Oral dydrogesterone vs. vaginal progesterone capsules for luteal-phase support in women undergoing embryo transfer: a systematic review and meta-analysis.

Progesterone is needed to maintain early pregnancy, and progesterone supplementation in assisted reproduction technology (ART) cycles is a well-accepted procedure ( Shapiro et al ., 2014 ; Holmdahl et al ., 1971 ; van der Linden et al ., 2015 ). Luteal phase deficiency affects women undergoing ART for many reasons. The most widely accepted theory posits that luteal phase deficiency originates from premature negative feedback on LH secretion in the pituitary caused by supra-physiological levels of steroids during controlled ovarian stimulation (COS) sustained after oocyte aspiration by multiple corpora lutea ( van der Linden et al ., 2015 ; Fatemi 2009 ). There is evidence that luteal phase support (LPS) with progesterone, human chorionic gonadotropin (hCG) or gonadotropin-releasing hormone (GnRH) agonists improves reproductive outcomes in women undergoing in vitro fertilization (IVF) ( Shapiro et al ., 2014 ; van der Linden et al ., 2015 ; Fatemi et al ., 2007 ; Vaisbuch et al ., 2012 ; Merriam et al ., 2015 ; Martins et al ., 2016 ). Since hCG correlates with higher risk of ovarian hyper stimulation syndrome ( van der Linden et al ., 2015 ; Fatemi et al ., 2007 ; Vaisbuch et al ., 2012 ) and evidence of the benefits of GnRH agonists is still of very low quality ( Martins et al ., 2016 ), progesterone appears to be the best option for LPS. Progesterone can be administered orally, intramuscularly, vaginally or rectally; all routes seem to present similar levels of efficacy ( Shapiro et al ., 2014 ; van der Linden et al ., 2015 ; Vaisbuch et al ., 2012 ; Merriam et al ., 2015 ). First-pass metabolism substantially reduces the bioavailability of oral progesterone to <10% ( Nahoul et al ., 1993 ). Intramuscular progesterone has been associated with pain caused by daily injections, inflammatory response, and local abscess ( van der Linden et al ., 2015 ; Fatemi et al ., 2007 ; Vaisbuch et al ., 2012 ; Ghanem & Al-Boghdady, 2012 ). Although fewer adverse events are observed with the vaginal route ( Maher et al ., 2013 ), vaginal progesterone causes local irritation, discharge, and bleeding; it is also affected by coitus, since absorption is decreased after intercourse ( Merriam et al ., 2015 ; Ghanem & Al-Boghdady, 2012 ). Dydrogesterone is a synthetic progestin with enhanced oral bioavailability, known for being highly selective for the progesterone receptor ( Kupferminc et al ., 1990 ; Domitrz et al ., 1999 ). It is effective in treating reproductive disorders such as threatened abortion and recurrent pregnancy loss, and has also been investigated in the prevention of gestational hypertension and preterm birth ( Carp, 2012 ; 2015 ; Hudic et al ., 2016 ; Mohamad Razi et al ., 2016 ). Dydrogesterone has also been described to provide similar reproductive results as vaginal progesterone ( van der Linden et al ., 2015 ; Barbosa et al ., 2016 ). The oral route of administration is thought to be a more patient-friendly regimen that might improve compliance to treatment. The objective of this systematic review and meta-analysis was to identify, appraise, and summarize the evidence from randomized controlled trials examining the efficacy, safety, and tolerability of oral dydrogesterone compared to vaginal progesterone capsules for LPS in women undergoing ART.

An electronic search run in June 7, 2017 retrieved 376 records (PubMed = 77; Scopus = 216; Embase = 83). Additional papers hand-searched from the references of the included studies or related reviews were not included. Three hundred and four papers were excluded after their titles and abstracts were read: 128 were duplicates and 238 clearly did not meet the eligibility criteria. Ten studies were further examined for eligibility: one study was excluded because it compared dydrogesterone with vaginal progesterone capsules for luteal support in IUI cycles ( Khosravi et al ., 2015 ). Nine studies were included in our quantitative analysis ( Chakravarty et al ., 2005a ; Ganesh et al ., 2011 ; Patki & Pawar, 2007 ; Rashidi et al ., 2016 ; Saharkhiz et al ., 2016 ; Salehpour et al ., 2013 ; Tournaye et al ., 2017 ; Zarei et al ., 2017 ; Zargar et al ., 2016 ); four of the nine studies had groups given medication other than oral dydrogesterone and vaginal progesterone capsules: vaginal progesterone gel ( Ganesh et al ., 2011 ); intramuscular progesterone ( Rashidi et al ., 2016 ; Zargar et al ., 2016 ); dydrogesterone combined with either GnRH agonist or hCG ( Zarei et al ., 2017 ). The individuals in these groups were not included in the quantitative analysis. Figure 1 shows the study flow diagram. Figure 1 Flowchart of study selection. Flowchart of study selection. The characteristics of the nine parallel studies included in the quantitative analysis are reported in Table 1 . One study was held in two centers ( Saharkhiz et al ., 2016 ), one in 38 different sites ( Tournaye et al ., 2017 ), and the remaining seven were carried out in single centers. Five studies were conducted in Iran ( Rashidi et al ., 2016 ; Saharkhiz et al ., 2016 ; Salehpour et al ., 2013 ; Zarei et al ., 2017 ; Zargar et al ., 2016 ), three in India ( Chakravarty et al ., 2005a ; Ganesh et al ., 2011 ; Patki & Pawar, 2007 ), and one in multiple centers in Austria, Belgium, Germany, Finland, Israel, Russia, and Spain ( Tournaye et al ., 2017 ). Despite attempts to contact the authors of all studies, additional details were collected from only one study ( Saharkhiz et al ., 2016 ). All studies were published as full articles. Only patients using vaginal progesterone capsules were used in the comparisons. Characteristics of the included studies. NR = not reported; ART = assisted reproductive technology; FET = frozen embryo transfer; VP = vaginal progesterone; * = study vs. control NOTES: One study reported a conflict of interest  ( Tournaye et al. , 2017 ); Two studies reported funding sources ( Tournaye et al. , 2017 ; Zarei et al. , 2017 ); All studies obtained approval from ethics committees, and one study did not provide informed consent  ( Chakravarty  et al. , 2005a ); All were parallel studies; Two studies reported an mean of three embryos transferred ( Ganesh et al. , 2011 ; Salehpour et al. , 2013 ), and one study reported an mean of two embryos transferred  ( Rashidi et al. , 2016 ). Participants: 4,061 women submitted to ART in nine studies were included; 1,905 were allocated to groups prescribed dydrogesterone for luteal phase supplementation, and 2,156 were allocated to groups on vaginal progesterone capsules. The eligibility criteria, and therefore the characteristics of the included participants, were different across studies and are reported on Table 1 . Interventions: The nine studies assessed the use of daily oral dydrogesterone in doses ranging from 20mg to 40mg versus vaginal progesterone capsules in doses ranging from 600 mg/day to 800 mg/day. Outcomes: Two of nine studies reported live births ( Rashidi et al ., 2016 ; Tournaye et al ., 2017 ); 3/9 reported ongoing pregnancies ( Chakravarty et al ., 2005a ; Saharkhiz et al ., 2016 ; Zarei et al ., 2017 ); 8/9 reported clinical pregnancies ( Ganesh et al ., 2011 ; Patki & Pawar, 2007 ; Rashidi et al ., 2016 ; Saharkhiz et al ., 2016 ; Salehpour et al ., 2013 ; Tournaye et al ., 2017 ; Zarei et al ., 2017 ; Zargar et al ., 2016 ); 7/9 reported miscarriages ( Chakravarty et al ., 2005a ; Ganesh et al ., 2011 ; Rashidi et al ., 2016 ; Saharkhiz et al ., 2016 ; Salehpour et al ., 2013 ; Zarei et al ., 2017 ; Zargar et al ., 2016 ); 2/9 reported female patient dissatisfaction ( Chakravarty et al ., 2005a ; Saharkhiz et al ., 2016 ); and 3/9 reported side effects ( Saharkhiz et al ., 2016 ; Salehpour et al ., 2013 ; Tournaye et al ., 2017 ). In one study, the number of clinical pregnancies was assumed to be equal to the summation of ongoing pregnancies and miscarriages ( Chakravarty et al ., 2005a ). In three studies, the number of ongoing pregnancies was assumed to be equal to the number of clinical pregnancies minus miscarriages ( Ganesh et al ., 2011 ; Salehpour et al ., 2013 ; Zargar et al ., 2016 ). Six studies described adequate methods of randomization ( Ganesh et al ., 2011 ; Rashidi et al ., 2016 ; Saharkhiz et al ., 2016 ; Salehpour et al ., 2013 ; Tournaye et al ., 2017 ; Zarei et al ., 2017 ; Zargar et al ., 2016 ) and two studies did not report the method used ( Chakravarty et al ., 2005a ; Patki & Pawar, 2007 ). Six studies described allocation concealment through sealed envelopes ( Ganesh et al ., 2011 ; Rashidi et al ., 2016 ; Saharkhiz et al ., 2016 ; Salehpour et al ., 2013 ; Tournaye et al ., 2017 ; Zargar et al ., 2016 ). One study blinded participants and care providers ( Tournaye et al ., 2017 ). In six studies outcome assessors were blinded to allocation ( Ganesh et al ., 2011 ; Rashidi et al ., 2016 ; Saharkhiz et al ., 2016 ; Salehpour et al ., 2013 ; Tournaye et al ., 2017 ; Zargar et al ., 2016 ) and the remaining three studies did not report whether outcome assessors were blinded. Saharkhiz et al ., 2016 was judged to be at high risk of attrition bias, since 24/234 (10.3%) participants were excluded after randomization; loss to follow up was unbalanced between groups, with 21/117 (17.9%) participants in the dydrogesterone group and 3/117 (2.6%) in the progesterone group. The other eight studies were judged to be at a low risk of attrition bias. Five studies analyzed all randomized women ( Chakravarty et al ., 2005a ; Ganesh et al ., 2011 ; Patki & Pawar, 2007 ; Salehpour et al ., 2013 ; Zargar et al ., 2016 ). Rashidi et al . (2016 ) excluded one of 120 participants from the analysis because she failed to come to embryo transfer due to a car accident; this study was deemed to present low risk of attrition bias since the withdrawal rate was low. Tournaye et al . (2017) excluded 57/1031 (5.5%) participants after randomization; loss to follow-up was balanced between groups - 23/520 (4.4%) in the dydrogesterone group and 34/511 (6.6%) in the progesterone group - and the study was considered to present low risk of attrition bias. For the same reasons the study by Zarei et al ., (2017) was assigned low risk of attrition bias: 22/222 (10%) participants were excluded after randomization, but loss to follow-up was balanced between groups, with 10/110 in the dydrogesterone group and 12/112 in the progesterone group. The study by Zargar et al . (2016 ) was judged to present high risk of selective reporting bias, as three outcomes described in the registered protocol were not reported (live birth, preterm delivery, and perineal irritation caused by vaginal progesterone). Four studies reported all outcomes described in the registered protocol ( Rashidi et al ., 2016 ; Saharkhiz et al ., 2016 ; Salehpour et al ., 2013 ; Tournaye et al ., 2017 ) and the remaining four were not assessed as presenting selective reporting bias. One study was deemed at high risk of bias for containing a larger proportion of women aged 40+ years in the dydrogesterone group ( Chakravarty et al ., 2005a ). There was no suspicion of other sources of bias in the other eight studies. Forest plots were used to show the results of each individual study and their respective possible biases ( Figures 2 - 4 ) Figure 2 Forest plot for live birth/ongoing pregnancy. Risk of bias legend: A = Selection bias (random sequence generation); B = Selection bias (allocation concealment); C = Performance bias; D = Detection bias; E = Attrition bias; F = Reporting bias; G = Other biases. Forest plot for live birth/ongoing pregnancy. Risk of bias legend: A = Selection bias (random sequence generation); B = Selection bias (allocation concealment); C = Performance bias; D = Detection bias; E = Attrition bias; F = Reporting bias; G = Other biases. Figure 4 Forest plot for miscarriage. Risk of bias legend: A = Selection bias (random sequence generation); B = Selection bias (allocation concealment); C = Performance bias; D = Detection bias; E = Attrition bias; F = Reporting bias; G = Other biases. Forest plot for miscarriage. Risk of bias legend: A = Selection bias (random sequence generation); B = Selection bias (allocation concealment); C = Performance bias; D = Detection bias; E = Attrition bias; F = Reporting bias; G = Other biases. Live birth / Ongoing pregnancy ( Table 2 ) Summary of findings. All outcomes, except miscarriage, were analyzed per randomized women. CI = confidence interval; RR = relative risk; a = The absolute risk in the Vaginal Progesterone group was determined as the mean risk in these groups; the absolute risk in the Oral Dydrogesterone group and its 95% CI was determined using the RR and its 95% CI; 1 . Downgraded one level because of imprecision. Overall, there was no evidence of relevant differences between oral dydrogesterone and vaginal progesterone on live birth/ongoing pregnancy rates: RR 1.08; 95% CI 0.92 to 1.26; I 2 =29%, 8 RCTs, 3,386 women; high quality evidence. In other words, considering a live birth/ongoing pregnancy rate of 24% in women using vaginal progesterone, this rate would be in the range of 22-30% in women using oral dydrogesterone. Sensitivity analysis excluding the three studies at high risk of bias did not change the estimate: RR 1.10; 95% CI 0.86 to 1.40; I 2 =48%, 5 RCTs, 2,334 women. Clinical pregnancy ( Figure 3 ) Figure 3 Forest plot for clinical pregnancy. Risk of bias legend: A = f4 Selection bias (random sequence generation); B = Selection bias (allocation concealment); C = Performance bias; D = Detection bias; E = Attrition bias; F = Reporting bias; G = Other biases. Forest plot for clinical pregnancy. Risk of bias legend: A = f4 Selection bias (random sequence generation); B = Selection bias (allocation concealment); C = Performance bias; D = Detection bias; E = Attrition bias; F = Reporting bias; G = Other biases. Overall, there was no evidence indicating that clinical pregnancy was affected by the use of oral dydrogesterone versus vaginal progesterone capsules: RR 1.10, 95% CI 0.95 to 1.27; I 2 =43%; 9 RCTs; 4,061 women; high quality evidence. If 28% of the women using vaginal progesterone became pregnant, 27-36% of the women using oral dydrogesterone might also be clinically pregnant. Sensitivity analysis excluding the three studies at high risk of bias did not change the estimate: RR 1.08; 95% CI 0.86 to 1.36; I 2 = 51%, 5 RCTs, 2,334 women. Miscarriage ( Figure 4 ) Overall, there was no evidence indicating that miscarriage was affected by the use of oral dydrogesterone versus vaginal progesterone: RR=0.92, 95%CI=0.68-1.26, I 2 =6%, 8 RCTs, 988 clinical pregnancies; moderate quality evidence. Two studies reported patient dissatisfaction with treatment ( Chakravarty et al ., 2005a ; Saharkhiz et al ., 2016 ) (26,30). Since the two studies were significantly heterogeneous (I 2 = 91%), their results were not pooled together. Saharkhiz et al . (2016) reported no difference in dissatisfaction between groups: 8% in women using dydrogesterone vs. 7% in women using vaginal progesterone capsules; RR 1.19, 95% CI 0.46 to 3.04; 210 women. This study was deemed to be at high risk of bias. Chakravarty et al . (2005a) described a great benefit of dydrogesterone in reducing patient dissatisfaction: 3% in women using dydrogesterone vs. 26% in women using vaginal progesterone capsules; RR 0.10, 95% CI 0.02 to 0.39; 430 women. Substantial heterogeneity (>50%) was found for all side effects reported by the three studies and therefore the results were not pooled together. While two studies did not describe differences in reported side effects between the two groups ( Saharkhiz et al ., 2016 ; Tournaye et al ., 2017 ), one study showed that dydrogesterone was associated with more cases of vaginal bleeding (RR 2.38; 95% CI 1.18 to 4.78), nausea (RR 21.00; 95% CI 1.27 to 346.66), and abdominal pain (RR 13.00; 95% CI 0.76 to 223.33) when compared to vaginal progesterone capsules. Although suboptimal, since fewer than 10 studies were included, the funnel-plot analysis for the only outcome reported in the nine studies - clinical pregnancy - was not suggestive of publication bias ( Figure 5 ). Figure 5 Funnel plot analysis for clinical pregnancy. Funnel plot analysis for clinical pregnancy. Sensitivity analysis was reported along with the synthesis of the results.

Nine studies were included in the comparison between oral dydrogesterone and vaginal progesterone capsules. Oral dydrogesterone was generally as effective as vaginal progesterone capsules for luteal phase support in women undergoing embryo transfers after IVF/ICSI. The assessment of patient dissatisfaction with treatment revealed an important inconsistency between the two studies reporting this outcome: one reported a significant difference favoring dydrogesterone ( Chakravarty et al ., 2005a ) while the other found no differences between the regimens ( Saharkhiz et al ., 2016 ). Possible explanations for this discrepancy are the different doses of dydrogesterone and the potential differences in the characteristics of the two patient populations. Our findings were in agreement with the latest Cochrane review on the subject, which suggested a significant effect in favor of synthetic progesterone versus natural progesterone ( van der Linden et al ., 2015 ). Four studies were included in the comparison, three of which also included in our review ( Chakravarty et al ., 2005 a; Ganesh et al ., 2011 ; Patki & Pawar, 2007 ). The other study was not included in our review because it compared oral chlormadinone acetate to intramuscular progesterone ( Iwase et al ., 2008 ). Another recent review showed that dydrogesterone provides similar reproductive results when compared to vaginal progesterone ( Barbosa et al ., 2016 ). Seven studies were included in this review ( Chakravarty et al ., 2005a ; b; et al . , 2006 ; Ganesh et al ., 2011 ; Patki & Pawar, 2007 ; Saharkhiz et al. , 2016 ; Salehpour et al ., 2013 ), five of which were also included in our review ( Chakravarty et al ., 2005a ; Ganesh et al ., 2011 ; Patki & Pawar, 2007 ; Saharkhiz et al ., 2016 ; Salehpour et al ., 2013 ). Two of the studies were not included in our review because they were published as abstracts, thus yielding a high risk of bias to the comparison ( Chakravarty et al ., 2005 b ; 2006 ). Four other studies were included in our review ( Rashidi et al ., 2016 ; Tournaye et al ., 2017 ; Zarei et al ., 2017 ; Zargar et al ., 2016 ). One of these studies was sponsored by a pharmaceutical company ( Tournaye et al ., 2017 ), but its results were similar to the one described in other trials. The only difference between the study by Tournaye et al . (2017) and the others included was the double-blinding procedure, which in fact minimizes the risk of bias ( Lexchin et al ., 2003 ). With the addition of more studies in our review, and by excluding the abstracts, the authors believe that this review provides a robust body of evidence for the comparison between dydrogesterone and vaginal progesterone capsules for LPS in women undergoing embryo transfers. In terms of dissatisfaction with treatment, our review included the same studies as the cited review ( Barbosa et al ., 2016 ). The discrepancy between the two studies in regards to this outcomes makes it difficult to draw firm conclusions. Different side effects were reported in three studies ( Saharkhiz et al ., 2016 ; Salehpour et al ., 2013 ; Tournaye et al ., 2017 ), and two of them did not report differences between the two groups ( Saharkhiz et al ., 2016 ; Tournaye et al ., 2017 ). Additionally, a systematic review on the use of dydrogesterone for recurrent miscarriage found 13 studies reporting apparently minimal adverse effects ( Carp, 2015 ). The evidence available suffers from the limitations inherent to the included studies: five of the nine studies had high risk of bias in at least one domain; and the use of different doses in case and control groups along with different durations of LPS may have introduced some heterogeneity in the analysis. This issue was addressed with a random-effects model and by the incorporation of observed heterogeneity in the interpretation of the findings and in the assessment of the quality of the evidence. The quality of the evidence was considered to be high for live birth/ongoing pregnancy and clinical pregnancy. It was downgraded one level for miscarriage because of imprecision: there was a relatively low number of events and a broad confidence interval.

Oral dydrogesterone is as effective as vaginal progesterone capsules for luteal-phase supplementation in ART cycles. Oral dydrogesterone might be a good option in clinical practice, since oral administration is more patient-friendly than the vaginal route. The choice for either should be based mainly on availability, cost, and side effects.

The protocol of this review was registered at PROSPERO(CRD42017071571). True randomized controlled trials (RCTs) comparing oral dydrogesterone to vaginal progesterone capsules for LPS in women undergoing ART (fresh or frozen embryo transfer following IVF/ICSI) were included. Quasi and pseudo-randomized trials were not included. The following electronic databases were searched for RCTs: PubMed, Scopus, and Embase. The references of the included studies and related reviews were also hand-searched. The following terms were used, adjusting for each database as necessary: (IVF OR ICSI OR embryo OR blastocyst OR oocyte OR egg OR retrieval OR luteal) AND (dydrogesterone OR duphaston OR isopregnenone OR dehydrogesterone). There was no limitation regarding language, publication date or publication status. Two authors (MWPB and CON) independently screened publications for titles and abstracts based on the pre-established inclusion criteria and checked for duplicates. The same authors examined the full text articles of the studies selected for inclusion in the review; a third author (WPM) was involved to solve disagreements as needed. The authors corresponded with original study authors to clarify study eligibility when required. A data extraction form designed and pilot-tested by the authors was used to extract data from the included trials. In the event of studies with multiple publications, the main trial report was used as reference and additional details were supplemented from secondary reports. The authors corresponded with trial authors to get clarification when required. Data were extracted independently in a standardized manner by two authors (MWPB and CON); a third author (WPM) was involved to solve disagreements as needed. The following data were collected to characterize the included trials: authors; country; institution; funding sources; conflicts of interest; informed consent; approval by ethics committees; study design; period of enrollment; eligibility criteria; number of participants in each group at each stage; age and BMI (mean±SD) of participants; COS protocol and trigger; number of embryos transferred per woman; and implantation rate. The primary outcomes for effectiveness were live birth and/or ongoing pregnancy rates, while the primary outcome for adverse effect was dissatisfaction. Ongoing pregnancy was used a surrogate indicator of live birth in trials not reporting the latter. Ongoing pregnancy was defined as evidence of fetal cardiac activity on ultrasound examination after 10-12 weeks of gestation ( Daya, 2003 ). Ongoing pregnancy was calculated as the number of clinical pregnancies minus the number of miscarriages in the trials in which it was not described. Secondary outcomes were clinical pregnancy; miscarriage per clinical pregnancy (single fetal demise in twin or triplet pregnancies was not counted as miscarriage); and any reported side effects. Additional unreported data were collected from the authors of the studies. Where data could not be obtained, clinical pregnancy (and subsequent miscarriage or live birth) was assumed not to have occurred in women with cycle cancellation. No assumption was made for women lost to follow up for other reasons. Two authors (MWPB and CON) independently assessed the risk of selection bias (random sequence generation and allocation concealment); performance bias (blinding of participants and personnel); detection bias (blinding of outcome assessors); attrition bias (incomplete outcome data); reporting bias (selective outcome reporting), and other potential sources of bias (e.g.: difference in the number of embryos transferred, age of participants, co-interventions, early stopping). A third author (WPM) was involved to solve disagreements as needed. The Cochrane Collaboration criteria for judging risk of bias was used in this review ( Higgins & Green, 2011 ): the trials were assigned 'low', 'high' or 'unclear' risk of bias. Blinding was not considered as a factor likely to affect the risk of performance and detection bias on reproductive outcomes, but it might be detrimental to the evaluation of participant satisfaction with treatment, since the main adverse effects related to the route of drug administration. Dichotomous variables were expressed as risk ratios (RR) and the precision of the estimates was evaluated by the 95% confidence interval (CI). The clinical relevance of all comparisons was assessed based on the precision of the estimates. A random effects model was used to address the differences in true effect size across studies, since doses were different. The random effects model also incorporated the heterogeneity observed among studies and thus produced more conservative confidence intervals ( Higgins & Green, 2011 ). Review Manager 5.3.5 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014) was used to combine the results comprised in the meta-analysis. The I 2 index was used to assess heterogeneity. Increases in the risk of positive (e.g.: live birth) or negative (e.g.: miscarriage) outcomes in the meta-analysis were plotted to the right of the centerline, while decreases in the risk such outcomes were plotted to the left of the centerline. Since one multi-arm study was included, we were careful not to double count controls. In view of the difficulty detecting and correcting for publication bias and other reporting biases, the authors aimed to minimize the potential impact by performing a comprehensive search for eligible studies and by preventing the duplication of data. Additionally, a funnel plot was used to assess the presence of small-study effects suggestive of publication bias. Sensitivity analysis was performed for primary outcomes to verify whether the conclusions would have been different if eligibility was restricted to studies at low risk of bias. A table was generated to summarize the review findings. The quality of the evidence for the main outcomes was evaluated following the Grading of Recommendations Assessment, Development and Evaluation (GRADE) Working Group recommendation ( Guyatt et al ., 2011 ): the limitations of included studies, inconsistency of effect, imprecision, indirectness, and risk of publication bias were considered. The quality of the evidence was graded in the following levels ( Balshem et al ., 2011 ): High quality = We are very confident that the true effect lies close to the effect observed in this review; Moderate quality = We are moderately confident in the effect estimate: the true effect is likely to be close to the effect observed in this review, but it might be substantially different; Low quality = Our confidence in the effect estimate is limited: the true effect may be substantially different from the effect observed in this review; Very low quality = We have very little confidence in the effect estimate: the true effect is likely to be substantially different from the effect observed in this review.