Chromosomal Aberrations in Lymphocytes Predict Human Cancer Independently of Exposure to Carcinogens

Cohorts and Cytogenetic Methods.

A database was established of 3541 subjects (1968 from the Nordic countries and 1573 from Italy) examined for CAs in PBLs at adult age in 10 Nordic and 10 Italian laboratories from 1965 to 1988 (9) . For each subject, a personal identification code and the year and result of the CA test (total percentage of aberrant cells) were registered. CAs were classified according to the International System for Human Cytogenetic Nomenclature (10) , and at least 100 metaphases from each individual were scored. Gaps were not counted. To standardize for interlaboratory variation, the CA scores were trichotomized within each laboratory into three levels, low (1st to 33rd percentiles), medium (34th to 66th percentiles), and high (67th to 100th percentiles). The CA data based on 48- and 72-h culture times were trichotomized separately. More details on the cohort studies and the laboratory protocols can be found in previously published reports (6, 7, 8, 9) .

Cases and Controls.

In the Nordic cohorts, information on malignant tumors diagnosed from the date of CA testing until the end of 1993 (Denmark), 1994 (Sweden and Norway), or 1995 (Finland) was obtained from national cancer registries. Overall, the Nordic cohorts comprised 93 incident cancer cases [2 more Swedish cases were included than in the cohort analysis (8) due to an additional year of follow-up]. In the Italian cohort, the specific causes of death until April 30, 1996, were obtained from the municipality of residence. In total, the Italian cohort comprised 62 deceased cancer cases (two cases included in the cohort analysis were excluded because their tumors had been diagnosed before the date of CA testing). The distribution of subjects in the whole database by major cancer site and by the other variables considered in the analysis is given in Table 1 ⇓ .

The study protocol was approved by ethics committees and legal authorities in all participating countries.

For each case, our aim was to select four controls from the cohort (11) . Controls were retrospectively and randomly selected from the subjects in the cohort who were at risk for the outcome event in the calendar year of each case occurrence (12) . Controls were matched with their corresponding case by country (only the Nordic data), sex, year of birth, and year of CA test. The maximum differences permitted between the matched subjects’ year of birth and year of CA test were ±15 and ±5 years, respectively. Thus, the matched sets generally consisted of one case and four controls, except for eight Nordic and three Italian sets, which had three controls.

Ascertainment of Occupational Exposure and Smoking Habits.

Occupational hygienists performed partly structured telephone interviews with cases and controls or, if deceased, with next-of-kin (widows, widowers, children, parents, or siblings). Exposure measurements performed for the original cytogenetic studies or for other purposes were used. Other information sources included contacts with companies at which the subjects had been employed, former co-workers, company records, and medical records. In some instances, when the medical or company records were reliable and covered the whole occupational history, the subjects were not interviewed. Exposure data for 4 cases and 10 controls in the Nordic countries and for 2 cases and 20 controls in Italy were missing or were too poor for exposure classification. Subjects who had quit smoking were classified as nonsmokers when the available data implied that the subject had quit smoking cigarettes more than 5 years before the CA test year. Eight cases and 13 controls in the Nordic countries and 10 cases and 30 controls in Italy lacked smoking data.

Occupational Exposure Matrix.

Scientific publications and other information on all original cytogenetic studies from which the cohorts were recruited were scrutinized for occupational exposures of potential interest (6 , 7 , 13) . A matrix comprising 22 categorized exposure indices was constructed (Table 2) ⇓ . Exposure assessment was semiquantitative except for three categories classified as exposed versus nonexposed.

The exposures in the matrix were ultimately classified into three groups to optimize statistical analysis (Table 2) ⇓ . Group A: exposures evaluated in the original cytogenetic study to agents classified by the IARC as human carcinogens (class 1); Group B: exposures to other IARC class 1 agents not described in the original papers and revealed by the exposure assessment process, and Group C: all other agents in the matrix.

Occupational exposure was assessed annually starting from the year the subject left school until the end of follow up. Only exposure periods exceeding 1 year were assessed.

For subjects classified as nonexposed, an accuracy assessment of their exposure information was performed. Subjects for whom occupational exposure to any agent in the matrix was unlikely, either because of the nature of their reported job or because information available totally excluded any such exposure, were classified as high-accuracy nonexposed. The rest of the subjects, for whom exposure had been possible although not likely, were classified as moderate-accuracy nonexposed.

Quality Control of Occupational Exposure Assessment.

After the first national exposure assessments, a description of working histories, including types and levels of exposures, was sent to an independent occupational hygienist. Based on those descriptions, the hygienist selected 27 histories to be used in the harmonization procedure (5% of the whole study group). Discrepancies were discussed at a meeting, and consensus on all assessments was reached, which resulted in a harmonization of the exposure categorization. The final exposure assessments were thereafter performed on a national basis. At the end of the assessment, a quality control round testing interassessor repeatability was carried out. The round included 55 (10%) randomly selected subjects evenly distributed among the countries. For the occupational exposure classification presented in Table 4 ⇓ , three different classifications were possible. All five occupational hygienists were in agreement for 33 subjects (60%), and four of five were in agreement for 13 subjects (23.6%). For the remaining nine subjects (16.4%), a lower degree of agreement was reached, although for eight of nine subjects, the disagreement was restricted to contiguous exposure classifications.

Statistical Methods.

Associations between total cancer incidence and mortality, CAs at test, and occupational exposure and smoking habits were modeled by means of conditional logistic regression (14) , whereby the matching factors [i.e., country (only the Nordic data), sex, year of birth, and CA test] were controlled in the analyses. The ORs between the medium and low groups and between the high and low CA groups were estimated. This measure can be interpreted as the total cancer incidence or mortality ratio between the compared CA groups (15) . The main purpose of the statistical analysis was to examine whether stratification of data by occupational exposure or smoking habit generated different stratum-specific ORs. Homogeneity of stratum-specific ORs was tested by the likelihood ratio test, comparing models without and with the relevant interaction term (14) . We also performed restricted analyses based on matched sets in which the case had occurred within a limited time period after the date of CA testing, as well as on matched sets in which the case tumors belonged to a certain category of diagnosis in the three time periods described in Fig. 1 ⇓ . The threshold of 5 years before CA test, which discriminates recent from past exposures, was chosen in consideration of the half-life of PBLs (16) . Moreover, subgroup analyses among nonexposed subjects and among nonsmokers were performed by means of unconditional logistic regression.

Cancer Predictivity of CAs.

The adjusted ORs for developing cancer for subjects with a high level of CAs indicated a statistically significant increase in risk compared to those with a low level of CAs (Table 3) ⇓ . These estimates did not differ substantially from the crude ORs and were not significantly modified by country (only considered for the Nordic countries), sex, age at test, or time since test (all P > 0.11; likelihood ratio test).

Impact of Occupational Exposure on the Cancer Predictivity of CAs.

The ORs associated with CA frequency reported in Table 3 ⇓ did not change noticeably, when occupational exposure level was included in the conditional logistic regression models. Table 4 ⇓ shows the data for the period of 5 years or less before CA test; results from the other two periods were similar.

No marked confounding by occupational exposure was found when the analysis was restricted to cases (and their controls) occurring within a limited period of time after the CA test (7 and 10 years, median values for Nordic countries and Italy, respectively).

An informative data subset is represented by those subjects classified as nonexposed during the whole assessment period. Crude ORs of 1.22 (0.30–5.22) for the medium versus low CA group and 2.86 (0.85–9.66) for the high versus low CA group were found for Nordic countries (18 cases and 80 controls). The estimates were not noticeably changed when adjustment was made for the matching factors and the accuracy scoring of the nonexposed subjects. For Italy, the corresponding subgroup analysis could not be performed due to the small number of cases (n = 4) and controls (n = 20).

Table 5 ⇓ shows the maximum occupational exposure levels of the cases and controls assessed for different time periods relative to the time of CA testing. For most individuals, the exposure classification remained the same during all three time periods. Regression models considering changes of exposure status between time periods or based on alternative definitions of exposure (duration of occupational exposure and different time periods relative to the CA test) were performed. None of these models revealed any evidence of modification or confounding effect.

Effect of Cigarette Smoking on the Cancer Predictivity of CAs.

The association between CA frequency and cancer risk did not change markedly for the Nordic countries or for Italy when cigarette smoking status assessed for the time period ≤5 years before CA testing was included in the conditional logistic regression models (Table 6) ⇓ .

For Italy, the relatively large number of deaths from respiratory tract cancer (codes 161–162 according to 7th International Classification of Diseases), i.e., 5, 7, and 11 in the low, medium, and high CA groups, respectively, allowed restriction of the analysis to this site. When smoking was factored in, the ORs changed from 1.80 (CI, 0.50–6.51) to 1.36 (CI, 0.33–5.70) for medium versus low CA group and from 2.75 (CI, 0.87–8.62) to 2.40 (CI, 0.67–8.56) for high versus low CA group. In the Nordic countries, there were too few respiratory tract cancer cases (two, one, and eight cases in the low, medium, and high CA groups, respectively) to allow such a subgroup analysis.

The crude ORs within each stratum of the cigarette smoking variable revealed no obvious modification of the total cancer predictivity of the CA biomarker (P = 0.6 for the Nordic countries; P = 1.0 for Italy; likelihood ratio test; Table 6 ⇓ ). The group of Nordic subjects (28 cases and 134 controls) classified as nonsmokers during the whole assessment period (including non-pipe- and non-cigar-smokers) showed crude ORs of 1.28 (CI, 0.47–3.49) for medium versus low CA group and 2.52 (CI, 0.92–6.94) for high versus low CA group. These estimates were not markedly changed when adjustment was made for the matching factors. For Italy, the corresponding subgroup analysis was not performed because of the small number of cases classified as nonsmokers during the whole assessment period (8 cases and 63 controls).

Finally, ORs adjusted for matching factors and both occupational exposure and smoking habit were 0.86 (CI, 0.45–1.66) for medium versus low CA group and 2.25 (CI, 1.24–4.09) for high versus low CA group for the Nordic countries and 1.59 (CI, 0.72–3.48) and 2.56 (CI, 1.18–5.56), respectively, for Italy.