Data source

Patients with prostate cancer were identified from the Mount Sinai Hospital at New York City data warehouse. Original EMR data from the hospital data warehouse were processed using the Sema4 Centrellis^TM platform, consisting of automated abstraction engine, curation platform, patient dashboard, and cohort builder. Along with information from structured data fields, data from unstructured clinical notes were automatically abstracted, curated and integrated into structured fields. Cancer diagnosis metastases, medication, treatment, PSA values and other related data elements were stored in the database

Patient cohort identification

We identified prostate cancer patients diagnosed between January 2014 and April 2019 to capture the time frame when docetaxel and NHA were approved for mHSPC. 11,358 patients were assigned with at least one prostate cancer related ICD codes (185 in ICD-9 code and C61 in ICD-10 code) between January 1^st, 2014 and April 30^th, 2019 in the EMR system. However, not all patients with prostate cancer ICD codes assigned have confirmed prostate cancer. In addition, for patients diagnosed prior to visiting Mount Sinai hospital, their first time ICD code recorded in the EMR system didn’t always match the date of diagnosis. To correctly identify patients with pathologically confirmed cancer diagnosis dates, we used the automated oncology data retrieval and curation platform developed by Sema4 as demonstrated in S1A Fig. Both structured information and extracted data elements from unstructured clinical note using heuristic rules were used to assign cancer diagnosis dates as demonstrated in S1B Fig [9]. Patients with their extracted cancer diagnosis dates and other diagnosis related data elements, such as stage, histology, TNM, were classified into three groups using a stratification algorithm. Selected patients’ data were manually reviewed by domain experts to evaluate the quality of the results We excluded patients diagnosed after April 2019 to ensure at least 12 months of follow-up. We defined de novo metastatic disease as clinical documentation of metastatic disease within 3 months of the diagnosis date. Upfront docetaxel and NHA use were determined by chart review. After the cohorts were identified, chart review was carried out to confirm duration of upfront docetaxel or NHA therapy, adverse events, identification of 2^nd line therapy, and start date of 2^nd line therapy. Additional clinical and demographic variables such as PSA values, age, sex, race, performance status, Gleason score, and metastasis burden at time of diagnosis were also evaluated. This study was approved by the Mount Sinai Institutional Review Board (Protocol number: STUDY-21-00442). All experimental protocol for involving human data is in accordance with the Mount Sinai Health System (institutional) and Declaration of Helsinki guidelines. Participant informed consent was obtained in writing before they were included in the database for analysis and the study was approved by the Mount Sinai Health System ethics committee as mentioned above.

Time of next treatment and overall survival analysis

Time to next treatment (TTNT) was evaluated for this study, defined as time from initiation of either docetaxel or NHA to the next line of treatment. For the purposes of this metric, any change or addition of systemic therapy–including chemotherapy, NHA, immunotherapy, and clinical trials–was considered a failure of therapy, regardless of the reason for the change (e.g. progression, toxicity, patient preference, physician discretion). Patients not known to receive a 2^nd line treatment were censored at the time of their last follow-up. TTNT was represented as Kaplan-Meier curves as detailed under Statistical Analysis. Overall survival (OS) was evaluated for this study, defined as time from diagnosis to death (event) or last follow-up (censor). OS was represented as Kaplan-Meier curves as detailed under Statistical Analysis.

PSA progression free survival (PFS) analysis

PSA PFS was defined as the time from initiation of therapy to first PSA increase ≥25% and ≥2 ng/mL above that patient’s nadir, confirmed by a second value at least 3 weeks later, per PCWG3 criteria [10]. Patients who were started on 2^nd line therapy before criteria for PSA progression were met were excluded from this analysis. Patients who did not have PSA progression and who had not started 2^nd line therapy were censored at the time of their last follow-up. PSA PFS was represented as Kaplan-Meier curves as detailed under Statistical Analysis.

Metastasis burden and time of next treatment

High metastasis burden was defined as visceral metastases or ≥4 bone lesions with ≥1 beyond the vertebral bodies and pelvis, per the CHAARTED trial [2]. Patients were assigned: high, low, or uncertain metastatic burden based on chart review by an oncologist. Differences in TTNT between patient with high and low burden disease were compared in the 1. overall mHSPC cohort; 2. docetaxel treated cohort; and 3. NHA treated cohort. We further compared TTNT between patients treated with docetaxel or NHA, stratified by metastasis burden. TTNT were represented as Kaplan-Meier curves as detailed under Statistical Analysis.

PSA nadir and time to next treatment

The post-treatment PSA in the docetaxel cohort was defined as the PSA value following completion of at least 4 cycles of docetaxel therapy. Patients who started 2^nd line treatment prior to finishing 4 cycles of docetaxel (e.g. due to toxicity or refractory disease) were excluded from this analysis. For the NHA cohort, post-treatment PSA value was recorded after patients had received at least 7 months of ADT and at least 3 months of NHA. Patient were categorized in the docetaxel and NHA cohorts into three separate groups based on post-treatment PSA values: ≤0.2, 0.2–4.0, or ≥4.0 ng/mL. We compared TTNT according to post-treatment PSA within each treatment cohort.

Statistical analysis

Failure-free survival and PSA progression free survival data were presented as Kaplan-Meier curves. Survival curves were created with R (version 3.5.0; http://www.R-project.org/.) using the survival package (version 2.44–1; https://cran.r-project.org/web/packages/survival/index.html) and report the log rank p-value according to default parameters. We examined potential confounding variables that may also be associated with clinical outcome using single variable and multivariable Cox proportional hazards regression, reporting the hazard ratios and the Wald test statistic p-value for each variable. PSA values were log transformed due to the heavy right tail, and 2 individuals with missing values were assigned to the median value. Patients with unknown metastasis burden status were excluded in the Cox regression models.

Patient cohort

Total 5725 patients were identified by the platform with pathologically confirmed diagnosis dates for prostate cancer between January 1, 2014 to April 30, 2019. Of these, 240 were determined to have had de novo metastatic disease, based on clinical documentation of metastatic disease within 3 months of the diagnosis date. The diagnosis date, metastatic status, and date of metastasis detection were based on terms extracted from clinical notes and pathology reports via the abstraction engine described above. We then identified patients who had been treated with docetaxel (N = 67) or an NHA (abiraterone or enzalutamide) (N = 99) at any time. Finally, we identified those who had been treated with upfront docetaxel (N = 52) or an NHA (N = 42), defined as treatment within 7 months of diagnosis in conjunction with ADT and in the absence of any prior systemic treatment for prostate cancer.

The two groups of de novo mHSPC patients had no significant differences in age, race/ethnicity, PSA at diagnosis, Gleason score, ECOG performance status, and metastatic burden. Table 1 report the standardized mean difference (SMD) value. Variables with SMD<0.1 are considered well matched between the two groups. The docetaxel cohort has a higher frequency of patients with a Gleason score of 10 compared to NHA cohort (N = 12 versus N = 1). However, when combining Gleason scores for aggressive disease (9 and 10), the cohorts are well matched with frequencies of 57.7% and 57.1% for the docetaxel and NHA cohorts, respectively. We further evaluated the high burden patients for visceral metastasis, 6 patients with high metastasis burden in each cohort had visceral metastasis. Hence the cohorts were well matched for the variables include in the study (Table 1). The patient selection process is presented in S2 Fig.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Table of co-variants for the docetaxel and NHA cohorts.

https://doi.org/10.1371/journal.pone.0264800.t001

Docetaxel treatment course

In the docetaxel cohort, 47 of 52 (90.4%) patients began treatment at the standard 75 mg/m2 dosing; 10 patients had subsequent dose reduction to 60 mg/m2. The remainder of patients, 5/52 (9.6%), started at a reduced dose of 60 mg/m2; 1 patient required a dose reduction to 45 mg/m2 (S1A Table). The adverse effects leading to dose reduction in these 11 patients included: fatigue (3), neuropathy (3), febrile neutropenia (3), infection (1), and bradycardia (1) (S1B Table).

Forty-five patients (86.5%) completed 6 cycles of treatment. Six patients (11.5%) terminated docetaxel early due to adverse effects, most commonly leg edema, fatigue, LFT abnormalities, anaphylactic infusion reaction. One patient (1.9%) had docetaxel terminated before 6 cycles and moved to 2^nd line treatment due to refractory disease.

Forty-two docetaxel-treated patients (82.7%) initiated subsequent therapy during the follow-up period. The most common subsequent therapy was an NHA such as abiraterone (14/52, 26.9%) or enzalutamide (11/52, 21.2%). Eight (15.4%) patients enrolled in clinical trial following docetaxel therapy. Other agents included bicalutamide (3), sipuleucel-T (3), cabazitaxel (2), and apalutamide (2) (S2 Table).

Novel hormonal agent (NHA) treatment course

In the NHA cohort, 40 of 42 (95.2%) patients received abiraterone, and 2 of 42 (4.8%) received enzalutamide. All patients were started at each agent’s respective approved dosing: abiraterone 1000 mg daily with prednisone 5 mg daily, or enzalutamide 160 mg daily. Dose reductions occurred in 4 (9.5%) patients, primarily due to abnormal LFTs (2) or hot flashes (1). (S3 Table)

Of the 42 patients treated with upfront NHA, 20 (47.6%) patients transitioned to 2^nd line therapy during the follow-up period. 10% (2/20) of patients who moved onto 2^nd line therapy did so within first 3 months of 1^st line therapy due to rising PSA levels (S4 Table). The most common subsequent therapy following upfront NHA was an alternative NHA, e.g. enzalutamide (13/20, 65%). Three (15%) patients received docetaxel, one patient each received radium-223 (5%), olaparib (5%), Sipuleucel-T (5%) and bicalutamide (5%) (S4 Table).

Time of next treatment

Patients who received upfront NHA had a significantly longer median TTNT of 20.7 months (95% CI = 17.0-NE [not estimable]) compared to 10.1 months (95% CI 8.75–18.2) in docetaxel treated patients (p = 0.023) (Fig 1A). Single and multivariable analyses are presented in Table 2. NHA, low metastasis burden, and lower baseline PSA levels are all associated with longer TTNT. Variables with pvalues<0.15 were retained for the multivariable regression model. In the multi-variable model adjusting for age, baseline PSA, and metastasis burden, Gleason score, docetaxel was associated with shorter TTNT compared to NHA (HR 2.15, 95% CI 1.19−3.89, p = 0.011) (Table 2). None of the other variables remain statistically significant, but age below 65 at time of diagnosis, high baseline PSA, high Gleason score and high metastatic burden are still modestly associated with shorter TTNT, irrespective of treatment received (Table 2). However, OS was not different between the two treatment groups (p = 0.773) (S3 Fig). These findings from our real-world analysis replicate findings from clinical trial [11].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1.

Kaplan-Meier analysis comparing mHSPC patients treated with upfront docetaxel versus novel hormonal therapy (NHA) using A. Failure-free survival and B. PSA Progression free survival.

https://doi.org/10.1371/journal.pone.0264800.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Univariable and multivariable analysis of TTNT comparing upfront docetaxel to upfront NHA and the other co-variants.

https://doi.org/10.1371/journal.pone.0264800.t002

Clinical trials have suggested that high metastatic disease burden is predictive for clinical benefit from upfront treatment strategies for mHSPC [1, 2, 11]. We sought to determine whether there was evidence for heterogeneous treatment effects based on disease burden. Patients with high metastasis burden had a significantly longer TTNT with NHAs than docetaxel (25.12 [95% CI 17.03-NE] vs. 9.63 [95% CI 8.38–14.53] months, p = 0.014) (Fig 2A), while there was no significant difference in TTNT among patients with low metastasis burden (NHA 20.71 [95% CI 12.92-NE] months vs. Docetaxel 26.5 [95% CI 8.65-NE] months, p = 0.9) (Fig 2B). However, in a cox regression model predicting TTNT, the interaction between upfront therapy and metastasis burden was not significant (p = 0.28).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2.

Kaplan-Meier analysis for TTNT comparing patients treated with upfront docetaxel versus upfront NHA in A. High metastasis B. Low metastasis burden patients. Kaplan-Meier analysis for time to next treatment (TTNT) comparing patients with high and low metastasis burden in C. all mHSPC patients, D. Patients treated with upfront docetaxel and E. Patients treated with upfront NHA.

https://doi.org/10.1371/journal.pone.0264800.g002

We further analyzed the prognostic implications of metastasis burden in the combined mHSPC cohort and in the individual docetaxel and NHA cohorts separately. In the combined cohort of docetaxel and NHA treated patients, a high metastatic burden of disease was associated with a shorter median TTNT as compared to patients with low burden of disease, 11.57 [95% CI 9.24–20.12] vs. 20.71 [95% CI 12.92-NE] months (p = 0.034) (Fig 2C). In the docetaxel group, high burden disease is associated with shorter median TTNT as well, 9.63 [95% CI 8.38–14.53] vs. 26.5 [95% CI 8.65-NE] months (p = 0.026) (Fig 2D). However, metastasis burden was not associated with a significant TTNT difference for patients that received NHA (25.12 [95% CI 17.03-NE] vs. 20.71 [95% CI 12.92-NE] months, p = 0.79, Fig 2E).

PSA progression

Since time to next treatment may be influenced by decisions to change therapy due to toxicity rather than disease progression, we also evaluated PSA progression between the two treatment groups. Of note, 7 of 52 docetaxel-treated patients were started on a subsequent line of therapy prior to completing 6 cycles of docetaxel treatment (due to toxicity [6 patients] or refractory disease [1 patient], as mentioned above); PSA progression after docetaxel alone could not be evaluated in these patients. Among evaluable patients, we found that those treated with docetaxel were more likely to start a subsequent therapy in the context of rising PSA levels prior to meeting PCWG criteria for PSA progression (14/45 vs. 4/42, p = 0.02). The median PSA difference between PSA nadir and PSA at the start of 2nd treatment was 0.68 vs 1.58 ng/ml among those treated with docetaxel vs NHAs (p = 0.55).

The co-variants were reasonably well matched between the remaining 30 docetaxel-treated patients and 38 NHA-treated patients with respect to age, race, log PSA at time of diagnosis, Gleason score, ECOG score, and metastasis burden (S5 Table). Among these patients, there was a trend towards longer PSA PFS with NHA compared to docetaxel, though the difference was not statistically significant (19.0 [95% CI 15.6-NE] vs. 13.2 months [95% CI 7.7–22.6], p = 0.069) (Fig 1B). S6 Table presents the results for cox regression for both the univariable and multivariable analyses. In multi-variable analysis, treatment (docetaxel vs. NHA) was independently associated with PSA PFS (HR 2.06, 95% CI 1.04–4.08; p = .0.039) (S6 Table). Higher log PSA at time of diagnosis was also independently associated with worse PSA PFS (S6 Table).

Post-treatment PSA as prognostic marker in de novo mHSPC

Prior to the advent of combination therapies for de novo mHSPC, when such patients were treated with ADT alone, post-ADT PSA levels were shown to be independent predictors of overall survival [12]. It is not known whether post-treatment PSA also holds prognostic value in the context of modern combination treatments.

To evaluate the prognostic significance of post-treatment PSA, we examined TTNT stratified by post-treatment PSA in the overall study cohort, as well as in the individual docetaxel and NHA cohorts. Patients were stratified into 3 groups based on post-treatment PSA: ≤0.2, 0.2–4, and ≥4. In the combined docetaxel and NHA cohorts evaluable for PSA response, patients with a post-treatment PSA ≤0.2 had the longest TTNT (58.95 [95% CI 20.71-NE] vs. 11.57 [95% CI 9.63–25.12] vs. 9.40 [95% CI 7.76–20.84] months, p<0.001) (Fig 3A). Similar outcomes were observed in the individual docetaxel and NHA cohorts (Fig 3B and 3C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3.

Kaplan-Meier analysis for time to next treatment (TTNT) comparing patients stratified by drop in PSA after treatment as mentioned in methods in A. All mHSPC patients; B. upfront Docetaxel; C. upfront NHA treated patients.

https://doi.org/10.1371/journal.pone.0264800.g003

In a multivariable analysis of the combined cohort, higher post-treatment PSA was independently associated with worse TTNT (0.2–4 vs. ≤0.2 HR 5.06, 95% CI 1.98–12.97; ≥4 vs. ≤0.2 HR 6.42, 95% CI 2.43–16.95) (Table 3). In the NHA cohort, similar analysis demonstrated that the PSA 0.2–4 and PSA ≥4 groups have significantly worse outcomes compared to the patients in the PSA ≤0.2 group (HR 6.01, 95% CI 1.54 − 23.44; HR 6.83, 95% CI 1.33–35.09) (S7 Table). For docetaxel, there is a trend towards worse outcomes among those with higher post-treatment PSA levels, (0.2–4 vs. ≤0.2 HR 3.42, 95% CI 0.84–13.88, ≥4 vs. ≤0.2 HR 4.35, 95% CI 1.02–18.55) with the outcome between post-treatment PSA levels ≥4 vs. ≤0.2 being statistically significant (p = 0.046) (S7 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Univariable and multivariable analysis of TTNT in mHSPC patients stratified by post treatment PSA and other co-variants.

https://doi.org/10.1371/journal.pone.0264800.t003

Time to next treatment and metastasis burden

Upfront NHA was associated with longer median TTNT than docetaxel and remained an independent predictor controlling for demographic and clinical factors, however there was no change in OS. This corresponds with findings from STAMPEDE, where the NHA cohort also showed a significantly longer FFS than the docetaxel cohort [11].

Time to next treatment was shorter with docetaxel than NHAs. Several factors may have contributed to this observation. First, more docetaxel-treated patients changed therapy due to toxicity (6 vs. 0, p = 0.022). Second, there appeared to be a lower clinical threshold for docetaxel-treated patients to be initiated on a 2^nd line therapy compared to NHA-treated patients. For example, 14 docetaxel-treated patients were started on a 2^nd line therapy before meeting formal criteria for PSA progression, compared to only 4 of NHA-treated patients (p = 0.03). In addition, the median increase in PSA from nadir at the time of 2^nd line treatment initiation was 0.68 ng/ml among docetaxel-treated patients compared to 1.58 ng/ml among NHA treated patients; this difference was not statistically significant (p = 0.55), but suggests that clinicians were more inclined to initiate 2^nd line therapy in docetaxel-treated patients than NHA-treated patients. The fact that docetaxel treatment is limited to 6 cycles whereas NHAs are taken on an ongoing basis may influence clinicians’ response to rising PSA levels.

High burden of metastatic disease has been associated with greater overall survival (OS) benefit from upfront docetaxel in post-hoc analyses of clinical trials. In our TTNT analysis, NHAs were associated with median TTNT of over 20 months regardless of disease burden (high: 25.12 vs. low: 20.71 months) while median TTNT after docetaxel was significantly shorter in patients with high rather than low metastatic burden (9.63 vs. 26.5 months, p = 0.014). These observations raise the possibility of treatment effect heterogeneity between docetaxel and NHAs by metastatic burden. We assessed the interaction term, which was not significant (p = 0.28), but this analysis was limited by sample size. Thus, while NHAs were associated with longer TTNT in high-burden patients, it does not seem to translate into an OS benefit.

Second-line therapy

At the end of follow-up, 17.3%, of patients remained on ADT alone after completion of docetaxel, whereas 52.4% of the NHA cohort remained on 1^st line therapy. This was due in part to earlier approval for docetaxel in this setting and, consequently, longer follow-up in the docetaxel cohort (maximum of 64 months) compared to the NHA cohort (maximum of 28 months). The apparent lower clinical threshold to initiate 2^nd line treatment among docetaxel-treated patients, described above, may also have contributed.

Patterns of toxicity leading to dose reductions of treatment drug reflect otherwise well-described adverse effect profiles. Most patients, 86.5%, were able to successfully complete 6 cycles of docetaxel, with the main reasons for early termination being fatigue, peripheral edema, and LFT abnormalities. Dose reductions in docetaxel were recommended for fatigue, neuropathy, and febrile neutropenia. Meanwhile, no clear cases of early termination of an NHA was observed, and dose reductions occurred only in a minority of patients, primarily in response to abnormal LFT’s or hot flashes.

The prevailing practice pattern following upfront docetaxel was to transition to an NHA, with a relatively even split between abiraterone and enzalutamide. Following upfront NHA, most practitioners opted to transition to a second NHA. However, understanding that sequencing two different AR-pathway targeting agents back to back is usually associated with relatively low rates of response [13], more recent studies have explored transitioning to chemotherapy [14], radium, or targeted therapy with PARP inhibitor [15].

PSA as a prognostic indicator

In the pre-CHAARTED era, where standard of care for mHSPC was ADT alone, post-ADT induction PSA level was shown to have strong prognostic significance, with lower PSA levels associated with better survival outcomes [12]. To our knowledge, there is no data examining the prognostic value of post-treatment PSA in the context of modern combination therapies for mHSPC. In the overall study cohort, we observed that patients with a post-treatment PSA of ≤0.2 enjoyed the longest TTNT, whereas those with PSA 0.2–4 and ≥4 were associated with significantly shorter TTNT. This suggests that the post-treatment PSA after combination treatment in mHSPC patients has prognostic value, though further evaluation of its association with overall survival is necessary.

Limitations

There are inherent limitations in our data given the retrospective nature of this study. Comparisons between the docetaxel and NHA-treated patients may be confounded by indication. We were unable to perform propensity-score matching due to limited sample size. However, baseline characteristics were not significantly different between the two groups.

Enzalutamide and apalutamide, which were recently approved for mHSPC, were not well-represented in our dataset. Future studies should address these newer agents as experience with them accrues.

Time to next treatment may be affected by several factors including frequency of follow-up, toxicity, and patient or provider-preference. We provided information on the decision to change treatment where available and found that most treatment changes were prompted by PSA progression. We conducted a separate PSA PFS analysis among a subset of evaluable patients which showed a trend towards longer PSA PFS with NHAs compared to docetaxel. Larger observational cohorts or prospective data are needed to compare long-term survival outcomes between these two treatment strategies for mHPSC.