Sexual Activity and Weekly Contraceptive Discontinuation and Selection Among Young Adult Women in Michigan.

Understanding young women's contraceptive and pregnancy prevention behaviors is important for helping women and their partners control if and when they have children. Prior research on associations between patterns of sexual activity and contraceptive behaviors is limited. We assessed the influence of recent sexual activity on discontinuation and selection of specific contraceptive methods. We used weekly data from the Relationship Dynamics and Social Life (RDSL) study, a longitudinal 2.5-year population-based project that sampled woman ages 18 and 19 (N = 1,003) in one Michigan county. We estimated logistic and multinomial regression models that accounted for clustering of weekly observations within partnerships and women. Weekly discontinuation of longer-acting methods declined with increasing sexual activity in the past month, as did discontinuation of shorter-acting hormonal methods. Sexual activity was associated with decreased selection of condoms relative to other methods. Future research into life events that lead to changes in the frequency of sexual activity may provide insight into times when women are at risk of contraceptive discontinuation. These findings underscore the importance of anticipatory guidance in contraceptive counseling so that when women change their contraceptive behavior they are equipped in advance with resources to make safe transitions between methods.

Almost half of pregnancies in the United States are unintended, and young adults ages 20 to 24 have the highest unintended pregnancy rate (Finer & Zolna, [10]), even though the majority of young adults at risk of unintended pregnancy use contraception (Mosher, Moreau, & Lantos, [31]). Approximately 40% of women with unintended births report using contraception at the time of conception (Mosher, Jones, & Abma, [30]). The large differences between perfect use and typical use failure rates for the pill and condom (Sundaram et al., [39]; Trussell, [40]) reflect unintended pregnancies that arise due to inconsistent and incorrect method use.

One of the reasons that women often give for not using a contraceptive method is infrequent sexual activity (Frost, Singh, & Finer, [13]; Mosher et al., [30]). Positive associations between frequency of sexual activity and contraceptive use have been identified in national cross-sectional surveys (Frost, Singh, & Finer, [12]; Wu, Meldrum, Dozier, Stanwood, & Fiscella, [45]). Prior research also finds that more frequent sexual activity is associated with use of more effective methods (Frost & Darroch, [11]; Gibbs, Kusunoki, Colantuoni, & Moreau, [15]; Kusunoki & Upchurch, [25]).

Changes in sexual activity may influence contraceptive behavior via changes in perceived risk of unintended pregnancy, as conceptualized in multiple health behavior theories that relate perceived risk of a health outcome to behavior change (Brewer & Rimer, [6]). The literature on sexual activity and contraceptive use, however, does not distinguish the potentially differential influences of sexual activity on contraceptive method selection versus contraceptive continuation. The behavioral processes for initiating a health behavior, such as method selection, are likely distinct from the processes for maintaining a health behavior (Rothman, [35]). For example, contraceptive discontinuation and inconsistent use may be influenced by changes in unintended pregnancy risk perceptions, while method selection may be influenced by characteristics of specific methods.

The link between changes in sexual activity and discontinuation of specific contraceptive methods has been acknowledged in the literature but with limitations across studies (Inoue, Barratt, & Richters, [21]). Some show that women acknowledge changes in frequency of sexual activity as a reason for discontinuing or switching contraceptive methods (Huber et al., [20]; Jaccard et al., [22]; Rosenberg & Waugh, [34]), but findings are mixed with respect to an association between frequency of sexual activity and discontinuation or intermittent use of contraceptive methods. Two studies have identified an association between less frequent sexual activity and discontinuing the pill and condoms (Huber et al., [20]; Sanders, Graham, Bass, & Bancroft, [36]). Frost and Darroch ([11]), however, provided methodologically compelling evidence for a possible null association between sexual activity and inconsistent use, finding no association between average frequency of sexual activity and less than perfect condom or pill use over the past three months. These inconsistent findings highlight the potential limitations of data that summarize behaviors over the course of months or years. Little research has been conducted specifically on sexual activity and method selection, particularly with longitudinal data sets where individual women can be followed over time.

The complex interplay between sexual activity and contraceptive behaviors is further complicated by characteristics of partners and partnerships. Changes in partners and partnerships are important situational factors that contribute to changes in both sexual activity and contraceptive behaviors over time. Patterns of sexual activity, for example, change over the course of relationships (Rao & Demaris, [33]) and vary across different types of relationships (Wildsmith, Manlove, & Steward-Streng, [43]). These same factors have also been found to influence contraceptive use (Frost & Darroch, [11]; Manlove, Ryan, & Franzetta, [26]; Manlove et al., [27]), and risk of contraceptive discontinuation appears to vary by marital status (Grady, Billy, & Klepinger, [17]; Trussell & Vaughan, [41]). Furthermore, events such as a breakup may precipitate contraceptive discontinuation and increase risk of a subsequent unintended pregnancy and abortion (Bajos et al., [1]; Moreau, Beltzer, Bozon, & Bajos, [29]). While the saliency of partners and relationships to understanding how sexual activity relates to contraceptive discontinuation is clear, many prior studies have not fully accounted for these dynamics.

This study addresses the dearth of research on the influence of patterns of sexual activity on contraceptive discontinuation and method selection. An improved understanding of young adult women's contraceptive and pregnancy prevention behaviors is important for helping young women and their partners gain control over if and when they conceive. This information is also important for clinicians and counselors providing advice to help women select a contraceptive method based on past and current patterns of sexual behavior in a way that anticipates future changes in sexual activity and contraceptive behaviors. We used a longitudinal population-based sample of young adult women, ages 18 to 22, acknowledging the dynamic nature of sexual and contraceptive behaviors. Specifically, we investigated (a) how recent frequency of sexual activity is associated with weekly contraceptive discontinuation and (b) the association between recent sexual activity and subsequent method selection.

We used data from the Relationship Dynamics and Social Life (RDSL) study, a population-based sample of a Michigan county (Barber, Kusunoki, & Gatny, [4]). Women ages 18 and 19 were randomly sampled at baseline (N = 1,003) and followed for 2.5 years or until loss to follow-up. Baseline surveys were collected in 2008–2009, and women completed brief weekly follow-up surveys. These weekly surveys included partner-specific questions that allowed researchers to group data collected about specific partnerships over time. Follow-up surveys were completed by 95% (n = 953) of the baseline cohort and resulted in a total of 57,602 weeks of data. Most women remained in the study, with a high retention rate (75%) at 18 months of follow-up (Barber, Kusunoki, & Gatny, [3]). Continued participation, however, varied by sociodemographic characteristics. Women who were African American and those with less education, for example, remained in the study for fewer days on average (Barber, Kusunoki, Gatny & Schulz, [2]). This secondary analysis was limited to sexually active weeks when women were not pregnant and did not want to become pregnant (N = 664 women; n = 14,300 weeks). The RDSL received ethical approval from the University of Michigan Institutional Review Board (IRB) and the National Institutes of Health (NIH). This secondary analysis was exempted from full review by the Johns Hopkins Bloomberg School of Public Health Institutional Review Board.

Women were asked about contraceptive use each week through a series of questions that minimized response burden. First, they were asked if since the last interview they had used or done "anything that can help people avoid becoming pregnant." If they answered yes to this question, they were asked sequentially about use of specific noncoital methods until indicating an affirmative response. Later in the interview, women who reported sexual activity were then asked about use of specific coital methods. We classified weekly contraceptive method use into four categories according to the most effective method reported: longer-acting methods (intrauterine device [IUD], implant, sterilization, partner vasectomy, and injectable contraception); pill, patch, and ring (referred to as shorter-acting hormonal methods); condoms; and all less effective methods (e.g., withdrawal, spermicides, diaphragm), including no method. A discontinuation was identified if the reported method type was different than the method in the prior week. This weekly measure of discontinuation captures long-term discontinuations as well as short-term interruptions in contraceptive use, often classified as inconsistent use. The second outcome was the four-category measure of contraceptive method selection following each discontinuation. Switches from dual-method use (condoms plus a more effective method) to condom use were not considered a switch for method selection analyses.

Any penile–vaginal sexual activity was assessed dichotomously each week in the survey and was defined in the question as "when a man puts his penis into a woman's vagina." We defined recent sexual activity as the proportion of sexually active weeks in the month prior to each week of sexual activity. Because of missed and late surveys, we used up to four recent prior weeks of data reported within 60 days of observation. Weeks with less than two prior weeks of data within 60 days were excluded (n = 1,575 weeks). We scaled past-month sexual activity by a factor of four in regression models so that resulting odds ratio (OR) estimates indicate relative increases per additional week sexually active.

The RDSL consists of data structured within three levels of measurement. The most granular perspective comes from the Level 1 weekly follow-up survey data. Weekly data are grouped into women's reported partnerships at Level 2, and partnerships are clustered within individual women at Level 3. To identify the short-term past-month association between sexual activity and contraceptive behaviors, we decomposed past-month sexual activity information from partnership- and woman-level measures of sexual activity (Begg & Parides, [5]). We first calculated the proportion of weeks sexually active for each partnership and for each woman overall. We then centered short-term and partnership-level measures of sexual activity by subtracting the partnership-level measure from the short-term measure and subtracting the partnership-level measure from the woman-level measure. We included these two centered measures, as well as the woman-level measure, in each regression model. The resulting regression coefficients for the centered short-term measures provide the effects of short-term sexual activity, regardless of the partnership- and woman-level sexual activity, such that a significant short-term effect is truly related to variation over time for individual women rather than an artifact of her overall level of sexual activity. This approach has been taken for other similar analyses (Harvey, Washburn, Oakley, Warren, & Sanchez, [19]; Weir & Latkin, [42]) and is described in greater detail elsewhere for our study (Gibbs et al., [15]).

Potential individual-level confounders that we considered included age, race/ethnicity, high school grade point average (GPA), enrollment in postsecondary education at baseline, ever having received public assistance prior to enrollment, religiosity, sexual debut by age 15, more than one prior lifetime sexual partner at enrollment in the study, and any pregnancies prior to enrollment in the study.

Relationship qualities included relationship type, classified as casual, nonexclusive dating, long distance, exclusive dating, cohabiting, and married or engaged (Kusunoki, [24]). Stability was assessed according to whether the relationship had experienced a previous breakup and reconciliation. We also considered prior pregnancies with the current partner. Demographic characteristics of partners included age difference of three or more years, education differential, partners' prior children, and partners' current fertility desires.

We used multilevel logistic regression models with three levels (weeks, partnerships, and women) to estimate the association of recent and longer-term sexual activity with method type discontinuation. Models were estimated with random intercepts to account for correlation between weeks of data within partnerships and partnerships within women. We included a main effect for specific method type in each model because of the differences in discontinuation between different methods (Trussell, [40]). We also assessed modification of the effect of recent sexual activity on discontinuation by specific method type. We multiplied interaction coefficients by main effects to obtain the effects of recent sexual activity on discontinuation for each specific contraceptive method type.

For method selection analyses, we estimated three-level marginal multinomial logistic regression models. least effective methods were selected as the reference group for all models. Each model was estimated with robust standard errors (SEs) to account for correlated weekly data. We accounted for clustering of weeks within partnerships and partnerships within women by estimating bootstrapped bias-corrected and accelerated (BCa) confidence intervals (CIs) (Carpenter & Bithell, [7]) with resampling at the level of the individual woman, so that all partnerships and weeks of data for a selected woman were resampled together in blocks. Each model was estimated with 1,000 bootstrap replications. For these multinomial analyses we estimated marginal models that account for the correlated data structure, rather than conditional random intercept models as for the discontinuation analyses, because of the complexity of the three-level multinomial model. Although the magnitude of an effect size can differ between a correctly specified marginal model and a conditional random intercept model, the direction and significance level is the same and leads to consistent statistical inference (Neuhaus, [32]).

All models included the centered measures of past-month and partnership-level sexual activity, as well as the measure of woman-level sexual activity. For each outcome we estimated unadjusted models as well as models adjusting for potential individual-level and partnership-level confounders as appropriate, which were selected both theoretically and empirically. Age was included in all models on a theoretical basis. Additional confounders were selected separately for discontinuation and method selection models based on empirical associations with measures of sexual activity and contraceptive behavior. For discontinuation analysis, these confounders were race/ethnicity, postsecondary enrollment, high school GPA, religiosity, public assistance, multiple past partners, sexual debut by age 15, pregnancy prior to study enrollment, relationship type, relationship duration, pregnancy with current partner, partner age difference, and partners' children. Method selection analyses included race/ethnicity, postsecondary enrollment, religiosity, public assistance, multiple past partners, relationship type, relationship duration, and pregnancy with current partner. All analyses were conducted using StataMP 14.2.

The sample included 664 women and a total of 14,300 weeks of data. Women's total number of weeks in the sample ranged from 1 to 123 and averaged 22 (SD = 24). Across weeks in the analytic sample, past-month sexual activity averaged 74% of weeks in the past month that were sexually active (SD = 33 percentage points) (Table 1). Women were sexually active for an average of 68% (SD = 26%) of weeks. Women reported discontinuing their current type of method in 14% of weeks overall. There were a total of 430 women who selected a new method during the study. Among these women there were 1,799 new method selections, with an average of 4.2 selections per woman. Of these method selections, 9% were longer-acting methods (6% injectable; 3% others), while 25% were shorter-acting hormonal methods, 27% were condoms, and 40% were least effective methods.

Table 1. Past-Month Sexual Activity and Weekly Contraceptive Behaviors

1 Note. Past-month sexual activity is the percentage of weeks sexually active in the past month, measured each week.

The majority (60%) of women were white (Table 2), and just over half of women were in postsecondary education at enrollment (57%). Approximately one in three women reported sexual debut at age 15 or earlier (36%), and one in four reported ever having been pregnant (26%) at baseline. The majority of women had more than one prior sexual partner at baseline (68%). The greatest proportion of weeks was spent in exclusive dating relationships (39%), followed by cohabiting (21%) and married or engaged (18%) relationships (Table 3). Just under 20% of weeks were spent in relationships with a prior breakup.

Table 2. Baseline Characteristics of Women (n = 664) During the Study

Table 3. Partnership Characteristics (n = 1,379 Partnerships; n = 14,300 weeks)

2 Note. Relationship duration as of the last week recorded in the analytic sample.

Our first unadjusted model indicated an association between recent sexual activity and a lower chance of subsequent method discontinuation (Table 4, Model 1). Each additional sexually active week in the past month was associated with a 21% reduction in the odds of discontinuing a method (OR = 0.79, 95% CI [0.76, 0.83]). We found that discontinuation was lowest when women were using shorter-acting hormonal methods relative to each of the other three types of methods. For example, the odds of discontinuing a longer-acting method were twice that of discontinuing a shorter-acting hormonal method in any given week (OR = 2.01 [1.54, 2.61]). The odds of discontinuing condoms were more than three times the odds of discontinuing a shorter-acting hormonal method (OR = 3.23 [2.73, 3.82]), as were the odds of discontinuing a least effective method (OR = 3.19 [2.70, 3.77]).

Table 4. Logistic Models for Weekly Odds of Method Discontinuation Among Sexually Active Weeks (N = 664 Women; N = 1,379 Partnerships; n = 14,300 weeks)

• 3 Note. Past-month sexual activity is scaled by 4 so that OR estimates refer to relative increases per additional week sexually active. Model 3 is adjusted for partnership-level sexual activity, woman-level sexual activity, age, race/ethnicity, postsecondary enrollment, high school grade point average, religiosity, public assistance, multiple past partners, sexual debut by age 15, pregnancy prior to enrollment, relationship type, relationship duration, pregnancy with current partner, partner age difference, and partner's children.
• 4 *p < 0.05; ***p < 0.001.

Effect modification analyses indicated that the association between recent sexual activity and method discontinuation varied depending on the type of method that women were using (Table 4, Model 2). Specifically, for condom use the association was attenuated to a 9% decline in odds of discontinuation per additional sexually active week (OR = 0.91, 95% CI [0.84, 0.99]). The associations for the other three types of methods, however, were stronger, reflecting an approximately 25% decline in odds of discontinuation per additional sexually active week (longer-acting: OR = 0.71 [0.60, 0.83]; shorter-acting hormonal: OR = 0.77 [0.70, 0.84]; least-effective: OR = 0.72 [0.66, 0.78]).

Adjusting the effect modification model for potential confounders resulted in minor differences in the associations between recent sexual activity and discontinuation of specific methods (Table 4, Model 3). Each of the method-specific associations between recent sexual activity and discontinuation was somewhat attenuated. The association for condom discontinuation was attenuated to the point of becoming not statistically significant (aOR = 0.94, 95% CI [0.87, 1.02]). Each of the associations with discontinuation of the other three methods remained highly statistically significant (p < 0.001). The main effects of specific method type on discontinuation were somewhat attenuated, but all remained highly statistically significant (p < 0.001).

Overall, greater frequency of sexual activity in the past month was associated with a lower chance of selecting condoms relative to selecting each other specific method (Table 5). Specifically, the risk ratio of selecting condoms relative to the least effective methods was 23% reduced for each additional week sexually active in the past month (RRR = 0.77, 95% CI [0.71, 0.84]). The risk ratio for sexual activity and selection of condoms relative to shorter-acting hormonal methods (RRR = 0.80 [0.70, 0.90]) was similar to the risk ratio relative to longer-acting methods (RRR = 0.81 [0.70, 0.95]).

Table 5. Multinomial Logistic Models for Contraceptive Method Selection (N = 430 Women; n = 1,799 Switches)

• 5 Note. CI = bootstrapped bias-corrected and accelerated confidence intervals. Past-month sexual activity is scaled by 4 so that RRR estimates refer to relative increases per additional week sexually active in prior month. Unadjusted models include measures of partnership-level sexual activity and woman-level sexual activity. Adjusted models also include age, race/ethnicity, postsecondary enrollment, religiosity, public assistance, multiple past partners, relationship type, relationship duration, and pregnancy with current partner.
• 6 ***p < 0.001.

Relative selection among the nonbarrier methods (i.e., methods other than condoms) did not relate to past-month sexual activity. There was no significant difference in selection of a shorter-acting hormonal method relative to the least effective methods (RRR = 0.96, 95% CI [0.87, 1.08]), selection of longer-acting methods relative to least effective methods (RRR = 0.95 [0.81, 1.09]) or selection of longer-acting methods relative to shorter-acting hormonal methods (RRR = 0.98 [0.84, 1.15]) per additional week sexually active in the past month. Adjusting for potential confounders left the associations between sexual activity and method selection almost unchanged.

Patterns of sexual activity have a complex association with contraceptive behaviors. Several prior studies have indicated positive associations between more frequent sexual activity and use of more effective contraceptive methods (Frost & Darroch, [11]; Frost et al., [12]; Gibbs et al., [15]; Kusunoki & Upchurch, [25]; Wu et al., [45]). Our study expands this literature, indicating that the relative influence of sexual activity on discontinuation and method selection is heterogeneous by specific method type.

Women appear to take into account their recent sexual activity when deciding whether to select condoms as their contraceptive method or to select another method. Specifically, women who have recently been less sexually active tend to select condoms. This finding is generally consistent with prior research suggesting an association between less frequent sexual activity and greater condom use (Katz, Fortenberry, Zimet, Blythe, & Orr, [23]; Marshall et al., [28]; Sayegh, Fortenberry, Shew, & Orr, [37]; Sheeran, Abraham, & Orbell, [38]). To our knowledge, what is lacking in this prior literature is a distinction between the influence of sexual activity on selection of condoms as a new contraceptive method versus influence on continuation of condom use. The variety of measures of condom use, such as condom use at last sex (Marshall et al., [28]) or the number of condom-unprotected sex acts (Sayegh et al., [37]), do not distinguish between the decision to start using condoms and the decision to continue using condoms.

We found that the mechanism for the influence of sexual activity on hormonal and longer-acting contraceptive method use was different than the mechanism for condom use. The influence of sexual activity on use of more effective methods appeared to be through its association with discontinuation, rather than through an association with initial method selection. These findings are consistent with the reasons that women give for switching methods (Huber et al., [20]; Jaccard et al., [22]; Rosenberg & Waugh, [34]). The heterogeneity of the association with selection and discontinuation across more effective methods may explain some of the heterogeneity in prior studies (Wildsmith et al., [43]; Wilson & Koo, [44]).

Future research into life events that can lead to changes in the frequency of sexual activity may provide insight into times when women are at risk of discontinuing a contraceptive method. Preliminary research, for example, has identified housing instability and residential moves as times when women are at risk of not using contraception (Clark, Kusunoki, & Barber, [8]). There is also evidence that partnership instability, which could result in disrupted patterns of sexual activity, may increase the risk of unintended pregnancy (Moreau et al., [29]).

There were limitations to this study. While this was a randomly selected population-based sample, the target population was narrow, reflecting young women ages 18 to 22 in a single county in Michigan. These findings may not apply to older women, particularly those in much longer-term partnerships, and those in other geographic settings within or outside the United States. In addition, our measures of sexual activity were limited to a dichotomous weekly measure, designed to reduce the weekly burden of response and increase retention rates. While a more detailed account of coital frequency within a specific week would add valuable information to this type of analysis, the loss to follow-up due to burdensome weekly surveys would likely not be worth the gained information.

Regardless of these limitations, our research has important implications for public health practice. National guidelines recommend that a medical history be taken regarding information that may be relevant to selection of a contraceptive method, including taking a history of recent sexual activity (Gavin et al., [14]). There is not, however, specific guidance on how to use this sexual history in the context of contraceptive counseling beyond a note in the medical eligibility criteria for contraceptive use specific only to the use of withdrawal, which the criteria state "might be appropriate for couples ... who have intercourse infrequently" (Division of Reproductive Health, Centers for Disease Control and Prevention et al., [9]). While clinicians and counselors are in a position to discuss contraceptive use and sexual activity, they need guidance on what to discuss.

Contraceptive counseling frequently occurs in the context of new method selection, but we found that this is not where the greatest influence of sexual activity occurs. Counselors need to be equipped to provide anticipatory guidance to women about how method discontinuation can occur in the context of future life events and changes in life circumstances that result in changes in sexual activity. Counselors can support women who make changes in their contraceptive behavior when they are out of clinical contact by equipping women in advance with resources to make safe transitions between contraceptive methods. The goal of contraceptive counseling should be not to urge women to choose a specific method but to enable them to make an informed and autonomous decision about what method to select (Gomez, Fuentes, & Allina, [16]; Gubrium et al., [18]). Taking into account our study findings, we suggest that an additional goal of contraceptive counseling should be to enable women to make future informed decisions about contraceptive continuation and selection.