In the United States, a majority of pregnancies (64–82%) among females aged 15–19 years and 20–24 years are unintended, a higher proportion than in other age groups.1 More than half of unintended pregnancies occur among females who were using contraception when they became pregnant.2 Females whose pregnancies are unintended are more likely to experience adverse pregnancy outcomes, including low-birth-weight neonates, preterm delivery, and preterm premature rupture of membranes.3–5
Coitus interruptus or withdrawal is a widely used but understudied contraceptive method. Most prior studies of this method have focused on older, married couples.6 Withdrawal use might be quite common, with between 10% and 60% of recent study cohorts reporting ever-use of withdrawal.7–13 Despite the high proportion of females who use withdrawal, most practitioners do not mention withdrawal when discussing contraception options with their patients.11 When withdrawal is used correctly every time a couple has sexual intercourse, the pregnancy rate is approximately 4% per year, a perfect use rate comparable to that of male condoms.14 However, the pregnancy rate, 18–28% per year of use, associated with typical use of withdrawal exceeds other methods.15,16
In this study, we examined whether withdrawal use and unintended pregnancy are correlated, as well as elucidated characteristics of females who choose withdrawal. Because withdrawal is typically less effective at preventing pregnancy than other methods, we hypothesized that withdrawal users will be more likely to report an unintended pregnancy. We focused on females aged 15–24 years because this age group is at highest risk of experiencing unintended pregnancy and accounts for an estimated 73% of unintended pregnancies in the United States.2,17
MATERIALS AND METHODS
Our analysis used the first public release file of the continuous version of the National Survey of Family Growth, a nationally representative survey of U.S. females aged 15–44 years conducted from June 2006 to December 2008. The National Survey of Family Growth is a population-based survey drawing participants from the entire United States, which collects data on pregnancy incidence, pregnancy outcomes, contraceptive use, marriage, divorce, cohabitation, sexual behavior, economic status, and many other topics. Same-gender interviewers collected most of the data through in-person interviews. Interviewees self-administered a portion of the survey regarding sensitive information such as abortion history, using a computer to gain more accurate responses. The survey sampling design was based on geographic areas and oversampled for multiple subgroups, including African Americans and Hispanics, to render the sample truly representative. Complete information on survey design, implementation, coverage, refusal rates, and recoding can be found at: http://www.cdc.gov/nchs/NSFG.htm.
We used retrospective calendar data to construct our variable for contraceptive use. Starting with the January 3 years before the year of survey and continuing until the month before survey (up to 47 months), respondents were asked to choose from a preprinted list up to four contraceptive methods they had used in each month, including the option “withdrawal, pulling out.” Respondents could change methods from month to month. We classified participants into two groups of contraceptive users: females who reported that they used withdrawal in at least 1 month (“withdrawal users”) and females who used contraception in at least 1 month but did not report withdrawal as one of their methods in any month (“other contraception users”).
For each pregnancy, participants reported the month of conception as well as their thoughts about the timing of the pregnancy. Unintended pregnancies included both those pregnancies that were unwanted (the woman did not want a child at the time of pregnancy or in the future) and those that were mistimed (the woman wanted a child in the future but not at the time she actually got pregnant). We categorized pregnancies about which participants indicated “indifferent” or “did not care” as well as pregnancies that were reported to occur “too late” rather than too early, as intended. We used these questions to construct our measure of unintended pregnancy rather than whether the pregnancy was terminated, both because abortions are underreported in the National Survey of Family Growth and because many unintended pregnancies are carried to term.18
We also included several control variables in our analyses that prior studies have indicated are associated with withdrawal use, unintended pregnancy, or both.1,19 Some National Survey of Family Growth variables are only measured at the time of survey, including educational attainment (ranging from 9 to 19 years), whether the woman was currently in the labor force, number of lifetime sex partners, whether she had been tested or treated for a sexually transmitted infection in the past 12 months, whether she had ever used emergency contraception, and whether she wanted more children in the future. Retrospective calendar data on births and relationship histories enabled us to measure some variables at the time of censoring rather than survey, including age, relationship status (married, cohabitating, or single noncohabitating), and total number of live births. Finally, some variables we included remained constant over time, including race and ethnicity (white non-Hispanic, African American non-Hispanic, Hispanic, and other) and age at first sex. In keeping with the approach of Ranjit et al,15 we measured economic status using a dichotomous variable indicating whether an individual reported that her household income was at or below 200% of the federal poverty line in the year before the survey.
We used calendar data on contraception and pregnancy for all months reported by sexually experienced females (defined as having had sex at least once since menarche) aged 15–24 years. Participants were right-censored (no longer contributed data to our study) at the month of their first pregnancy during the study period, at the month of tubal ligation, at the month in which they turned 25 years of age, or at the time of survey. We excluded participants who did not use contraception in any month, because these females are likely quite different from others in terms of their pregnancy intentions and sexual behaviors. We excluded participants who were missing any contraceptive information, those for whom the month and year of their tubal ligation were unknown, those who received a tubal ligation before the start of the study, and those who were pregnant at the start of the study. We also excluded participants missing data for at least one of the variables in our multivariable models.
We estimated either the proportion or the mean value of each variable to delineate descriptive statistics for our sample. We used bivariate analysis to compare proportions (for categorical variables) or mean values (for continuous variables) of key predictors in two mutually exclusive populations: females who had an unintended pregnancy and females who did not have an unintended pregnancy (which included females who had an intended pregnancy). We determined statistical significance of the association in the bivariate comparisons using χ2 tests for categorical variables and modified Wald tests for continuous variables.
We performed multivariable analysis using Cox proportional hazard models to estimate hazard ratios and 95% confidence intervals (CIs) for correlates of an unintended pregnancy. We used a multivariable logit model to estimate odds ratios (ORs) and 95% CI associated with predictors of withdrawal use during the study period. We included covariates in both multivariable models if they were associated with either unintended pregnancy or withdrawal use at P<.05 in bivariable comparisons.
We carried out all analyses in STATA 11.0; we considered associations statistically significant at the P<.05 level. We weighted all regressions using the “svy” series of commands in STATA as well as the included variables for probability sampling weights, strata, and principle sampling units to account for survey design effects given that the National Survey of Family Growth is a survey that uses a multistage sampling design and is not a simple random sample. Although accounting for survey design effects results in increased standard errors in our estimated models, we believe that because the geographic and demographic variables used to define the sampling frame are likely correlated with unintended pregnancy and withdrawal use, these survey effects must be accounted for to produce unbiased conclusions for this study.20
The research reported in this article was registered and deemed exempt from review by the University of Chicago Biological Sciences institutional review board.
The National Survey of Family Growth included 2,640 sexually active female respondents who were aged 15–24 years during at least some part of the study period. We excluded 88 nonusers of contraception, three females who were missing contraceptive information, 33 females for whom the month and year of their tubal ligation were unknown, 31 females who received a tubal ligation before the start of the study, and 211 females who were pregnant at the start of the study. Fifty-four participants were missing data on covariates examined in bivariate and multivariable analysis; χ2 tests indicated that these females were no more likely to experience an unintended pregnancy (P=.89) or differ in their use of withdrawal (P=.28) from those with complete data. These 54 participants were also excluded from our final study sample, which included 2,220 females who contributed a mean value of 22.4 calendar months of data.
Within our final study sample, 31.0% of females used withdrawal for at least 1 month during the study, and 69.0% relied solely on other methods of contraception (Table 1; we calculated all percentages using weights to account for survey design effects and thus those presented differ from raw percentages). Most withdrawal users (89.8%) also used other methods during the course of the study. Of the females who used withdrawal as their only form of contraception, 11.3% had an unintended pregnancy compared with 22.6% of those who used withdrawal among other contraceptive methods (P=.11). Almost all (56/62) of the females who relied solely on withdrawal during the study period had tried another contraceptive method at some point in their life.
During the study period, 665 females (26.6% of the weighted sample) experienced a pregnancy; 414 of these pregnancies (59.2% of the weighted sample) were unintended. Overall, 15.7% of females experienced an unintended pregnancy during the study period. Females who used withdrawal were at higher risk of an unintended pregnancy compared with females who solely used other forms of contraception (21.4% compared with 13.2%), even after controlling for other factors (adjusted hazard ratio 1.75, 95% CI 1.23–2.49; Table 2). Females tested or treated for a sexually transmitted infection in the year before their interview were more likely to experience an unintended pregnancy (24.1% compared with 12.2%) as were females living in poverty, African American females, and those in cohabitating relationships (as compared with noncohabiting singles). Parity, a desire to have more children, and years of education were inversely associated with reporting an unintended pregnancy (Table 2). Removing participants who reported intended pregnancies did not change our finding of increased risk of unintended pregnancy among females who use withdrawal (adjusted hazard ratio 1.78, 95% CI 1.24–2.55).
We compared characteristics of females who reported withdrawal use with those of females who used only other contraceptive methods (Table 3). Withdrawal users were 7.5% more likely to have used emergency contraception (adjusted OR 1.57, 95% CI 1.13–2.20). Married participants were 14.8% less likely than single respondents to report withdrawal use.
We demonstrated that withdrawal use is quite common among young U.S. females, because nearly one-third of females aged 15–24 years in our nationally representative sample indicated that they had recently used withdrawal. Females who used withdrawal were at a higher risk of unintended pregnancy compared with those who used other forms of contraception. The majority of pregnancies in this study were unintended. Withdrawal use in lieu of effective methods of contraception may contribute to high rates of unintended pregnancies among U.S. females.
Using withdrawal as contraception requires determination and communication by both the male and female partners21 and can be an effective method of contraception, particularly among long-term, monogamous couples committed to consistent use.6 Previous qualitative research indicates that young females typically use withdrawal as a substitute contraceptive method when they have forgotten to use a hormonal method, have difficulty using condoms, or as an “emergency” form of contraception for unanticipated intercourse, often during a female's very first sexual experience.11,12 Females in this age group who use withdrawal are also more likely to engage in other risk behaviors such as having multiple sex partners.22 These contextual factors as well as fertility status may reflect why withdrawal use is particularly ineffective for young females. In this study, females who solely used withdrawal were less likely to experience an unintended pregnancy, perhaps because they were more consistent in their use of this method, although this difference was not significant. Females in long-term marital or cohabitating relationships were less likely to choose withdrawal.
Most females in this study, however, did not use withdrawal as their only contraceptive method. Whittaker and colleagues23 indicated that many females who use withdrawal do so because they are dissatisfied with hormonal methods or condoms. These same authors noted that their study participants almost never discussed withdrawal with their reproductive health care provider. Given that almost all withdrawal users in our study also used more effective methods of contraception at some point, practitioners might build on their patients' desire to prevent pregnancy to support their use of more effective contraceptive methods. As Whittaker and colleagues23 also noted, most young females learned about withdrawal from sources such as peers, parents, and the media and often harbored misperceptions regarding the basic mechanism and efficacy of this practice.
The limitations of our study must be noted. First, it is likely that withdrawal users are heterogeneous. Our category of “withdrawal user” included females who used withdrawal in 1 month only as well as those who used withdrawal more regularly. Females in these two categories likely have different risks of unintended pregnancy. Nevertheless, we found that exclusive withdrawal users had a similar risk of unintended pregnancy when compared with females reporting use of withdrawal among other methods. Second, others have noted withdrawal use is underreported in the National Survey of Family Growth and other large surveys, in part because females may not consider withdrawal a contraceptive method.11 Underreporting of withdrawal among females in our “other contraception” category would cause us to overestimate the risk of an unintended pregnancy among withdrawal users. Although participants could only list up to four methods each month, only one to two females each month reported a fourth contraceptive method, indicating that any underreporting of withdrawal was likely not because respondents ran out of space. Third, surveys in general are subject to recall bias and these biases may be exaggerated for withdrawal. Although most females likely reported the dates of their pregnancies correctly, one would imagine recalling each contraceptive method is difficult and intermittent withdrawal use even more difficult. Fourth, we use calendar data, allowing us to know that contraceptive use preceded pregnancy. However, other covariates (eg, educational attainment, ever-use of emergency contraception, economic status) were measured at the time of survey and may not reflect conditions at the time of pregnancy or contraceptive choice. Finally, we likely included in our study some females who, for a portion of observation time, were not truly at risk of an unintended pregnancy as a result of changes in pregnancy intention, although removing participants who reported intended pregnancies during the study period did not change our main conclusion.
Despite these limitations, this study demonstrated a high prevalence of withdrawal use and an association with unintended pregnancy. Although unintended pregnancy and withdrawal can be difficult to classify, even intermittent use of withdrawal is associated with an increased risk of unintended pregnancy. Withdrawal users are more likely to use postcoital methods, an additional marker for risk of undesired pregnancy. Ultimately, withdrawal use is an important marker of pregnancy risk and even females who report contraceptive use may also use withdrawal. Furthermore, these females are at high risk of sexually transmitted infections. Our study suggests that clinicians should routinely inquire about withdrawal use even among females reporting condom and combined hormonal contraceptive use.
1. Finer L, Zolna M. Unintended pregnancy in the United States: incidence and disparities. 2006. Contraception 2011;84:478–85.
2. Finer L, Henshaw S. Disparities in rates of unintended pregnancy in the United States, 1994 and 2001. Perspect Sex Reprod Health 2006;38:90–6.
3. Kost K, Landry D, Darroch J. The effects of pregnancy planning status on birth outcomes and infant care. Fam Plann Perspect 1998;30:223–30.
4. Mohllajee AP, Curtis KM, Morrow B, Marchbanks PA. Pregnancy intention and its relationship to birth and maternal outcomes. Obstet Gynecol 2007;109:678–86.
5. Eggleston E, Tsui A, Kotelchuck M. Unintended pregnancy and low birthweight in Ecuador. Am J Public Health 2001;91:808–10.
6. Rogow D, Horowitz S. Withdrawal: a review of the literature and an agenda for research. Stud Fam Plann 1995;26:140–53.
7. Chandra A, Martinez GM, Mosher WD, Abma JC, Jones J. Fertility, family planning, and reproductive health of U.S. women: data from the 2002 National Survey of Family Growth. Vital Health Stat 23 2005:1–160.
8. Kaiser Family Foundation. National survey of adolescents and young adults: sexual health knowledge, attitudes, and experiences. Menlo Park (CA): Kaiser Family Foundation; 2003.
9. Santelli JS, Morrow B, Anderson JE, Lindberg LD. Contraceptive use and pregnancy risk among U.S. high school students, 1991–2003. Perspect Sex Reprod Health 2006;38:106–11.
10. Woods J, Hensel D, Fortenberry J. Influences on contraceptive withdrawal in adolescent women. J Pediatr Adolesc Gynecol 2007;20:S134–5.
11. Jones RK, Fennell J, Higgins JA, Blanchard K. Better than nothing or savvy risk—reduction practice: the importance of withdrawal. Contraception 2009;79:407–10.
12. Higgins J, Hirsch J. Pleasure and power: incorporating sexuality, agency, and inequality into research on contraceptive use and unintended pregnancy. Am J Public Health 2008;98:1803–13.
13. Anderson J, Santelli J, Morrow B. Trends in adolescent contraceptive use, unprotected and poorly protected sex, 1991–2003. J Adolesc Health 2006;38:734–9.
14. Hatcher RA, Trussell J, Stewart F, Cates W Jr, Stewart G, Guest F, et al.. Contraceptive technology. 19th ed. New York (NY): Ardent Media; 2007.
15. Ranjit N, Bankole A, Darroch JE, Singh S. Contraceptive failure in the first two years of use: differences across socioeconomic subgroups. Fam Plann Perspect 2001;33:19–27.
16. Kost K, Singh S, Vaughan B, Trussell J, Bankole A. Estimates of contraceptive failure from the 2002 National Survey of Family Growth. Contraception 2008;77:10–21.
17. Finer L. Unintended pregnancy among U.S. adolescents: accounting for sexual activity. J Adolesc Health 2010;47:312–4.
18. Jones R, Kost K. Underreporting of induced and spontaneous abortion in the United States: an analysis of the 2002 National Survey of Family Growth. Stud Fam Plann 2007;38:187–97.
19. Frost J, Darroch J. Factors associated with contraceptive choice and inconsistent method use, United States, 2004. Perspect Sex Reprod Health 2008;40:94–104.
20. Korn E, Graubard B. Analysis of large health surveys: accounting for sampling design. J R Stat Soc Ser A Stat Soc 1995;158:263–95.
21. Camelo T, Landry D. Young unmarried men and women discuss men’s role in contraceptive practice. Fam Plann Perspect 1994;26:222–7.
22. Cavazos-Rehg PA, Krauss MJ, Spitznagel EL, Schootman M, Peipert JF, Cottler LB, et al.. Type of contraception method used at last intercourse and associations with health risk behaviors among US adolescents. Contraception 2010;82:549–55.
23. Whittaker PG, Merkh RD, Henry-Moss D, Hock-Long L. Withdrawal attitudes and experiences: a qualitative perspective among young urban adults. Perspect Sex Reprod Health 2010;42:102–9.