Comparison between cancers identified by state cancer registry, self-report, and death certificate in a prospective cohort study of US radiologic technologists

When medical records or pathology reports to confirm diagnoses are unavailable, epidemiological studies often rely on self-reported diagnoses or death certificates to identify cancer outcomes. In assessing the validity of these sources, some studies focus on the positive predictive value, the percentage of reported cases in which cancer is correctly identified.¹ A second important measure of validity is false negative response, i.e. the percentage of cases, according to a gold standard, that are not identified in the self-report or death certificate or because of survey non-response. This is the complement of sensitivity, which is the percentage of cancers that are captured by self-reports and other sources. Here we present the extent of under-ascertainment due to self-report, death certificate, and survey non-response for cancers overall and six major cancer sites in a prospective, largely female cohort, the US Radiologic Technologists (USRT) Study, based on state cancer registries as the gold standard. A previous analysis of the USRT study presented the positive predictive values for cancers reported by cohort participants and death certificates.²

Self-administered questionnaires were sent to members of the USRT Cohort in three surveys carried out during 1983–89, 1994–98, and 2003 to the present. More than 110 000 study participants completed at least one of the first two questionnaires. Previous published reports analysing USRT data have relied on several sources to ascertain cancer outcomes: self-report, with cancers validated by medical records to the extent feasible; death certificates or the National Death Index, which include cancers identified as underlying or contributing causes of death; and additional subject contacts for ongoing studies.

To estimate the extent of under-reported cancers, we matched the 110 000 study participants with state-wide cancer registries in four states, California, Pennsylvania, Michigan, and Florida; about one-fourth of participating members reside in these four states. In 1990 the self-reported completeness of these state cancer registries ranged from 88 to 98%, although no completeness summary was reported for Michigan.³ Moreover, limited participant mobility contributed to the comprehensiveness of registry ascertainment, with almost 90% of participants living in the same state during the first survey (1983–89) as at the time of diagnosis in the registry state. Matches between persons with cancer diagnoses listed in registries and cohort members relied on a combination of probabilistic⁴ and deterministic algorithms and were based on participant social security numbers (97% available), names (100%), birthdates (100%), and gender (100%). We assessed under-reporting for cancers diagnosed between 1990 and the completion of the second questionnaire (1994–98), the date of death, or August 31, 1998 (for living participants who did not complete the second questionnaire), whichever occurred first.

There were a total of 761 invasive cancers reported in the registries for study participants during this period; 42% were diagnosed in California, 26% in Florida, 13% in Michigan, and 19% in Pennsylvania. The majority was diagnosed in women (69%) and those under age 60 in 1990 (65%).

A total of 196 of the 761 (25.8%) cancers identified by the registries were not captured by self-report, death certificate, or other follow-up (Table 1). Self-reports (from the second questionnaire) failed to identify 12.2% of total cancer diagnoses, although respondents failed to report only 2.6% of female breast cancers and <5% each of prostate and colon cancers. Ascertainment was less complete for death certificates, which did not identify 35.2% of cancers identified by the registries in those participants who had died before completing the second questionnaire.

Table 1

Under-ascertainment of cancers in the US Radiologic Technologists Study based on linkage with four State Cancer Registries (1990–98)^a

Type of cancer	Cancer registry-identified cancers by subject status at second survey	Cancer registry-identified cancers not identified in second survey by subject status
All malignancies	761
Questionnaire responder	508	62 (12.2)b
Deceased	165	58 (35.2)
Non-responder/lost to follow-up	88	76 ()c
Female breast	243
Questionnaire responder	189	5 (2.6)
Deceased	21	2(9.5)
Non-responder/lost to follow-up	33	31 ()
Prostate	84
Questionnaire responder	62	3 (4.8)
Deceased	14	8 (57.1)
Non-responder/lost to follow-up	8	7 ()
Colon	33
Questionnaire responder	22	1 (4.5)
Deceased	9	4 (44.4)
Non-responder/lost to follow-up	2	1 ()
Lung	67
Questionnaire responder	25	4 (16.0)
Deceased	38	12 (31.6)
Non-responder/lost to follow-up	4	3 ()
Melanoma	31
Questionnaire responder	18	4 (22.2)
Deceased	5	1 (20.0)
Non-responder/lost to follow-up	8	8 ()
Endometrial	38
Questionnaire responder	27	5 (18.5)
Deceased	5	4 (80.0)
Non-responder/lost to follow-up	6	5 ()

a

The four states are California, Florida, Pennsylvania, and Michigan.

b

Numbers in parentheses are percentages.

c

Some cancers were identified through follow-up related to studies within the cohort, even though the participant did not respond to the second questionnaire.

Estimates of the under-reporting of self-reported cancer in other studies vary widely, with high rates, 39%, for example, in a community-based study⁵; and more modest rates, 21%, in an American Cancer Society cohort.⁶ The extent of false-negative reporting has also varied substantially by cancer site across studies.^5,7–9 Yet several studies, including this one, reflect a similar pattern of ascertainment, i.e. higher ascertainment, for breast, colon, and prostate cancers, and lower ascertainment for melanoma and uterine/endometrial cancers. Levels of under-reporting by site are relatively low in the USRT cohort, with no more than ∼23% of the cancers under-reported by respondents in any of the six sites presented (Table 1). Because USRT study members work in health care settings, they may be more knowledgeable about their history of serious diseases, and this may at least partially account for their low rates of under-reported cancers.

This research was supported by the Intramural Research Program of the NIH, National Cancer Institute.