Return-on-investment for interventions to reduce adolescent pregnancy and child marriage in Southeast Asia: a five-country modelling study

Abstract

Introduction

Adolescent pregnancy remains a concern in Southeast Asia, with pregnancy both a driver and an outcome of child marriage or union. For Cambodia, Indonesia, the Philippines, Lao People’s Democratic Republic (PDR) and Malaysia, this study aimed to calculate the benefit-cost ratio of investing in evidence-based interventions to prevent adolescent pregnancy and child marriage, considering their complex interdependencies.

Methods

The population of girls aged 15–19 years was modelled at a national level, stratified by marriage, poverty and school enrolment status (known risk factors for pregnancy). Time-varying inputs for sociodemographics, contraception use, pregnancy and marriage were taken from population surveys (eg, Demographic and Health Survey [DHS]) and census data. A targeted review identified interventions with sufficient cost and effectiveness evidence for inclusion in modelling. A baseline scenario (constant intervention coverage over 2024–2030) was compared with investment scenarios, with interventions scaled up to reach an aspirational 95% coverage by 2030.

Results

Evidence-based interventions included in-school comprehensive sexuality education, cash transfers, educational support, life skills training, livelihood training and community dialogue. Across the five countries, baseline scenario intervention costs were US$1.4 billion over 2024–2030, with a projected 2.4 million unintended adolescent pregnancies and 2.1 million child marriages among girls 15–19 years. Scale-up of all interventions to reach 95% coverage by 2030 would cost an additional US$5.0 billion and, in return, avert 1.3 million unintended pregnancies and 1.1 million child marriages. This was estimated to generate US$13.2 billion in economic benefits by 2050 with a benefit-cost ratio of 2.7, increasing to 9.7 if interventions were maintained to 2050 and lifetime benefits considered. Achieving a more modest 50% intervention coverage by 2030 produced similar benefit-cost ratios but with reduced total impact.

Conclusion

For every US$1 invested to scale up and maintain high coverage of evidence-based interventions to reduce child marriages and adolescent pregnancies in Cambodia, Indonesia, the Philippines, Lao PDR and Malaysia, US$9.7 could be returned to the economy in the longer term.

WHAT IS ALREADY KNOWN ON THIS TOPIC.

WHAT THIS STUDY ADDS

• This study estimated the return-on-investment from scaling up evidence-based interventions to prevent adolescent pregnancy and child marriage in Cambodia, Indonesia, the Philippines, Lao People’s Democratic Republic and Malaysia, to inform investment decisions. It extends previous work by considering demand generation interventions rather than just the supply side, by having a focus on adolescents only, and by considering adolescent pregnancy and child marriage together, capturing their complex interdependencies.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Investing in evidence-based interventions to reduce adolescent pregnancy and child marriage is likely to have substantial health and economic benefits. There are significant gaps in the literature for interventions to reduce adolescent pregnancies and child marriage, suggesting that additional programme impact evaluation studies would be valuable.

Introduction

Adolescent pregnancy and child marriage are associated with adverse health and socioeconomic outcomes for girls and their children, including higher rates of maternal and perinatal morbidity and mortality, lower educational attainment and reduced workforce participation and productivity.¹ It is estimated that annually, there are over 3.7 million births to girls aged 15–19 years in the Asia and Pacific region. In low- and middle-income countries (LMICs) in Southeast Asia, adolescent fertility rates are relatively high (43 births per 1000 girls aged 15–19 years), and 1 in 14 adolescent girls have begun childbearing by 18 years of age.^{2 3} In Southeast Asia, 43% of pregnancies among adolescent girls are unintended, and a considerable proportion of them end in induced abortion,² which is legally restricted and mostly unsafe, causing significant maternal morbidity and mortality. In 2020, the maternal mortality ratio (MMR) was estimated at 134 women per 100 000 live births in Southeast Asia.⁴

In Southeast Asia, adolescent pregnancy and child marriage share a number of risk factors and are related to one another in complex ways. Non-school enrolment is a risk factor for adolescent pregnancy and child marriage. A survey conducted by UNICEF and UNFPA in 2022 found that, depending on the country, the prevalence of child marriage in Southeast Asia was 1.3–3.1 times greater among 15–19-year-old girls out of school compared with 15–19-year-old girls in school.⁵ Poverty is also a risk factor for adolescent pregnancy and child marriage, as well as being a risk factor for non-school enrolment. The same study by UNICEF and UNFPA reported that, depending on the country, adolescent girls in Southeast Asia living below the poverty line are 2.7–11.0 times more vulnerable to child marriage compared with those who are above the poverty line, suggesting that there is a large but heterogeneous impact of poverty on child marriages in the region.⁵ Poverty is also a significant factor influencing school dropout among adolescent girls, with 44% of girls from the poorest wealth quintile in LMICs never being enrolled in school or not continuing into secondary education.⁶

Much of the focus in this region has been on addressing traditional forms of child marriage (typically arranged or forced) through national strategies and legislation raising the legal age of marriage, with the assumption that reducing child marriage would also reduce adolescent pregnancy.⁵ Yet, over the last two decades, child marriage prevalence has not declined significantly, and in several countries adolescent fertility rates have stalled or increased.³ There is increasing recognition that not all early unions are non-consensual or formal,⁵ and recent studies have found that adolescent pregnancy can be an outcome but also a driver of child marriage.⁷ In Southeast Asia, three main typologies of child marriage and early union have been identified (noting that these may overlap and exist concurrently within countries): (1) traditional (arranged or forced) child marriage, which is often non-consensual though can involve consent by a girl; (2) love unions, where peers are married or live together as if married by choice and engage in sexual activity and (3) circumstantial child marriage, where union is pursued (with or without the girl’s consent) in response to unintended pregnancy or sexual abuse.⁵ Furthermore, a significant proportion of adolescent pregnancies occur before marriage; particularly so in the Philippines (36%), Viet Nam (31%) and Lao People’s Democratic Republic (PDR) (27%).⁸ Pregnancy (often unintended) in these settings can lead to a circumstantial child marriage or more formal ‘love marriage’ to avoid social sanctions associated with premarital pregnancy or sexual activity in conservative settings.^8–10 Therefore, when considering interventions and their impacts, it is important to consider different drivers and pathways; that is, prevention of some pregnancies can prevent child marriage, and prevention of some marriages can prevent adolescent pregnancies.

Previous work has estimated the return-on-investment of family planning considering only the supply side (ie, the return-on-investment of reaching zero unmet need, without specifying how to do so), as well as interventions to reduce child marriage in isolation. While certain countries have made progress in reducing adolescent pregnancies, the stalled overall progress in Southeast Asia^{2 5} suggests that investment is also needed in interventions to increase demand (eg, in-school comprehensive sexuality education (CSE) to achieve increases in modern contraception use in practice). In addition, the complex interactions between adolescent pregnancy and child marriage suggest there may be benefits of considering these outcomes and shared risk factors together. Therefore, the aim of this study was to calculate the return-on-investment from scaling up evidence-based interventions to prevent adolescent pregnancy and child marriage in Cambodia, Indonesia, the Philippines, Lao PDR and Malaysia to inform investment decisions. This study extends previous work by considering demand generation interventions, by having a focus on adolescents only and by considering adolescent pregnancy and child marriage together through an economic model that captures the complex interdependencies described above. The evidence generated can provide a scientific basis for the health and economic impact of investing in interventions to prevent adolescent pregnancies and child marriages in Southeast Asia.

Methods

Model structure and population

Independent compartmental models were used to replicate the national population of girls aged 15–19 years over time in Cambodia, Indonesia, the Philippines, Lao PDR and Malaysia. The population was restricted to girls aged 15–19 years, as opposed to the full range of adolescence defined by the WHO (10–19 years),¹¹ as data are limited for 10–14 years (eg, questions not asked in DHS), and fertility rates are low in that age group in the Asia region. The 15–19-year-old population was categorised as either married versus unmarried; poor or not poor (defined as individuals living below or above the poverty line based on country-specific poverty estimates) and enrolled in school or not (additional details in online supplemental appendix C).

Each year, girls turning age 15 enter the model (based on demographic projections), and girls turning age 20 leave the model. The model includes a rate of pregnancy for 15–19-year-old girls, which varies according to marriage, poverty and school enrolment status (online supplemental tables A3, A4). Given the overall pregnancy rate for each subgroup, as well as their unmet need for family planning, contraception use prevalence, methods mix and average failure rate, the annual risk of pregnancy can be calculated for the proportion who are sexually active with unmet need and the proportion who are using some form of contraception. Over time, the unmarried groups can become married, which may lead to dropping out of school for those enrolled (figure 1). A proportion of marriages are assumed to occur as a direct result of pregnancy and hence are reduced when contraception use increases or sexual initiation is delayed, whereas the remaining proportion of marriages are independent of pregnancy rates.

Figure 1. Model schematic. The population of 15–19-year-old girls is categorised according to marriage, poverty and school enrolment status. Each group has a pregnancy rate (red arrow) that varies according to marriage, poverty and school enrolment status. Over time, the unmarried group can become married, which may lead to dropping out of school for those enrolled. Interventions can be provided to some subpopulation groups that influence parameters for rates of pregnancy, marriage and school enrolment, which will have flow-on effects to other outcomes. CSE, comprehensive sexuality education.

Further details, including parameters and time-varying inputs related to demographics, contraception use, pregnancy and marriage, are in Appendix A (online supplemental tables A1-A3).

Settings

The modelling was undertaken for five countries in the Southeast Asia region: Cambodia, Indonesia, Lao PDR, Malaysia and the Philippines. The country selection was based on the availability of data and prior evidence mapping studies that had been undertaken by UNICEF and UNFPA.^{5 12 13}

Interventions

Several interventions are available that can reduce unmet need for family planning (leading to reduced pregnancies), increase school enrolment or reduce the marriage rate (figure 1). As well as having direct effects on these parameters, interventions can have additional indirect effects due to model dynamics. For example, because the rates of marriage and pregnancy are lower among those enrolled in school, interventions that increase school enrolment will also indirectly reduce pregnancies and marriages. Similarly, a proportion of marriages are estimated to occur because of adolescent pregnancy, and interventions to reduce pregnancy will therefore also indirectly reduce marriage.

In order to be modelled, interventions require quantitative evidence and parameters for a target population (eg, married or unmarried, poor or not-poor and enrolled or not enrolled); a baseline coverage among 15–19-year-olds in the target population; a unit cost (cost per person reached per year) and an effect size (eg, OR or relative risk (RR)) for one or more of increased contraception prevalence rate, delayed sexual initiation, reduced marriage rate or increased school enrolment.

For the purposes of this study, a targeted literature review was undertaken to identify interventions with sufficient evidence for inclusion in economic modelling. The review considered adolescent pregnancy and child marriage prevention interventions highlighted in recent systematic reviews^{14 15} and child marriage modelling studies conducted in other settings.^{16 17} These key publications were supplemented with targeted searches on PubMed, MEDLINE (Ovid) databases and Google Scholar to ensure that any relevant literature published after 2016 (for adolescent pregnancy interventions) and 2021 (for child marriage/early union interventions) was included in the review (including grey literature). Studies reporting at least one quantitative measure of the impact/effect size and/or cost of an intervention were considered; effect sizes and costs from studies in LMICs were prioritised in the final list of interventions. The search strategy, inclusion criteria, keywords/search terms and additional details of the review of evidence are in Appendix B (online supplemental tables B1, B2).

Interventions identified were in-school CSE, cash transfers, educational support, life skills training, livelihood training and community dialogue. Target populations, unit costs and effect sizes of included interventions are summarised in table 1, with intervention specifications described further in online supplemental appendix B3. Baseline coverage of interventions is in Appendix D (online supplemental table D1). Baseline contraceptive use is assumed to be through a mix of modern and traditional methods (online supplemental table D2), but any increased demand satisfied as a result of increased intervention coverage is assumed to be modern methods only.

Table 1. Interventions used in this analysis.

Intervention	Target population (among 15–19-year-old girls)	Unit cost (cost per person per year, inflated to 2023 US$)	Effects: unmarried			Effects: married		Effect: school enrolment/retention	Sources
			Demand met/CPR	No demand/delayed sexual initiation	Marriage	Demand met/CPR	No demand/delayed sexual initiation
School-based comprehensive sexuality education (CSE)	Enrolled	US$3.74							Cost: Cambodia country estimates34
			RR=1.48			RR=1.48			Taylor et al (South Africa)27
				OR=2.797					Ramirez-Villalobos et al (Mexico)28 OR for not being sexually active, applied as a 6-month delay
Life skills training	All (enrolled+not enrolled)	US$40.75							Cost: average of Amin et al (Bangladesh)26 and Bandiera et al (Uganda)35
			RR=1.504			RR=1.504			Bandiera et al (Uganda) (50.4% increased likelihood of consistent condom use)
					RR=0.69				Average of Amin et al (Bangladesh), Bandiera et al (Uganda) and Mehra et al (India)26 35 36
								OR=1/0.84	Amin ī (Bangladesh) (16% reduction in school dropout at endline)26
Conditional cash/asset transfers	Below poverty line	US$77.78 (KHM), 87.50 (IDN), 110.53 (PHL), 100.00 (LAO), 116.67 (MYS)							Cost: average of Austrian et al (Kenya),37 Baird et al (Malawi)38 and Erulkar et al (Ethiopia),23 divided by average household size to get cost per person per year.
				RR=1.38					Baird et al (Malawi)39(38% reduced risk of sexual initiation)
					RR=0.56				Average of Austrian et al (Kenya),37 Baird et al (Malawi)38 and Erulkar et al (Ethiopia)23
								RR=1/0.46	Baird et al (Malawi)38
Educational support	Below poverty line	US$56.50			RR=0.69			OR=1/0.86	Amin et al (Bangladesh) (31% reduction in risk of child marriage; 14% reduction in school dropout at endline)26
Livelihood training	Unmarried	US$56.50							Cost: Amin et al (Bangladesh)26
			RR=1.133						Jensen (India) (13.3% reduction in pregnancy at endline)40
					OR=0.767			OR=1/0.84	Amin et al (Bangladesh) (23% lower likelihood of being married as children at endline vs baseline; 16% reduction in school dropout at endline)26
Community dialogue	Unmarried	US$18.00							Cost: Erulkar et al (Burkina Faso, Ethiopia, Tanzania)23
			RR=1.31						Diez et al (Spain) (31% reduction in adolescent fertility rate)24
					RR=0.33				Erulkar et al (67% reduction in risk of being married among girls aged 15–17 years (arm 1, Burkina Faso) (aRR=0.33))23

Summary of target populations, costs, effects, and sources. Further details are available in online supplemental tables B1, B2 and in intervention specifications (online supplemental file B3).

aRR, adjusted relative risk; CPR, contraceptive prevalence rate; OR, odds ratio; RR, realtive risk.

Scenarios

Three scenarios have been considered for each country for the period 2024–2030:

1. Baseline: assuming no changes to intervention coverage.

2. Individual intervention scale-up: linearly scaling up each intervention in isolation from 2024 to reach 95% of their target population by 2030.

3. All intervention scale-up: linearly scaling up all interventions from 2024 to reach 95% of their target population by 2030.

Health and social outcomes

For each scenario, the main outcomes compared with the baseline are the number of unintended adolescent pregnancies and child marriages averted over 2024–2030. Secondary outcomes include unmet need for family planning (all methods, although interventions only increase the use of modern methods), maternal deaths and stillbirths averted (due to unintended pregnancies averted, used in the calculation of economic benefits) and increases in school enrolment.

Costs

Economic costs of interventions were considered from a service provider’s perspective for the period 2024–2030 inclusive, presented in 2023 US$ and discounted at 3% per annum. The cost of the interventions for each year was calculated as the unit cost multiplied by coverage. The costs of contraception for each scenario were estimated using an ingredients-based approach, based on the estimated number of users and methods mix. For the married and unmarried population groups, this is based on estimates of the number of users, methods mix and unit cost per person per year for each method (calculated in the Lives Saving Tool,¹⁸ including commodities/consumables, human resources, training and overhead costs; online supplemental appendix D).

Economic benefits

Health benefits from reduced unintended adolescent pregnancies and child marriages averted were converted to economic benefits across the following domains (figure 2):^{16 19–22}

Figure 2. Health and economic benefits of interventions. Modelled relationship between interventions, improvement in outcomes (reduced adolescent pregnancy, reduced child marriage and increased school enrolment) and economic benefits. CSE, comprehensive sexuality education; GDP, gross domestic product.

1. Direct costs averted reduced cost of antenatal and delivery care due to unintended pregnancies averted.

2. Education resulting from unintended pregnancies averted: average increase in years of school completed due to unintended pregnancies averted, leading to increased future productivity and lifetime earnings.

3. Workforce participation: increased participation in the workforce due to years of life gained (from unintended pregnancies multiplied by the MMR) or time out of the workforce due to unintended pregnancies.

4. Workforce productivity resulting from child marriages averted: prevention of child marriages leading to increased education, which (a) increases productivity and earnings in each form of employment and (b) increases likelihood of being in formal versus informal employment.

5. Social: disability-adjusted life years averted converted to economic benefits from mortality (a small percentage of unintended pregnancies) averted.

Total economic benefits were calculated based on the difference in outcomes between scale-up and baseline scenarios. Benefits were calculated up to 2050, but only for the cohort receiving the interventions over 2024–2030. Additional details are in online supplemental appendix E.

The return-on-investment from the individual intervention and all intervention scale-up scenarios compared with the baseline scenario was measured using the benefit-cost ratio (economic benefits divided by additional costs). A benefit-cost ratio greater than one suggests that, from a societal perspective, the benefits are greater than the costs.

Sensitivity analysis

Three sensitivity analyses were run. First, coverages of the interventions were only assumed to reach a maximum coverage of 50% by 2030, accounting for potential implementation challenges in each country. Second, a more optimistic scenario was considered where all the interventions were scaled up to reach 95% coverage within 1 year (by the end of 2025), assuming all individuals in the cohort were potentially exposed to interventions. Third, a longer time horizon, where interventions were scaled up to reach 95% coverage by 2030 and then maintained through to 2050, with lifetime benefits considered (ie, the entire future trajectories of girls receiving the interventions between 2024 and 2050).

Ethics

Human Research Ethics Committee approval was not required, as the study had no participants and used only aggregate publicly available data as model inputs.

Patient and public involvement

It was not applicable to involve patients in the research. However, consumers (ie, stakeholders from UN agencies and governments in countries being modelled) were consulted regarding the research design and reporting of research outcomes.

Results

Baseline scenario

In aggregate across all five countries, the baseline scenario of maintaining constant intervention coverage was estimated to cost US$1.4 billion over 2024–2030 (discounted), and there were 2.4 million unintended pregnancies and 2.1 million child marriages (table 2 and figure 3).

Table 2. Summary of full intervention scale-up scenario for each country.

Scenario	Total cost (millions)	Unintended pregnancies	Child marriages
Baseline	US$1385	2 609 000	2 139 000
	Total additional cost (millions)	Unintended pregnancies averted	Child marriages averted	Economic benefits (millions)	Benefit-cost ratio	Longer-term BCR*
In-school CSE	US$141	129 000	3000	US$512	3.6	8.4
Cash transfers	US$408	64 000	124 000	US$1128	2.8	10.8
Life skills training	US$2069	935 000	311 000	US$5903	2.9	8.7
Educational support	US$99	8000	84 000	US$519	5.3	17.3
Livelihood training	US$1683	73 000	227 000	US$1990	1.2	5.8
Community dialogue	US$ 638	178 000	686 000	US$4413	6.9	37.2
Full scale-up	US$ 4946	1 295 000	1 114 000	US$13 209	2.7	9.7

Values were rounded before addition or division and may lead to discrepancies across rows or columns.

^*Interventions were maintained up to 2050, and lifetime benefits were considered.

BCR, benefit-cost ratio; CSE, comprehensive sexuality education.

Figure 3. Impact and annual cost of the full intervention scale-up scenarios compared with the baseline (aggregated across five countries). Panels show annual unintended pregnancies (top-left); annual child marriages (top-right); annual number of girls enrolled in school (middle-left); weighted averaged unmet need for family planning (middle-right) and annual cost (bottom). Total costs are in 2023 US$, discounted at 3% per annum. CSE, comprehensive sexuality education.

Scale-up of individual interventions

When interventions were scaled up in isolation, they influenced different outcomes depending on their target population and effect parameters (table 2). Life skills training had the greatest total impact for preventing unintended pregnancies (and the second best for preventing child marriages), influenced by both its effectiveness and the fact that it is available to everyone in the model. Community dialogue had the greatest impact on preventing child marriages. Educational support had the lowest total costs, followed by in-school CSE. Life skills training had the highest total cost, largely due to its bigger target population. The benefit-cost ratios of individual interventions ranged from 1.2 for livelihood training to 6.9 for community dialogue.

Combined intervention scale-up

In aggregate across all five countries, scaling up all interventions to reach 95% coverage by 2030 was estimated to cost an additional US$5.0 billion more than business-as-usual over 2024–2030 but could avert 1.3 million unintended pregnancies and 1.1 million child marriages among girls aged 15–19 years. This investment could generate US$13.2 billion in economic benefits by 2050, with an overall benefit-cost ratio of 2.7 (table 2).

Differences between countries

There was also some variation between countries in the benefit-cost ratio of the full intervention scale-up scenario, which ranged from 1.9 by 2050 in the Philippines to 3.0 in Indonesia and 4.0 in Malaysia (online supplemental appendix F). This variation was a result of differences in baseline school enrolment, pregnancy and marriage rates, as well as GDP per capita and workforce participation rates. For example, Malaysia has relatively lower adolescent pregnancy and child marriage rates but the highest GDP per capita, which greatly influences economic benefits and workforce productivity calculations and hence resulted in the highest benefit-cost ratio; this makes a strong case for closing the gaps, even in upper-middle-income countries. In contrast, the Philippines has both the lowest adolescent pregnancy and child marriage rates, resulting in relatively lower (but still favourable) potential impacts from the package of interventions. Cambodia, on the other hand, has the lowest GDP per capita and rates of pregnancy and child marriage around the middle of the other countries; this resulted in the second lowest benefit-cost ratio for the full intervention scale-up scenario. Lao PDR has both the highest adolescent pregnancy and child marriage rates and hence the greatest opportunity for relative impact of interventions, but the lowest workforce participation and second lowest GDP per capita.

Sensitivity analyses

If implementation challenges meant that intervention coverage was only scaled to reach 50% coverage by 2030, the overall benefit-cost ratio was similar (3.0 vs 2.7 for the full scale-up analysis), but the total impact was lower. If intervention scale-up occurred immediately rather than gradually over time, then the benefit-cost ratio was the same, but the number of adolescent unintended pregnancies and child marriages averted was increased by 37% and 47%, respectively, compared with the full scale-up scenario in the main analysis.

The benefit-cost ratio was increased to 9.7 if the 95% intervention coverage achieved by 2030 were maintained up to 2050 and lifetime benefits considered.

Discussion

This study estimated that if the six identified evidence-based interventions could be scaled up in Cambodia, Indonesia, the Philippines, Lao PDR and Malaysia to reach 95% coverage by 2030, an additional 1.3 million unintended pregnancies and 1.1 million child marriages could potentially be averted among girls aged 15–19 years. This intervention scale-up was estimated to cost US$5.0 billion more than business-as-usual over 2024–2030, but could generate US$13.2 billion in economic benefits by 2050 with an overall benefit-cost ratio of 2.7, increasing to 9.7 if high intervention coverages were maintained up to 2050 and lifetime benefits considered.

When interventions were scaled up in isolation for comparative purposes, the greatest return-on-investment was estimated to come from community dialogue in Cambodia, Indonesia and Malaysia, in-school CSE in Lao PDR and educational support in the Philippines. The evidence suggests that the community dialogue intervention could have a significant impact on reducing marriage rates (RR=0.33)²³ as well as increasing contraception prevalence (RR=1.31)²⁴ among adolescents. It is important to note that this intervention is usually implemented as part of a package, rather than in isolation, and may impact synergistically with other interventions already underway, for example, health system-based interventions, including cost reduction or subsidies and greater on-site availability of contraceptive commodities.²⁵ The intervention with the next best return-on-investment varied by country but was either in-school CSE or educational support for adolescents below the poverty line. Educational support can increase school enrolment (RR=1/0.86)²⁶ as well as reduce marriage rates (RR=0.69),²⁶ whereas in-school CSE can increase contraception use (RR=1.48)²⁷ as well as delay sexual initiation (OR=2.8);²⁸ therefore, the greatest return-on-investment varied depending on the complete set of baseline indicators, including epidemiological and social variables (child marriage rate, adolescent pregnancy rate, school enrolment rate and percentage of adolescents living below the poverty line) as well as the economic variables (GDP per capita and workforce participation rate) for each country. Among the interventions considered, livelihood training had the lowest benefit-cost ratio, as the evidence mapping (online supplemental tables B.1, B.2 and B.3) identified smaller effect size parameters (eg, RR=0.767 on marriages compared with RR=0.33 for community dialogue);^{23 26} however, the benefit-cost ratio was still greater than 1 (ie, returning more than US$1 to the economy for every US$1 invested).

The interventions considered in our study target different subpopulations, such as those below the poverty line or those enrolled in school, which influence the total cost and total benefits. For example, life skills training had the greatest total cost, because it is modelled to be available for all 15–19-year-olds. This targeting of interventions may also influence impact, since in the model, rates of marriage and unintended pregnancies are higher for those below the poverty line and hence the model captures the fact that cash transfers and educational support are directed to those most at risk. However, because not everyone is eligible for every intervention there are still adolescents who may benefit from additional interventions, and as new evidence emerges, more interventions could be included.

Achieving 95% intervention coverage is aspirational and may not be realistic across the five countries with diverse sociocultural contexts. As well as limited financial and human resources, cultural and religious resistance to CSE could result in teachers being hesitant to talk about sexual health with students and a lack of engagement of stakeholders outside of school, such as parents of adolescents.²⁹ Interventions like education support or life skills training would also require trusted and skilful mentors, and adolescent girls who participate in these programmes should be provided with safe places to meet. Implementation barriers of this type could result in interventions not achieving the intended efficacy or only minimal coverage being achieved,³⁰ and therefore our estimated impacts represent an upper bound. However, our sensitivity analyses show that while lower intervention coverage would achieve reduced impacts, the benefit-cost ratio is likely to be similar.

Previous work has estimated the return-on-investment of increased contraception use along with improved maternal health services among all women of reproductive age (ie, the return-on-investment of reaching zero unmet need and scaling up of maternal health interventions), finding benefit-cost ratios range from 15 to 24 in Pacific and Caribbean countries and 8.3 for 120 LMICs aggregated.^{19 22} However, those studies have only considered the supply side of contraception and not how increased contraception use could be achieved in practice. This study extends that work by considering demand generation interventions that could achieve increases in modern contraception use; hence, it includes more realistic costs and has a lower (but still favourable) benefit-cost ratio. It is also not directly comparable, as the scope is focused on just adolescent girls, rather than all women, and hence workforce participation benefits do not begin to accrue immediately.

This study extends previous work, estimating the return-on-investment of interventions to prevent child marriage by considering both adolescent pregnancy and child marriage together, as well as the complex interactions between them. Previous studies include an economic analysis of interventions to reduce child marriage in 31 countries from different world regions, which estimated an unweighted average benefit-cost ratio across countries of 7.4, considering lifetime benefits for scaling up a package of interventions (9.4 for low-income countries and 5.8 for middle-income countries).¹⁶ In our study, the estimated longer-term benefit-cost ratio of 9.7 across five countries was slightly higher; however, it considers a different subset of countries and hence baseline conditions, a different set of interventions, and the inclusion of benefits from adolescent pregnancies averted. Consistent across studies is that investment in evidence-based interventions to reduce child marriage represents good value.

There are some important limitations to this study. First, there may be other interventions that could reduce child marriages and adolescent pregnancies; however, due to limited quantitative evidence of cost and effectiveness, they could not be included in this analysis. Second, effect sizes of interventions came from a review of global literature, and in the model, interventions were largely parametrised from studies conducted in countries outside of those considered in this analysis. While these studies were also conducted in low-income settings, this could still have implications on the benefit-cost ratios. Third, baseline coverage estimates for some interventions were not available and assumed to be zero. This may overstate total cost and impact of scaling up these interventions but would have less influence on the benefit-cost ratios. Where interventions have been partially implemented (eg, CSE, cash transfers—see online supplemental table D1), evaluations of their impact were generally lacking, highlighting an important area for future research. Fourth, unit costs for interventions were based on cost and cost-effectiveness analyses that were conducted in other low-resource settings because country-specific cost information was not available from reliable sources. Local costing studies would be recommended as a next step towards implementation. Fifth, data to inform pregnancy and marriage rates, contraception use and unmet need for family planning were taken from population surveys, which have their own bias. For example, survey participants were found to be less likely to report details of their sexual behaviour (eg, ever having sexual activity) when an interviewer is 10 or more years older than the participant.³¹ These biases may have significant implications when interpreting health indicators such as unmet need for family planning and fertility rates that may also influence the benefit-to-cost ratios. Sixth, the benefits associated with increased education are included in this analysis; however, the cost of providing the education is not included (ie, there would be additional costs borne by the education system due to increased enrolment that are not included in this study). Seventh, 15–19-year-old girls who are not enrolled in school will not be reached by the school-based interventions in this model. Further work could consider the value of early-childhood education strategies as a complementary policy option. Eighth, preventing adolescent unintended pregnancies may have extra benefits of reducing maternal and neonatal morbidities, and there could also be indirect benefits to adolescent boys, such as harm reduction and preventing sexually transmissible diseases. These were not included in the economic benefits; therefore, the benefit-cost ratio might be underestimated. Ninth, our model may underestimate the total benefits of interventions because it only considers effects for 15–19-year-olds, whereas the risk window for unintended adolescent pregnancy and child marriage extends beyond this. For example, in Southeast Asia and the Pacific, birth rates are approximately 2.2 times higher among women aged 20–24^{32 33} years compared with those aged 15–19 years, and while many pregnancies may be intended among women aged 20–24 years, unintended pregnancies may still be substantial. It is plausible that interventions such as in-school CSE may have enduring effects for older age groups, which may reduce unintended pregnancies among 20–24-year-old women; however, these effects are not captured in the model.

Conclusion

This study found that investing in interventions to reduce adolescent pregnancy and child marriage could have substantial health and economic benefits. For every US$1 invested to scale up and maintain high coverage of evidence-based interventions to reduce child marriages and adolescent pregnancies in Cambodia, Indonesia, the Philippines, Lao PDR and Malaysia, US$9.7 could be returned to the economy in the longer term. Economic modelling studies like this one greatly rely on the availability of accurate evidence of programme costs and effectiveness; however, there are significant gaps in the literature for interventions to reduce adolescent pregnancies and child marriage, and additional programme impact evaluation studies would be valuable.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.