Effect of Postmenopausal Hormone Therapy on Glucose Regulation in Women With Type 1 or Type 2 Diabetes: A Systematic Review and Meta-analysis

Diabetes affects approximately 10% of adult females worldwide, of whom the majority (>90%) have type 2 diabetes (1). In women with diabetes, blood glucose regulation may change during and after menopause. It is clinically recognized that these changes often result in more hyperglycemic episodes and higher glucose variability, which lead to increased risk of diabetes-related complications (2,3), underlining the importance of optimizing glycemic control around menopause.

Menopause is a physiological process defined as the final menstrual period in a woman’s life and is characterized by a rapid decrease of circulating estradiol (E2) and progesterone (P4) (4). While metabolic and endocrine changes after menopause may be partly caused by aging (5), they may also be the result of lower estrogen levels, specifically, lower E2 levels. The reduction of E2 after menopause leads to accumulation of visceral fat, decreased energy expenditure, and decreased lipolysis (6), as well as decreased non–insulin-dependent glucose uptake and decreased insulin secretion (7). These metabolic and endocrine changes may be tempered by hormone therapy (HT).

Previous meta-analyses demonstrated that HT improves insulin sensitivity and fasting glucose levels (8,9) in postmenopausal women without diabetes. However, much less is known about the effects of HT on glucose regulation in postmenopausal women with type 1 and type 2 diabetes. Authors of a meta-analysis from 2006 (9) reported that postmenopausal HT caused a greater reduction of insulin resistance in women with diabetes (30%) in comparison with women without diabetes (12.9%). However, effects related to HbA_1c, different types of HT preparations, and different types of diabetes were not studied. In 2013, authors of a systematic review investigating the impact of HT on glycemic control in women with type 1 diabetes identified only one relevant study (10). In the current study, we systematically reviewed the literature to determine the effect of postmenopausal HT on glucose regulation in women with type 1 and women with type 2 diabetes.

The protocol of this systematic review and meta-analysis was registered with International Prospective Register of Systematic Reviews (PROSPERO) (record CRD42021258615). The terms “hormonal replacement therapy” and “peri-postmenopausal women” in the protocol have been changed to “postmenopausal hormone therapy” and “postmenopausal women” in the article, as these terms represent the correct nomenclature for the studied population and intervention.

A systematic literature search was conducted on 24 November 2021 in Embase, MEDLINE, Scopus, and the Cochrane Library. We searched ClinicalTrials.gov to identify unpublished trials. There were no restrictions in language or publication year. Full search terms are presented in Supplementary Table 1. The search was repeated on 28 February 2023 with the original search terms, for identification of new published and unpublished trials between November 2021 and February 2023.

Randomized controlled trials (RCTs) of parallel or crossover design were eligible for inclusion if they met the following criteria:

1. Postmenopausal women as subjects, defined according to cessation of the menstrual cycle for at least 12 months

2. Diagnosis with type 1 or type 2 diabetes according to standard criteria valid at the time of the trial commencing

3. Intervention of systemic HT with at least estrogen monotherapy, in comparison with placebo, standard treatment, or observation

4. One or more of the following as outcomes: HbA_1c, fasting glucose, postprandial glucose, and differences in use of glucose-lowering drugs

Prior to the selection process, we used the deduplicate tool (EndNote v.20) to remove duplicates from the list of retrieved studies. Two authors (E.M.S. and G.V.t.N.d.B.) independently performed title and abstract screening, followed by a full-text eligibility assessment with Rayyan QCRI software. Conflicts were resolved through discussion with a third author (S.E.S.).

A data extraction table was constructed based on standard templates (11). Data collection was performed independently by two authors (E.M.S. and G.V.t.N.d.B.). Results were sought for baseline and posttreatment outcome data. Study authors were contacted if outcome data were missing. Other collected data are listed in Supplementary Table 2.

Quality assessment was completed independently by two reviewers (E.M.S. and G.V.t.N.d.B.), using version 2 of the Cochrane risk-of-bias tool (RoB2) for parallel-group RCTs and crossover RCTs. Judgement about the risk of bias (high, low, or some concerns) from each domain was generated by an automated algorithm, including an overall risk-of-bias judgement result. Conflicts were discussed between the two reviewers, and any remaining conflicts were resolved through discussion with a third author (S.E.S.). Funnel plots were constructed to assess for publication bias.

Effect measures for the outcomes were mean difference in HbA_1c (% and mmol/mol), fasting glucose (mmol/L), postprandial glucose (mmol/L), and odds ratio for differences in use of glucose-lowering drugs. Studies were grouped for data synthesis by tabulation of the study design and by type of estrogen used in the study. Studies with multiple intervention groups were combined with use of the method described in the Cochrane Handbook for Systematic Reviews of Interventions (11).

Meta-analyses were performed with a random-effects model in Review Manager (RevMan) (version 5.4) and were presented as forest plots. Outcome data of included trials were pooled with mean differences based on changes from baseline and change-from-baseline SDs. If the change-from-baseline SD was not available, postintervention means and SDs were pooled. To correct for within-subject correlations in crossover trials, a correlation coefficient of 0.85 was applied based on the average correlation coefficient of two included studies (1999, Samaras et al. [12], correlation coefficient 0.8, and 1997, Brussaard et al. [13], correlation coefficient 0.9). Means and SEs of parallel-group trials and crossover trials were then combined in a meta-analysis and pooled with use of the generic inverse variance method. Subgroup analyses were predefined and grouped by type of diabetes, active substance, and administration route. Statistical significance was set at a P value of <0.05. Heterogeneity was assessed with the I² test for heterogeneity.

To explore whether certain patient characteristics could explain some of the heterogeneity in the data, post hoc meta-regression analyses were performed in SPSS (version 28) with a random-effects model. Mean difference data and corresponding SEs were included as the effect estimate. We hypothesized that the mean age of participants and mean baseline HbA_1c values would explain most of the heterogeneity; therefore, these two predictors were included as covariates.

Sensitivity analyses were performed in RevMan (version 5.4), with the test for comparison of subgroups, to assess whether the parallel-group and crossover trials rendered different results, the effect of studies with high risk of bias on the results, and the effect of imputing correlation coefficients for the crossover trials. Quality of the evidence was evaluated with the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework.

The data sets generated during or analyzed in the current study are available from the corresponding author on request.

A total of 3,376 records were identified, 975 duplications were removed with DedupEndNote, and an additional 53 duplications were removed by the reviewers. The remaining 2,348 records were screened, of which 2,283 records could be excluded by title and abstract screening. The full text of three records could not be retrieved. Attempts were made to retrieve the full text by contacting the authors; nevertheless, no response was received. The remaining 62 records were screened for full text eligibility. A total of published RCTs (12 parallel-group trials and 7 crossover trials) and no unpublished RCTs met the inclusion criteria and were included (Supplementary Fig. 1). For one crossover trial, two reports were identified: the report of the first arm of the crossover trial was included in the meta-analysis as a parallel-group trial (14). We excluded the report of the second arm of the trial (15) to prevent unit-of-analysis error.

Overall, 1,412 participants were included, of whom 4.0% had type 1 diabetes. The median intervention duration was 26 weeks (interquartile range [IQR] 12–46) for parallel-group trials and 12 weeks (IQR 8–26) in crossover trials. Characteristics of included studies are presented in Table 1.

In the risk of bias assessment, 10 studies were categorized as “some concerns” (13,14,16–23) and 9 studies were categorized as “high risk” (Supplementary Tables 4 and 5) (12,24–31). Only one study (20) published a prespecified protocol. Therefore, the remaining studies were all categorized as “some concerns” or “high risk” of bias regarding the selective reporting domain. Six studies (24–28,31) had high risk of bias due to missing outcome data, and three crossover trials (12,24,30) had high risk of bias arising from carryover effects. However, there was no indication for risk of bias based on the measurement methods of HbA_1c, fasting glucose, or postprandial glucose in the included studies. The funnel plots did not show an asymmetric pattern, suggesting low risk of publication bias (Supplementary Fig. 2).

Nine parallel-group trials (n = 352 participants) and six crossover trials (n = 136 participants) included investigation of the effect of postmenopausal HT on HbA_1c, with a total of 488 participants. Seven trials had high risk of bias, and there were some concerns about eight other trials (Supplementary Tables 4 and 5).

In eleven studies oral preparations were used and in three studies transdermal preparations, and one study included both transdermal and oral preparations (Fig. 1 and Table 1). Twelve studies used combined HT, of which three studies used conjugated equine estrogens (CEE) with medroxyprogesterone acetate (MPA), two studies used 17-β-estradiol (E2) with micronized P4, six studies used E2 with norethisterone acetate (NRA), and one study used E2 with drospirenone (D). Three studies used unopposed estrogen therapy (ET), of which two studies used CEE and one study used oral E2.

Postmenopausal HT reduced HbA_1c of patients with type 2 diabetes, with a mean difference of −0.56% (95% CI −0.80, −0.31) (−6.08 mmol/mol [95% CI −8.80, −3.36]) (I² = 68%) (Fig. 1). No studies investigated the effect of HT on HbA_1c in patients with type 1 diabetes.

Seven parallel-group trials (n = 275 participants) and five crossover trials (n = 107 participants), with a total of 382 participants, investigated the effect of oral HT on HbA_1c. Oral HT reduced HbA_1c, with a mean difference of −0.61% (95% CI −0.89, −0.32) (−6.63 mmol/mol [95% CI −9.74, −3.52]) (I² = 73%) (Supplementary Fig. 3).

Seven studies used preparations containing oral E2 (13,16,19,22,25,27,28), which reduced HbA_1c, with a mean difference of −0.74% (95% CI −1.16, −0.33) (−8.14 mmol/mol [95% CI −12.62, −3.66]) (I² = 78%). Oral E2 reduced HbA_1c in high doses of 2 mg/day (mean difference −0.80% [95% CI −1.35, −0.25], −8.76 mmol/mol [95% CI −14.81, −2.72], I² = 89%), as well as in low doses of 1 mg/day (mean difference −0.61% [95% CI −1.16, −0.07], −6.68 mmol/mol [95% CI −12.65, −0.71], I² = 0%).

Five studies investigated the effect of HT containing CEE on HbA_1c (12,14,17,30,31). CEE reduced HbA_1c, with a mean difference of −0.43% (95% CI −0.74, −0.12) (−4.71 mmol/mol [95% CI −8.11, −1.32]) (I² = 38%).

Three parallel-group trials (n = 90 participants) and one crossover trial (n = 19 participants) investigated the effect of transdermal HT on HbA_1c, with a total of 109 participants. No significant differences in HbA_1c following transdermal HT were found (mean difference −0.34% [95% CI −0.75, 0.08], −3.66 mmol/mol [95% CI −8.16, 0.85], I² = 0%) (Fig. 2).

Twelve studies reported the effect of combined HT on HbA_1c (12,14,16,18,19,21,22,25,27–29,31). The analysis included eight parallel-group trials (n = 312 participants) and four crossover trials (n = 91 participants), with 403 participants in total. Six studies used E2 + NRA, two studies used E2 + P4, one study used E2 + D, and three studies used CEE + MPA. Combined HT reduced HbA_1c, with a mean difference of −0.58% (95% CI −0.91, −0.25) (−6.30 mmol/mol [95% CI −9.90, −2.70]) (I² = 68%) (Supplementary Fig. 4).

One parallel-group trial (n = 20 participants) (13) and two crossover trials (n = 45 participants) (17,30) reported the effect of ET on HbA_1c. The parallel-group trial used oral E2, and the two crossover trials used CEE. Compared with the control group, ET reduced HbA_1c, with a mean difference of −0.42% (95% CI −0.61, −0.22) (−4.59 mmol/mol [95% CI −6.76, −2.42]) (I² = 0%) (Supplementary Fig. 5).

All trials with investigation of the impact of postmenopausal HT on HbA_1c were included in the meta-regression model (Fig. 1 and Supplementary Fig. 6). No association was found between the mean difference in HbA_1c and baseline HbA_1c of participants (slope −0.095 [95% CI −0.405, 0.214], P = 0.513) or mean age of participants (slope 0.006 [95% CI −0.90, 0.101], P = 0.896) (Supplementary Fig. 6).

Nine parallel-group trials (n = 1,147 participants) and five crossover trials (n = 112 participants), with 1,259 participants in total, investigated the effect of HT on fasting glucose. Five trials had high risk of bias, and there were some concerns about nine other trials (Supplementary Tables 4 and 5).

Eleven studies had oral preparations, two studies had transdermal preparations, and one study had both transdermal and oral HT (Table 1 and Fig. 3). Twelve studies included combined HT, of which one included E2 + P4, seven E2 + NRA, and one E2 + D. Four studies used CEE + MPA, of which one study investigated both combined CEE + MPA and unopposed CEE in patients with type 2 diabetes. One study investigated the effect of unopposed CEE on fasting glucose (Fig. 3).

Postmenopausal HT decreased fasting glucose in patients with type 2 diabetes (mean difference −1.15 mmol/L [95% CI −1.78, −0.51], I² = 73%) (Fig. 3). Only Scott et al. (26) investigated whether HT had an effect on fasting glucose in both patients with type 1 diabetes and patients with type 2 diabetes. The authors combined the mean difference results for patients with type 1 and type 2 diabetes (Fig. 3).

Eight parallel-group trials (n = 1,095 participants) and four crossover trials (n = 93 participants), with a total of 1,188 participants, investigated the effect of oral HT on fasting glucose (14,16,17,19,20,22–28). Oral HT reduced fasting glucose, with a mean difference of −1.34 mmol/L [95% CI −2.03, −0.65], I² = 75%) (Supplementary Fig. 7).

When stratified by active substance, oral E2 reduced fasting glucose (mean difference −1.79 mmol/L [95% CI −2.91, −0.66], I² = 77%). Oral E2 reduced fasting glucose in high-dose preparations (2 mg/day), with a mean difference of −1.77 mmol/L (95% CI −3.42, −0.13; I² = 85%), as well as in low-dose preparations of 1 mg/day (mean difference −1.70 mmol/L [95% CI −2.68, −0.71], I² = 0%). CEE reduced fasting glucose, with a mean difference of −0.73 mmol/L (95% CI −1.12, −0.35; I² = 0%).

Two parallel-group studies (n = 65 participants) and one crossover study (n = 19 participants), with a total of 109 participants, used transdermal HT (18,21,25). Two studies used E2 + NRA and one study used E2 + P4. Transdermal HT did not affect fasting glucose (mean difference −0.06 mmol/L [95% CI −1.09, 0.97], I² = 0%) (Supplementary Fig. 8).

In 13 studies, participants were treated with combined HT (14,16,18–28): 9 parallel-group trials (n = 1,147 participants) and 4 crossover trials (n = 87 participants), with 1,234 participants in total. Seven studies used E2 + NRA, one study used E2 + P4, one study used E2 + D, and four studies used CEE + MPA. Combined HT reduced fasting glucose, with a mean difference of −1.15 mmol/L (95% CI −1.84, −0.45; I² = 75%) (Supplementary Fig. 9).

Two crossover studies (n = 45 participants) investigated the effect of ET on fasting glucose (17,24). ET was given in oral CEE preparations, which reduced fasting glucose, with a mean difference of −0.96 mmol/L (95% CI −1.76, −0.16; I² = 0%) (Supplementary Fig. 10).

The meta-regression model of included studies with investigation of the effect of HT on fasting glucose (Fig. 3) showed no significant association between treatment effect and mean age of participants (slope 0.062 [95% CI −0.138, 261], P = 0.497) or baseline HbA_1c of participants (−0.493 [95% CI −1.20, 0.213], P = 0.146) (Supplementary Fig. 11).

Postprandial glucose was measured in one crossover study (n = 25) (17). No differences in postprandial glucose concentrations were found between the estrogen and placebo arms in the study. In another crossover study glucose was measured after an oral glucose tolerance test (n = 23) (27). The authors reported that the sum of glucose concentrations at 0, 30, 60, 90, and 120 min after an oral glucose tolerance test did not differ between treatment arms.

In one crossover trial (17) investigators reported that, in 3 of 25 participants (12%), diabetes medication was added: one participant was in the estrogen arm (n = 1) and two were in the placebo arm (n = 2) when diabetes medication was added. In six studies investigators reported that glucose-lowering medication was not altered throughout the study (14,19,22,24,28,31). In the remaining 12 studies, no information on changes of glucose-lowering medication was reported.

We conducted a sensitivity analysis to assess whether the parallel-group and crossover trials rendered significantly different results. There was a small, but significant, difference for the effect of HT on HbA_1c between parallel-group trials and crossover trials (−0.31 vs. −0.77, P = 0.04) (Fig. 1). However, both types of RCTs showed a significant reduction in HbA_1c (Fig. 1). The effect on fasting glucose was comparable between parallel-group studies and crossover studies (mean difference −0.78 vs. −1.37, P = 0.36).

We performed another sensitivity analysis to assess whether imputing a correlation coefficient of 0.85 for crossover studies would change the results, in comparison with a correlation coefficient of 0. As expected, the correlation coefficient of 0.85 increased heterogeneity for HbA_1c (I² 83% vs. 13%) and fasting glucose (I² 73% vs. 56%). However, the correlation coefficient did not significantly change the overall effect of HT on HbA_1c (P = 0.96) or fasting glucose (P = 0.96). In addition, studies with high risk of bias did not change the overall effect on HbA_1c (P = 0.91) or fasting glucose (P = 0.34).

In this systematic review, we investigated the effect of postmenopausal HT on glucose regulation in women with type 1 or type 2 diabetes. We demonstrated that postmenopausal HT significantly improved HbA_1c and fasting glucose in patients with type 2 diabetes in the short-term. However, the sample size of patients with type 1 diabetes was insufficient to determine the effect of postmenopausal HT in women with type 1 diabetes. In addition, our results were underpowered to draw conclusions on the effect of transdermal HT on glucose regulation, and the evidence for the effect of HT on postprandial glucose and on use of glucose-lowering drugs was scarce.

The mechanisms by which HT reduces fasting glucose and HbA_1c are not entirely understood (7). Rodent studies suggest that estrogens decrease insulin-mediated suppression of hepatic glucose production, leading to lower plasma glucose levels (32). Results of previous studies have suggested that, as transdermal HT avoids the first-pass effect in the liver, the suppressive effect of transdermal HT on hepatic glucose production may be weaker in comparison with oral HT (33). However, due to the lack of available data, we were unable to conclusively determine the impact of transdermal HT on glucose regulation. Furthermore, it has been observed that estrogens improve β-cell survival and insulin secretion, as assessed on the basis of increased C-peptide levels (7), which may contribute to estrogen-mediated reduction of hyperglycemia in women with type 2 diabetes with some remaining β-cell function. Further research is needed for investigation of whether postmenopausal HT also leads to favorable glucose-lowering effects in women with type 1 diabetes, who have little remaining β-cell function.

Results of previous studies suggested that the estrogen-mediated effects on glucose regulation may be attenuated when progestogens are added (7). For example, in insulin-resistant mice, subcutaneous P4 injections increased hepatic glucose production in the liver (34). However, our findings demonstrate that the reductions of fasting glucose and HbA_1c were maintained in combined HT preparations with progestogens and that the extent of the reduction in HbA_1c and fasting glucose was no different in unopposed HT preparations.

In line with our work, investigators of previous meta-analyses observed similar beneficial effects of postmenopausal HT on glucose regulation in women with type 2 diabetes: HT significantly reduced fasting glucose (9,35) and HbA_1c (35). Likewise, findings of the Women’s Health Initiative (WHI) trial (36), in which 16,608 healthy postmenopausal women were included and treated with CEE + MPA for a mean of 5.2 years, demonstrated favorable changes in glucose metabolism. However, the increased risk of venous thromboembolism, breast cancer, and ovarian cancer resulted in changes to practice guidelines, which now advise against prescription of HT for chronic disease prevention (37).

Notwithstanding, for postmenopausal women with bothersome vasomotor symptoms, the benefits of HT might outweigh the risks (37). Our findings show that postmenopausal HT is expected to have a neutral-to-beneficial impact on glucose regulation in women with type 2 diabetes for whom postmenopausal HT is considered to treat menopausal symptoms. Postmenopausal HT in women with type 2 diabetes should be prescribed in accordance with the guidelines for healthy postmenopausal women (37). Transdermal HT preparations may be preferred, as the risk of venous thromboembolism is lower in comparison with oral HT (38,39).

This systematic review has some limitations. Firstly, half of the included studies had high risk of bias, due to the lack of predefined protocols and analysis plans. However, the studies in the present systematic review were published before protocol registration became a requirement in many peer-reviewed journals. The risk of bias arising from faulty measurement of the outcomes was low. Therefore, the significant reductions of ∼0.4–0.8% HbA_1c and ∼1–2 mmol/L fasting glucose are clinically relevant. However, there was not sufficient power to examine differential effects of different estrogen or progestogen formulations. In addition, parallel-group trials and crossover trials were combined in the meta-analyses to increase conciseness and statistical power; however, methodological differences of crossover and parallel-group trials may introduce bias in the effect estimate. To account for this limitation, we separated crossover and parallel-group studies into subgroups within the forest plots. Furthermore, we did not have access to individual participant data; therefore, in the findings of the meta-analysis and meta-regression analysis at the group level there may be overestimation or underestimation of the association between patient-level characteristics and treatment effect. In addition, we used the I² statistic to report heterogeneity; however, an important limitation of I² is that it does not show how much the effect size varies (40). To explore whether certain patient characteristics could explain some of the heterogeneity in the data, we performed meta-regression analyses. The meta-regression analyses were conducted post hoc and therefore findings should be interpreted with caution. Only six studies reported that glucose-lowering medication was unchanged during the study, and changes in glucose-lowering medication could make our findings less generalizable. Lastly, the sensitivity analyses showed that investigators of crossover trials found a more pronounced reduction of HbA_1c in comparison with parallel-group trials. However, both types of RCTs demonstrated that HT significantly reduced HbA_1c. Despite the difference in reduction of HbA_1c between parallel-group trials and crossover trials, the overall results of the sensitivity analyses provide evidence for the robustness of our findings.

In conclusion, we report that short-term postmenopausal HT reduced fasting glucose and HbA_1c in women with type 2 diabetes. When postmenopausal HT is considered for menopausal symptoms in women with type 2 diabetes, HT is expected to have a neutral-to-beneficial impact on glucose regulation. Further research is needed for investigation of the effect of transdermal HT on glucose regulation among postmenopausal women with diabetes and for understanding of the impact of postmenopausal HT on glucose regulation in patients with type 1 diabetes.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. G.V.t.N.d.B. and S.E.S. designed the protocol of the study. A.M. designed the search strategy and performed the search. E.M.S. and G.V.t.N.d.B. performed screening, quality assessment, data collection, and data synthesis. E.M.S., P.H.B., D.J.S., and S.E.S. contributed to interpretation of data in the manuscript. All authors contributed to writing the manuscript.