Association of Hormonal Contraceptive Use With Reduced Levels of Depressive Symptoms

A National Study of Sexually Active Women in the United States

Katherine M. Keyes; Keely Cheslack-Postava; Carolyn Westhoff; Christine M. Heim; Michelle Haloossim; Kate Walsh; Karestan Koenen


Am J Epidemiol. 2013;178(9):1378-1388. 

In This Article

Materials and Methods


Data were drawn from the National Longitudinal Study of Adolescent Health,[23] an ongoing nationally representative study designed to address health, behaviors, and relationships. The study began with approximately 90,000 adolescents aged 12–21 years who were sampled at school in 1994–1995; approximately 20,000 of these adolescents underwent in-depth home interviews and longitudinal follow-up. Follow-up interviews were conducted in 1996 (wave 2), 2001–2002 (wave 3), and 2007–2008 (wave 4).

The present study focused on the subset of participants who completed a wave 4 interview in 2007–2008, at ages 25–34 years. Figure 1 outlines the exclusion criteria for the present sample. Women who were not sexually active in the past year were not asked about contraceptive use; thus, we did not include them in the analytical sample. The sample size for analysis was 6,654.

Figure 1.

Selection of the study sample from participants in the wave 4 interview (ages 25–34 years) of the National Longitudinal Study of Adolescent Health (Add Health), 2007–2008.


Contraceptive Use in Wave 4 Women who had had a sex partner in the past year were asked whether they used any form of birth control or disease prevention. We divided the sample into groups based on World Health Organization categories of contraception effectiveness.[24]

The first group included women using methods that are rated as highly effective, and included vasectomy, intrauterine devices (IUDs), and tubal ligation (n = 831; 12.5%). We did not have information on use of hormonal IUDs versus nonhormonal IUDs; however, hormonal IUDs would probably comprise a small proportion of total IUD use in this sample, which was already low (4.5%). A second category included methods rated as effective; these included combination hormonal contraceptives in the form of the contraceptive pill, patch, or ring (n = 2,393; 36.0%). We analyzed women using known progestin-only forms of contraception (i.e., Depo-Provera (Pfizer Inc., New York, New York) or Norplant (Wyeth Pharmaceuticals, Radnor, Pennsylvania)) separately (n = 298; 4.5%) and report results separately from those for the pill, patch, and ring group. Among women who did not use a highly effective or effective method, a third category included women who used condoms, a diaphragm, a vaginal sponge, or the withdrawal method (n = 1,913; 28.8%). Among women who did not use any of the above methods, a fourth category included the least effective contraceptive methods: periodic abstinence, spermicides, and contraceptive film (n = 91; 1.4%). Finally, 17.0% (n = 1,128) of women in the sample did not report any contraceptive use with a current partner. Because of low prevalence, women using the least effective contraceptive methods were combined with women using no contraception.

Contraceptive Uuse in Waves 1–3 In waves 1 and 2, women were asked whether they were currently taking birth control pills (wave 1: 8.2%; wave 2: 8.6%). We note that the wave 1 and 2 interviews were conducted between 1994 and 1996, before the contraceptive patch and ring formulations were introduced. We did not have information on known progestin-only forms of contraception in waves 1 and 2. However, fewer than 2% of women have ever used these forms of contraception;[25] thus, while there may have been some women in the referent group who were using a hormonal form of contraception, the number is unlikely to have been appreciable.

In wave 3, questions on contraceptive use mirrored the wave 4 measures. We included anyone using a pill, patch, or ring and users of known progestin-only products as hormonal contraceptive users (50.4%). We created dichotomous indicators for these hormonal contraceptive users versus all other women. As noted above, information on hormonal IUDs was not included, but the number of women using these devices was probably negligible.

Depressive Symptoms and Suicide Attempts in Waves 1–4 The presence of depressive symptoms during the past 7 days was assessed in all waves using the Center for Epidemiologic Studies Depression Scale (CES-D).[26,27] In waves 1 and 2, the 20-item CES-D Scale was used, with each item scored 0–3, for a maximum score of 60. In waves 3 and 4, the 10-item CES-D Scale was used, with each item scored 0–3, for a maximum score of 30. Both the 10-item and the 20-item CES-D have well-documented reliability and validity.[26–28] In waves 3 and 4, a CES-D score greater than or equal to 11 was considered a high level of depressive symptoms.[29] In each wave, respondents were also asked how many times they had attempted suicide in the past year (dichotomized at 1 or more attempts).

Other Covariates We assessed a range of covariates for capturing information on the health and lifestyles of sexually active women. All covariates included are shown in Table 1 . Demographic factors included age, race/ethnicity, education, and parity. Relationship factors included past-year monogamy and total lifetime number of vaginal sex partners. Health-related factors included current body mass index (weight (kg)/height (m)2), derived using weight and height measured by the interviewer at the time of the interview or imputed with self-report data when measured data were unavailable; participation in any individual sports during the past week (e.g., running or jogging); visiting a dentist during the last 2 years; undergoing a routine checkup by a physician during the last 2 years; and current smoking status. We also included the centrality of religion to the respondent, given that religion may be important in the decision to use contraceptives[30,31] and is also associated with depression.[32]

Statistical Analysis

First, we examined whether women using various forms of contraception or no contraception (4-level categorical variable) varied with regard to health-related covariates. We used chi-square tests for categorical outcomes and t tests for continuous outcomes. We created a propensity score using the binary variable of hormonal contraceptive use versus none as the outcome. Covariates used to create the propensity score included health-related covariates as well as design variables (study stratum (geographic area), primary sampling unit, and sample weight), as is recommended for propensity score analysis in complex surveys.[33] We examined the distribution of contraception categories within quintiles of the propensity score for sufficient sample sizes to conduct the analysis and controlled for propensity score in additional regression models.[34] Second, we tested whether women differed with regard to wave 4 depression symptoms, a high level of depressive symptoms, and past-year suicide attempts based on wave 4 contraceptive use. Women using the least effective contraceptives or no contraceptives were designated the referent group. Linear regression was used for depressive symptoms and logistic regression for high levels of depressive symptoms and past-year suicide attempts. Results of regression analyses were first unadjusted and then adjusted for all of the health-related variables listed in Table 1; analyses were also carried out with adjustment for propensity score. In addition, analyses of depression outcomes were also controlled for prior-wave highest depression score, and analyses of suicide attempts were also controlled for any prior suicide attempt. Users of known progestin-only products were analyzed separately. Third, longitudinal associations were assessed examining wave 3 hormonal contraceptive use (versus all other wave 3 respondents), wave 4 hormonal contraceptive use (versus all other wave 4 respondents), and high levels of depressive symptoms and past-year suicide attempts in both wave 3 and wave 4—all with logistic regression models. Analyses were conducted using SAS-callable (SAS Institute Inc., Cary, North Carolina) SUDAAN (RTI International, Research Triangle Park, North Carolina) and incorporated design weights.