Efficacy of Oral Medications or Intrauterine Device-Delivered Progestin in Patients with Endometrial Hyperplasia with or without Atypia: A Network Meta-Analysis

The aim of this systematic review was to evaluate the efficacy of oral medication or intrauterine device-delivered progestins in patients with endometrial hyperplasia (EH) with or without atypia. We systematically examined PubMed, EMBASE, the Cochrane Library, and clinicaltrials.gov to identify studies reporting the regression rate of patients with EH who received progestins or non-progestins. The regression rates after different treatments were compared using a network meta-analysis in terms of the relative ratios (RRs) and 95% confidence intervals (CIs). Begg–Mazumdar rank correlation and funnel plots were performed to evaluate the publication bias. Five non-randomized studies and 21 randomized controlled trials involving 2268 patients were included in the network meta-analysis. The levonorgestrel-releasing intrauterine system (LNG-IUS) was associated with a higher regression rate than medroxyprogesterone acetate (MPA) (RR 1.30, 95% CI 1.16–1.46) in patients with EH. Among those without atypia, the LNG-IUS was associated with a higher regression rate than any of the three types of oral medications (MPA, norethisterone, or dydrogesterone (DGT)) (RR 1.35, 95% CI 1.18–1.55). According to the network meta-analysis, combining the LNG-IUS with MPA or metformin increased regression rate, while DGT was associated with the highest regression rate among all oral medications. The LNG-IUS may be the best choice for patients with EH, and combining it with MPA or metformin may further improve its efficacy. DGT may be the preferred choice for patients who are unwilling to use the LNG-IUS or who cannot tolerate its side effects.


Introduction
Endometrial hyperplasia (EH) is a non-invasive, abnormal proliferation of endometrial glands or stroma of the uterus, and it increases the risk of endometrial cancer [1]. The World Health Organization defines two types of EH, with or without atypia [2]. The type with atypia is thought to be highly precancerous, with the condition becoming malignant in almost 60% of cases within five years of diagnosis [3]. Both types rarely occur in women younger than 30 years; the type with atypia is diagnosed most often in women 60-64 years old, and the type without atypia in women 50-54 years old [4]. Among premenopausal women, the incidence rate of EH with atypia is 7 per 100,000 woman-years, while that of EH without atypia is 30 per 100,000 woman-years [5].
The most common symptom of EH is abnormal uterine bleeding, which means more frequent or severe bleeding in the case of premenopausal women or any uterine bleeding in the case of postmenopausal women [6]. A systematic review of studies on premenopausal women concluded that the risk of endometrial carcinoma is higher among women who experience inter-menstrual bleeding than among women who experience heavy menstrual

Search Strategy
The following electronic databases were searched: PubMed, EMBASE, the Cochrane Library, and clinicaltrials.gov. We searched all databases from their respective inception to 17 May 2022. The search algorithm was based on the key terms: (endometrial hyperplasia) AND ((levonorgestrel) OR (medroxyprogesterone acetate) OR (megestrol acetate) OR (progesterone) OR (norethisterone) OR (dydrogesterone)). Reference lists of included studies and previous systematic reviews were manually reviewed in order to identify additional studies. In cases of multiple studies reporting on the same patient population, only the largest study was included.
We excluded studies if: (1) the participants were diagnosed with endometrial carcinoma; (2) the original data were not reported, such as in the case of reviews, study protocols, comments, or letters; (3) necessary data could not be obtained; (4) the studies had a single-arm cohort design; or (5) the studies were published in a language other than English.

Study Selection
All literature searches were screened independently by two reviewers, and any discrepancies were resolved by discussion between them or together with the corresponding author. The studies were screened for eligibility initially based on their titles and abstracts, then based on a review of the full text.

Quality Assessment
The quality of the RCTs was evaluated using the Cochrane Risk of Bias (RoB) assessment tool 2.0 (RoB 2.0) [18], and the risk of bias in the following domains was classified as low, high, or unclear: randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. The quality of non-randomized studies was assessed using the Risk of Bias in Non-Randomized Studies of Interventions (ROBINS-I) tool, and the risk of bias in the following domains was classified as low, moderate, high, or critical: confounding, selection of participants, classification of interventions, deviation from intended intervention, missing data, measurement and reporting of outcomes [19]. The publication bias was assessed using Begg-Mazumdar rank correlation and funnel plots [20]. Any discrepancies during the quality assessment were resolved through discussion with the corresponding author.

Data Extraction and Calculations of Outcome
Two reviewers independently extracted the following data from each study: name of authors, publication year, study design, EH subtypes (with or without atypia), numbers of total patients and patients who experienced regression, and follow-up. Regression was defined as when the endometrial biopsy during follow-up was described as appearing "proliferative", "secretory", "inactive", or "atrophic", or as indicating a "progesterone effect" [9]. The regression rate was calculated as the number of patients with regression, divided by the total number of patients who received medication [13].

Statistical Analysis
The meta-analysis was performed using Stata 14.0 (StataCorp, College Station, TX, USA). Results associated with p values < 0.05 were considered significant. The rates of regression were compared in terms of the relative ratios (RRs) and 95% confidence intervals (CIs) using the random-effect and DerSimonian-Laird methods [21]. Heterogeneity was assessed based on I 2 values and a visual analysis of forest plots. We considered I 2 > 50% as high heterogeneity, in which case we conducted subgroup and sensitivity analyses, and drew Galbraith plot to obtain more detailed insights and to assess potential sources of heterogeneity [22]. The subgroup analyses were based on the EH subtype, country, and study design. The sensitivity analyses were performed by removing one study at a time and repeating the meta-analysis.
A network meta-analysis was performed using Aggregate Data Drug Information System (ADDIS) 1.16.8, which uses a Bayesian approach and allows comparisons among all treatment arms of studies, including direct and indirect comparisons simultaneously [23].

Study Selection
Our search found a total of 3885 published articles-625 in PubMed, 2971 in Embase, 219 in the Cochrane Library, and 70 on clinicaltrials.gov. We removed 433 duplicate articles and excluded another 3340 based on the titles or abstracts. A full-text review of the remaining 112 articles led to the inclusion of 26 in the systematic review and network meta-analysis ( Figure 1).
Our search found a total of 3885 published articles-625 in PubMed, 2971 in Emb 219 in the Cochrane Library, and 70 on clinicaltrials.gov. We removed 433 dupli articles and excluded another 3340 based on the titles or abstracts. A full-text review the remaining 112 articles led to the inclusion of 26 in the systematic review and netw meta-analysis ( Figure 1).  Table 1 shows the characteristics of the 26 studies, of which 5 were non-random and 21 were RCTs . Altogether, the trials involved 2268 patients with EH wit without atypia. The samples in the studies ranged from 40 to 215 patients, and the stu were carried out in the following countries: Egypt (n = 6), Iran (n = 7), Turkey (n = Norway (n = 4), Italy (n = 1), China (n = 1), Russia (n = 1), India (n = 1), and Pakistan 1). The baseline patient characteristics were similar among the studies, allowing a netw of comparisons involving 11 treatments to be analyzed ( Figure 2).  Table 1 shows the characteristics of the 26 studies, of which 5 were non-randomized and 21 were RCTs . Altogether, the trials involved 2268 patients with EH with or without atypia. The samples in the studies ranged from 40 to 215 patients, and the studies were carried out in the following countries: Egypt (n = 6), Iran (n = 7), Turkey (n = 4), Norway (n = 4), Italy (n = 1), China (n = 1), Russia (n = 1), India (n = 1), and Pakistan (n = 1). The baseline patient characteristics were similar among the studies, allowing a network of comparisons involving 11 treatments to be analyzed ( Figure 2).  In the figure, the line width is proportional to the number of trials comparing treatments. The node size is proportional to the number of participants randomly assig to that treatment.

Quality Assessment of Included Studies
The quality of the RCTs was evaluated using the RoB 2.0 tool. Nearly all RCTs (1 21) were classified as being at low risk of bias ( Figure 3A). Two of the 21 RCTs had so concerns regarding the randomization process. One of the 21 RCTs had some conce regarding the deviations from intended intervention and the selection of the repor results.
The quality of the non-randomized studies was assessed using the ROBINS-I t Five non-randomized studies were classified as being at moderate risk of bias ( Figure 3 All of them were evaluated as having a moderate risk of bias in the measurement of outcome, and two of them were also evaluated as having a moderate risk of bias in selection of the reported result. In the figure, the line width is proportional to the number of trials comparing the treatments. The node size is proportional to the number of participants randomly assigned to that treatment.

Quality Assessment of Included Studies
The quality of the RCTs was evaluated using the RoB 2.0 tool. Nearly all RCTs (18 of 21) were classified as being at low risk of bias ( Figure 3A). Two of the 21 RCTs had some concerns regarding the randomization process. One of the 21 RCTs had some concerns regarding the deviations from intended intervention and the selection of the reported results.
The quality of the non-randomized studies was assessed using the ROBINS-I tool. Five non-randomized studies were classified as being at moderate risk of bias ( Figure 3B). All of them were evaluated as having a moderate risk of bias in the measurement of the outcome, and two of them were also evaluated as having a moderate risk of bias in the selection of the reported result.   [26][27][28]32,33,[36][37][38]41,42,44,49]. The LNG-IUS was associated with a significantly higher regression rate (RR 1.30, 95% CI 1.16-1.46, p < 0.001; I 2 = 82.0%; Figure 4A). Given the high heterogeneity of the pooled data, we did subgroup analyses and sensitivity analyses but failed to uncover clear differences among subgroups. The Galbraith plot showed four studies might be the potential sources of heterogeneity [27,32,44,49] (Supplementary Figure S1).  Figure 4A). Given the high heterogeneity of the pooled data, we did subgroup analyses and sensitivity analyses but failed to uncover clear differences among subgroups. The Galbraith plot showed four studies might be the potential sources of heterogeneity [27,32,44,49] (Supplementary Figure S1).

LNG-IUS vs. NET
Four RCTs involving 357 patients reported regression rates for the LNG-IUS (86.5%, 154/178) and NET (64.2%, 115/179) [24,25,32,41]. The LNG-IUS was associated with a significantly higher regression rate (RR 1.37, 95% CI 1.07-1.74, p = 0.012; I 2 = 72.6%; Figure 5A). Given the high heterogeneity of the pooled data, we did subgroup analyses but failed to uncover clear differences among subgroups. Sensitivity analyses showed one study might be the potential source of heterogeneity [32]. Excluding this study led to the same result as the full meta-analysis, but with lower heterogeneity (RR 1.18, 95%CI 1.06-1.32, p = 0.003; I 2 = 0.0%; Supplementary Figure S3).  Figure 5B). Given the high heterogeneity of the pooled data, we did subgroup analyses and sensitivity analyses but failed to uncover clear differences among subgroups. The Galbraith plot showed four studies might be the potential sources of heterogeneity [32,35] (Supplementary Figure S4).

Network Meta-Analysis of Regression Rates
Twenty-six studies involving 2268 patients contributed to our network meta-analysis of regression rates after treatment with one of 11 regimens (Table 2 and Figure 6). The rank probability of regression across all patients showed the following trend: LNG-IUS+MPA > LNG-IUS+MET > LNG-IUS > DGT > NET > MP > MPA > LYN > LET > MA+MET > MA. The LNG-IUS was ranked higher than any of the oral medications on their own, and combining the LNG-IUS with oral MPA or MET shifted the LNG-IUS to the two highest rank positions. Among the oral medications on their own, DGT was ranked at the top.  [32,35,39,41]. The rates did not differ significantly between the two groups (RR 1.26, 95% CI 1.00-1.60, p = 0.055; I 2 = 86.2%; Figure 5B). Given the high heterogeneity of the pooled data, we did subgroup analyses and sensitivity analyses but failed to uncover clear differences among subgroups. The Galbraith plot showed four studies might be the potential sources of heterogeneity [32,35] (Supplementary Figure S4).

Network Meta-Analysis of Regression Rates
Twenty-six studies involving 2268 patients contributed to our network meta-analysis of regression rates after treatment with one of 11 regimens (Table 2

Publication Bias
The Begg-Mazumdar rank correlation test showed no evidence of publication bias in the meta-analysis of regression rates (p = 0.152), and the funnel plot was symmetrical ( Figure 7).

Discussion
In this network meta-analysis, we evaluated the effectiveness of different medications in the treatment of EH with or without atypia. Among the oral medications used individually, DGT may be superior to other progestins or non-progestins. Among any of the medications used individually, the LNG-IUS seems to be associated with a higher regression rate than oral progestins or non-progestins. Consistently, two previous

Publication Bias
The Begg-Mazumdar rank correlation test showed no evidence of publication bias in the meta-analysis of regression rates (p = 0.152), and the funnel plot was symmetrical (Figure 7).

Publication Bias
The Begg-Mazumdar rank correlation test showed no evidence of publication bias in the meta-analysis of regression rates (p = 0.152), and the funnel plot was symmetrical ( Figure 7).

Discussion
In this network meta-analysis, we evaluated the effectiveness of different medications in the treatment of EH with or without atypia. Among the oral medications used individually, DGT may be superior to other progestins or non-progestins. Among any of the medications used individually, the LNG-IUS seems to be associated with a higher regression rate than oral progestins or non-progestins. Consistently, two previous

NET
The values in each cell are the relative ratio (and associated 95% confidence interval) of regression rates after the indicated treatment. Each box represents a comparison of the row-defining treatment versus the column-defining treatment. Odds ratios of more than 1 favor the column-defining treatment and odds ratios of less than 1 favor the row-defining treatment.

Discussion
In this network meta-analysis, we evaluated the effectiveness of different medications in the treatment of EH with or without atypia. Among the oral medications used individually, DGT may be superior to other progestins or non-progestins. Among any of the medications used individually, the LNG-IUS seems to be associated with a higher regression rate than oral progestins or non-progestins. Consistently, two previous studies suggested that the LNG-IUS was more effective than other oral progestins [10,50]. The present review substantially extends those findings by examining a much larger sample and by comparing the LNG-IUS and non-progestins.
Among patients with EH without atypia, the present meta-analysis showed that the LNG-IUS was associated with a significantly higher regression rate than other treatments, similar to another meta-review involving fewer studies, only a few of which were also included in the present analysis [50]. In contrast, another meta-analysis, which involved only single-arm studies that conducted purely indirect comparisons of the LNG-IUS and oral medications, found no significant difference in regression rates between oral medications and the LNG-IUS [51]. Our network meta-analysis suggests that the combinations of the LNG-IUS with MPA or MET are superior to the LNG-IUS or MPA on their own. Our work supports the use of the LNG-IUS+MPA for young women with early-stage endometrial carcinoma who want to preserve their fertility [52,53]. Our work also supports previous studies that concluded that adding MET or LET can increase the efficacy of the LNG-IUS or oral progestins [45,54]. Non-progestins may increase the efficacy of progestins on their own by upregulating the progesterone receptor in the endometrium [55].
Our findings should be interpreted with caution because of its limitations. First, the samples were quite small for some treatment regimens, especially the LNG-IUS+MPA (27 of 2268 patients), LET (92), LYN (55), LNG-IUS+MET (25), MA+MET (73), and MP (61) treatments. Second, our meta-analysis pooled data from RCTs and non-randomized studies, which differed substantially in size and potentially in heterogeneity. Indeed, the included studies differed markedly in their medication dose and the usage method of drugs, as well as the duration of follow-up. Patient compliance also plays an important role in the treatment of EH, while women treated with the LNG-IUS need to go to the hospital regularly to make sure the location of the IUD and women treated with oral medication need to take their medicine according to their physician's orders. Third, our study did not compare the safety of the different treatments. For patients receiving the LNG-IUS, the IUD may migrate to any place other than the uterine cavity. Patients with EH who take oral progestins for a long time may have bloating, nausea, headaches, and mood swings or even depression, and the use of oral progestins may lead to liver function damage. Our study also did not compare the curative effect and safety between oral progestins and non-progestins as a result of lacking relevant data. More studies are needed. Fourth, the included studies did not uniformly report sufficient data for us to compare the treatments in terms of other clinically important outcomes, such as menstrual blood loss or other symptomatic improvements. The same was true for patient characteristics such as obesity or the presence of diabetes mellitus, whose potential influence we could not assess in the subgroup analyses [50,56]. We thought that the same characteristics as the risk factors of EH were related to the treatment effect. Fifth, most of the patients in the included studies were younger than 50 years old, and there were no suitable studies about the therapies of EH with the women aged around 50-60, even though these women are more susceptible to EH [5]. Although we did not get enough information about the race of the women included, we did find that the LNG-IUS is more used in economically advanced countries based on the countries of each study. Sixth, the treatment of patients with EH is a long-term treatment, and our study did not compare the recurrence rate and resistant rate of different therapies. Seventh, we observed the high heterogeneity in regression rate between LNG-IUS and MPA groups, LNG-IUS and oral medications groups, LNG-IUS and NET groups, MA and MA+MET groups, and we found the potential sources by using subgroup analyses, sensitivity analyses and Galbraith plots. After removing those studies, we obtained similar results, this suggests that even our more heterogeneous meta-analyses are reliable. We also observed the high heterogeneity in regression rate between MA and MA+MET groups, and we thought the reason might be there were only two studies and too few patients. More research is needed to support whether MET can increase the efficacy of MA. Lastly, we may have introduced bias by including only English-language studies, yet our analysis suggests a low risk of such bias.
Nowadays, people are beginning to pay more attention to their own health, and when there is a sign something's wrong, such as abnormal uterine bleeding, many people will go to hospitals for professional instruction and treatment. Additionally, with the development of medical technology, the diagnosis rate of EH has been increasing gradually in recent years, especially in perimenopausal women and postmenopausal women. Some studies have suggested that calculating the endometrial thickness of postmenopausal women using an endovaginal color dopple has similar diagnostic accuracy compared with a histopathologic diagnosis [57]. The diagnosis methods of EH include dilatation and curettage (D&C) and an endometrial biopsy with a hysteroscopy [58,59]. The treatment of EH should be individualized according to age, fertility demands, personal conditions, and other factors. For postmenopausal women with atypical EH, hysterectomy is the most suitable treatment. For premenopausal women with EH who want to preserve their fertility, the LNG-IUS and oral medication are more suitable [8]. Our study has compared the therapeutic effect to EH with or without atypia between the LNG-IUS and oral medication, and we hope the outcomes of our study can help clinicians find the most appropriate treatment for these patients.
In spite of these limitations, the present work appears to be the first network metaanalysis suggesting that the LNG-IUS may be the most appropriate choice for patients with EH with or without atypia, and that combining the LNG-IUS with MPA or MET may further increase the regression rate. Our work also suggests that among the oral medications, DGT may be the most appropriate choice for women who are unwilling to use the LNG-IUS or who cannot tolerate its side effects.