Global, regional, and national prevalence and disability-adjusted life-years for infertility in 195 countries and territories, 1990–2017: results from a global burden of disease study, 2017

To provide comprehensive estimates of the global, regional, and national burden of infertility from 1990 to 2017, using findings from a 2017 study on the global burden of disease (GBD), we assessed the burden of infertility in 195 countries and territories from 1990 to 2017. DisMod-MR 2.1 is a Bayesian meta-regression method that estimates non-fatal outcomes using sparse and heterogeneous epidemiological data. Globally, the age-standardized prevalence rate of infertility increased by 0.370% per year for females and 0.291% per year for males from 1990 to 2017. Additionally, age-standardized disability-adjusted life-years (DALYs) of infertility increased by 0.396% per year for females and 0.293% per year for males during the observational period. An increasing trend to these burden estimates was observed throughout the all socio-demographic index (SDI) countries. Interestingly, we found that high SDI countries had the lowest level of prevalence and DALYs in both genders. However, the largest increasing trend was observed in high-SDI countries for females. By contrast, low-SDI countries had the largest increasing trend in males. Negative associations were observed between these burden estimates and the SDI level. The global disease burden of infertility has been increasing throughout the period from 1990 to 2017.


INTRODUCTION
Infertility is the inability to conceive within 1 year of unprotected intercourse, and it has been identified as a public health priority [1]. The Centers for Disease Control and Prevention of the United States emphasizes that infertility is more than a quality-of-life issue, with considerable public health consequences including psychological distress, social stigmatization, economic strain, and marital discord [2,3]. Globally, infertility affects 15% of couples of reproductive age [4,5]. A report from the 2006-2010 National Survey of Family Growth estimated that 6% of married females aged 15-44 years in the United States are infertile, and 12% have impaired fecundity, defined as the inability to conceive and carry a baby to term [6]. By contrast, among couples of reproductive age in China, the prevalence of infertility was 25% [7]. Furthermore, infertility is associated with increased risk of subsequent chronic health conditions such as cardiovascular disease [5].
A woman who is unable to bear a child is classified as having primary infertility. A woman who has previously conceived and successfully given birth yet is unable to do so subsequently is classified as having secondary infertility. Using survey data from 277 demographic and reproductive health surveys a study showed differences in the prevalence of primary and secondary infertility between 1990 and 2010 in 190 countries and territories [8]. Some regions have a high prevalence of primary infertility, but a low prevalence of secondary infertility, such as North Africa and the Middle East, notably Morocco and Yemen. However, some areas have a high prevalence of secondary infertility but a low prevalence of primary infertility, such as Central and Eastern Europe and Central Asia. Additionally, several previous studies provided information regarding the prevalence of infertility according to sex. For example, the reported prevalence of infertility in Britain was 12.5% among females but 10.1% among males [9]. Of note, among these published studies, some focused only on females [10][11][12]. Others exclusively examined males registered at infertility clinics [13,14]. As such, these studies were based on relatively small groups, unrepresentative of the larger population of infertile people [15,16].
Infertility affects both sexes across the globe. On a global scale, accurate information regarding the burden of infertility is sorely lacking. Without accurate national and regional data on infertility, it is impossible to identify and comprehensively treat infertile patients. Therefore, in this systematic analysis, we assessed the global burden of infertility from 1990 to 2017 based on prevalence and disability-adjusted life-years (DALYs), and we assessed its relationship to the level of development, using the socio-demographic index (SDI; a composite indicator of income per person, years of education, and fertility).

Infertility prevalence
Globally, the age-standardized prevalence rate of female infertility increased by 14.962% from 1366.85 per 100,000 (95% UI: 988.34, 1819.86) in 1990 to 1571.35 per 100,000 (95% UI: 1115.30, 2121.94) in 2017, representing a shift of 0.370% per year (95% CI: 0.213, 0.527) (Figure 1). The age-standardized prevalence rate of male infertility increased by 8.224% from 710.19 per 100,000 (95% UI: 586.08, 848.94) in 1990 to 768.59 per 100,000 (95% UI: 623.20, 929.91) in 2017, with an increasing rate of 0.291% per year (95% CI: 0.241, 0.341) ( Figure 2). Among those aged 15-44 years in 2017, the 35-39 age group had the highest prevalence rate, and the 15-19 age group had the lowest (Figures 3 and 4). When stratified by SDI quintiles, we observed an increasing trend in all SDI countries. Of note, although high-SDI countries had the lowest prevalence rate throughout the observational period among both genders (Figures 1 and  2), the high-SDI quintile had the largest increasing trend (annual percentage change (APC) = 0.766%) in females, with a 51.41% contribution rate to the total increasing trend (Supplementary Tables 1 and 2). By contrast, low-SDI countries had the largest increasing trend (APC = 0.385%) in males, with a 33.75% contribution rate to the total increasing trend (Supplementary Tables 1 and 2).
Among females, 14 regions showed an increasing trend among the 21 regions ( Figure 1). The largest APC was observed in Andean Latin America (2.129%), followed by Tropical Latin America (1.504%) and North Africa and the Middle East (1.352%), which contributed 53.78% to the overall increasing trend (Supplementary  Tables 1 and 2). Among males, increasing trends were observed in 16 of the 21 regions ( Figure 2). The largest APC was detected in Andean Latin America (1.558%), followed by Tropical Latin America (0.926%) and Southeast Asia (0.660 %), which contributed 47.39% to the overall increasing trend (Supplementary Tables 1  and 2).
Among females, an increasing trend was observed in 14 of the 21 regions (Figure 7). Similar to prevalence, Andean Latin America (2.200%), Tropical Latin America (1.487%) and North Africa and the Middle East (1.273%) were the top three regions, contributing 54.34% to the overall increasing trend (Supplementary  Tables 4 and 5). Among males, we observed an increasing trend in 16 of the 21 regions (Figure 8). The top three regions were Andean Latin America (1.436%), Tropical Latin America (0.871%), and Central Latin America (0.543%), contributing 46.97% to the overall increasing trend (Supplementary Tables 4 and 5).

Global burden estimates of infertility in relation to SDI levels
We illustrated the associations between global burden estimates of infertility and the SDI levels for each of the 21 global burden of disease (GBD) regions for all individual years between 1990 and 2017 ( Figures 11   and 12). General negative associations were observed between burden estimates and the SDI level. In brief, burden estimates tended to be stable when the SDI was limited to < 0.4. Subsequently, when the SDI was over 0.4, we observed negative associations between burden estimates and the SDI level. For Western Sub-Saharan Africa, we observed a U-shape association between prevalence and DALYs, and the SDI level. Similar patterns were observed in the Eastern and Central Sub-Saharan Africa.

DISCUSSION
To the best of our knowledge, this is the first study to provide a comprehensive assessment of the values and trends of burden estimates of infertility by sex in 195 countries and territories from 1990 to 2017 on the basis of GBD 2017 [17,18]. The burden estimates of male and female infertility, as measured by prevalence and DALYs, increased globally between the observational period, and it increased in all countries regardless of the SDI. Of note, we observed the largest increasing burden estimates in low-SDI countries for males but in high-SDI countries for females. We expect that our findings will be invaluable to health professionals toward their   This study demonstrated that the prevalence of female infertility is relatively higher than that of male infertility. However, limited studies have focused on infertility by gender. Nevertheless, our findings are consistent with these studies [9,19]. Meanwhile, an etiological study that included community-based females and their husbands or male partners and clinically-based patients showed that risk factors accounted for 65.9% of female infertility etiology, whereas this number was a mere 6.8% for male infertility [19]. It can be seen that the potential for infertility in females is greater than it is in males. The reason why the prevalence of female infertility is higher than male infertility might be attributed to two reasons. First, unlike female infertility, male infertility is not     well reported in general, especially in countries where cultural differences and patriarchy prevent accurate statistics from being collected and compiled. Second, a study has shown that tubal factor infertility was the most common cause [19]. Reproductive health is of special importance to females, particularly during their reproductive years. Males also have reproductive health concerns and needs, but their general health is affected by their reproductive health to a lesser extent than in females [20]. Infertility caused by female reproductive health problems is more common. This helps to explain why the prevalence of infertility in females is higher than in males.
Among global infertile females and males aged 15-44 years from 1990 to 2017, the 35-39 age group had the highest prevalence and the 15-19 age group had the lowest. Researchers estimated the cumulative incidence of infertility for 1,037 males and females using a longitudinal birth cohort study in Dunedin, New Zealand. The results showed that the most pronounced incidence of infertility occurred during the mid-to late-30s [21]. In another study, researchers analyzed data from the infertility component of the 2009-2010 Canadian Community Health Survey for married and common-law couples with a female partner aged 18-44. Couples with lower parity (0 or 1 child) had significantly higher odds of being infertile when female partners were aged 35-44 years, compared to those 18-34 years old [22]. Another cross-sectional population survey showed that the ageadjusted odds of experiencing infertility were significantly higher among females who first gave birth at age 35 or older compared with those who did so before the age of 25 [9]. A similar, though slightly weaker, association was observed among males. These studies are very similar to our results. As far as we know, age at marriage can play an important role in causing infertility [23]. Over the past decades, conjugal unions have been delayed, resulting in couples starting to live together or getting married at an older age. This has led to a delay in childbearing, with females being older when first attempting pregnancy. A quantitative cross-sectional survey showed that a longer duration of infertility is associated with a significant decrease in the live-birth rate [24]. Meanwhile, females in their mid-to late-30s are nearing the end of their reproductive spans, when males may be experiencing an age-related decline in fertility. Because patients are older, the disease is more serious and the success rate of treatment is lower. Moreover, younger patients are prioritized for publicly funded infertility treatment in countries such as New Zealand [23, 24]. As such, older patients have less access to treatment.
We found that the largest increasing burden estimates were in low-SDI countries for males and in high-SDI countries for females. This may be attributed to the increasing rate of infertility detection, especially in males with low SDI levels, due to the gradual development of national economies. Of note, high-SDI countries had the lowest prevalence rate for both sex. To the best of our knowledge, disparities in infertility are likely due to differential distributions of factors such as education, socioeconomic status, health behavior, access to quality infertility services, and service-seeking behavior. Studies in Europe, North America, and Australia show that the large majority of research participants who experienced infertility but did not seek medical help. This is of concern, as are the marked inequalities in seeking help among those who are well qualified and employed in high-status jobs compared to those who are not [25][26][27]. A study has shown that the proportion of couples seeking medical care was 56% in developed countries and 51% in developing countries [28]. Although it is not possible to treat all these couples successfully, treatment will lead to a decline in infertility rates in economically developed regions. Thus, we found the lowest prevalence in areas with high-SDI countries. It is quite surprising that Datta et al. found that infertility was most common among females with a post-secondary degree and lowest among those with no academic qualifications, whereas no statistically significant association was observed among males in this regard. A large body of literature describes a trend among females in developed countries of delaying procreation, and it is expected that this changing tempo to fertility is becoming a global phenomenon [29]. Meanwhile, with overall improvements to the economy and changes to lifestyle, the number of overweight (and underweight) individuals is increasing, where obesity is an important factor leading to infertility [30]. Esmaeilzadeh et al. found in their study that infertile females had a 4.8-fold increased risk of obesity and an almost 3.8-fold increased risk of being overweight compared to fertile females [31].
Our investigation has several strengths. First, to the best of our knowledge, this is the first comprehensive overview of the epidemiological situation and trends regarding the female and male infertility burden around the world. Second, the GBD 2017 [17, 18] approach to estimating the prevalence of infertility is novel and can be repeated with relative efficiency. Our findings will be useful to resource allocation and health services planning for the growing number of patients with infertility. However, GBD 2017 [17, 18] methods have several limitations. First, data are absent or extremely sparse for some regions of the world. As such, the models we used to predict prevalence and DALYs might lead to unusual changes in segments of the data. We cannot ignore that the relatively low burden of infertility in developing countries is related to the under-diagnosis of the condition due to limited access to specialized medical care, imaging resources, and laboratory investigations. Until such information becomes available, however, we maintain that the results from our model are valid. Second, the data lacks robust predictive covariates for infertility to aid in population-based risk assessments. GBD is actively seeking access to medical claims data in other countries to improve the accuracy of estimates for diseases such as infertility, for which every patient can be expected to be in contact with the health-care system if there are no major barriers to accessing care. Through our network of collaborators, we expect that future iterations of GBD will be able to add such sources from other countries. Third, there is no relevant data on risk factors of infertility in the GBD database. As such, we cannot compare the magnitude of the risk factors for infertility. Finally, reports on intentional injuries (especially self-harm and legal intervention) are subject to underreporting or even being covered up in many countries. Many of the countries involved in conflicts do not have a reliable health information system even in their preconflict states. We did not evaluate the indirect effects of collective violence (war) on total population. For example, Africa is affected by war, political and economic instability, resulting in population decrease [32,33].
In summary, the burden estimates of infertility increased globally for both genders between 1990 and 2017. This report provides an integrated, contemporary understanding of the global infertility disease burden. Our findings can inform policymakers regarding the health care priority of infertility, and preventive and managerial interventions must be implemented to address the growing burden of infertility in these regions. More studies are needed to investigate the risk factors of infertility in order to carry out efficient preventive and managerial strategies to reduce the burden of this disease.

Data sources
The Global Burden of Diseases, Injuries, and Risk Factors Study, 2017 (GBD 2017) employed a standardized analytical method that used all eligible sources to estimate epidemiological data, including prevalence and DALYs

Modeling
For GBD 2017, the following case definitions were used for infertility: primary infertility was defined as existing in a couple who have not had a live birth, who wanted a child, and had been in a relationship for more than 5 years without using contraceptives. Secondary infertility was defined as existing in a couple who wanted a child and have been in a relationship for more than 5 years without using contraceptives since a previous live birth. Estimation was completed in three steps [17]. First, we estimated the total primary and secondary infertility in couples. This was accomplished by first quantifying the rate of infertility among married survey respondents and then quantifying how this married population related to the overall population. Second, we modeled the proportion of primary and secondary infertility due to female and male factors, respectively, to estimate four "envelopes" of infertility: male primary infertility, male secondary infertility, female primary infertility, and female secondary infertility. Third, we executed a "causal attribution" process to assign cases of each envelope to likely underlying causes and assigned the remainder to idiopathic infertility. Non-fatal modeling, using DisMod-MR 2.1, was performed to estimate the prevalence of infertility [34]. DisMod-MR 2.1 is a Bayesian metaregression method that estimates non-fatal outcomes using sparse and heterogeneous epidemiological data. It also pools data from different sources, adjusts them for variations in study methods across sources, and enforces consistency between different epidemiological parameters. Binary study-level covariates were used to minimize the residual errors of the estimated prevalence and years lived with disability (YLD). Using mixedeffects nonlinear regression on all the available data at the global level, super-region Bayesian priors were generated; likewise, the super-region regression model was then used to generate regional Bayesian priors, and so on down the cascade [34, 35]. YLD were calculated by multiplying the prevalence of each sequela by its disability weight and adding the procedure-related morbidity associated with infertility treatment [34]. Years of life lost (YLL) due to infertility were calculated using normative global life expectancy. DALYs were calculated by summing the YLD and YLL [36].

Socio-demographic Index
The SDI is a summary measure that estimates a location's position on a spectrum of development. The SDI and epidemiological transition SDI is a summary measure that places all GBD locations on a spectrum of socioeconomic development [37]. SDI, expressed on a scale of 0 to 1, is a summary measure that identifies where GBD locations sit on the spectrum of socioeconomic development [37]. The SDI is calculated based on the geometric mean of lag-distributed income, average years of schooling among populations aged 15 years or older, and total fertility rate. More details regarding the calculation of the SDI are provided in previous GBD publications [17,18,38]. All 195 countries and territories were then categorized into five regions in terms of the SDI; low, low-middle, middle, high-middle, and high. The cutoff values used to determine quintiles for analysis were then computed using country-level estimates of SDI for 2017, excluding countries with populations of less than 1 million. These quintiles are used to categorise and present GBD 2017 results on the basis of sociodemographic status. Additional details on and results from the SDI calculation are available in the supplementary file (Supplementary Table 1)

Statistical analysis
We ran DisMod-MR 2.1 models to estimate the proportion of primary and secondary infertility by sex, proportion of primary female infertility, proportion of secondary female infertility, proportion of primary male infertility, and proportion of secondary male infertility. We model sex-specific infertility as a proportion [17]. Prevalence was estimated for nine impairments, defined as sequelae of multiple causes for which better data were available to estimate the overall occurrence than for each underlying cause: Infertility and eight other diseases [17]. We assumed that infertility does not lead to mortality and, therefore, DALYs of infertility are equal to their YLD [34]. So we used the agestandardized prevalence rate and DALYs as well as the annual percentage change (APC) to quantify female and male infertility burden estimated trends [39]. Restricting the age range to 15 to 44 years and divided six 5-year age groups. All measures were age-standardized using the GBD standard population. The age-standardized rates (per 100,000 people) in accordance with a direct method were calculated by summing the products of age-specific rates and the number of individuals in the same age subgroup of the selected reference standard population and subsequently dividing the sum of standard population weights. The APC is a widely used measure of trends in an age-standardized rate over a specific time interval. A regression line was fitted to the natural logarithm of the rates. The APC and 95% confidential interval (CI) values can also be obtained from a linear regression model [40,41]. We employed a generalized additive model with locally estimated scatterplot smoothing to the SDI to estimate the associations between SDI and the age-standardized prevalence rate and DALYs using GBD estimates from all national locations from 1990 to 2017 [42]. All statistical analyses were performed using SPSS (Version 23, SPSS Inc.) and the R program, Version 3.4.4 (ggplot2, readxl, dplyr), with P values <.001 considered significant. R program Version 3.4.4 was used to generate figures of the final estimates of prevalence and DALYs from data available from ghdx. healthdata. org/ gbd-results-tool.

Abbreviations
DALYs: disability-adjusted life-years; SDI: sociodemographic index; GBD: global burden of disease; YLD: years of life lived with disability; YLL: years of life lost; APC: annual percentage change; CI: confidential interval; PC: percentage change.

AUTHORS CONTRIBUTION
Hui Sun and Ting-Ting Gong contributed equally to this work. Hui Sun, Ting-Ting Gong, Yu-Hong Zhao, and Qi-Jun Wu contributed to the study conception and design; Yu-Ting Jiang, Shuang Zhang contributed to acquisition, analysis, or interpretation of data; Hui Sun, Ting-Ting Gong, Yu-Hong Zhao, and Qi-Jun Wu contributed to the manuscript drafting and approval of the final version of the manuscript.

Statement of GATHER compliance
This study complies with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) recommendations. We have documented the steps involved in our analytical procedures and detailed the data sources used in compliance with the GATHER.
The GATHER recommendations may be found here: http://gather-statement.org/

GBD results overview
Results from the Global Burden of Disease Study (GBD) are now measured in terabytes. Results are available in an interactive data downloading tool on the Global Health Data exchange (GHDx). Data and underlying code used for this analysis will be made publicly available pending manuscript acceptance.
The core summary results include years of life lost (YLLs), years lived with disability (YLDs), and disability-adjusted life-years (DALYs). The GHDx includes data for causes, risks, cause-risk attribution, aetiologies, and impairments.

Data input sources overview
GBD 2017 incorporated a large number and wide variety of input sources to estimate mortality, population, fertility, causes of death and illness, and risk factors for 195 countries and territories from 1990-2017. These input sources are accessible through an interactive citation tool available in the GHDx.
Users can retrieve citations for a specific GBD component, cause or risk, and geography by choosing from the available selection boxes. They can then view and access GHDx records for input sources and export a CSV file that includes the GHDx metadata, citations, and information about where the data were used in GBD. Additional metadata for each input source are available through the citation tool, as required by the GATHER statement.

Infertility Outcome estimation
Conceptually, the estimation effort is divided into eight major components: (1) compiling data sources through data identification and extraction; (2) data adjustment; (3) estimation of prevalence by cause and sequelae using DisMod-MR 2.1 or alternative modelling strategies for selected cause groups; (4) estimation by impairment; (5) severity distributions; (6) incorporation of disability weights; (7) comorbidity adjustment; and (8) the estimation of YLDs by sequelae and causes. Other national health surveys were identified based on survey series that had yielded usable data for past rounds of GBD, sources suggested to us by in-country collaborators, and surveys identified in major multinational survey data catalogs, such as the International Household Survey Network and the World Health Organization (WHO) Central Data Catalog, as well as through country Ministry of Health and Central Statistical Office websites. Case notifications reported to the WHO were updated through 2017. Citations for all data sources used for nonfatal estimation in GBD 2017 are provided in searchable form through a web tool (http://ghdx.healthdata.org/).

Survey data preparation
For GBD 2017, survey data for which we have access to the unit record data constitute a substantial part of the underlying data used in the estimation process. During extraction, we concentrate on demographic variables (such as location, sex, age), survey design variables (such as sampling strategy and sampling weights), and the variables used to define the population estimate (such as prevalence or a proportion) and a measure of uncertainty (standard error, confidence interval or sample size and number of cases).

Nonfatal disease registries
For GBD 2017 nonfatal estimation, disease registries were an important source for a select number of conditions such as cancers, end-stage renal disease, and congenital disorders.
Registry data is particularly key in the estimation of neoplasms given the increasing attention to noncommunicable diseases, particularly cancers, in low and middle-income areas of the world. The GHDx source tool (http://ghdx.healthdata.org/data-type/ disease-registry) provides a comprehensive list of registry data used in GBD estimation processes.

Data adjustment
In addition to the corrections applied to claims and hospital data, a number of other adjustments were applied to extracted nonfatal sources in order to make the data more consistent and suitable for modelling. In this second step of nonfatal estimation, commonly applied adjustments included age-sex splitting, bias correction, adjustments for underreporting of notification data, and computing expected values of excess mortality. Age-sex splitting was commonly applied to literature data reported by age or sex but not by age and sex. For GBD 2017, we split all data reported in age groups with a width greater than 20 years, using age patterns from available survey microdata or regional patterns derived from an initial run of main modelling tool, DisMod-MR 2.1. We relied on the meta-regression component of DisMod-MR 2.1 for most of the bias correction of data for variations in study attributes such as case definitions and measurement method. DisMod-MR 2.1 calculates a single adjustment that is applied regardless of age, sex, or location. If enough data were available to differentiate these adjustments by age, sex, or location, or if detailed survey data were available to make more precise adjustments between different thresholds on a biochemical measure, we applied bias corrections to the data before entry into DisMod-MR 2.1. For instance, we crosswalked between 12 different case definitions with different thresholds of fasting plasma glucose or glycated hemoglobin levels for diabetes mellitus based on available survey data with individual records of the actual measurements. In another example, we corrected data on COPD from surveys applying different thresholds on spirometry measurements using studies that had reported on prevalence of COPD for the reference and alternative thresholds. As this relationship varied with age, age-specific correction factors were derived. The correction of notification data for underreporting relied on studies that had examined the gap between true incidence and notified cases.

IMPAIRMENT AND UNDERLYING CAUSE ESTIMATION
Impairments in GBD are conditions or specific domains of functional health loss which are spread across many GBD causes as sequelae and for which there are better data to estimate the occurrence of the overall impairment than for each sequela based on the underlying cause. Overall impairment prevalence was estimated using DisMod-MR 2.1. We constrained cause-specific estimates of impairments, as in the 19 causes of blindness, to sum to the total prevalence estimated for that impairment. Estimates were made separately for primary infertility (those unable to conceive), secondary infertility (those having trouble conceiving again), and whether the impairment affected men and/or women.

Disability weights
To compute YLDs for a particular health outcome in a given population, the number of people living with that outcome is multiplied by a disability weight that represents the magnitude of health loss associated with the outcome. Disability weights are measured on a scale from 0 to 1, with 0 implying a state that is equivalent to full health and 1 a state equivalent to death. Disability weights used in GBD studies prior to GBD 2010 have been criticized for the method used (person trade-off), the small elite panel of international public health experts who determined the weights and the lack of consistency over time as the GBD cause list expanded and additional disability weights from a study in the Netherlands24 were added or others derived by ad-hoc methods.

YLD computation, uncertainty, and residual YLDs
For GBD 2017, we computed YLDs by sequela as prevalence multiplied by the disability weight for the health state associated with that sequela. The uncertainty ranges reported around YLDs incorporates uncertainty in prevalence and uncertainty in the disability weight. To do this, we take the 1,000 samples of comorbidity-corrected YLDs and 1,000 samples of the disability weight to generate 1,000 samples of the YLD distribution. We assume no correlation in the uncertainty in prevalence and disability weights. The 95% uncertainty interval is reported as the 25th and 975th values of the distribution. Uncertainty intervals for YLDs at different points in time (1990,1995,2000,2005,2010, and 2016) for a given disease or sequela are correlated because of the shared uncertainty in the disability weight. For this reason, changes in YLDs over time can be significant even if the uncertainty intervals of the two estimates of YLDs largely overlap as significance is determined by the uncertainty around the prevalence estimates.

Socio-demographic Index (SDI) analysis and epidemiological transition
The Socio-demographic Index (SDI) is a composite indicator of development status strongly correlated with health outcomes. In short, it is the geometric mean of 0 to 1 indices of total fertility rate under the age of 25 (TFU25), mean education for those aged 15 and older (EDU15+), and lag distributed income (LDI) per capita.

Development of revised SDI indicator
SDI was originally constructed for GBD 2015 using the Human Development Index (HDI) methodology, wherein a 0 to 1 index value was determined for each of the original three covariate inputs (total fertility rate in ages 15 to 49, EDU15+, and LDI per capita) using the observed minima and maxima over the estimation period to set the scales.
In response to feedback from collaborators and the evolution of the GBD, we have refined the indicator with each GBD cycle. For GBD 2017, in conjunction with our expanded estimation of age-specific fertility, we chose to replace the total fertility rate as one of the three component indices with the total fertility rate under 25 (TFU25). The TFU25 provides a better measure of womens status in society, as it focuses on ages where childbearing disrupts the pursuit of education and entrance into the workforce.
During GBD 2016 we moved from using relative index scales to absolute scales to enhance the stability of SDIs interpretation over time, as we noticed that the measure was highly sensitive to the addition of subnational units that tended to stretch the empirical minima and maxima. We selected the minima and maxima of the scales by examining the relationships each of the inputs had with life expectancy at birth and under-5 mortality and identifying points of limiting returns at both high and low values, if they occurred prior to theoretical limits (e.g., a TFU25 of 0).
Thus, an index score of 0 represents the minimum level of each covariate input past which selected health outcomes can get no worse, while an index score of 1 represents the maximum level of each covariate input past which selected health outcomes cease to improve. As a composite, a location with an SDI of 0 would have a theoretical minimum level of development relevant to these health outcomes, while a location with an SDI of 1 would have a theoretical maximum level of development relevant to these health outcomes.
The composite Socio-Demographic Index is the geometric mean of these three indices for a given location year. The cutoff values used to determine quintiles for analysis were then computed using country-level estimates of SDI for the year 2017.

INFERTILITY CASE DEFINITION AND MODELLING SUMMARY
For GBD 2017, the following case definitions were used for infertility: 1. Primary infertility is defined as a couple who have not had a livebirth, who wish a child, and have been in a union for more than five years without using contraceptives.
2. Secondary infertility is defined in a couple who wish a child and have been in a union for more than five years without using contraceptives since the last livebirth.
Estimation is completed in three steps. First, we estimate total primary (unable to have any child) and secondary (unable to have an additional child) infertility in couples. This is accomplished by first quantifying the rate of infertility among survey respondents who are married (the subset to whom such questions are directed) and then quantifying how the married population relates to the overall population. Second, we model which proportion of primary and secondary infertility is due to female and male factor, respectively, to estimate four "envelopes" of infertility: male primary infertility, male secondary infertility, female primary infertility, and female secondary infertility. Third, we execute a "causal arrtibution" process to assign cases of each envelope to likely underlying causes and assingn the remainder to idiopathic infertility (ie, unknown causes).

Input data
Our primary data sources are population surveys. The desire to have a child is the crucial determinant of whether a couple is labeled as infertile (ie, if no child is wanted, infertility is not present).
The combination of variables in surveys that were used to construct each of the four datasets (primary "impairment" and "exposure" and secondary "impairment" and "exposure")are illustrated in the table below. As described below, overall primary and secondary infertility are estimated by multiplying prevalence among those with the "impairment" of infertility (married women who desire a[nother] child) by the prevalence of the "exposure" (being married for 5+ years, not using contraception for 5+ years). The majority of excluded studies were excluded because of the latter criterion. In total, 15 studies were included in our analysis for the sex breakdown among infertile couples. Infertility among couples was reported as due to one of the following causes: male factor, female factor, both, or unknown. Couples with infertility due to both partners were allocated to both male factor and female factor, and couples with infertility of unknown cause were allocated to male and female factors based on the proportion observed in other couples in the study. We estimated the proportion of couples' infertility due to male factors and female factors separately in DisMod-MR 2.1. The quantity modelled was the proportion of couples' infertility due to each sex for each of primary and secondary infertility. The table below shows the dataset contents for these four models, each of which used the same set of sources.

Modelling strategy
For GBD 2017, we estimated the prevalence of primary and secondary infertility by sex and cause in three steps: 1) estimation of couples infertility [four DisMod-MR 2.1 models], 2) estimation of infertility by sex [four AGING DisMod-MR 2.1 models], and 3) causal attribution of infertility. We assumed zero infertility prior to age 15 or after age 50 years as fertility is not expected to be desired outside these age ranges in women; an assumption that was therefore carried over to men as well. All DisMod-MR 2.1 models were run as single parameter models. No study or country covariates were used in any models.

Estimation of couples' infertility
To estimate the prevalence of primary and secondary infertility among couples, we first run four DisModMR 2.1 models to estimate the four parameters detailed above, prevalence of primary infertility (1), prevalence of primary infertility exposure (2), prevalence of secondary infertility (3), and prevalence of secondary infertility exposure (4). For prevalence of infertility (models 1 and 3), we tried using the natural log of the age-standardised death rate (lnASDR) of sexually transmitted infections (STIs), but it was not statistically significant so we did not use it in the final model. We did not use any study-or country-level covariates for these models. Next, we estimated primary and secondary couples' infertility form DisMod-covariates for these models. Next, we estimated primary and secondary couples' infertility form DisMod-MR 2.1 models by multiplying the estimates for prevalence of infertility among exposed women by the prevalence of exposure to infertility to obtain prevalence of infertility among all women and all men.

Estimation of infertility by sex
After running the four models estimating overall infertility, described above, we ran four DisMod-MR 2.1 models to estimate the proportion of primary and secondary infertility by sex, proportion of primary female infertility, proportion of secondary female infertility, proportion of primary male infertility, and proportion of secondary male infertility. We model sexspecific infertility as a proportion. Because infertility in some couples is attributable to both partners rather than just one, the sum of the proportions due to each partner is greater than one when both partners are infertile. When the sum of the proportions is lower than one, we scale it to be equal to one through custom code. Again, we tried using lnASDR of STIs as a covariate, but it was not statistically significant so we did not use it in the final model. We did not use any study-or countrylevel covariates for these models. We multiplied our prevalence of primary and secondary infertility derived in step 1 by the proportion due to male and female factors to estimate primary and secondary infertility by sex.

Causal attribution
There are seven identified causes of female infertility in the GBD 2017 cause list: pelvic inflammatory disease (PID) due to chlamydia, PID due to gonorrhoea, PID due to other sexually transmitted diseases, maternal sepsis, polycystic ovarian syndrome, endometriosis, and Turner syndrome. For each of these diseases, we determined the prevalence of infertility by a literature review of the probability of becoming infertile due to that disease. For STIs, we applied a proportion with infertility derived from Westrom and colleagues1 to incident cases of PID and used DisMod-MR 2.1 to calculate corresponding prevalence for each subsequent age group through the fertile years, assuming zero remission or excess mortality. For the others, we added all the diseasespecific estimates of prevalence and assigned the remaining proportion to categories of "female primary infertility due to other causes" and "female secondary infertility due to other causes." We assumed all infertility form Turner syndrome is primary infertility and all infertility following maternal sepsis is secondary infertility. The only recognized cause of male infertility in the GBD 2018 cause list is Klinefelter syndrome. We assigned all other male infertility to "male" infertility due to other causes.

Sequelae/disability weights
Every person with infertility was assumed to experience the health state as determined from the GBD disability weights survey. The lay descriptions of primary and secondary are listed below.

Computing DALYs
To estimate DALYs for GBD 2017, we started by estimating cause-specific mortality and non-fatal health loss. For each year for which YLDs have been estimated (1990,1995,2000,2007,2010