Mortality in men with castration‐resistant prostate cancer—A long‐term follow‐up of a population‐based real‐world cohort

Abstract Objectives The objective of this study is to find clinical variables that predict the prognosis for men with castration‐resistant prostate cancer (CRPC) in a Swedish real‐life CRPC cohort, including a risk group classification to clarify the risk of succumbing to prostate cancer. This is a natural history cohort representing the premodern drug era before the introduction of novel hormonal drug therapies. Methods PSA tests from the clinical chemistry laboratories serving health care in six regions of Sweden were retrieved and cross‐linked to the National Prostate Cancer Registry (NPCR) to identify men with a prostate cancer diagnosis. Through further cross‐linking with data sources at the Swedish Board of Health and Welfare, we retrieved other relevant information such as prescribed drugs, hospitalizations, and cause of death. Men entered the CRPC cohort at the first date of doubling of their PSA nadir value with the last value being >2 ng/ml, or an absolute increase of >5 ng/ml or more, whilst on 3 months of medical castration or if they had been surgically castrated (n = 4098). By combining the two variables with the largest C‐statistics, “PSA at time of CRPC” and “PSA doubling time,” a risk group classification was created. Results PSA‐DT and PSA at date of CRPC are the strongest variables associated with PC specific survival. At the end of follow‐up, the proportion of men who died due to PC was 57%, 71%, 81%, 86%, and 89% for risk categories one through five, respectively. The median overall survival in our cohort of men with CRPC was 1.86 years (95% CI: 1.79–1.97). Conclusion For a man with castration‐resistant prostate cancer, there is a high probability that this will be the main cause contributing to his death. However, there is a significant difference in mortality that varies in relation to tumor burden assessed as PSA doubling time and PSA at time of CRCP. This information could be used in a clinical setting when deciding when to treat more or less aggressively once entering the CRPC phase of the disease.


| Study population
We obtained all PSA measurements during the period 2005-2014 from the Uppsala/Örebro PSA cohort (UPSAC) including five counties in the region of Uppsala/Örebro. 10 Similarly, we retrieved data for the Stockholm region from the STHLM-0 cohort 11  We differed between CRPC patients treated with neoadjuvant ADT prior to radiotherapy from CRPC patients treated with ADT due to disease progression and ADT as primary treatment, as described in Hemelrijck's publication for Prostate Cancer data Base Sweden (PCBaSe). 13 Figure 1 presents the patient selection flow chart.

| Medical castration
Medical castration was reached when patients had been treated with ADT for at least 3 months within a total period of 6 months according to records in the Swedish National Prescribed Drug Registry.

| Surgical castration
These patients have a record of surgical castration in the Swedish National Prostate Cancer Registry. The hospital admission date related to the surgical castration was used as an estimate of the castration date.

| Castration-resistant prostate cancer
The date of CRPC is based on an increase in PSA (first date of doubling of nadir PSA value with the last value being greater than 2 ng/ml, or an absolute increase of 5 ng/ml or more) despite medical or surgical castration.

| Primary versus secondary ADT
Primary ADT is defined as ADT initiated as first treatment for PC. Secondary ADT is defined as ADT initiated following recurrence of PC that occurs after curative intent prostatectomy or radiotherapy.
This group also includes patients on deferred ADT.

| M status
M1/M0 status is based upon PC diagnosis.

| Risk group classification
We decided to apply measures of PSA-kinetics combining the two variables with the largest C-statistics (see Section 2.4) that remained mainly unchanged in a full multivariable model, which were "PSA at time of CRPC" and "PSA-DT." Interaction effects between the log of PSA at date of CRCP against log PSA-DT were investigated in a Cox model by the use of a restricted two-dimensional cubic spline and the gam-function in the R-package mgcv. 14,15 The equation for the stratification into risk groups 1-5 is as follows: Score = log (PSA at CRPC) À 1.40 * log (PSA-DT).
As we knew that our M-stage variable was only reflecting Mstage at PC diagnosis, this study was not intended to create a complete validated risk score of CRPC. Therefore, no other covariates were used when creating our score. CRPC men having low risk characteristics with low PSA at the date of CRPC and long PSA-DT belong to risk group 1, whereas these values are vice versa for risk group 5 patients (Figure 3).

| Statistical analysis
Univariable and multivariable Cox proportional hazards regression was performed for overall mortality and PC specific mortality. 16 Cumulative incidence of death was computed considering death from PC and other causes as competing risks, censoring for end of follow-up or date of move from the region. Harrell's C-index, also known as the concordance index, is a goodness-of-fit measure for models that produce risk scores. 17 C-index was calculated for each variable in Figure 2 to validate those being most representative for predicting mortality.

| Baseline characteristics
A total of 4098 patients were eligible for analysis (Table 1) Table 1.

| Study limitations
The algorithm used to identify men with CRPC, using increase in PSA whilst on hormonal treatment, and without the use of testosterone may lead to minor misclassification and loss of some patients where clinical action was taken prior to an increase in PSA, which led to the inclusion in this cohort.

| Study strengths
We performed sensitivity analysis in which PSA measurements taken after a long period without GnRH were ignored. The result did not change as could be expected given the high adherence presented in Gincy George's publication on long-term medical adherence to GNRH agonist treatment in men with PC in Sweden. 31 The source population of this study includes almost 40% of the Swedish population and a total of 4098 CRPC patients, which is one of the largest CRPC cohorts to our knowledge. The length of the follow up and the high-quality data in our population-based registers 32 are also major strengths of this study. It is a real-life study including patients with a larger burden of comorbidity and higher age, thus being more representative for clinical purpose.
Our aim with this study was to explore if and how clinically commonly used variables were associated with the risk of mortality in this F I G U R E 4 Stacked cumulative incidence of PC death and death of other causes in men with primary ADT and secondary ADT by the defined CRPC-risk categories in Figure 3 group of patients. The findings of this study are mainly of hypothesis generating nature, and our aim is to further explore the results in future separate cohorts to eventually create a clinically useful tool to predict the outcome of patients with CRPC.