Postoperative hormonal treatment for prevention of endometrioma recurrence after ovarian cystectomy: a systematic review and network meta‐analysis

Background The efficacy of hormonal regimens for the prevention of endometrioma recurrence in women who have undergone conservative surgery is still controversial. Objective To compare the efficacy of different hormonal regimens in this context and to rank them. Search strategy MEDLINE and Scopus databases were searched through January 2020. Selection criteria Randomised controlled trials (RCTs) or cohorts, comparing the effect of any pair of interventions (i.e. cyclic oral contraceptives [OC], continuous OC, gonadotropin‐releasing hormone agonist [GnRHa], dienogest [DNG], levonorgestrel‐releasing intrauterine system [LNG‐IUS] and expectant management) on endometrioma recurrence were selected. Data collection and analysis Data were independently extracted by two reviewers. Relative treatment effects were estimated using network meta‐analysis (NMA) and ranked in descending order. Main results Six RCTs (675 patients) and 16 cohorts (3089 patients) were included. NMA of the RCTs involving expectant management, cyclic OC, continuous OC, GnRHa and GnRHa + LNG‐IUS, showed that all hormonal regimens had a nonsignificant lower risk of endometrioma recurrence compared with expectant management. NMA of the cohorts involving expectant, cyclic OC, continuous OC, GnRHa, DNG, LNG‐IUS, GnRHa + OC, and GnRHa + LNG‐IUS indicated that LNG‐IUS, DNG, continuous OC, GnRHa + OC and cyclic OC had a significantly lower risk of endometrioma recurrence than expectant management. LNG‐IUS was ranked highest, followed by DNG and GnRHa + LNG‐IUS. Long‐term use of hormonal treatment either OC or progestin had a significantly lower risk of endometrioma recurrence than expectant treatment. Conclusion In the NMA of RCTs, there was no evidence supporting hormonal treatment for postoperative prevention of endometrioma recurrence. This was at odds with the cohort evidence, which found the protective effect of OC and progestin regimens, especially long‐term treatment. Large‐scale RCTs of these agents are still required. Tweetable abstract Hormonal regimens given as long‐term treatment tend to reduce risk of endometrioma recurrence after conservative surgery.


Introduction
Ovarian endometrioma is a major subtype of endometriosis, found up to 55% of women with endometriosis. 1 Symptoms include dysmenorrhoea, dyspareunia, infertility or pelvic fullness/mass, resulting in adverse effects (AEs) on fertility, personal relationships and quality of life. [2][3][4] Ovarian cystectomy is one of the conservative surgical procedures for endometrioma and results in decreased ovarian reserve, especially after re-operating for recurrent disease. 5,6 Endometrioma recurrence rates range between 30 and 50%, 7,8 due to the regrowth of residual lesions or de novo lesion formation. 9 Therefore, postoperative hormonal regimens that suppress ovarian function or reduce menstrual flow may play a role in the prevention of disease recurrence. 10 Combined oral contraceptives (OC) are the regimens most commonly prescribed for prevention of endometrioma recurrence in clinical practice, and have been recommended as the first-line treatments in clinical practice guidelines. [11][12][13] In addition, other types of hormonal treatments (e.g. gonadotropin releasing hormone agonist [GnRHa], depot progestin, dienogest [DNG] or levonorgestrel intrauterine system [LNG-IUS]) are alternative options for women who are sensitive, or have contraindications, to OC.
The efficacy of these treatments in prevention of endometrioma recurrence has been assessed by two pairwise meta-analyses. 14,15 The first meta-analysis (MA), conducted in 2013, 14 combined evidence from one randomized controlled trial [RCT] and three cohorts, which indicated the benefit of long-term (>12 months) use of either cyclic or continuous OCs compared with expectant management. The second meta-analysis, conducted in 2016, 15 pooled evidence from three RCTs and one cohort, indicating no difference in endometrioma recurrence but a significantly lower recurrence of dysmenorrhoea in continuous OC users, compared with cyclic regimens given for at least 6 months postoperation. Although these MAs showed a possible benefit of OCs in the prevention of endometrioma recurrence, evidence was based on small numbers and was not robust. In addition, other hormonal regimens such as DNG, LNG-IUS and GnRHa were not considered in the previous meta-analyses. Given the increased concern about long-term use of various hormonal regimens, the harm-benefit ratios of different OC regimens for prevention of endometrioma recurrence are uncertain. Therefore, the present systematic review and network meta-analysis (NMA) was conducted to estimate the treatment efficacy and safety of the different hormonal regimens used in the prevention of endometrioma recurrence.

Methods
This systematic review and NMA was conducted according to the preferred reporting items for systematic reviews and meta-analyses (PRISMA), extension for network metaanalyses. 16 The review protocol was registered in PROS-PERO (CRD42018105271).

Search strategy
A literature search was performed using MEDLINE and Scopus databases for identification of relevant articles published from inception to 31 January 2020. Search terms and search strategies were constructed based on population (P), intervention (I), comparator (C) and outcomes (O) (Appendix S1). Identified studies were selected by RW and SR based on information from the title and abstract according to selection criteria. Disagreements were resolved through discussion.

Study selection and criteria
All RCTs and cohorts conducted in humans were included if they met all the following criteria: studied in patients with ovarian endometrioma who underwent ovarian cystectomy; compared any pair of the following interventions regardless of dosage, duration of treatment and drug discontinuation: OC, DNG, LNG-IUS, GnRHa and expectant treatment; had any of the following outcomes: endometrioma recurrence, dysmenorrhoea/pelvic pain recurrence or adverse hormonal effects. Studies were excluded if they provided insufficient data for analysis after three attempts to contact the author.
The primary outcome was endometrioma recurrence, defined as ultrasound identification of a round mass with diffuse homogeneous ground-glass echoes, in an individual who had undergone ovarian cystectomy. 17 The secondary outcome was dysmenorrhoea/pelvic pain recurrence, defined as presence of menstrual pain or pain in the pelvic area occurring any time after postoperative pain relief. 18 Adverse hormonal effects including metrorrhagia and amenorrhoea were also considered.

Data extraction
Two reviewers (RW and SR) independently extracted data using a standardised data extraction form; this captured age, revised American Society for Reproductive Medicine (rASRM) score, rASRM staging, characteristics of endometrioma (i.e. size of cyst, bilateral cyst), presence of pelvic adhesion, duration of treatment and follow up. Type of interventions and outcomes of interest along with definitions reported in each study were extracted. Frequency data or summary statistics with standard errors were also extracted for data pooling. Disagreements were discussed and resolved by a third reviewer (AT).

Risk of bias assessment
Two reviewers (RW and SR) independently assessed the quality of the studies. The Cochrane collaboration tool for assessing risk-of-bias version 5.1.0 19 was used for assessment of RCTs in six domains, including selection bias, performance bias, attrition bias, detection bias, reporting bias and other sources of bias. Each item was classified as low, high or unclear risk of bias.
Cohort studies were assessed using the risk of bias in nonrandomised studies of interventions (ROBINS-I) assessment tool, 20 which assessed seven domains: confounding, selection of participants, classification of interventions, deviations from intended interventions, missing data, measurement of outcomes, and selection of the reported result. Each domain was classified as low, moderate, serious, critical risk of bias or no information. Disagreement was resolved by consensus after discussion between both reviewers.

Statistical analysis
Direct meta-analysis (DMA) RCTs and cohort studies were analysed separately. A risk ratio (RR) along with its 95% confidence interval (CI) of each study was estimated from frequency data. In the studies with zero events, a continuity correction was performed by adding 0.5 to all cells, allowing estimation of RR. The RRs were pooled across studies using a fixed-effect model by an inverse-variance method if there was no heterogeneity, otherwise a random-effect model using the DerSimonian and Laird method was applied. Heterogeneity was assessed by Cochrane's Q test and Higgins's I 2 statistic. 21 Publication bias was assessed using funnel plots, Egger's tests 22 and contour-enhanced funnel plots. 23 Network meta-analysis Relative treatment effects between hormonal regimens were compared using an NMA framework. Treatment regimens were numerically coded for expectant management (1), cyclic OC (2), continuous OC (3), GnRHa (4), DNG (5), LNG-IUS (6), GnRHa + OC (7) and GnRHa + LNG-IUS (8). A two-stage approach using multivariate random effects meta-analysis with consistency model was applied to compare treatment efficacy across the network. 24 Multiple treatment comparisons were then estimated.
The inconsistency assumption was checked using an adjusted design-by-treatment interaction model. 25 If there was evidence of inconsistency (P-value for global test < 0.05), a loop-specific approach was used to identify the treatment arms and studies which contributed most to the inconsistency. The probabilities of being the best treatment (lowest RR for recurrence of endometrioma, dysmenorrhoea/pelvic pain and AEs) were estimated and ranked using a rankogram and the surface under the cumulative ranking curve (SUCRA) method. 26 Predictive intervals were estimated and plotted by considering heterogeneity within and between treatment comparisons. Subgroup analysis by treatment continuity up to follow-up time was performed. Publication bias for NMA was assessed using comparisonadjusted funnel plots. 27 If this was present, sensitivity analysis was performed by excluding the studies with low precision (i.e. standard error of effect size larger than 75 percentiles) to see robustness of results.
All analyses were performed using STATA software package, version 15.0 (Stata Corp, College Station, TX, USA). A two-sided P-value of <0.05 was set as the threshold for statistical significance, except for the test of heterogeneity, in which a P-value <0.10 was used.

Patient involvement
There was no patient involvement because this metaanalysis used data from published studies.

Results
A total of 2152 studies were identified from PubMed and Scopus. After deleting duplicates, 24 plus two additional studies identified from reference lists met the inclusion criteria (i.e. eight RCTs and 18 cohorts) and were included for qualitative analysis. After two studies (i.e. RCT 28 and cohort 29 ) were excluded due to insufficient data for pooling, 24 studies remained (i.e. seven RCTs and 17 cohorts) for the quantitative analysis ( Figure 1).

Risk of bias assessment
For RCTs, all studies had low risk for selective outcome reporting. Seven studies (87.5%) had low risk for random sequence generation. Allocation concealment was appropriately performed in five trials (62.5%). All studies had low risk for blinding of participants for both endometrioma and pain recurrence outcomes. For detection bias domain, all studies had low risk for blinding of endometrioma recurrence assessment, whereas all had high risk for blinding of pain recurrence assessment. Five studies (62.5%) had incomplete outcome data. Seven studies (87.5%) were potentially biased from applying per protocol analysis or allowing cross-over of subjects ( Figure S1).   Figure S2).
A sensitivity analysis was performed by excluding two studies 33,36 with continued treatments of continuous OC, 33 cyclic OC 33 (Table S3). This could be interpreted as short-term effects of GnRH (i.e. 3-6 months) being to be more effective than continuous OC or cyclic OC (i.e. 6 months).
Transitivity was further assessed by exploring characteristics for five pairwise comparisons (i.e. cyclic OC versus expectant management, continuous OC versus expectant management, GnRHa versus expectant management, continuous OC versus cyclic OC and GnRHa versus continuous OC) across studies. The results indicated that percent rASRM stage, bilateral cyst, duration of treatment and duration of follow-up were quite different across studies, whereas the rest did not show much difference (Table S4). A comparison-adjusted funnel plot showed no asymmetry, reflecting the absence of any association between study size and study effect ( Figure S6).

Cohorts
DMA was performed and showed that cyclic OC and DNG had respectively about a 64% (RR 0.36, 95% CI 0.18-0.72;  Figure S7). Sources of heterogeneity for comparisons of cyclic OC versus expectant management and GnRHa versus expectant management were explored across studies. The results indicated that none of the baseline characteristics was a source of heterogeneity (Table S5). The results of Egger's test and funnel plot showed no evidence of publication bias for either pooling ( Figure S8 Pooled relative treatment effects demonstrated that LNG-IUS, DNG, continuous OC, GnRHa + OC and cyclic OC had a significantly lower endometrioma recurrence compared with expectant management, with pooled RRs of 0.05 (95% CI 0.00-0.98), 0.14 (95% CI 0.05-0.43), 0.30 (95% CI 0.11-0.77), 0.33 (95% CI 0.15-0.71), and 0.35 (95% CI 0.20-0.60), respectively. LNG-IUS and DNG seemed to be better than other active regimens in lowering endometrioma recurrence, but their effects were significant only when compared with GnRHa, with a pooled RR of 0.05 (95% CI 0.00-0.95) and 0.14 (95% CI 0.04-0.46), respectively. Continuous and cyclic OCs had a significantly lower risk of endometrioma recurrence than GnRHa alone with pooled RRs of 0.29 (95% CI 0.11-0.79) and 0.34 (95% CI 0.17-0.67), respectively. Addition of GnRHa to either OC regimen or LNG-IUS did not significantly alter the treatment effect ( Table 3).

Main findings
We conducted a systematic review and NMA of RCTs and cohorts comparing endometrioma recurrence among various hormonal regimens and expectant management. The pooled RR point estimates from the RCT-NMA indicated that all hormonal regimens could lower endometrioma recurrence compared with expectant management, but none reached statistical significance. Pooled relative treatment effects from cohort NMA found a significantly lower endometrioma recurrence risk in LNG-IUS, DNG, continuous OC, GnRHa + OC and cyclic OC compared with expectant management. Summarising the evidence from RCT NMA, GnRHa + LNG-IUS was the most effective regimen followed by continuous OC and GnRHa. Evidence from the cohort NMA suggested that LNG-IUS ranked first in lowering endometrioma recurrence, followed by DNG and GnRHa + LNG-IUS.
None of the hormonal regimens given as a short-term treatment for about 3-6 months lowered endometrioma recurrence compared with expectant management. However, long-term or continuous use of any hormonal regimen inhibiting ovulation could significantly lower endometrioma recurrence compared with expectant management in which DNG was the most effective regimen, followed by GnRHa + OC.
Relative treatment effects estimated by RCT and cohort NMAs were inconsistent in two comparisons: GnRHa versus expectant management and GnRHa versus cyclic OC (Table S10). Among these comparisons, only GnRHa versus expectant management was directly compared in both RCTs and cohorts, with opposite relative treatment effects. For RCTs, based on reported data, chanracteristics of patients in expectant management and GnRHa groups were quite comparable, except for the percentage of bilateral cyst, which was a bit lower in expectant management than in GnRHa groups. For cohorts, cyst size and percentage of bilateral cyst were higher in GnRHa than in expectant management groups, whereas bilateral cyst and rASRM stage IV were higher in GnRHa than in cyclic OC. This explains why GnRHa decreased the recurrence of endometrioma in RCTs but was a higher risk in cohorts (Table S11). However, this was only considered based on the data available, as not all RCTs/cohorts reported baseline risks (Table S12).

Strengths and limitations
The strength of our study is that we considered most current regimens for prevention of endometrioma recurrence. Treatments other than OC were added to those from the previous studies; 14,15 application of the NMA framework allowed multiple relative treatment comparisons and ranked the best agents for preventing endometrioma recurrence given the evidence to date from both RCTs and cohorts.
There were some limitations that we could not avoid. First, there was a limited number of relevant studies and their sample sizes were rather small, especially in the RCT network, which might cause uncertainty of the estimated treatment effect, low power to detect the consistency assumption by global test, and limited generalisability of our findings.
Second, duration of treatment and follow up among regimens varied greatly, which may potentially affect the outcome. However, subgroup analysis by continuity of treatment up to end of follow up was performed to assess effects of continuity of treatments.
Third, other few important outcomes such as endometrioma pain and recurrence that required surgery were not considered due to a small number of studies reporting these outcomes.
Finally, most RCTs were potentially biased because they applied per protocol analysis.

Interpretation
Our findings from RCT and cohort NMA confirm and extend the previous two meta-analyses 14,15 that supported the benefit of OCs in the prevention of endometrioma recurrence. Apart from cyclic and continuous OCs, hormonal regimens involving GnRHa + LNG-IUS and GnRHa seemed to decrease endometrioma recurrence better than expectant management does, although none of these was statistically significant. However, the duration of GnRHa administration should be no longer than 6 months due to unfavourable effects (e.g. decreased bone mineral density, menopausal symptoms, etc.). Their effects in the suppression of ovarian function will vanish as soon as the treatment are discontinued. Therefore, continuity and safe longterm treatments are still required. Large-scale RCTs considering both efficacy and safety are still required to confirm The evidence from the cohort NMA found that LNG-IUS and DNG were the top two ranked treatments, lowering the risk of endometrioma by 95 and 86%, compared with expectant management, respectively. Although the estimated treatment effects were significant, the results were derived from few studies. Moreover, cohort designs are more susceptible to selection bias. To confirm the protective effect of these regimens, more high-quality studies are needed. Combining RCT and cohort data using a hierarchical mixed-effect logit model and adjusting for study design, 53 suggested that DNG was still better than expectant management and other hormonal regimens (Table S13).
Our findings complied with clinical practice guidelines 11-13 that only long-term hormonal treatment can prevent endometrioma recurrence. All hormonal regimens inhibiting ovulation are better than no treatment, therefore cost and AEs of each regimen should be considered in practice. The ovulation inhibition effect of LNG-IUS is theoretically less than 50% after 3 months of insertion, 54 which is the main mechanism of endometrioma. 55,56 However, our findings showed that continuous treatment with LNG-IUS might be beneficial. Therefore, there may be other possible mechanisms that LNG-IUS might work through and which would be of benefit in the prevention of endometrioma recurrence. 57,58

Conclusion
The best evidence derived from the RCT network suggested that GnRHa plus LNG-IUS was the most effective regimen in lowering risk of endometrioma recurrence, followed by continuous OC and GnRHa. Nevertheless, this was based on nonsignificant treatment effects, perhaps because of the small number of RCTs. Long-term use of DNG had a favourable effect in prevention of endometrioma recurrence, but the evidence was derived from a cohort network. Therefore, applying these treatments for prevention of endometrioma recurrence should be considered carefully for individual patients.

Disclosure of interests
None declared. Completed disclosure of interests forms are available to view online as supporting information.
Contribution to authorship RW, SR, CC and AT were involved in the conception and design of the protocol. RW, SR and AT designed the search strategy. RW and SR selected eligible studies, extracted data and assessed the quality of the included studies. Data analysis and interpretation were carried out by RW, SR and AT. Drafts of the manuscript were prepared by RW and SR, and critically revised by JA and AT. All authors contributed to the drafts and final version of the manuscript and approved the final review.

Details of ethics approval
Not applicable.

Funding
This study received no funding.

Supporting Information
Additional supporting information may be found online in the Supporting Information section at the end of the article.
Appendix S1. Search strategies and results MEDLINE via PubMed and Scopus. Figure S1. Risk of bias assessments for RCTs. Figure S2. Forest plot of pairwise meta-analysis of endometrioma recurrence based on RCT network. Figure S3. Network map of RCTs comparing hormonal treatments for prevention of endometrioma recurrence. Figure S4. Network contribution plot of RCTs for endometrioma recurrence outcome. Figure S5. Rankograms for hormonal network of RCTs showing the probability for each regimen being at a particular order in lowering endometrioma recurrence. Figure S6. Comparison-adjusted funnel plot for network meta-analysis of RCTs on endometrioma recurrence outcome. Figure S7. Forest plot of pairwise meta-analysis of endometrioma recurrence based on cohort network. Figure S8. Funnel plot of cyclic OC versus expectant management and GnRHa versus expectant management for endometrioma recurrence based on cohort data. Figure S9. Network map of cohorts comparing among hormonal treatments for prevention of endometrioma recurrence. Figure S10. Network contribution plot of cohorts for endometrioma recurrence outcome. Figure S11. Rankograms for hormonal network of cohorts showing the probability for each regimen being at a particular order in lowering endometrioma recurrence. Figure S12. Comparison-adjusted funnel plot for network meta-analysis of cohorts on endometrioma recurrence outcome. Table S1. Characteristics of included cohorts. Table S2. Risk of bias assessment of included cohort studies. Table S3. Multiple treatment comparison of RCTs network for endometrioma recurrence outcome: a sensitivity analysis by excluding two studies with continued hormonal treatments  Table S4. Distribution of effect modifiers between each pairwise comparison across network of RCTs for endometrioma recurrence. Table S5. Explore sources of heterogeneity: meta-regression Table S6. Distribution of effect modifiers between each pairwise comparison across network of cohorts for endometrioma recurrence. Table S7. Multiple treatment comparison of cohort network for endometrioma recurrence outcome in a subgroup continued hormonal treatment group. Table S8. Multiple treatment comparison of cohort network for endometrioma recurrence outcome in a subgroup continued hormonal treatment group with removing LNG-IUS. Table S9. Multiple treatment comparison of cohort network for endometrioma recurrence outcome in a subgroup with discontinued treatment up to last follow up. Table S10. Relative effect estimates of treatment comparisons of RCT and cohort network for endometrioma recurrence.
Table S11. Characteristics of subjects and RRs (95% CI) in GnRHa versus expectant management comparison; RCT and cohort study. Table S12. Characteristics of subjects in expectant management, cyclic OC and GnRHa intervention among RCTs and cohort studies. Table S13. Estimation of treatment effects on recurrence adjusting for study designs: a multilevel, mixed-effect logit model.