Introduction

Dopamine agonists (DAs) are the first-line treatment for hyperprolactinemia, they can effectively reduce prolactin (PRL) level and tumor size [1,2,3,4,5]. Bromocriptine (BRC) and cabergoline (CAB) are the most commonly used DAs. However, after treatment, sudden stop of DA administration may run the risk of tumor re-growth and recurrence [6,7,8]. Therefore, the question of how to reasonably reduce the dose of DAs and achieve a complete withdrawal is a critical clinical issue of great significance.

The 2005 edition Guidelines of the Pituitary Society for the diagnosis and management of prolactinomas [9] pointed out that a complete drug withdrawal in some patients was possible, and in patients with normalized PRL levels, and significant reduction in tumor volume, DA treatment should last more than 1 year. The end point of this guideline is to suggest the best timing of withdrawal in order to minimize recurrences.

Recently, the United States Endocrine Society emphasized the importance of “withdrawal standard” [1]. According to the systematic review by Dekkers et al. [10] and other three supportive studies [11,12,13], patients should receive DA therapy for at least 2 years until their serum PRL levels got normalized and there was no residual tumor on magnetic resonance imaging (MRI). Obviously, the withdrawal standard of this guideline emphasizes the importance of prolonged DA treatment and tumor disappearance. In addition, the use of CAB prior to other DAs is also recommended, since CAB can help reduce the PRL level and tumor volume in a more efficient way.

The aim of this systematic review and meta-analysis is to discuss the optimal timing of DA withdrawal in hyperprolactinemia patients, by reviewing the past and the current studies to provide useful clues for decision making of DA withdrawal in clinical practice.

Subjects and methods

Search strategy

First search was performed in PubMed, Cochrane, and EMBASE databases. The references in the relevant papers were also examined. Unpublished results were not included in this analysis. The assessment of successful withdrawal was mainly based on PRL levels. Clinical symptoms and tumor size were regarded as auxiliary references.

Inclusion criteria

The validity of studies included in this review can be referred to the criteria in Table 1.

Table 1 Inclusion criteria
Full size table

Methods

Whenever possible, each article was stratified by idiopathic hyperprolactinemia, microprolactinomas and macroprolactinomas, BRC, and CAB, respectively. If a subgroup of a cohort was withdrawn from DA treatment, data was extracted for this subgroup only.

These are two main units when measuring prolactin levels (mIU/l and ng/ml). All PRL levels in this meta-analysis were reported in nanograms per milliliter. It is impossible to acquire the conversion factors from all the assays. The conversion factor between them is 21.2 (as shown in http://prolactinemia.com/info/levels-units/). The tumor regression during DA treatment was determined by MRI and computed tomography (CT). Two independent reviewers (M.Y.X. and X.H.L.) were responsible for the study selection and data extraction. Any disagreement that occurred would be resolved by a group discussion and negotiation.

Statistical methods

The weighted average of the proportion of patients with persisting normoprolactinemia after DA withdrawal was the main outcome of this meta-analysis. We used Cohen’s as the standardized mean difference calculation method. Since the original data did not conform to the normal distribution, all the statistical results were processed by double inverse sine statistical method. Multivariate regression analysis would be used to determine which of the independent variables are significant, and which are not. Sensitivity analysis would be conducted on low-quality studies. Heterogeneity degree was carried out for each meta-analysis using the I-squared statistic (I-squared < 25% and I-squared > 50% reflect small and large inconsistencies, respectively [14]).The Q statistic test would be used to assess the heterogeneity. And the Egger’s test would be used to investigate whether there was a publication bias. A p-value < 0.05 was considered to be statistically significant. Based on the results of meta-analysis, several subgroup analyses were performed. Review Manager (version 5.3.5; the cochrane collaboration) and STATA (version 13.0; Stata Corporation, College Station, TX) were used for carrying out these statistical analyses.

Results

Search results

First search in PubMed, Cochrane, and EMBASE databases up to May 2016 retrieved a total of 795 articles by using following free terms: “dopamine agonist, withdrawl”, “Bromocriptine, withdrawal”, “Cabergoline, withdrawl”, and “hyperprolactinemia, withdrawal”. Among them, 519 articles were excluded because of insufficient correlations with the theme. And an additional 198 papers were excluded based on the duplication. Moreover, another 49 articles were also excluded from further analysis due to the lack of specific experimental data, preliminary experimental results, or patient information. Also, two articles were excluded because they focused on the second time of drug withdrawal [15, 16]. Three more studies were eliminated for using different drugs: quinagolide, dihydroergocriptine, and lisuride [17,18,19]. One paper was added after check of ref. [20]. One paper mixed with patients with empty sella was also removed [21]. Finally, 24 potentially relevant papers [11, 12, 20, 22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42] were retrieved for full assessment (Fig. 1).

Fig. 1
figure 1

Summary of study assessment and exclusion. The search in PubMed, Cochrane, and EMBASE databases yielded a total of 794 articles. After screening, 24 potentially relevant papers were retrieved for full assessment consequently

Full size image

Study characteristics

Details of the 24 articles are summarized in Tables 2a, 2b and 2c. The time span ranged 37 years from 1979 to 2016, and the number of patients involved in each study varied from 2 to 221. The total number of patients in this meta-analysis was 1106. Stratified data was available for a total of 727 microprolactinoma patients and 306 macroprolactinoma patients. While a paper [32] engaged a total of 32 patients, the patients could not be separated concerning different etiologies. The proportion of patients with persisting normoprolactinemia after DA withdrawal ranged from 0 to 85% in different studies.

Table 2a Characteristics for included studies in patients with BRC
Full size table
Table 2b Characteristics for included studies in patients with CAB
Full size table
Table 2c Characteristics for included studies with BRC and CAB being both used in patients
Full size table

Risk of bias analysis

Publication bias exists, due to Egger’s test (Fig. 2) showing p = 0.006 < 0.05. The clipping method (Fig. 3), was then used to evaluate the impact of the publication bias on the results. Our findings indicate that it has little effect, which proves the validity of the result. Sensitivity analysis (Fig. 4) shows that no study which was included had a large impact on the results. Due to the high test heterogeneity (Q = 194.556 on 34 degrees of freedom [p = 0.000]), the random-effects model (random: I–V heterogeneity) would handle the summary statistics.

Fig. 2
figure 2

The Egger’s test. It was used to investigate whether there was a publication bias. Publication bias exists, due to Egger’s test showing p = 0.006 < 0.05

Full size image
Fig. 3
figure 3

The clipping method. It was used to evaluate the impact of the publication bias on the results. Our findings indicate that it has little effect, which proves the validity of the result

Full size image
Fig. 4
figure 4

Sensitivity analysis. Sensitivity analysis was conducted on low-quality studies. This analysis shows that no study which was included had a large impact on the results

Full size image

Meta-analysis

The primary aim of this meta-analysis was to explore the overall success rate of DA withdrawal. Surprisingly, the overall success rate after DA withdrawal reached 36.6% (95% CI 29.4–44.2%) (Fig. 5). The secondary aim was to investigate the favorable factors related to the high success rate of DA withdrawal, including the serum PRL concentration, treatment duration, cause of hyperprolactinemia, type, and dose of DAs, regression of tumor size, and follow-up period. These results are shown in Tables 3 and 4 (multivariate regression analysis)

Fig. 5
figure 5

Forest map. This forest map contains 24 articles included in this analysis. The proportion of patients with persisting normoprolactinemia after DA withdrawal reached 36.6% in a random effects model (95% CI, 29.4–44.2%; I-squared: 82.5%)

Full size image
Table 3 Results of meta-analysis
Full size table
Table 4 Multivariate meta-regression analysis
Full size table

Serum PRL concentrations

We analyzed differences in serum PRL concentrations between patients with PRL ≤  200 μg/l and those with PRL >200 μg/l before treatment (36.5%; 95% CI 27.8–45.7 vs. 28.8%; 95% CI 16.4–43%), and between patients with PRL ≤10 μg/l and those with PRL >10 μg/l before drug withdrawal (36.6%; 95%CI 17.8–57.7 vs. 40.7%; 95% CI 27.9–54.2%).

Dosage of DAs

In this paper, a low maintenance dose of CAB was defined as 0.5 mg/week of CAB before withdrawal [11,12,13]. In patients with a low maintenance dose of CAB before withdrawal, the proportion of patients with normoprolactinemia was higher (51.5%; 95% CI 40–62.8%), compared with patients without low-dose CAB for maintenance therapy (21.5%; 95% CI 12.2–32.5%, p = 0.007). However, the success rate of drug withdrawal was not correlated with the dose of BRC treatment. Whether it was less than 7.5 mg or more than 7.5 mg, the results remained similar (17.5%; 95% CI 10.7–25.6 vs. 23.1%; 95% CI 9–41.3%).

Tumor size reduction

Tumor regression on CT or MRI during treatment is another important indicator. There were patients with tumor shrinkage (40.5%; 95% CI 31.4–49.9%), and patients without tumor shrinkage (22.1%; 95% CI 10.7–36.2%). A dramatic reduction in tumor size before withdrawal was generally associated with success of the treatment (p = 0.032). Further classification and calculation of cases with tumor volume reduction by more than 50% showed a 49.4% remission rate.

Types of DAs and treatment duration

The overall therapeutic outcome was better in patients using CAB as the only treatment (41.2%; 95% CI 32.3–50.4%) than in patients treated only with BRC (22.4%; 95% CI 14.8–31%). It was found that the success rate of drug withdrawal was lower in patients using DAs for less than 24 months than in patients treated for more than 24 months (20.8%, 95% CI 12.9–30 vs. 41.3%, 95% CI 32.1–50.8%; p = 0.037). Based on this outcome, four subgroups were further designed: patients treated with BRC less than 24 months (17.2%; 95% CI 8.9–27.5%) and CAB less than 24 months (24.8%; 95% CI 13–38.8%); patients treated with BRC more than 24 months (27.3%; 95% CI 15.6–40.9%) and patients treated with CAB more than 24 months (48.7%; 95% CI 38.9–58.5%). Clearly, patients treated with CAB for more than 2 years have the best outcome.

Types of DAs and causes of hyperprolactinemia

Our results showed no significant difference in causes of hyperprolactinemia: idiopathic hyperprolactinemia (48.7%; 95% CI, 6.6–92.1%): microprolactinomas (37.3%; 95% CI, 28–47%), and macroprolactinomas (25%; 95% CI, 15.3–36.3%) (p = 0.067). These three causes of hyperprolactinemia were matched with two DAs into six subgroups: idiopathic with BRC (23.2%; 95% CI 6.1–47.1%); idiopathic with CAB (73.2%; 95% CI 55.6–87.7%); microprolactinomas with BRC (25.9%; 95% CI 12.1–42.6%); microprolactinomas with CAB (40.8%; 95% CI 28.4–53.9%); macroprolactinomas with BRC (15.1%; 95% CI 9.1–22.3%); and macroprolactinomas with CAB (33.5%; 95% CI 17.6–51.7%). These results indicate that idiopathic hyperprolactinemia, microprolactinoma, or macroprolactinoma had no significant impact on the success rate; but again, the fact that better results were achieved with CAB was confirmed.

Follow-up periods

The follow-up periods were classified into three groups: a follow-up period: follow-up period ≤12 months (30.8%; 95% CI 22.4–39.8%); 12 months < a follow-up period ≤ 24 months (41.4%; 95% CI 20.9–63.6%); and a follow-up period > 24 months (41.3%; 95% CI 28.9–54.3%). The success rate of persisting hyperprolactinemia after DA withdrawal was not related to the follow-up period (p = 0.906).

Year of publication

We analyzed differences in year of publication, paper published after 2000 get better success rate than those published before 2000 (43.1%; 95% CI 34.7–51.9% vs. 20.3%; 95% CI 13–28.7%).

Discussion

This meta-analysis showed that a normal PRL level could be successfully maintained after DA withdrawal in quite a proportion of patients. Compared with the success rate in meta-analysis by Dekkers et al. [10], the result of our analysis showed that the success rate of DA withdrawal has greatly increased. According to the analysis by Dekkers et al. the probability of treatment success was highest when CAB was used for at least 2 years. Our analysis found the same conclusion. Low-dose maintenance therapy with CAB and a significant reduction in tumor size were two independent factors affecting the success rate of DA withdrawal. We also found that the normoprolactinemia maintenance rate in patients who used CAB as the only treatment was increased compared with that reported by Hu et al. in 2015 [43]. In their study, the overall success rate of normoprolactinemia maintenance after CAB withdrawal was 35%. The greatly improved success rate of withdrawal was in relationship with the increased number of patients included, common use of CAB, longer durations of DA treatment, and patient selection according to several criteria.

The duration of DA treatment is critical in maintaining a normal PRL level, and treatment for more than 2 years is strongly recommended [1]. The guideline [1] formulated by the American Endocrine Society in 2011 also recommended that patients should be treated by DAs for at least 2 years with normalized serum PRL levels and with no residual tumor on MRI. However, the duration of treatment was not considered as a favorable contributive factor for successful DA withdrawal in the meta-analysis reported by Hu et al. [43]. In contrast, our meta-analysis showed that the length of treatment is still a factor affecting the successful withdrawal of DAs.

There are few studies considering whether significant reduction in tumor size is associated to success withdrawal. In the meta-analysis by Hu et al. [43], they divided the cases into two groups: tumor shrinkage (39%) and no tumor shrinkage (29%). Similarly, Colao et al. [13] observed that if the tumor was still visible, the recurrence rate was almost twice as high as that in patients with no visible tumor [44], and the recurrence rate after DA withdrawal increased to 40% for microprolactinomas and 58% for macroprolactinomas. But if no tumor was visible, the recurrence rate after DA withdrawal decreased to 24% for microprolactinomas and 26% for macroprolactinomas. Based on these results, we further classified and calculated cases with tumor volume reduction by more than 50% and found that nearly half of the patients could stop using DAs without recurrence. These findings indicate that significant reduction in tumor size may be a contributive factor.

According to the results obtained from the present meta-analysis, more than half of the patients with low-dose CAB maintenance could safely stop using DAs. According to the guidelines of the American Endocrine Society [1], DAs should be tapered gradually before complete withdrawal. In line with other reports, Buyukbayrak et al. [28], Colao et al. [13], and Hu et al. [43] emphasized the importance of switching to maintenance treatment after PRL restored to normal levels. The remission rate reported in Colao et al.’s research was surprisingly high in patients with either microprolactinemia or macroprolactinemia (66.1% and 46.8%) after the CAB dosage was tapered to 0.5 mg/week. As a result, a gradual drug dose reduction is suggested. Nevertheless, the criteria for switching to maintenance treatment, the duration of maintenance treatment, and accurate dose adjustment needs to be further discussed and studied [28].

Limitations and strengths of this study

The main limitation of the present study is the high heterogeneity of the patients involved. Data derived from the articles included mostly are of average value. In each document, the extent of the tumor shrinkage before drug withdrawal was not absolutely accurate, devoid of mean values. These variables were given qualitative judgments with “no tumor shrinkage”, and “tumor shrinkage” according to these original articles. In some articles, with a large number of cases, there were few patients with pregnancy, or undergoing surgery or radiotherapy, and these variables cannot be easily detached and might affect the statistical results. These issues highlight the inherent limitations in any meta-analysis based on observational data and the need for larger prospective studies in the future.

Conclusions

Although the results of each individual patient cannot be easily predicted, there are a few points that may help clinicians make sensible selection of patients in routine clinical practice: (1) the patient has used CAB for at least 2 years; (2) a low-maintenance dose of CAB has been used for at least 1 year; and (3) restoration of a normal serum PRL level and a significant reduction in tumor size have been achieved before drug withdrawal. The degree of tumor invasion may be associated with recurrence. In patients, whose PRL levels remain normal for at least 2 years and the tumor volume shrinks by more than 50% or even no tumor residue, the dose of DAs can be reduced step by step, knowing that low-dose medication can maintain the stability of PRL level and tumor size.

Both guidelines [1, 9] agree that CAB is preferred to BRC and a more-than-24-month treatment duration is conducive to withdrawal. However, our meta-analysis showed that the success rate of DA withdrawal has greatly increased in recent years. Compared with previously mentioned guidelines [1, 9] this article also addresses three other points to be considered in clinical practice: a low dose of CAB maintenance before withdrawal, a significant reduction in tumor size and the use of CAB in patients with idiopathic hyperprolactinemia, which may make sense in increasing the success rate.