European Journal of Cancer (2013) xxx, xxx xxx

Available at www.sciencedirect.com

ScienceDirect
journal homepage: www.ejcancer.com

Choosing the net survival method for cancer survival

estimation
Karri Seppa a,b, Timo Hakulinen a, Arun Pokhrel a,
a
b

Finnish Cancer Registry, Institute for Statistical and Epidemiological Cancer Research, Pieni Roobertinkatu 9, FI-00130 Helsinki, Finland
Department of Mathematical Sciences, University of Oulu, Oulu, Finland

KEYWORDS
Epidemiologic methods
Models
Neoplasms
Prognosis
Relative survival
Net survival

Abstract Background: A new net survival method has been introduced by Pohar Perme et al.
(2012 [4]) and recommended to substitute the relative survival methods in current use for
evaluating population-based cancer survival.
Methods: The new method is based on the use of continuous follow-up time, and is unbiased
only under non-informative censoring of the observed survival. However, the populationbased cancer survival is often evaluated based on annually or monthly tabulated follow-up
intervals. An empirical investigation based on data from the Finnish Cancer Registry was
made into the practical importance of the censoring and the level of data tabulation. A systematic comparison was made against the earlier recommended Ederer II method of relative
survival using the two currently available computer programs (Pohar Perme (2013) [10] and
Dickman et al. (2013) [11]).
Results: With exact or monthly tabulated data, the Pohar-Perme and the Ederer II methods
give, on average, results that are at ve years of follow-up less than 0.5% units and at 10
and 14 years 12% units apart from each other. The Pohar-Perme net survival estimator is
prone to random variation and may result in biased estimates when exact follow-up times
are not available or follow-up is incomplete. With annually tabulated follow-up times, estimates can deviate substantially from those based on more accurate observations, if the actuarial approach is not used.
Conclusion: At 5 years, both the methods perform well. In longer follow-up, the Pohar-Perme
estimates should be interpreted with caution using error margins. The actuarial approach
should be preferred, if data are annually tabulated.
2013 Elsevier Ltd. All rights reserved.

Corresponding author: Tel.: +358 9 135 33 274; fax: +358 9 135 5378.

E-mail address: arun.pokhrel@cancer. (A. Pokhrel).

0959-8049/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ejca.2013.09.019

Please cite this article in press as: Seppa K. et al., Choosing the net survival method for cancer survival estimation, Eur J Cancer (2013), http://
dx.doi.org/10.1016/j.ejca.2013.09.019

K. Seppa et al. / European Journal of Cancer xxx (2013) xxxxxx

1. Introduction
The population-based cancer registries have used relative survival to give estimates of patients net survival,
i.e. as far as the patients cancer is concerned when eliminating the eects of the other causes of death [1,2]. In
this way, no information on causes of death has been
needed as the mortality from the other causes (often
called expected mortality) has been estimated from life
tables of the underlying general population. Recently,
a recommendation of using the Ederer II relative survival method was made based on both theoretical and
empirical arguments [3]. This recommendation has been
also followed, e.g. by the pan-European EUROCARE-5
study (European cancer registry based study on survival
and care of cancer patients).
Even more recently, a new method to estimate net survival has been proposed by Pohar Perme et al. [4] as a substitute of the relative survival approach. This method is
not based on a direct comparison of an observed survival
proportion of the patients against an expected survival
proportion in the comparable general population group
as the relative survival methods. It still uses the general
population mortality as an estimate of mortality due to
the other causes, so that no information on the actual
causes of death is needed. This method, unlike the relative
survival methods, has been shown to provide an unbiased
estimator of the true net survival, if there is no informative censoring of the observed survival (e.g. censoring that
would vary by patients age [5]) and continuous time is
used in survival calculations. The international CONCORD-2 (Global surveillance of cancer survival) study
will use the Pohar-Perme net survival method.
Also the relative survival methods, including the Ederer II method, aim to estimate net survival. The Ederer II
estimator calculates the cumulative product of the interval-specic relative survival ratios, which are based on
unweighted observations of patients alive at the beginning of the corresponding intervals. Therefore, patients
who have a high probability of dying due to other causes
than cancer get too small weights in estimation of net survival, as a patients contribution to net survival is omitted
in subsequent intervals after dying. Because net survival
depends almost always on the same demographic variables as the expected hazard due to other causes than cancer, the estimator of the Ederer II method becomes
biased. In the classical relative survival methods, stratied analyses and their summarisations, e.g. by (age-)standardisation, have been conducted to reduce this bias.
In the method of Pohar Perme et al., a patients contribution to net survival is weighted on the basis of the
patients expected survival, i.e. the probability of being
alive for a healthy person in the national or other population (comparable with respect to demographic variables e.g. sex, age and calendar year). The method
may be viewed also as a generalisation of the gold

standard used in an earlier study [3] into a situation

where each patient makes her own group dened by
sex, age and year of diagnosis. The choice of weights
for each group can also be viewed natural, as in a true
gold standard, depending on the cancer-related excess
hazard of death only.
The present study investigates systematically, using
data from the population-based Finnish Cancer Registry and the two publicly available computer programs,
how crucial these two assumptions (no informative censoring of the observed survival and use of continuous
time) are, particularly the latter one, when a change of
method from the traditional relative to the new net survival is done. It is important to know, for national and
international population-based cancer survival analyses,
how much results obtained by the two methods dier
and under which conditions the new method can be recommended in practice.
2. Patients and methods
Patients diagnosed in Finland in 19811995 and followed-up until the end of 2010 were included in the analysis with stratication by the most common 26 sites.
Table 1 shows the list of the sites and the numbers of
Table 1
The 26 cancer sites included in the analyses and the numbers of
patients diagnosed in Finland in 19811995 by site and sex.
Cancer site

International
Total number of
Classication of patients
Diseases (ICD)-10
code
Males
Females

Oesophagus
Stomach
Colon
Rectum, rectosigma, anus
Liver
Gall bladder, bile ducts
Pancreas
Larynx
Lung, trachea
Skin, melanoma
Skin, non-melanoma
Soft tissues
Breast
Cervix uteri
Corpus uteri
Ovary
Prostate
Testis
Kidney
Bladder, ureter, urethra
Central nervous system
Thyroid
Hodgkin lymphoma
Non-Hodgkin lymphoma
Multiple myeloma
Leukaemia

C15
C16
C18
C1920
C22
C2324
C25
C32
C3334
C43
C44
C4849
C50
C53
C54
C56
C61
C62
C6465
C6768
C7072
C73
C81
C8285, C96
C90
C9195

1545
8071
5905
5006
1555
1020
4266
1672
25,992
3331
3538
901

21,359
893
4626
7235
3747
783
1007
4274
1625
3600

1516
7297
8449
4991
1340
2762
5166

5260
3577
4236
970
35,399
2420
7777
6043

3867
2389
5102
3128
775
4620
2035
3299

Please cite this article in press as: Seppa K. et al., Choosing the net survival method for cancer survival estimation, Eur J Cancer (2013), http://
dx.doi.org/10.1016/j.ejca.2013.09.019

K. Seppa et al. / European Journal of Cancer xxx (2013) xxxxxx

diagnosed patients by site and sex. Cancer sites with less

than 500 patients were not included. The eect of censoring was studied by using the ends of 1995 and 1999
as the alternative closing dates of follow-up. The overall
non-standardised net survival estimates were obtained
by the Ederer II relative survival method [6] and by
the method proposed by Pohar Perme et al. [4]. The
results of applying the methods were compared at 5,
10 and 14 years of follow-up by using exact follow-up
times as well as by applying annual and monthly follow-up intervals as a basis of grouping the data. In addition to the point estimates, also the precision of the
point estimates was evaluated by investigating the
lengths of condence intervals.
In an empirical comparison, a true gold standard is
not available, as even an unbiased method with no censoring is prone to give estimates with random error.
Nevertheless, due to unbiasedness and the recent recommendation [7], the results of the Pohar-Perme method
with exact and uncensored follow-up times (i.e. followed-up until the end of 2010) were selected as the gold
standard against which the other approaches were compared. The other approaches included the use of
monthly or annually grouped observations, also subject
to empirical patterns of censoring due to earlier common closing dates (1995 or 1999) and the use of the
Ederer II method instead of the Pohar-Perme method.
Results were calculated by site, but the estimates of
site-specic gold standards are very unstable. Overall
survival combining patients of all sites is more stable
but a less reasonable measure in practice [8]. Therefore,
we focused on results averaged over the various sites
with equal weights. The average gives a summary measure that treats the estimation for each site equally
important and has a smaller random error than the
site-specic results. Age-standardised results were produced by using internal age-standardisation [3,9] based
on ve age groups: 044, 4554, 5564, 6574 and
75+ years.
The calculations for the monthly and annually
grouped observations were conducted by using the both
available computer programs: the original program in R
by Pohar Perme [10] (version 2.0-4) and another program in STATA by Dickman et al. [11] (version 1.3.8).
The most accurate follow-up time was called the exact
follow-up time, although it was based on the exact date
at exit (day of death, emigration or the 31st December
2010) and an approximated date of diagnosis, as the
exact date of diagnosis is not available. The date of diagnosis was set to be the 15th day of a month of diagnosis,
or, if the month of diagnosis and exit were the same, the
day in the middle between the 1st day and the day of
exit. As the R program had been designed for exact
observations, in its grouped data application, following
the traditional life table practice, all the deaths and censoring events were placed in mid-points of the respective

follow-up intervals. We slightly modied the variance of

the actuarial estimator of the Pohar-Perme method in
STATA program to obtain better approximation for
the weighted person-time at risk. Implementations of
these dierent approaches in R and STATA are presented in the Supplementary Web Appendix.

3. Results
Colon cancer in males is shown as an example on the
comparisons (Fig. 1). The results depend quite a lot on
the choice of the method, level of grouping of the data
and on patterns of censoring. With annually grouped
follow-up times, the Pohar-Perme and the Ederer II
methods tend to give much higher values in R, particularly when the data are censored (closing year 1995). The
actuarial approach in STATA provides estimates that
are much closer to those based on exact follow-up times.
Age-standardisation does not remove these dierences
although it brings the Ederer II and the Pohar-Perme
estimates closer to each other when the data are censored. In incomplete follow-up with exact observations,
the Pohar-Perme method tends to underestimate longterm net survival, whereas the estimates of Ederer II
method are closer to the gold standard. The site-specic
results of males are summarised in Supplementary
Figs. 13.
Averaged over the sites, the gold standards of net survival for males were 49.9%, 42.6% and 38.9% for the 5-,
10- and 14-year follow-up, respectively. Annually
grouped data in R caused a marked overestimation, particularly when the data were censored and the follow-up
was long (Table 2). At 10 and 14 years, the average overestimations in the most heavily censored situation were
3.4% and 6.0% units, respectively, in the Pohar-Perme
estimates and 3.3% and 5.3% units, respectively, in the
Ederer II estimates. Even with no censoring, the average
overestimations were 1.5% and 2.5% units in the PoharPerme estimates and 1.8% and 2.8% units in the Ederer
II estimates. The analyses based on the actuarial
approach in STATA virtually removed the large dierences to the gold standard observed in R with annually
grouped data. The use of monthly grouped data mostly
reproduced the same average dierences as the exact
data.
With censored observations, the Pohar-Perme
method using exact follow-up times tended to underestimate net survival at 10 and 14 years. With the heaviest
censored data (closing year 1995), the underestimation
was 2.0% units at 14 years (Table 2). The Ederer II
method based on exact follow-up times did not give
the same negative dierences to the gold standard as
the Pohar-Perme method, particularly when the follow-up was long and the data heavily censored. But
when there was no censoring, this approach

Please cite this article in press as: Seppa K. et al., Choosing the net survival method for cancer survival estimation, Eur J Cancer (2013), http://
dx.doi.org/10.1016/j.ejca.2013.09.019

K. Seppa et al. / European Journal of Cancer xxx (2013) xxxxxx

Closing year 1995

Age-standardised
0.70
0.60

0.65

Anl, R (EdII)
Anl, STATA (EdII)
Exact, R (EdII)
Anl, R (PP)
Anl, STATA (PP)
Exact, R (PP)

0.55

0.60

0.35

0.35

0.40

0.45

0.45

0.50

0.50

0.55

Anl, R (EdII)
Anl, STATA (EdII)
Exact, R (EdII)
Anl, R (PP)
Anl, STATA (PP)
Exact, R (PP)

0.40

Cumulative net survival

0.65

0.70

Non-standardised

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Years from diagnosis

Years from diagnosis

Closing year 2010

Anl, R (EdII)
Anl, STATA (EdII)
Exact, R (EdII)
Anl, R (PP)
Anl, STATA (PP)
Exact, R (PP)

0.55

0.60

0.65

0.70

Age-standardised

0.40

0.45

0.45

0.50

0.50

0.55

0.60

Anl, R (EdII)
Anl, STATA (EdII)
Exact, R (EdII)
Anl, R (PP)
Anl, STATA (PP)
Exact, R (PP)

0.35

0.35

0.40

Cumulative net survival

0.65

0.70

Non-standardised

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Years from diagnosis

Years from diagnosis

Fig. 1. Cumulative net survival curves of male colon-cancer patients diagnosed in Finland 19811995 and followed-up until the end of 1995 and
until the end of 2010 by using the Pohar-Perme (solid lines) and the Ederer II method (dashed lines), two dierent levels of grouping the data
(annually grouped (Anl) and exact follow-up times (Exact)) and the programs in R and STATA. In R, all the events have been placed in the midpoints of the follow-up intervals. Both non-standardised and internally age-standardised curves are shown.

overestimated the gold standard on average with 1.0%

and 1.8% units at 10 and 14 years of follow-up,
respectively.
Lengths of condence intervals (CIs) of the gold standard of net survival for males were, on average, 4.1%,
5.7% and 8.1% units at 5, 10 and 14 years, respectively.
With censored data, the Pohar-Perme method tended to
give longer condence intervals than the gold standard

(Table 2). The Ederer II method did so only with the

heaviest censored data, whereas otherwise the condence intervals were shorter than those of the gold standard. Within data of the same closing year, the average
lengths of the condence intervals of the Pohar-Perme
method were 58%, 2839% and 6675% longer at 5,
10 and 14 years, respectively, than those of the Ederer
II method.

Please cite this article in press as: Seppa K. et al., Choosing the net survival method for cancer survival estimation, Eur J Cancer (2013), http://
dx.doi.org/10.1016/j.ejca.2013.09.019

K. Seppa et al. / European Journal of Cancer xxx (2013) xxxxxx

Table 2
Dierences (in % units, average of 20 sitesa) to the point estimate and the length of condence interval of gold standard of net survival (Pohar
Perme, exact follow-up times for closing year 2010) at 5, 10 and 14 years by method, program and level of data grouping, for male cancer patients
diagnosed in Finland in 19811995 and followed-up until the end of three dierent closing years.
Follow-up time
(years)

Closing
year

Ederer II

Pohar Perme

Annualb
R

Monthly
STATA

Dierence to the point estimate of gold standard

5
1995
1.93
1999
0.95
2010
0.74

(49.88%, 42.60% and

0.20
0.15
0.26
0.45
0.23
0.40

STATA

Exact

Annual

Monthly
STATA

38.91% at 5, 10 and 14 years, respectively)

0.00
0.02
1.82
0.36
0.42
0.39
0.76
0.03
0.40
0.37
0.51
0.02

Exact
STATA

0.14
0.11
0.05

0.32
0.06
0.02

0.32
0.04
0

10

1995
1999
2010

3.27
2.49
1.77

0.37
0.42
0.60

0.35
0.97
1.11

0.10
0.86
1.07

0.08
0.82
1.04

3.41
2.36
1.45

0.95
0.35
0.15

0.51
0.06
0.13

0.85
0.27
0.01

0.86
0.29
0

14

1995
1999
2010

5.33
4.14
2.80

0.24
0.92
1.21

0.68
1.75
1.94

0.27
1.55
1.89

0.19
1.49
1.84

6.02
4.35
2.48

1.77
0.35
0.15

1.24
0.01
0.22

1.91
0.36
0.03

2.02
0.42
0

respectively)
0.45
0.49
0.01
0.03
0.03
0.01

0.48
0.02
0

Dierence to the length of condence

5
1995
1999
2010

interval of gold standard (4.11%,

0.16
0.17
0.15
0.54
0.24
0.27
0.54
0.26
0.28

5.66% and 8.09% units

0.18
0.18
0.23
0.24
0.25
0.25

at 5, 10 and 14 years,
0.13
0.42
0.29
0.03
0.29
0.05

10

1995
1999
2010

0.07
1.10
1.47

0.06
0.93
1.34

0.10
0.90
1.31

0.13
0.88
1.29

0.12
0.89
1.30

2.13
0.63
0.12

2.13
0.67
0.12

2.26
0.80
0.01

2.36
0.85
0.02

2.27
0.82
0

14

1995
1999
2010

0.54
2.14
3.41

0.90
2.07
3.37

1.11
1.97
3.31

0.54
2.64
4.02

1.18
1.96
3.30

7.00
2.53
0.12

6.83
1.95
0.28

7.49
2.38
0.01

7.96
2.46
0.07

7.60
2.32
0

Cancers of the liver and gallbladder not included as no results were estimable for them at 14-year follow-up with closing date at the end of 1995.
Levels of data grouping: annually grouped, monthly grouped and exact follow-up times. In R, all the events (deaths and censorings) were placed
in mid-points of the respective follow-up intervals.
b

The results obtained for females were quite comparable with those obtained for males (available from the
rst author on request).

4. Discussion
The recommendation [7] to use the Pohar-Perme
method [4] is based on the fact that, unlike the traditional relative survival methods, it gives unbiased estimates. This, however, holds true provided that the
follow-up times are recorded accurately and used as
such and when there is no informative censoring of the
observed survival. The former condition cannot always
be met in practical applications due to, e.g. non-availability or condentiality of the data whereas the latter
condition can be guaranteed only with a complete follow-up. Fortunately, with the Finnish Cancer Registrys
data, both of these two conditions can be met, and thus
it is possible to study the importance of these conditions
when they are not met in practice.
The site-specic gold standards were prone to random variation. Therefore, it was more dicult to assess
the magnitude of bias by site. Averaging the net survival
estimates over the sites retains the unbiasedness of the

gold standard and gives case-mix (site) adjusted comparisons, in which each site has the same weight.
As the computer program [10] in R requires accurate
follow-up times it was necessary to decide how to produce data grouped into follow-up intervals by year or
month of follow-up. The old actuarial choice was to
place all the events in the middle of the interval. With
annually grouped data, this approach proved to be
clearly unacceptable. There were many ties between
observed and censored survival times, and patients
whose survival times were censored at the mid-point of
the interval were assumed to remain at risk of dying at
the mid-point according to the practice suggested originally by Breslow [12] for handling tied survival observations in the Cox proportional hazards analyses.
Exact dates of deaths and diagnosis may not be available or accessible (e.g. due to condentiality and data
protection regulations). On the other hand, the closing
date of the study is known allowing more accurate follow-up times for censored patients. In an alternative
analysis, ties were removed by using exact follow-up
times for patients alive at the end of follow-up. The estimates of this alternative were very close to those based
on the exact data in the heaviest censored situation.
However, the corresponding estimates were deviating,

Please cite this article in press as: Seppa K. et al., Choosing the net survival method for cancer survival estimation, Eur J Cancer (2013), http://
dx.doi.org/10.1016/j.ejca.2013.09.019

K. Seppa et al. / European Journal of Cancer xxx (2013) xxxxxx

when there was no censoring. It seemed that the R program worked better, when event times were more heterogeneous and not centred at the mid-points of follow-up
intervals. A feasible solution in R might be to draw
event times of each interval from a uniform distribution.
In a routine use of the net survival method, however,
these kinds of tricks are not really applicable.
Censoring of the observed survival was informative,
because patients were diagnosed over a long calendar
period during which the distributions of covariates that
aect survival (e.g. age at diagnosis) have changed. This
emerges, e.g. in ageing populations, when the mean age
of diagnosed patients increases over the period of diagnosis, and therefore, older patients have on average
shorter times from their diagnosis to the end of the
study. The impact of this type of informative censoring
on various net survival methods has been studied using
simulated accurate follow-up data under various scenarios [13]. Recently, Rebolj Kodre and Pohar Perme proposed a method of inverse probability weighting that
allows this type of informative censoring [14]. The
method requires estimating probabilities of censoring
times and was not used in our study, as it is not available
in the current computer programs. Moreover, the Ederer II method was not considered in that paper.
A second reason for informative censoring was that
the prognosis of the patients changed over the period
of diagnosis: patients whose follow-up times were censored earlier had often better prognosis than earlier
diagnosed patients who remained under follow-up. This
type of informative censoring cannot be corrected for
without extrapolation of survival beyond the closing
date, and therefore, it is not a problem of the methods,
as any corrections are subject to pure guessing [14]. The
dierent sources of bias can be controlled for in simulation-based studies. In empirical data, the second type of
informative censoring cannot be eliminated without
eliminating the rst type of informative censoring, too.
In this study, results are based on real data with real
progressive censoring owing to early closing dates of follow-up. These real patterns of censoring may well be different in dierent countries, but a good net survival
method should be resistant against biases any pattern
of informative censoring might cause. As opposed to
simulated data, however, the targeted gold standard
under complete follow-up and accurate follow-up times
is still a random quantity with a standard error.
When the data are censored, the Ederer II method
could be preferred as it gives results closer to the gold
standard. This may, however, be a characteristic due
to the particular censoring pattern in the Finnish data,
as the positive bias in the Ederer II method was compensated by the negative bias due to informative censoring
of the observed survival.
In the setting of cause-specic survival, deaths due to
other causes than cancer are considered as censoring

events. The KaplanMeier estimator is biased under

informative censoring but can be corrected for by following the idea of inverse probability weighting [15] that
was adapted to the framework of relative survival by
Pohar Perme et al. [4]. Of course, informative censoring
caused by changes in patients prognosis cannot be corrected for in cause-specic survival, either. In cause-specic survival, cause of death is not always correct,
whereas, in the framework of relative survival, the
expected survival estimated from the mortality rates of
national population may not always be relevant for
the patients [8]. This is the main reason for dierences
between results of the two approaches which both aim
to estimate net survival.
In the Pohar-Perme method the few observations in
the old age groups get large weights, because the competing risks of death do not leave for older ages sufciently sizable materials on which to base reliable
estimation [16]. This can be seen in the standard
errors and the condence intervals based on them.
The Ederer II method gives estimates that have distinctively narrower condence intervals than those
derived by the Pohar-Perme method. It is likely that
in this respect the gold standard is far from a true
gold standard.
The age-standardisation is not a solution for removing biases related to the level of grouping and informative censoring or inaccuracies related to interval
estimation. Statistical modelling [1719] is capable of
nding the essence also in net survival analyses and
should be developed into a standard for routine use on
a large scale. Otherwise, it is crucial to report in which
way the net survival results have been obtained.
Net survival is especially useful for evaluating dierences in cancer survival between population groups and
over time, when the expected mortality diers across the
groups we wish to compare. Thus, in future studies, it
would be important to assess whether the choice of the
approach could actually aect results of comparisons
between population groups.
A clear recommendation to use the net survival
method by Pohar Perme et al. is conditional on the completeness of follow-up and the time point of follow-up at
which the net survival is wished to be estimated. Both
methods perform well in the estimation of net survival
until 5 years. In complete follow-up, the Pohar-Perme
estimator can be preferred in terms of bias but point estimates of the long-term net survival should be interpreted
with due caution, because the estimator becomes prone to
random variation. In incomplete follow-up, the estimator
of the Pohar-Perme method may be biased if censoring of
the observed survival is informative, even if the recently
developed weighting method [14] was used. Irrespective
of the method, the actuarial approach in STATA should
be utilised, if data are grouped into annual follow-up
intervals. Following this recommendation, the results

Please cite this article in press as: Seppa K. et al., Choosing the net survival method for cancer survival estimation, Eur J Cancer (2013), http://
dx.doi.org/10.1016/j.ejca.2013.09.019

K. Seppa et al. / European Journal of Cancer xxx (2013) xxxxxx

given by dierent approaches dier on average by 12%

units depending on the context.
Conict of interest statement
None declared.
Acknowledgement
This work was supported by a grant from the Finnish
Cancer Foundation.
Appendix A. Supplementary data
Supplementary data associated with this article can
be found, in the online version, at http://dx.doi.org/
10.1016/j.ejca.2013.09.019.
References
[1] Ries LAG, Melbert D, Krapcho M, et al., editors. SEER cancer
statistics review, 19752004. National Cancer Institute: Bethesda,
MD; 2007.
[2] Coleman MP, Quaresma M, Berrino F, et al. Cancer survival in
ve continents: a worldwide population-based study (CONCORD). Lancet Oncol 2008;9:73056.
[3] Hakulinen T, Seppa K, Lambert PC. Choosing the relative
survival method for cancer survival estimation. Eur J Cancer
2011;47:220210.
[4] Pohar Perme M, Stare J, Este`ve J. On estimation in relative
survival. Biometrics 2012;68:11320.
[5] Hakulinen T. Cancer survival corrected for heterogeneity in
patient withdrawal. Biometrics 1982;38:93342.
[6] Ederer F, Heise H. Instructions to IBM 650 programmers in
processing survival computations. Methodological note no. 10.
Bethesda, MD: End Results Evaluation Section, National Cancer
Institute; 1959.

[7] Roche L, Danieli C, Belot A, et al. Cancer net survival on registry

data: use of the new unbiased Pohar-Perme estimator and
magnitude of the bias with the classical methods. Int J Cancer
2013;132:235969.
[8] Dickman PW, Adami HO. Interpreting trends in cancer patient
survival. J Intern Med 2006;260:10317.
[9] Pokhrel A, Hakulinen T. How to interpret the relative survival
ratios of cancer patients. Eur J Cancer 2008;44:26617.
[10] Pohar Perme M. relsurv: Relative survival. R package version 2.04; 2013. Available from: http://CRAN.R-project.org/package=relsurv. [Accessed on 26 June 2013].
[11] Dickman PW, Coviello E, Hills M. STATA computer program
strs.ado, version 1.3.8 (29 March 2013); 2013. Available from:
http://www.pauldickman.com/rsmodel/stata_colon. [Accessed on
26 June 2013].
[12] Breslow NE. Discussion on the paper by D.R. Cox. J R Stat Soc
Ser B 1972;34:2167.
[13] Danieli C, Remontet L, Bossard N, et al. Estimating net survival;
the importance of allowing for informative censoring. Stat Med
2012;31:7586.
[14] Rebolj Kodre A, Pohar Perme M. Informative censoring in
relative survival. Stat Med 2013. http://dx.doi.org/10.1002/
sim.5877.
[15] Robins JM. Information recovery and bias adjustment in
proportional hazards regression analysis of randomized trials
using surrogate markers. In: Proceedings of the biopharmaceutical section. San Francisco, CA: American Statistical Association; 1993. p. 2433.
[16] Hakulinen T. On long-term relative survival rates. J Chron Dis
1977;30:43143.
[17] Dickman PW, Sloggett A, Hills M, Hakulinen T. Regression
models for relative survival. Stat Med 2004;23:5164.
[18] Nelson CP, Lambert PC, Squire IB, Jones DR. Flexible
parametric models for relative survival with application in
coronary heart disease. Stat Med 2007;26:548698.
[19] Remontet L, Bossard N, Belot A, Este`ve J, FRANCIM. An
overall strategy based on regression models to estimate relative
survival and model the eects of prognostic factors in cancer
survival studies. Stat Med 2007;26:221428.

Please cite this article in press as: Seppa K. et al., Choosing the net survival method for cancer survival estimation, Eur J Cancer (2013), http://
dx.doi.org/10.1016/j.ejca.2013.09.019