Johansson J, Andrén O, Andersson S, Dickman PW, Holmberg L, Magnuson A, Adami H. Natural History of Early, Localized Prostate Cancer. JAMA. 2004;291(22):2713-2719. doi:10.1001/jama.291.22.2713
Author Affiliations: Departments of Urology and Clinical Medicine (Drs Johansson, Andrén, and Andersson) and Statistical Unit, Center for Clinical Research (Mr Magnuson), Örebro University Hospital, and Center for Assessment of Medical Technology (Drs Johansson and Andersson), Örebro, Sweden; Department of Medical Epidemiology and Biostatistics, Karolinska Institute, Stockholm, Sweden (Drs Dickman and Adami); Regional Oncologic Center, University Hospital, Uppsala, Sweden (Dr Holmberg); and Department of Epidemiology and the Harvard Center for Cancer Prevention, Harvard School of Public Health, Boston, Mass (Dr Adami).
Context Among men with early prostate cancer, the natural history without initial
therapy determines the potential for survival benefit following radical local
treatment. However, little is known about disease progression and mortality
beyond 10 to 15 years of watchful waiting.
Objective To examine the long-term natural history of untreated, early stage prostatic
Design Population-based, cohort study with a mean observation period of 21
Setting Regionally well-defined catchment area in central Sweden (recruitment
March 1977 through February 1984).
Patients A consecutive sample of 223 patients (98% of all eligible) with early-stage
(T0-T2 NX M0 classification), initially untreated prostatic cancer. Patients
with tumor progression were hormonally treated (either by orchiectomy or estrogens)
if they had symptoms.
Main Outcome Measures Progression-free, cause-specific, and overall survival.
Results After complete follow-up, 39 (17%) of all patients experienced generalized
disease. Most cancers had an indolent course during the first 10 to 15 years.
However, further follow-up from 15 (when 49 patients were still alive) to
20 years revealed a substantial decrease in cumulative progression-free survival
(from 45.0% to 36.0%), survival without metastases (from 76.9% to 51.2%),
and prostate cancer–specific survival (from 78.7% to 54.4%). The prostate
cancer mortality rate increased from 15 per 1000 person-years (95% confidence
interval, 10-21) during the first 15 years to 44 per 1000 person-years (95%
confidence interval, 22-88) beyond 15 years of follow-up (P = .01).
Conclusion Although most prostate cancers diagnosed at an early stage have an indolent
course, local tumor progression and aggressive metastatic disease may develop
in the long term. These findings would support early radical treatment, notably
among patients with an estimated life expectancy exceeding 15 years.
Without understanding the natural history of prostate cancer diagnosed
at an early, localized stage, patient counseling and clinical management are
difficult. The challenge is to maximize the possibilities for survival without
extensive overtreatment. Even without initial treatment, only a small proportion
of all patients with cancer diagnosed at an early clinical stage die from
prostate cancer within 10 to 15 years following diagnosis.1- 3 However,
to our knowledge, no study has hitherto adequately analyzed whether patients
who escaped metastasis and death during those 10 to 15 years without treatment
continue to have an indolent, nonfatal disease course or whether in the long
term, tumor progression takes a more aggressive course. Recently, a randomized
trial4 demonstrated that radical prostatectomy
may further reduce the low-death rate in early prostate cancer by approximately
50%. Because it takes several years after operation for this benefit to emerge,
age at diagnosis, comorbidity that influences life expectancy, and long-term
natural history will determine the potential advantage with radical primary
This study focuses on information that aids clinical decision making,
namely the association between prognostic factors available at diagnosis and
the long-term natural history in patients without initial treatment. Such
knowledge can help us understand whether there are men with prostate cancer
and a long life expectancy in whom early radical treatment might be justified
despite the fact that they have favorable prognostic signs seen from a perspective
of 5 to 10 years of follow-up. We studied these issues prospectively in the
largest population-based cohort ever impaneled to analyze survival following
watchful waiting of patients with early prostate cancer. Complete follow-up
of this cohort has now been achieved during an average of 21 years, and only
9% of the patients are still alive.
The patients comprised a population-based cohort of patients with early,
initially untreated prostate cancer as previously described in detail.3 The TNM system5 and
the World Health Organization6 classification
of malignant diseases were used. At the time of diagnosis, all patients underwent
a clinical examination, excretory urography, chest radiography, bone scan,
and skeletal radiography (if needed) and had routine blood samples taken.
The nodal status was not known for any of the patients. Prostate-specific
antigen (PSA) testing was not available, and no screening activities for prostate
cancer took place during the period when this cohort was recruited.
From March 1977 through February 1984, a total of 654 new cases of prostate
cancer were diagnosed among residents in the regionally well-defined catchment
area in central Sweden. Patients were given no initial treatment if the tumor
growth was localized to the prostate gland as judged by digital rectal examination
(T0-T2) and no distant metastases were present (306 patients). This was in
accordance with the standard management at the time in Sweden. The following
restrictions were applied, however, among those with palpable tumors (T1-T2).
From March 1977 through February 1979, only patients with a highly differentiated
tumor (grade 1) were included in the untreated group. From March 1979 through
the end of the recruitment period, patients younger than 75 years at diagnosis
and with moderately or poorly differentiated tumors (grades 2-3) were randomly
allocated to receive local radiation (10 patients) or no treatment, and only
the latter group was included in this cohort study. Patients 75 years or older
were not treated and included in the study.
Among the 227 eligible patients, 4 (2%) were given initial treatment
and had to be excluded from the analyses. The distribution of the study group
of 223 patients by age, stage, and grade at the time of diagnosis is shown
in Table 1. The mean age at diagnosis
was 72 years (range, 41-91 years). Altogether 106 (48%) cases were detected
by histopathologic examinations of specimens obtained at operations for suspected
benign prostatic hyperplasia. The remaining 117 patients had a palpable clinical
disease localized to the prostate gland. A review by an experienced histopathologist
confirmed the initial diagnosis in all cohort members. Approximately two thirds
of patients had highly differentiated tumors, whereas only 9 (4%) had a poorly
differentiated tumor (Table 1).
All 223 patients were followed up from diagnosis until death or the
end of the observation period (September 1, 2001). No patient was lost to
follow-up. Clinical examination, laboratory tests, and bone scans were performed
every 6 months during the first 2 years after diagnosis and subsequently once
a year during the first 10 years of observation and thereafter at least once
every second year. Those in whom the cancer progressed to symptomatic disease
were treated with exogenous estrogens or orchidectomy.
Local progression was defined as tumor growth through the prostate capsule
(T3) as judged by digital rectal examination. Development of distant metastasis
(M1) was classified as generalization. If both local progression and metastatic
disease were present, the patient was classified as having generalized disease.
During the first 6 years of follow-up, all patients who were still alive
and consented underwent a new fine-needle biopsy every other year. We obtained
such biopsy specimens from 178 (80%) of the 223 patients. Although this procedure
has lower than 100% sensitivity, notably for impalpable tumors, remaining
cancer growth was confirmed cytologically in most patients: 45 (73%) among
those with T01 disease, 24 (92%) among those with T0d disease, and 90 (100%)
among those with T1-2 disease. Altogether 31 (17%) of 178 patients showed
evidence of dedifferentiation. Twenty-one patients (18%) changed from high
to moderate differentiation, 7 (13%) from moderate to low differentiation,
and 3 (3%) from high to low differentiation.
The medical records of all deceased patients were reviewed. In most
instances, the cause of death determined in real time was obvious on clinical
grounds alone. An autopsy was performed if the cause of death was not clear.
Prostate cancer was recorded as the underlying cause of death, a contributory
cause of death, or unrelated to death as described in detail in a previous
report.3 If treatment of the prostate cancer
was related to death (chiefly due to cardiovascular complications following
estrogen administration), prostate cancer was recorded as a contributory cause.
As a validation, we compared our own classification of causes of death with
those recorded in the Swedish Death Register. This information was obtained
through record linkage between our study cohort and the Swedish Death Register
based on the individually unique national registration number assigned to
all Swedish residents. There was agreement in 90% of the patients and no evidence
of systematic overascertainment or underascertainment of prostate cancer as
cause of death in our data. Although based on small numbers, there was no
evidence that the disagreement was larger in older than in younger patients.
We estimated various measures of patient survival using the actuarial
(life-table) method.7 Cause-specific survival
was estimated by considering only deaths due to prostate cancer as events
of interest (deaths due to other causes were considered censored), observed
survival by considering deaths due to any cause as events, and progression-free
survival by considering progression as the event of interest. We also estimated
relative survival, defined as the ratio of observed survival to the expected
survival of a comparable group from the general population assumed to be free
of prostate cancer. We estimated expected survival using the Hakulinen method8 based on Swedish population life tables stratified
by age, sex, and calendar time. We also calculated prostate cancer–specific
mortality rates (deaths per 1000 person-years at risk) and associated 95%
confidence intervals (CIs).9 We estimated Poisson
regression models9 to study the association
between prostate cancer mortality and time since diagnosis while adjusting
for age at diagnosis, stage, and grade. Relative survival was estimated using
software developed at the Finnish Cancer Registry.10 All
other analyses were performed using Stata statistical software (Stata Corporation,
College Station, Tex). All reported P values are
2-sided. Statistical significance was P<.05.
During a mean observation period of 21 years, 89 patients (40%) experienced
progression of disease, and of these 39 (17% of the entire cohort) developed
generalized disease. A total of 203 patients (91% of the entire cohort) died
during follow-up, with prostate cancer considered the cause of death in 35
(16% of the entire cohort; Table 1).
Among patients who were 70 years or younger at diagnosis, 22 (22%) died from
prostate cancer during follow-up, whereas this proportion decreased markedly
at higher ages. The proportion of patients dying from prostate cancer was
strikingly similar among those with nonpalpable (T0) tumors detected at transurethral
resection (18 patients [17%]) and those with a palpable tumor (17 patients
[15%]). In contrast, poor differentiation was a strong predictor of prostate
cancer–specific death (Table 1).
Although based on small numbers, the progression and mortality rates
remained fairly constant during the first three 5-year periods following diagnosis
(Table 2). Averaged over the first
15 years, the rate of progression to metastatic disease was 18 per 1000 person-years
(95% CI, 13-25) and the prostate cancer mortality rate was 15 per 1000 person-years
(95% CI, 10-21). In contrast, an approximately 3-fold higher rate was found
both for progression and death during follow-up beyond 15 years (Table 2). This increase was almost statistically
significant for progression (P = .06) and statistically
significant for death (P = .01).
Table 3 shows various measures
of survival after 15 and 20 years of follow-up. During this 5-year period,
the progression-free survival among all patients decreased from 45.0% to 36.0%.
The low and rapidly decreasing observed survival reflects chiefly the impact
of causes of death other than prostate cancer. Most notably, however, we found
a substantial decline by approximately 25 percentage points in both the relative
and the cause-specific survival rate during the last 5 years of follow-up. Figure 1 further illustrates how a gradual
decline in relative and cause-specific survival seemingly occurred more rapidly
after approximately 16 years of follow-up. This change seemed to affect tumors
regardless of initial stage and also to affect tumors that were both initially
highly and moderately differentiated (Figure
2). The gloomy outlook among patients with poorly differentiated
tumors became manifested already within the first 5 years of follow-up.
Prostate cancer mortality was slightly higher among patients whose cancer
was diagnosed at 70 years or younger than among those whose cancer was diagnosed
at older ages (Table 2). Strikingly
similar mortality rates were found among patients who had localized nonpalpable
cancer compared with those who had a cancer in stage T1 or T2. In contrast,
the mortality rate was 70% higher among men with a nonpalpable diffuse cancer.
With regard to differentiation, the mortality rate increased drastically from
highly to poorly differentiated tumors (Table 2).
Multivariable Poisson regression models were fitted to quantify the
independent effects of follow-up time, age at diagnosis, grade, and stage
(Table 4). Our analyses showed
a significant (approximately 6-fold) higher mortality rate after 15 years
of follow-up compared with the first 5 years. The strong prognostic impact
of grade, notably of poorly differentiated tumors, was also confirmed. In
contrast, neither age at diagnosis nor stage of disease was significantly
associated with risk of death due to prostate cancer. Although we had limited
power to test interaction, the risk of death due to prostate cancer after
15 or more years of follow-up compared with 0 to 14 years was similar among
patients with cancer diagnosed before (relative risk, 4.1) and after (relative
risk, 3.1) 70 years of age.
A separate model was fitted in which the event of interest was local
progression. In this analysis, we disregarded if and when regional and/or
distant progression or metastases were ascertained. Except for age at diagnosis,
the pattern for local progression was strikingly different from that of death
due to prostate cancer (Table 4).
Hence, the risk of local progression did not increase over follow-up time,
and the association with grade was weak. Moreover, compared with T01 tumors,
growth beyond the prostate capsule was 2 or 3 times more likely in patients
with T0d and T1-2 tumors, respectively.
Although our cohort of patients with early stage, initially untreated
prostate cancer has been previously followed up in great detail during an
average of 15 years,3 the additional 6 years
included in this analysis revealed an unexpected change in prognostic outlook;
the cause-specific survival rate decreased by almost 25 percentage points,
reflecting an approximate 3-fold increase in prostate cancer mortality rate
compared with the first 15 years of follow-up. This change occurred consistently
across stage and grade except for poorly differentiated cancers in which excess
mortality becomes manifest already during early follow-up. We were unable
to conceive of any bias that could have spuriously generated these recent
findings. Indeed, the internal validity of our population-based study should
be high because we achieved complete follow-up and used standardized procedures
for clinical examination, ascertainment of disease progression, and classification
of death. Moreover, the slight difference between cause-specific and relative
survival estimates were largely consistent over time. This argues against
any shifting criteria for classification of cause of death, since the relative
survival rate reflects excess mortality (compared with mortality in the general
population) and is thus unaffected by any subjective judgment. Prostate cancer
mortality rates were mirrored closely by rates of disease progression to metastatic
disease. Hence, chance is the only realistic alternative to a real deterioration
in prognosis after long-term follow-up, and the level of statistical significance
argues against this explanation.
If our data reflect a real phenomenon, they would imply that the probability
of progression from localized and indolent to metastatic mortal disease increases
markedly after long-term follow-up. This progression is not restricted to
cancers diagnosed due to clinical symptoms but includes also tumors detected
incidentally at transurethral resection due to presumed benign prostatic hyperplasia.
Our survival data, supported by biopsy specimens taken during follow-up, would
further imply that these latter lesions are either incompletely removed or
multifocal with malignant clones left at a transurethral resection. Contrary
to emerging views,11,12 our data
also suggest that metastases may arise as a consequence of late mutations
rather than being determined already by the early mechanisms of malignant
transformation. According to a rival interpretation, the phenomenon we observed
reflects transformation of new, more aggressive cancer clones rather than
progression of those initially detected. Empirical testing of these complementary,
but not mutually exclusive, theories seems difficult.
It may be difficult to validate our survival data in any new cohort
study of watchful waiting since aggressive treatment of localized prostate
cancer has become more routine now than it was 25 years ago when we started
to assemble our cohort.13- 16 Indeed,
it has been estimated that approximately 60 000 men undergo radical prostatectomy
yearly in the United States alone, and the number performed annually in England
increased nearly 20-fold between 1991 and 1999.17 This
development toward treatment with a curative intent may further accelerate
following recent documentation that radical prostatectomy reduces prostate
cancer mortality by approximately 50%.4 Hence,
support for our findings has to be found chiefly in existing studies of watchful
waiting. Other such cohorts are, however, few and small, none of them are
population based and prospective,2,18- 20 and
virtually no follow-up data are available beyond 15 years after diagnosis.
Within these constraints, experience during the first 10 years after diagnosis
is strikingly similar in existing cohorts of patients with early stage prostate
cancer left without initial treatment, with a favorable course of the disease
for men with highly or moderately differentiated tumors.1
From a public health perspective, implications of late progression from
early stage to mortal disease may not be significant because without PSA testing,
average age at diagnosis of prostate cancer is so high that competing causes
of death predominate (Table 1).
Although it is well established21,22 that
an excess death rate continues long term in population-based cohorts of prostate
cancer patients, these data do not enable distinction of deaths generated
by patients initially diagnosed as having localized disease. In our entire
cohort, 25 (11%) of 223 patients died from prostate cancer within 15 years
of diagnosis and an additional 10 during subsequent follow-up until a time
when only 9% of all patients in the cohort were still alive and therefore
at risk of progression. Assuming that radical prostatectomy prevents approximately
50% of prostate cancer deaths,4 approximately
18 patients (8%) in our entire cohort (that is, 0.5 × 35) would have
experienced a survival benefit, whereas the remaining 205 would not. However,
among elderly men, reducing the risk of death from prostate cancer by a certain
amount may have limited impact on their overall survival.
Our data may be important for counseling and clinical management of
individual patients. Postponement of death is not the only treatment objective
because local progression may create substantial suffering. Indeed, many of
our patients experienced symptomatic local growth without generalized disease
(Table 1), requiring treatment
with estrogens or orchidectomy. Obviously, radical prostatectomy is a major
procedure with substantial adverse effects, chiefly impotence and incontinence.23,24 Because these complications are surprisingly
well tolerated,25 many patients may prefer
a radical prostatectomy even if prolonged survival is an uncertain consequence.
Our data may be particularly relevant to otherwise healthy men diagnosed as
having prostate cancer at an early age. If such patients are in their 60s
or younger, disease progression that occurs after 15 or more years may be
a real concern, arguing for early local treatment with a curative intent.
In patients with a PSA-detected cancer,26 such
counseling is, however, complicated by the fact that a lead time that cannot
be individually determined has to be added to the approximately 15 years that
may precede more rapid tumor progression.
One important and complicated question is how the findings of this study
relate to the current era when many patients are detected by means of PSA
testing. The results are directly relevant for patients with clinical disease
diagnosed before the PSA era and also to preclinical disease detected at transurethral
resection for presumed benign prostatic hyperplasia. Indeed, as shown in Table 4, these 2 categories of patients
experienced similar risk of dying from prostate cancer. The natural history
we have described reflects also what would happen among PSA-detected cancer
if the lead time could be accommodated and the patients were left without
early therapeutic intervention. However, a substantial proportion of PSA-diagnosed
cancers represents overdetection of subclinical disease. These cancers would
never have surfaced clinically during the patient's lifetime, either because
they are indolent or because death occurs from competing causes before clinical
manifestation of the malignancy. By definition, these cancers do not generate
In conclusion, our data indicate that the probability of progression
to a more aggressive and lethal phenotype may increase after long-term follow-up
of prostate cancers that are diagnosed at an early stage and initially left
without treatment. These findings argue for early radical treatment of patients
with long life expectancy. Not only would such surgical intervention potentially
prevent deaths, it would also convey prevention from disability caused by
local tumor growth.