic intercept	.06	.003	1.02	<.001	–	–
Individual intercept	4.39	3.66	5.26	4.12	3.03	5.61
Residual variances	8.31	7.62	9.05	8.84	7.87	9.93
Time × intervention interaction null hypothesis	χ2 (df)	p		χ2 (df)	p
Before 24 months (β4 = β5 = 0)	5.22 (2)		.074	7.86 (2)		.020
After 24 months (β4 + β5 = 0)	1.38 (1)		.241	3.80		.051

All results were adjusted for within-clinic clustering using robust Huber-Whiter sandwich variance estimation.

CI = confidence interval.

^aDepression subgroup: children who had a total score above the Children’s Depression Inventory Scale at baseline.

Table 4 shows the results of multilevel piecewise mixed effect models assessing hypothesized time × intervention interaction in changes in self-concept. There was no statistically significant time × intervention interaction on self-concept either among the entire sample or in the low self-concept group during the entirety of follow-up, and the difference of the effect sizes between the intervention and control groups was also similar in the subgroup analysis from baseline to Month 24. Specifically, we found that for each month participants received the intervention, the mean self-concept score increased by .37 points (β₂ + β₄ = .37, 95% CI [.29, .45]) for each month. This suggests that after receiving the intervention for 24 months, the mean self-concept score increased by 8.88 points. For participants in the control condition, the mean self-concept score increased by .36 for each month and 8.64 in total from the baseline to Month 24. From Months 24 to 48, the difference between the intervention and control was still similar. Specifically, for the participants in the intervention condition, the mean self-concept score decreased by .41 points (β₃ + β₅ = −.41, 95% CI [−.54, −.28]) for each month and 9.84 in total from Months 24 to 48. For the participants in the control condition, the mean self-concept score decreased by .44 for each month and 10.56 in total from Months 24 to 48.

Discussion

ALHIV experiencing mental health challenges are a highly vulnerable population in SSA. Yet, there are few existing interventions that address their mental health needs. In response, the present study examines the extent to which an FE intervention has a positive mental health impact among ALHIV in SSA, many of whom experience poverty. Our findings demonstrate improvements in depression and hopelessness among adolescents who were randomly assigned to the Suubi + Adherence intervention and who showed symptoms of depression and hopelessness at baseline (i.e., depression and hopelessness subgroups); specifically, their improvements occurred during the 24 months of postintervention initiation. Our results corroborate past studies on economic interventions [36,37] and support theories that suggest that hopefulness, well-being, and other psychological states can impact and be impacted by economic growth, warranting the use of such FE programs with this population [38]. For adolescents who were randomly assigned to the Suubi + Adherence intervention but did not endorse mental health challenges at baseline, we found the FE intervention had no substantial impact on their mental health. In addition, the FE program did not have a significant impact on adolescents’ self-concept. This may be due to nuanced differences between self-concept, depression, and hopelessness, and further study may work to design intervention components that target the improvement of self-concept for ALHIV. Overall, however, this study expands on the body of knowledge regarding long-term economic-based interventions, such as the FE intervention, demonstrating that such interventions can have a positive impact on the mental health of ALHIV in SSA.

This study demonstrated mental health improvement during the 24 months of the FE intervention; however, these mental health improvements were not maintained at follow-up time points. This may be because of participants experiencing increased economic hardships/instability upon the discontinuation of our intervention or that this tangible loss of benefits after the intervention was removed could have induced hopelessness. Maturation may have also played a role in our findings. Specifically, without the FE intervention, over time, participants in the intervention condition may have slowly reverted to their past state of mind and experienced an increase in depression and hopelessness as a result. Recurrence of depression is a relatively common phenomenon with about one-third of people having another episode of depression once the first episode is experienced [39]. Our findings indicate further research is needed in this area, including examination of whether a longer duration of the economic intervention could help to sustain the mental health improvements we initially observed. Relatedly, it is possible that brief booster sessions of the key intervention components could be beneficial for maintaining the treatment gains after the initial intervention period has ended, as booster sessions are often a recommended component of mental health interventions including cognitive behavioral treatment for adolescents [40]. Further mediation analyses are needed to investigate which FE intervention components most impact mental health outcomes such as depression, hopelessness, and self-concept to inform how these specific strategies can be used to improve and sustain mental health for ALHIV over time.

A key strength of this study is its longitudinal study design that facilitates causal inferences; however, several limitations are relevant to consider when interpreting our findings. In particular, the self-report measures we used could cause social desirability bias despite their wide use in mental health research. Furthermore, a comprehensive mental health diagnostic interview, while more burdensome for our participants, could have potentially yielded more fine-tuned results. Adolescents’ mental health may have been affected by a variety of unidentified factors such as life events for within-subject variations and cohort effect or environment changes for between-subject variations, which were unaccounted for in the present study—although given that this is a randomized control trial, these counterfactuals should be relatively minimized. The randomized control trial method for this study reduces variability between sample clusters and minimizes selection bias, therefore increasing the reliability of our statistical outcomes.

Despite our limitations, this study adds to the currently scant literature available about the long-term mental health impact of economic-based interventions on adolescents in SSA living with HIV. Our results indicate that FE interventions can effectively improve the mental health struggles among ALHIV, an important group to target for prevention and intervention efforts, given the vulnerable nature of this population during a crucial period of development [41]. Notably, the promising positive effects of these interventions are relatively short term and signal a potential need for continued psychological support for the sustainability of mental health benefits.

IMPLICATIONS AND CONTRIBUTION.

The study adds to knowledge on long-term family-based economic interventions, demonstrating that these kinds of interventions have the potential to effectively improve the mental health struggles among adolescents living with HIV in Sub-Saharan Africa. Future studies should explore ways to further the promising positive effects of these interventions to promote the sustainability of mental health benefits long term.