Survey data for key family planning indicators.

Survey-based estimates of contraceptive prevalence by method (any, modern and traditional) and of the unmet need for family planning among adolescent women were compiled in World Contraceptive Use 2019 [10], a comprehensive collection of family planning indicators, by marital status and by five-year age group, for 195 countries or areas. The observations come mainly from nationally representative household-based surveys and were obtained from the associated micro datasets or derived from tabulations published in reports or specific tabulations provided by the institutions responsible for data collection. For girls and young women aged 15–19 who are married or in a union, the data set includes 754 observations of contraceptive prevalence from 160 countries and 354 observations of unmet need from 101 countries. For adolescent women aged 15–19 who were not married and not in a union, 432 observations from 114 countries on contraceptive prevalence, and 245 observations from 74 countries on unmet need were available for the most recent compilation. The input data set is presented as supplementary information S2 Data. Data for women younger than 15 years of age were not included in this analysis. Data for this age range are rarely collected in surveys or reported separately: the data compilation of World Contraceptive Use 2019 includes only nine data points from nationally-representative surveys for age groups 10–14 or 12–14.

There are fewer observations for unmarried women, usually because (1) survey questions on current contraceptive use often exclude unmarried women, especially for older surveys, (2) published tables in reports rarely present results for unmarried women, and (3) microdata needed for generating data disaggregated by age and marital status are not available for some surveys [22].

The major sources of family planning data for adolescents used in this study are the Demographic and Health Surveys (DHS) and the Multiple Indicator Cluster Surveys (MICS). Other international survey programmes providing at least 25 observations are the Reproductive and Health Surveys, the World Fertility Survey, and the Performance Monitoring and Accountability Surveys. The proportions of observations obtained from these sources are given in Table 2. One consequence of relying on data from these survey programmes is that geographic coverage is best in their focus regions, namely sub-Saharan Africa, Central and Southern Asia, Western Asia and Northern Africa, Eastern and South-Eastern Asia, and Latin America and the Caribbean (data availability by regions are presented as supplementary information, Figure A in S1 Supplementary Figures in S1 File). There are only a few observations of the unmet need for family planning among adolescents in Northern America and Europe, Oceania, and Eastern and South-Eastern Asia.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Distribution of observations by survey type, for married and unmarried adolescent women aged 15–19 (in per cent).

https://doi.org/10.1371/journal.pone.0247479.t002

The survey data on contraceptive use among adolescents cover the period 1969 to 2018, with the majority of surveys (84% for married, 94% for unmarried) conducted after 1990. Furthermore, 57% of countries had no data from before 1990 for married adolescents, and 80% had no data before 1990 for unmarried adolescents. At the regional level, 47 countries in sub-Saharan Africa had at least one observation of contraceptive prevalence for married adolescents; however, before 1990 70% of these countries had no observations and a further 21% had only one. For unmarried adolescents, 42 sub-Saharan African countries had at least one observation; of these, 76% had no observations before 1990. Countries in Latin American and the Caribbean had more early observations, 82% of countries in this region had a least one observation before 1990 for married women, and 42% for unmarried women. Additionally, the number of survey observations varies by country. Of the 160 countries with at least one observation for married women, 41% had five or more observations, and 30% of the 114 countries with at least one observation for unmarried women had five or more observations.

Estimates and projections of the proportion and number of adolescents aged 15–19 years who are married or in a union.

The estimated population of adolescent women by marital status and country or area were used to transform model-based estimates of proportions into estimates of counts of women using family planning. These population counts were derived from the 2018 revision of Estimates and Projections of Women of Reproductive Age Who Are Married or in a Union [23], including estimates and projections of the proportion of women who are married or in a union disaggregated by five-year age group for 201 countries or areas over the period 1970 to 2030. Estimates are based on selected census- or survey-based observations of the proportion of female population by marital status and age from World Marriage Data 2017 [24].

Changes in the proportions of adolescents who are married or in a union influence the proportions of contraceptive users and those with unmet need for family planning among all adolescents since contraceptive use and needs vary considerably by marital status. The proportions of adolescents aged 15–19 that are married or in a union in the regions are given in Table 3 and Fig 1. The proportions married have declined in all regions and are projected to continue declining to 2030 [23]. Worldwide, 18.5% of women in this age group were married or in a union in 1990, but this was estimated to be 12.3% in 2019 and is projected to be 9.3% in 2030. Central Asia and Southern Asia, Sub-Saharan Africa, and Western Asia and Northern Africa had the greatest proportions married in 1990 (37.5%, 32.1% and 17.9%, respectively) and are projected to continue fast declines through 2030. Conversely, Eastern and South-Eastern Asia, Oceania, and Northern America and Europe had low, and still declining, proportions married or in a union. Latin America and the Caribbean has seen only a little change in the proportion married among adolescents aged 15–19, from 15.3% in 1990 to 14.4% in 2019; this proportion is projected to decline only moderately to 11.7% in 2030, which will be the highest level for any region other than sub-Saharan Africa.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Proportion of women aged 15–19 years who are married or in a union, for the world and SDG level 1 regions, 1990–2030.

https://doi.org/10.1371/journal.pone.0247479.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Proportion of women aged 15–19 years old who are married or in a union, globally and by region, 1990–2030.

https://doi.org/10.1371/journal.pone.0247479.t003

Some trends in proportions married by regions were due primarily to the dynamics in subregions. In particular, the decline in Central and Southern Asia was driven primarily by the trend in Southern Asia. Results by subregion, development group and income group are presented as S2 Data.

Modelling trends in contraceptive use and unmet need for family planning.

A Bayesian hierarchical model was used to estimate and project contraceptive prevalence and the unmet need for family planning among women aged 15–19 years. The estimates for women who are married or in a union were derived separately from those for women who are not married or in a union. We used the same model for estimating and projecting family planning indicators developed in earlier research for all women of reproductive age (15–49 years) [19–21], except we applied it to data only on women aged 15–19 years. We refer the reader to the supplementary material of [21] for the full model specification; we give a conceptual overview below.

For each country, the model generates estimates of the proportion of adolescents aged 15–19 belonging to each of the following four categories: i) using a traditional contraceptive method, ii) using a modern contraceptive method, iii) having an unmet need for family planning, and iv) having no need for family planning. The shares across the four groups must sum to one. In the modelling process [19], an equivalent set of quantities is modelled: i) the proportion using a contraceptive method of any kind, ii) among those using any method of contraception, the proportion using any modern method, iii) among those not using any method of contraception, the proportion who have an unmet need. From any three of these indicators, the fourth can be calculated, bearing in mind that the four indicators must sum to one.

The time trends for the proportion of women using contraception (any method) and the proportion using modern contraceptive methods were modelled using logistic curves. These are “S”-shaped curves that capture the entire transition from a state of very low contraceptive prevalence, through a period of growth, to an eventual maximum level. A separate pair of curves was estimated for each country. Thus, for example, the rate of growth in prevalence during the transition period, the timing, and the long-run maximum level could, and did, vary by country. Moreover, the model did not assume, nor impose the condition, that all countries move through the entire curve during the period for which estimates and projections were prepared. Many countries, for example, remained in the state of low prevalence across the whole time period of interest, including to the end of the projection. In these cases, only a portion of the logistic curve was used to model the components. The extent of the curve used was determined by the country-level data through the statistical model.

The proportion experiencing unmet need for family planning, the third of the modelled parameters, was not modelled as a logistic curve over time but as a function of the total contraceptive prevalence. This was based on theoretical reasoning about the relationship between these two indicators. The reasoning was that, as prevalence increases, there will be a concomitant increase in unmet need due to an increased need for, and awareness of, contraception as it becomes more common in the population. At a certain level of contraceptive prevalence, unmet need reaches a turning point and begins to decline as availability, accessibility, and acceptability of contraception improves and use increases further. The model used encapsulates this relationship.

Accounting for bias due to sampling of non-baseline groups and misclassifications.

Some surveys targeted groups outside the baseline national populations of women aged 15–19. Examples include the Gender and Generations Surveys (GGSs) that surveyed only women aged 18 and over. Several other surveys excluded some populations, for example non-nationals or people living in particular provinces or areas. Additional sources of bias were due to misclassification of women as users of modern or traditional methods. These sources of bias were accounted through the use of additional parameters, multipliers and misclassification parameters, that adjusted the estimated proportions to counter the sources of bias. These were estimated from the data through the statistical model in the same way as all the other model parameters. A detailed explanation of these parameters is given in previous work [21]. The input data file, including biases and misclassifications, is presented as supplementary information S3 Data.

Hierarchical structures of models for married and unmarried women.

Many countries had few data points and few countries had no data available. To improve the precision and accuracy of estimations and projections, hierarchical models were used [25]. These models allowed for ‘borrowing of strength’ across data points such that estimates for countries with little-to-no data were based partly on data for other ‘similar’ countries. What constituted ‘similar’ was defined through the hierarchical structure of the model. For married women, a geographic hierarchy wherein countries were clustered into sub-regions (e.g., Eastern Africa), and sub-regions into regions (e.g., Africa) was used [19]. All countries belonged to one, and only one, sub-region and region. All regions together constituted the world. The clustering was based on geographic location, regardless of the availability of data. For further details on how hierarchies are used in the model, see S1 Appendix.

For unmarried adolescent women, need for family planning is closely related to level of sexual activity and sexual activity among unmarried adolescent women varies considerably between countries [26]. To account for this association, a different hierarchical structure was used for unmarried women. Countries were first clustered into geographic sub-regions similar to those used in the model for married women. The sub-regions were then clustered into one of two sexual activity groups [20]. Group 0 consisted of countries where sexual activity among unmarried women was estimated to be very low, defined as the proportion of unmarried women of reproductive age who report sexual activity in past 28 days was not more than 2%. Group 1 consisted of all other countries. To estimate the extent of sexual activity among unmarried women was estimated from the question on recent sexual activity in DHS, MICS and other available surveys. In countries without this data, we used information about the acceptance of sex between unmarried adults from other surveys and the level of religiosity. For further details on the information used for creating the clusters, see S1 Appendix.

Projections to 2030.

Estimates for years outside the periods of data availability were obtained from the fitted Bayesian model. The parameters of the systematic trend component are time-invariant [19, Technical Appendix] and thus provide estimates of the trend at all years. The auto-correlated error process is also parameterized by time-invariant parameters. Once these were obtained from the model fit, extrapolations beyond the period of data availability were obtained by sequentially sampling from the error process conditional distributions.

Parameter estimation and software.

A Markov chain Monte Carlo (MCMC) algorithm [25,27] was used to draw a large sample from the posterior joint probability distribution of the parameters described above (i.e., the proportion of women using any method of contraception, the proportion of users using modern methods, the proportion of non-users experiencing unmet need), by country, by year. These were transformed into country-level estimates of the indicators of interest, such as contraceptive prevalence of modern methods or unmet need for family planning.

The MCMC samples were used to generate summary statistics for any parameter of interest. For example, the posterior median contraceptive prevalence, which was used as the measure of central tendency, was obtained by taking the empirical median of the marginal MCMC sample for that parameter. We followed Kantorová et al. [22, S1 Appendix, Subsection 3.7.6] and used small adjustments to medians only to ensure that basic identities held, such as that the proportion using any method equal the sum of the proportions using modern and traditional methods.

Out-of-sample validation exercises were performed to assesses model fit; these are reported in S1 Appendix. The exercises indicated that the model fitted the data well and had good out-of-sample predictive validity.

All computation was done using the R Environment for Statistical Computing, version 3.5.1 (R Core Team, 2018) and JAGS version 4.2.0 [28,29]. The source code is available at https://github.com/FPcounts/FPEMglobal.

Indicators reported.

Estimates and projections of family planning indicators among adolescent women aged 15–19 are presented for: contraceptive prevalence (any method, modern methods), the unmet need for family planning, the proportion of need for family planning satisfied by modern contraceptive methods (SDG indicator 3.7.1), and total need for family planning. These are given for the groupings by regions and subregions (the SDG country groupings at levels 1 and 2, supplementary information S1 and S2 Results), income levels and development groups (supplementary information S3 and S4 Results) as the proportions and the counts. For the proportion of need for family planning satisfied by modern contraceptive methods (SDG indicator 3.7.1.), the results for the regions of “Western Asia and Northern Africa” and “Central Asia and Southern Asia” are not presented for unmarried adolescent women separately due to small overall counts of women with need for family planning; however, they are included in the all adolescent estimates.

The results are presented for individual countries in supplementary information S2 Appendix. Estimates and projections for countries with fewer than two survey-based observations of the proportion of married adolescents aged 15–19 using any method of contraception are not presented individually because they are subject to substantial uncertainty. However, they are included in the aggregates.

The 95% uncertainty intervals (UI) are computed for all indicators using the empirical 2.5th and 97.5th percentiles of the marginal posterior distributions and reported in tables and figures. Empirical medians (50th percentiles) of the marginal posterior distributions were used as single-number summaries of central tendency.

As an example, modelled estimates for Colombia are shown with the underlying source data in Fig 2. The modelled estimates do not necessarily pass through all data points because they account for systematic bias and uncertainty due to measurement error.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Estimates and projections of contraceptive prevalence (any method) (left panels) and unmet need for family planning (right panels), and source data, among married (top panels) and unmarried (bottom panels) women aged 15–19 years in Colombia, 1970–2030.

Note: Estimates and projections are indicated by the solid line (posterior medians), dashed lines (lower and upper limits of 80% uncertainty/prediction interval), and grey ribbon (extent of 95% uncertainty/prediction interval). The source data are indicated by the red (DHS), green (other international surveys) and orange (national surveys) circles.

https://doi.org/10.1371/journal.pone.0247479.g002

In many countries, the need for family planning and contraceptive use are substantially different among married and unmarried adolescents. Therefore, we present the results by marital status, as well as for all adolescents.

Changes over time in the proportion of adolescents who are married or in a union can have a strong effect on overall need for family planning and contraceptive use. Estimates for all women were derived for each indicator by weighting and summing the respective estimates for married and unmarried women. The proportions of adolescent women who were married or unmarried were used as weights. Summing was done at the level of the MCMC samples by country. This yielded a joint posterior distribution for all women which was summarized using quantiles in the same way as described above for married and unmarried women.

Estimates for geographical aggregates were derived by summing MCMC samples at the country level to produce joint posterior distributions.

For each indicator, estimates of the proportion of all women of reproductive age that are adolescents were calculated by taking the ratio of the number of adolescent women in each indicator category to the number of all women of reproductive age [11].

Joint visualization of three family planning indicators by country.

In addition to reporting trends, visualizations are used to display family planning indicators jointly on maps. This was possible because the three indicators form a complete partition of the population, with each woman being counted under one, and only one, of the indicators. For this reason, at the population level the indicator values sum to 100%. In general, indicators with this property are called compositions [30], as illustrated in Fig 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Family planning indicators as categories of a composition (top panel) and illustration of family planning compositions plotted on a ternary diagram (bottom panel).

Note: The bars represent compositions; vectors of components that sum to 100%. The points inside the triangle plot the values for specific hypothetical countries. The right edge is the axis for contraceptive prevalence, the left edge for unmet need, and the bottom edge is the axis for no need. All axes range from 0 to 100%. The arrows adjacent to the triangle’s edges indicate the direction of increasing values; they point to the vertex at which that component is 100%. The open circle marks the centre of the triangle; at this point all components are equal at 33¹/₃%.

https://doi.org/10.1371/journal.pone.0247479.g003

Compositional indicators were formed from the posterior medians for each marital group, for the year 2019. These were displayed as choropleth world maps with ternary plots and ternary-balance colour schemes [31]. These visualizations show the distribution of countries among the three components in true multivariate fashion and illustrate the relationships among the three component indicators simultaneously across countries.

The ternary balance colour scheme assigns a basic colour to each of the three components and represents compositions by mixtures of these colours in the same proportions. This gives each composition a single, unique colour. We assigned green to contraceptive prevalence, pink to unmet need, and blue to no need. In the extreme, countries with a proportion of no need close to 100% would be coloured almost purely blue. Countries with proportions more evenly allocated would be coloured an appropriate blend of the base colours, where the proportion of each base colour is equal to the proportion of each component. For example, a country with the proportions of women using contraception of 65%, having an unmet need for family planning of 25%, and having a no need for family planning of 10%, would be indicated on the map by the colour obtained by blending green, pink, and blue in relative proportions of 65%, 25%, and 10%.

The legend for the ternary balance choropleth maps is the ternary diagram overlaying the ternary colour scheme. A ternary diagram is a triangular plotting space, resulting in a legend that is a triangle with a colour contour inside it. Each edge of the triangle is an axis for one of the components. The interior regions of the triangle near the vertices are made up entirely of the base colours corresponding to the component that is at its maximum at that vertex. In the deep interior of the triangle, the colours become a blend as described above.

The posterior median estimates for each country are plotted as points inside this triangular legend just as they are in the rectangular plotting space of a standard scatter plot. Points close to the vertices represent countries where most women fall into the component associated with that vertex. Points near the centre have proportions for each component close to 1/3. Values of the three components for any point are found by reading back to each edge in the directions indicated by the orientations of the axis labels, as illustrated in Fig 3.

Contraceptive use (modern methods).

Globally in 2019, the estimated proportion of adolescent women (ages 15–19 years) who were using modern methods of contraception was 9.1% (UI = 7.4–12.8) (Table 4). Majority of adolescent users report using a modern method, 89.2% (95%UI = 83.0–91.9) in 2019 increasing from 68.5% (95%UI = 54.7–77.5) in 1990. The proportion of modern contraceptive use among all use among adolescents was highest in Northern America and Europe, at 94.6% (95%UI = 83.0–91.9) and it was lowest in Western Asia and Northern Africa, at 71.0% (95%UI = 59.5–79.8).

Unmet need for family planning.

Worldwide, 5.1% (95%UI = 4.2–10.0) of adolescent women aged 15–19 years are estimated to have an unmet need for family planning, with the highest proportions of unmet need among adolescents in sub-Saharan Africa at 10.2% (95%UI = 9.2–11.5) (Table 4 and Fig 4).

The proportion of married adolescents with unmet need for family planning worldwide in 2019 was estimated to be 20.7% (95%UI = 18.7–23.0) and among unmarried adolescents 3.0% (95%UI = 1.9–8.5). In view of the wide prediction intervals, the projections to 2030 are inconclusive concerning likely future changes in the proportion of adolescents with an unmet need for family planning at the global level: the projected level of this indicator in 2030 is 19.5% (95%UI = 16.6, 22.8) for married adolescent women and 3.3% (95%UI = 2.1–8.7) for unmarried adolescent women.

The proportion of adolescents with an unmet need for family planning in 2019 among married adolescent women was highest in sub-Saharan Africa at 23.7% (95%UI = 21.4–26.4). Among unmarried adolescent women in the same year, the regions with the greatest unmet need for family planning were sub-Saharan Africa at 6.7% (95%UI = 5.8–8.0), and Latin America and the Caribbean at 5.3% (95%UI = 2.5–11.6).