Assessing the cost-effectiveness of economic strengthening and parenting support for preventing violence against adolescents in South Africa: an economic modelling study using non-randomised data

Introduction

Annually, 1.4 billion children are estimated to be affected by violence worldwide, incurring more than 5.1 million disability-adjusted life years (DALYs) (1,2). To support countries in their efforts to reduce child violence the World Health Organization (WHO) and eight other global agencies have endorsed seven evidence-based strategies to address violence against children, known collectively as INSPIRE (3). These seven strategies include Implementation and enforcement of laws, Norms and values, Safe Environments, Parent and caregiver support, Income and economic strengthening, Response and support services, and Education and life skills (3). The concrete recommendations included in the INSPIRE handbook mark a defining step forward for ending violence against children. Further evidence around the cost-effectiveness of INSPIRE strategies could help stakeholders to prioritise resources to strategies that are likely to have the greatest return on investment (4).

Economic evaluations of violence prevention interventions remain relatively uncommon, particularly in low- and middle-income countries (5–8). Experimental studies are recommended for evaluating cost-effectiveness (9–11). However, the cost and complexity of such studies, especially when more than one intervention is under consideration, means that another methodology may be better suited to filling the urgent need for estimates of cost-effectiveness. Decision analytic modelling is a much cheaper alternative approach for generating estimates of cost-effectiveness that uses data from secondary sources including published trials and observational studies (7,9). By modelling different scenarios, this approach is also suited to generating early estimates of cost-effectiveness for novel interventions that have not yet been implemented. The validity and transferability of findings from decision analytic models is often limited by the quality of published evidence that is available in the literature.

Decision-analytic modelling provides an opportunity to use evidence from published studies to generate early estimates of cost-effectiveness for INSPIRE strategies (12,13). A previous study to guide the choice of ‘best-buy’ INSPIRE strategies in South Africa found that across seven INSPIRE-aligned protective factors food security, caregiver supervision, and positive caregiving are likely to be promising targets for addressing multiple adolescent violence outcomes simultaneously (14). Consistent with positive youth development theories, these three protective factors were also found to combine additively, such that experiencing two or three of them together was associated with a significantly lower probability of experiencing multiple forms of violence, compared to experiencing one of them alone (14). Findings from an experimental trial in Tanzania also support the additive combination of parenting support and economic strengthening, compared to either of the two interventions alone (15).

Identification of positive and supervisory caregiving and food security as promising intervention targets for preventing multiple violence outcomes marks a step forward towards helping stakeholders identify INSPIRE strategies that are likely to be particularly effective (16). However, to inform resource-prioritisation further work is needed to select candidate interventions that could be used to strengthen food security, and/or positive and supervisory caregiving, and draw together sufficient data to compare the expected effects of selected interventions against their associated cost. Aiming to do this, we had three objectives in this study 1) Select expert- and evidence-based interventions for reducing violence via improving food security, caregiver supervision, or positive caregiving in South Africa; 2) Draw together cost and effectiveness data relating to selected interventions from a variety of high-quality sources; 3) Combine this data in a decision-analytic model to estimate the cost-effectiveness of selected interventions.

Methods

This economic modelling study builds on a non-randomised analysis of seven INSPIRE-aligned protective factors for violence victimisation in South Africa (14). The analysis found that food security, caregiver supervision, and positive caregiving were associated with lower odds of physical abuse, sexual abuse, emotional abuse, community violence victimisation, youth lawbreaking, and transactional sex. It also found that these protective factors combined additively, such that compared to experiencing a single protective factor alone, experiencing two, or all three simultaneously was associated with significantly lower odds of violence outcomes.

Our methodological approach of modelling the impact of interventions on violence outcomes had three steps. First, we consulted with regional experts in violence prevention to identify the best candidate interventions for reducing violence via improving food security, caregiver supervision, positive caregiving in South Africa. Second, we drew together cost and effectiveness data relating to the selected interventions from a variety of high-quality sources including published literature and online survey data. Third we combined these data in a decision-analytic model to estimate the total DALYs averted, cost, and cost-effectiveness of our selected interventions. Our choice to use DALYs averted was based on generating a single estimate of cost-effectiveness for selected interventions. We include full details of data-sources and assumptions throughout the analysis and report our study using the Consolidated Health Economic Evaluation Reporting Standard (CHEERS) checklist, S1 Appendix (17,18).

Choice of model

To estimate the cost-effectiveness of selected interventions, we constructed a probability tree model that modelled i) the effect of selected interventions on intermediary protective factors, and ii) the corresponding effect of the estimated improvement in intermediary protective factors on adolescent violence victimisation, Fig 1. Mixed evidence for whether group-based parenting support improves positive caregiving, meant that we did not consider parenting support or parenting support ‘plus’ grant linkage acting via this intermediary protective factor. Furthermore, our choice to measure intervention effectiveness in disability-adjusted life years (DALYs) averted meant that we could only consider violence outcomes with evidence for their attributable DALYs in our model. Of the six violence outcomes investigated in the non-randomised analysis of seven INSPIRE-aligned protective factors in South Africa, this included emotional abuse, physical abuse, and sexual abuse; but excluded youth lawbreaking, community violence victimization, and transactional sex (14). Finally, no evidence for an association between caregiver supervision and sexual abuse in the analysis of INSPIRE aligned protective factors, meant that we did not consider parenting support or parenting support ‘plus’ grant linkage acting on this outcome (14). With these considerations, we modelled community grant outreach acting on physical, emotional, and sexual abuse via food security, parenting support acting on physical and emotional abuse via caregiver supervision, and parenting support plus grant linkage intervention acting on the same outcomes via both food security and caregiver supervision.

• Download figure
• Open in new tab

Fig 1. Graphical representation of our probability tree model for estimating the effectiveness of interventions on adolescent violence outcomes.

*Estimate synthesized across two studies examining the impact of grants on household food security in sub-Saharan Africa and crosschecked using data from South Africa’s 2018 General Household Survey. +Estimate obtained from the primary analysis of a randomised evaluation of PLH Teen in South Africa.

Results

Details on the demographics of Mpumalanga Province, and the target population for interventions are summarized in Table 1. We observed that one out of three households with an adolescent in Mpumalanga live below the food poverty line, corresponding to 212,000 adolescents. We estimated that approximately 66,000 adolescents living in 64,300 households would benefit from the grant outreach intervention. We also estimated that over 10 years of intervention implementation, roughly 70,000 adolescents living in 68,000 households would benefit from the parenting support or parenting support plus grant linkage interventions.

Intervention cost-effectiveness

Intervention ACERs are summarised against a South African specific WTP threshold for each of the combinations of population-average versus high prevalence of violence and routine service delivery and trial-based intervention costs in Fig 2. An intervention is considered cost-effective if its ACER lies below the WTP threshold. The further an ACER lies to the right of the plane while remaining below the WTP threshold the higher its indicated cost-effectiveness.

• Download figure
• Open in new tab

Fig 2. Cost-effectiveness plane scatter plots of costs per adolescent over effectiveness (in DALYs averted) for four scenarios at A) Population-average prevalence of violence and routine service delivery costs, B) Population-average prevalence of violence and trial-based costs C) high prevalence of violence and routine service delivery costs, and D) high prevalence of violence and trial-based costs.

Diagonal lines indicate an evidence-based SA-specific willingness to pay threshold for health interventions estimated by Edoka et al. Strategies become increasing cost-effective as they lie to the north easternmost part of the cost-effectiveness plane below the willingness to pay threshold. Strategies above the line are not cost-effective at the stated threshold. The values underlying this graph are reported in S8 Appendix. Abbreviations: DALYs, Disability-Adjusted Life Years; USD, United States Dollar.

For a population-average prevalence of violence

Under a routine service delivery cost scenario, ACERs for community grant outreach and parenting support were both below the WTP threshold at USD 2,650 and USD 2,830 per DALY averted respectively. Integrated parenting support plus grant linkage had an ICER of USD 271 per DALY averted compared to the parenting support intervention.

Under a trial-based cost scenario, ACERs for community grant outreach and parenting support were both above the WTP threshold at USD 4,800 and USD 16,880 per DALY averted respectively. The integrated parenting support plus grant linkage was also not cost-effective with an ICER of USD 12, 850 per DALY averted.

For a high prevalence of violence

Under a routine service delivery cost scenario, ACERs for community grant outreach and parenting support were both below the WTP threshold at USD 1,330 and USD 1,310 per DALY averted. The integrated parenting support plus grant linkage intervention had an ICER of USD 177 per DALY averted compared to the parenting support intervention.

Under a trial-based cost scenario, the ACER for community grant outreach was below the WTP threshold at USD 2,400 per DALY averted, but the ACER for parenting support was above the WTP threshold at USD 7,780 per DALY averted. The ICER for integrated parenting support plus grant linkage was USD 6,430 per DALY averted and above the WTP threshold.

Discussion

We find that decision-analytic modelling is a valuable and low-cost approach for assessing the cost-effectiveness of three expert- and evidence-informed interventions aimed at reducing adolescent violence in South Africa. Our model indicates that implementing each of our three selected interventions using a routine service delivery model is likely to be cost-effective for reducing emotional, physical, and sexual abuse against adolescents. This finding is robust to all of the modelled combinations of low, average, and high intervention effectiveness with an average or high prevalence of violence, except for the combination of low intervention effectiveness and an average prevalence of violence in the target population. Across the three interventions, the most cost-effective option is likely to be the community grant outreach intervention. Thereafter, the integrated parenting support plus grant linkage intervention is estimated to be more cost-effective than the parenting support intervention alone. We also find that if our three selected interventions are implemented using trial-based costings, parenting support and parenting plus grant linkage support are unlikely to be cost effective, and community grant outreach is only likely to be cost-effective assuming average to high intervention effectiveness, and higher than average prevalence of violence in the target population.

Our study provides early evidence around the cost-effectiveness of a community outreach intervention to support household’s access to social grants if they are eligible, but not yet receiving them. The cost of the proposed grant outreach intervention per adolescent is equivalent to the value of three household child support grants and falls on the lower end of the scale of home-visiting interventions in high-income settings (42). While our results are promising, evidence around the effectiveness of cash transfers alone for preventing adolescent violence victimisation remains limited, and this is an important area for further research (43,44).

There are still limited economic evaluations of adolescent violence prevention interventions in low- and middle-income countries, particularly of large-scale interventions (45,7). Our findings around the cost-effectiveness of parenting support implemented using routine-delivery costs matches a previous economic evaluation of the original PLH Teen intervention in South Africa that also modelled a similar ‘lean’ scenario (6). An assumption in both of these studies is that the modelled routine-service delivery intervention is just as effective as the intervention implemented using trial-based costings. However, is increasingly supported by evidence that when implemented in the real-world at much lower cost PLH Teens is just as effective as that observed in experimental studies (46). The increasing roll-out and scale-up of parenting support programmes across sub-Saharan Africa will present further opportunity to validate this in the future (47).

Unlike for routine service delivery, our findings around the cost-effectiveness of trial-based delivery of parenting support do match the previous economic evaluation of PLH. Possible explanations for this discrepancy may be our focus on the indirect effects of parenting support via caregiver supervision alone rather than the total effect via all possible pathways. Other impact pathways have been found to include improved caregiver mental health, caregiver alcohol/drug avoidance, and improved economic welfare (48). Another explanation could be the previous evaluation modelled a higher prevalence of adolescent violence victimisation among families targeted by the PLH Teen intervention than we considered in either our population-average or high prevalence of violence scenarios(49).

While there is growing evidence on the enhanced effectiveness of combining multiple interventions with a ‘plus’ approach, there is still very little evidence around the cost-effectiveness of these interventions (50,51). Our findings of the incremental cost-effectiveness of combining an additional grant linkage session with PLH Teen supports the added value of integrated services for reducing adolescent violence victimisation. Building on this, future research should consider other possible approaches for combining parenting support with an economic strengthening intervention to simultaneously act via caregiver supervision and food security (52–54).

Our modelling study enabled us to consider important questions around the design of three interventions for reducing violence against children. These included how best to build on existing social policies such as South Africa’s CSG, the resources needed for implementing interventions at scale, and approaches for combining two interventions to simultaneously promote food security and caregiver supervision (3). Strengths of our study include its significantly lower cost compared to running a randomised evaluation of the three evaluated interventions, and also our use of a widely applicable methodology that could be used to generate similar evidence for other interventions and/or settings. However, our study also had limitations. While many of our model parameters were drawn from high-quality randomised studies, estimates of association between protective factors and violence outcomes were based on observational research and may be affected by sources of bias associated with this type of research design including unmeasured confounding (14). Gaps in the academic literature also meant that we had to make assumptions about some of the parameters in our model, for example the post-intervention decay in effectiveness of cash grants and parenting programmes. Several methodological choices also mean that our estimates of cost-effectiveness are likely to be conservative. These include our focus on DALYs averted for measuring cost-effectiveness, which while allowing us to generate a single estimate of cost-effectiveness for selected interventions, also meant we could not consider outcomes that lacked estimates of avoidable DALYs in our model including community violence victimisation, youth lawbreaking, or transactional sexual exploitation. Also, our focus on the impact of interventions via food security and caregiver supervision alone did not consider the full complement of protective factors via which selected interventions could act on adolescent violence outcomes. There is growing evidence that parenting support is also likely to act on adolescent violence victimisation via improving caregiver mental health, caregiver alcohol/drug avoidance, and improved economic welfare (48,55,56). Due to a lack of evidence in the wider literature, we were also unable to consider the wider societal benefits of preventing adolescent violence victimisation. Such benefits might include reduced health service use, social service use, and court case time. Finally, all three of the interventions considered in this study are likely to have benefits for adolescent wellbeing beyond reducing violence victimisation, for example by boosting school enrolment, or promoting mental health. Impacts across these additional domains of wellbeing should be considered for accurately valuing intervention cost-effectiveness in the future. Doing so may also support cross-sectoral buy-in from multiple government departments including health, education, and social development (57).

In South Africa, 40% of young people are estimated to experience some form of sexual, physical, or emotional abuse in their lives (58). Reviewing and analysing the literature on violence against children in South Africa, our study provides novel and policy-relevant evidence for much needed action to reduce child abuse. Our findings suggest that all three of our selected interventions are likely to be cost-effective so long as they are implemented using routine service delivery rather than a trial-based costing. The study also provides a comprehensive summary of evidence gaps that should be investigated as a priority for further informing the scale-up of interventions in this field. These include limited estimates of the prevalence of adolescent violence victimisation among the most vulnerable groups, the effects of cash transfers on adolescent violence victimisation or the mechanisms via which they act on this outcome, and the wider benefits of reducing violence against children to communities and society. Future attempts to quantify the cost-effectiveness of interventions like community grant outreach and parenting support studies should also account for their spill over effects on other areas of adolescent wellbeing including education and health.

Conclusion

In conclusion, we find that investments in community grant outreach, and parenting support interventions are likely to be cost-effective for preventing violence against adolescents when these programmes are implemented as routine services. Adding a grant linkage component to parenting support is likely to enhance the cost-effectiveness of this intervention and supports the value of integrated services to accelerate progress in preventing violence against children. Further research would strengthen the accuracy of our model estimates.

Supporting Information

S1 Appendix. CHEERS Checklist.

S2 Appendix. Description of hypothesised interventions, their duration, and staffing structure.

S3 Appendix. Summary of data sources used to estimate the effectiveness and cost of selected interventions.

S4 Appendix. Synthesis of the effect of Cash Grants on Household Food Insecurity Access Scale (HFIAS) from selected studies in Sub-Saharan Africa.

S5 Appendix. Assumptions about the modelled annual decay in intervention effectiveness and exit rates.

S6 Appendix. Estimated DALY per case of physical, emotional, and sexual abuse.

S7 Appendix. Sensitivity analysis of the effect of cash transfers on food security in South Africa using data from the 2018 General Household Survey.

S8 Appendix. Average and Incremental Cost Effectiveness Ratios.

S9 Appendix. Cost-effectiveness plane scatter plots of costs per adolescent over effectiveness (in DALYs averted) for four scenarios with the SA GDP per capita as the willingness-to-pay threshold at: A) Population-average prevalence of violence and routine service delivery costs, B) Population-average prevalence of violence and trial-based costs, C) High prevalence of violence and routine service delivery costs, and D) High prevalence of violence and trial-based costs.

S10 Appendix. Cost-effectiveness plane scatter plots of costs per adolescent over effectiveness (in DALYs averted) for four scenarios with the SA-specific threshold per capita as the willingness-to-pay threshold at: A) Population-average prevalence of violence and routine service delivery costs, B) Population-average prevalence of violence and trial-based costs, C) High prevalence of violence and routine service delivery costs, and D) High prevalence of violence and trial-based costs.