European

Journal of
Cancer
European Journal of Cancer 40 (2004) 1361–1372
www.ejconline.com

Up-to-date long-term survival of cancer patients: an evaluation

of period analysis on Swedish Cancer Registry data
ack *, Magnus Stenbeck, M

Mats Talb€ ans Rosen
The Swedish Cancer Registry, Centre for Epidemiology, The National Board of Health and Welfare, SE-106 30 Stockholm, Sweden

Received 6 November 2003; received in revised form 28 January 2004; accepted 3 February 2004
Available online 7 May 2004

Abstract

The natural development of cancers as well as the measures to fight the disease are often long processes that require decades of
follow up. Available information on long-term survival will thus often appear outdated and irrelevant. A few years ago, period-
survival analysis was proposed as a means to obtain more up-to-date information on long-term cancer survival.
This article assesses period and conventional cohort-based survival analyses on their ability to predict future survival. Based on
historical data from the nationwide Swedish Cancer Registry 5-, 10- and 15-year relative survival actually observed for patients
diagnosed at one particular point in time are compared to the most recent period and cohort-based survival estimates available at
that point in time. The study shows that period analysis can, in most cases, be used to provide more up-to-date long-term estimates
of cancer survival. Period analysis reduces the time lag of the survival estimates by some 5–10 years for all cancers combined and
especially affects the survival estimates for small intestine carcinoids, meningioma and intracranial neurinoma of the brain, non-
seminoma testicular cancer, chronic lymphocytic leukaemia and Hodgkin’s lymphoma.
Ó 2004 Elsevier Ltd. All rights reserved.

Keywords: Cancer registries; Statistical methods; Survival analysis; Period analysis; Prognosis

1. Introduction often pertain to clinical methods no longer in use. This

time lag may be reduced with period survival analysis,
Up-to-date information on cancer survival is impor- which was introduced into cancer research some 6 years
tant as reference material for clinicians, oncologists and ago [1,2], since it aims to provide more up-to-date esti-
scientists involved in clinical work, medical auditing or mates of survival than can be derived from conventional
research. It provides a basis for evaluation of decision- cohort-based analysis.
making, and is used in debate on the effects of cancer Empirical evaluation of period survival analysis on
treatment and prevention. Long-term follow up periods large data has so far been mainly performed on material
are generally needed to evaluate changes in cancer sur- from the Finnish Cancer Registry [3–6]. These analyses
vival since the natural development of cancers, as well as show that period analysis in general provides more up-to-
the measures to fight the disease, often are long pro- date estimates of long-term cancer survival than cohort-
cesses that sometimes require decades of follow up. based survival estimates. They also suggest that period
However, survival estimates of long-term follow up may estimates from a given time period in most cases quite
often appear irrelevant since the time lag between di- accurately predict the long-term survival for patients di-
agnosis and evaluation is considered too long, and they agnosed during that particular period. The period ap-
proach has been used to estimate long-term survival for
patients diagnosed in recent years (e.g. [7–10]).
*
Corresponding author. Tel.: +46-8-555-530-00; fax: +46-8-555-533- The aim of the present study is to provide an em-
27. pirical evaluation of period survival analysis on Swedish
E-mail address: [email protected] (M. Talb€ack). Cancer Registry data. If this proves successful, the

0959-8049/$ - see front matter Ó 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ejca.2004.02.004
1362 M. Talb€ack et al. / European Journal of Cancer 40 (2004) 1361–1372

ultimate goal is to incorporate period survival estimates cohort-based analysis [13]. The Hakulinen method [14]
in the regular statistical publications from the Cancer adjusting for potential heterogeneous follow up time is
Registry. implemented in the macro and was used for the cumu-
lative relative survival rate (RSR) estimates. Prior to
analysis, the macro was updated to facilitate the use of
2. Materials and methods annual population survival probabilities. The results
were compared to SURV3 developed at the Finnish
2.1. The Swedish Cancer Registry Cancer Registry [15,16]. Cohort-based RSRs can be
estimated directly by SURV3 and the period estimates
Since 1958 every clinician, pathologist and cytolo- can be calculated with partial result from SURV3 and
gist in Sweden has been required by law to notify the an extension of the modified life-table approach de-
Cancer Registry at the National Board of Health and scribed by Brenner and Gefeller [1]. The updated SAS
Welfare of each new cancer diagnosed. The non-re- macro and SURV3 yield equivalent results (data not
porting rate has been estimated at less than 2% [11]. shown). Relative survival is defined as the observed
The Swedish Cancer Registry is population based and survival among the cancer patients divided by the ex-
today covers 8.9 million people. For the years 1958– pected survival for a comparable group from the general
1998 the register has accumulated 1.6 million tumours population with respect to the main factors affecting
for 1.4 million persons. survival, in this case sex, age and calendar year.
For all sites combined and for each of the 40 different
2.2. Data material forms of cancer analysed the 5-, 10-, and 15-year RSRs
actually observed for different cohorts of patients were
The study was based on all cancer cases reported calculated. These observed RSRs were then compared
between 1960 and 1998. Forty different forms of with the most up-to-date survival estimates available at
cancer and all sites combined were analysed. This that particular cohort’s time of diagnosis, using two
selection is the same data that have previously been conventional cohort-based methods for survival esti-
used in a comprehensive overview of the development mates, denoted ‘cohort’ and ‘complete’ analysis [2], as
of cancer survival in Sweden during the past four well as for period analysis. ‘Cohort analysis’ evaluates
decades, amended with one additional year of follow survival for cohorts of patients diagnosed at close
up [12]. Excluded from the analyses were multiple proximity in time to each other (e.g. within the same
tumours at the same anatomical site, autopsy findings, calendar year). Hence, all patients have the potential to
cases with zero survival time, and patients who were be under observation for the entire period of follow up.
90 years or older at diagnosis. In the definition of a ‘Complete analysis’ additionally includes in the cohort
few sites, some histopathological groups were excluded patients diagnosed in later years, providing a mixture of
from the analyses due to low incidence and/or due to patients with short and long potential follow up time.
survival probabilities that differ from the dominating This is illustrated in Fig. 1 for the year 1996, which is the
pattern for that particular site. For small intestine, last available period (dashed frame) at the time of
testis and brain and nervous system, different histo- analysis for which it is also possible to calculate the
pathological groups within the same site were analysed actual observed 5-year RSR (solid thick frame). Fig. 1
separately. Basal cell carcinoma has not been a part of also depicts the corresponding cohort (solid thin frame)
the cancer registration historically, and is not included and complete analyses (squares frame) that were avail-
in the non-melanoma skin cancer group. able in 1996.
A total of 512,133 men and 509,288 women were in- The RSRs were calculated for 1-year intervals, and
cluded in the analysis. The Cancer Register is linked an- for 3- and 5-year moving averages. Single-year estimates
nually by personal identification numbers to the Cause of were used for common cancers. For less common sites,
Death Register at the National Board of Health and 3- or 5-year moving averages were used, making a nec-
Welfare and to the Migration and Population Registries essary trade off between up-to-date information and
at Statistics Sweden, to obtain dates of death or censoring precision. The analysis was done for males and females
and to confirm continued residency in Sweden. At the separately as well as for both sexes combined and for
time of analysis the follow up was completed up to and different age groups. This article reports only results
including 31 December 2001. Complete follow up was for males and females combined, with the exception for
available for 99.9% of the recorded cases. breast, prostate, testis and gynaecological cancers, and
for the age group 0–89 years at diagnosis.
2.3. Statistical analysis In period analysis, follow up is restricted to a narrow
calendar time period (e.g. one calendar year). The esti-
The analyses were performed with a publicly avail- mates are obtained by left truncation of the data at the
able SAS macro that can be used for both period and beginning of the period and by right censoring at its end.
 M. Talb€ack et al. / European Journal of Cancer 40 (2004) 1361–1372 1363

Year of follow up
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001

1986 4/5 5/6 6/7 7/8 8/9 9/10 10/11 11/12 12/13 13/14 14/15 15/16
1987 3/4 4/5 5/6 6/7 7/8 8/9 9/10 10/11 11/12 12/13 13/14 14/15
1988 2/3 3/4 4/5 5/6 6/7 7/8 8/9 9/10 10/11 11/12 12/13 13/14
Year of diagnosis

1989 1/2 2/3 3/4 4/5 5/6 6/7 7/8 8/9 9/10 10/11 11/12 12/13
1990 1 1/2 2/3 3/4 4/5 5/6 6/7 7/8 8/9 9/10 10/11 11/12
1991 1 1/2 2/3 3/4 4/5 5/6 6/7 7/8 8/9 9/10 10/11
1992 1 1/2 2/3 3/4 4/5 5/6 6/7 7/8 8/9 9/10
1993 1 1/2 2/3 3/4 4/5 5/6 6/7 7/8 8/9
1994 1 1/2 2/3 3/4 4/5 5/6 6/7 7/8
1995 1 1/2 2/3 3/4 4/5 5/6 6/7
1996 1 1/2 2/3 3/4 4/5 5/6
1997 1 1/2 2/3 3/4 4/5
1998 1 1/2 2/3 3/4

Fig. 1. Data included to calculate observed 5-year survival of patients diagnosed 1996 (solid thick frame), and the corresponding most up-to-date
estimates that could have been obtained in 1996 by cohort (solid thin frame), complete (squares frame), and period analysis (dashed frame). Numbers
within cells indicate years of follow up since diagnosis.

Only those at risk and events (death or censoring) oc- 110

Interval-specific RSR

100
curring during this particular period are considered, and 90
survival experiences during other years are discarded. 80
Hence, the period estimates reflect only survival expe- 70
60
rienced during the chosen calendar period. For the ex-
50
ample in Fig. 1, the survival probability for the first year 40
1964-1966

of follow up is solely estimated from patients diagnosed 30

1974-1976
1984-1986
in 1995–1996, and conditional probabilities for the sec- 20
1994-1996
10
ond, third, fourth and fifth year of follow up are esti-
0
mated by patients diagnosed in 1994–1995, 1993–1994, 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1992–1993 and 1991–1992, respectively (dashed frame). Years since diagnosis
The interval-specific survival probabilities are then
Fig. 2. Graph illustrating the way the interval-specific relative survival
multiplied to obtain cumulative survival probabilities. rates (RSR) commonly shift upward for more recent years of diagnosis
The alternatives to period analysis would have been to during the first years of follow up, after which the difference between
calculate the survival for the last available cohort with 5 the survival curves decrease. Observed RSR of the rectum. Males and
years of follow up, patients diagnosed in 1991 (solid thin females, 0–89 years of age at diagnosis (site chosen for illustrative
purposes).
frame), or to use the complete available information
from patients diagnosed in 1991–1996 (squares frame).
If the actual 1991 cohort is not the particular focus of a shift upward for more recent years of diagnosis during
the investigation, the complete approach seems to be the the first years of follow up after which the difference
best choice since it utilises the available data in a more between the survival curves decreases and, in this ex-
efficient way and provides more updated information. ample, becomes negligible after 5 years.
Most of the increased mortality experienced by can- In order to evaluate the period analysis approach
cer patients occurs during the first few years after di- compared to the cohort-based methods the mean dif-
agnosis, on which the cumulative RSRs to a large extent ference and the mean squared difference between the
depend. Since the time lag for period analysis for the observed RSR and the latest cohort, complete and pe-
first years after diagnosis is short, and the later years of riod RSRs available at the time of diagnosis were cal-
follow up have a more limited impact on the shape of culated. The mean difference is a measure of systematic
the cumulative survival curve, the period analysis will over- or underestimation of the RSRs, and the mean
respond more quickly to changes in survival. Fig. 2 il- squared difference quantifies the degree of deviation and
lustrates the common shape for interval-specific curves: is determined both by systematic and random variability
1364 M. Talb€ack et al. / European Journal of Cancer 40 (2004) 1361–1372

in the RSRs. This allows us to address the question of to creases with increasing length of follow up and the
what extent the relative survival observed for patients estimated RSRs are in most cases an underestimation of
diagnosed at one particular point in time could have the future observed RSR. The latter indicates that for
been predicted by period analysis and by the cohort and many forms of cancer there have been continuing im-
complete methods at the time of diagnosis. The mean provements in survival during the past four decades.
squared difference is not a formal test statistic and For the 5-year RSRs (Table 1) the advantage of the
should only be used to compare the relative size of the period approach compared to the cohort and complete
differences for the same form of cancer. analyses in early detection of the future survival, ex-
pressed as the mean difference in percent units to the
observed RSR, is most noticeable for small intestine
3. Results carcinoids ()4.5, )3.0, )0.2), meningioma and intra-
cranial nerve neurinoma of the brain ()6.2, )4.1, )0.3
An evaluation of the cohort, complete and period and )4.4, )2.7, )0.5, respectively), non-seminoma tes-
RSRs available at a given point in time compared to the ticular cancer ()8.4, )5.1, )1.8), chronic lymphocytic
RSRs observed for patients diagnosed at this particular leukaemia ()5.7, )4.0, )1.9), and Hodgkin’s lymphoma
point in time are shown in Tables 1–3 for a follow up of ()8.0, )5.6, )2.1), for the cohort, complete and period
5, 10 and 15 years, respectively. RSRs, respectively. The deviation between the period
The period width at which each form of cancer has and observed RSRs, as measured by the mean squared
been evaluated was chosen by considering the number at difference, are for these same cancers only some 6–30%
risk together with a visual inspection of the stability of and 20–65% of the deviation for the cohort and com-
the respective 5-, 10-, and 15-year RSR curves. For each plete RSRs, respectively. This indicates a considerably
individual form of cancer a common width was chosen closer relation between the observed and period RSRs
for all lengths of follow up. All sites combined, as well as than was achieved by conventional cohort-based sur-
breast and prostate cancer, were large and stable enough vival analysis. The same pattern holds also over 10 and
to be evaluated using single-year periods. A period width 15 years of follow up for the cancer sites exhibiting
of 3 years (3-year moving averages) was chosen for e.g. survival good enough to have stable estimates for that
colon, rectum, lung, cervix uteri, kidney, skin and non- length of time (Tables 2 and 3).
Hodgkin’s lymphoma. Less common cancers were According to the cohort, complete and period RSRs
evaluated at a period width of 5 years (5-year moving that are comparable in time to the last available
averages), e.g. lip, small intestine, primary liver, testis, observed RSRs (Tables 1–3), some forms of cancer had
eye, bone, and acute lymphocytic and acute myeloid a continuing improvement in survival, whereas others
leukaemia. All the RSR curves underlying the results in appear to have levelled off. Others still, such as
Table 1–3 appear stable in terms of percentage units. In oesophagus, primary liver, gallbladder, pancreas and
some cases, e.g. primary liver cancer, this is more due to lung, ought to be evaluated at a shorter follow up in-
very low survival probabilities than to the number at risk. terval since only a small fraction of the patients survive
The highest mean annual increases in 5-year RSR for 5 years. For the same reason these cancers also ex-
during the past decades were for acute myeloid and hibit the smallest mean and mean squared difference
lymphocytic leukaemia, 7.2% and 4.3%, respectively, between the estimated and observed RSRs.
primary liver (4.7%), chronic myeloid leukaemia (3.5%), A good example of a cancer site where period analysis
gallbladder with biliary tract (3.3%) and non-seminoma would have detected early the exceptional improvement
testicular cancer (3.0%). The same cancers were also in long-term survival during the recent decades is non-
amongst those with the highest mean annual increase in seminoma testicular cancer. The variability, mean
10- and 15-year RSR, and amongst those with the squared difference, between the period and observed
largest percent unit increase over the decades: non- RSRs is only some 4–8% of the variability between the
seminoma testicular cancer (48.5, 36.4 and 24.6), observed and cohort curves, indicating a considerably
Hodgkin’s lymphoma (45.1, 30.9 and 21.9), acute lym- closer relation between the observed and period RSRs.
phocytic leukaemia (43.2, 37.4 and 17.0), meningioma of Despite this close relation the period analysis underes-
the brain (36.3, 23.2 and 10.8), and chronic lymphocytic timates the observed RSR by on average some 2%, 3%,
leukaemia (35.0, 18.6 and 11.3), for 5-, 10-, and 15-year and 6% units for a follow up of 5, 10 and 15 years, re-
RSRs (Tables 1–3). spectively. The last available estimate shows that the
In general, the period RSR is a better estimate of the 5-year RSR had levelled of at about 94% in the late
ultimately observed RSR than the cohort and complete 1980s (Table 1). For 10-, and 15-year RSR the equal
estimates. The variability between the period and ob- period and observed RSRs of 94% and 89%, respec-
served RSRs is in most cases much smaller than for the tively, are considerably higher than the cohort and
corresponding cohort and complete estimates. The dif- complete RSRs of 73% and 86%, and 47% and 68%,
ference between the estimated and observed RSRs in- respectively (Tables 2 and 3).
Table 1
Five-year relative survival rate (RSR) for patients diagnosed between 1965–1996 at 0–89 years of age
Site Sex Period Mean First Mean Mean difference Mean squared difference Last available comparable RSRs
width in number available annual
Cohort Complete Period Cohort Complete Period Cohort Complete Period Observed
years at risk RSR change
All sites combined M&F 1 13415 40.3 1.4 )3.1 )2.6 )0.7 11.0 7.1 0.8 57.6 58.2 59.6 61.0
Lip M&F 5 711 95.5 )0.2 0.6 0.5 0.7 5.8 3.6 4.1 94.2 93.1 92.8 88.9
Oral cavity and M&F 5 613 46.7 0.0 )0.5 )0.5 )0.3 10.0 5.5 3.8 45.4 47.5 49.8 52.1

M. Talb€ack et al. / European Journal of Cancer 40 (2004) 1361–1372

mesopharynx
Oesophagus M&F 5 89 4.4 2.4 )0.6 )0.2 )0.2 1.8 0.6 0.3 8.5 9.8 10.3 10.5
Stomach M&F 3 652 12.4 2.0 )1.1 )0.7 )0.2 2.7 1.2 0.5 19.2 19.2 19.1 20.4
Small intestine, M&F 5 50 22.3 1.3 )1.1 )0.5 0.3 23.9 11.8 11.9 30.3 27.6 26.6 22.2
adenocarcinoma
Small intestine, M&F 5 192 43.9 1.8 )4.5 )3.0 )0.2 48.4 21.4 14.0 63.4 66.1 68.7 71.6
carcinoids
Colon, adenocarcinoma M&F 3 2826 40.6 1.3 )2.6 )2.0 )0.5 11.2 5.2 0.7 53.6 54.9 56.6 56.8
Rectum, adenocarcinoma M&F 3 1644 38.9 1.5 )3.1 )2.4 )1.0 13.0 7.8 2.8 50.6 52.6 55.1 56.7
Liver, primary M&F 5 38 1.5 4.7 )0.7 )0.3 )0.1 1.0 0.3 0.2 4.4 5.3 6.1 6.2
Gall bladder, biliary tract M&F 5 142 4.0 3.3 )0.9 )0.4 )0.1 1.7 0.6 0.4 8.7 8.8 9.6 9.1
Pancreas M&F 5 86 2.6 )1.2 0.3 0.3 0.2 0.3 0.2 0.1 2.3 2.2 2.0 1.7
Nose and nasal sinuses M&F 5 142 45.5 1.2 )1.8 )1.1 )0.1 34.0 15.3 11.5 57.3 58.1 57.8 55.9
Larynx M&F 5 596 71.5 )0.3 0.1 )0.1 0.1 9.3 5.2 4.7 68.8 68.3 69.4 68.3
Lung M&F 3 565 8.1 1.3 )0.5 )0.2 )0.0 1.2 0.4 0.3 9.7 11.0 11.9 12.9
Breast Females 1 2890 63.1 0.9 )3.2 )2.5 )1.2 16.4 9.8 4.2 83.3 83.1 83.6 84.7
Cervix uteri Females 3 1176 67.8 0.2 )0.8 )0.8 )0.3 4.9 2.4 1.7 70.6 69.6 70.3 70.9
Corpus uteri Females 3 1994 74.8 0.3 )1.0 )0.9 )0.2 3.0 2.1 1.7 79.2 80.4 82.3 83.4
Ovary Females 3 1088 34.5 0.9 )1.3 )0.8 0.3 6.4 2.8 1.4 41.5 42.9 45.6 45.4
Prostate Males 1 1877 42.3 1.6 )4.7 )3.5 )1.5 29.5 15.9 6.0 67.9 69.2 70.2 72.4
Testis, seminoma Males 5 366 88.4 0.7 )2.3 )1.4 )0.4 15.8 7.1 3.8 96.4 97.1 97.3 97.3
Testis, non-seminoma Males 5 298 45.1 3.0 )8.4 )5.1 )1.8 121.8 44.4 9.9 94.5 93.7 93.5 93.6
Kidney excluding renal M&F 3 1026 33.7 1.6 )3.1 )2.2 )0.3 13.7 5.9 1.1 48.8 50.0 51.7 54.1
pelvis
Urinary bladder and M&F 3 2529 57.7 0.8 )2.6 )2.0 )0.6 11.1 6.3 2.4 70.5 71.0 71.9 72.0
urethra
Malignant melanoma of M&F 3 2106 71.5 0.8 )4.2 )2.8 )1.4 26.2 12.1 4.5 87.4 87.3 87.3 87.9
skin
Malignant skin cancer, M&F 3 2095 87.8 0.1 )0.3 )0.3 0.1 3.2 2.4 3.3 87.6 87.8 88.8 88.4
excl. melanoma

1365
 1366
Tabel 1 (Continued )
Site Sex Period Mean First Mean Mean difference Mean squared difference Last available comparable RSRs
width in number available annual
Cohort Complete Period Cohort Complete Period Cohort Complete Period Observed
years at risk RSR change
Eye M&F 5 332 70.0 0.3 )1.2 )1.0 )0.6 16.5 9.4 6.4 74.0 74.1 75.0 75.2
Brain, excl. cranial M&F 3 484 21.5 1.9 )2.3 )1.5 )0.6 10.1 4.1 1.4 30.7 32.0 33.1 34.3

M. Talb€ack et al. / European Journal of Cancer 40 (2004) 1361–1372

nerves, meningioma
Brain, meningioma M&F 3 454 56.9 2.0 )6.2 )4.1 )0.3 65.9 22.4 6.3 90.4 92.0 93.1 93.2
Brain, intracranial nerves M&F 5 242 72.5 1.6 )4.4 )2.7 )0.5 65.1 20.0 4.0 99.1 99.3 99.3 99.0
neurinoma
Thyroid gland M & F 3 623 64.9 0.9 )3.4 )2.6 )0.9 15.8 8.7 2.7 83.6 83.3 83.8 85.2
Endocrine glands M & F 3 1337 75.3 0.8 )3.9 )2.4 )0.4 20.9 7.9 1.2 93.2 93.7 94.5 95.1
Bone M & F 5 207 37.5 1.8 )4.8 )3.2 )1.4 43.9 21.0 14.5 63.2 64.0 65.0 65.8
Connective tissue, muscle M & F 5 549 41.8 0.7 )2.0 )1.4 )0.4 14.8 7.2 4.8 55.2 54.6 54.7 55.6
Non-Hodgkin’s M & F 3 1142 29.0 2.2 )3.7 )2.6 )1.0 24.3 10.8 3.1 52.3 53.0 54.6 54.8
lymphoma
Hodgkin’s lymphoma M&F 3 343 35.9 2.9 )8.0 )5.6 )2.1 74.4 36.4 10.1 76.9 78.8 80.8 81.0
Multiple myeloma M&F 3 422 22.0 1.2 )1.9 )1.0 )0.0 8.3 3.8 2.9 30.8 31.7 32.9 33.6
Acute lymphocytic M&F 5 221 20.2 4.3 )6.8 )5.0 )2.8 85.1 44.6 18.9 59.0 59.4 60.3 63.4
leukaemia
Chronic lymphocytic M&F 3 450 33.4 2.5 )5.7 )4.0 )1.9 43.6 21.0 7.6 63.6 65.2 68.3 68.4
leukaemia
Acute myeloid leukaemia M&F 5 92 2.4 7.2 )2.6 )1.6 )0.9 9.7 4.3 1.7 14.1 14.9 15.5 18.2
Chronic myeloid M&F 5 144 17.3 3.5 )4.9 )3.6 )2.1 47.8 27.8 17.6 36.1 40.2 43.5 50.7
leukaemia
Period width in years: 1 ¼ 32 one-year intervals between 1965 and 1996; 3 ¼ 3-year moving averages, 30 intervals between 1966 and 1995; 5 ¼ 5-year moving averages, 28 intervals between
1967 and 1994. Mean number at risk: Mean number at risk at the beginning of the fifth year after diagnosis for the 1-, 3- and 5-year periods, respectively. First available RSR: First available cohort/
observed RSR. Period width ¼ 1, diagnostic year 1960; period width ¼ 3, diagnostic year 1960–1962; period width ¼ 5, diagnostic year 1960–1964. Mean annual change: Mean annual change of
cohort/observed RSR in percent. Mean difference: Cohort ¼ mean of all (cohort ) observed); complete ¼ mean of all (complete ) observed); period ¼ mean of all (period ) observed). Mean
squared difference: Cohort ¼ mean of all ((cohort ) observed)2 ); complete ¼ mean of all ((complete ) observed)2 ); period ¼ mean of all ((period ) observed)2 ). Last available comparable RSRs:
Period width ¼ 1, Cohort (diagnostic year) ¼ 1991, Complete (diagnostic year) ¼ 1991–1996, Period (Period) ¼ 1996 and Observed (diagnostic year) ¼ 1996. Period width ¼ 3, Cohort ¼ 1989–
1991, Complete ¼ 1989–1996, Period ¼ 1994–1996 and Observed ¼ 1994–1996. Period width ¼ 5, Cohort ¼ 1987–1991, Complete ¼ 1987–1996, Period ¼ 1992–1996 and Observed ¼ 1992–
1996.
First and last available cohort/observed RSR with its mean annual change in percent. Mean difference and mean squared difference between the cohort, complete and period RSRs available at a
given time and the RSR later observed for patients diagnosed in this interval. Last available cohort, complete, period and observed RSRs that are comparable in time.
Table 2
Ten-year relative survival rate (RSR) for patients diagnosed between 1970–1991 at 0–89 years of age
Site Sex Period Mean First Mean Mean difference Mean squared difference Last available comparable RSRs
width in number available annual
Cohort Complete Period Cohort Complete Period Cohort Complete Period Observed
years at risk RSR change
All sites combined M&F 1 9053 38.2 1.4 )5.4 )3.4 )0.5 30.4 12.3 0.9 44.8 46.6 49.6 51.0
Lip M&F 5 503 90.7 )0.3 3.2 2.3 1.3 14.5 9.3 8.1 88.0 86.7 86.3 88.8

M. Talb€ack et al. / European Journal of Cancer 40 (2004) 1361–1372

Oral cavity and M&F 5 377 40.4 )0.3 3.1 1.9 1.6 24.3 12.0 10.1 34.3 35.4 37.3 37.6
mesopharynx
Oesophagus M&F 5 42 4.6 1.6 )0.5 )0.3 )0.2 0.8 0.6 0.8 4.1 4.7 4.6 6.5
Stomach M&F 3 398 10.6 2.6 )2.7 )1.7 )0.4 9.9 4.0 0.9 13.2 15.3 16.2 17.1
Small intestine, M&F 5 36 16.9 1.4 )5.9 )4.1 )2.5 80.1 48.9 36.3 30.6 26.8 27.4 25.0
adenocarcinoma
Small intestine, M&F 5 122 36.6 1.8 )6.8 )2.5 2.3 71.6 17.9 24.0 49.1 51.7 54.1 52.5
carcinoids
Colon, adenocarcinoma M&F 3 1834 37.6 1.4 )4.7 )2.6 0.1 31.1 10.1 1.0 45.6 46.7 48.7 47.7
Rectum, adenocarcinoma M&F 3 999 34.8 1.3 )4.3 )2.5 )0.2 21.4 7.1 1.2 37.4 40.5 42.9 43.7
Liver, primary M&F 5 15 1.4 5.4 )0.3 0.0 0.3 1.0 0.8 0.7 1.1 1.6 2.0 3.6
Gall bladder, biliary tract M&F 5 82 4.4 2.5 )0.7 )0.1 0.3 1.1 0.5 0.7 4.7 5.5 6.2 7.3
Pancreas M&F 5 43 1.8 0.2 0.5 0.3 0.1 0.6 0.2 0.0 1.3 1.6 1.6 1.4
Nose and nasal sinuses M&F 5 86 34.7 2.7 )5.1 )3.0 )0.9 77.6 34.7 18.5 36.0 39.3 40.8 52.2
Larynx M&F 5 422 64.8 )0.8 3.9 2.8 2.1 20.2 9.7 9.1 60.5 58.5 57.4 56.2
Lung M&F 3 322 7.1 0.1 )0.3 0.1 0.4 0.6 0.2 0.6 7.6 7.7 7.6 7.2
Breast Females 1 2042 54.1 1.5 )7.1 )5.1 )2.5 59.4 32.6 12.7 65.8 65.8 68.5 74.6
Cervix uteri Females 3 927 62.2 0.4 )1.2 )1.0 )0.5 9.7 4.8 2.5 60.0 62.7 66.2 67.4
Corpus uteri Females 3 1650 75.9 0.2 )2.6 )1.7 )0.4 11.5 6.1 2.7 79.4 78.8 79.1 75.9
Ovary Females 3 856 33.4 0.7 )2.9 )1.3 0.9 11.6 4.7 3.4 37.3 38.7 39.1 35.0
Prostate Males 1 815 30.1 1.5 )7.2 )4.4 )1.3 64.7 30.2 13.0 39.2 41.7 44.7 47.6
Testis, seminoma Males 5 327 84.2 0.8 )5.2 )3.0 )1.0 36.8 11.4 6.6 86.7 91.9 93.6 94.9
Testis, non-seminoma Males 5 286 57.3 3.4 )23.1 )12.1 )2.6 571.8 166.7 22.2 73.2 86.2 93.6 93.7
Kidney excluding renal M&F 3 699 32.0 1.2 )4.6 )2.4 0.3 22.7 7.0 1.7 36.7 38.8 40.5 42.0
pelvis
Urinary bladder and M&F 3 1720 57.9 0.5 )5.9 )3.3 )0.7 44.7 15.9 3.8 62.4 62.5 63.8 64.4
urethra
Malignant melanoma of M&F 3 1688 68.7 1.1 )8.8 )4.9 )2.1 92.5 29.0 8.5 74.9 76.9 79.4 84.1
skin
Malignant skin cancer, M&F 3 1238 82.5 )0.2 )0.4 )0.0 0.2 8.3 8.6 13.5 78.8 80.0 80.9 80.8
excl. melanoma

1367
 1368
Table 2 (Continued )
Site Sex Period Mean First Mean Mean difference Mean squared difference Last available comparable RSRs
width in number available annual
Cohort Complete Period Cohort Complete Period Cohort Complete Period Observed
years at risk RSR change
Eye M&F 5 236 60.2 0.2 )1.7 )1.1 )0.4 11.7 8.2 7.8 65.5 63.4 60.6 64.6

M. Talb€ack et al. / European Journal of Cancer 40 (2004) 1361–1372

Brain, excl. cranial M&F 3 352 18.2 3.2 )4.3 )2.4 )0.8 31.5 12.2 3.4 23.4 25.3 26.9 28.2
nerves, meningioma
Brain, meningioma M&F 3 376 64.3 1.7 )12.4 )5.8 2.5 167.0 36.8 49.2 78.6 79.6 82.2 87.5
Brain, intracranial nerves M&F 5 220 70.7 2.0 )11.4 )4.9 0.9 256.3 70.1 22.3 94.1 97.2 98.7 100.4
neurinoma
Thyroid gland M & F 3 561 69.3 1.0 )7.5 )4.1 )0.2 61.1 19.3 5.5 78.1 78.9 81.7 84.4
Endocrine glands M & F 3 1142 76.7 0.6 )8.2 )3.7 )0.1 83.2 17.7 2.9 84.2 85.2 86.5 87.4
Bone M & F 5 169 41.9 2.4 )10.2 )6.4 )2.4 112.9 44.2 11.8 47.9 51.3 54.2 59.3
Connective tissue, muscle M & F 5 413 45.4 0.3 )4.8 )2.4 )0.1 46.2 15.6 6.5 48.3 49.5 50.0 49.5
Non-Hodgkin’s M & F 3 679 30.1 2.1 )6.9 )3.9 )0.8 56.8 19.9 7.3 31.9 37.3 40.5 42.0
lymphoma
Hodgkin’s lymphoma M&F 3 282 42.4 3.0 )18.1 )11.7 )3.3 333.5 140.1 19.1 55.6 62.7 72.7 73.3
Multiple myeloma M&F 3 128 12.8 0.9 )1.0 )0.3 0.3 4.0 3.4 5.5 11.6 11.7 12.7 13.4
Acute lymphocytic M&F 5 186 19.5 5.9 )23.4 )17.3 )10.7 588.3 327.7 161.1 40.9 49.0 55.3 56. 9
leukaemia
Chronic lymphocytic M&F 3 206 22.9 3.5 )11.0 )7.3 )3.6 133.1 60.4 22.0 34.6 36.9 41.6 41.5
leukaemia
Acute myeloid leukaemia M&F 5 52 3.3 9.6 )3.6 )2.2 )1.0 17.7 7.7 3.3 5.5 8.3 11.5 12.8
Chronic myeloid M&F 5 41 7.7 5.6 )3.3 )2.6 )1.6 29.0 22.2 15.8 6.4 8.4 10.9 21.9
leukaemia
Period width in years: 1 ¼ 22 one-year intervals between 1970 and 1991; 3 ¼ 3-year moving averages, 20 intervals between 1971 and 1990; 5 ¼ 5-year moving averages, 18 intervals between 1972
and 1989. Mean number at risk: Mean number at risk at the beginning of the tenth year after diagnosis respectively for 1-, 3- and 5-year periods. First available RSR: First available cohort/observed
RSR. Period width ¼ 1, diagnostic year 1960; period width ¼ 3, diagnostic year 1960–1962; period width ¼ 5, diagnostic year 1960–1964. Mean annual change: Mean annual change of cohort/
observed RSR in percent. Mean difference: Cohort ¼ mean of all (cohort ) observed), Complete ¼ mean of all (complete ) observed), Period ¼ mean of all (period ) observed). Mean squared
difference: Cohort ¼ mean of all ((cohort ) observed)2 ); complete ¼ mean of all ((complete ) observed)2 ); period ¼ mean of all ((period ) observed)2 ). Last available comparable RSRs: Period
width ¼ 1, Cohort (diagnostic year) ¼ 1981, Complete (diagnostic year) ¼ 1981–1991, Period (Period) ¼ 1991 and Observed (diagnostic year) ¼ 1991. Period width ¼ 3, Cohort ¼ 1979–1981,
Complete ¼ 1979–1991, Period ¼ 1989–1991 and Observed ¼ 1989–1991. Period width ¼ 5, Cohort ¼ 1977–1981, Complete ¼ 1977–1991, Period ¼ 1987–1991 and Observed ¼ 1987–1991.
First and last available cohort/observed RSR with its mean annual change in percent. Mean difference and mean squared difference between the cohort, complete and period RSRs available at a
given time and the RSR later observed for patients diagnosed in this interval. Last available cohort, complete, period and observed RSRs that are comparable in time.
Table 3
Fifteen-year relative survival rate (RSR) for patients diagnosed between 1975 and 1986 at 0–89 years of age
Site Sex Period Mean First Mean Mean difference Mean squared difference Last available comparable RSRs
width in number available annual
Cohort Complete Period Cohort Complete Period Cohort Complete Period Observed
years at risk RSR change
All sites combined M&F 1 6459 38.7 1.2 )7.0 )3.8 0.6 50.5 14.9 0.5 37.3 41.4 45.5 44.9
Lip M&F 5 334 82.3 )1.3 10.8 8.9 6.6 120.6 84.9 59.4 87.8 88.1 89.4 76.3

M. Talb€ack et al. / European Journal of Cancer 40 (2004) 1361–1372

Oral cavity and M&F 5 238 29.8 0.8 5.3 3.6 0.9 29.4 15.2 7.5 37.4 33.6 30.5 31.9
mesopharynx
Oesophagus M&F 5 24 3.8 1.2 )0.1 0.3 0.5 0.3 0.1 0.3 3.7 4.2 4.4 4.4
Stomach M&F 3 271 11.8 4.0 )3.7 )2.9 )0.9 17.8 10.7 2.3 10.3 11.9 15.1 16.1
Small intestine, M&F 5 28 17.2 2.2 )9.5 )7.7 )6.1 109.2 71.0 46.9 12.3 18.4 21.1 23.2
adenocarcinoma
Small intestine, M&F 5 80 36.1 1.8 )5.0 )2.1 1.1 74.6 32.1 84.2 26.5 34.7 36.6 40.8
carcinoids
Colon, adenocarcinoma M&F 3 1274 39.3 1.4 )6.1 )3.6 1.3 39.7 13.2 2.3 37.6 41.9 47.1 44.5
Rectum, adenocarcinoma M&F 3 692 36.0 1.3 )5.7 )3.5 )0.6 35.2 15.0 2.8 33.6 35.9 40.0 39.9
Liver, primary M&F 5 8 1.9 )2.4 )0.2 0.0 )0.1 0.2 0.3 0.3 1.8 1.6 1.2 2.2
Gall bladder, biliary tract M&F 5 55 4.2 2.5 )0.5 )0.2 0.1 0.9 0.4 0.4 3.1 3.9 4.3 5.0
Pancreas M&F 5 27 1.2 3.3 0.7 0.3 )0.0 0.7 0.1 0.0 2.0 1.5 1.3 1.5
Nose and nasal sinuses M&F 5 47 33.0 3.5 )0.5 )0.6 0.5 16.2 14.5 10.3 32.9 34.1 37.6 39.6
Larynx M&F 5 274 50.4 )0.3 5.0 3.6 1.3 27.6 13.1 3.4 56.6 52.6 49.1 48.1
Lung M&F 3 212 6.3 0.1 )0.4 0.2 0.9 0.5 0.2 1.4 5.9 6.9 6.9 6.7
Breast Females 1 1457 50.8 1.3 )9.7 )6.5 )2.2 98.5 44.1 7.1 47.2 53.4 58.6 60.0
Cervix uteri Females 3 796 59.6 1.0 )1.7 )1.6 )1.3 8.7 6.9 4.6 60.2 60.7 63.6 63.1
Corpus uteri Females 3 1419 76.6 0.1 )5.9 )3.9 )1.3 43.1 18.6 6.6 75.2 76.2 77.0 77.7
Ovary Females 3 736 36.8 1.0 )5.1 )2.8 0.3 26.4 8.3 1.2 33.9 36.3 39.7 39.9
Prostate Males 1 332 27.9 0.5 )8.7 )3.3 1.6 89.4 26.0 19.9 23.3 31.7 32.4 32.9
Testis, seminoma Males 5 293 84.7 1.0 )2.5 )0.9 2.8 12.6 3.6 9.7 84.6 88.0 92.2 91.9
Testis, non-seminoma Males 5 260 64.3 4.8 )35.4 )20.6 )6.4 1325.6 441.1 52.5 46.8 67.6 88.5 88.9
Kidney excluding renal M&F 3 492 29.5 1.3 )5.9 )2.6 1.6 40.9 9.9 5.5 31.5 33.8 36.0 32.5
pelvis
Urinary bladder and M&F 3 1153 57.3 0.7 )11.8 )5.9 )1.0 145.1 36.4 2.2 53.5 57.5 60.7 60.8
urethra
Malignant melanoma of M&F 3 1413 69.2 0.8 )14.6 )7.1 )2.5 231.6 53.6 9.6 66.1 71.3 74.9 76.9
skin
Malignant skin cancer, M&F 3 690 77.2 0.4 )1.5 )0.3 0.1 5.6 2.2 14.5 79.1 78.1 77.1 77.3
excl. melanoma

1369
 1370
Table 3 (Continued )
Site Sex Period Mean First Mean Mean difference Mean squared difference Last available comparable RSRs
width in number available annual
Cohort Complete Period Cohort Complete Period Cohort Complete Period Observed
years at risk RSR change
Eye M&F 5 185 60.3 0.4 )5.7 )5.3 )3.1 35.1 30.2 16.0 56.6 59.1 62.7 62.4

M. Talb€ack et al. / European Journal of Cancer 40 (2004) 1361–1372

Brain, excl. cranial M&F 3 287 17.5 4.4 )4.3 )2.9 )0.5 21.5 12.1 1.9 18.2 20.4 25.0 25.9
nerves, meningioma
Brain, meningioma M&F 3 307 69.0 1.4 )17.5 )6.4 7.9 319.0 46.1 99.3 68.4 77.3 85.2 79.8
Brain, intracranial nerves M&F 5 192 87.2 1.3 )12.6 )6.7 5.1 171.3 50.3 52.3 77.3 93.0 107.8 94.9
neurinoma
Thyroid gland M & F 3 514 75.9 0.8 )10.4 )4.2 2.4 113.4 19.9 8.2 70.6 78.3 85.1 81.7
Endocrine glands M & F 3 904 79.6 0.1 )13.5 )3.8 1.5 200.4 23.3 6.2 70.7 79.0 81.3 82.1
Bone M & F 5 156 47.6 1.4 )15.9 )9.9 )4.2 263.5 99.2 29.7 40.1 44.4 52.1 51.7
Connective tissue, muscle M & F 5 322 51.9 )1.1 )8.7 )3.4 1.3 76.2 19.1 2.6 40.9 46.9 50.3 49.2
Non-Hodgkin’s M & F 3 413 26.3 3.1 )6.2 )2.3 1.7 60.6 9.2 7.9 25.7 29.4 32.5 33.5
lymphoma
Hodgkin’s lymphoma M&F 3 227 45.1 4.5 )27.2 )15.9 )3.9 749.8 259.7 26.4 34.3 46.3 56.3 67.0
Multiple myeloma M&F 3 43 8.5 )2.5 1.9 2.8 3.3 5.0 8.7 13.3 9.5 8.8 8.5 7.2
Acute lymphocytic M&F 5 175 37.4 4.8 )39.3 )31.5 )22.3 1573.8 994.8 502.6 6.0 19.6 34.1 54.4
leukaemia
Chronic lymphocytic M&F 3 112 19.3 5.3 )13.1 )8.7 )4.0 184.3 81.7 25.4 15.0 20.2 23.5 30.6
leukaemia
Acute myeloid leukaemia M&F 5 38 4.1 8.5 )4.8 )3.4 )2.6 24.6 12.9 8.0 2.6 3.9 4.7 7.8
Chronic myeloid M&F 5 14 2.8 10.3 )2.0 )1.6 )1.0 5.0 3.1 3.1 3.0 3.0 2.9 5.9
leukaemia
Period width in years: 1 ¼ 12 one-year intervals between 1975 and 1986; 3 ¼ 3-year moving averages, 10 intervals between 1976 and 1985; 5 ¼ 5-year moving averages, 8 intervals between 1977
and 1984. Mean number at risk: Mean number at risk at the beginning of the 15th year after diagnosis respectively for 1-, 3- and 5-year periods. First available RSR: First available cohort/observed
RSR. Period width ¼ 1, diagnostic year 1960; period width ¼ 3, diagnostic year 1960–1962; period width ¼ 5, diagnostic year 1960–1964. Mean annual change: Mean annual change of cohort/
observed RSR in percent. Mean difference: Cohort ¼ mean of all (cohort ) observed); complete ¼ mean of all (complete ) observed); period ¼ mean of all (period ) observed). Mean squared
difference: Cohort ¼ mean of all ((cohort ) observed)2 ); complete ¼ mean of all ((complete ) observed)2 ); period ¼ mean of all ((period ) observed)2 ). Last available comparable RSRs: Period
width ¼ 1, Cohort (diagnostic year) ¼ 1971, Complete (diagnostic year) ¼ 1971–1986, Period (Period) ¼ 1986 and Observed (diagnostic year) ¼ 1986. Period width ¼ 3, Cohort ¼ 1969–1971,
Complete ¼ 1969–1986, Period ¼ 1984–1986 and Observed ¼ 1984–1986. Period width ¼ 5, Cohort ¼ 1967–1971, Complete ¼ 1967–1986, Period ¼ 1982–1986 and Observed ¼ 1982–1986.
First and last available cohort/observed RSR with its mean annual change in percent. Mean difference and mean squared difference between the cohort, complete and period RSRs available at a
given time and the RSR later observed for patients diagnosed in this interval. Last available cohort, complete, period and observed RSRs that are comparable in time.
 M. Talb€ack et al. / European Journal of Cancer 40 (2004) 1361–1372 1371

4. Discussion duced in the late 1960s. But the improvement is also

likely to be partly explained by elimination of previous
To the best of our knowledge this is the largest misclassification. Fewer cases of non-Hodgkin lym-
evaluation of the period approach to survival analysis so phoma (NHL) have been diagnosed as Hodgkin’s lym-
far, both in terms of the number of cases included and phoma during the past two decades. NHL has poorer
the number of cancer sites analysed. Evaluations on survival than Hodgkin’s lymphoma. For patients diag-
large data have previously been performed mainly on nosed with acute lymphocytic leukaemia, better cyto-
material from the Finnish Cancer Registry [3–6]. static treatments are responsible for the improved
Improvements in survival may reflect a variety of survival. For chronic lymphocytic leukaemia, the im-
different factors such as increased and/or earlier diag- provements are likely to be a result of better cytostatic
nosis, a shift towards more favourable histological drug treatment, but long-term cure is still not possible
subtypes or sites (for all sites combined), or advances in for this disease with the regimens that have been avail-
treatment. But regardless of the origin, period analysis able so far. For acute myeloid leukaemia, more effective
allows for more up-to-date estimates, and if additional chemotherapy regimens, together with improved sup-
knowledge of the possible reasons behind the improve- portive care allowing for safer administration of inten-
ments is available this will lead to more accurate inter- sive treatment, are responsible for the improved
pretation of the results. There have been substantial survival. These treatments were introduced in the late
improvements in long-term cancer survival in Sweden 1960s, but continuous improvements have taken place.
during the past four decades. The possible reasons for The survival improvements for chronic myeloid leu-
the improvements have recently been reported elsewhere kaemia can be attributed to improved treatments, che-
[12]. For all sites combined, cancers with poor survival motherapy, interferon and, more recently, high-dose
have decreased and cancers with better survival have therapy followed by stem-cell transplantation [12].
increased their relative shares of the total cancer inci- As shown with the historical data evaluated in this
dence across the decades, and cancers have become study, period survival estimates are generally more ac-
relatively more common in older than younger groups. curate for describing the current situation than the
The individual forms of cancer discussed in Section 3 corresponding complete and cohort estimates. There are
above have had a variety of different reasons for their only a few exceptions to this general rule. The cohort,
improvements in survival. The increase in survival for complete and period RSRs all tended to underestimate
patients diagnosed with small intestine carcinoids can, at future observed RSR, but the underestimation was
least during the first decade, be attributed to improved considerably smaller for the period approach. If the
diagnostics and staging, centralisation of treatment for purpose of the analysis is to estimate/predict survival for
this rare disease, increased surgical aggressiveness, and a recently diagnosed group of patients, and not to
better antitumour drugs, particularly interferon and evaluate some particular cohort that has already com-
somatostatin. The improved survival for patients diag- pleted its follow up time of interest, then the period
nosed with gallbladder and biliary tract cancers is estimates seems to be the best available choice.
mainly due to better imaging techniques, surgical ag- When survival improves over time the period esti-
gressiveness and postoperative care. For primary liver mates will be higher than the corresponding cohort and
cancer, better imaging and earlier diagnosis have con- complete estimates, as is often the case in this study. The
tributed to the improvements, together with better op- opposite is expected if the survival is declining. No
portunities to evaluate the extent of liver disease and systematic difference would be seen if the survival was
liver function, and improved surgical techniques and constant over time. Period analysis will lead to an
postoperative care. For non-seminoma testicular cancer, overestimation of the long-term survival compared to
better staging procedures and, above all, more effective earlier years if improvements in early diagnosis or
treatments have caused the marked improvement. The treatment merely identify cancer at an earlier point in
introduction of cisplatin in 1978 has made the greatest time without prolonging life, i.e. lead-time bias. This is
contribution. Improved opportunities to diagnose and exemplified with an hypothetical example in Fig. 3,
operate safely on patients are likely causes of the sur- which depicts cumulative observed and period RSRs,
vival improvements for patients diagnosed with menin- and period RSRs that have an arbitrary lead-time of 6
gioma of the brain. Misclassification may also have been and 12 months, respectively, introduced for 50% of the
prevalent in the early years, which partly explains poor cases for the last five diagnostic years included in the
survival for a predominantly benign tumour. In recent period analysis. In the presence of lead-time the period
years the increased use of radiotherapy may also have RSRs will shift upward and thus overestimate the actual
contributed to improved survival. The major reason for survival. The same phenomenon is not seen for the
the improved survival for patients diagnosed with complete estimates if the same lead-time is introduced
Hodgkin’s lymphoma is better treatment. A major step (Fig. 4). The complete estimates would in this case be
was taken when combination chemotherapy was intro- the best alternative.
1372 M. Talb€ack et al. / European Journal of Cancer 40 (2004) 1361–1372

110
as a means to provide more up-to-date estimates of
100
Observed long-term survival for cancer patients. It is, however,
Cumulative RSR

90
Period
80
Lead-time 6m
important to recognise that both short- and long-term
70
Lead-time 12m survival should be considered. Otherwise, there is a risk
60
50 that short-term improvements will be missed, since the
40 focus so often is directed at 5- and 10-year survival.
30
20
10
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
5. Conflict of interest statement
Years since diagnosis
The authors have no financial or personal relation-
Fig. 3. The effect of lead-time bias in period analysis (an hypothetical ships or obligation to any person or organisation that
example). Cumulative observed and period relative survival rates
(RSR), and period RSRs that have an arbitrary lead-time of 6 and 12
have influenced our work with this article.
months introduced for 50% of the cases for the last 5 years of diag-
nosis. Cancer of the rectum. Males and females, 0–89 years of age at
diagnosis (site chosen for illustrative purposes).
References

1. Brenner H, Gefeller O. An alternative approach to monitoring

110 cancer patient survival. Cancer 1996, 78, 2004–2010.
100 2. Brenner H, Gefeller O. Deriving more up-to-date estimates of long-
Observed
Cumulative RSR

90
Complete term patient survival. J Clin Epidemiol 1997, 50, 211–216.
80
Lead-time 6m 3. Brenner H, Hakulinen T. Advanced detection of time trends in
70 Lead-time 12m long-term cancer patient survival: Experience from 50 years of
60
50 cancer registration in Finland. Am J Epidemiol 2002, 156, 566–577.
40 4. Brenner H, Hakulinen T. Up-to-date long-term survival curves
30 of patients with cancer by period analysis. J Clin Oncol 2002, 20,
20 826–832.
10 5. Brenner H, S€ oderman B, Hakulinen T. Use of period analysis for
0 providing more up-to-date estimates of long-term survival rates:
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
empirical evaluation among 370 000 cancer patients in Finland. Int
Years since diagnosis J Epidemiol 2002, 31, 456–462.
6. Brenner H, Hakulinen T. Very long-term survival rates of patients
Fig. 4. The effect of lead-time bias in complete analysis (an hypo-
with cancer. J Clin Oncol 2002, 20, 4405–4409.
thetical example). Cumulative observed and complete relative survival
7. Brenner H, Hakulinen T. Long-term cancer patient survival
rates (RSR), and complete RSRs that have an arbitrary lead-time of 6
achieved by the end of the 20th century: most up-to-date estimates
and 12 months introduced for 50% of the cases for the last 5 years of
from the nationwide Finnish cancer registry. Br J Cancer 2001, 85,
diagnosis. Cancer of the rectum. Males and females, 0–89 years of age
367–371.
at diagnosis (site chosen for illustrative purposes).
8. Brenner H. Long-term survival rates of cancer patients achieved by
the end of the 20th century: a period analysis. The Lancet 2002,
360, 1131–1135.
It is the stability of the RSR estimates over time, and 9. Aareleid T, Brenner H. Trends in cancer patient survival in Estonia
before and after the transition from a Soviet republic to an open-
not primarily the number at risk, that determines if market economy. Int J Cancer 2002, 102, 45–50.
period analysis will be successful for a given form of 10. Smith LK, Lambert PC, Jones DR. Up-to-date estimates of long-
cancer. There is, of course, a relation between the term cancer survival in England and Wales. Br J Cancer 2003, 89,
number at risk and stable estimates, but there is nothing 74–76.
11. Mattson B, Wallgren A. Completeness of the Swedish Cancer
per se that indicates that cancers with relatively few
Register. Non-notified cancer cases on death certificates in 1978.
cases cannot be analysed with the period approach. One Acta Radiol Oncol 1984, 23, 305–313.
disadvantage with period analysis compared to the 12. Talb€ack M, Stenbeck M, Rosen M, Barlow L, Glimelius B. Cancer
complete approach is its increase in random variation survival in Sweden 1960–1998 – developments across four decades.
due to the reduction in the number of patients at risk, as Acta Oncol 2003, 42, 637–659.
the analysis is restricted to events during one particular 13. Brenner H, Hakulinen T, Gefeller O. Computational realization of
period analysis for monitoring cancer patient survival. Epidemiol-
time period, commonly 1–5 years. This disadvantage ogy 2002, 13, 611–612.
seems in reality to be outweighed by the reduced vari- 14. Hakulinen T. Cancer survival corrected for heterogeneity in patient
ability between the observed and period RSRs, and if, as withdrawal. Biometrics 1982, 38, 933–942.
in this case, applied to population-based registry data 15. Voutilainen ET, Dickman PW, Hakulinen T. SURV2: Relative
this problem will probably be limited to rare cancer Survival Analysis Program version 2.02b. Helsinki, Finland,
Finnish Cancer Registry, 1998.
sites. 16. Information about SURV3 and the program can be downloaded
To conclude, our evaluation confirms earlier studies from the Finnish Cancer Registry web site. http://www.cancerreg-
[3–6] and shows that period survival analysis can be used istry.fi/surv3/.