MRI: An explanation for the increasing incidence in childhood brain tumors?

The National Cancer Institute's SEER program first initiated population-based cancer surveillance across selected areas of the United States in 1973. From 1973 through 1994 incidence rates of primary malignant brain tumors in children increased by 35%. Brain cancer incidence rates in adults, specifically in the very elderly, increased even more substantially during that time period. There is considerable debate regarding the reasons for the trends. One concern is that exposure to environmental toxins leading to brain cancer development has increased in recent decades. A less ominous explanation is that improved diagnostic imaging technology has lead to artifactual increases in rates due to improved detection. 

Smith and several colleagues from NCI recently explored the latter conjecture in a statistical analysis of childhood brain cancer rates using the national SEER data [J Natl Cancer Inst 1998;90: 1269-77]. Rather than averaging yearly rate changes over the 1973-1994 time period using a model with a constant annual rate of increase (linear model), they accounted for a sharp increase in rates that occurred in 1984 and 1985 using a step function model. Smith and colleagues showed that the step function model provided a statistically significantly better fit to the incidence data than the linear model. From the step function model, they report that childhood brain cancer incidence was essentially stable before 1984, a jump in rates occurred in 1984-85, and a new baseline rate was established after 1985 with stable rates through 1994. In other words, the average annual increase in childhood brain cancer rates that was shown in previous reports using linear models was due primarily to the sharp rate increase that occurred in 1984-85. 

Smith's a priori hypothesis was that the introduction and dissemination of MRI technology in the mid-1980s resulted in earlier detection and reporting of pediatric brain cancer, thus largely explaining the observed brain cancer trends. In the Smith et al report, and in an accompanying editorial by W. Black [J Natl Cancer Inst 1998;90:1249-51], the evidence presented in support of the MRI-detection effect included the timing of MRI introduction; the rapid nature of the diffusion and application of MRI technology in the US; the fact that focal low grade tumors, especially of the brain stem and cerebrum for which MRI has considerable detection benefits over CT scans, substantially accounted for the increased rates; and that mortality rates did not mirror the increase in incidence rates. Changes in the mid-1980s in other diagnostic capabilities (e.g., stereotactic biopsy) and in case reporting may also account for some portion of the observed increase in incidence. 

COMMENT: Smith's carefully presented report, and Black's cogent additional explanation for how the results of the analysis are consistent with the MRI hypothesis, lends strong support to the contention that recent increases in pediatric brain malignancies are due to improved detection and reporting coincident with the advent of MRI in the mid-1980s. If they are correct, then some consolation can be taken that the increase in the reported incidence of childhood brain cancer in the U.S. over the past two decades did not represent increasing exposure to environmental or behavioral hazards. This consolation must be balanced, however, by our continuing lack of knowledge about factors (environmental and/or genetic) causally associated with the majority of brain cancers in children. 

The controversy related to childhood cancer trends points out an important issue: Interpretation of temporal trends is something of a tricky business. Any analysis of changes in incidence rates over time will be influenced by temporal changes in population characteristics, the accuracy of census estimates, screening practices, diagnostic technology, morphology classifications, and other practices related to case ascertainment. One or more of these factors can effectively conspire to show increasing incidence over time that is not reflective of more cancer, but rather of better case identification. Because of the relative rarity of childhood cancer, especially when stratified by site or histology, incidence rates are considerably more sensitive to changes in case identification than are most adult cancers. The difficulty in properly understanding the contribution of confounding factors when looking for true changes in disease occurrence, however, absolutely does not mean that we should stop surveillance efforts. Rather, the lesson serves to reinforce the need for a cautious and thoughtful approach to the interpretation of all data. As Dr. Black points out in his editorial, Smith's report should lead us to other questions that need to be answered in relation to brain cancer: "How many children have unrecognized disease, what is the full spectrum of the disease, how should the disease be managed, and what signs or symptoms should prompt an MRI examination in the first place?" I would add to his comments that we also need to continue research efforts toward identifying genetic factors related to susceptibility and identifying exogenous factors that increase risk for childhood brain cancer occurrence. James G. Gurney 

Don't bring home the bacon (dead or alive)!

As reported previously (C3, Vol 7, No 5), parental farming (in some studies, not all) has been positively associated with childhood brain tumors and leukemias. Holly et al [Cancer Epi Biomark Prev 7:77-802, 1998] explored farm-related exposures in a population-based study of childhood brain tumors (CBT). Subjects were recruited from 19 counties in the San Francisco, Los Angeles, and Seattle areas in a study that focused on exposure to N-nitroso compounds. A total of 540 cases (diagnosed with a benign or malignant brain tumor under the age of 20 between 1984 and 1991) and 801 frequency-matched controls participated. Data were collected on mother and child exposures to beef and dairy cattle, pigs, sheep, goats, poultry, and horses, as well as living or working on a farm. The exposures had to be at least one month in duration and occur either 1 month prior to or during the index pregnancy or between the birth and diagnosis of the index child. The authors found an elevated risk of CBTs associated with mothers' exposure to pigs and horses on a farm during the index pregnancy (OR=3.8; 95% CI=1.2-12; OR=2.2; 95%CI=1.0-4.8; respectively). Moreover, children diagnosed with PNETs (107 cases) showed an elevated risk associated with maternal or personal exposure to pigs (OR=4.0; 95% CI=1.2-13; OR=11.9; 95%CI=2.8-51, respectively). The authors suggest further studies using molecular markers or clinical measures of exposure to pesticides and animal viruses to support or refute their results. 

COMMENT: In delineating the potential limitations of their study, the authors raised the issue of recall and reporting bias among the participants (where memories could falter or mothers of cases could over or under-report exposures). The authors also raised the possibility of participation bias, and admitted that limited information was available on the non-participants in this study. However, the authors did not elaborate on how this lack of information could markedly affect the interpretation of their results. The control mothers in this study were notably more educated than the case mothers; 34% of control mothers had at least a college degree compared with 24% of case mothers. Moreover, 21% of case mothers did not graduate from high school compared with only 14% of control mothers. Of the 1079 eligible control mothers identified for this study by random digit dialing, only 801 (74%) agreed to participate. The analyses that examined pigs, chickens, horses, etc. did not adjust for education. It is entirely possible that a proportion of the control parents who farm (possibly of lower education and income) refused to participate in this study. This would lead to artificially inflated positive associations with farming. In the authors previous report on this same study [CEBP, 5:599-606, 1996], they found an increased risk of CBTs associated with increasing consumption of cured meats (i.e., the nitrosamine hypothesis), which has been supported by other studies . It is also possible that children who are raised on farms tend to eat more meats. Finally, the authors quote two supporting studies in their abstract: one was essentially a record-linking study with no information about individual exposures ((Kristensen et al; Int J Cancer 65:39-50, 1996; see C3 Vol 7, no 5); the other study was an exploratory analysis of a multitude of several potential risk factors for childhood astrocytic gliomas and PNETs (Bunin et al Cancer Causes Control 5:177-187, 1994)). The study reported by Holly et al should be considered as an exploratory analysis, with the potential limitations clearly delineated. It is entirely possible that exposure to pigs, etc., could be associated with an increased risk of CBTs. However, before parents start blaming the farm animals, a more rigorous and detailed study (with an equal likelihood of farmers participating from both case and control populations) is necessary. Julie A. Ross

Childhood leukemia in Taiwan - Moving up to the big leagues?

There have been several studies conducted (and hypotheses formulated) about the peak in incidence that occurs in childhood ALL between the ages of two and five. Factors associated with modernization have been implicated, especially since the age peak is more apparent in industrialized and economically affluent populations. Since Taiwan has undergone substantial economic development in the 1980's, Li et al [Int J Epidemiology, 27:587-91, 1998] explored childhood leukemia incidence with respect to urbanization during this period. Between 1981 and 1990, there was a statistically significant increase (approximately 20%) in the incidence of childhood leukemia. When compared to rural areas, metropolitan (urban) areas had a significantly higher incidence rate, which was most notable for children under the age of 5. There was also a dose-response increase in incidence associated with increasing level of urbanization (rural, town, city, metropolitan).The authors suggest that these increases may be due to population mobility from rural areas into more metropolitan areas and could provide support for an infectious etiology. 

COMMENT: Despite the 20% increase in childhood leukemia incidence during this ten year period in Taiwan, it is important to note that the overall rates of childhood leukemia are still considerably lower when compared to the United States (28/million vs approx 42/million, respectively). Kinlen [Lancet 1988; 2:1323-7] has suggested that childhood leukemia may be a rare response to direct exposure to an infectious agent, which is most apparent when herd immunity is deregulated by population mixing. While this ecologic study may provide modest support for this theory, it cannot directly test this hypothesis. Moreover, although the authors described the distribution of leukemia by age and type in their study (ALL, AML, CL), they did not perform their analysis separately for the ALL cases, which comprise the age peak of interest for the infection theories. Julie A. Ross 

C3 Quarterly Newsletter
Children's Cancer Research Fund
Epidemiology Research Unit
Division of Pediatric Epidemiology
Clinical Research
University of Minnesota
420 Delaware St. SE, Box 422
Minneapolis, MN 55455
pedsepi@umn.edu

Editors: 
Stella M. Davies, MD, PhD, and Julie A. Ross, PhD