- Email: [email protected]
A self-assessment efficacy tool for spermicide contraceptive users
Original Research
ajog.org
GYNECOLOGY
A self-assessment efficacy tool for spermicide contraceptive users Courtney A. Schreiber, MD; Sarah J. Ratcliffe, PhD; Mary Dupuis Sammel, ScD; Paul G. Whittaker, DPhil
BACKGROUND: Easily accessible contraceptive methods, such as chemical and barrier methods, are used currently by approximately 1 in 6 women who use contraception in the United States. Even in the face of suboptimal effectiveness, coitally dependent methods likely will always have a role in fertility management. Because most contraceptive efficacy stratifications use population-based data, for women to make informed decisions about the individual fit of a contraceptive method, better evidence-based, user-friendly tools are needed. OBJECTIVES: Spermicides are a readily available, over-the counter, woman-controlled contraceptive method, but their effectiveness is userdependent. Patient-decision aids for spermicides and other barrier methods are not well-developed, and overall failure rates could be improved by aids that account for individual characteristics. We sought to derive a prediction rule for successful use of spermicides for pregnancy prevention and to convert those data to a point-of-care instrument that women can use when they are considering spermicide use during contraceptive decision-making. STUDY DESIGN: We pooled local data from 3 randomized clinical trials that were published in 2004, 2007, and 2010 that tested spermicide efficacy. We constructed a prediction rule for unintended
P
atient-centered health care and decision-making may improve adherence to medical therapy,1 but family planning service provisions in the United States generally do not utilize patient-centered models. The persistently high unintended pregnancy rates in the United States may be due to systemic barriers that impede patientfriendly access.2 With the exception of some barrier methods and some emergency contraceptive methods, women must interact with a health care provider to obtain contraception; unless she has good insurance coverage, contraceptives are expensive, and several steps are often required to obtain the method. Best Cite this article as: Schreiber CA, Ratcliffe SJ, Sammel MD, et al. A self-assessment efficacy tool for spermicide contraceptive users. Am J Obstet Gynecol 2016;214:264.e1-7. 0002-9378/$36.00 ª 2016 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.ajog.2015.10.018
pregnancy using bootstrap validation and developed a scoring system. RESULTS: Data from 621 women showed a mean age of 29 years; 49% of the women were African American, and 43% were white. The overall pregnancy rate was 10.3% (95% confidence interval, 7.9e12.7) over 6 months. In adjusted logistic regression, age >35 years was protective against pregnancy (odds ratio, 0.19; 95% confidence interval, 0.06e0.58; P ¼ .003), and multigravidity was associated with high failure rates (odds ratio, 7.24; 95% confidence interval, 3.04e17.3; P < .001). These risk factors (together with frequency of unprotected sex) were used in a model that maximized sensitivity for pregnancy prediction to compute the predicted probability of unintended pregnancy for each woman. This model was 97% accurate in predicting women who had a <5% pregnancy risk while using spermicides. CONCLUSION: Using prospectively collected data, we built a simple risk calculator for contraceptive failure that women can consult when considering spermicide use. This instrument could support patientcentered contraceptive decision-making. Key words: contraceptive efficacy, decision, spermicide, tool
practices for contraceptive counseling and decision-making have not been wellestablished. Data suggest, however, that women may prefer more autonomy when it comes to contraceptive use than other medical interventions3 but that we have much to learn in this domain.4 The most effective long-acting reversible contraceptive (LARC) methods are emphasized increasingly by providers. LARC methods are very low maintenance over time and highly effective for all women, regardless of user characteristics. But some women are deterred by the need to have the method placed and removed by a provider.5 For these and other reasons, easily accessible contraceptive methods, such as chemical and barrier methods, currently are used by approximately 17% of US women who use contraception6 and likely will remain a mainstay of US contraceptive use. In theory, vaginal spermicides possess many positive attributes: they are woman-controlled, are inexpensive, are
264.e1 American Journal of Obstetrics & Gynecology FEBRUARY 2016
available directly to the consumer overthe-counter, possess a limited sideeffect profile, and are safe.7,8 The main limitation of spermicides is that they are not highly effective. Pregnancy rates in the first year of spermicide use are estimated at 10-20%, but in typical use may be even higher.9 To contextualize this, highly effective methods such as the LARC class are associated with <1 of 100 pregnancies annually.10 Even in the face of suboptimal effectiveness, womancontrolled coitally dependent methods likely will always have a role in fertility management. Because most contraceptive efficacy stratifications use population-based data, for women to make informed decisions about the individual fit of a contraceptive method, better evidence-based, user-friendly tools are needed. Our aim was to use prospectively collected data to build a user-friendly scoring system to predict individualized contraceptive efficacy for women who
ajog.org
GYNECOLOGY
TABLE 1
Comparison of baseline demographic and clinical characteristics for the trials in the combined data set Variable
C31G11 (n ¼ 138)
Mean age, y standard deviationa
29.8 6.1
BufferGel12 (n ¼ 300) 27.6 5.3
FHI Spermicide7 (n ¼ 183) 28.9 5.9
Race/ethnicity, n (%) African-American
78 (56.5)
153 (51.0)
78 (42.6)
White
50 (36.2)
125 (41.7)
93 (50.8)
Other
10 (7.2)
22 (7.3)
12 (6.6)
Never married
99 (71.7)
243 (81.0)
127 (69.4)
Married or living with partner
32 (23.2)
54 (18.0)
39 (21.3)
Separated, divorced, widowed
7 (5.1)
3 (1.0)
17 (9.3)
0
46 (33.3)
114 (38.0)
74 (40.4)
1
31 (22.5)
69 (23.0)
39 (21.3)
2
24 (17.4)
45 (15.0)
25 (13.7)
3
37 (26.8)
72 (24.0)
45 (24.6)
0
76 (55.1)
171 (57.0)
108 (59.0)
1
30 (21.7)
63 (21.0)
30 (16.4)
2
14 (10.1)
37 (12.3)
21 (11.5)
18 (13.0)
29 (9.7)
24 (13.1)
Marital status, n (%)
Gravidity, n (%)
Parity, n (%)
3 Mean body mass index, kg/m standard deviationb 2
Mean no. of contraceptive methods used at baseline, n standard deviationb
26.8 6.9 N/A 1.6 0.8
1.0 1.0
27.3 6.3 1.6 0.9
Used barrier method at baseline, n standard deviationb
65 (47.1)
199 (66.3)
75 (41.0)
Pregnancy during study, n (%)
16 (11.6)
29 (9.7)
19 (10.4)
C31G, a spermicide and microbicide, not commercially available; FHI, Family Health International; N/A, not available. P < .01; b P < .001, reflects significant statistical difference between studies (Mann-Whitney U-test or chi-square test). Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
a
use vaginal spermicides. To accomplish this goal, we pooled data from clinical trials to yield estimates of a woman’s risk of pregnancy while using spermicides. We included both pregnancies that were characterized as user-failures and those that were characterized as methodfailures to increase the generalizability of our results.
Material and Methods The data came from 3 recent multicenter randomized controlled trials; the
methods and primary results of all the trials have been published elsewhere.7,11,12 Briefly, 2 studies compared the efficacy, safety, and acceptability of the spermicide under study with an active placebo,11,12 and 1 study compared 5 different doses/delivery systems of Nonoxynol-9.7 Women who were recruited for each study met the following inclusion criteria: they were healthy, sexually active, 18-42 years of age, had no history of infertility, no sexually transmitted disease diagnosis in previous 6
Original Research
months, had normal length menstrual cycles (defined as 24-35 days in 2 studies and 21-35 days in the third study), and had vaginal intercourse with regularity. The median coital act frequency was 2.3e3 times per week. Women in the trials were observed for at least 6 months or until a pregnancy occurred. From these 3 clinical trials, we extracted the local data that were collected prospectively at the University of Pennsylvania Center for Clinical Research in Women’s Health, created a new database, and analyzed these data to elucidate risk factors for pregnancy during these trials. This study was exempt from institutional review board approval, because it was a secondary data analysis of deidentified data. The overall failure rates for the 3 studies were similar (Table 1), which enabled us to pool the data. Testing for stratification by trial indicated that the studies were sufficiently homogenous to be combined. Each clinical trial collected detailed data on menstrual cycles, intercourse timing, and contraceptive methods that had been used (or not used) during each act of intercourse. Two trials reported that 97% of coital acts used spermicide alone7,11; in the third trial, all the women used spermicide in conjunction with a diaphragm.12 Pregnancies were each carefully dated by the Principal Investigator of each clinical trial who used information about last menstrual period, intercourse timing, and early pregnancy ultrasound scanning. We characterized incident pregnancies by variables that are potential risk factors for pregnancy such as age, race, gravidity/parity, and previous contraceptive use. Because the overall pregnancy rate was 10.3%, odds ratios were used to approximate incidence rate ratios for dichotomous risk factors, and Mann Whitney U tests were used to compare pregnant cases and control cases, with respect to continuous variables. We used statistical software (STATA; Stata Corporation, College Station, TX) to perform descriptive analyses; logistic regression was used to construct the optimal predictive model for incident pregnancy, with emphasis given to high sensitivity. The final model
FEBRUARY 2016 American Journal of Obstetrics & Gynecology
264.e2
Original Research
ajog.org
GYNECOLOGY
was simplified to generate a scoring algorithm.
Results A total of 621 women were enrolled in the 3 barrier method studies. The baseline demographics and clinical characteristics were similar among participants across the 3 studies, with statistical differences having no meaningful clinical difference (Table 1; MannWhitney U- or chi-square tests). The women had a mean age of 29 years; 43% of them were white, and 49% were black; 58% of them were nulliparous, and 38% of them were nulligravid. The overall pregnancy rate over 6 months was 10.3% (95% confidence interval [CI], 7.9e12.7). In unadjusted analyses (Table 2; Mann-Whitney Uor chi-square tests), the women who were of black race (P ¼ .046), obese (P ¼ .058), of higher gravidity (P < .001) or parity (P ¼ .026), and had unprotected sex more frequently (P ¼.026) were at higher risk for unintended pregnancy during the trial periods. The demographic and clinical characteristics were used to generate (with the use of logistic regression) the optimal predictive model with respect to sensitivity to contraceptive failure (Table 3). With the assumption of the use of the spermicide with 100% of coital acts, women who had been pregnant previously and were <35 years old were found to be at higher risk (>5%) for an unintended pregnancy; nulligravid women who were >35 years old were at low risk (<5%) for pregnancy (Figure 1). The area under the receiver operating characteristic curve for this model is 0.74. The multivariable model (Table 3) was then used to calculate a predicted probability of unintended pregnancy prevention or contraceptive efficacy for each woman; a woman was classified as low risk for unintended pregnancy if her probability of avoiding pregnancy with this model was >95%. This cutpoint was used to ensure that pregnancy prediction had >90% sensitivity (actual, 90.2%) because the cost of an unintended pregnancy is high. With the use of this model, only 6 of the unintended pregnancies (9.8%) were classified
TABLE 2
Demographic and clinical characteristics by pregnancy status during the study period Pregnancy during study period? Yes (n ¼ 64)
No (n ¼ 557)
<25 Y
20 (32.8)
165 (30.2)
25-<35 Y
36 (59.0)
291 (53.3)
5 (8.2)
90 (16.5)
White
19 (29.7)
247 (44.3)
Black
41 (64.1)
266 (47.8)
Other
4 (6.3)
44 (7.9)
1 (1.6)
27 (4.9)
Characteristic Age, n (%)
35 Y Race, n (%)
a
Education, n (%)
12 (19.8)
113 (20.4)
Some college
32 (50.0)
236 (42.7)
College graduate
19 (29.7)
177 (32.0)
<10
10 (28.6)
89 (21.1)
10-29
11 (31.4)
93 (32.5)
30-49
9 (25.7)
62 (21.7)
50
5 (14.3)
42 (14.7)
27 (62.8)
215 (56.0)
9 (20.9)
91 (23.7)
7 (16.3)
78 (20.3)
14 (43.8)
76 (27.6)
18 (56.3)
199 (72.4)
0
10 (16.4)
223 (40.8)
1
15 (24.6)
125 (22.9)
2
12 (19.7)
73 (13.4)
24 (39.3)
126 (23.0)
0
25 (41.0)
326 (59.6)
1
18 (29.5)
102 (18.7)
2
11 (18.0)
58 (10.6)
7 (11.5)
61 (11.2)
Household income ($1000), n (%)
Desire future children, n (%) Yes Maybe No Obese, n (%)
b
Yes No Gravidity, n (%)
c
3þ Parity, n (%)
a
3
Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
incorrectly in the current sample as low risk of pregnancy (false positives). The specificity was low at 42%, but the
264.e3 American Journal of Obstetrics & Gynecology FEBRUARY 2016
(continued)
negative predictive value was high, 97%. A total of 233 women (38%) were classified as low risk of unintended
ajog.org
GYNECOLOGY
TABLE 2
Demographic and clinical characteristics by pregnancy status during the study period (continued) Pregnancy during study period? Yes (n ¼ 64)
Characteristic
No (n ¼ 557)
Average cycle length, n (%) <25 D
1 (1.6)
11 (2.0)
25-31 D
60 (93.8)
509 (91.4)
>31 D
3 (4.7)
37 (6.6)
Percentage of unprotected sex during study, mean standard deviationa
12.1 18.9
5.7 10.9
P < .05; b Body mass index data available only on 307 subjects; c P < .001, reflects significant statistical difference between groups (Mann-Whitney U test or chi-square test). Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016. a
pregnancy. The ability of this model to discriminate among women who become pregnant from those who do not is measured by the area under the receiver-operator-characteristic curve, which was 0.737 (95% CI, 0.672, 0.802). A bootstrap validation13 was conducted to assess the model’s robustness. With a thousand bootstrap samples, the average area under the receiveroperator-characteristic curve was 0.75
with the empirical 95% CI [0.728, 0.814]. 10-fold cross-validation was performed and the predictive error rate from that method was 0.70 (95% CI [0.629e0.771]). Based on these results, a user-friendly scoring algorithm was generated to predict those women who were at low risk of becoming pregnant while using spermicides. The simplification was achieved by (1) recalculating the predictive score using values of the
TABLE 3
Final logistic regression model for spermicide failure (pregnancy) Variable
Odds ratio
95% Confidence interval
P value
Pointsa
Age, y <25
Reference
25-35
0.56
0.29e1.09
.089
2
35
0.19
0.06e0.58
.003
0
3
Gravity 0
Reference
1
3.38
1.43e7.99
.005
2
2
5.97
2.33e15.3
< .001
3
7.24
3.04e17.3
< .001
3
1.03
1.02e1.05
< .001
—
3 Unprotected sex acts, %
a
b
0
<10
0
10-29.9
2
30
4
This column includes the weighting or points used to calculate the simplified scoring system; b Continuous percentages were used for the logistic models, but discreet categories were used for the computation of the score; analyses were adjusted for race. Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
Original Research
coefficients, rounded to the nearest integer, and (2) by calculating the score simply as the sum of dichotomized “risk factors” that were identified by the model. For an individual woman, a risk score can be calculated easily by summing the points for each risk factor (Table 3; Figure 2). This simplified model performed comparably to the full model: the area under the receiveroperator-characteristic curve was 0.73 (95% CI, 0.669e0.793; Figure 1). The scores in Table 4 rank women with respect to their risk of contraceptive failure; the absolute risk of pregnancy, along with 95% CIs, is presented for each value. A woman who calculates her score to be “4” has twice the risk of pregnancy as a woman who scores “3.” Given these estimates of absolute risk and the test characteristics, a cutoff score of 4 was chosen to classify women as high risk for pregnancy. In the sample used to derive the rule, this threshold resulted in a total of 267 of the women (44%) classified as low risk, with 97% accuracy in the lowrisk prediction (negative predictive value); 8 of the women (9.8%) with unintended pregnancies were classified incorrectly as low risk. The 2 strongest influences on the probability of incidental pregnancy in our model were the number of previous pregnancies and age 35 years. When the method was used with only 50% of sex acts, the risk of pregnancy was also increased greatly (odds ratio, 5.36; 95% CI, 2.15e13.3), but occasional nonuse alone was not an important predictor of pregnancy.
Comment We conducted a secondary analysis of data combined from 3 different spermicide efficacy trials to build a prediction rule for successful method use. We used our findings to create a simple scoring system that women can incorporate into their contraceptive decisionmaking. We found that nulliparous women 35 years old are at low risk of pregnancy. We also found that, although perfect use was an important enhancer of contraceptive success, it was not more important than the biologic markers of fertility, such as age or gravidity. These data provide evidence that a woman’s
FEBRUARY 2016 American Journal of Obstetrics & Gynecology
264.e4
Original Research
ajog.org
GYNECOLOGY
FIGURE 1
Predicted method failure rates
These predicted method failure rates were based on age, the number of previous pregnancies, and the assumption of 100% protected sex. The numbers by each data point illustrate the calculated “score” with the use of our model. Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
personal characteristics and her behaviors are important in the facilitation of contraceptive success, as emphasized in a recent Cochrane review on spermicide use.14 There are women for whom the efficacy of spermicides is comparable with LARC, and for whom the decision to use spermicides would be medically sound, should this fit best with their values. The generalizability and validity of our results are supported by the fact that increasing age and nulliparity have been shown previously to be protective against pregnancy in the National Survey of Family Growth15 and other studies.16 Although older age is a well-known risk factor for decreased fertility, contraceptive trials enroll only women of reproductive age who are regularly sexually active and who are still at risk of pregnancy. Even within this fertile group that engages in regular sexual intercourse, age makes a difference. We found similar results among women who were enrolled in a microbicide trial: sexually active women in that study had linearly decreased risk of unintended pregnancy during the trial per each advancing year of age.17 An advantage of our method is that we were able to use prospective data efficiently to identify women who were most at risk of pregnancy while using
this class of contraceptives. Previous investigations of predictors of unintended pregnancy while using a contraceptive method have been limited by design challenges. Most existing data on risk factors for unintended pregnancy and contraceptive effectiveness are derived from the National Survey of Family Growth. This survey relies on recall of behaviors around the time of pregnancy and recall of the classification of the pregnancy as intended or unintended, (modified by socialdesirability bias18-20). Although this retrospective survey has produced
important estimates over decades, prospective data provide us with more accurate risk-estimates to develop a clinical tool to aid in self-assessment and contraceptive decision-making. In our study, all pregnancies were, by definition, unintended because all women who were enrolled in the spermicide clinical trials were screened rigorously for pregnancy avoidance before enrollment. In addition, the risk of misclassification bias is low because, within the parent clinical trials, menstrual cycles and pregnancy tests were monitored closely and prospectively. In
FIGURE 2
Decision aid for patient use when assessing pregnancy risk while using spermicides
Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
264.e5 American Journal of Obstetrics & Gynecology FEBRUARY 2016
ajog.org
GYNECOLOGY
TABLE 4
Absolute risk of pregnancy by risk assessment score Score
Frequency, n
Chance of pregnancy, %
95% Confidence interval, %
0
6
0
0.0e45.9
1
0
N/A
2
106
2.8
0.59e8.0
3
155
3.2
1.1e7.4
Risk Low
—
4
77
6.5
2.1e14.5
5
171
15.8
10.7e22.1
6
32
15.6
5.3e32.8
7
39
18.0
7.5e 33.5
8
22
40.9
20.7e63.6
High
N/A, not applicable. Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
our secondary analysis, we included all pregnancies that were conceived during the trial period, which allowed us to capture documented user-failures in addition to method-failures, much like the real world typical use.
Limitations One limitation is the error that is inherent in our moderate sample size with regard to pregnancy events. However, the fact that all 3 studies had a pregnancy rate between 10% and 12% may lend validity to our overall point estimate. Additionally, the limited number of observed pregnancies limited our ability to perform cross-validation of the model and proposed cut-offs. We have presented error rates for misclassification using the same data that were used for model development. These error rates may over-state the performance of the proposed rule. External validation of the rule is highly recommended to obtain more accurate performance information before implementation. Another limitation is that the direction of biases that are associated with unprotected sex reporting is unknown. Study patients may have under-reported unprotected sex; in doing so without consequences (ie, pregnancy), the decision thresholds may be overestimates of risk. Alternatively, the women may have truly behaved more carefully because they were in a trial. However, the data
showed that user-failure (unprotected sex) still occurred and was a significant risk factor in the prediction of unintended pregnancy. Our use of clinical trial data may limit generalizability. Women who were willing to participate in clinical trials could be inherently different from women in the general population, but we believe these differences are likely to be minimal given that, before the Affordable Care Act, for many women, trial participation was the only way they could access contraception and basic gynecologic care, because of lack of insurance coverage. Because women who choose spermicides also are often un- or under-insured, we surmise that the differences between our study population and the women in the general population who may benefit from our findings are small. The clinical trial population included here reflects the sociodemographic factors that already are known to be associated with unintended pregnancy. The average age, socioeconomic, and marital statuses of the participants here are demographically similar to the very women for whom we need to improve our contraceptive counseling and access.21
Implications Metaanalysis indicates that decision aids improve informed patient choices and that decision aids have a positive effect on patient-practitioner
Original Research
communication.1 Providers have used scoring schemes successfully to assess contraceptive method risks22 and the need for Chlamydia trachomatis screening.23 A common reproductive health scoring scheme used by providers and patients is the Edinburgh Postnatal Depression Scale.24 To be useful, a prediction rule should be adaptable to the appropriate clinical setting.25 Our prediction rule was developed within a cohort of women that is representative of the population for whom it can be most useful; 50% of study subjects had used barrier methods before this study for contraception. We believe that, by focusing on concrete demographic and behavioral characteristics, we have built a rule that can be used easily by the women themselves. With the use of Table 4 and Figure 2, for example, a 30-year-old woman with 1 previous pregnancy (score, 4) may decide that a 6.5% risk of pregnancy is not only acceptable, given the advantages and cost of a spermicide, but also she can know that her risk of pregnancy will increase if she does not use protection consistently (add 2-4 points). Her 20-year-old friend with 2 children (score, 6) may decide that a 15% risk of pregnancy is too great, even if she uses the method perfectly. The utility of a prediction rule is dependent on its cost and on balancing the potential harm its introduction could cause. The rule presented here is free of cost and helps women to make personalized choices about fertility control without introducing harm. Recent publications from the prospective Contraceptive Choice Project have illustrated the important role that LARC methods play in increasing the proportion of US pregnancies that are intended.2,26 Although this landmark research provides us with the first independent longitudinal data of the longterm outcomes of hormonal and LARC contraceptive users, women who use barrier methods were not included in the Contraceptive Choice Project. The data we present here help to fill an important gap in the science of unintended pregnancy prevention and show that, for some women, spermicides can be highly effective as well. We are likely to have
FEBRUARY 2016 American Journal of Obstetrics & Gynecology
264.e6
Original Research
GYNECOLOGY
more sustained success in decreasing the unintended pregnancy rate in the United States if we work with patient-centered, individualized models. Although future studies could be done to test the efficacy of our self-assessment tool on contraceptive decision-making and ultimately unintended pregnancy prevention, its use is a low-risk, low-cost, individualized approach to help women avoid unintended pregnancy. n Acknowledgments We acknowledge Dr Kurt Barnhart (Department of Obstetrics and Gynecology, Perelman School of Medicine, University of Pennsylvania) for his helpful editing of this manuscript and thank Family Health International and the National institute of Child Health and Human Development for allowing us access to the data that were used in this study.
References 1. Stacey D, Bennett CL, Barry MJ, et al. Decision aids for people facing health treatment or screening decisions. Cochrane Database Syst Rev 2011;10:CD001431. 2. Peipert JF, Madden T, Allsworth JE, Secura GM. Preventing unintended pregnancies by providing no-cost contraception. Obstet Gynecol 2012;120:1291-7. 3. Dehlendorf C, Diedrich J, Drey E, Postone A, Steinauer J. Preferences for decision-making about contraception and general health care among reproductive age women at an abortion clinic. Patient Educ Couns 2010;81:343-8. 4. Dehlendorf C, Levy K, Kelley A, Grumbach K, Steinauer J. Women’s preferences for contraceptive counseling and decision making. Contraception 2013;88:250-6. 5. Foster DG, Grossman D, Turok DK, et al. Interest in and experience with IUD self-removal. Contraception 2014;90:54-9. 6. Jones J, Mosher W, Daniels K. Current contraceptive use in the United States, 2006e2010, and changes in patterns of use since 1995.
National health statistics reports; no 60. Hyattsville (MD): National Center for Health Statistics; 2012. 7. Raymond EG, Chen PL, Luoto J; Spermicide Trial Group. Contraceptive effectiveness and safety of five nonoxynol-9 spermicides: a randomized trial. Obstet Gynecol 2004;103: 430-9. 8. Schreiber CA, Meyn LA, Creinin MD, Barnhart KT, Hillier SL. Effects of long-term use of nonoxynol-9 on vaginal flora. Obstet Gynecol 2006;107:136-43. 9. Raymond EG, Trussell J, Weaver MA, Reeves MF. Estimating contraceptive efficacy: the case of spermicides. Contraception 2013;87:134-7. 10. American College of Obstetricians and Gynecologists. Increasing use of contraceptive implants and intrauterine devices to reduce unintended pregnancy: ACOG Committee Opinion, no. 450. Obstet Gynecol 2009;114: 1434-8. 11. Burke AE, Barnhart K, Jensen JT, et al. A randomized trial of the contraceptive efficacy, acceptability, and safety of C31G and nonoxynol-9 spermicidal gels. Obstet Gynecol 2010;116:1265-73. 12. Barnhart KT, Rosenberg MJ, MacKay HT, et al. Contraceptive efficacy of a novel spermicidal microbicide used with a diaphragm: a randomized controlled trial. Obstet Gynecol 2007;110:577-86. 13. Efron B, Tibshirani R. Improvements on cross-validation: the .632þ bootstrap method. J Am Stat Assoc 1997;92:548-60. 14. Grimes DA, Lopez LM, Raymond EG, Halpern V, Nanda K, Schulz KF. Spermicide used alone for contraception. Cochrane Database Syst Rev 2013;12:CD005218. 15. Finer LB, Zolna MR. Unintended pregnancy in the United States: incidence and disparities, 2006. Contraception 2011;84:478-85. 16. Menken J, Trussell J, Larsen U. Age and fertility. Science 1986;233:1389-94. 17. Schreiber CA, Whittington S, Cen L, Maslankowski L. Good intentions: risk factors for unintended pregnancies in the US cohort of a microbicide trial. Contraception 2011;83: 74-81.
264.e7 American Journal of Obstetrics & Gynecology FEBRUARY 2016
ajog.org 18. Moos MK, Petersen R, Meadows K, Melvin CL, Spitz AM. Pregnant women’s perspectives on intendedness of pregnancy. Womens Health Issues 1997;7:385-92. 19. Santelli J, Rochat R, Hatfield-Timajchy K, et al. The measurement and meaning of unintended pregnancy. Perspect Sex Reprod Health 2003;35:94-101. 20. Petersen R, Moos MK. Defining and measuring unintended pregnancy: issues and concerns. Womens Health Issues 1997;7: 234-40. 21. Finer LB, Henshaw SK. Disparities in rates of unintended pregnancy in the United States, 1994 and 2001. Perspect Sex Reprod Health 2006;38:90-6. 22. Spellacy WN, Kerr DA, White B, Keith GE, Waggoner BL. A family planning risk scoring system for health care providers. Obstet Gynecol 1984;63:846-9. 23. Lavoué V, Morcel K, Voltzenlogel MC, et al. Scoring system avoids Chlamydia trachomatis overscreening in women seeking surgical abortions. Sex Transm Dis 2014;41: 470-4. 24. Cox JL, Holden JM, Sagovsky R. Detection of postnatal depression: development of the 10-item Edinburgh Postnatal Depression Scale. Br J Psychiatry 1987;150:782-6. 25. Grady D, Berkowitz SA. Why is a good clinical prediction rule so hard to find? Arch Intern Med 2011;171:1701-2. 26. Winner B, Peipert JF, Zhao Q, et al. Effectiveness of long-acting reversible contraception. N Engl J Med 2012;366:1998-2007.
Author and article information From the Penn Family Planning and Pregnancy Loss Center (Drs Schreiber and Whittaker) and the Department of Biostatistics and Epidemiology, Center for Clinical Epidemiology and Biostatistics (Drs Ratcliffe, Sammel, and Whittaker), Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA. Received March 15, 2015; revised Oct. 14, 2015; accepted Oct. 18, 2015. The authors report no conflict of interest. Corresponding author: Courtney A. Schreiber, MD, MPH. [email protected]
ajog.org
GYNECOLOGY
A self-assessment efficacy tool for spermicide contraceptive users Courtney A. Schreiber, MD; Sarah J. Ratcliffe, PhD; Mary Dupuis Sammel, ScD; Paul G. Whittaker, DPhil
BACKGROUND: Easily accessible contraceptive methods, such as chemical and barrier methods, are used currently by approximately 1 in 6 women who use contraception in the United States. Even in the face of suboptimal effectiveness, coitally dependent methods likely will always have a role in fertility management. Because most contraceptive efficacy stratifications use population-based data, for women to make informed decisions about the individual fit of a contraceptive method, better evidence-based, user-friendly tools are needed. OBJECTIVES: Spermicides are a readily available, over-the counter, woman-controlled contraceptive method, but their effectiveness is userdependent. Patient-decision aids for spermicides and other barrier methods are not well-developed, and overall failure rates could be improved by aids that account for individual characteristics. We sought to derive a prediction rule for successful use of spermicides for pregnancy prevention and to convert those data to a point-of-care instrument that women can use when they are considering spermicide use during contraceptive decision-making. STUDY DESIGN: We pooled local data from 3 randomized clinical trials that were published in 2004, 2007, and 2010 that tested spermicide efficacy. We constructed a prediction rule for unintended
P
atient-centered health care and decision-making may improve adherence to medical therapy,1 but family planning service provisions in the United States generally do not utilize patient-centered models. The persistently high unintended pregnancy rates in the United States may be due to systemic barriers that impede patientfriendly access.2 With the exception of some barrier methods and some emergency contraceptive methods, women must interact with a health care provider to obtain contraception; unless she has good insurance coverage, contraceptives are expensive, and several steps are often required to obtain the method. Best Cite this article as: Schreiber CA, Ratcliffe SJ, Sammel MD, et al. A self-assessment efficacy tool for spermicide contraceptive users. Am J Obstet Gynecol 2016;214:264.e1-7. 0002-9378/$36.00 ª 2016 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.ajog.2015.10.018
pregnancy using bootstrap validation and developed a scoring system. RESULTS: Data from 621 women showed a mean age of 29 years; 49% of the women were African American, and 43% were white. The overall pregnancy rate was 10.3% (95% confidence interval, 7.9e12.7) over 6 months. In adjusted logistic regression, age >35 years was protective against pregnancy (odds ratio, 0.19; 95% confidence interval, 0.06e0.58; P ¼ .003), and multigravidity was associated with high failure rates (odds ratio, 7.24; 95% confidence interval, 3.04e17.3; P < .001). These risk factors (together with frequency of unprotected sex) were used in a model that maximized sensitivity for pregnancy prediction to compute the predicted probability of unintended pregnancy for each woman. This model was 97% accurate in predicting women who had a <5% pregnancy risk while using spermicides. CONCLUSION: Using prospectively collected data, we built a simple risk calculator for contraceptive failure that women can consult when considering spermicide use. This instrument could support patientcentered contraceptive decision-making. Key words: contraceptive efficacy, decision, spermicide, tool
practices for contraceptive counseling and decision-making have not been wellestablished. Data suggest, however, that women may prefer more autonomy when it comes to contraceptive use than other medical interventions3 but that we have much to learn in this domain.4 The most effective long-acting reversible contraceptive (LARC) methods are emphasized increasingly by providers. LARC methods are very low maintenance over time and highly effective for all women, regardless of user characteristics. But some women are deterred by the need to have the method placed and removed by a provider.5 For these and other reasons, easily accessible contraceptive methods, such as chemical and barrier methods, currently are used by approximately 17% of US women who use contraception6 and likely will remain a mainstay of US contraceptive use. In theory, vaginal spermicides possess many positive attributes: they are woman-controlled, are inexpensive, are
264.e1 American Journal of Obstetrics & Gynecology FEBRUARY 2016
available directly to the consumer overthe-counter, possess a limited sideeffect profile, and are safe.7,8 The main limitation of spermicides is that they are not highly effective. Pregnancy rates in the first year of spermicide use are estimated at 10-20%, but in typical use may be even higher.9 To contextualize this, highly effective methods such as the LARC class are associated with <1 of 100 pregnancies annually.10 Even in the face of suboptimal effectiveness, womancontrolled coitally dependent methods likely will always have a role in fertility management. Because most contraceptive efficacy stratifications use population-based data, for women to make informed decisions about the individual fit of a contraceptive method, better evidence-based, user-friendly tools are needed. Our aim was to use prospectively collected data to build a user-friendly scoring system to predict individualized contraceptive efficacy for women who
ajog.org
GYNECOLOGY
TABLE 1
Comparison of baseline demographic and clinical characteristics for the trials in the combined data set Variable
C31G11 (n ¼ 138)
Mean age, y standard deviationa
29.8 6.1
BufferGel12 (n ¼ 300) 27.6 5.3
FHI Spermicide7 (n ¼ 183) 28.9 5.9
Race/ethnicity, n (%) African-American
78 (56.5)
153 (51.0)
78 (42.6)
White
50 (36.2)
125 (41.7)
93 (50.8)
Other
10 (7.2)
22 (7.3)
12 (6.6)
Never married
99 (71.7)
243 (81.0)
127 (69.4)
Married or living with partner
32 (23.2)
54 (18.0)
39 (21.3)
Separated, divorced, widowed
7 (5.1)
3 (1.0)
17 (9.3)
0
46 (33.3)
114 (38.0)
74 (40.4)
1
31 (22.5)
69 (23.0)
39 (21.3)
2
24 (17.4)
45 (15.0)
25 (13.7)
3
37 (26.8)
72 (24.0)
45 (24.6)
0
76 (55.1)
171 (57.0)
108 (59.0)
1
30 (21.7)
63 (21.0)
30 (16.4)
2
14 (10.1)
37 (12.3)
21 (11.5)
18 (13.0)
29 (9.7)
24 (13.1)
Marital status, n (%)
Gravidity, n (%)
Parity, n (%)
3 Mean body mass index, kg/m standard deviationb 2
Mean no. of contraceptive methods used at baseline, n standard deviationb
26.8 6.9 N/A 1.6 0.8
1.0 1.0
27.3 6.3 1.6 0.9
Used barrier method at baseline, n standard deviationb
65 (47.1)
199 (66.3)
75 (41.0)
Pregnancy during study, n (%)
16 (11.6)
29 (9.7)
19 (10.4)
C31G, a spermicide and microbicide, not commercially available; FHI, Family Health International; N/A, not available. P < .01; b P < .001, reflects significant statistical difference between studies (Mann-Whitney U-test or chi-square test). Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
a
use vaginal spermicides. To accomplish this goal, we pooled data from clinical trials to yield estimates of a woman’s risk of pregnancy while using spermicides. We included both pregnancies that were characterized as user-failures and those that were characterized as methodfailures to increase the generalizability of our results.
Material and Methods The data came from 3 recent multicenter randomized controlled trials; the
methods and primary results of all the trials have been published elsewhere.7,11,12 Briefly, 2 studies compared the efficacy, safety, and acceptability of the spermicide under study with an active placebo,11,12 and 1 study compared 5 different doses/delivery systems of Nonoxynol-9.7 Women who were recruited for each study met the following inclusion criteria: they were healthy, sexually active, 18-42 years of age, had no history of infertility, no sexually transmitted disease diagnosis in previous 6
Original Research
months, had normal length menstrual cycles (defined as 24-35 days in 2 studies and 21-35 days in the third study), and had vaginal intercourse with regularity. The median coital act frequency was 2.3e3 times per week. Women in the trials were observed for at least 6 months or until a pregnancy occurred. From these 3 clinical trials, we extracted the local data that were collected prospectively at the University of Pennsylvania Center for Clinical Research in Women’s Health, created a new database, and analyzed these data to elucidate risk factors for pregnancy during these trials. This study was exempt from institutional review board approval, because it was a secondary data analysis of deidentified data. The overall failure rates for the 3 studies were similar (Table 1), which enabled us to pool the data. Testing for stratification by trial indicated that the studies were sufficiently homogenous to be combined. Each clinical trial collected detailed data on menstrual cycles, intercourse timing, and contraceptive methods that had been used (or not used) during each act of intercourse. Two trials reported that 97% of coital acts used spermicide alone7,11; in the third trial, all the women used spermicide in conjunction with a diaphragm.12 Pregnancies were each carefully dated by the Principal Investigator of each clinical trial who used information about last menstrual period, intercourse timing, and early pregnancy ultrasound scanning. We characterized incident pregnancies by variables that are potential risk factors for pregnancy such as age, race, gravidity/parity, and previous contraceptive use. Because the overall pregnancy rate was 10.3%, odds ratios were used to approximate incidence rate ratios for dichotomous risk factors, and Mann Whitney U tests were used to compare pregnant cases and control cases, with respect to continuous variables. We used statistical software (STATA; Stata Corporation, College Station, TX) to perform descriptive analyses; logistic regression was used to construct the optimal predictive model for incident pregnancy, with emphasis given to high sensitivity. The final model
FEBRUARY 2016 American Journal of Obstetrics & Gynecology
264.e2
Original Research
ajog.org
GYNECOLOGY
was simplified to generate a scoring algorithm.
Results A total of 621 women were enrolled in the 3 barrier method studies. The baseline demographics and clinical characteristics were similar among participants across the 3 studies, with statistical differences having no meaningful clinical difference (Table 1; MannWhitney U- or chi-square tests). The women had a mean age of 29 years; 43% of them were white, and 49% were black; 58% of them were nulliparous, and 38% of them were nulligravid. The overall pregnancy rate over 6 months was 10.3% (95% confidence interval [CI], 7.9e12.7). In unadjusted analyses (Table 2; Mann-Whitney Uor chi-square tests), the women who were of black race (P ¼ .046), obese (P ¼ .058), of higher gravidity (P < .001) or parity (P ¼ .026), and had unprotected sex more frequently (P ¼.026) were at higher risk for unintended pregnancy during the trial periods. The demographic and clinical characteristics were used to generate (with the use of logistic regression) the optimal predictive model with respect to sensitivity to contraceptive failure (Table 3). With the assumption of the use of the spermicide with 100% of coital acts, women who had been pregnant previously and were <35 years old were found to be at higher risk (>5%) for an unintended pregnancy; nulligravid women who were >35 years old were at low risk (<5%) for pregnancy (Figure 1). The area under the receiver operating characteristic curve for this model is 0.74. The multivariable model (Table 3) was then used to calculate a predicted probability of unintended pregnancy prevention or contraceptive efficacy for each woman; a woman was classified as low risk for unintended pregnancy if her probability of avoiding pregnancy with this model was >95%. This cutpoint was used to ensure that pregnancy prediction had >90% sensitivity (actual, 90.2%) because the cost of an unintended pregnancy is high. With the use of this model, only 6 of the unintended pregnancies (9.8%) were classified
TABLE 2
Demographic and clinical characteristics by pregnancy status during the study period Pregnancy during study period? Yes (n ¼ 64)
No (n ¼ 557)
<25 Y
20 (32.8)
165 (30.2)
25-<35 Y
36 (59.0)
291 (53.3)
5 (8.2)
90 (16.5)
White
19 (29.7)
247 (44.3)
Black
41 (64.1)
266 (47.8)
Other
4 (6.3)
44 (7.9)
1 (1.6)
27 (4.9)
Characteristic Age, n (%)
35 Y Race, n (%)
a
Education, n (%)
12 (19.8)
113 (20.4)
Some college
32 (50.0)
236 (42.7)
College graduate
19 (29.7)
177 (32.0)
<10
10 (28.6)
89 (21.1)
10-29
11 (31.4)
93 (32.5)
30-49
9 (25.7)
62 (21.7)
50
5 (14.3)
42 (14.7)
27 (62.8)
215 (56.0)
9 (20.9)
91 (23.7)
7 (16.3)
78 (20.3)
14 (43.8)
76 (27.6)
18 (56.3)
199 (72.4)
0
10 (16.4)
223 (40.8)
1
15 (24.6)
125 (22.9)
2
12 (19.7)
73 (13.4)
24 (39.3)
126 (23.0)
0
25 (41.0)
326 (59.6)
1
18 (29.5)
102 (18.7)
2
11 (18.0)
58 (10.6)
7 (11.5)
61 (11.2)
Household income ($1000), n (%)
Desire future children, n (%) Yes Maybe No Obese, n (%)
b
Yes No Gravidity, n (%)
c
3þ Parity, n (%)
a
3
Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
incorrectly in the current sample as low risk of pregnancy (false positives). The specificity was low at 42%, but the
264.e3 American Journal of Obstetrics & Gynecology FEBRUARY 2016
(continued)
negative predictive value was high, 97%. A total of 233 women (38%) were classified as low risk of unintended
ajog.org
GYNECOLOGY
TABLE 2
Demographic and clinical characteristics by pregnancy status during the study period (continued) Pregnancy during study period? Yes (n ¼ 64)
Characteristic
No (n ¼ 557)
Average cycle length, n (%) <25 D
1 (1.6)
11 (2.0)
25-31 D
60 (93.8)
509 (91.4)
>31 D
3 (4.7)
37 (6.6)
Percentage of unprotected sex during study, mean standard deviationa
12.1 18.9
5.7 10.9
P < .05; b Body mass index data available only on 307 subjects; c P < .001, reflects significant statistical difference between groups (Mann-Whitney U test or chi-square test). Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016. a
pregnancy. The ability of this model to discriminate among women who become pregnant from those who do not is measured by the area under the receiver-operator-characteristic curve, which was 0.737 (95% CI, 0.672, 0.802). A bootstrap validation13 was conducted to assess the model’s robustness. With a thousand bootstrap samples, the average area under the receiveroperator-characteristic curve was 0.75
with the empirical 95% CI [0.728, 0.814]. 10-fold cross-validation was performed and the predictive error rate from that method was 0.70 (95% CI [0.629e0.771]). Based on these results, a user-friendly scoring algorithm was generated to predict those women who were at low risk of becoming pregnant while using spermicides. The simplification was achieved by (1) recalculating the predictive score using values of the
TABLE 3
Final logistic regression model for spermicide failure (pregnancy) Variable
Odds ratio
95% Confidence interval
P value
Pointsa
Age, y <25
Reference
25-35
0.56
0.29e1.09
.089
2
35
0.19
0.06e0.58
.003
0
3
Gravity 0
Reference
1
3.38
1.43e7.99
.005
2
2
5.97
2.33e15.3
< .001
3
7.24
3.04e17.3
< .001
3
1.03
1.02e1.05
< .001
—
3 Unprotected sex acts, %
a
b
0
<10
0
10-29.9
2
30
4
This column includes the weighting or points used to calculate the simplified scoring system; b Continuous percentages were used for the logistic models, but discreet categories were used for the computation of the score; analyses were adjusted for race. Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
Original Research
coefficients, rounded to the nearest integer, and (2) by calculating the score simply as the sum of dichotomized “risk factors” that were identified by the model. For an individual woman, a risk score can be calculated easily by summing the points for each risk factor (Table 3; Figure 2). This simplified model performed comparably to the full model: the area under the receiveroperator-characteristic curve was 0.73 (95% CI, 0.669e0.793; Figure 1). The scores in Table 4 rank women with respect to their risk of contraceptive failure; the absolute risk of pregnancy, along with 95% CIs, is presented for each value. A woman who calculates her score to be “4” has twice the risk of pregnancy as a woman who scores “3.” Given these estimates of absolute risk and the test characteristics, a cutoff score of 4 was chosen to classify women as high risk for pregnancy. In the sample used to derive the rule, this threshold resulted in a total of 267 of the women (44%) classified as low risk, with 97% accuracy in the lowrisk prediction (negative predictive value); 8 of the women (9.8%) with unintended pregnancies were classified incorrectly as low risk. The 2 strongest influences on the probability of incidental pregnancy in our model were the number of previous pregnancies and age 35 years. When the method was used with only 50% of sex acts, the risk of pregnancy was also increased greatly (odds ratio, 5.36; 95% CI, 2.15e13.3), but occasional nonuse alone was not an important predictor of pregnancy.
Comment We conducted a secondary analysis of data combined from 3 different spermicide efficacy trials to build a prediction rule for successful method use. We used our findings to create a simple scoring system that women can incorporate into their contraceptive decisionmaking. We found that nulliparous women 35 years old are at low risk of pregnancy. We also found that, although perfect use was an important enhancer of contraceptive success, it was not more important than the biologic markers of fertility, such as age or gravidity. These data provide evidence that a woman’s
FEBRUARY 2016 American Journal of Obstetrics & Gynecology
264.e4
Original Research
ajog.org
GYNECOLOGY
FIGURE 1
Predicted method failure rates
These predicted method failure rates were based on age, the number of previous pregnancies, and the assumption of 100% protected sex. The numbers by each data point illustrate the calculated “score” with the use of our model. Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
personal characteristics and her behaviors are important in the facilitation of contraceptive success, as emphasized in a recent Cochrane review on spermicide use.14 There are women for whom the efficacy of spermicides is comparable with LARC, and for whom the decision to use spermicides would be medically sound, should this fit best with their values. The generalizability and validity of our results are supported by the fact that increasing age and nulliparity have been shown previously to be protective against pregnancy in the National Survey of Family Growth15 and other studies.16 Although older age is a well-known risk factor for decreased fertility, contraceptive trials enroll only women of reproductive age who are regularly sexually active and who are still at risk of pregnancy. Even within this fertile group that engages in regular sexual intercourse, age makes a difference. We found similar results among women who were enrolled in a microbicide trial: sexually active women in that study had linearly decreased risk of unintended pregnancy during the trial per each advancing year of age.17 An advantage of our method is that we were able to use prospective data efficiently to identify women who were most at risk of pregnancy while using
this class of contraceptives. Previous investigations of predictors of unintended pregnancy while using a contraceptive method have been limited by design challenges. Most existing data on risk factors for unintended pregnancy and contraceptive effectiveness are derived from the National Survey of Family Growth. This survey relies on recall of behaviors around the time of pregnancy and recall of the classification of the pregnancy as intended or unintended, (modified by socialdesirability bias18-20). Although this retrospective survey has produced
important estimates over decades, prospective data provide us with more accurate risk-estimates to develop a clinical tool to aid in self-assessment and contraceptive decision-making. In our study, all pregnancies were, by definition, unintended because all women who were enrolled in the spermicide clinical trials were screened rigorously for pregnancy avoidance before enrollment. In addition, the risk of misclassification bias is low because, within the parent clinical trials, menstrual cycles and pregnancy tests were monitored closely and prospectively. In
FIGURE 2
Decision aid for patient use when assessing pregnancy risk while using spermicides
Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
264.e5 American Journal of Obstetrics & Gynecology FEBRUARY 2016
ajog.org
GYNECOLOGY
TABLE 4
Absolute risk of pregnancy by risk assessment score Score
Frequency, n
Chance of pregnancy, %
95% Confidence interval, %
0
6
0
0.0e45.9
1
0
N/A
2
106
2.8
0.59e8.0
3
155
3.2
1.1e7.4
Risk Low
—
4
77
6.5
2.1e14.5
5
171
15.8
10.7e22.1
6
32
15.6
5.3e32.8
7
39
18.0
7.5e 33.5
8
22
40.9
20.7e63.6
High
N/A, not applicable. Schreiber et al. Contraceptive efficacy for spermicide users. Am J Obstet Gynecol 2016.
our secondary analysis, we included all pregnancies that were conceived during the trial period, which allowed us to capture documented user-failures in addition to method-failures, much like the real world typical use.
Limitations One limitation is the error that is inherent in our moderate sample size with regard to pregnancy events. However, the fact that all 3 studies had a pregnancy rate between 10% and 12% may lend validity to our overall point estimate. Additionally, the limited number of observed pregnancies limited our ability to perform cross-validation of the model and proposed cut-offs. We have presented error rates for misclassification using the same data that were used for model development. These error rates may over-state the performance of the proposed rule. External validation of the rule is highly recommended to obtain more accurate performance information before implementation. Another limitation is that the direction of biases that are associated with unprotected sex reporting is unknown. Study patients may have under-reported unprotected sex; in doing so without consequences (ie, pregnancy), the decision thresholds may be overestimates of risk. Alternatively, the women may have truly behaved more carefully because they were in a trial. However, the data
showed that user-failure (unprotected sex) still occurred and was a significant risk factor in the prediction of unintended pregnancy. Our use of clinical trial data may limit generalizability. Women who were willing to participate in clinical trials could be inherently different from women in the general population, but we believe these differences are likely to be minimal given that, before the Affordable Care Act, for many women, trial participation was the only way they could access contraception and basic gynecologic care, because of lack of insurance coverage. Because women who choose spermicides also are often un- or under-insured, we surmise that the differences between our study population and the women in the general population who may benefit from our findings are small. The clinical trial population included here reflects the sociodemographic factors that already are known to be associated with unintended pregnancy. The average age, socioeconomic, and marital statuses of the participants here are demographically similar to the very women for whom we need to improve our contraceptive counseling and access.21
Implications Metaanalysis indicates that decision aids improve informed patient choices and that decision aids have a positive effect on patient-practitioner
Original Research
communication.1 Providers have used scoring schemes successfully to assess contraceptive method risks22 and the need for Chlamydia trachomatis screening.23 A common reproductive health scoring scheme used by providers and patients is the Edinburgh Postnatal Depression Scale.24 To be useful, a prediction rule should be adaptable to the appropriate clinical setting.25 Our prediction rule was developed within a cohort of women that is representative of the population for whom it can be most useful; 50% of study subjects had used barrier methods before this study for contraception. We believe that, by focusing on concrete demographic and behavioral characteristics, we have built a rule that can be used easily by the women themselves. With the use of Table 4 and Figure 2, for example, a 30-year-old woman with 1 previous pregnancy (score, 4) may decide that a 6.5% risk of pregnancy is not only acceptable, given the advantages and cost of a spermicide, but also she can know that her risk of pregnancy will increase if she does not use protection consistently (add 2-4 points). Her 20-year-old friend with 2 children (score, 6) may decide that a 15% risk of pregnancy is too great, even if she uses the method perfectly. The utility of a prediction rule is dependent on its cost and on balancing the potential harm its introduction could cause. The rule presented here is free of cost and helps women to make personalized choices about fertility control without introducing harm. Recent publications from the prospective Contraceptive Choice Project have illustrated the important role that LARC methods play in increasing the proportion of US pregnancies that are intended.2,26 Although this landmark research provides us with the first independent longitudinal data of the longterm outcomes of hormonal and LARC contraceptive users, women who use barrier methods were not included in the Contraceptive Choice Project. The data we present here help to fill an important gap in the science of unintended pregnancy prevention and show that, for some women, spermicides can be highly effective as well. We are likely to have
FEBRUARY 2016 American Journal of Obstetrics & Gynecology
264.e6
Original Research
GYNECOLOGY
more sustained success in decreasing the unintended pregnancy rate in the United States if we work with patient-centered, individualized models. Although future studies could be done to test the efficacy of our self-assessment tool on contraceptive decision-making and ultimately unintended pregnancy prevention, its use is a low-risk, low-cost, individualized approach to help women avoid unintended pregnancy. n Acknowledgments We acknowledge Dr Kurt Barnhart (Department of Obstetrics and Gynecology, Perelman School of Medicine, University of Pennsylvania) for his helpful editing of this manuscript and thank Family Health International and the National institute of Child Health and Human Development for allowing us access to the data that were used in this study.
References 1. Stacey D, Bennett CL, Barry MJ, et al. Decision aids for people facing health treatment or screening decisions. Cochrane Database Syst Rev 2011;10:CD001431. 2. Peipert JF, Madden T, Allsworth JE, Secura GM. Preventing unintended pregnancies by providing no-cost contraception. Obstet Gynecol 2012;120:1291-7. 3. Dehlendorf C, Diedrich J, Drey E, Postone A, Steinauer J. Preferences for decision-making about contraception and general health care among reproductive age women at an abortion clinic. Patient Educ Couns 2010;81:343-8. 4. Dehlendorf C, Levy K, Kelley A, Grumbach K, Steinauer J. Women’s preferences for contraceptive counseling and decision making. Contraception 2013;88:250-6. 5. Foster DG, Grossman D, Turok DK, et al. Interest in and experience with IUD self-removal. Contraception 2014;90:54-9. 6. Jones J, Mosher W, Daniels K. Current contraceptive use in the United States, 2006e2010, and changes in patterns of use since 1995.
National health statistics reports; no 60. Hyattsville (MD): National Center for Health Statistics; 2012. 7. Raymond EG, Chen PL, Luoto J; Spermicide Trial Group. Contraceptive effectiveness and safety of five nonoxynol-9 spermicides: a randomized trial. Obstet Gynecol 2004;103: 430-9. 8. Schreiber CA, Meyn LA, Creinin MD, Barnhart KT, Hillier SL. Effects of long-term use of nonoxynol-9 on vaginal flora. Obstet Gynecol 2006;107:136-43. 9. Raymond EG, Trussell J, Weaver MA, Reeves MF. Estimating contraceptive efficacy: the case of spermicides. Contraception 2013;87:134-7. 10. American College of Obstetricians and Gynecologists. Increasing use of contraceptive implants and intrauterine devices to reduce unintended pregnancy: ACOG Committee Opinion, no. 450. Obstet Gynecol 2009;114: 1434-8. 11. Burke AE, Barnhart K, Jensen JT, et al. A randomized trial of the contraceptive efficacy, acceptability, and safety of C31G and nonoxynol-9 spermicidal gels. Obstet Gynecol 2010;116:1265-73. 12. Barnhart KT, Rosenberg MJ, MacKay HT, et al. Contraceptive efficacy of a novel spermicidal microbicide used with a diaphragm: a randomized controlled trial. Obstet Gynecol 2007;110:577-86. 13. Efron B, Tibshirani R. Improvements on cross-validation: the .632þ bootstrap method. J Am Stat Assoc 1997;92:548-60. 14. Grimes DA, Lopez LM, Raymond EG, Halpern V, Nanda K, Schulz KF. Spermicide used alone for contraception. Cochrane Database Syst Rev 2013;12:CD005218. 15. Finer LB, Zolna MR. Unintended pregnancy in the United States: incidence and disparities, 2006. Contraception 2011;84:478-85. 16. Menken J, Trussell J, Larsen U. Age and fertility. Science 1986;233:1389-94. 17. Schreiber CA, Whittington S, Cen L, Maslankowski L. Good intentions: risk factors for unintended pregnancies in the US cohort of a microbicide trial. Contraception 2011;83: 74-81.
264.e7 American Journal of Obstetrics & Gynecology FEBRUARY 2016
ajog.org 18. Moos MK, Petersen R, Meadows K, Melvin CL, Spitz AM. Pregnant women’s perspectives on intendedness of pregnancy. Womens Health Issues 1997;7:385-92. 19. Santelli J, Rochat R, Hatfield-Timajchy K, et al. The measurement and meaning of unintended pregnancy. Perspect Sex Reprod Health 2003;35:94-101. 20. Petersen R, Moos MK. Defining and measuring unintended pregnancy: issues and concerns. Womens Health Issues 1997;7: 234-40. 21. Finer LB, Henshaw SK. Disparities in rates of unintended pregnancy in the United States, 1994 and 2001. Perspect Sex Reprod Health 2006;38:90-6. 22. Spellacy WN, Kerr DA, White B, Keith GE, Waggoner BL. A family planning risk scoring system for health care providers. Obstet Gynecol 1984;63:846-9. 23. Lavoué V, Morcel K, Voltzenlogel MC, et al. Scoring system avoids Chlamydia trachomatis overscreening in women seeking surgical abortions. Sex Transm Dis 2014;41: 470-4. 24. Cox JL, Holden JM, Sagovsky R. Detection of postnatal depression: development of the 10-item Edinburgh Postnatal Depression Scale. Br J Psychiatry 1987;150:782-6. 25. Grady D, Berkowitz SA. Why is a good clinical prediction rule so hard to find? Arch Intern Med 2011;171:1701-2. 26. Winner B, Peipert JF, Zhao Q, et al. Effectiveness of long-acting reversible contraception. N Engl J Med 2012;366:1998-2007.
Author and article information From the Penn Family Planning and Pregnancy Loss Center (Drs Schreiber and Whittaker) and the Department of Biostatistics and Epidemiology, Center for Clinical Epidemiology and Biostatistics (Drs Ratcliffe, Sammel, and Whittaker), Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA. Received March 15, 2015; revised Oct. 14, 2015; accepted Oct. 18, 2015. The authors report no conflict of interest. Corresponding author: Courtney A. Schreiber, MD, MPH. [email protected]