Optimizing Time to Treatment to Achieve Durable Biochemical Disease Control after Surgery in Prostate Cancer: A Multi-Institutional Cohort Study

Time to receive treatment or treatment delays is an important quality metric in patient-centered care that has been shown to affect outcome among various cancers (1), including breast, head and neck, colorectal, and melanoma (2–6), with patients receiving treatment sooner after initial diagnosis being more likely to experience favorable treatment outcomes. However, the implications of treatment delays on prostate cancer prognosis remain unclear in contemporary literature. Unlike other cancers, prostate cancer usually has slow progression and does not present immediate risk to patient survival (7), rendering both clinicians and patients ample time to make decisions on the appropriate treatment choice. In addition, concerns related with overtreatment of prostate cancer have prompted clinicians to adapt more pragmatic approaches toward treatment delivery which may increase the response time to treatment initiation, leading to increased time to treatment (8).

Few studies have demonstrated the possibility of adverse outcomes related with increasing time to treatment in prostate cancer (9–11); however, the length of delays prior to the initiation of treatment which is sustainable to achieve durable prostate cancer outcomes, without losing window of cure, remains debatable (12, 13). In addition, studies have argued against the use of early intervention in prostate cancer (14) and have shown that delays up to several years do not adversely affect prostate cancer outcomes (15–17). Consequently, a lack of consensus among previously published studies has resulted in significant ambiguity in the association between time to treatment and prostate cancer outcomes (14, 15, 18–21), which warrants further investigation. This study leverages the availability of 25 years’ worth of data from two large tertiary referral centers, with detailed clinicopathologic information, on prostate cancer patients who received prostatectomy as part of their primary treatment. This study seeks to determine the optimal time to start treatment that maximizes favorable prostate cancer outcomes and the clinical implications of treatment delays on biochemical disease control after prostatectomy.

This retrospective analysis includes a total of 1,891 patients with histologically confirmed localized nonactive surveillance prostate cancer who received prostatectomy as a primary treatment at two large tertiary referral centers—H. Lee Moffitt Cancer Center and Research Institute (MCC) and the University of Pennsylvania Health System (UPHS, Philadelphia, PA)—from 1987 to 2015 (Table 1). Patients who received prostatectomy at MCC were identified from the Health Research and Informatics platform, an enterprise-wide data warehouse. Patients from UPHS were those recruited to the Study of Clinical Outcomes, Risk and Ethnicity (SCORE) between 1990 and 2012. In addition, patients who received neoadjuvant treatment prior to their surgery were excluded from the analysis (n = 84). Therefore, a total 1,807 patients were used in the final analysis. Institutional Review Board approval was obtained before the commencement of the study.

All the patients included in the study received prostatectomy as definite primary treatment for their localized prostate cancer. Detailed clinicopathologic and demographic information including pretreatment variables such as clinical Gleason score (GS), PSA, and National Comprehensive Cancer Network (NCCN) risk grouping on patients from both institutes was recorded (Tables 1 and 2). GS for the cohort was categorized using International Society of Urological Pathology and was divided in group 1 (GS 3+3), Group 2 and 3 (GS 7), group 4 (GS 8), and group 5 (GS 9 and 10; ref. 22). Patients’ pretreatment risk status was determined by NCCN and was categorized as low (PSA ≤ 10 ng/mL and GS 3+3 and clinical T stage ≤ 2A), intermediate (PSA >10–20 ng/mL or GS 7, or clinical T stage 2b–2c), and high risk (PSA > 20 ng/mL or GS ≥ 8 or clinical T stage > 3b; ref. 23). In addition, MCC cohort had additional information on patients’ marital, insurance, and history tobacco exposure status. Date of diagnosis was used as an index date to determine time to treatment (delays between diagnosis and prostatectomy) and was calculated as months elapsed since the date of diagnosis to the date of prostatectomy.

Biochemical disease control was used as the primary outcome of the study and was defined as biochemical recurrence (BCR) after prostatectomy. To capture BCR, patients were followed after their prostatectomy for postsurgical PSA. BCR was determined by calculating the time between prostatectomy and PSA failure. PSA failure was determined by clinician-documented single PSA ≥ 0.2 ng/mL or two consecutive PSA values of 0.2 ng/mL after prostatectomy (23, 24).

In order to identify the optimal cutoff; time to treatment from the date of diagnosis and prostatectomy was introduced as a continuous variable in a multivariate cox regression model. Restricted cubic spline (RCS) function with four knots was used in cox model to allow nonlinear association between HR and time to treatment (2, 25). RCS-based cox model was adjusted using all the pertinent pretreatment clinicodemographic variables including clinical GS, PSA, clinical T stage, race, age, and year of prostatectomy. RCS in cox regression allowed us to identify not only the time point at which the risk of BCR increased (Fig. 1) but also the precise subcategorization of the different time interval for the treatment start time for further comparison. Differences between final subcategories of time to treatment and clinicopathologic/demographic characteristics were analyzed using the χ² test. Methods of survival analysis including Kaplan–Meier (KM) analysis and cox proportional hazard (CPH) model were used to analyze the association between time to treatment subcategories and biochemical disease control. Both univariate and multivariate CPH models were used to estimate the risk of BCR. In the models, cGS, PSA, time to treatment, and race category were used as categorical variables, whereas age at diagnosis and year of prostatectomy were used as numeric variables. A P value of ≤ 0.05 was considered as statistically significant to make clinically meaningful inference from the analysis. Analysis was completed using SAS 9.4 (26).

Comparison of baseline characteristics from the two institutions is provided in Table 1. Median follow-up time of the study cohort was 46 [interquartile range (IQR), 18–86] months, with an overall median time to treatment of 3 months. In multivariate cox regression model with RCS function, adjusted for all the pretreatment variables (Fig. 1), optimal cutoffs were provided for the time to treatment subcategories. Based on RCS model (Fig. 1), around 3 to 4 months after the date of diagnosis, inflection in the risk of BCR was observed (HR started to increase) which further crossed the reference line (HR, 1) near 6 months. Using this model-based approach, time to treatment was divided as 0 to 3, >3 to 6, and >6 months for further comparisons. A large proportion (61.5 %) of the patients received prostatectomy in 0 to 3 months, followed by 31.1% in >3 to 6 months, and 7.4% after 6 months. Higher proportions of African-American men (AAM) were likely to receive treatment after >6 months (Table 2). Delays beyond 6 months were also more prominent among patients with low-risk NCCN group. There was no apparent association among time to treatment categories and other clinicopathologic characteristics (Table 2).

Overall, 5-year freedom from biochemical failure (FFbF) rates were 78% [95% confidence interval (CI), 75%–81%], 82% (95% CI, 78%–85%), and 69% (95% CI, 59%–77%) for time to treatment of 0 to 3, >3 to 6, and >6 months, respectively (P < 0.001; Fig. 2A). Similarly, the trends in unfavorable BCR rate within the strata of time-to-treatment categories were also preserved when KM curves were stratified by NCCN especially among intermediate risk (Fig. 2B–D). In adjusted cox model, treatment delays after 6 months were significantly associated with an increased risk of BCR compared with those with 0 to 3 months (HR, 1.84; 95% CI, 1.30–2.60; P = < 0.001). Conversely, time to treatment >3 to 6 months did not show a significant effect on BCR. Both unadjusted and adjusted estimates for the risk of BCR are provided in Table 3. To explain the possible heterogeneity within the strata of GS, PSA, race, and NCCN risk groupings, the cross-product interaction between these covariates and time to treatment subcategories was analyzed separately. There was no interaction (likelihood ratio P > 0.05) observed between these covariates and time to treatment (result not presented). We also ran a sensitivity analysis in the MCC cohort for the multivariate cox model (Supplementary Table S1) to test whether inclusion of sociodemographic variables such as marital status, health insurance status, and history of tobacco exposure can affect the association between risk of BCR and time to treatment. Inclusion of these variable to the multivariate model adjusted for pretreatment clinical variables does not affect the hazard estimates as treatment delays beyond 6 month continue to negatively affect BCR (HR, 2.41; 95% CI, 1.39–4.17; P = 0.001).

This study optimizes the ideal time to treatment by applying model-based approach and assesses the implications of treatment delays on prostate cancer outcome. In our analysis, we carefully included all the pretreatment or clinical variables in multivariate RCS cox model, that are available to clinicians in order to make treatment decision, to identify the time point beyond which the risk of BCR was higher. Based on our results, delays in treatment after diagnosis for up to 6 months may not adversely affect prostate cancer outcome; however, significant delays beyond 6 months may lead to poor biochemical disease control. These results were consistent with another retrospective study by O’Brien and colleagues who reported a significantly higher risk of BCR among patients with surgical delays beyond 6 months. It is important to note that in our RCS-based model, risk of BCR increased around the same time point as O’Brien and colleagues reported (9). Interestingly, when stratified by NCCN risk groups, rate of BCR was significantly higher among intermediate-risk patients, as there was no statistical difference observed among low-risk patients. Our finding aligns perfectly with a large retrospective study by Abern and colleagues, who reported a high risk of BCR among intermediate-risk patients with delayed prostatectomy, whereas no apparent risk of poor outcomes among low-risk patients (10). Given that higher proportions of low-risk patients were likely to wait for surgery beyond 6 months, observed nonsignificant effect of treatment delays on BCR can be reassuring. However, attempts to perform restaging after 6 months may be necessary to avoid any disease upgrade as delays negatively affect BCR.

Contrary to our findings, there are studies which argue against the association of treatment delays and its implications on prostate cancer outcomes. In a study by Korest and colleagues (14), the authors reported negligible effect of treatment delays on BCR-free survival and prostate cancer pathology. In addition, a review by Van Den Bergh and colleagues reported nonsignificant effect of treatment delays on prostate cancer outcomes and reported that delays up to several years may not inferiorly affect prostate cancer outcome (15). In another study by Morini and colleagues on 908 prostate cancer patients, authors reported nonsignificant association of treatment delays and BCR among low-/intermediate-risk patients (16). However, the study design used in most of these contemporary studies utilized arbitrary or a priori–determined cutoff points to categorize time to treatment and then used these categories to predict prostate cancer outcome (14, 15, 18–21). This approach is susceptible to misclassification bias as it ignores the important factors such as clinical characteristics which have a direct association with treatment planning and may affect time to treatment (2, 10, 27). Furthermore, most studies often fail to acknowledge that prostate cancer progression is not fixed and alterations in the tumor characteristics as a result of biological changes are inevitable due to lack of definite treatment. Vickers and colleagues draw attention to this issue and point to the frequent use of null hypothesis–based approach in reporting that delayed treatment does not lead to any change in prostate cancer pathology (17), which can be misleading because it does not render any clinical utility toward treatment recommendation. While analyzing our data, we addressed this issue differently and applied a model-based approach to optimize time to treatment and provided an optimal time point beyond which risk of BCR starts to increase. In addition, with the inclusion of all the key clinical variables that are used in prostate cancer treatment decision-making, we further increased the clinical relevance of our estimates to better guide prostate cancer treatment planning. In addition, we included pooled data from two large cancer institutes and had a significantly large sample size with higher statistical power to assess the association treatment delays and prostate cancer.

Our results will aid in the shared decision-making process of prostate cancer treatment for both clinicians and patients. Although the cohort we used in analysis consisted of nonactive surveillance patients, our results may still have implications on the existing practices related to repeat biopsy especially among those who are likely to experience treatment delays. Currently, NCCN guidelines identify active surveillance as viable strategy for very low, low-risk, and a small subset of favorable intermediate-risk patients and recommend repeat biopsy in 12 months (23, 28). However, observed outcome deficits with increasing time to treatment may necessitate the restaging before 12 months especially among intermediate-risk patients. Although the absence of any statistical association of treatment delays and BCR among low-risk patients can be reassuring, attempts to avoid undue delays must be ensured to obviate any possibility of losing window of cure and prevent inferior outcomes. Our results underscore the clinical utility of time to treatment in targeting the subset of patients who are likely to benefit from initiation of treatment soon after diagnosis. This result necessitates the inclusion of time to treatment in the decision-making of prostate cancer treatment recommendation. Although this study does not corroborate whether Gleason progression or pathologic upstaging could have resulted in poor outcome among those with treatment delays beyond 6 months, future studies should focus on understanding the interaction between prostate cancer pathology and treatment delays to explain outcomes’ heterogeneity with increasing time to treatment.

The presence of detailed clinicopathologic information along with longer follow-up from two large tertiary care referral centers adds significant value to the generalizability of our results. In addition, patients across the different surgical era (1987–2015) were included in the analysis, therefore allowing us to capture data spanning over 25 years. Compared with other studies evaluating similar associations (14, 15, 18–21), our study consisted of a higher percentage of a traditionally underrepresented population of AAM (∼16% of the study population), thus making our results more generalizable to AAM who are at risk of experiencing inferior outcomes. In our analysis, application of the model-based approach over arbitrary cut point selection methods, to select optimal time point for time to treatment, reduced the likelihood of selection biases among those with delayed treatment. This study also has few limitations. First, our cohort only consisted of prostatectomy patients, thus lacking the information on those who received or contemplated receiving radiotherapy as a primary treatment. Therefore, our results may lack the generalizability across the treatment spectrum of prostate cancer, and further investigation may be warranted to confirm our finding in patients that received radiotherapy as their primary treatment. Second, in order to significantly increase our sample size and provide generalizability of our findings, the analysis for this study was conducted using combined data from two different tertiary referral centers. Although we controlled for the health centers’ effects in multivariate analysis, unmeasured confounding resulting from the differential treatment practices and diagnostic ascertainment (lack of standardization in pathologist review across the patient population) remains a potential limitation. It is also important to note that UPHS was relatively older institution and contributed a large proportion of older cases in our cohort (diagnosed before 2000). Because PSA screening also began around the same time, treatment delays in the older cases were more prominent among patients from UPHS than MCC. This variation can explain the higher proportion of pT3b-pT4 disease (advance stage) in the UPHS cohort (Table 1). Furthermore, loss to follow-up was another limitation of this study. Likelihood that many patients after receiving their primary treatment at one of the two tertiary centers (MCC and UPHS) would have moved on to seek care in their local facility remains higher and can result in inadequate follow-up. Compared with other categories, there were fewer patients with treatment delays beyond 6 months. Smaller numbers in this category may reduce the precision in the observed risk estimate and their corresponding confidence interval. Lastly, BCR was the only surrogate used in the analysis compared with metastasis and prostate cancer–specific mortality. Use of BCR as a surrogate may potentially limit our ability to associate our finding with long-term survival endpoints. As highlighted by McLaughlin and colleagues (3), ethical issues remain one of the major limitations in understanding the optimal time to initiate treatment, resulting in a lack of interventional prospective studies to determine optimal time to treatment. Continuing research will be required to advance existing guidelines on the delivery of prostate cancer treatment.

Time to treatment >6 months has detrimental effects on biochemical disease control after surgery. In order to achieve sustainable outcomes, definitive treatment for prostate cancer must be delivered within 6 months. This study identifies time to treatment as an independent predictor of prostate cancer outcome, and therefore, careful planning may be warranted while considering treatment delays.

D.I. Lee received honoraria from the speakers’ bureau of Intuitive Surgical. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Awasthi, R. Balkrishnan, P. Lal, T.R. Rebbeck, K. Yamoah

Development of methodology: S. Awasthi, F.A. Asamoah, T.R. Rebbeck, K. Yamoah

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Awasthi, D.I. Lee, S.B. Malkowicz, P. Lal, T.R. Rebbeck, K. Yamoah

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Awasthi, T. Gerke, V.L. Williams, S.B. Malkowicz, K. Yamoah

Writing, review, and/or revision of the manuscript: S. Awasthi, T. Gerke, J.Y. Park, F.A. Asamoah, V.L. Williams, A.K. Fink, R. Balkrishnan, D.I. Lee, J. Dhillon, J.M. Pow-Sang, T.R. Rebbeck, K. Yamoah

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Awasthi, T. Gerke, K. Yamoah

Study supervision: R. Balkrishnan, K. Yamoah

Others (pathology review and database management): P. Lal

This work was supported by the Prostate Cancer Foundation Young Investigator Award (to K. Yamoah), American Cancer Society MRSG-17-108-01-TBG (to K. Yamoah), and V Foundation (to K. Yamoah and J.Y. Park).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.