Content uploaded by Paul Blood
Author content
All content in this area was uploaded by Paul Blood on Dec 03, 2016
Content may be subject to copyright.
Available via license: CC BY-NC
Content may be subject to copyright.
4
CONJ • 15/1/05 RCSIO • 15/1/05
By Shannon Scott-Findlay, Jeff A. Sloan, Anne Nemecek,
Paul Blood, Cheryl Trylinski, Heather Whittaker,
Samy El Sayed, Jennifer Clinch, and Kong Khoo
Abstract
This is the third in a series of articles relating results from a
line of research whose intent was to construct a complete history
of patient in teracti ons with the healt h c are sy stem using
available data sources for all patients diagnosed in 1990 with a
primary breast, colorectal, or lung tumour in Manitoba. This
article presents details of the development and application of
methods to produce TNM staging data on the roughly 2,000
patients in th is po pu la tion. T he op er at io nal d ef in itions
constructed for this research can be adapted for other tumour
sites and data sources. Findings include methods developed to
overcome the sometimes ambiguous and inconsistent available
documentation, which ultimately produced reliable TNM staging
data. Survival data for this population by stage of disease are
given.
In the can cer care litera ture, stag ing is a cri tically im portant
covariate and prognostic for survival. Stagin g is includ ed in
virtually every published cancer study. A search through the
CancerLit database 1991 to 1995 reveals more th an 4,00 0
articles with the term ‘st aging’ in the title and 3 24 ar ticles with
a spec ific focus on neoplas m staging. The objective o f this
ma nu scrip t is to d es cribe how we ove rcame a ma jo r
methodol ogical hurdle to produce patho logical staging for
2, 00 0 ca ses re trosp ec tivel y. Altho ugh st aging data are
important for comparisons o f incidenc e and outcome, it is
diffic ult to apply a uniform stagi ng sys tem in practice with
consistent interpretat ion. A need for a co mprehensive method to
comp il e stag in g info rm at ion be ca me a pp ar ent du ri ng t he
implementation of the research. The goal of the study was to
detail histories on all patients diagnosed with breast, colorectal,
or lung ca ncer in Manitoba in th e year 1990 using exis ting
documentation and computerized data so urces. Culling the
staging information in an objective and reliable fashion from
these sources became a major challenge and focus of the
research program. This paper delineates the major problems
faced and the operational procedures de veloped to cir cumvent
or overcome them.
Motivation for this work on staging came from an exist ing
gap in the recording system for cancers in Man itoba. The
Ma ni toba Ca nc er Treat ment and Re search Fo undatio n
(MCTRF) has a legislated mandate to collect data on
malignancies diag nosed in Manitoba. A spe cial form, Form IV
‘Report of Ma lignant Neoplasm,’ is available to recor d all the
basic informatio n needed to construct a complete picture of a
cancer patient’s statu s and subsequent treatment. The form
states that it ‘should be completed at the firs t cancer di agnosis
an d agai n for each ne w p rimar y c ancer. Unfortu natel y,
compl iance is poor (S cott-Co nner & Christ ie, 1995). The
MCTRF estim ates that fewer than 15% o f n ewly diagnosed
cancers in Manitoba have a Form IV completed. The study team
had n o other choice but to atte mp t to cu ll th e required
information from the retrospectiv e chart data availa ble and
affili ated computer dat abases.
Staging data were gathered primarily through an abstraction
process involving patient information from MCTRF patient
Mapping the journey of
cancer patients through
the health care system
Part 3: An approach to staging
Shannon Scott-Findlay, RN, PhD(c), at the time of the project
was Research Assistant/Nurse for Community Cancer Programs
Network (CCPN) at the Manitoba Cancer Treatment and
Research Foundation (MCTRF). She is currently a Doctoral
Candidate, Faculty of Nursing, University of Alberta, Edmonton,
AB. Jeff A. Sloan, PhD, at the time of the project was
Biostatistician at the Faculty of Nursing, University of Manitoba.
He is currently Lead Biostatistician at the Mayo Clinic in
Rochester, MN. Anne Nemecek, RN, is the Previous Program
Director of Community Cancer Programs Network at MCTRF,
Winnipeg, MB. Paul Blood, MD, at the time of project was
Radiation Oncologist at MCTRF. Currently, he is a Radiation
Oncologist at the B.C. Cancer Agency, Victoria, BC. Cheryl
Trylinski, HRT, at the time of the project was Data Analyst for the
Community Cancer Programs Network. She is currently an
Outcomes Associate with the Cross Cancer Institute, Edmonton,
AB. Heather Whittaker, HRA, at the time of the project was
Director of Records & Registry at MCTRF. Currently, she is
Director of Health Records & Privacy Officer, CancerCare
Manitoba, Winnipeg, MB. Samy El Sayed, MD, at the time of the
project was Radiation Oncologist at MCTRF and is currently a
Radiation Oncologist at Ottawa Regional Cancer Centre, Ottawa,
ON. Jennifer Clinch, BSc, MA, at the time of the project was Co-
Director, Research Analysis at the WHO Collaborating Centre for
Quality of Life in Cancer Care. Currently, she is a biostatistician
at the Clinical Epidemiology Unit, Ottawa Health Research
Institute. Kong Khoo, MD, at the time of the project was a Medical
Oncologist at MCTRF and is currently a Radiation Oncologist at
Cancer Centre of the Southern Interior, Kelowna, BC.
doi:10.5737/1181912x15148
5
CONJ • 15/1/05 RCSIO • 15/1/05
records. The MCTRF Cancer Registry was used to identify the
patient population. The Cancer Registry registered 6,662 new
cases of cancer diagnosed in 1990. Approximately 30% of these
new cancer cases have been included in this study. The portion of
the 2,015 cases accorded to each type of cancer is split evenly
among the three sites of breast (654), colorectal (673), and lung
(688).
Staging data collection
A controversy that arose early in the development of the
stagi ng data collection system left us with an open
methodological and perhaps philosophical question: Can anyone
other than physicians produce accurate staging data? Several
clinicians commented that, for our data to be believed, the staging
data should be created by physicians, even though other health
care professionals have demonstrated the capacity to stage cancer
(Fehr, 1994).
It is mandatory under the Cancer Act for the MCTRF to collect
data on cancer diagnoses. The reality of the present record-keeping
system is that staging data are not easily obtainable from the
medical chart. Although it is logical to assume that the physician
involved in each cancer case is aware of the relevant case
characteristics, more often than not, they do not document it in a
fashion sufficient to produce a staging variable by retrospective
analysis of chart notes.
It was impractical to have attending physicians stage the more
than 2,000 cases, so we compromised by using physicians to
train and monitor the research assistant (a registered nurse) who
put the staging classification system into practice. During the
abstraction p rocess, the r es ea rch assistant a ss igned the
pathological stage for each case using the American Joint
Committee on Cancer (AJCC) staging criteria (AJCC, 1988).
Detailed operational rules were developed for the research
assistant to apply in producing the T, N, and M classifications
(tumour, nodes, metastases) on each case.
We took extra pains to ensure the data’s veracity. An expert
in oncology was identified for each cancer site and met with the
research assistant. These oncologists instructed the research
assistant on what to look f or an d how to c lassify chart
information. An initial series of 10 test cases from each of the
three chosen disease sites was run to check if, given the same
information, the research assistant would come to the same
conclusion as the clinician. An iterative process involving
further test cases followed until all oncologists involved were
satisfied with the research assistant’s ability to abstract the
required data consistently. Ultimate agreement rates between
clinicians and the research assistant were in excess of 90% for
the mor e than 60 cases reviewed initially. Any ‘difficult’ cases
were sent to the oncologists. For some tumours, the pathologist
had indicated the stage on the pathology report, in which case
the staging classification was not used. Instead, the research
assistant would independently stage the case, and then compare
the results. If the results were different, that particular case
would be giv en to one of the oncologists on the team for his/her
determination.
Once the data collection process had been completed, three
oncologists independently audited at least 10% of the staged
cases and then met with the research assistant to compare staging
resul ts. Agreem ent in a ll three sites was abov e 90 %.
Discrepancies were limited to minor interpretational issues. At
worst, a misclassification between the substage type would result
(e.g., IIa versus IIb). Discrepancies that did exist were restricted
to minor interpretational issues. Typically, this occurred when the
clinician had supplementary knowledge that was not obtainable
from the chart.
Staging data types
The type of stagin g to be implemented in the study wa s a
major issue for discussion. Pat hologic al sta ging was use d
based o n the assumption t hat it woul d provide a more acc urate
and consistent description of t he tum our than clini cal st aging.
Patho logical data are often available due to the substan tial
propo rtion of tumours that are resected. Ano ther system ic
difficulty is that roug hly 40% of all cancer case s are treated
ou ts ide the MCTRF. Tog et her, the se c halleng es m ade it
difficult to obtain stagin g data for a large propo rtion of c ases.
Breas t and c olorect al cases proce eded wel l usin g pathol ogical
stagi ng, but lung cance r was difficul t becaus e not many case s
had lobe ctomies . O nly about one-th ird of lung tumo urs are
resec ted and no patholo gy report was avai lable. The lac k of
infor mation for pathol ogical staging in lung cancer cases
requi red clinic al staging to be collec ted as well for cros s-
valid ation.
Staging operational definitions
In order to assess the quality and quantity of data available in
the MCTRF charts, the team implemented additional measures.
These guidelines deviated slightly from the criteria outlined in the
American Joint Committee of Cancer Care manual (AJCC, 1988).
Several operational rules had to be developed and implemented to
account for the state of available data. As each disease site under
study had unique challenges, different procedures were used for
each.
Th e code “X” wa s o nly im pleme nt ed if t here w as
information available to stage either T, N, or M, but the
in fo rmati on wa s amb ig uous, o r the re wa s ins uf fi cient
information to assign a stage. For inst ance, if the pathology
report in a breast cancer case sa id there were several nodes
affect ed, this would be indicated as “NX.” If no information
wa s av ailab le u po n wh ich to c on struc t a T, N, o r M
classification, th e field was left blank.
For breast cancer cases, if macroscopic residual tumour was
pres en t an d th e di me ns ions w er e st at ed, we added the
dimensions together to give the maximum tumour size. If the
Mark your calendar …
2nd Annual Canadian
Oncology Nursing Day
Tuesday, April 19, 2005
“Speak Up. Be An Advocate.”
doi:10.5737/1181912x15148
6
CONJ • 15/1/05 RCSIO • 15/1/05
size of the involvement was stated in the biopsy specimen and
there was residual, but the dimen sions were not state d, a stage
was determine d based on the word ing of the report. For
instance, if the biopsy contained 1.8 cm of tumour and there was
macroscopic residual involvement, the tumour was upstaged to
T2. If the pathology report indicated that there was microscopic
residual tumour, only the size of the biopsy was used to stage
the cancer. If the tumour size was not given, it wou ld be
recorded as TX.
Bo th pa tholo gical a nd c lin ic al st aging
me c han ism s we r e u sed for det ermi nin g
meta st at ic involvement. F or i nstance, a bone
scan that indica te d m et as ta sis w as su ff ic ie nt
for ou r st udy to discer n a po sitive metastat ic
re s ult , a nd a c o nfi rma t ory pa t hol ogic al
samp le was not req ui red for the metastatic
ca t ego riz a tio n. I n th e ev e nt o f an y
vagu en es s in a patholo gy report, th e chart
wa s f orwa rde d t o a n o nco logi st for
cons ul ta ti on a nd compl et io n of the TN M
stag in g proce ss .
There were special challenges in staging
br eas t ca nce r. Many wom en hav e an
as pir ati on/ bio psy f irs t, f oll owe d by a
lu mp ectomy/ ma stectom y. T he s tructur e of
the pathology report does not differ en ti at e
be twe en t he amo unt of int rad uct al and
invasive involvemen t. T hus, a 2 .5 c m ar ea of
carcinoma may be 1.5 cm intraductal and 1.0
cm invasive, but is staged as a T2 tu mo ur
be cau se th e pat hol ogy rep o rt d oes n ot
separate int raductal and invasive tum ou rs .
Here “over-st ag in g” may occur because the
st ruc tur e of p ath olo gy re por ts do es n ot
facilitate following the staging rules which
only us e the i nv as iv e portion of the specimen
for the stage. For lymph nodes, the degree of
mobility was alm os t nev er mentioned in the
pathology re po rt, so the l ymph nodes we re
assumed to b e mobile. Hence, i t was assumed
th at the lym ph n od es w er e mo vable and
un der tw o c ent ime tre s u nle ss oth erw ise
stated.
Th e pat hol ogy r epo rts f or color ect al
ca nce r we re the mo st deta il ed, but
te rmi no log y w as varie d a nd som etime s
ambig uous. It was d iffic ult to differentiate
betwe en T3 and T4 categories for so me c ases
becau se of the vagu eness in so me of the
patho logy repo rts with re spect to th e extent
of the cancer in the layers of the intestine. It
was decide d, in the case of a large invasive
tumou r, to assume a T3 c lassifi cation if there
was no operat ive report to state f urther organ
in vol ve men t. To as ses s meta sta si s fo r
colorectal cancer, CT scans, liver function
tests , and/or chest x-rays had to have bee n
compl eted. If none of these dia gnostic tests
were perform ed, “MX ” was recorded . If the
treatment chart indicated that the diagnostic
tests had not been comp leted, the chart was
pa sse d o n to a n o ncolo gis t for furth er
determination.
Lung ca ncer lymph no de s were clas si fi ed
ipsilateral to the affected lung unless
otherwise specified. Lung cancer cases were
scre en ed a f ur th er tim e to obt ai n a cli nical T, N, a nd M
stag in g fr om t he c ha rt r ec or d du e to a n in it ial fi nding that the
majo ri ty of l un g cance rs were n ot resect ed . Oncologis ts once
agai n pr ov id ed guidance and experti se to en su re rel iability.
Quali ty c he cks were aga in done and any que st io ns raised
duri ng th e c ha rt ab st raction were sent for revie w b y an
onco lo gi st .
These oper ational definitions made possib le th e sta ging of a
nu mber of c as es w hic h would oth erwis e ha ve r emain ed
Table One: Tumour classification by cancer site
T Breast Colorectal Lung Total
Tis 23 (4%) 48 (8%) 0 71 (5%)
1 295 (47%) 41 (8%) 82 (28%) 418 (21%)
2 212 (34%) 90 (13%) 116 (39%) 418 (21%)
3 29 (5%) 335 (59%) 10 (3%) 374 (25%)
4 22 (3%) 24 (4%) 24 (8%) 70 (5%)
X 48 (8%) 28 (5%) 65 (22%) 141 (9%)
Missing 25 (4%) 107 (16%) 391 (57%) 523 (26%)
Total 654 (32%) 673 (31%) 688 (34%) 2,015
Table Two: Node classification by cancer site
N Breast Colorectal Lung Total
0 326 (61%) 299 (59%) 144 (49%) 769 (58%)
1 193 (36%) 102 (20%) 74 (25%) 369 (28%)
2 7 (1%) 58 (12%) 55 (18%) 120 (9%)
3 0 (0%) 5 (1%) 13 (4%) 18 (1%)
X 9 (1%) 34 (7%) 8 (3%) 51 (4%)
Missing 119 (18%) 175 (26%) 394 (57%) 688 (34%)
Total 654 673 688 2,015
Table Three: Metastases classification by cancer site
M Breast Colorectal Lung Total
0 171 (90%) 73 (47%) 105 (33%) 349 (53%)
1 16 (8%) 69 (44%) 194 (62%) 279 (42%)
X 4 (2%) 14 (9%) 15 (5%) 33 (5%)
Missing 463 (71%) 517 (77%) 374 (54%) 1354 (67%)
Total 654 673 688 2,015
Table Four: Staging results for 2,015 cancer cases
Breast Colorectal Lung Overall
Stage Tumours Tumours Tumours Results
I 223 (40%) 85 (16%) 111 (29%) 419 (29%)
II 252 (45%) 182 (34%) 36 (9%) 470 (32%)
III 40 (7%) 148 (28%) 40 (11%) 228 (16%)
IV 16 (3%) 69 (13%) 194 (51%) 279 (19%)
Tis 23 (4%) 48 (9%) 0 71 (4%)
Missing 100 (15%) 141 (21%) 307 (45%) 548 (27%)
Total 654 673 688 2,015
doi:10.5737/1181912x15148
7
CONJ • 15/1/05 RCSIO • 15/1/05
missi ng. We estima te th e addi tional number of c ases to be
be low 5 % of the t otal cas es. The m ain i mpa ct of th e
defin itions, as spec ified above, more likely was to introduc e a
sligh t bias towards over-staging of some tu mours b y one level.
Again, this bias is estimated to be in the o rder of less th an 5%
of all cases.
Staging results
Breakdown by site and T classification is found in Table One.
Percentages for the classifiable cases are given exclusive of the
missing data while the missing data percentages reflect the
portion of the total number of cases. Similar results for node (N)
and metastases classification (M) are given in Tables Two and
Three respectively.
A computer algorithm used the rules set out in the AJCC
manual (AJCC , 19 88) to take the T, N, and M results from the
chart abs traction proce ss and produce a TNM classifica tion.
Even after a thorough review of the available chart record s and
ex te nsive ope ra tiona lizatio ns, stagi ng d at a we re s till
unobtainable for 27% of the 2,015 br east, colorect al, and lung
ca nc er mal ig nanci es diagn osed in 1 99 0 (Tabl e F our).
Percentages in Table Four sum to 100% exclusive of th e
missing cases. For example , the 228 stage III malignancies
represent 16% of the 1,467 cases for whic h a TNM stage wa s
obta in ed. The p ercentage reported be si de the number of
mi ss ing c ases is relat iv e to the to ta l num ber o f 2 ,0 15
malignancies.
More than 40% of breast ma lignancies were stage I with a
furth er 45% be ing stag e II. In total, four out of every five
breas t cancer cases were in early sta ge of dis ease. Only 3%
were stage IV. Bre ast cance r cases had the bes t documen tation
in terms of being abl e to sta ge all but 15% of the cas es.
Color ectal cases wer e un stageable in just ove r on e-fifth of the
673 case s. Stages II and III acco unted for two-thirds of thes e
malig nancies . The 252 stag e II breast tu mours comp rised 170
stage IIB and 82 stage IIB. The 40 stage III breast tu mours
equal ly divided into stag e III A and stage IIIB classi ficatio ns
with 19 and 21 ca ses respect ively. The 40 stage III lung
tumou rs had 11 stag e IIIA and 29 stage IIIB class ificati ons.
Lung can cer cases were unstag eable 45% of the time and, in
fact, account ed for 56 % of the cases for which insufficient
docum entatio n was available to pr oduce a TNM st age. Of the
st age ab le lun g cas es, h alf we re cla ssifi ed as sta ge IV,
highl ighting the s everity o f th e disease at di agnosis r elative to
the other two cancer sites .
Grouping disease stages into an early/late dichotomy, with
early defined as I, II, or Tis and late as III or IV, revealed
differences in the disease site stage distributions. Almost 90% of
breast tumours diagnosed were early stage, roughly half of the
colorectal cases and a third of the lung cancer cases appeared in
the early stage of disease.
Disease stage and survival
Figures One, Two, and Three demonstrate the diffe rence
between the early and late stage cancer patients by disease and
age at death. F ive-year Kaplan-Meier survival rate estimates for
early/late stage breast cancer patients are above 90% and just
under 50% respectively (Figure One). Late-stage breast cancer
patients are at increased risk of death, especially in the first
three months post-diagnosis, but the risk is small relative to
other disease sites. Survival curves for early - and late-stage
colorectal cancer (Figure Two) indicate the prognosis for this
disease site is better than the lung cancer, but worse than breast
cancer.
Lung cancer patients in late stage of disease can expect to live
three times shorter from diagnosis than those in early stage
cancer (Table Five). The lung cancer survival curves (Figure
Three) indicate that those in early stage of disease have a better
than 50% chance of surviving five years, while those in late-
stage disease have only a 10% chance of survival to five years
post-diagnosis. Lung cancer patients can expect to live an
average of just under three years if the disease is diagnosed early
(Table Five).
Figure One: Survival in breast cancer patients by disease
stage. Manitobans diagnosed in 1990 (N=643)
Figure Two: Survival in colorectal cancer patients by disease
stage. Manitobans diagnosed in 1990 (N=655)
Figure Three: Survival in lung cancer patients by disease
stage. Manitobans diagnosed in 1990 (N=681)
doi:10.5737/1181912x15148
8
CONJ • 15/1/05 RCSIO • 15/1/05
Breast cancer patients with early stage of disease can expect to
live into their 90s on average (Table Six). Even those with late-
stage disease averaged well into their 70s before death. Colorectal
patients live a full 10 years longer than lung cancer patients if
diagnosed early and six years if diagnosed in the latter stages
(Table Six). Their age at death is roughly three years less than that
of breast cancer patients regardless of disease stage. For lung
cancer, Figure Six highlights the discrepancy between early- and
late-stage disease. Late-stage cancer patients live an average of six
years less than those in early stage of disease at time of
presentation.
Discussion
This segment of the research into constructing complete
histories of patient interactions with the Manitoba health care
system represe nted a major hurdle. Through c onsiderable
discussion and operational definition, T, N, M staging was
produced for the majority of tumours from available pathology
information in a reliable and consistent manner. The approach
could be adapted to other disease sites. For example, to repeat the
process for prostate cancers would only involve an examination of
staging peculiarities for the particular disease relative to breast,
colorectal, and lun g. The abstraction, validat ion, and
amalgamation process to produce the T, N, M staging data remains
the same. Ultimately, any cancer tumour could be staged from the
existing data sources using our methods.
The staging system we developed will provide as reliable
retrospective data as is possible to be obtained from the present
charting system. While it would be desirable that every clinician enter
the precise staging information so that others may use this important
clinical variable, it is not reasonable to assume that it will become
achievable in the near future. As such, our approach provides a means
for the optimal amount of staging data to be abstracted from available
information.
The approach employed in this study was to have only one person
carry out the staging determination and, thereby, become as
intricately aware of staging as any physician/oncologist in terms of
using data available from charts. Fehr (1994) came to the conclusion
that physicians are not consistent among themselves. We, thus,
circumvented the issue of staging data consistency in terms of inter-
rater reliability by using a single rater with reliability checks
provided by clinicians auditing the results. Many meetings with
oncologists were essential to produce clinically relevant and reliable
information. The training program developed for the research
assistant combined with the quality control checks of the clinicians
formed a model that can be used by other researchers. The inter-rater
consistency achieved was, in our opinion, higher than what would
have been obtained if complete staging data from physicians had
been available. The provincial physician variability in staging cancer
is undoubtedly higher than the variability of our data due to the
extensive data verification procedures.
The success of this systematic staging construction system is made
more remarkable in that the databases incorporated into this project
were built with a different intent in mind than building patient
histories or carrying out clinical research. As such, the quality of the
data for research purposes was somewhat lacking initially. There are
gaps in the data with missing, incorrect, and unusable data in all
sources. A large part of the challenge and success, therefore, became
the separating of the wheat from the chaff to salvage usable clinical
data for analysis. Even with the extensive measures taken to develop
a staging collection methodology and detailed chart review to recover
the information, 27% of the cases were unstageable in this population.
This finding has helped create changes in the MCTRF data collection
process so that staging data will be incorporated in the future. As
treatment planning is based to a great degree on stage of disease, this
alteration to the content of available data is an important improvement
in the documentation process.
The critical nature of staging data to cancer treatment and research
cannot be overstated. Results indicate that there is a need for better
data collection of basic variables to be carried out at the clinician
level. Complete basic data collected during the course of clinical care
often reside mainly in the minds of the physicians/oncologists. The
operationally defined data collection tools developed for this project
provide an easily completed mechanism to ensure that the basic data
are readily available. The standard data collection instrument
developed in this study is convenient for clinicians to complete and
for researchers to use as support for the veracity of any research study
that includes staging information. With careful construction of staging
information, one can put greater stock in the subsequent statistical
analyses because they are based on reliable classifications.
Acknowledgements
Financial support for this study was provided by the Manitoba
Cancer Treatment Research Foundation and a grant from the
Manitoba Medical Service Foundation.
At the time of this study, the Manitoba Oncology Centre was called
the Manitoba Cancer Treatment and Research Foundation, it is at
present called CancerCare Manitoba.
American Joint Committee on Cancer, TNM Committee of the
International Union Against Cancer. (1988). Handbook for staging
of cancer. In O.H. Beahrs et al. (Eds), Manual for staging of
cancer (4th ed.). Philadelphia, PA: J.B. Lippincott Company.
Fehr, C. (1994). Comparison of TNM Staging for female breast
cancers by clinical oncologists and by a registry coder. Cancer
Record, Autumn, 12, 3-5.
Scott-Conner, C., & Christie, D. (1995). Cancer staging using the
American Joint Committee on Cancer TNM System. Journal of
the American College of Surgeons, 181(2), 182-8.
References
Table Five: Mean (median) days censored survival from
diagnosis by cancer site
Cancer Site Early Stage Late Stage
Breast 1,476 (1,550) 1,008 (1,451)
Colorectal 1,351 (1,638) 793 (596)
Lung 1,012 (1,240) 301 (125)
Total 1,327 (74%) 465(26%)
Table Six: Estimated mean (median) censored age at death in
years by cancer site
Cancer Site Early Stage Late Stage
Breast 90 (93) 75 (79)
Colorectal 86 (89) 75 (76)
Lung 76 (76) 69 (70)
Total 1,327 (74%) 465 (26%)
doi:10.5737/1181912x15148