Total Recall? Hasta la vista, baby! 

Recall bias, or the differential recall of events based on case or control status, is very much a concern in epidemiology studies. In case-control studies of childhood cancer, parents of children diagnosed with cancer may actively search for causes, or they may have a preconceived idea of whether specific events in the past may have led to the development of their child's cancer. One advantage, however, of conducting childhood cancer epidemiology studies is that the time period from a putative exposure and disease development is relatively short compared with adult malignancies. Nevertheless, it is critical for epidemiologists to explore ways to both measure and minimize the potential for recall bias. In a recent study from Quebec by Infante-Rivard et al [Am J Epidemiology, 152:480-6, 2000], two validation sub-studies were conducted within a case-control study of childhood ALL. Briefly, the overall case-control study consisted of 491 newly diagnosed cases of ALL diagnosed less than 9 years of age, and two sets of controls (healthy population controls and hospital controls with severe hematological disorders) matched on age, sex, and region of residence. The authors chose to assess the validity of two variables: reported distance of the home from a power line (a surrogate for exposure to EMF, a putative risk factor for childhood ALL) and reported number of x-rays during pregnancy (a recognized risk factor for ALL). For the distance to power lines sub-study, the measured distance as determined by a transportation and energy network map was compared to the parent's response to the following question: Within a radius of 1 km of your house, was there a high voltage power line? Only residents of Montreal Island were included in this sub-study. One caveat unrelated to the case-control study occurred during this time period: residents from one of the geographical areas of Montreal Island were concerned about a perceived excess of childhood ALL potentially attributable to the proximity to power lines. Because of this concern, cases from this area of Montreal Island were considered separately in the analysis and referred to as 'geographical area' (GA) cases. The authors compared the sensitivity and specificity of the reported measures of distance to power lines for the 'geographical area cases' (n=32), the other cases on Montreal Island (n=95), the population controls (n=110) and the hospital controls (n=122). With the exception of the GA cases, where there was higher sensitivity (62%) and lower specificity (55%), the other three groups demonstrated similar sensitivities (35% for cases, 36% for hospital controls, and 22% for healthy controls) and high specificities (91%, 89%, 90%, respectively). For maternal radiographic examinations, where self-report was compared to what was in the medical record, the sensitivity for cases was similar to population controls (65% and 71%, respectively), although it was less comparable to hospital controls (50%). These data suggest that while sensitivity may be low, parental recall is probably most often non-differential (with the exception of when there is current public concern). 

COMMENT: This is an important study for several reasons. First, it provides evidence (which until now was mostly intuitive) that current events can influence parental reporting. In this instance, in an area where there was concern about power lines contributing to an excess of leukemia, there was differential reporting of the distance to power lines compared to an area where there was no public concern. Second, there was evidence of non-differential reporting in the absence of public concern, which suggests a less influential role for recall bias. However, the overall low sensitivity reported here should be of concern to epidemiologists as it means we are not adequately measuring 'the truth' for certain exposures. For example, if the distance to power lines was etiologically relevant, a case-control analysis of this study's data would produce an odds ratio biased toward the null. It should be noted, however, that the validation of recall of some past events including birth weight and gestational age has been quite good [Olson et al, Am J Epidemiol 145;58-67, 1997]. Furthermore, analytical methods that address hypothesized etiologic differences among cases can be used in an attempt to address recall bias. If, for example, a putative risk factor is associated with childhood AML and not ALL, it provides more credence to a true association, since it is highly unlikely that a mother of a child with AML would recall her exposures differently than a mother of a child with ALL. Clearly, further methodologic investigations of exposure validation are needed. Third, the selection of appropriate control groups (in this case, hospital-based controls) may help in increasing comparable accuracy of recall, since parents of affected children (both cases and controls) would perhaps be thinking more about their exposures. However, this study demonstrated that depending on the exposure, the sensitivity of recall was different for the hospital-based controls. Thus, challenges for epidemiologists include finding ways to improve methods of exposure assessment as well as to identify appropriate control groups. One last note about this specific study: the selection of hospital-based controls with 'severe hematological disorders' may not be the best hospital-based comparison group for children with leukemia as both groups may share similar risk factors. Nonetheless, further studies of this type are needed to determine the magnitude of these problems. 
- Julie A. Ross 

A mole Down Under? 

The association of multiple moles (melanocytic nevi) with melanoma is well known, but little attention has been paid to their natural history in the very young child. Harrison et al [JNCI 92: 1436-1438, 2000] compared rates of development of nevi from birth to 3 years of age in two cohorts of Caucasian children of similar ethnicity from the contrasting climates of Townsville, Australia (high levels of ambient UV light, n=115) and Glasgow, UK (low levels of ambient UV light, n=157). The proportion of Australian children with nevi (counted in a standardized manner) increased rapidly in the first 2 years of life from 2.3% at birth to 65.2% at 12 months and 100% at 24 months. Corresponding proportions for the Scottish cohort were similar at age 6 weeks (3.2%; p=0.781) but they were considerably less at 12, 24 , and 36 months of age, when 30.5% (p=0.0001), 61.7% (p= 0.0001) and 83.6% (p=0.0002) of children, respectively, had nevi. Median total nevus counts were considerably higher in Australian subjects than in Scottish subjects, but did not differ by sex in either cohort. 

COMMENT: The incidence of melanoma is much higher in Australia than in the UK, and previous studies have shown an increased frequency of nevi in young Australians compared with their British counterparts. Despite this, there is little difference in nevus frequency between older Caucasian individuals from contrasting climates [Bataille et al Br. J Cancer, 77:505-10, 1998; Crijns et al, Melanoma Res, 7:407-416, 1997]. The incidence of melanoma appears to be highest in populations who develop nevi early in life, and the authors of this study postulate that early nevi may be a risk factor for melanoma. In agreement with this, 44% of melanomas diagnosed in Scottish subjects aged less than 30 years of age develop in small, early onset nevi [MacKie et al, Br. J Dermatol. 124:560-564, 1991]. This study identifies an interesting subset of healthy children- those 2-3% with nevi at birth. It would be fascinating to generate a cohort of these children and follow their melanoma experience late in life. This would require a great deal of patience and foresight on the part of investigators and funding agencies. 
- Stella M. Davies

Infectious Mononucleosis and Hodgkin's Disease: kissing cousins? 

Data regarding the association of infection with Epstein-Barr virus (EBV) and Hodgkin's Disease (HD) are conflicting, with some support from cohort studies, but less coming from case-control studies. In a cohort study of impressive magnitude, Hjalgrim et al [JNCI 92:1522-8, 2000] have followed 38,562 infections (determined by a positive Paul Bunnell test) between 1964 and 1995 in Denmark and Sweden. A total of 1,381 cancers were observed during 689,619 person years of follow-up (SIR=1.03, 95% CI 0.98-1.09). A statistically significant excess of HD was noted (SIR 2.55, 95% CI 1.87-3.40), together with an excess of skin cancers (melanoma and non-melanoma both elevated) (SIR 1.27, 95% CI 1.13-1.43). The SIR for HD remained elevated for up to two decades after the occurrence of infectious mononucleosis, but decreased with time since the initial diagnosis of infectious mononucleosis (p for trend, <0.001). The SIR for HD tended to increase with age at diagnosis of infectious mononucleosis (p for trend, 0.05). Following infectious mononucleosis, the SIR for HD at ages 15-34 years was 3.46 (95% CI 2.46-4.81), which was statistically significantly higher than for any other age group. The SIR for lung cancer was reduced (SIR 0.71, 95% CI 0.58-0.86). 

COMMENT: This is the largest and the longest cohort study looking at infectious mononucleosis and HD, and demonstrates the power of population-based disease registries. The observation that the highest relative risk occurred in young adults is surprising, as pathological evidence of EBV in tumor tissue is most frequently found in young children and older adults with HD. However, direct studies of the tumors in this cohort study are not available, and they may represent a subgroup associated with higher frequencies of detectable EBV products. Alternatively, the association with EBV may be indirect. The reason for the reduction in risk of lung cancer is unclear, although it may reflect the known association between higher socioeconomic status (associated with a lower smoking prevalence) and infectious mononucleosis [Evans AE, Viral Infections of Humans, Epidemiology and Control, New York, Plenum Medical Book Co, 1989]. 
- Stella M. Davies

Briefly noted: imprinting in blood cells 

Genomic imprinting is defined as parent-of-origin specific gene expression. For example, the insulin-like growth factor-2 gene (IGF2) is, in most circumstances, only expressed from the copy inherited from one's father. The copy inherited from one's mother is silent. Relaxation of genomic imprinting (expression of both the maternal and paternal copies) of IGF2 is a common feature of pediatric embryonal neoplasms such as Wilms' tumor and rhabdomyosarcoma. In a different twist to the imprinting story, Morison et al [Blood 96:3023-3028, 2000] show that IGF2 is expressed biallelically (ie without imprinting) in human bone marrow, but is imprinted (expressed only from the paternal allele) in peripheral blood. This is in agreement with previous studies showing biallelic expression of IGF2 in AML [Wu et al, Biochem Biophys Res Comm 231:466-472, 1997]. These data suggest that IGF2 is an important growth factor in hematopoietic development, and that the imprint (or its recognition) differs at different stages of cell maturation. Further studies of different hematopoietic cell lineages would be fascinating. 
- Stella M. Davies

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