Clegg,
LX, EJ Feuer, DN Midthune, MP Fay and BF Hankey. 2002. Impact
of Reporting Delay and Reporting Error on Cancer Incidence Rates
and Trends. Journal
of the National Cancer Institute 94:1537–45.
Press
coverage of this study
For
years, optimistic messages coming from the National Cancer Institute
and the American Cancer Society had implied that after decades
and billions of dollars we were winning the war on cancer.
Mortality
rates were on the way down because of better treatment, and even
incidence rates appeared to be leveling off, meaning that fewer
people were developing cancer than expected from past patterns.
But this new analysis by Clegg et al. reveals that for
several of the most common cancers the incidence statistics are
far less rosy.
Incidence
rates are still rising, not falling, and the trends for
several cancers are alarming. Thus while fewer people are dying,
more people are having their lives profoundly disrupted
by cancer.
What
did they do? Clegg et al. focused on the fact
that it takes time for all cancer cases to be reported to the national
registry of cancer, NCI's SEER (Surveillance, Epidemiology, and
End Results), and for diagnosis corrections to be incorporated.
Previous estimates of cancer incidence rates had not accounted for
this delay adequately.
Some
adjustment for delayed reporting was built explicitly into the system
by simply delaying the first reports of a year's data for two years,
waiting beyond the "deadline" (19 months after diagnosis)
for cancer case reporting. But some cases take much longer to report,
sometimes many years. Periodically the cancer incidence rates would
be corrected to incorporate these delayed cases. Some of the corrections
also involved removal of cases after corrected diagnosis.
Clegg
et al. developed a computer model that allowed them, based
on past experience, to predict the number of cases that are missed
due to reporting delay or removed because of diagnosis correction.
To build the model, they analyzed the history of reporting delays
at nine cancer registries across the US, focusing on melanoma, prostate
cancer, female breast cancers, colorectal cancer and lung/bronchus
cancers. They then used the model to predict missed/removed cases
in the most recent SEER cancer incidence report, 1998 (published
in 2000), and compared rates reported there with rates calculated
by summing known with predicted cases.
What
did they find? Clegg et al. found that reporting
delays for cancer cases can last years. They calculate that it "would
take 4–17 years for 99% or more of the cancer cases to be
reported," and that the numbers in hand at the end of the current
reporting period, 2 years, amount to somewhere between 88% and 97%
of final cases counted.
These
delays, in turn, introduce a bias into comparisons of recent vs
less-recent years. More cases were removed than added using the
old approach, and while this led to a better understanding of what
had happened in earlier years, it created a inappropriate base of
comparison between older and recent years: the old data set was
complete, the more recent one was not. Any comparison of time trends
was then biased toward finding lower rates in the most recent years,
all other things being equal.
How
big an effect did that bias create? The corrected estimates for
female breast cancer in whites in 1998, for example, is 4% higher
than the uncorrected calculation.
Cancer |
1998
Adjustment |
Female
breast cancer |
+4% |
Prostate
cancer
(white males)
|
+12% |
Prostate
cancer
(black males) |
+14% |
colorectal
cancer |
+3% |
melanoma
(whites)
|
+14% |
lung
cancer |
4% |
These
adjusments alter not only the assessment of cancer incidence in
1998, they change the trend analysis. For example, prior to the
adjustment, the trend analysis of female breast cancer for the years
1987 to 1998 was flat: the trend was not statistically distinguishable
from no annual change. The adjustment, however, revealed a statistically
significant 0.6% annual increase in breast cancer risk during recent
years.
Cancer |
Annual
Trend before |
Annual
Trend after |
Female
breast cancer
(white)
|
+0.4
ns |
+0.6% |
Prostate
cancer
(white males)
|
-0.1
ns |
+2.2%
ns |
colorectal
cancer
(white females)
|
+0.9%
ns |
+2.8%
ns |
melanoma
(white males)
|
-4.2%
ns |
+4.1% |
lung
cancer
(white females) |
0.5%
ns |
1.2% |
["ns"
indicates that the trend is statistically indistinguishable from
no annual change]
What
does it mean? According to the authors, "our results suggest
that ignoring reporting delay and reporting error may result in
the false impression of a recent decline in cancer incidence
when the apparent decline is, in fact, caused by delayed reporting
of the most recently diagnosed cases."
This
is important because the SEER data are key signposts used by the
policy and health community to gauge how well we are faring in the
war against cancer. As long as the incidence trends were perceived
to be headed downward, people could argue that we are winning. This
new analysis suggests that we continue to lose, and should heighten
pressure to direct new resources toward understanding the causes
of cancer and toward prevention.
|