Hormonal contraceptives for contraception in overweight or obese women

Types of studies

We included studies of hormonal contraceptive effectiveness among overweight or obese women. Reports had to identify the specific contraceptive methods used. Because we did not anticipate finding randomized controlled trials stratified by body weight, we included all study designs. The main comparisons were overweight or obese women versus women of lower BMI or weight. Therefore, the comparisons of interest were possible in single‐arm studies, i.e. those with only one intervention. All languages of publication were eligible for inclusion.

We eliminated studies focused on women with specific health problems, such as HIV or diabetes. We also excluded studies of contraceptives as treatment for specific disorders, e.g. acne, hirsutism, or polycystic ovary syndrome.

Types of participants

Participants were the women in the studies who used the hormonal contraceptive for contraception. Overweight or obese women must have been identified by an analysis cutoff for BMI or weight. The comparison group could have been women in a lower BMI or weight group. As noted earlier, several criteria for defining overweight or obese are still used, including a BMI (kg/m²) ≥ 25 for overweight and ≥ 30 for obesity (CDC 2016) as well as the NHANES II cutoffs (> 27.3 for overweight and > 32 for obesity) (Burkman 2009). Some investigators analyzed the outcome by body weight quartiles, deciles or groups, e.g. < 70 kg or ≥ 70 kg. We included studies with differing criteria, as practices differ across time periods and by country and we initially anticipated finding few studies. The report had to provide the BMI or weight cutoff points.

Types of interventions

The study must have examined the use of one or more hormonal contraceptives. The focus could have been any hormonal contraceptive, i.e. an oral contraceptive, a transdermal skin patch, a vaginal ring, an injectable contraceptive, a subdermal implant, or hormonal intrauterine contraception. Treatment duration must have been at least three cycles or three months.

Types of outcome measures

Electronic searches

Until 4 August, we searched PubMed (MEDLINE), the Cochrane Central Register of Controlled Trials (CENTRAL), POPLINE, and Web of Science. We also searched for current trials through ClinicalTrials.gov and ICTRP. Appendix 1 shows the recent search strategies and Appendix 2 contains the earlier strategies.

Searching other resources

We examined reference lists of relevant articles to identify additional studies. For the initial review, we contacted investigators in the field to seek additional unpublished trials or published studies.

Selection of studies

We assessed for inclusion all titles and abstracts identified during the literature searches with no language limitations. One author reviewed the search results and identified reports for inclusion or exclusion. A second author examined the reports for appropriate categorization.

Studies could have been randomized controlled trials (RCTs) or non‐randomized studies (NRS), e.g. prospective single‐arm or multi‐arm studies, case‐control studies, or observational studies of contraceptive users. We considered post hoc analysis from these types of studies as long as the analysis met the eligibility criteria.

Data extraction and management

One author extracted the data and entered the information into RevMan. Another author conducted a second data extraction and verified correct data entry. We resolved any discrepancies by discussion or with a third author if necessary.

Assessment of risk of bias in included studies

We examined RCTs for methodological quality, according to recommended principles (Higgins 2011). The randomization was unrelated to the BMI or weight comparisons, but provides an indicator of study quality. We considered randomization method, allocation concealment, blinding, and losses to follow‐up and early discontinuation.

For NRS, we used the Newcastle‐Ottawa Quality Assessment Scale (NOS) (Higgins 2011; Wells 2014). Of the two NOS versions for case‐control and cohort studies, the latter was more pertinent here (Appendix 3). The NOS investigators are examining the criterion validity and construct validity of this scale as well as the inter‐rater and intra‐rater reliability. The scale does not yet have an overall scoring or threshold for 'good' or 'poor' quality. The NOS has eight items within three domains: selection (representativeness), comparability (due to design or analysis), and outcomes (assessment and follow‐up). A study can receive one star (✸) for meeting each criterion. The exception is comparability, which can have two stars (for design and analysis). We adapted NOS items for this project as suggested by the developers (Wells 2014).

We recorded whether pregnancies and body weight were measured or self‐reported. Pregnancies may be under‐reported when relying on self reports from interviews or questionnaires rather than testing. Such under‐reporting is unlikely to differ by BMI or weight group. However, body weight is frequently underestimated by a few pounds (Holt 2005). The result would be categorizing more women in a lower weight group, which would bias the effect estimate toward no difference.

Measures of treatment effect

The main comparisons for this review were between overweight or obese women and women of lower BMI or weight. The comparisons were possible in studies having only one intervention. For example, for the primary outcome of pregnancy, we compared pregnancies among 'overweight' women with those of 'normal' or 'healthy' weight women who used the specific contraceptive. Definitions of overweight and normal, or the cutoffs for BMI or weight, depended on the analytic methods used for the study reports. For two‐arm studies, we compared the BMI or weight groups within each contraceptive method group.

Oral contraceptive studies tend to have relatively high discontinuation rates. Time‐to‐event measures such as life‐table or incidence rates are commonly used, as they are based on actual exposure to the contraceptive and prevent an imbalance in discontinuations from distorting the comparisons. We extracted life‐table rates (actuarial or continuous) where available. We also used unadjusted pregnancy rates, relative risk (RR), or rate ratio when those were the only results reported by the investigator.

With non‐randomized studies, investigators need to control for confounding factors. When available, we used adjusted measures that the investigators reported. The effect measure may have been hazard ratio, rate ratio, or relative risk. Where only the crude number of events was available for dichotomous outcomes, we computed the Mantel‐Haenszel odds ratio (OR) with 95% confidence interval (CI). An example is the proportion of women that reported bleeding or spotting problems.

Dealing with missing data

We wrote to the study investigators for missing data and clarification of issues related to participants and methods. Responses and any data provided are in Characteristics of included studies. We limited our requests to studies less than 10 years old. Investigators are unlikely to have access to information from older studies.

Assessment of heterogeneity

We expected study populations, designs, and interventions to be heterogeneous. We described the clinical and methodological diversity (or heterogeneity) of the studies. We did not pool data from studies that had different contraceptive methods (e.g. COC and transdermal patch), different doses of the same method, or different criteria for analyzing BMI or body weight. Therefore, we did not conduct meta‐analysis. Statistical heterogeneity is not an issue when a comparison has a single study.

Data synthesis

To assess the quality of evidence and address confidence in the effect estimates, we applied principles from GRADE (Grades of Recommendation, Assessment, Development and Evaluation) (Higgins 2011; GRADE 2013). If meta‐analysis is not viable because of varied interventions or outcome measures, a typical 'Summary of findings' table is not feasible. Also, the criteria for quality assessment differ for non‐randomized versus randomized comparisons. We do provide 'Summary of findings' tables for the main results, although we did not conduct a formal GRADE assessment for all outcomes (GRADE 2013).

We based our assessment of the body of evidence on the quality of evidence from the studies. In 2016, we revised the Risk of bias tables to accommodate the criteria for both RCTs and NRS. We used the Newcastle‐Ottawa Quality Assessment Scale for all studies (Appendix 3). The comparisons were non‐randomized, since participants could not be randomized by BMI or weight, though the sample could have been stratified. Evidence quality included the design, implementation, and reporting of the study. We examined the evidence for the primary outcome of pregnancy. We list the criteria for downgrading below.

1. BMI not analyzed

2. NRS: high risk of bias in selection (NOS)

3. NRS: not controlling for relevant confounding in design or analysis, e.g. age, education, prior contraceptive use

4. Follow‐up less than 12 months for pregnancy outcome

5. Loss to follow‐up > 20%, combined loss to follow‐up and discontinuation > 50%, or retrospective chart review of selected cases

Sensitivity analysis

In the initial review in 2010, we synthesized results from the studies that had confirmed pregnancies and measured body weight (not self‐reported). In 2016, we conducted the quality assessment using the NOS in lieu of the earlier sensitivity analysis.

Results of the search

The 2012 searches produced 421 references: 395 from the database searches, 8 from other sources such as reference lists, and 18 trials from searches of clinical trials sites. We included two new studies (with four reports) (Westhoff 2012a; Xu 2012). We excluded six (with seven reports) after examining the full text (Excluded studies). A conference presentation was awaiting classification due to limited data; we included the full report in 2016 (Kaunitz 2014). After reviewing the abstracts, we retained 22 references for background information and discarded 366 for not meeting the eligibility criteria.

In 2016, the database searches yielded 681 references. After removing 280 duplicates, we had 401 unduplicated references (Figure 1). Through other searching, we identified four articles with relevant outcome and design information from three older studies. With those 4 articles plus 4 reports from other sources, the new total was 409 unduplicated references. After discarding 393 based on title or abstract, we examined the full text of 16 articles and then excluded two studies. The remaining 14 articles represented 8 new studies. Searches of ClinicalTrials.gov and ICTRP produced 26 unduplicated listings of clinical trials. No ongoing trial appeared to meet our eligibility criteria.

Included studies

With the 8 studies added in this update, 17 studies met our inclusion criteria and had a total of 63,813 women. The median sample size was 1736. The hormonal contraceptives that they studied varied.

Six studies examined COCs.

• Burkman 2009

  • norgestimate 180/215/250 µg + ethinyl estradiol (EE) 25 µg

  • norethindrone acetate (NETA) 1 mg + EE 20 µg

• Mansour 2012

  • nomegestrol acetate 2.5 mg + 17β‐estradiol (E₂) 1.5 mg

  • drospirenone 3 mg + EE 30 μg

• Westhoff 2012a: levonorgestrel 100 µg + EE 20 µg (84 days)/ EE 10 µg (7 days)

• Kaunitz 2014: levonorgestrel 100 µg + EE 20 µg

• Nakajima 2016: norethindrone acetate 1 mg + EE 10 µg

• Yamazaki 2015 (data from seven trials)

  • 3 norgestimate (NGM) formulations combined in analysis (NGM 180/215/250 µg + EE 25 µg; NGM 180/250 µg + EE 25 µg; NGM 60/180 µg + EE 20 µg)

  • norethindrone acetate 1 mg + EE 10 µg/ EE 10 µg

  • norethindrone 800 µg + EE 25 µg

  • levonorgestrel 90 µg + EE 20 µg

  • levonorgestrel 100 µg + EE 20/10 µg

  • levonorgestrel 150 µg + EE 20/25/30/ 10 µg

  • desogestrel 150 µg + EE 20/10 µg

Two studies examined different transdermal patches. One analyzed data from three trials (Zieman 2002); the other had a COC comparison group listed above (Kaunitz 2014).

• Zieman 2002: norelgestromin 150 µg + EE 20 µg daily release (data from three studies)

• Kaunitz 2014; levonorgestrel 120 µg + EE 30 µg daily release (experimental; not yet marketed)

Of seven implant studies, six examined levonorgestrel implants; one of which included two different types (Sivin 1998b). The seventh study examined the etonogestrel implant.

• levonorgestrel 6 capsules, 216 mg (Grubb 1995; Gu 1995; Sivin 1998b; Sivin 1998c)

• levonorgestrel 2 rods, 150 mg (Sivin 1997; Sivin 1998a; Sivin 1998b)

• etonogestrel 1 rod, 68 mg (Xu 2012)

Three studies examined other methods.

• Jain 2004: depot medroxyprogesterone acetate, subcutaneous formulation (DMPA‐SC) containing 104/0.65 mL

• Gemzell‐Danielson 2015: levonorgestrel intrauterine contraception (LNG‐IUC), with one releasing 8 µg/day (LNG‐IUS 8) and one releasing 13 µg/day (LNG‐IUS 13)

• WHO 1990: experimental vaginal ring (not marketed); levonorgestrel (LNG) 20 µg daily release, intended for 90‐day use

The treatment duration varied. Four studies lasted one year or 13 cycles (WHO 1990; Jain 2004; Westhoff 2012a; Nakajima 2016) and four had durations of 6 and 13 cycles within the same study or analysis (Zieman 2002; Burkman 2009; Kaunitz 2014; Yamazaki 2015). The IUC study lasted three years (Gemzell‐Danielson 2015). Implant studies had durations of three years (Xu 2012), five years (Grubb 1995; Sivin 1998a; Sivin 1998b; Sivin 1998c), and seven years (Gu 1995; Sivin 1997).

Seven reports used data from RCTs (Sivin 1997; Sivin 1998b; Zieman 2002; Burkman 2009; Kaunitz 2014; Gemzell‐Danielson 2015; Yamazaki 2015). The main comparisons in those studies were between contraceptive methods, and the randomization was not stratified by weight. Zieman 2002 pooled data from two RCTs and one uncontrolled study. Yamazaki 2015 was a meta‐analysis of individual participant data from seven trials. The remaining reports were from prospective non‐comparative trials.

Nine studies used BMI cutoffs for overweight or obesity (Jain 2004; Burkman 2009;Westhoff 2012a; Mansour 2012; Xu 2012; Kaunitz 2014; Gemzell‐Danielson 2015; Nakajima 2016; Yamazaki 2015). Burkman 2009 and Westhoff 2012a also used a body weight dichotomy at 70 kg and deciles for BMI and weight. The results for cycle control under Burkman 2009 came from an earlier report in which the investigators used quartiles of body weight (lb). Zieman 2002 presented results by body weight deciles and a dichotomy at 90 kg. Four older studies used body weight groups of 10 kg each (WHO 1990; Grubb 1995; Gu 1995; Sivin 1997). Three older studies reported body weights of women who became pregnant (Sivin 1998a; Sivin 1998b; Sivin 1998c).

Excluded studies

We excluded 14 studies. Common reasons were no pregnancy outcome data or insufficient analysis by BMI or weight group. Studies of emergency contraception did not have sufficient treatment duration. Details are in Characteristics of excluded studies.

Allocation

For the RCTs, we extracted information on how the randomization sequence was generated and the allocation concealment. The randomization method indicates overall study quality, but is unrelated to our comparisons of interest.

Blinding

Blinding was not applicable for single‐arm studies (WHO 1990; Grubb 1995; Gu 1995; Sivin 1998a; Sivin 1998c; Jain 2004; Westhoff 2012a; Nakajima 2016). Nor was it feasible in some studies due to visible differences in the contraceptive methods (Sivin 1998b; Xu 2012; Kaunitz 2014). In Gemzell‐Danielson 2015, blinding investigators was not feasible due to differences in the devices but participants were blinded to device type.

Of the remaining studies, Sivin 1997 was blinded. Burkman 2009 had blinding of the NGM arm; the trial originally had three NGM arms. The trials included in Zieman 2002 were open label, as was Mansour 2012. Yamazaki 2015 did not mention blinding for the included trials.

Incomplete outcome data

Three reports had some evidence of incomplete outcome data. The investigators in WHO 1990 dropped women from the study if they had three expulsions in one week or more than five within four weeks. Burkman 2009 excluded cycles with incorrect pill intake as well as cycles lacking data on dosing and bleeding. Mansour 2012 excluded those who did not take any study drug.

Three studies had loss to follow‐up greater than 20% or combined loss to follow‐up and discontinuation greater than 50%. High losses to follow‐up threaten validity (Strauss 2005).

• Sivin 1998b: discontinuation by 5 years, 55% 2‐rod and 60% Norplant; loss to follow‐up, 14% 2‐rod and 18% Norplant

• Sivin 1998c: loss to follow‐up 42% by 5 years

• Xu 2012: loss to follow‐up 23% by 3 years

Yamazaki 2015 did not have information on losses in the seven trials included in those analyses.

Other potential sources of bias

Combined oral contraceptives

Five trials examined one or two COC formulations each (Burkman 2009; Mansour 2012; Westhoff 2012a; Kaunitz 2014; Nakajima 2016). We grouped the results by type of progestin. A sixth report was a meta‐analysis of individual participant data from seven trials (Yamazaki 2015). We present the results separately. Although some of the COC formulations overlap with some in the review, others are not used in practice. Also, the analysis combined three norgestimate formulations from one trial.

Burkman 2009 studied norgestimate (NGM) 180/215/250 µg + EE 25 µg (N = 1671) as well as a COC with norethindrone acetate (below). For NGM, pregnancy risk did not differ by body weight or BMI groups (Analysis 1.1). The report stated that women with a BMI greater than 32.4 were excluded from participation. However, that may have been an error, as the report also stated the highest BMI was 47.6. An earlier publication reported on cycle control (Hampton 2008). Of eight comparisons for breakthrough bleeding or spotting, three showed a significant difference. Women above the 75th percentile for weight (more than 155 lb) were less likely to report breakthrough bleeding or spotting at cycle 13 than those below the 25th percentile (123 lb or less) (Analysis 1.4) or those in the 25th to 75th percentiles (124 to 155 lb) (Analysis 1.5). Burkman 2009 had adherence data from daily diaries, but the investigators reported the results by COC not BMI group.

Estrogen dose differed for two trials of a COC containing norethindrone acetate (NETA). With NETA 1 mg + EE 20 µg (Burkman 2009) (N = 1139), women with a BMI of 25 or more had a higher pregnancy risk compared with those who had a BMI less than 25 (reported relative risk [RR] 2.49, 95% CI 1.01 to 6.13) (Analysis 2.1). Pregnancy risk did not differ significantly between groups when using a BMI cutoff of 27.3. Women above the 75th percentile (more than 155 lb) were less likely to report breakthrough bleeding or spotting at cycle 6 than the women below the 25th percentile (123 lb or less) (Analysis 2.2). The other trial used NETA 1 mg + EE 10 µg (Nakajima 2016) (N = 1581). According to the Pearl Index, obese women (BMI > 30) did not have decreased efficacy compared with women in the lower BMI groups (Analysis 3.1). For those aged 18 to 35 years old, the Pearl Index was 2.50 versus 3.04 for the overweight group (BMI 25 to 30) and 3.06 for those with a lower BMI (< 25). The pattern was similar in the analysis of those aged 18 to 45 years. The adherence data show that obese women took an average of 25.8 pills per cycle compared with 26.2 pills each of the groups with a lower BMI (Analysis 3.2). Bleeding data came from participant diaries. For unscheduled bleeding or spotting in cycle 13, the mean number of days was 2.2 for obese women, 1.9 for the overweight group, and 1.6 for those with a lower BMI (Analysis 3.3). Some of the more common treatment‐related adverse events were more common among obese women, i.e. headache, upper respiratory tract infection, nasopharyngitis, and nausea (Analysis 3.4).

Two trials examined LNG 100 µg + EE 20 µg in different regimens. With an extended regimen (84 days) (Westhoff 2012a ) (N = 1736), pregnancy rates did not differ significantly by BMI groups of less than 25 versus 25 or greater (Analysis 4.1; Analysis 4.2). Similarly, with 70 kg as the cutoff for weight, pregnancy rates did not differ significantly (Analysis 4.3; Analysis 4.4). The investigators also examined deciles for weight and BMI and found no trends in the crude pregnancy rates. Kaunitz 2014 used a standard regimen of LNG 100 µg + EE 20 µg (N = 1129) as the comparison in a trial of a transdermal patch releasing levonorgestrel. Results for the patch are in the relevant section below. With the LNG COC, the reported Pearl Index for nonobese women was 5.59 (95% CI 0.70 to 10.47) compared with zero for obese women (Analysis 5.1). For the subgroup of women who were "treatment‐compliant," the investigators found a similar pattern. The nonobese women had a Pearl Index of 5.26 (95% CI 0.12 to 10.41) versus zero for the obese women (Analysis 5.2). The BMI groups did not differ significantly in adherence to the treatment regimens, according to self‐reported data and plasma levels of LNG and EE (Analysis 5.3).

Mansour 2012 examined pregnancies by BMI group with nomegestrol acetate 2.5 mg + 17β‐estradiol (E₂) 1.5 mg (N = 1613) and with drospirenone 3 mg + EE 30 µg (N = 539). The investigators used four BMI groups: < 18.5; 18.5 to < 25; 25 to < 30; ≥ 30. Four pregnancies occurred with nomegestrol acetate + E₂ (Analysis 6.1) and three with drospirenone + EE (Analysis 7.1). All pregnancies were in women with BMI 18.5 to < 25, a group considered not overweight. That normal weight group included 73% (1556/2121) of the women. The conference abstract did not include the distribution by BMI within treatment group. The investigators reported that BMI was not associated with pregnancy but did not provide any statistical test.

Transdermal contraceptive patch

Zieman 2002 studied the patch containing norelgestromin 150 µg + EE 20 µg (N = 3318). The investigators stated that baseline body weight was associated with pregnancy risk in a proportional hazards model, which included age, race, BMI, and body surface area (reported P < 0.001). Pooled data from three trials showed 15 pregnancies over one year in 3319 women (Analysis 15.1). The top three deciles of women weighed 69 kg or more and were about 30% of the sample. Of the 15 pregnancies, 7 were in the top decile (80 kg or more). Five of the seven were among those weighing 90 kg (198 lb) or more, who constituted 3% of the study population. Reportedly, pregnancies were not clustered in any BMI subgroup (no data provided).

Kaunitz 2014 examined an experimental patch releasing levonorgestrel 120 µg + EE 30 µg (N = 1129). The comparison was an LNG COC as noted above. For obese women, the Pearl Index was 4.58 (95% CI 0.57 to 8.59) compared with 4.40 (95% CI 1.92 to 6.89) for nonobese women (Analysis 16.3). Of those who were "treatment‐compliant," the Pearl Index for obese women was higher than that for nonobese women, i.e. 4.63 (95% CI 0.10 to 9.17) versus 2.15 (95% CI 0.27 to 4.04), respectively (Analysis 16.4). The BMI groups did not differ significantly for adherence to the treatment regimens, according to self‐reported data and plasma levels of LNG and EE (Analysis 16.5).

Depot medroxyprogesterone acetate, subcutaneous (DMPA‐SC)

With data from two trials, Jain 2004 examined a subcutaneous form of depot medroxyprogesterone acetate (DMPA‐SC) containing 104 mg/0.65 mL. No pregnancies occurred in one year (Analysis 17.1). Overweight or obese women were 27% of the sample in the European and Asian trial, but only 6% had a BMI greater than 30. In contrast, 44% of the women in the 'Americas' trial were overweight or obese with nearly 18% having a BMI greater than 30.

Levonorgestrel‐releasing intrauterine contraception (LNG‐IUC)

Gemzell‐Danielson 2015 was a subanalysis from a trial of low‐dose LNG‐IUC. One method released 8 µg/day (LNG‐IUS 8) (N = 1432) and the other 13 µg/day (LNG‐IUS 13) (N = 1452). Within both LNG‐IUC groups, the Pearl Index was similar for nonobese (BMI < 30) and obese (BMI ≥ 30) women at year 1 and year 3 (Analysis 18.1; Analysis 18.2). Based on the 95% CI for each BMI group, the investigators concluded that BMI was not significantly associated with the cumulative failure rates in either LNG‐IUC group at one year or three years (Analysis 18.3; Analysis 18.4). The report did not include a statistical test for the comparisons.

Implants

Progestin‐only vaginal ring

An experimental ring with levonorgestrel 5 mg was never marketed (WHO 1990) (N = 1005). The group weighing 70 kg or more had a cumulative discontinuation rate for pregnancy of 8.2 (Analysis 22.1), and represented about 10% of the sample. Rates for the other weight groups were 1.8 (≤ 49 kg), 2.6 (50 to 59 kg), and 5.3 (60 to 69 kg). Weight was associated with pregnancy (reported P = 0.0013). Cox proportional hazards indicated pregnancy risk increased by 61% with a 10 kg increase in body weight. The risk of pregnancy more than doubled with a 20 kg increase (e.g. from 60 to 80 kg). The estimated hazards ratio was 2.60 for this increase in body weight, but the report did not provide information on statistical significance. The cumulative life‐table rate for a woman weighing 80 kg was 9.8, i.e. more than twice the 4.4 rate for a woman weighing 60 kg (Analysis 22.2).