Results of the searches are presented in Fig. 1. Of 11,573 results returned across all five database search sets, there were 114 studies (127 publications) that assessed the effect of physical activity or exercise on sex steroid hormones or SHBG. These included 28 RCTs (24–57), 81 nonrandomized interventions (58–145), and six observational studies (6, 7, 146–149). The pre-post studies included 30 that examined the acute effects of exercise (58–110) and 51 that examined the effects of more than a single exercise session on sex steroid hormones (111–145).

PRISMA flow diagram of literature search, screening, and study selection.

Study characteristics are presented in Supplementary Methods and Material (Supplementary Tables S2A–S2D). There were nine RCTs that included premenopausal (24–37), 18 RCTs that included postmenopausal (38–56), and one that included perimenopausal (57) women. Sample sizes ranged from 18 to 391 in studies of premenopausal, and from 16 to 382 in studies of postmenopausal women. The intervention consisted of predominantly aerobic exercise in 14 studies (25–27, 31, 32, 34–36, 38, 40, 44, 46, 49, 50), strength training in six (29, 30, 41, 54–56), a combination of aerobic and strength in seven (24, 39, 42, 47, 52, 53, 57), and yoga in one (48), with an intervention duration ranging from 8 weeks to 12 months. An inactive control group served as the comparator in 21 studies (24–26, 29–32, 34–36, 39, 40, 42, 46–49, 53–57). Four studies offered stretch, flexibility, or group information classes during the intervention period (38, 41, 50, 52). Three studies had a different exercise intensity or dose as a comparator (27, 28, 44). Outcomes included circulating SHBG (n = 9; refs. 24, 25, 34, 44, 46, 49, 52, 53), estradiol (n = 15; refs. 26, 27, 34, 35, 40–42, 44, 46, 50, 53, 55–57), estrone (n = 6; refs. 40, 42, 44, 46, 50, 53), estrogen (n = 5; refs. 30–32, 36, 47), free estradiol (n = 6; refs. 34, 40, 44, 46, 50, 53), estrone sulfate (n = 2; refs. 34, 53), bioavailable estradiol (n = 1; ref. 34), 2-OH-E1 (n = 3; refs. 25, 33, 38), 16a-OH-E1 (n = 3; refs. 25, 33, 38), progesterone (n = 5; refs. 32, 34, 48, 56, 57), androstenedione (n = 5; refs. 40, 42, 46, 51, 53), testosterone (n = 14; refs. 24, 27, 29, 34, 40, 41, 46, 51–55, 57), free or bioavailable testosterone (n = 5; refs. 34, 40, 46, 51, 53), dehydroepiandrosterone (DHEA, n = 3; refs. 51, 53, 57), dehydroepiandrosterone sulfate (DHEAS, n = 4; refs. 51, 53, 54, 57), and cortisol (n = 7; refs. 24, 29, 37, 39, 42, 52, 55). In studies of premenopausal women, steroid hormone levels were assessed in the follicular (n = 5; refs. 26, 29–31, 34), luteal (n = 1; ref. 25), or both (n = 3) phases of the menstrual cycle.

Nonrandomized interventions included studies that examined acute hormonal responses to a single exercise session (n = 51; refs. 58–110), or multiple exercise sessions (n = 30; refs. 111–145), with interventions ranging from 5 days to 6 months. Samples included premenopausal women only (acute n = 43, multiple sessions n = 17), postmenopausal women only (acute n = 4, multiple sessions n = 9), or both pre- and postmenopausal women (acute n = 4, multiple sessions n = 4). Sample sizes ranged from n = 5 to n = 75 in acute exercise interventions, and n = 6 to n = 148 in interventions of a longer duration. Exercise interventions included aerobic exercise (acute n = 32, multiple sessions n = 18), anaerobic exercise (acute = 1) strength exercise (acute n = 15, multiple sessions n = 8), and combined aerobic and strength training (acute n = 3, multiple sessions n = 4). Twenty-eight studies had a relevant comparison condition, which included menstrual cycle phase (acute n = 8), participant menopause status (acute n = 3, multiple sessions n = 3), participant fitness or body composition (acute n = 4, multiple sessions n = 1), exercise type or dose (acute n = 6, multiple sessions n = 1), or time of day of exercise (acute n = 3). Outcomes included SHBG (acute n = 2, multiple sessions n = 14), estradiol (acute n = 25, multiple sessions n = 15), estrone (multiple sessions n = 4), estrogen (acute n = 2, multiple sessions n = 3), progesterone (acute = 14, multiple sessions = 12), testosterone (acute = 19, multiple sessions = 13), free testosterone (acute = 5, multiple sessions = 2), androstenedione (acute = 2, multiple sessions = 2), DHEA (acute = 3), DHEAS (acute = 3, multiple sessions = 1), and cortisol (acute = 26, multiple sessions = 4).

Prospective cohort studies included pre- (n = 3; refs. 7, 146, 149) and post- (n = 3; refs. 6, 147, 148) menopausal women and had sample sizes ranging from n = 104 to n = 623 participants. Each used self-reported measures of physical activity, with follow-up duration ranging from one or two menstrual cycles to four years. Outcomes included SHBG (n = 2; refs. 6, 148), estradiol (n = 5; refs. 6, 7, 146–148), estrone (n = 3; refs. 6, 147, 148), free estradiol (n = 2; refs. 6, 148), bioavailable estradiol (n = 2; refs. 6, 148), estrone sulfate (n = 1; ref. 6), progesterone (n = 1; ref. 146), testosterone (n = 4; refs. 6, 147–149), androstenedione (n = 3; refs. 6, 147, 148), DHEA (n = 1; ref. 6), and DHEAS (n = 1; ref. 6).

Risk of bias results are presented in Supplementary Materials and Methods (Supplementary Tables S3A–S3D). Sources of bias in RCTs included performance bias (all RCTs), due to the inability to blind participants to the fact they are completing exercise, as well as attrition bias, with 12 studies having greater than 10% attrition or noncompliance (25, 27, 32, 34, 35, 39, 42, 48, 54, 57). There was insufficient information regarding selection bias in 19 studies (24, 26, 27, 29–31, 34–36, 39, 41, 42, 47, 49, 52–56) and insufficient information on the measurement, accuracy, reliability, or sensitivity of hormone assays in two studies (24, 39). All nonrandomized interventions had at least a moderate risk of bias owing to the presence of confounding and seven scored serious for confounding as participant body composition was not considered or reported when selecting or describing participants (58, 59, 64, 73, 79, 101, 109, 125). Studies also scored moderate for participant selection (n = 2; refs. 115, 117), intervention classification and the potential for intervention deviation (n = 13; refs. 63, 69, 71, 116, 117, 120, 121, 124, 131, 133, 138, 142, 144), outcome assessment (n = 1), as well as the number of outcomes reported (n = 2; refs. 129, 138). Five observational studies had moderate risk of bias overall, owing to the potential for confounding, self-reported assessment of physical activity, or the number of analyses performed and reported (6, 7, 146, 148, 149). One observational study had serious risk of bias as it did not adjust for important confounders and as the assessment of estradiol lacked adequate sensitivity for approximately half the participants (147).

Meta-analysis results are presented in Fig. 2 (estrogens and progesterone) and Fig. 3 (androgens, SHBG, and cortisol) and in Supplementary Methods and Materials (Supplementary Figures 1–4). Results from individual studies that were not included in meta-analyses are presented in Supplementary Methods and Materials (Supplementary Tables S4A–S4D).

Abbreviated forest plots for estrogens and progesterone. The x-axis represents the SMD between exercise and control groups.

Abbreviated forest plots for androgens and SHBG. The x-axis represents the SMD between exercise and control groups.

A meta-analysis of RCTs (8 studies, n = 1,353) identified a small increase in SHBG following exercise (SMD = 0.13; 95% CI = 0.02, 0.24; I2 = 0%). Only one individual RCT examined exercise dose and SHBG and it did not identify a clear dose–response relationship (44). Nonrandomized interventions that examined the response of SHBG to multiple exercise sessions mostly reported no significant changes from baseline (113, 115, 116, 119, 120, 127, 128, 131, 132, 137, 139, 140, 145). The acute response of SHBG to a single session of exercise was described by two studies. These observed a small increase (∼10%) in SHBG at the conclusion of exercise, which returned to baseline upon recovery (73, 98). Findings from observational studies were consistent with those from the meta-analysis, with 2 of 2 studies identifying higher levels of SHBG in postmenopausal women who reported higher levels of physical activity (6, 148).

Meta-analyses of RCTs identified small decreases in estradiol (12 studies, n = 1,452; SMD = −0.22; 95% CI = −0.37, −0.08; I2 = 37%) and free estradiol (5 studies, n = 1,033; SMD = −0.20; 95% CI = −0.32, −0.09; I2 = 0%) in response to exercise. There was a suggestion of a small, but nonsignificant, decrease in estrone (4 RCTs, n = 878; SMD = −0.10; 95% CI = −0.24, 0.03; I2 = 0%). There was no effect of exercise on estrogen or estrogen metabolites 2-OH-E1 or 16a-OH-E1 identified via meta-analysis. As moderate heterogeneity was evident in the estradiol meta-analysis, subgroup analysis was performed to identify any differences in effect according to exercise type and menopausal status. A decrease in estradiol was evident in studies that prescribed both aerobic (6 studies, n = 1,060; SMD = −0.15; 95% CI = −0.27, −0.03; I2 = 0%) and resistance exercise (3 studies, n = 116; SMD = −0.80; 95% CI = −1.17, −0.43; I2 = 0%), and in studies that enrolled only postmenopausal women (9 studies, n = 1,070; SMD = −0.28; 95% CI = −0.49, −0.06; I2 = 60%). There was no clear evidence of an effect identified in studies that prescribed combined training (2 studies) or yoga (1 study), or studies that included only pre- (2 studies) or peri- (1 study) menopausal women. In individual RCTs, performing a higher quantity of moderate-activity exercise (300 minutes compared with 150 minutes per week) did not result in a greater effect on hormone levels (32, 44), nor did high-intensity interval exercise compared with continuous moderate-to-vigorous aerobic exercise (27, 28).

Of 14 nonrandomized interventions that had more than a single exercise session, seven reported a reduction in resting estradiol levels (117, 121–124, 130, 133, 143). This was more frequently detected in the luteal rather than the follicular phase of the menstrual cycle (121, 124, 130, 133). Two studies that examined differences in hormone responses between pre- and postmenopausal women did not identify an effect of exercise on sex steroid hormones (136, 141). In studies of acute exercise, estradiol appeared to increase following a single session before returning to baseline upon recovery (60–65, 67, 69–72, 75, 79, 81, 83, 84, 89, 96, 97, 107, 110). This relationship was influenced by exercise type, duration and intensity, recent exercise history, and cycle phase. Longer, more intense exercise led to greater increases in estradiol (69, 79, 83). The response was also greater following a week of intense training (75) and greater in the luteal phase of the menstrual cycle (60–63, 68, 81, 89, 94, 96, 97). Although there was no difference between pre- and postmenopausal women in acute responses to prolonged endurance exercise (a 50–100 km run; ref. 71), acute changes in estradiol levels were more evident in pre-menopausal women following strength training (72, 104).

In prospective cohort studies, more physical activity was associated with less estradiol in 1 of 2 studies of premenopausal women and in 1 of 3 studies in postmenopausal women (6, 7, 146–148). Consistent with the meta-analysis results, higher levels of physical activity were associated with less bioavailable estradiol and free estradiol in 2 of 2 studies of postmenopausal women (6, 148). No association between physical activity and estrone levels were identified (6, 147, 148).

In RCTs, progesterone levels decreased in response to exercise (5 studies, n = 548; SMD = −0.19; 95% CI = −0.36, −0.02; I2 = 0%). Only 2 of 12 nonrandomized intervention studies that examined ongoing exercise identified a decrease in progesterone after exercise training (133, 143), with the remaining studies showing no change from baseline. Nonrandomized interventions that examined the acute response of progesterone to a single session of exercise described a brief increase that returned to baseline following recovery (60–62, 67, 79, 81, 83, 94, 96). This was more evident in the luteal phase (81, 83, 94, 96). There was only one relevant observational study, which did not identify an association between weekly physical activity and progesterone levels (146).

Meta-analyses of RCTs identified small reductions in testosterone (11 studies, n = 1,434; SMD = −0.11; 95% CI = −0.21, −0.01; I2 = 0%), free testosterone (5 studies, n = 1,187; SMD = −0.12; 95% CI = −0.23, −0.01; I2 = 0%), and DHEA (3 studies, n = 312; SMD = −0.23; 95% CI = −0.46, −0.01; I2 = 0%) levels following exercise. Androstenedione (5 studies, n = 868; SMD = −0.10; 95% CI = −0.23, 0.03; I2 = 0%) and DHEAS (4 studies, n = 309; SMD = −0.28; 95% CI = −0.56, 0.01; I2 = 8%) also declined following exercise, although these effects were not statistically significant.

In nonrandomized interventions, there was no change in resting testosterone levels after ongoing exercise in 10 of 12 studies. In response to acute exercise, testosterone levels increased briefly following exercise then returned to baseline following recovery in some (68, 70, 72, 84, 85, 91, 98, 99, 101, 104), but not all studies (65, 90, 93, 96, 100, 108, 109). The increase was evident in response to both aerobic and strength exercise and during both phases of the menstrual cycle. Both DHEA and DHEAS increased in response to exercise before decreasing in 3 of 5 studies (70, 72, 82).

Results from the prospective cohort studies were inconsistent. One study identified lower levels of testosterone and androstenedione in more physically active postmenopausal women (148). This contrasted with studies that did not identify an association between physical activity and testosterone or androstenedione (6, 147). Physical activity was not linearly associated with a decline in either DHEA or DHEAS in one study (6).

Little evidence of a change in cortisol was evident from a meta-analysis of RCTs (5 studies, n = 328; SMD = 0.08; 95% CI = −0.19, 0.35; I2 = 26%). Following a single session of exercise, some studies documented no change (58, 59, 88, 100), some documented an increase in cortisol during or immediately following exercise (66, 68, 74, 82, 90, 103, 105, 106), and others identified a decrease in cortisol compared with baseline at the conclusion of exercise or following a recovery period (70, 72, 79, 85, 87, 92, 96, 104). The increase in cortisol levels observed in some studies was greater when exercise was performed in the morning rather than the afternoon and was linearly related to exercise intensity (66, 102). The decrease in levels following exercise observed in other studies was more evident in premenopausal than postmenopausal women and was independent of exercise type and intensity (72, 79, 104). No prospective cohort studies that investigated the effect of physical activity on cortisol were identified.