| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Section of Endocrinology and Metabolism (M.S.M, E.S.-F., B.J.C., A.B.S.), University of Illinois, Chicago, Illinois 60612; and Division of Cancer Epidemiology and Genetics (J.L., E.R.), National Cancer Institute, Bethesda, Maryland 20892
Address all correspondence and requests for reprints to: Dr. Arthur B. Schneider, Section of Endocrinology and Metabolism, University of Illinois, 1819 West Polk Street, MC 640, Chicago, Illinois 60612. E-mail: abschnei{at}uic.edu.
 |
Abstract |
|---|
 Top Abstract IntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
The study of this cohort has allowed us to look for patterns of tumors that would suggest that there is variable susceptibility to radiation in the general population. It is reasonable to expect such variation because some rare hereditary diseases, such as ataxia telangiectasia, are associated with a striking increase in susceptibility to radiation (10). However, it is less clear whether heterozygous carriers of these diseases or polymorphisms in other genes, such as DNA repair genes, are associated with radiation susceptibility (11, 12, 13). We hypothesized that if there were such susceptibility, an individual who developed one radiation-associated neoplasm might have an increased risk of developing a second. In a recent analysis of our cohort we did not find this (14).
Here, we analyze the occurrence of head and neck tumors in irradiated sibling pairs to look for evidence of familial susceptibility. For the 251 informative sibling pairs in the cohort, we determined whether the distribution of tumors was random (i.e. based only on known risk factors, e.g. dose of radiation) or whether familial factors are involved.
Fifteen years ago, we carried out a more limited analysis (15). We revisited this question now because we have substantially improved the quality and quantity of the data as well as the statistical analysis. Many more thyroid and other head or neck neoplasms have developed during the 15 yr of follow-up, individual organ doses have been estimated (1), and there is a better understanding of risk factors for parathyroid, salivary, and neural tumors, including dose-response relationships for parathyroid and salivary neoplasms. We also devised new methods to include a wider range of neoplasms in the analysis.
As we observed an apparent excess of tumor concordance (presence or absence of neoplasms) in sibling pairs, we tested the hypothesis that after taking known risk factors into account, this excess was not a result of chance alone. We also examined thyroid cancers to determine whether their characteristics and behavior in one sibling were related to the presence of a radiation-related neoplasm (concordance) in the other sibling.
|
Subjects and Methods |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Of 5373 patients who received external radiation treatment at Michael Reese Hospital between 1939 and the early 1960s for benign conditions of the head and neck area, 4296 were treated with conventional external radiation before their 16th birthday (1). These 4296 patients comprise the study cohort. Radiation doses to specific organs were estimated for 3842 patients in the cohort (1). Follow-up was continued through December 31, 2000.
Sibling groups
There were 677 subjects in sibling groups (two or more individuals) in the cohort of 4296 patients. Sixty-four individuals (32 pairs) were excluded because of the absence of follow-up information for one or both members of the sibling group. Of the 613 remaining subjects, 88 individuals (44 pairs) were removed for lack of thyroid-specific doses in one or both siblings. Among the remaining 525 individuals there were several sibling groups of three or four individuals. In these cases the elder two siblings were included in study. As a result, from three family groups with four members, six individuals were removed, and for 17 family groups with three members, 17 individuals were removed. This resulted in 502 individuals in 251 sibling pairs for the present study (Fig. 1
). The mean (±SD) follow-up for these 502 individuals was 44.3 ± 9.4 yr (range, 13.7–59.5).
|
Information for this family study was obtained from the general surveys conducted for the entire cohort. Self-administered questionnaires were sent by mail. Included with the surveys was information about the potential risks of childhood radiation exposure. When a neoplasm was reported to us, based on surgery or other diagnostic findings, we obtained and reviewed the relevant medical records and specimens.
In analyzing thyroid cancer, years at risk was the time from initial radiation treatment to the date of first diagnosis for cancer (for those who developed cancer) or to the end of last known follow-up (for those who did not). The same model was used for neural (meningiomas, acoustic neuromas, and other Schwann cell tumors of the head and neck area) and salivary gland (benign and malignant) tumors.
Hyperparathyroidism was defined as elevated calcium levels with elevated PTH levels and/or surgical correction of hypercalcemia by removal of one or more parathyroid glands. For those with hyperparathyroidism, years at risk were calculated from exposure until the date of surgery or, for those who did not have surgery, the date when the diagnosis was confirmed by an inappropriately high PTH level.
All 83 observed thyroid cancers, regardless of size, were included in the analysis. For seven of them, the size was not known, 32 were less than 10 mm (including 11 considered microscopic), and the remaining 44 (53%) were 10 mm or more. For some analyses, benign and malignant thyroid neoplasms were considered as a single end point. Previously, we showed that approximately 90% of the members of this cohort have ultrasound-detected thyroid nodules (16). Therefore, we limited this end point to neoplasms that were 10 mm or more in largest dimension determined at the time of surgery, by ultrasound, or by other diagnostic means. For this end point years at risk was defined as the interval between the date of initial radiation treatment and the date of surgery for nodules or, for those who did not have surgery, the date of diagnosis. The group consisted of 180 individuals with thyroid neoplasms (nodules) confirmed by surgical pathology to be malignant or benign and 10 mm or larger. Another 17 individuals with nodules were included based on the following findings: 14 had ultrasound documented nodules 10 mm or larger, and three had nodules 10 mm or larger by palpation.
Organ-specific radiation doses were estimated as reported previously (1). The dose for the thyroid gland was also used for the parathyroid glands. A weighted average of the doses to the different salivary glands was used as described previously (7). It is not possible to determine an average dose for the whole group of neural tumors given the wide range of tumor locations, so no dose estimates for these tumors are available.
Risk factor analysis
For each individual, we needed to calculate the risk that they would develop each of the tumor end points (thyroid cancer, thyroid nodules, hyperparathyroidism, and salivary or neural tumors). We adapted statistical methods that were previously described in detail (14, 15). In brief, Cox proportional hazards analysis was used to define significant risk factors, taking years at risk into account (17). Each tumor end point was analyzed separately, and years at risk was specific to each end point. For example, individuals who developed thyroid cancer remained at risk for the other tumors until the end of their follow-up periods. For each risk factor (covariate) the regression coefficient and upper and lower confidence intervals were calculated, and a factor was considered significant if the range of the upper and lower 95% confidence intervals did not include 1.0. For both thyroid end points, dose, sex, and age at exposure were significant risk factors; for salivary tumors and hyperparathyroidism, dose was a significant risk factor, and no significant factors were found for the neural tumors. Risks are expressed as excess relative risk. Excess relative risk (ERR) and relative risk (RR) are related as follows: ERR = RR – 1.
The risk that an individual would develop each of the tumor end points was estimated as the cumulative hazard derived from the Cox analysis as follows. For each individual for each tumor, based on the number of years at risk, a basal hazard, with all covariates set at zero, was determined. Then the cumulative hazard was calculated, using the individual’s actual set of risk factors. As a check, we confirmed that the sum of the cumulative hazards for all individuals for each tumor equaled the number of the tumors observed.
Methods for defining familial factors
We are testing the hypothesis that familial effects influence the risk of development of radiation-induced head and neck neoplasms. If this were true, then the distribution of tumors within family pairs would not be accounted for by known risk factors. On the other hand, if familial factors do not influence the risk of development of these tumors, after taking into account years at risk, sex, age at exposure, and radiation dose, then the distribution of tumors would be independent of family pairs.
To test this hypothesis, for each individual we determined a residual for each tumor. The residual was defined as zero minus the cumulative hazard for subjects who did not develop the tumor and 1 minus the cumulative hazard for those who did. Each individual was then given a score equal to the sum of the residuals for the four tumors included in the analysis. Two sets of four end points were used: thyroid cancer or thyroid nodules plus the other three tumors (parathyroid, salivary, and neural). The presence of neoplasms in both siblings or the absence of neoplasms in both indicated concordance. The more similar the residuals in the siblings (in either the positive or the negative direction), the stronger the concordance. Therefore, the product of the two residuals for each pair, referred to as the concordance score, was used as a quantitative measure of concordance. Concordant pairs have positive scores (the product of two positive or two negative numbers), and discordant pairs have negative scores.
The population of the 251 sibling pairs was characterized by the sum of the concordance scores. This sum has the property that when it is positive it indicates an excess of concordance, and the greater the number of concordant pairs, the higher it is. The sum of concordance scores was used to test the null hypothesis that concordance occurred randomly. To do this, the 502 individuals were used to randomly reconstruct 251 sibling pairs and to calculate the sum of concordance for the hypothetical population. This random pairing was performed 103 times, and 103 sums of concordance scores, with an average value of zero, were obtained. The null hypothesis could be rejected at level 0.05 if the sum of concordance scores for the actual sibling pairs was greater than 95% of the sums of concordance scores for the randomly generated permutations. Given the number of sibling pairs and the observed prevalence of neoplasms, and assuming that the Cox modeling removes effects due to sex, dose, and age at exposure, there is a greater than 80% power to detect an attributable risk of 4% (correlation coefficient: r = 0.2; r2= 0.04) using either thyroid cancer or thyroid nodules.
As a check on this methodology, familial effects were also assessed using interclass correlation of the residuals (18).
Analysis of thyroid cancer behavior with respect to sibling concordance
We used the following analytical methods to determine whether thyroid cancers in concordant sibling pairs differed from thyroid cancers among patients whose sibling had not developed a radiation-related tumor. The thyroid cancers were divided into two groups, defined by the status of the sibling, and were compared for presenting features by t test, for categorical variables by 
2 analysis, and for recurrence by comparing the Kaplan-Meier plots using the log-rank and Breslow tests (SPSS, Inc., Chicago, IL).
|
Results |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Analysis of the distribution of neoplasms in families
Tables 1 and 2 show the distributions of neoplasms in the sibling pairs and the observed frequency of concordance. In Table 1, cancer is the thyroid-related end point, whereas in Table 2, nodules, including cancer, as defined in Subjects and Methods, is the thyroid-related end point. The other three end points (hyperparathyroidism, benign and malignant salivary neoplasms, and neural tumors) are the same in both tables. As a preliminary analysis of concordance, we compared observed vs. expected using simple binomial probabilities to determine the latter for all specific tumor combinations. The expected values shown in the tables, therefore, depend on the proportion of different tumors in the cohort, but not the presence of known risk factors or the length of follow-up. With thyroid cancer as the end point, the number of pairs concordant for no tumors and the pairs concordant for tumors were similar to those expected (152 vs. 146.36 and 13 vs. 13.91, respectively). With thyroid nodules, including cancer, as the end point, there was an apparent excess of sibling concordance for the absence of tumors (87 vs. 77.35). There was no apparent excess of concordance for both siblings having tumors (50 vs. 49.22), except when one or both had multiple tumors (11 vs. 7.19).
|
|
and 3 show the results of the analyses taking the length of follow-up and risk factors into account. In Fig. 2, cancer is the thyroid-related end point, whereas in Fig. 3, nodules, including cancer, is the thyroid-related end point. The top panels of these figures show the individual residuals, the difference between the observed and the expected number of tumors for each of the 502 individuals comprising the sibling pairs. The middle panels of these figures show the products of the sibling pair residuals, the measure of concordance in each of the 251 sibling groups. The bottom panels show the sum of products (the population concordance score) of actual sibling pairs in relation to the 103 concordance scores generated by random permutations and the fitted normal curve centered on zero, where zero is the distribution showing no concordance. The probability that the observed distribution is not due to chance (the level of significance) is shown by shaded areas under the curve. Using cancer as the thyroid-related end point, the null hypothesis could not be rejected (P = 0.32). Similarly, using thyroid nodules, including cancer, although the P value was borderline, the null hypothesis could not be rejected (P = 0.075).
|
|
). Familial variation accounts for only 2.4% of the total variance when thyroid cancer is one of the end points and for only 9.1% of the variance when thyroid nodules is used as one of the end points. The estimates of the significance of between-family significance agree very closely with the permutation method, as described above.
|
To determine whether the presence of a radiation-related neoplasm in a sibling influenced the characteristics or clinical behavior of the 83 thyroid cancers included in this study, we made two comparisons. For each comparison, we divided the cancers into two groups depending on whether the sibling had a tumor (was concordant) or not. In the first comparison, a concordant sibling was considered one with thyroid cancer, hyperparathyroidism, a salivary tumor, or a neural tumor. In the second comparison, a concordant sibling was considered one with a thyroid nodule, including cancer, hyperparathyroidism, a salivary tumor, or a neural tumor. In the first comparison there were 20 concordant thyroid cancers and 63 discordant ones, whereas in the second comparison there were 38 concordant and 45 discordant thyroid cancers (Table 4).
|
). However, there were no significant differences in latency, age at development, size, lymph node involvement, and multifocal or bilateral presentation of the cancers. Using Kaplan-Meier analysis, the disease-free survival times were compared for the thyroid cancer groups (Fig. 4). Although the concordant cancers tended to show a higher rate of early recurrence, the differences were not significant and diminished with time.
|
|
Discussion |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
Using these methods, we found that there was no excess concordance in sibling pairs; in other words, no evidence for familial risk factors. In addition to shared genetic factors, siblings would be exposed to common environmental factors during childhood and adolescence, as it is highly likely that they would have grown up in the same household. Thus, the findings do not support the existence of heritable or environmental factors that substantially alter radiation susceptibility in this group.
Could biases in the reporting of data to the study or in the intensity of medical attention received by the participant account for the findings? Although these factors cannot be discounted, it is expected that they would account for increased, rather than decreased, concordance. If one sibling develops a radiation-related neoplasm, it would be more likely that the other sibling would have a heightened awareness of the risk and would seek medical attention. Also, when one sibling provides information to the study, it facilitates obtaining information from the other sibling, again increasing the chance of observing concordance. Other factors could contribute to concordance. Radiation-associated thyroid cancer may be more common in Jews, and most of the individuals in this cohort are Jewish. However, our information about religion is incomplete, so an analysis taking this into account was not possible. Family history and shared environment could also contribute to concordance, but would be expected to increase, rather than decrease, it.
Using cancer as the thyroid-related end point had the advantage of being unambiguous. In these analyses, there was no evidence of a susceptibility effect, and the estimated (nonsignificant) risk ascribable to family factors was only 2.4% (intraclass correlation). In an earlier study, we found more support for concordance when thyroid nodules, including cancer, were considered as the end point (15). However, as in the initial observations more than 15 yr ago, we have recognized, with the use of ultrasound examinations, that about 90% of the subjects in this study have thyroid nodules (16). Therefore, we adopted the criteria described above to include only the larger and potentially clinically significant benign nodules. Although there is some support for excess concordance in these analyses, with an ascribable risk of 9.1%, the findings are not statistically significant. It should be noted that in addition to the large number of tumors that accumulated in the intervening 15-yr interval, the current analysis uses specific organ-dose exposure estimates that were not available for the earlier study.
There is good reason to expect that there are genetic radiation susceptibility factors (19). The syndrome of ataxia telangiectasia, caused by a recessive mutation in the ATM gene, includes a marked increase in radiation sensitivity. It is estimated that about 1% of the general population is heterozygous for ATM mutations. ATM plays a central role in the response to radiation damage, either promoting apoptosis or facilitating DNA repair. There is some evidence, based on a small number of cases, that heterozygous women are at increased risk for radiation-related breast cancer (20). Polymorphisms in the genes directly active in DNA repair, presumably producing small quantitative changes in function, are candidates for producing variations in radiation susceptibility (12). Evidence supporting this concept comes from epidemiological studies of patients who develop second cancers after treatment of a first one (21, 22). Information about cancer and other neoplasms in first degree relatives (especially parents and siblings) not exposed to radiation will be obtained in the future, as these may shed additional light on genetic susceptibility factors.
In considering radiation susceptibility, it is tempting to assume that exposed individuals who develop tumors are more susceptible than similarly exposed individuals who do not. However, if this were true, it would be expected that an individual who develops one tumor would be more likely to develop a second. We did not find this, as reported previously (14). Similarly, it is often assumed that susceptibility has a genetic component. If this were true, then concordance in exposed sibling groups would exceed what is predicted based on known risk factors. Again, we did not find this. Thus, it appears that although radiation has been a potent carcinogen in this group, it has acted rather randomly.
The findings show the difficulty in assigning a susceptibility phenotype and the associated difficulty of testing candidate susceptibility genes. Our hypothesis and analyses are related to general radiation susceptibility factors. It remains possible that there are organ-specific factors, but given the smaller numbers, these will be even more difficult to find. Even with pan-genomic methodology, the uncertainty about what phenotype represents increased susceptibility needs to be taken into account. The findings do not exclude subtle variations in susceptibility that could only be detected in larger cohorts, by case-control studies, or by multigeneration pedigree studies.
A secondary goal of the present study was to determine whether there were any differences in the presentation or clinical behavior of the thyroid cancers in patients with siblings affected by radiation-related neoplasms compared with patients whose siblings did not have radiation-related neoplasms. Among the presenting characteristics, age at surgery and latency (years from radiation exposure to surgery) are of interest because one might expect cancers in individuals with increased susceptibility to occur at younger ages and with shorter latency. This was not seen. With respect to behavior, the tendency for concordant cancer to recur more quickly was not significant. Therefore, there is no reason to expect a different outcome or to alter the management of thyroid cancer based on the tumor status of an irradiated sibling.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Received November 3, 2003.
Accepted February 12, 2004.
|
References |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Endocrinology | Endocrine Reviews | J. Clin. End. & Metab. |
| Molecular Endocrinology | Recent Prog. Horm. Res. | All Endocrine Journals |
