Few comparative analyses of global infertility have been conducted, and none, to our knowledge, have applied a consistent algorithm to demographic and reproductive health survey data from both developing and developed countries, nor used these data to estimate regional and global trends in infertility prevalence. Boivin et al. estimated global infertility by summarizing prevalence data from seven studies: five from developed countries and two from developing countries [5]. A Demographic and Health Surveys (DHS) report also estimated infertility for developing countries using survey data from 47 national DHS surveys [6]. The report's estimate of infertility and analysis of trends did not apply to developed countries, nor to China. Ericksen and Brunette [7] and Larsen [8] applied consistent definitions of infertility in their analyses of household survey data, but considered only Sub-Saharan African countries.

The main challenges in generating global estimates of infertility are the scarcity of population-based studies and the inconsistent definitions used in the few high-quality studies available [9],[10]. In population-based studies of infertility, there has been little consistency in how prevalence is calculated [9],[11]. An explicit detailing of the numerator and denominator of each definition is needed to make clear what is being measured. The authors of a recent literature review concluded that it is not possible to synthesize infertility prevalence data in the published literature because of the incomparable definitions used [9].

An alternative to synthesizing data found in the literature is to apply a consistent definition to regularly collected demographic and reproductive health survey data. In this paper, we used a consistent algorithm to measure infertility using household survey data. Our measure is a demographic definition that uses live birth as the outcome and a 5-y exposure period based on union status, use of contraceptives, and desire for a child [6]–[8],[12]. There are challenges associated with inferring prevalence from household survey data. Few household surveys ask how long the respondent has tried to get pregnant, and none include a comprehensive medical history and clinical examination. Instead, these surveys may collect information on births, couple status, fertility preferences, and contraceptive use. In a previous analysis we performed sensitivity analyses around each of these components to identify important biases that may arise when information is incomplete [13]. We found that a 5-y exposure period is needed to accommodate the time it takes to become pregnant and give birth, and helps prevent unreported temporary separations, periods of postpartum sexual abstinence, or lactational amenorrhea from unduly affecting the infertility measure. Births, rather than pregnancies, are the preferred outcome, as information on live births is collected more often and reported more accurately: neither pregnancies in the first trimester nor voluntary terminations are reliably reported in household surveys [14]–[16]. Lastly, we argued previously that the intent to have a child serves as a proxy for regular, unprotected sexual intercourse, and may correct for underreporting of contraceptive use [13],[17].

Clinical and epidemiologic infertility definitions are also used to monitor infertility; however, they are not appropriate when making population-based estimates of infertility using household surveys. The clinical definition of infertility used by the World Health Organization (WHO) is “a disease of the reproductive system defined by the failure to achieve a clinical pregnancy after 12 months or more of regular unprotected sexual intercourse” [18], while the WHO's epidemiologic definition is “women of reproductive age at risk of becoming pregnant who report unsuccessfully trying for a pregnancy for more than two years” [19]. Clinical definitions are designed for early detection and treatment of infertility [18]–[20]. A definition and assessment of infertility based on medical histories and diagnostic tests is appropriate for clinical settings, where the aim is to understand causes and provide treatment as soon as it is indicated. However, measuring patterns and trends in infertility at the population level necessitates a measure that may be elicited using a standard set of survey questions [17]. The WHO's epidemiologic definition is more closely aligned with clinical practice than demographic definitions are, and may be measured using survey data. However, few household surveys determine whether a couple is trying to become pregnant, and the majority do not collect information on past pregnancies, only on previous live births.

Predictive validity statistics are shown in Table C in Text S1. Root mean square prediction errors for countries for which data were left out were 1.3% for primary prevalence, 6.1% for secondary prevalence, and 9.1%–13.1% for exposure to primary and secondary fertility (see Figures D–I in Text S1 for graphical presentation of model fit). The models' 95% uncertainty intervals contained 93%–96% of left-out data points.

In 2010, 1.9% of child-seeking women aged 20–44 y were unable to have a first live birth (primary infertility; 95% uncertainty interval 1.7%, 2.2%), and 10.5% of child-seeking women with a prior live birth were unable to have an additional live birth (secondary infertility; 9.5%, 11.7%). Levels of infertility were similar in 1990 and 2010, decreasing 0.1 (−0.1, 0.3) percentage points for primary infertility (from 2.0% [1.9%, 2.2%] in 1990; pp=0.84) and increasing 0.4 (−0.8, 1.6) percentage points for secondary infertility (from 10.2% [9.3%, 11.1%] in 1990; pp of increase=0.71).

Figure 1 presents the prevalence of primary and secondary infertility by age (see Figure J in Text S1 for age pattern of exposure to primary and secondary infertility). The prevalence of primary infertility was higher among women aged 20–24 y (2.7% [2.4%, 3.0%] in 2010) compared to women aged 25–29 y (2.0% [1.8%, 2.2%]) and women aged 30–44 y (ranging from 1.6% to 1.7% in 2010). Prevalence of secondary infertility increased sharply with age, from 2.6% (2.3%, 3.0%) in women aged 20–24 y to 27.1% (24.7%, 29.9%) in women aged 40–44 y. Both age patterns are less pronounced when calculated as a percent of all women (Figure 1).

Our estimate of the global number of couples affected by infertility is lower than that of Boivin et al. [5] or Rutstein and Shah [6]. Boiven et al. estimated 72.4 million women were currently infertile in 2006 [5]. They used the median prevalence reported by seven published infertility studies that used a 12- or 24-mo definition of infertility; our estimates differ because we used a larger dataset and a different algorithm to calculate infertility [5],[10]. Rutstein and Shah presented a variety of infertility measures using DHS data from the late 1990s, demonstrating the importance of choices in defining infertility [6]. They estimated that 186 million ever-married women in developing countries (excluding China) were infertile in 2002; this larger number is a result of definitional differences: they included women who may not have been exposed to the risk of pregnancy and women aged 15–20 y and 45–49 y, age groups that have higher prevalences of infertility than women aged 20–44 y.

The strengths of this study were the application of consistent algorithms to calculate primary and secondary infertility from 277 survey datasets, most of which were nationally representative; our use of a Bayesian hierarchical model to estimate infertility prevalence and trends; and our systematic quantification of uncertainty. We identified where survey data did not collect information on past contraceptive use or marital status, and corrected for biases that arose when information on contraceptive use or marriage was incomplete. We used definitions of primary and secondary infertility that allowed us to disentangle trends in ability to have a child from trends in fertility preferences [25]. Specifically, women who were not in a union, had used any contraceptive in the previous 5 y, or did not wish to have a child were excluded from both the numerator and the denominator when calculating the prevalence of infertility. This allowed us to calculate trends in infertility that were independent from worldwide declines in the preferred number of children and independent of population growth in that time period.

The major limitations of our study are gaps in data for certain countries, the use of proxies to assess exposure to pregnancy, potential reporting inaccuracies, and the inability of our definition to capture all instances of infertility. Despite extensive data seeking, data gaps remained, especially in high-income countries and in Central and Eastern Europe. The use of demographic and reproductive health surveys to infer infertility prevalence requires several assumptions. First, we assume that women who are in a union, wish to have a child, and are not using contraceptives are engaged in regular, unprotected sexual intercourse. We also rely on women's reported couple status, births, contraceptive use, and desire for a child. These assumptions may be violated, as women may not report accurately on sensitive topics, such as past voluntary abortions [26],[27]. Women might also report non-biological children as their own. Furthermore, the reporting of the date of marriage and date of last birth may not be accurate in some settings [7]. Several studies have found that, in China, reporting of births in household surveys may be suppressed or the timing of births may be misreported because of policy considerations, which could affect our infertility estimates [28]–[30]. Finally, infertile women may state that they do not want a child, as a coping mechanism [17],[31]. Our correction of incomplete contraceptive and marriage information, use of birth as the outcome, and use of a 5-y infertility definition reduced the susceptibility of our estimates to these biases [13]. Some types of infertility are not measured using our algorithm [32]. The algorithm cannot capture any infertile men whose female partners conceive and give birth to a child with another man, nor primary infertility in men who have had multiple partners. It is not possible to capture infertile couples trying to have a child but using condoms intermittently for sexually transmitted infection (STI) prevention [21]. Lastly, our 5-y definition excludes from the prevalence estimation men and women who do not maintain a union for 5 y. Our prevalence estimate of infertility, however, is applied to all couples in a union, independent of the length, to calculate absolute numbers of couples affected. To the extent that infertile unions are more likely to dissolve than fertile unions, we expect our estimate to be biased downwards because we only measure infertility in unions that last for 5 y [33].

There are several important implications of the algorithm we use to measure infertility. We measure current infertility using a 5-y exposure with birth as an outcome. An infertility measure based on ability to become pregnant may have different patterns, trends, and levels than those presented in this paper. Infertility prevalences measured using a shorter exposure period would have a similar geographic and temporal pattern, but would be approximately twice as high as our estimates (see Figure Q in Text S1; [13]) The shorter exposure period identifies couples affected by temporary separations or periods of abstinence or lactational amenorrhea, infertility that resolves at between 2 and 5 y, and infertile unions that dissolve after 2 y but before 5 y without a birth. Our algorithm does not capture childlessness experienced by couples who are no longer of reproductive age or infertility experienced by women aged less than 20 y. Infertility that is identified and successfully treated within a 5-y period is not captured by this definition. Finally, men and women who use contraceptives, choose to be childless, or are not in a union, may indeed be infertile. However, these individuals are not included in our estimate of the number of infertile unions. We aimed to calculate the number of couples currently affected by infertility, and these individuals are not currently attempting to have a child, or, in the case of those not in a union, it is not possible to determine whether they are attempting to have a child.

Multiple factors—infectious, environmental, genetic, and even dietary in origin—can contribute to infertility [34]. These factors may affect the female, the male, or both partners in a union, resulting in an inability to become pregnant or carry a child to term. Current evidence, mostly from clinical studies with few exceptions [35], indicates that differences in the incidence and prevalence of infectious diseases, leading to fallopian tube blockage in women, are the main reason for changes over time and differences between populations [36]–[39]. Some have hypothesized that sperm quality is declining [40], but the evidence is not conclusive [41].

Increasing age at childbearing could also increase the prevalence of infertility, as the ability to become pregnant and deliver a live birth reduces with age in all populations. Globally, the mean age at childbearing has remained the same (about 28 y) since the 1970s, although this masks regional and temporal heterogeneity in trends [42]. In low- and middle-income countries, age at first birth has increased, although first birth still occurs at young ages: in 40 countries with one DHS survey in the 1990s and another survey during 2000–2011, the overall median of the median age at first birth among women aged 25–49 y increased from 19.8 to 20.3 y [42]. While the age at first birth has increased, the average number of children has decreased, and thus, the mean age at childbearing has not changed in these countries [42]. On the other hand, mean age at first birth and mean age at childbearing have increased in all developed countries since the 1990s [42],[43]. This does not appear to have affected primary infertility levels in those countries. However, it may have contributed to the modest increase in secondary infertility that we estimated.

The geographic pattern of infertility prevalence we found is consistent with previous estimates of infertility in Sub-Saharan Africa, specifically high prevalence in some West, Central, and Southern African countries, and low prevalence in most East African countries [7],[8],[44]. This pattern has mainly been attributed to the consequences of untreated reproductive tract infections, including both STIs such as Neisseria gonorrhoeae and Chlamydia trachomatis, and, to a lesser extent, infections from unsafe abortions or obstetric practices [34],[36],[45]. The improved trends for the region as a whole may be due to reduced prevalence of STIs, possibly associated with changes in sexual behavior and STI treatment in response to the HIV epidemic. There are, however, no reliable data on regional trends in the prevalence of STIs. WHO estimated that the prevalence of C. trachomatis infections among adult females in 2005 was 4%–6% in all regions of the world, except the WHO Eastern Mediterranean and South East Asia regions, where prevalence was below 2% [46]. N. gonorrhoeae was considerably more prevalent in the WHO African region than all other regions among adult women and men. If the prevalence of maternal syphilis has decreased since 1990, it may have reduced the risk of stillbirths and therefore increased the ability to have a live birth, which is our definition of fertility [47]–[50]. Infection is also associated with reduced fertility. Infertile women, especially those with primary infertility, are more likely to acquire HIV infection because of greater marital instability [51], and HIV is also associated with reduced fertility in the later stages of infection [52]. However, the population effect of the HIV epidemic on fertility is likely small: despite the epidemic, infertility declined in all Sub-Saharan African subregions.

Post-abortion complications are also an important factor contributing to infertility. The risk is higher for unsafe practices than for safe abortion procedures. The relatively high levels of secondary infertility in the Central/Eastern Europe and Central Asia region may be associated with the higher incidence of abortion. In these regions, the abortion rate declined between 1995 and 2003, but stayed at levels higher than the global average [16]. Both induced abortions and higher levels of STIs/HIV may play a role in explaining the elevated levels of secondary infertility in the Caribbean. Declines in unsafe abortion rates in Sub-Saharan Africa between 1995 and 2003 may have contributed to declines in infertility rates [16].

Among women who have had a pregnancy or birth, pregnancy complications may cause infections of the reproductive tract that result in infertility. Maternal mortality ratios—an indicator of obstetric risk—are estimated to have declined slightly in Sub-Saharan Africa and more substantially Southern Asia since 1990, and it is possible that injuries/infections caused or aggravated by childbirth declined together with decreases in maternal mortality [2].

Dataset S1

Dataset S2

Text S1