The prognosis benefits of adjuvant versus salvage radiotherapy for patients after radical prostatectomy with adverse pathological features: a systematic review and meta-analysis

The search strategy was initiated by two reviewers respectively. To retrieve the required literature as completely as possible, a great number of databases have been searched, included PubMed (from 1950 to June 2019), EMBASE (using Ovid as the main search engine from 1974 to June 2019), Web of Science (from 1900 to June 2019) and the Cochrane Library electronic databases (from 1948 to June 2019). Combinations of the following MeSH and keywords were used in databases: (prostate neoplasms or prostate cancer or prostatic cancer) and (adjuvant radiotherapy or adjuvant RT) and (salvage radiotherapy or salvage RT).

The criteria for included studies were: (1) all potential studies concerning the comparison of the prognosis of postoperative ART and SRT; (2) articles published in English; (3) at least one of the following outcomes was reported: overall survival (OS), BRFS and distant metastases-free survival (DMFS).

All patients must undergo RP and had at least one of APFs, including PSM, extracapsular extension, seminal vesicle invasion, and high GS. The specific eligibility criteria for the ART group were as follows: (1) postoperative RT was initiated when serum PSA was undetectable; (2) RT should be performed within 6 months after RP. Correspondingly, the SRT group should meet the following conditions: (1) RT was started when serum PSA rising constantly from undetectable level; (2) patients have been shown to develop PSA recurrence.

Conference abstracts which did not provide enough information were excluded. Case reports, review articles and editorial comments were not in our consideration. Neoadjuvant therapy should not be administered to these patients. The patient had other malignancies other than prostate cancer should be also ruled out.

Two authors carried out the procedure of data extraction independently. The titles and abstracts of articles retrieved by the proposed strategy firstly were screened to rule out irrelevant articles. Then, the full texts of selected articles were elevated in complying with the inclusion and exclusion criteria. The necessary data of the finally included articles were extracted, included the type of study, authors, publication year, the characteristics of participants in the ART and SRT groups (number, age, GS, staging, and follow-up time), outcomes (OS, BRFS, DMFS and related hazard ratio (HR)), etc.

The Newcastle-Ottawa scale (NOS), included three items: Selection, Comparability, and Outcome, was used to elevate the methodological quality of each study [10]. Each article was scored on a scale of 0 to 9. A study that achieved a score of 8 or 9 was considered high quality and a score of 5 to 7 were regarded as moderate quality [11]. Discussion and consultation assisted in resolving an existed disagreement between two authors during the procedure.

The primary outcomes were OS and BRFS. The secondary outcome was DMFS. OS was defined as the time from RP/RT to death, irrespective of the reason of death. BRFS was calculated from RP/RT to a detectable PSA value, or a serum PSA > 0.2 ng/ml on two consecutive detections for post-RP patients, or a rise of PSA > 0.2 ng/ml above nadir for post-RT patients. DMFS was identified as the time from RP/RT to evidence of local recurrence or distant metastasis from imaging examination.

The Review Manager software (RevMan version 5.3, The Cochrane Collaboration 2014) was used to analyze the data. Two reviewers input the data and performed the analysis. The other reviewers verified it in order to minimize the chance of error and bias.

As OS, BRFS and DMFS were time-to-event outcomes and were most appropriately analyzed using HR [12], we used HR between two survival distributions as a summary statistic. For a study which reported HR and corresponding 95% confidence interval (CI), we extracted it directly. We also calculated these HRs and the corresponding 95% CIs of the included studies which provided sufficient data using the methods outlined by Tierney and colleagues [12]. In accordance with the contract, an overall HR of less than 1 favored the ART group. The survival beneficial effect of ART compared to SRT was considered statistically significant if the 95% CI of HR did not overlap 1 [13]. The reported odds ratio (OR) with 95% CI was also calculated in the analysis.

The heterogeneity among the studies was assessed using the Chi-squared test and the I² statistic. A p-value of < 0.1 and an I² value > 50% were considered as statistical heterogeneity. If significant heterogeneity was indicated, a random-effect model was used; instead, a fixed-effect model was used. Funnel plots would be used to investigate publication bias if enough studies were available.

5-year and 10-year OS and BRFS rates of these included studies were further extracted and a subgroup analysis was conducted in terms of the starting point of follow-up time. We also separately compared the survival benefits of ART with early salvage radiotherapy (ESRT), which was defined as RT administered at a postoperative serum PSA ≤ 0.5 ng/ml.

According to our search strategy, 1139 articles were screened. A total of sixteen studies met with the predefined inclusion criteria; however, fifteen studies were finally included in the meta-analysis [14‐28]. One study was excluded due to lack of necessary data, failure to report the results of interest, and unsuccessful HR extract [29]. Five studies were excluded because some subjects received neoadjuvant therapy [30‐34]. Since two clinical trials paid more attention to the comparison of ART and wait-and-see, but the percentage of patients who received SRT in the observation group was very small, they were both not in our consideration [7, 35]. The selection process of qualified studies was shown in Fig. 1.

All included articles were cohort studies published between 2002 and 2018. Seven studies took RP as the starting point of follow-up and reported corresponding results [14‐18, 21, 22]. Six studies followed RT as the basis of follow-up and reported related results [19, 20, 23, 24, 27, 28]. The remaining two studies reported respectively results at different follow-up starting points [25, 26]. It was worth mentioning that four articles compared the prognosis of postoperative ART and ESRT [14‐17]. The detailed description of these studies was shown in Table 1.

A total of 5586 patients with APFs after RP were enrolled in this meta-analysis, including 2419 patients received ART and 3167 patients got SRT. Of those patients received SRT, 1253 patients specifically were stated that they were treated with ESRT. The age of subjects ranged from 59 to 66 years old in different studies. The PSM rate of patients in the ART group ranged from 50 to 96%, while it ranged from 23 to 87% in the SRT group. Detailed information of included patients was summarized in Table 2.

The HRs and corresponding 95% CIs of OS in five articles were pooled and the results showed that OS in the ART group was better than that in the SRT group (HR: 0.58; 95%CI: 0.42–0.79; p = 0.0006) with no statistically significant heterogeneity (I² = 0, p = 0.64) [15‐17, 21, 23] (Fig. 2a). Subgroup analysis also indicated a statistically significant difference between postoperative patients received ART and SRT when follow-up time was calculated from RP (HR: 0.55; 95%CI: 0.39–0.78; p = 0.0007). No significant heterogeneity was found among studies (I² = 0, p = 0.57). However, the subgroup analysis of follow-up from RT was not applicable (Fig. 2a).

We also compared the OS of patients who received post-prostatectomy ART versus ESRT, which were extracted from three studies [15‐17]. The meta-analysis of these studies showed that ART still had an advantage over ESRT in terms of OS (HR:0.59; 95%CI: 0.40–0.87; p = 0.008). Likewise, there was no evidence to reveal a significant heterogeneity between studies (I² = 0, p = 0.60) (Fig. 2B).

Three studies reported 5-year OS rates of the ART and SRT groups [18‐20], and four studies reported 10-year OS rates of these patients [15, 18, 20, 21]. The pooled results of 5-year OS rate showed there was a significant statistical difference between the ART group and the SRT group (OR: 0.19; 95%CI: 0.07–0.49; p = 0.0006; I² = 45%) (Fig. 3a). Similarly, subgroup analysis presented a statistically significant difference between the ART group and the SRT group when follow-up from RT (OR: 0.08; 95%CI: 0.02–0.44; p = 0.003; I² = 20%) (Fig. 3a). But the subgroup analysis of follow-up from RP was not applicable (Fig. 3a).

Nevertheless, the pooled result of 10-year OS rate suggested there was no significant statistical difference between the ART group and the SRT group (OR:0.70; 95%CI: 0.48–1.03; p = 0.07; I² = 30%) (Fig. 3B). Subgroup analysis indicated that there was also no statistically significant difference between two groups in terms of 10-year OS rate for studies calculated from RP (OR: 0.81; 95%CI: 0.53–1.23; p = 0.33; I² = 0%) (Fig. 3b). The subgroup analysis of follow-up from RT was also not applicable (Fig. 3b).

The HRs and corresponding 95% CIs of BRFS between the ART and SRT groups were extracted from nine articles [14, 15, 17, 20, 22‐25, 27]. A meta-analysis of these studies by a random effect model showed that postoperative patients who received ART had better control of BCR compared to those received SRT (HR: 0.50; 95%CI: 0.37–0.68; p < 0.0001) with a statistically significant heterogeneity (I² = 76%, p < 0.0001) (Fig. 4a). The pooled result of data reported in four studies [14, 15, 17, 22], which took RP as the starting point of follow-up, showed there was a significant statistical difference between the ART and the SRT groups (HR: 0.56; 95%CI: 0.34–0.92; p = 0.02) with a statistically significant heterogeneity (I² = 88%, p < 0.0001) (Fig. 4a). The subgroup analysis of follow-up from RT revealed the similar tendency that the ART group owned a longer BCR process (HR: 0.45; 95%CI: 0.31–0.66; p < 0.0001) (Fig. 4a). No statistically significant heterogeneity was found among these studies (I² = 46%, p = 0.11) (Fig. 4a).

In the same way, we compared the BRFS of patients who received post-prostatectomy ART versus ESRT, which were obtained from three studies [14, 15, 17]. Conversely, the meta-analysis of these studies illustrated there was no statistical difference in BRFS between the ART and ESRT groups (HR: 0.60; 95%CI: 0.31–1.17; p = 0.14; I² = 91%). (Fig. 4b).

The 5-year BRFS rate of ART and SRT groups were reported in nine studies [14, 18‐20, 24‐28]. Data were pooled with a random-effect model since there was a significant difference in heterogeneity among these studies (I² = 78%, p < 0.00001). The pooled result of the 5-year BRFS rate showed there was a significant statistical difference between the ART and the SRT groups (OR: 0.46; 95%CI: 0.30–0.71; p = 0.0003) (Fig. 5a). Of these studies, two calculated BRFS from the time of RP, five calculated BRFS from the time of RT and the remaining two reported 5-year BRFS rate both at these two different time points. The subgroup analysis showed 5-year BRFS rate between the ART and SRT groups were comparable (OR: 0.90; 95% CI: 0.60–1.35; p = 0.60) with no statistically significant heterogeneity (I² = 50%, p = 0.11) when follow-up started from RP (Fig. 5a). However, the pooled data followed up from RT indicated that there was a significant statistical difference between the ART and the SRT groups (OR: 0.31; 95%CI: 0.24–0.40; p < 0.00001), which was consistent with the overall outcome (Fig. 5a). No significant heterogeneity was found among studies (I² = 0, p = 0.78) (Fig. 5a).

Three studies reported 10-year BRFS rate of these patients [15, 18, 20]. Random effect model meta-analysis was conducted since there was a significant difference in heterogeneity among these studies (I² = 53%; p = 0.12). The pooled analysis showed that there was a significant statistical difference between the ART and the SRT groups (OR: 0.45; 95%CI: 0.29–0.69; p = 0.0003) (Fig. 5b). The subgroup analysis of follow-up started from RP indicated that there was also a significant statistical difference in 10-year BRFS rate between the ART and the SRT groups (OR: 0.55; 95% CI: 0.40–0.76; p = 0.0003) with no statistically significant heterogeneity (I² = 0, p = 0.88) (Fig. 5b).

The HRs and corresponding 95% CIs of DMFS were available in five studies [15‐17, 20, 22]. A meta-analysis of these studies by a fixed effect model demonstrated that patients in the ART group had a lower risk of distant metastasis than those in the SRT group (HR: 0.51; 95%CI: 0.36–0.71; p < 0.0001) with no statistically significant heterogeneity (I² = 11%, p = 0.34) (Fig. 6a). The subgroup analysis also produced a similar result that there was significant statistical difference in DMFS between the ART and the SRT groups (HR: 0.52; 95%CI: 0.37–0.73; p = 0.0002) without statistically significant heterogeneity (I² = 10%, p = 0.34) when follow-up from RP in four studies [15‐17, 22] (Fig. 6a).

When compared the data from three studies focused on DMFS of patients who treated with post-prostatectomy ART versus ESRT [15‐17], the pooled analysis showed that there was a significant statistical difference between the ART and the ESRT groups (HR: 0.63; 95% CI: 0.42–0.95; p = 0.03). No significant heterogeneity was found among studies (I² = 0, p = 0.6) (Fig. 6b).

The NOS scores of included studies ranged from 5 to 8 (median score: 7), which meant that all references rated at least moderate quality. The distribution of NOS scores of all included studies in this meta-analysis was presented in Table 3. A funnel plot of BRFS was made to evaluate the publication bias. No significant asymmetry can be found and it suggested there was no significant publication bias in this study (Fig. 7).