Epidemiological problems
From "Human Cancer: epidemiology and environmental causes" by
J Higginson, CS Muri & N Mun~oz, Cambridge Monograph, 1992.
Epidemiological data can be subject to many problems. In the gathering
of cancer statistics the relative frequency of the different cancer sites
can more often reflect the accessibility of neoplasms or the
interests of local surgeons. Unless all histological material in
an area is pooled, considerable distortion can be caused
by hospital admission policies, facilities, the existence of
specialist departments, etc. This is readily seen by comparing data
from the Tata Memorial Hospital in Bombay, which draws cancer
patients from all over India, notably those with head and neck
tumours, with data covering the total Bombay
population.
Relative frequency of selected cancer sites in population-based
vs hospital registry, expressed as a percentage of all tumours observed.
| Bombay Cancer Registry
1968--1972
| Tata Memorial Hospital
1970--1972
|
|---|
| M | F | M | F
|
|---|
| Tongue | 9.2 | 2.4 | 15.1 | 2.2
|
| Mouth | 5.7 | 4.6 | 10.3 | 6.6
|
| Pharynx | 10.8 | 2.7 | 23.6 | 4.5
|
| Oesophagus | 9.6 | 7.8 | 11.3 | 8.1
|
| Stomach | 5.7 | 3.8 | 1.7 | 1.0
|
| Colon-rectum | 5.7 | 4.0 | 2.9 | 1.7
|
| Larynx | 9.4 | 2.0 | 1.8 | 0.8
|
| Lung | 8.9 | 2.1 | 5.7 | 0.9
|
| Breast | 0.1 | 17.2 | 0.1 | 17.9
|
| Cervix uteri | - | 21.7 | - | 35.5
|
| Prostate | 2.6 | - | 0.7 | -
|
| Bladder | 1.6 | 0.8 | 1.1 | 0.3
|
| Lymphoma | 3.3 | 1.8 | 4.3 | 1.9
|
| Leukaemia | 3.8 | 3.0 | 1.6 | 0.9
|
Autopsies
Autopsy series can be valuable but also are somewhat biased.
Not only are hospital admissions selected but not all deaths are
autopsied. In many countries, autopsy is rarely practised. Curable
cancers, e.g., skin and cervix, are poorly represented. Males are
more likely to be autopsied than females and the rate of autopsies decline
after the age of 40.
It may be possible to pool biopsy and autopsy material from
the same population to provide a more balanced picture. In Africa,
a high frequency of liver cancer was previously shown on autopsy,
but cancer of the cervix was rare; whereas biopsies showed the opposite
pattern. In contrast, both autopsy and biopsy studies demonstrated
a high frequency
...
the data, however, led to conclusions that were very similar to those
produced by later incidence studies.
Minimum incidence rates
If the population of a region or a city is known
approximately, and genuine residents can be identified in biopsy,
autopsy and other material, it may be possible to calculate a
minimum incidence rate, i.e. the true rate can not be less.
This may be useful in certain developing countries. Yaker (1980)
used this technique for Algiers, Constantine and Oran.
Comparison of incidence, mortality and relative frequency
Incidence and mortality are complementary. Incidence is not influenced
by survival but is frequently not available for entire populations.
Mortality data generally cover an entire country
and have been collected since the beginning of this century for much
of the developed world. However, cancer mortality statistics are, as
pointed out by Smithers, a `summary of what thousands of doctors of
varying skill have, under very different conditions and opportunities
for accurate diagnosis, seen fit to write as their opinion of the
cause of death' (Boyle et al, 1989). The accuracy of death
certificates has been a subject of study. It has been show that cancers are
not only dismissed, but also over-diagnosed, but that these errors
tend to cancel out (Heasman & Lipworth, 1966; Puffer &
Wynn-Griffith, 1967; James et al, 1955). The coding
of death certificates, for both the eighth and ninth revisions
of ICD also varies between countries, coders interpreting
the underlying cause of death for the same
certificate differently (Percy & Muir, 1989). Even when
the diagnosis of cancer has been made during life and
cancer mentioned as a cause of death, the death certificate tends
to be less precise, e.g. leukemia rather than myeloid leukemia.
Time trends and cohort analysis
The fact that cancer incidence and mortality change over time is
of the greatest significance. A change in burden of cancer
as represented by the absolute number of cases in a community resulting
from demographic factors, such as the increasing age of the population,
must be distinguished from a change in incidence as reflected
in a standardised rate. Changes may be artefactual. An `epidemic'
of cancer in the United States, described in the mid-1970s, was
eventually shown to be due to the late arrival of an
accumulation of death certificates at the National Centre for Health
Statistics.
...