You are here
Survival and clinical metastases among prostate cancer patients treated with androgen deprivation therapy in Sweden
Cancer Epidemiology, 4, 38, pages 442 - 447
- We study prostate cancer patients treated with androgen deprivation therapy.
- Cumulative incidences of metastases and castration resistance are estimated.
- Overall survival and bone metastasis-free survival is estimated.
- PSA characteristics are important risk determinants of metastasis and death.
To examine the incidence of metastases and clinical course of prostate cancer patients who are without confirmed metastasis when initiating androgen deprivation therapy (ADT).
Retrospective cohort study conducted using electronic medical records from Swedish outpatient urology clinics linked to national mandatory registries to capture medical and demographic data. Prostate cancer patients initiating ADT between 2000 and 2010 were followed from initiation of ADT to metastasis, death, and/or end of follow-up.
The 5-year cumulative incidence (CI) of metastasis was 18%. Survival was 60% after 5 years; results were similar for bone metastasis-free survival. The 5-year CI of castration-resistant prostate cancer (CRPC) was 50% and the median survival from CRPC development was 2.7 years. Serum prostate-specific antigen (PSA) levels and PSA doubling time were strong predictors of bone metastasis, any metastasis, and death.
This study provides understanding of the clinical course of prostate cancer patients without confirmed metastasis treated with ADT in Sweden. Greater PSA values and shorter PSA doubling time (particularly ≤ 6 months) were associated with increased risk of bone metastasis, any metastasis, and death.
Keywords: ADT, CRPC, Metastasis, Prostate cancer, PSA.
Internationally, prostate cancer is the second most common cancer diagnosed among men, and the sixth most common cause of cancer death among men  . Despite efforts at better screening and treatment, prostate cancer continues to be a major public health burden. In Sweden, prostate cancer is the most common cancer in men with an average annual increase in incidence of 2.7% over the last 20 years and 0.8% over the last 10 years  . The increased incidence of prostate cancer is attributed to widespread serum prostate-specific antigen (PSA) testing in the population and the majority of new cases are diagnosed with non-metastatic disease. Approximately 15–40% of men diagnosed with prostate cancer who undergo therapy with curative intent will develop recurrent disease. The initial presentation of recurrence is most frequently evidenced by an increase in serum PSA in the absence of metastases, , and . Androgen deprivation therapy (ADT) has become the standard of care for men with PSA recurrence after primary therapy, , and . ADT includes gonadotropin-releasing hormone (GnRH) analog therapy and, to a much less frequent extent, bilateral orchiectomy  . Although prostate cancers are usually initially “castration sensitive” and respond to ADT  , nearly all patients progress to castration resistant prostate cancer (CRPC), a disease state defined by rising PSA during effective ADT.
It is of importance to provide a better understanding of the clinical course and natural history of prostate cancer and the characteristics of patients at high risk of developing clinical metastases such as those developing CRPC; currently, there is little evidence available in the literature, particularly for European populations. In addition, the few studies that have examined predictors of metastasis have largely focused on patient and tumor characteristics at cancer diagnosis, and have not examined how time-varying factors, such as PSA levels, may influence a patient's risk of disease progression. PSA characteristics after diagnosis and after ADT may be equally or more important predictors of bone metastases and survival of men with non-metastatic prostate cancer treated with ADT  .
The objective of this study was to estimate the cumulative incidence (CI) and evaluate PSA-based predictors of (1) bone metastasis, (2) any metastasis, (3) death, and (4) bone metastasis-free survival (BMFS) in the population of non-metastatic prostate cancer patients treated with continuous ADT. The objectives also included estimating the CI of CRPC among ADT-treated patients and the CI of becoming at high risk of developing bone metastases among patients with non-metastatic CRPC using clinical trial-based definitions of high risk CRPC, , and .
2. Patients and methods
2.1. Study design
This retrospective cohort study was based on electronic medical record (EMR) and national registry data in Sweden. Subjects were followed longitudinally in EMR data in a Swedish outpatient urology clinic setting until December 31, 2010. We focused our study population on outpatient urology clinics since non-metastatic prostate cancer patients in Sweden are commonly seen in this setting, as opposed to in inpatient urology clinics or by oncologists. The unique personal identification number that is assigned to all Swedish citizens enabled individual-level linkage from EMRs to several national mandatory registers. The data extraction from EMRs at urology clinics was performed using the extraction platform Pygargus CXP, which was validated in a recent study  . The extraction of individual patient-level data is anonymized, and ensured by use of unique anonymous study identification numbers in place of the personal identification numbers. The study was approved by the Regional Ethics Committee in Uppsala in November 2011 (2011/406).
2.2. Study population
The study population included all men with prostate cancer initiating continuous ADT (defined as GnRH agonist/antagonist therapy for ≥6 months, or bilateral orchiectomy) between 2000 and 2010, who had not been diagnosed with bone or distant metastasis (hereafter “non-metastatic”), had at least two PSA values recorded after ADT, and who were monitored at eight outpatient urology clinics across Sweden. Selection criteria employed for the urology clinics ensured that study sites had sufficient history and follow-up of prostate cancer patients, adequate and consistent data contained in EMRs for the time period of the study, and geographical representation in Sweden.
Prostate cancer patients who had a diagnosed metastasis (ICD-9 code 196, 197,198 or ICD-10 code C78-79, C77 at distant sites, or M1 status) on or before the date of continuous ADT were excluded from the study. Consistent with TNM classification, men whose prostate cancer diagnosis included local or regional lymph node involvement (ICD-10 C77) were not excluded from the study on this basis alone. The population thus included men diagnosed with stage I-III prostate cancer, or stage IV where the spread of disease was limited to regional lymph nodes. Subset analyses were performed on patients with evidence of CRPC.
The Swedish Cancer Registry was used to identify prostate cancer diagnosis, date of diagnosis, and the presence of metastasis and lymph node involvement at diagnosis. The overall completeness of this registry is high and comparable to other high quality registers in Northern Europe  . The EMRs were used to collect PSA laboratory data, details on GnRH agonist therapy, and concurrent treatments. Data from in- and outpatient care in hospitals were collected from the Swedish National Patient Register (NPR). The NPR is a mandatory national register and a well validated data source that has been used in a wide range of research projects  . The NPR allowed for tracking of subjects that transitioned from the urology clinics to (typically) hospital-based oncology clinics and retrieval of C77-C79 diagnoses (metastases) and the dates of those diagnoses. The NPR was also used to collect information on comorbidities and orchiectomies. Finally, to estimate overall survival and incidence of death, EMR data were linked to the Swedish National Register of Cause of Death.
Because outpatient diagnoses were not available in the NPR for the entire study period, comorbidities were based on recorded inpatient diagnoses from 2 years prior to ADT index date until ADT index date. Comorbidities were used for calculation of a baseline Charlson comorbidity index, where each comorbidity has a specific weight and may be categorized into three levels (index score in parenthesis): low (0), medium (1–2) and high (>2) and .
2.4. Statistical analysis
CRPC was defined as two consecutive rises in PSA: PSA1 < PSA2 < PSA3, where PSA2 and PSA3 ≥ 1.0 ng/mL. PSA1 could occur at any time after index date (i.e., the date of fulfillment of the continuous ADT criteria). CRPC and high risk CRPC definitions were consistent with recent clinical trials and . Among CRPC patients, high risk was defined as PSA doubling time (DT) ≤6 months based on the highest PSA-DT risk category described by Smith et al.  as well as recent clinical trial data  . PSA-DT was defined as the natural log of two divided by the slope of the linear regression line of the natural log of PSA (ng/mL) against time (months)  . At each time point of a recorded PSA value, no more than the three most recent PSA values were used in the regression of log PSA against time.
The index date marked the start of exposure time at risk for the primary study objectives. In the analysis of any metastasis and bone metastasis, failure was defined as presence of a diagnosis code indicating any metastasis (ICD-9/ICD-10 codes 196-198/C78-C79) and bone metastasis (ICD-9/ICD-10 codes 198.5/C79.5), respectively. CI over time was estimated using CI functions, taking the competing risk of death into account  . In the analysis of death and BMFS, Kaplan–Meier survival analysis was employed. Patients were censored at end of data availability (December 31, 2010).
Differences in patient characteristics were evaluated using t-tests for continuous variables and Pearson's chi-squared test for dichotomous and categorical variables. Competing risks regression  (for bone metastases and any metastases) and Cox regression (for overall survival and BMFS) analyses were used to examine PSA characteristics as predictors of the outcomes and generate hazard ratios (HR). Patient characteristics at prostate cancer diagnosis, index date, and during ADT were evaluated as potential covariates. PSA values and PSA-DT were measured during follow-up and were thus time-varying exposures. Thus, the data were time-split at the time point where the predictor changed value  . Univariate analysis was performed for each of the predictors individually and then a multivariate model was estimated. Covariates were identified a priori, and those with sufficient sample size and that were associated with outcomes were included in the final multivariate models. Finally, we evaluated the CI of CRPC among the ADT treated patients who had at least 3 PSA values recorded after the index date, as well as the CI of high-risk CRPC among patients who were identified as having CRPC.
All analyses were performed using STATA 12 (StataCorp LP, College Station, TX, USA). Differences were considered statistically significant ifp-values were less than 0.05.
3.1. Description of the study population
A total of 90,751 male patients were identified in the EMRs, of which 7996 had a diagnosis of non-metastatic prostate cancer recorded in the Cancer Registry between 1968 and 2010. A total of 850 of the prostate cancer patients initiated continuous ADT between 2000 and 2010. Of these, 446 had at least two PSA values measured after ADT and comprised the main cohort of the study. The remaining 404 ADT patients were excluded from the study. The stepwise inclusion and exclusion of patients is illustrated in Fig. 1 .
There were a few notable differences between the ADT patients who were excluded based on the criterion of number of PSA measurements available after ADT and patients included in the study ( Table 1 ). While there was a statistically significant difference in many of the characteristics, the clinical difference between the groups was only marked in a couple of cases, however. The proportion of patients having undergone orchiectomy was markedly higher in the excluded patients (18.1% compared to 1%), which may reflect that monitoring of PSA in orchiectomy patients took place in a different setting from the urology clinics participating in the study. The proportion of patients initiating ADT late in the study period was also noticeably higher in the excluded patients, which simply reflects the lower likelihood of at least two PSA measurements being available for patients for whom the length of available follow-up is shorter.
|Included in study; main cohort, N = 446||Excluded from study; <2 PSA values, N = 404||p-Value|
|Age at PC diagnosis (years)||75.4||7.0||76.8||7.8||0.006|
|Age at ADT index date (years)||78.1||6.5||80.1||7.3||<0.001|
|Time from PC diagnosis to ADT index date (years)||2.7||3.2||3.4||4.2||0.008|
|Charlson comorbidity categories||N||%||N||%||<0.001|
|Calendar year at PC diagnosis||N||%||N||%||0.026|
|Prior to 2001||78||17.5||78||19.3|
|Calendar year at ADT index date||N||%||N||%||<0.001|
|Western and Southern Sweden||136||30.5||76||18.8|
|Local lymph node involvement at diagnosis||3||0.7||7||1.7||0.152|
|Concurrent treatments 12 months prior to ADT index date||N||%||N||%|
|Other treatment modalities prior to ADT index date||N||%||N||%|
|Type of ADT initiated||N||%||N||%||<0.001|
The mean follow-up of patients in the main cohort was 3.32 years, for a total of 1480 person years available for analysis. The mean time from prostate cancer diagnosis to continuous ADT was approximately 2.5 years (ranging from 0.1 to over 18 years). Ninety percent of patients in the main cohort had no registered comorbidity, corresponding to “low” Charlson index in Table 1 . Continuous GnRH treatment was primarily accomplished with either leuprorelin or goserelin. Concurrent anti-androgen therapy was common. The number of patients having been treated with prostatectomy or radiation therapy prior to initiating ADT was relatively low, which may reflect the outpatient urology clinic setting of the study population as patients treated with surgery or radiation therapy are typically followed in an inpatient setting in Sweden.
3.2. Mortality and clinical metastases
As presented in Fig. 2 (a), the 5-year CI of bone metastasis among patients treated with continuous ADT reached 17% (95% conf. int.: 12.4–21.8), corresponding to a 5-year incidence rate of 36.0 per 1000 person-years (95% conf. int.: 27.2–47.7). CI of any metastasis was 18% after 5 years (95% conf. int.: 14.2–24.0), corresponding to a 5-year incidence rate of 40.6 per 1000 person-years (95% conf. int.: 31.1–52.8). The majority (87%) of metastases observed in this population were to the bone. In unadjusted analyses, patients younger than 65 at PC diagnosis were at greater risk of any metastasis (HR = 3.43) as well as bone metastasis (HR = 3.55) compared to men ≥65.
Sixty-four percent of patients were alive 5 years after ADT-index date (95% conf. int.: 57.3–70.4) and median survival was reached after 6.5 years ( Fig. 2 (e)). The survival curves for mortality and BMFS were similar, with median BMFS reached 6.3 years after ADT-index date.
3.3. Predictors of mortality and clinical metastases
The variables included in the multivariate analyses based on univariate findings were age at index date, baseline Charlson comorbidity index score, receipt of anti-androgen treatment within 1 year before index date, and region/clinic. PSA characteristics were consistent and strong predictors of the four outcomes of interest even when controlling for the other above-mentioned factors. Table 2 presents the results of the univariate and multivariate analyses for PSA characteristics. Higher PSA values and shorter DTs were in general associated with increased risk of all outcomes. Specifically, a PSA doubling time ≤ 6 months and PSA ≥ 20 ng/mL were associated with highest adjusted HRs of 6.9 (95% conf. int.: 4.7–10.1) and 7.4 (95% conf. int.: 5.1–10.7) for BMFS, respectively. All evaluated PSA-based predictors were statistically significant (p < 0.001 in all cases) also when accounting for multiple testing using the Bonferroni correction.
|Variable a||Bone metastasis||Any metastasis||Mortality||BMFS|
|Hazard ratio||95% CI||Hazard ratio||95% CI||Hazard ratio||95% CI||Hazard ratio||95% CI|
|PSA ≥ 8 ng/mL||15.5||8.0–30.0||10.7||6.1–18.9||4.3||2.9–6.2||5.4||3.8–7.8|
|PSA ≥ 20 ng/mL||14.2||8.1–24.8||12.0||7.2–19.9||5.7||3.9–8.3||7.3||5.1–10.5|
|PSA doubling time|
|PSA-DT ≤ 6 months||14.5||8.3–25.3||10.8||6.5–18.1||6.0||4.1–8.8||7.3||5.0–10.7|
|PSA-DT ≤ 10 months||10.7||5.8–19.5||7.3||4.3–12.4||4.2||2.9–6.1||4.8||3.4–6.9|
|PSA ≥ 8 ng/mL||13.3||6.7–26.4||9.2||5.1–16.6||4.7||3.2–6.9||5.5||3.8–8.0|
|PSA ≥ 20 ng/mL||12.6||7.3–21.7||10.6||6.5–17.5||6.1||4.1–9.0||7.4||5.1–10.7|
|PSA doubling time|
|PSA-DT ≤ 6 months||13.9||8.0–24.2||10.4||6.3–17.2||5.7||3.9–8.5||6.9||4.7–10.1|
|PSA-DT ≤ 10 months||9.5||5.2–17.5||6.5||3.8–11.2||4.0||2.7–5.8||4.6||3.2–6.6|
a The reference category for each respective regression is the complement of the PSA determinant of interest. The reference category for regressions of PSA ≥ 8 is thus PSA < 8, and so forth.
b Multivariate models include age at index date, Charlson comorbidity index, receipt of anti-androgen treatment, and region/clinic. Each row in this table shows the result for a specific PSA characteristic on the four outcomes of interest in such a model. Thus, the table corresponds to 16 estimated multivariate regression models.
3.4. Disease progression in patients treated with ADT
The analysis of CI of CRPC development was performed on patients who had at least three PSA values measured after ADT (N = 359). As seen in Fig. 2 (c), the 5-year CI of developing CRPC was 50%, reaching close to 80% after 8 years, indicating that almost all patients treated with continuous ADT will eventually see rising PSA values. The median survival in patients with CRPC was 2.7 years. The analysis of high risk development is shown in Fig. 2 (d). At 2 years, the CI of becoming at high risk of bone metastases (PSA-DT ≤ 6 months) for patients who developed CRPC was 70%, reaching over 80% after 5 years. The CI of becoming high risk within 2 years according to an alternative definition used in a recent clinical trial (PSA ≥ 8 ng/mL and/or a PSA doubling time ≤10 months)  was approximately 80% for patients with CRPC.
We further evaluated PSA-DT as a risk factor for BMFS among CRPC patients. Out of the 151 CRPC patients, 3 met the criteria for CRPC after diagnosis of bone metastasis. Among 148 patients evaluated for risk of BMFS after CRPC, 89 had a PSA-DT > 6 months at baseline (time of CRPC) and 59 had a DT ≤ months at baseline. In the univariate analysis, patients with a PSA-DT ≤ 6 months had a 2-fold increased risk of BMFS (HR = 2.01; log-rankp-value = 0.008; 95% conf. int.: 1.19–3.41) ( Fig. 3 ).
This study provides the first understanding of the clinical course of men diagnosed with prostate cancer without evidence of clinical metastases who are treated with ADT in Sweden. Real-world data from outpatient urology clinics were used to identify the cohort (N = 446) of non-metastatic prostate cancer patients actively receiving ADT from 2000 to 2010. The median survival of ADT-treated men was reached after 6 years, and 80% of patients eventually demonstrated evidence of CRPC. The CI of diagnosed metastasis was slightly above 5% 2 years after continuous ADT, reaching 20% after 6 years. Diagnosed metastases were predominantly bone metastases. In clinical practice, the monitoring of this patient population is focused primarily on identification of bone metastases, and screening for non-bone metastases is not routine. Hence, the incidence of non-bone metastases may be under-estimated.
This study also demonstrates the importance of PSA and PSA-DT as risk determinants of metastasis and death in patients receiving ADT and corroborates previous studies in this patient population. In a study by Abouassaly et al. in patients from the US  , higher baseline PSA and PSA velocity were both shown to be associated with the development of metastatic disease after ADT. Our study further demonstrates that PSA values after initiation of ADT are also strong predictors of metastases and death. Smith et al.  analyzed a sample of patients in the placebo arm of a randomized, placebo-controlled trial who had PSA rises during ADT treatment. Baseline PSA and PSA doubling time were identified as predictors of bone metastasis, death and metastasis-free survival  ; specifically, CRPC patients with PSA-DT < 6.3 months and those with PSA > 24 ng/mL represented patients in the top tertile of risk of bone metastasis or death. In our study, we were able to corroborate these findings to a great extent in the real-world setting: PSA-DT ≤ 6 months and PSA ≥ 20 ng/mL were associated with high hazard ratios of 6.9 (95% conf. int.: 4.7–10.1) and 7.4 (95% conf. int.: 5.1–10.7) for BMFS, respectively. Analysis of PSA-DT specifically within our CRPC cohort provides further support for PSA-DT ≤ 6 months as a high risk characteristic for BMFS.
A significant advantage of this EMR-based study was the ability to examine patient PSA values recorded over time. The majority of population-based sources of data commonly used for research purposes lack serial PSA measures, which are critical to evaluating risk determinants. It is important to note that these data reflect real-world PSA testing patterns in clinical practice (1.25 PSA tests per person-year), whereas PSA testing in clinical trials occur much more frequently at pre-specified time intervals (e.g., every 2 months)  . It is also worth noting that patients in Sweden may be administered treatment by a district nurse and do not need to see the treating physician each time treatment is prescribed, which could lead to PSA being monitored to a slightly lower extent in Sweden than in other countries. Another limitation of this study is that we were not able to confirm that castration levels of testosterone were achieved for patients treated with continuous ADT because testosterone values were not consistently available in the EMRs. In addition, the study population may not be completely representative of the entire prostate cancer population due to the identification of patients in the outpatient urology clinic setting and the inclusion criteria of having at least two PSA values available after ADT. Finally, because the recording of metastases may not be complete in the NPR in Sweden, it is possible that some patients were in fact metastatic at time of ADT.
In conclusion, almost all prostate cancer patients treated with androgen deprivation therapy will eventually see rising PSA values. PSA characteristics are important risk determinants of metastases and death in prostate cancer patients treated with androgen deprivation therapy and may thus help to guide treatment decisions in these patients. Real-world data has so far been used sparingly to study this patient population. Our findings using real-world data corroborate and complement those of clinical trials that have identified PSA characteristics as predictors of risk of prostate cancer related outcomes.
Conflict of interest statement
A Liede and R Hernandez are employees of and hold stock in Amgen, Inc. J Banefelt and J Mesterton are employees of Quantify Research, a contract research organization acting as consultants to the pharmaceutical industry. P Sobocki is an employee of Pygargus, a contract research organization acting as consultant to the pharmaceutical industry. J Banefelt, J Mesterton, P Sobocki and B-E Persson report personal fees from Amgen, Inc. during the conduct of the study. J Stålhammar has no conflict of interest to declare.
This study was supported by Amgen, Inc.
-  A. Jemal, F. Bray, M.M. Center, J. Ferlay, E. Ward, D. Forman. Global cancer statistics. CA Cancer J Clin. 2011;61(2):69-90 Crossref
-  Socialstyrelsen (National Board of Health and Welfare). Cancer Incidence in Sweden. (, 2010) 1 http://www.socialstyrelsen.se/Lists/Artikelkatalog/Attachments/18204/2010-12-17.pdf [accessed 21.08.11]
-  C.R. Pound, A.W. Partin, M.A. Eisenberger, D.W. Chan, J.D. Pearson, P.C. Walsh. Natural history of progression after PSA elevation following radical prostatectomy. JAMA. 1999;281(17):1591-1597 Crossref
-  D.V. Makarov, E.B. Humphreys, L.A. Mangold, M.A. Carducci, A.W. Partin, M.A. Eisenberger, et al. The natural history of men treated with deferred androgen deprivation therapy in whom metastatic prostate cancer developed following radical prostatectomy. J Urol. 2008;179(1):156-161 discussion 61–62
-  M.J. Donovan, S. Hamann, M. Clayton, F.M. Khan, M. Sapir, V. Bayer-Zubek, et al. Systems pathology approach for the prediction of prostate cancer progression after radical prostatectomy. J Clin Oncol. 2008;26(24):3923-3929 Crossref
-  J. Kawakami, J.E. Cowan, E.P. Elkin, D.M. Latini, J. DuChane, P.R. Carroll, et al. Androgen-deprivation therapy as primary treatment for localized prostate cancer: data from Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE). Cancer. 2006;106(8):1708-1714 Crossref
-  N. Sharifi, J.L. Gulley, W.L. Dahut. Androgen deprivation therapy for prostate cancer. JAMA. 2005;294(2):238-244 Crossref
-  S.M. Gilbert, Y.F. Kuo, V.B. Shahinian. Prevalent and incident use of androgen deprivation therapy among men with prostate cancer in the United States. Urol Oncol. 2009;29(6):647-653
-  V.B. Shahinian, Y.F. Kuo, S.M. Gilbert. Reimbursement policy and androgen-deprivation therapy for prostate cancer. New Engl J Med. 2010;363(19):1822-1832 Crossref
-  M.J. Shulman, E.A. Benaim. The natural history of androgen independent prostate cancer. J Urol. 2004;172(1):141-145 Crossref
-  R. Abouassaly, A. Paciorek, C.J. Ryan, P.R. Carroll, E.A. Klein. Predictors of clinical metastasis in prostate cancer patients receiving androgen deprivation therapy: results from CaPSURE. Cancer. 2009;115(19):4470-4476 Crossref
-  M.R. Smith, F. Saad, R. Coleman, N. Shore, K. Fizazi, B. Tombal, et al. Denosumab and bone-metastasis-free survival in men with castration-resistant prostate cancer: results of a phase 3, randomised, placebo-controlled trial. Lancet. 2012;379(9810):39-46
-  M.R. Smith, F. Kabbinavar, F. Saad, A. Hussain, M.C. Gittelman, D.L. Bilhartz, et al. Natural history of rising serum prostate-specific antigen in men with castrate nonmetastatic prostate cancer. J Clin Oncol. 2005;23(13):2918-2925 Crossref
-  M.R. Smith, F. Saad, S. Oudard, N. Shore, K. Fizazi, P. Sieber, et al. Denosumab and bone metastasis-free survival in men with nonmetastatic castration-resistant prostate cancer: exploratory analyses by baseline prostate-specific antigen doubling time. J Clin Oncol. 2013;31(30):3800-3806 Crossref
-  M. Martinell, J. Stalhammar, J. Hallqvist. Automated data extraction – a feasible way to construct patient registers of primary care utilization. Ups J Med Sci. 2012;117:52-56 Crossref
-  L. Barlow, K. Westergren, L. Holmberg, M. Talback. The completeness of the Swedish Cancer Register: a sample survey for year 1998. Acta Oncol. 2009;48(1):27-33 Crossref
-  J.F. Ludvigsson, E. Andersson, A. Ekbom, M. Feychting, J.-L. Kim, C. Reuterwall, et al. External review and validation of the Swedish national inpatient register. BMC Public Health. 2011;11(450)
-  M.E. Charlson, P. Pompei, K.L. Ales, C.R. MacKenzie. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chron Dis. 1987;40(5):373-383 Crossref
-  H. Quan, V. Sundararajan, P. Halfon, A. Fong, B. Burnand, J.-C. Luthi, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care. 2005;43(11):1130-1139 Crossref
-  MSKCC. Available from: http://nomograms.mskcc.org/Prostate/PsaDoublingTime.aspx [cited 17.12.12].
-  T.A. Gooley, W. Leisenring, J. Crowley, B.E. Storer. Estimation of failure probabilities in the presence of competing risks: new representations of old estimators. Stat Med. 1999;18:695-706
-  J.P. Fine, R.J. Gray. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc. 1999;94(446)
-  S. Suissa. Immortal time bias in pharmacoepidemiology. Am J Epidemiol. 2008;167(4):492-499 Crossref
a Quantify Research, Hantverkargatan 8, 112 21 Stockholm, Sweden
b Amgen Inc., Center for Observational Research, South San Francisco, CA, United States
c Institute of Environmental Medicine, Karolinska Institute, 171 77 Stockholm, Sweden
d Department of Public Health and Caring Sciences, Uppsala University, 751 22 Uppsala, Sweden
e Amgen Inc., Center for Observational Research, One Amgen Center Drive 24-2-A, Thousand Oaks, CA 91320, United States
f IMS Health/Pygargus, Sveavägen 155, 113 46 Stockholm, Sweden
g LIME/Medical Management Centre, Karolinska Institute, 171 77 Stockholm, Sweden
h Uppsala University, 751 05 Uppsala, Sweden
© 2014 Elsevier Ltd, All rights reserved.