We included studies that examined the timing of starting or resuming hormonal contraception after UPA use. The study population of interest was women of reproductive age (e.g., 15–49 years, but the specific age range varied based on the criteria used in individual studies). The exposure of interest was initiation or resumption (e.g., after missing pills) of hormonal contraception immediately or soon after UPA use. Hormonal contraception methods included: combined oral contraceptives (COCs), combined injectable contraceptives, contraceptive patch, contraceptive ring, progestin-only pills (POPs), progestin-only injectables, progestin-only implants, and LNG-releasing intrauterine devices (IUDs). The comparator was delayed initiation or resumption of hormonal contraception or no use of hormonal contraception after UPA use. The primary outcomes were 1) effectiveness of UPA and 2) effectiveness of regular hormonal contraception in preventing pregnancy, as measured by pregnancy rates and proxy measures for risk of pregnancy, such as ovulation, follicular rupture, ovarian activity, and hormone levels (e.g., progesterone, estradiol, and luteinizing hormone). Secondary outcomes included adverse events, side effects such as vaginal bleeding, and patient satisfaction. Eligible study designs included randomized clinical trials (RCTs), non-randomized trials, cohort studies (comparative and non-comparative), and case-control studies. We excluded cross-sectional studies, case reports, commentaries, and reviews.

We worked with a Centers for Disease Control and Prevention (CDC) research librarian to review the research question and develop a comprehensive search strategy to identify articles relevant to the review question. The search was conducted in MEDLINE, EMBASE, Cochrane, ClinicalTrials.gov, and CINAHL databases from database inception to December 15, 2022 (Online Appendix A). We also reviewed references in systematic reviews to optimize our search. Articles in all languages were considered for inclusion.

Two authors independently screened each title and abstract identified in the search to identify full-text articles for review using Covidence software [13]. Two authors then independently reviewed full-text articles to identify articles that met inclusion criteria. Any conflicts identified in the screening process were resolved through discussion among authors. We scanned reference lists of included articles and relevant review articles to identify any additional studies. We used a standardized evidence table to record study information, including authors, publication year, study location, years of data collection, study design and objective, study population, intervention, outcomes, measurement/definition of ovulation, results, and study strengths and limitations into the evidence table. The lead author extracted data from each study and a co-author performed quality control checks of the extracted data. Discrepancies were resolved through discussion among authors.

We used the Cochrane approach to assess risk of bias for RCTs and evaluated the following domains: randomization processes, deviations from intended interventions, missing outcome data, measurement of outcomes, and selection of reported results [14]. We used a modified version of the Cochrane approach to assess risk of bias in non-randomized studies, including assessment of selection bias, information bias (exposure and outcome), and confounding [15]. Two authors independently assessed the risk of bias of each included study and any conflicts were resolved through discussion among authors.

We summarized descriptive statistics, unadjusted and adjusted point estimates, results of significance tests, and interval estimates. We did not conduct meta-analyses because of the limited number of studies and the heterogeneity of exposures, outcomes, and study populations across the included studies.

We used the GRADE approach to assess the certainty of evidence [16–18]. Certainty of evidence was rated as high, moderate, low, or very low depending on criteria including risk of bias, indirectness, imprecision, and inconsistency. Initially, RCTs are considered to have high certainty and non-randomized trials and observational studies have low certainty; these ratings are adjusted according to the previously mentioned criteria. We evaluated certainty of evidence for the primary outcome of effectiveness (measured via pregnancy rates and proxy measures for risk of pregnancy, such as ovulation) and secondary outcomes of adverse events and side effects such as vaginal bleeding for the studies that reported these outcomes. Patient satisfaction was not reported by included studies and therefore not assessed as an outcome.

The risk of bias among the four included studies was low (Table 2). The three RCTs had low risk of bias because they had adequate randomization processes, no deviations from intended outcomes or missing outcome data, used appropriate measurements, and reported all pre-specified outcomes [7, 19, 21]. One RCT did not blind outcome assessors but assessment of the outcomes (hormone levels and follicle size) was not likely to be affected [21]. The non-randomized trial also had low risk of bias because participants served as their own controls which minimized selection bias, exposures and outcomes were measured appropriately, and participants served as their own controls to minimize confounding [20].

Three of the four studies examined initiation of hormonal contraception after UPA administration in the study cycle. Of these, two studies assessed the effect of UPA on the ability of hormonal contraception to suppress ovulation [7, 19]. An RCT enrolled 76 healthy women, ages 18–35 years, with body mass index (BMI) < 30 kg/m², and who had regular, normal menstrual cycles or were current COC users before the study cycle [19]. During the study cycle, each participant was followed until a lead follicle of > 13 mm in mean diameter was observed and then randomized to receive either UPA or placebo. All participants started 21 days of COCs (EE 30 μg and LNG 150 μg) on the day following UPA or placebo administration. Ovarian activity was measured using a Hoogland score [22], which includes follicle growth measured through transvaginal ultrasound and serum concentrations of estradiol and progesterone, with Hoogland scores of 1–3 indicating ovarian quiescence, a score of 4 indicating active follicle like structure, a score of 5 indicating luteinized unruptured follicle, and a score of 6 indicating ovulation. Results indicated no significant differences overall in ovarian activity between the UPA and placebo groups after 21 days of COC use (p=0.95). No significant differences in the proportions in each group who ovulated (33% in the UPA group and 32% in the placebo group) nor in the proportions who reached ovarian quiescence (62% in both groups) were observed, suggesting that UPA did not have an effect on the ability of the COC to inhibit ovulation. When separately assessing ovulatory activity among those who were initiating COCs as part of the study and those who had been current users of COCs (i.e., had used COCs in the cycle previous to the study cycle), no significant differences were observed, although the proportion reaching quiescence was higher among those previously using COCs (66%) than those initiating COCs (52%). Five women in the UPA group and eight in the placebo group experienced unscheduled vaginal bleeding during COC use; duration was similar in both groups. No serious adverse events were reported. Another RCT examined the pharmacokinetic and pharmacodynamic interactions between UPA and a POP (DSG 75 μg) [7]. This study enrolled 49 healthy women ages 18–35 years with BMI < 30 kg/m², who had not been using hormonal contraception for at least one complete menstrual cycle. Participants were randomized to two of three treatment sequences separated by a washout cycle: 1) UPA followed the next day by POP (DSG) for 20 days, 2) UPA followed the next day by placebo for 20 days, and 3) placebo followed the next day by POP (DSG) for 20 days. Treatment or placebo was initiated in each treatment sequence when a lead follicle was ≥ 14 to < 16 mm diameter on transvaginal ultrasound. The occurrence of ovulation within the first five days following UPA administration (to approximate ovulation around the time of unprotected intercourse) and within 21 days of UPA administration (to assess ovulation that could lead to pregnancy later in the cycle) was measured via ultrasound assessment of follicular rupture and serum progesterone levels. Cervical mucus permeability was also measured, as well as pharmacokinetic parameters for UPA and DSG. No significant differences were found between the UPA+POP (DSG) and placebo+ POP (DSG) sequences within 21 days of UPA use in the occurrence of ovulation (52% for UPA+POP (DSG) and 38% for placebo+POP (DSG); p=0.6753) or median days to ovulation (day 4 for UPA+POP (DSG) and day 3 for placebo+POP (DSG); p=0.5485), suggesting that UPA use did not affect the ability of POP (DSG) to inhibit ovulation later in the cycle. Cervical mucus scores were similar in all three groups (p=0.421), and there were no differences in the pharmacokinetic parameters for UPA or POP (DSG) between groups. This study did not report on adverse events or side effects.

Two studies assessed the effect of initiating hormonal contraception on the ability of UPA to delay ovulation [7, 20]. The RCT examining the pharmacokinetic and pharmacodynamic interactions between UPA and a POP (DSG 75 μg) also assessed the effect of initiating POP (DSG) versus placebo after taking UPA [7]. Ovulation within five days after treatment initiation occurred more frequently in the UPA+POP (DSG) sequences (45%) compared with the UPA+placebo sequences (3%; p=0.0054), suggesting that immediate POP (DSG) use after UPA affected the ability of UPA to delay ovulation. Additionally, the median time to ovulation was four days for the UPA+POP (DSG) sequences and eight days for the UPA+placebo sequences (p<0.0001). The second study was a non-randomized trial that examined the effect of COCs (EE 30 μg and LNG 150 μg) on the ability of UPA to delay ovulation [20]. This study included 36 healthy women, ages 18–35 years, with BMI < 28 kg/m², and regular menstrual cycles and followed them for three menstrual cycles. During cycle 1, participants received UPA when a dominant follicle size of 15 mm was observed, followed by a washout period for cycle 2. During cycle 3, participants again received UPA at the time of a 15 mm dominant follicle and two days later initiated COCs for five days. Participants were followed for seven days after UPA dosing in cycles 1 and 3 for follicle rupture. The primary outcome was the difference in the proportion of participants with a delay in follicle rupture of five days between cycle 1 and cycle 3. Follicle rupture was assessed via transvaginal ultrasound; if the date of follicle rupture was unclear, serum luteinizing hormone and progesterone levels were used to determine date of rupture. More participants demonstrated follicle rupture within five days during the UPA+COC cycle (27%) than during the UPA-only cycle (3%; p=0.008), suggesting that immediate COC use after UPA affected the ability of UPA to delay ovulation. No serious adverse events were reported.

An RCT examined the effect of resuming COCs (after missing three consecutive pills during the study cycle) either immediately or five days after UPA administration on ovarian activity at different time points in the cycle [21]. This study examined the effect of immediate versus delayed COC resumption on the ability of UPA to delay ovulation, in the setting of missed pills. Participants included 49 healthy women, ages 18–30 years, with BMI < 30 kg/m² who took COCs (EE 30 μg and LNG 150 μg) during a baseline cycle, followed by a 7-day hormone-free interval. Participants then took COCs on days 1–4 of the study cycle, deliberately missed pills on days 5–7, and then took UPA (30 mg) on day 8. Half of the participants resumed COCs immediately (on the same day as UPA) and half delayed resumption by five days. The primary outcome was the proportion of participants with ovulation associated with a theoretical risk of pregnancy (based on a modified Hoogland score with a postovulatory ultrasound image and progesterone concentrations >30 nmol/l on one occasion or between 10 and 30 nmol/l on two consecutive occasions) and time to ovulation through day 26 of the cycle. None of the participants in either group ovulated with theoretical risk of pregnancy during the first five days after taking UPA, suggesting that immediate resumption of COC use after UPA did not increase the risk of ovulation within five days of UPA administration compared with delayed COC use. Beyond the fifth day, four participants in the delayed start group experienced ovulation with theoretical risk of pregnancy (days 15–17) compared with no participants in the immediate start group (p=0.042), suggesting that delayed resumption of COCs after UPA use increased the risk of ovulation beyond five days after UPA administration. There was also higher risk of ovarian activity without theoretical risk of pregnancy through 22 days in the delayed restart compared with the immediate restart group (OR 7.78 [95% CI: 1.38; 43.95], p=0.02). This study did not report on adverse events or side effects.

The certainty of the evidence for the primary outcome of effectiveness (measured via ovulation) from the RCTs was assessed as moderate (Table 2) [23]. The RCTs were generally well designed with no serious concerns for risk of bias, inconsistency, or imprecision, but had serious concerns for indirectness due to the use of ovulation as a proxy measure of risk of pregnancy. The outcomes of adverse events and vaginal bleeding had moderate certainty of evidence because of serious concerns for imprecision; the studies had relatively small sample sizes and were not powered to detect differences in these outcomes. The non-randomized trial was also well designed with no serious concerns for risk of bias, imprecision, or inconsistency, but had very low certainty of evidence for effectiveness because of the use of ovulation as a proxy measure of risk of pregnancy and very low certainty for adverse events because of serious concerns for imprecision.