Prognostic significance of early changes in serum biomarker levels in patients with newly diagnosed metastatic prostate cancer

We evaluated the impact of early changes in serum biomarker levels on the survival of patients with metastatic hormone-sensitive prostate cancer (mHSPC) who were initially treated with androgen deprivation therapy (ADT). We retrospectively investigated 330 patients with mHSPC whose serum maker levels were at baseline and at 2–4 months. An optimal Cox regression model was established with the highest optimism-corrected concordance index based on 10-fold cross-validation. The median cancer-specific survival (CSS) and overall survival (OS) were 7.08 and 6.47 years (median follow-up, 2.53 years), respectively. In the final optimal Cox model with serum biomarker levels treated as time-varying covariates, prostate-specific antigen (PSA), hemoglobin (Hb), and alkaline phosphatase (ALP) significantly increased the risk of poor survival in the context of both CSS and OS. Kaplan–Meier curves stratified by the three risk factors of high PSA, low Hb and high ALP desmondtated that median OS were not reached with none of these factors, 6.47 years with one or two factors, and 1.76 years with all three factors.Early changes in serum biomarker levels after ADT may be good prognostic markers for the survival of patients with mHSPC.

with mHSPC and/or castration-resistant prostate cancer (CRPC) [6][7][8] . However, there is still less evidence regarding the impact of early changes in serum biomarker levels on clinical outcomes after ADT in recent diagnosed patients with mHSPC.
Therefore, this study aimed to investigate the prognostic impact of serum biomarker levels during early stages of ADT on the survival and develop a prognostic model for deployment among patients with newly diagnosed mHSPC. Table 1 shows the patients' characteristics. For the 330 patients analyzed, the median age was 72 years (interquartile range, 65-78 years). Approximately 90.9%, 58.5%, and 9.7% of patients had bone, lymph node, and visceral metastases, respectively. A combined androgen blockade was employed in 78.5% of the patients, whereas 21.5% were treated using LHRH antagonist.

Results
During the median 2.53-year follow-up, 111 (33.6%) patients died, of which 87 (26.4%) died due to progressive prostate cancer. The median cancer-specific survival (CSS) and OS were 7.08 and 6.47 years, respectively. Table 2 shows the univariable analysis for CSS and OS in patients with mHSPC treated with ADT. In both Cox models for CSS and OS, decreased body mass index (BMI) (p = 0.026 and p = 0.002), Gleason score of ≥9 (p = 0.035 and p = 0.043), extent of bone metastasis (EOD) score of ≥2 (both ps < 0.001), low Hb levels (p = 0.008 and p = 0.004 at baseline and p = 0.002 and p = 0.001 at 2-4 months) and high ALP levels (p = 0.023 and p = 0.046 at baseline and both ps < 0.001 at 2-4 months), high lactate dehydrogenase (LDH) level at baseline (both ps < 0.001), and high PSA level at 2-4 months (both ps < 0.001) were shown to be significant risk factors for worse survival. In addition, lymph node metastasis was an independent risk factor for poor CSS (p = 0.013).
The Kaplan-Meier curves of each binary serum biomarker PSA, ALP, and Hb at 2-4 months post-ADT are shown in Fig. 1 for CSS and in Fig. 2 for OS. The median CSS in patients with high PSA, high ALP, or low Hb levels at 2-4 months was shorter than that in patients with low PSA, low ALP, or high Hb levels (log-rank, p < 0.001, <0.001, and = 0.002, respectively). Additionally, the median OS in patients with high PSA, high ALP, or low Hb levels at 2-4 months was significantly shorter than that in patients with low PSA, low ALP, or high Hb levels at 2-4 months (log-rank, p < 0.001, <0.001, and = 0.001, respectively). Table 3 shows the model selection process with time-varying covariates of PSA, Hb, ALP, LDH, and albumin based on 10-fold cross-validation. Upon entering all significant variables selected at p < 0.2 in a univariate model into multivariable models, we identified the top five models for CCS and OS based on the observation of higher numbers of concordance indices (C-indices) ( Table 3). Among these, the final Cox model for CSS and OS was 0.771 with lymph node metastasis, EOD score, PSA, Hb, and ALP and 0.732 with BMI, Gleason score, lymph node metastasis, PSA, Hb, and ALP, respectively. These final Cox models demonstrated that p values for the proportional hazards score test 9 of selected variables calculated were all >0.05, indicating no statistical significant violation of the proportional hazard assumption. We also ran 10-fold cross-validation by using baseline serum biomarker levels instead of time-varying covariates of these biomarker levels. The highest C-indices shown in Supplementary Table 1 were all lower than those in the model with time-varying covariates (Table 3). Therefore, the Cox model with serum marker levels measured at 2-4 months as time-varying covariates had a better predictive ability than the model with serum marker levels measured at baseline.
The final Cox regression models for CSS and OS are shown in Table 4. The optimal models demonstrated that the time-varying covariates of PSA, Hb, and ALP significantly increased the risk of poor survival for both CSS and OS-i.e., >3.1 ng/mL of PSA [hazard ratio (HR): 2.47; 95% confidence interval (CI): 1.55-3.94 and HR: 1.88; 95% CI: 1.25-2.82, respectively], >350 IU/l of ALP (HR: 1.87; 95% CI: 1.13-3.08 and HR: 2.03; 95% CI: 1.30-3.17, respectively), and ≤12 g/dL of Hb (HR: 1.86; 95% CI: 1.20-2.90 and HR: 1.86; 95% CI: 1.25-2.76, respectively) measured at 2-4 months after ADT significantly increased the risk of poor survival. In the final model for CSS, in addition to PSA, Hb, and ALP, the presence of lymph node metastasis (HR: 1.61; 95% CI: 1.01-2.58) and the EOD score with ≥2 (HR: 2.02; 95% CI: 1.20-3.39) were also significantly associated with poor survival. When the measurement time for time-varying covariates was limited to between 70 and 110 days, the final Cox models with the highest C-index, which satisfied the proportional hazard assumption, suggested an influence of lymph node metastasis, PSA, Hb, and ALP for CSS (C-index: 0.774) and that of PSA, Hb, ALP, and albumin for OS (C-index: 0.753). Both the models demonstrated that PSA, Hb, and ALP were all significantly associated with poor CSS and OS. These sensitivity analysis results are further presented in Supplementary Tables 2 and 3.
According to the final Cox models for CSS and OS, the derived prognostic risk factors in common were high PSA and ALP levels and low Hb level at 2-4 months. Therefore, the risk groups were formed based on combinations of these three factors, with patients having zero, one or two, and three risk factors defined as the low-, intermediate-, and high-risk groups, respectively. The Kaplan-Meier curves and median CSS and OS were subsequently calculated for these groups, and the results are shown in Fig. 3 These results indicate that the risk stratification depending on the three risk factors with serum levels of PSA, Hb, and ALP at 2-4 months contributes to differences in the survival of patients with mHSPC.

Discussion
Various PSA-related "on therapy" biomarkers have been reported in mHSPC 10 . In PSA kinetic variables, PSA nadir and time to PSA nadir are well-known risk factors for outcomes in mHSPC 11,12 . In recent years, several studies demonstrated that PSA level ≤0.2 ng/mL at 7 months was a strong prognostic factor for longer OS in patients with ADT 13,14 . These results suggested that "on therapy" PSA-related biomarker levels were promising (2019) 9:12071 | https://doi.org/10.1038/s41598-019-48600-8 www.nature.com/scientificreports www.nature.com/scientificreports/ candidates for predicting treatment outcomes in patients with mHSPC. However, it was implied that the prognosis can be predicted more than half a year after ADT initiation. In recent phase III trials showing the superiority of upfront docetaxel and abiraterone acetate administration for mHSPC, ADT was commenced within 120 days before randomization 2,3 . Therefore, it may be reasonable to predict patients with poor prognosis at initial stages (up to 3 or 4 months) and detect very early changes in serum biomarker levels after ADT initiation. Park et al. reported that shorter PSA half-time calculated as log 2 divided by the slope of the linear regression of log PSA versus the time using pre-and post-treatment PSA levels assayed were independent risk factors for poor CSS in patients with mHSPC 7 . In our study cohort, PSA levels at 2-4 months increased from baseline in several patients (data not shown), indicating that it is not feasible to calculate values from chronological values, including PSA half-time, for such patients. Altogether, evaluation of absolute PSA levels after ADT may be a useful option for predicting poor prognosis as it is simple and can be applied to all patients.
Regarding the "on therapy" non-PSA serum biomarker levels, a decline in Hb levels after 3 months of ADT was independently associated with shorter survival (HR, 1.10 per 1 g/dL decline; p = 0.0035) and shorter progression-free survival (HR, 1.08 per 1 g/dL decline; p = 0.013) after adjusting for potential confounders, including baseline Hb levels in 827 of 1,286 patients enrolled in the SWOG Study S8894 (Intergroup Study 0105) 15 . Consistent with these findings, the current study showed that Hb levels at 2-4 months represented a strong risk factor for CSS and OS. Moreover, serum ALP at 2-4 months was also an independent prognostic factor for CSS and OS in the present study. A previous study suggested that increased ALP levels at 12 weeks is Orchidectomy + anti-androgen 18 (5.  a promising biomarker in bone metastatic CRPC patients treated with abiraterone acetate 8 . Although further validation study should be performed in the future, our study represents the first proposal of the impact of early changes in serum Hb and ALP levels in patients with mHSPC. Several limitations of the current study should be noted. First, we focused on patients whose serum data at an early period after ADT were available, thus indicating that we excluded patients who died at a very early period. In the entire study cohort, although only 21 (3.5%) of 605 patients died within 3 months, a selection bias based on this fact cannot be avoided. Second, our study did not consider the impact of sequential treatments after initial ADT. Although the treatment duration was not statistically associated with CSS and OS in the current study, sequential treatments after ADT failure may play a role in the patient's outcome. Furthermore, the cohort did not include patients who were treated with upfront docetaxel and abiraterone acetate, which are now considered to be the standard treatment for high-volume mHSPC. Novel risk factors for patients receiving upfront treatments should be elucidated in future studies. Finally, a retrospective study design and short follow-up duration were other limitations. Therefore, future studies with a longer follow-up period using a validation dataset are warranted.
In conclusion, high levels of PSA and ALP and low levels of Hb at 2-4 months are promising early "on therapy" prognostic biomarkers for survival in patients with newly diagnosed mHSPC who are treated with only ADT. Patients can be divided into different risk groups depending on the early changes in PSA, Hb, and ALP levels. Not only pretreatment risk factors but also early changes in serum biomarker levels may be useful for predicting poor survival in patients who require more aggressive treatment, including upfront chemotherapy and novel anti-androgens.

Methods
Patients. This retrospective multicenter study was conducted at nine medical institutes, in the Tohoku region, Japan. A consecutive group of 629 adult patients diagnosed with mHSPC between March 2008 and May 2016 was retrospectively identified at each institute. All patients initially received ADT, which consisted of orchiectomy, luteinizing hormone-releasing (LHRH) agonists/antagonists, alone or combined with bicalutamide. No patient www.nature.com/scientificreports www.nature.com/scientificreports/ received upfront docetaxel and/or abiraterone acetate as an initial therapy. Sequential treatments were administered after first-line ADT at the physician's discretion.
The study was conducted in accordance with the Helsinki Declaration. The study was also approved by each institute's ethical committee (Ethical committees of Akita University School of Medicine, Hirosaki University School of Medicine, Tohoku University School of Medicine, Yamagata University School of Medicine, Miyagi Cancer Center, Yamagata Prefectural Central Hospital, Iwate Prefectural Isawa Hospital, Aomori Prefectural Central Hospital, Sendai City Hospital). We applied an opt-out methodology is to provide accessible information to all patients to facilitate informed consent without interfering with the medical consultation 16 and patients were informed of their inclusion in the study and were provided information on the institution's website.
Variables. The variables in the data set contained patients' characteristics at the time of their diagnosis, including age; BMI (kg/m 2 ); medical institute; years of diagnosis; years of initial ADT; Eastern Cooperative Oncology Group performance status score (ECOG-PS); biopsy Gleason score; site of metastasis (visceral, lymph node, or bone); presence of bone pain; EOD score; types of initial hormonal therapy; levels of serum biomarker PSA, Hb, ALP, LDH, and albumin; and date of cause-specific death and all-cause death. ECOG-PS and the presence of bone pain were evaluated by inquiry and physical examination. EOD scores for each patient were classified using bone scintigraphy at the time of the initial diagnosis according to the definition of Soloway et al. 17 . To investigate the changes in serum biomarker levels in the early phase after ADT, we defined 2-4 months (70-144 days) as the early phase. Fig. 4. We first excluded 24 patients due to missing values on survival outcome. We also excluded 210 patients because they had serum biomarker levels evaluated beyond the  www.nature.com/scientificreports www.nature.com/scientificreports/ specified period of 2-4 months. For the remaining 395 patients, we interpolated the missing serum marker levels using an average value of each marker according to each cohort (medical institute). We interpolated BMI from the regression slope on age after stratification of cohorts. Finally, we excluded the missing values of EOD score (n = 1), bone pain (n = 33), Gleason score (n = 15), initial therapy (n = 1), and 17 patients whose primary site was treated, because we assumed that these variables were not interpolated for ethical considerations. The remaining 330 patients comprised the subjects in our analyses.

Statistical analyses.
The outcome of the present study included CSS and OS, which were calculated as the time from the diagnosis of mHSPC to death from prostate cancer or from any other cause. Patients known to be alive or lost to follow-up on the date of last contact were censored. Baseline characteristics were summarized using descriptive statistics (median and range for continuous variables, number and percentage for categorical variables). PSA level was divided at the median of its distribution at baseline (>345 ng/mL vs. ≤345 ng/mL) and at 2-4 months (>3.1 ng/mL vs. ≤3.1 ng/mL), otherwise the serum biomarkers were all divided into binary groups according to the normal ranges for Japanese males [18][19][20][21] : Hb (≤12 g/dL vs. >12 g/dL<), ALP (>350 IU vs. ≤350 IU), LDH (>220 IU vs. ≤220 IU), and albumin (≤3.5 mg/dL vs. >3.5 mg/dL).
The Kaplan-Meier method was applied to depict the CSS and OS curves between the binary groups of serum biomarkers at 2-4 months, and the comparison was statistically analyzed using a log-rank test. To investigate the impact of serum biomarker levels in the early phase after ADT initiation and to identify independent prognostic factors for CSS and OS, we applied a Cox proportional hazard model to calculate the HRs and 95% CIs. While building a Cox regression model, we considered PSA, Hb, ALP, LDH, and albumin levels measured at 2-4 months as time-varying variables because the levels of these markers were not constant over the follow-up duration. In order to select an optimal Cox model, while adjusting for all variables selected at p < 0.2 in the univariate models, we ran the 10-fold cross-validation. The highest optimism-corrected C-index value, which ranges from 0 to 1, can be used to measure and compare the discriminative power of prediction models 22 . Finally, we checked whether the best model chosen met the assumptions of the proportional hazard model based on the score test proposed  Table 4. Final Cox regression models for cancer-specific survival and overall survival in patients with mHSPC treated with ADT. ALP: alkaline phosphatase; BMI: body mass index; EOD: extent of bone disease; Hb: hemoglobin; PSA: prostate specific antigen.  We first excluded 24 patients due to missing data regarding survival outcome and then excluded 275 patients due to missing data for variables required for the analyses. The remaining 330 patients were used as the study subjects for our analyses.