Skip Navigation


American Journal of Epidemiology Advance Access originally published online on November 14, 2008
American Journal of Epidemiology 2009 169(3):257-266; doi:10.1093/aje/kwn363
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
169/3/257    most recent
kwn363v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Related articles in Am. J. Epidemiol.
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (3)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Parr, C. L.
Right arrow Articles by Veierød, M. B.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Parr, C. L.
Right arrow Articles by Veierød, M. B.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

American Journal of Epidemiology © The Author 2008. Published by the Johns Hopkins Bloomberg School of Public Health. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org.

PRACTICE OF EPIDEMIOLOGY

Recall Bias in Melanoma Risk Factors and Measurement Error Effects: A Nested Case-Control Study Within the Norwegian Women and Cancer Study

Christine L. Parr, Anette Hjartåker, Petter Laake, Eiliv Lund and Marit B. Veierød

Correspondence to Dr. Christine L. Parr, Department of Biostatistics, University of Oslo, P.O. Box 1122 Blindern, N-0317 Oslo, Norway (e-mail: c.l.parr{at}medisin.uio.no).

Received for publication October 16, 2007. Accepted for publication July 21, 2008.


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
Case-control studies of melanoma have the potential for recall bias after much public information about the relation with ultraviolet radiation. Recall bias has been investigated in few studies and only for some risk factors. A nested case-control study of recall bias was conducted in 2004 within the Norwegian Women and Cancer Study: 208 melanoma cases and 2,080 matched controls were invited. Data were analyzed for 162 cases (response, 78%) and 1,242 controls (response, 77%). Questionnaire responses to several host factors and ultraviolet exposures collected at enrollment in 1991–1997 and in 2004 were compared stratified on case-control status. Shifts in responses were observed among both cases and controls, but a shift in cases was observed only for skin color after chronic sun exposure, and a larger shift in cases was observed for nevi. Weighted kappa was lower for cases than for controls for most age intervals of sunburn, sunbathing vacations, and solarium use. Differences in odds ratio estimates of melanoma based on prospective and retrospective measurements indicate measurement error that is difficult to characterize. The authors conclude that indications of recall bias were found in this sample of Norwegian women, but that the results were inconsistent for the different exposures.

bias (epidemiology); case-control studies; cohort studies; epidemiologic measurements; melanoma; questionnaires; reproducibility of results; risk factors

Abbreviations: NOWAC, Norwegian Women and Cancer Study


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
The evidence implicating sun exposure and pigmentation factors in the etiology of cutaneous malignant melanoma (hereafter called melanoma) derives largely from case-control or nested case-control studies that assess risk factors retrospectively in people already diagnosed with melanoma (1–4). The small number of cohort studies includes the Women's Lifestyle and Health Cohort Study from Norway and Sweden (5) and a combined analysis of the Nurses’ Health Study and the Health Professionals Follow-up Study (6).

Melanoma risk factors are, like most exposure variables, prone to misclassification, which consequently may give biased risk estimates. Case-control studies of melanoma have the potential for recall bias, especially after much public information about the relation with ultraviolet radiation (7). Recall bias occurs when recall of prior exposures is misclassified differentially for those with and without disease (8). Differential errors may increase or decrease any true association, but the magnitude and direction of the bias in the estimates are usually difficult to predict (9).

Recall bias in relation to melanoma has been investigated in only a few studies and for a limited number of risk factors (4, 10–14). Several studies have assessed the reproducibility of melanoma risk factors separately for cases and controls (15), but to estimate recall bias the first measurement should ideally precede the development of melanoma (16). To our knowledge, only 2 previous studies have compared exposure data from the same individuals before and after a melanoma diagnosis. Both were nested case-control studies within the Nurses’ Health Study cohort (4, 12). Alternative approaches used to study recall bias in melanoma risk factors include comparison of confirmed and suspected cases (10, 11), cases and disease-free co-twins (13), and comparison of self-reported confidence ratings for cases and controls (14).

The purpose of the present methodological study was to study recall bias in a number of self-reported melanoma risk factors, including susceptibility host factors and history of sunburn, sunbathing vacations, and solarium use. Solarium implies artificial tanning and is commonly understood as a tanning bed. For some of the risk factors, including eye color, as well as history of sunburn, sunbathing vacations, and solarium use, we have not found previous studies of recall bias in the literature. Using a nested case-control design within the Norwegian Women and Cancer Study (NOWAC) cohort, we are able to compare questionnaire responses collected prospectively (at enrollment before a melanoma diagnosis) and retrospectively (after some women have developed melanoma). We also investigate if recall among the melanoma cases is affected by time since diagnosis. Finally, we compare odds ratio estimates of melanoma based on the prospective and retrospective exposure measurements to study the overall effect of differential and nondifferential measurement errors in data from cases and controls.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 References
 
Study design
NOWAC is a national population-based cohort study designed primarily to study risk factors for cancer. From 1991 to 1997, a total of 179,388 women aged 30–70 years were drawn at random from the national population register at Statistics Norway and invited to participate. Of the invited women, 102,433 (57%) were enrolled in the cohort, which has previously been described (17). The cohort has later been expanded (18), and updated information about NOWAC can be found on the website http://uit.no/kk/nowac/. Exposure data are collected by postal questionnaires, with prepaid envelopes and 1 or 2 reminders. Follow-up information on cancer is obtained from the Cancer Registry of Norway, and vital status (alive, dead, or migrated) and contact information are obtained from the population register at Statistics Norway. All linkages to national registers are done via a unique 11-digit national person number incorporating birth date and sex. NOWAC has been approved by the Regional Ethics Committee for Medical Research, Northern Norway. All data are stored and handled according to permissions issued by the national Data Inspectorate.

The present case-control study was nested within the NOWAC cohort. To study recall bias, we compared the answers to questions on melanoma risk factors collected at enrollment in 1991–1992 (age range, 34–49 years) or 1996–1997 (age range, 33–69 years) with the answers in a second questionnaire administered in April 2004 after some women had developed melanoma during 6–13 years of follow-up. The second data collection was conducted as a regular update of exposure information. A standard letter of invitation, explaining the need for an update as many lifestyle factors change with age, was mailed to 208 cases and 2,080 controls (case:control ratio = 1:10) together with a full-length NOWAC questionnaire (8 pages), including 1 page on melanoma risk factors. The questionnaire requested written, informed consent to participate in the study. There was no particular emphasis on melanoma, and subjects were not made aware of their case-control status or that recall bias would be studied, as this could influence the responses.

Study sample
Initially, 12% (n = 12,556) of the women enrolled in the NOWAC cohort were excluded because of inclusion in subcohorts without enrollment data on melanoma risk factors (n = 2,704), death (n = 2,138), migration (n = 689), unknown vital status (n = 6), withdrawal from the study (n = 21), participation in other methodological substudies (n = 499), melanoma diagnosed before enrollment (n = 470), or lack of informed consent to further contact (n = 6,029). Among the 89,877 women available for the present study, there were 208 women diagnosed with primary melanoma after enrollment (incident cases) and 89,669 noncases. All cases were histologically confirmed to be invasive melanoma. Ten controls per case were selected randomly from the noncases, matched on birth year and subcohort (proxy for same time of enrollment and questionnaire version). The number of controls was selected on the basis of calculations of statistical power, which increased up to 8 controls per case. Ten controls were selected to correct for an expected response of about 80%.

The response proportion for the nested case-control study was 77% (n = 1,759/2,288) overall, 78% (n = 163/208) for cases, and 77% (n = 1,596/2,080) for controls. Because of a lag in the cancer registration process, the cohort data were linked again with follow-up information through December 31, 2004, to verify the case-control status before the analysis. At the last linkage, 1 case had been reclassified as not having melanoma and was therefore excluded. Altogether, we excluded 354 controls: 8 matched to the excluded case, 337 matched to nonresponding cases, 7 without answers to the questions on melanoma risk factors in the prospective or retrospective questionnaire, and 2 diagnosed with melanoma after the return date of the retrospective questionnaire but within the same year. No cases were excluded because of the omission of all questions on melanoma risk factors. Thus, the final study sample included 162 cases and 1,242 controls. Seventeen of the cases were diagnosed with melanoma within the first year of follow-up.

Melanoma risk factors
The retrospective questionnaire in 2004 repeated some of the core questions about melanoma risk factors at enrollment. Recall bias was studied in 8 questions with identical wording in both questionnaires. The translated questions and precoded response options are given in Table 1. The susceptibility host factors studied included eye color, hair color, skin color after acute and chronic sun exposure, and the number of asymmetric nevi >5 mm on both legs. The number of nevi on the lower limbs or thighs has been found to correlate strongly with total body counts of nevi in women (19, 20). Scores are provided in Table 1 for eye, hair, and skin color variables in order to interpret the direction of shifts in responses. Ultraviolet exposures included history of sunburn, sunbathing vacations, and solarium use. In the question about sunbathing vacations (Table 1), "southern latitudes" imply various destinations but typically southern Europe (e.g., Spain or Greece). Separate answers were given for the age intervals <10, 10–19, 20–29, 30–39, and 40–49 years. Analyses of the interval 30–39 years were restricted to subjects aged ≥39 years at enrollment. Because the age interval 40–49 years was included only for those enrolled in 1991–1992 and few subjects were enrolled at 49 years, the analyses were restricted to subjects aged ≥40 years.


View this table:
[in this window]
[in a new window]

  		Table 1. Melanoma Risk Factors From the Norwegian Women and Cancer Study Questionnaire Included in the 2004 Recall Bias Study

 
Statistical analysis
Start of follow-up was defined as the date of registration for the incoming prospective (enrollment) questionnaire. The time between questionnaires was calculated from the start of follow-up to the date the retrospective questionnaire was mailed (April 13, 2004). The time since diagnosis was calculated from the date of diagnosis to April 13, 2004. Data on height, weight, education, and household income were taken from the prospective questionnaire. Body surface area was calculated according to the formula: weight (kg)0.425 x height (cm)0.725 x 71.84 (21).

All exposure variables were categorical with 3–7 response options (Table 1). The agreement between responses in the prospective (first) and retrospective (second) questionnaires and shifts in inconsistent (referred to as misclassified) responses were assessed from contingency tables stratified on case-control status. We present the proportion of women on the diagonal (absolute agreement) as "PA" and on the diagonal ± 1 category as "PA±1." As a summary measure of agreement, we estimated the weighted kappa coefficient ({kappa}w) using Cicchetti-Allison weights.

Our assessment of recall bias relies on the demonstration of a shift in self-reported exposure status associated with the diagnosis of melanoma similar to the method of Weinstock et al. (12). However, in the present study, shifts were calculated as the proportion of women with the first response lower than the second response among those with misclassified responses, rather than as the mean change in questionnaire scores (12). We also include differences in agreement (PA and {kappa}w) in our definition of recall bias. A 95% confidence interval was calculated for {kappa}w and the shift proportions for cases and controls. Confidence intervals for shifts were calculated for each variable by including the number of participants with misclassified responses (n = 7–68 for cases) using the Wilson score method, which has been shown to perform better than the exact or Wald (asymptotic) confidence intervals for small sample sizes (22). Shifts were termed borderline significant when the lower or upper confidence interval limit was 50%.

To investigate the effect of time since diagnosis on recall, we stratified measures of agreement and shifts for the melanoma cases (n = 162) on years since diagnosis using 2 (1–6, >6–13) and 4 (1–3, >3–6, >6–9, >9–13) categories. All age-specific variables were categorized into exposed versus nonexposed, and the number of nevi was categorized into null versus ≥1. For the dichotomized responses, we calculated a simple kappa coefficient ({kappa}) with 95% confidence interval and tested for equal {kappa} across time intervals.

To illustrate the potential effect of both differential and nondifferential measurement errors, we estimated odds ratios of melanoma on the basis of prospectively and retrospectively collected exposure data using conditional logistic regression with control for the matching variables. Each case had a minimum of 2 matched controls. We also performed analysis with additional control for the same covariates as in the prospective analysis of melanoma risk in the Norwegian-Swedish Women's Lifestyle and Health Cohort Study with about 50% NOWAC women (5): All melanoma risk factors were adjusted for region of residence (southern, middle, or northern Norway); ultraviolet exposure variables were also adjusted for hair color (measure of sun sensitivity), and solarium use was additionally adjusted for sunburn and sunbathing vacations in the respective age groups.

Because of small numbers, we combined the 2 upper response categories for skin color after acute and chronic sun exposure and the 3 upper categories for the age-specific variables related to sunburn, sunbathing vacations, and solarium use. No cases were exposed to a solarium before 10 years, and exposure in the age intervals 10–19 and 20–29 years was dichotomized into never versus ever exposed, which was also done to sunburn at 40–49 years. Nevi were grouped into 0, 1, 2–6, and ≥7. We tested for trend across categories of variables by treating the equally spaced response categories (e.g., 0, 1, 2, 3) as continuous variables in the logistic regression analysis.

The main statistical analyses were performed by using SAS, version 9.1.3, software (23). The Wilson score confidence intervals for the shift proportions were calculated in CIA [Confidence Interval Analysis], version 2.1.2, software (24). All P values are 2 sided, and a 5% significance level was used.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 References
 
Table 2 shows that the selected background characteristics of participating cases (n = 162) and controls (n = 1,242) were similar. The mean level of education was 12 years. The mean time between questionnaire administrations was 11 years (range, 6–13 years). For melanoma cases, the mean time since diagnosis at the second questionnaire dispatch was 6.3 years. The most frequent melanoma sites were lower limb (51%), followed by trunk (26%), and the most common histologic type was superficial spreading melanoma (67%).


View this table:
[in this window]
[in a new window]

  		Table 2. Selected Background Characteristics of Cases and Controls in the 2004 Study of Recall Bias in Melanoma Risk Factors Within the Norwegian Women and Cancer Study, 1991–2004

 
The responses to melanoma risk factors in the first and second questionnaires are compared in Table 3 stratified on case-control status. In summary, PA was somewhat lower for cases, with a median of 60% (range, 42 to 100) compared with controls with a median of 65% (range, 54 to 99). This result was reflected in {kappa}w with a case median of 0.40 (range, –0.01 to 0.89) and a control median of 0.46 (range, 0.10 to 0.87). {kappa}w was lower for cases than for controls for all age-specific exposures, except solarium use at age 30–39 years.


View this table:
[in this window]
[in a new window]

  		Table 3. Measures of Agreement and Shifts in Responses for Self-reported Melanoma Risk Factors Stratified on Cases (n = 162) and Controls (n = 1,242) in the 2004 Study of Recall Bias Within the Norwegian Women and Cancer Study, 1991–2004

 
Shift proportions >50% (Table 3) indicate reports of a lighter eye, hair, or skin color, more nevi, or more sun exposure in the second questionnaire. Among the host factors, a significant shift in cases only was observed for skin color after chronic sun exposure (shift proportions of 67% and 49% in cases and controls, respectively). For hair color, there appeared to be a shift in the case responses (shift proportion of 62% compared with 52% in controls), but the shift was not significant. For nevi, the shift proportion was larger in cases (71%) than in controls (57%) and significant in cases and borderline significant in controls. For skin color after acute sun exposure, there was a similar shift in cases (38%) and controls (35%) toward reporting a darker skin color, and the shift was borderline significant in cases and significant in controls.

For history of sunburn, sunbathing vacations, and solarium use, there was little difference in the shift proportions between cases and controls, but the direction of the shift seemed to change with age. For solarium use, the shift proportion and {kappa}w could not be calculated for cases at age <10 years, as PA was 100%. {kappa}w for controls at age <10 years and cases and controls at 10–19 years should be interpreted with caution because of PA values of 95%–99% (refer to Discussion on the properties of {kappa}w).

When agreement (PA and {kappa}w) and shift proportions (with 95% confidence intervals) were analyzed for all 162 cases stratified on time since diagnosis, no apparent trend was found. Exclusion of the 17 cases diagnosed with melanoma within the first year of follow-up had a negligible effect on the results for cases in Table 3.

Odds ratio estimates of melanoma were compared for exposure information from the prospective and retrospective questionnaires in analyses controlled for the matching variables (Table 4). The same risk factors were generally identified in both questionnaires: A lighter or red hair color, more skin redness after acute sun exposure, less tanning after chronic sun exposure, more nevi, and more sunburns before age 30 years were associated with significantly increased risk (Ptrend < 0.01) but with differences in magnitude. For risk factors where the estimated shift proportion (Table 3) appeared to be larger in cases (i.e., hair color, skin color after chronic sun exposure, and nevi), the odds ratio estimates for the retrospective data were consistently accentuated (higher) with a stronger trend. Among all risk factors, accentuation and attenuation, as well as a change in direction, were observed for the retrospective odds ratio estimates. These observed effects also differed between categories of the same variable, indicating measurement error which is difficult to characterize. Odds ratios could not be estimated for solarium use at age <10 years as no cases were exposed, but 6 and 7 controls reported to be ever exposed in the prospective and retrospective questionnaires, respectively. Exclusion of the 17 cases diagnosed with melanoma within the first year of follow-up and their matched controls did not change the significance of the trend estimates (Table 4).


View this table:
[in this window]
[in a new window]

  		Table 4. Odds Ratio Estimatesa of Melanoma According to Risk Factors Reported Prospectively and Retrospectively in the 2004 Study of Recall Bias in 162 Cases and 1,242 Controls From the Norwegian Women and Cancer Study, 1991–2004

 
When additional adjustments for relevant covariates were performed (refer to Statistical analysis), the same risk factors were identified (Ptrend < 0.01 in both the prospective and retrospective data). The difference between the prospective and retrospective odds ratios (accentuation, attenuation, or change in direction) did not change for most risk factors (data not shown). As measurement error in covariates may affect risk estimates (25), we present the conditional odds ratio estimates without these further adjustments. The matching variables can be assumed to be without error as birth year is from the national population register, and subcohort is a study variable.


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
References
 
In this study of recall bias in melanoma risk factors, we observed significant shifts in the responses of both cases and controls, but a shift in cases was observed only for skin color after chronic sun exposure, and a larger shift in cases was observed for nevi. {kappa}w was lower for cases than for controls for most age intervals of sunburn, sunbathing vacations, and solarium use. We therefore conclude that indications of recall bias were observed in this study sample of Norwegian women.

To our knowledge, only 2 other studies of recall bias in melanoma risk factors have a nested case-control design. The first study found recall bias in the ability to tan but not in hair color (12). The second study found no substantial recall bias in the tendency to tan or in hair color (4). For the tendency to burn, recall bias was observed for cases with squamous and basal cell carcinoma but not with malignant melanoma (4). A Canadian case-control study of sunbed/sunlamp usage and fluorescent light exposure assessed recall bias by comparing responses from confirmed and suspected cases (i.e., unconfirmed referral cases) of melanoma (10, 11). The authors found no evidence of recall bias in the exposure variables. A case-control study based on the International Twin Study indicated recall bias in sunbathing in childhood and adulthood, mole frequency, and freckling in childhood, but ease of burning and tanning appeared unbiased (13). A Canadian study, using confidence ratings to indirectly measure recall bias in time spent outdoors, reported similar ratings for cases and controls (14). Unlike the other nested case-control studies, the present study suggested recall bias in hair color, but the shift was not significant.

The effect of recall bias on the odds ratio estimates of melanoma may not be very pronounced in the present study, as the same risk factors were identified in the prospectively and retrospectively collected exposure data. However, the study was designed to minimize the potential for recall bias with no particular emphasis on melanoma, which could influence the questionnaire responses. The risk factors were included in a larger questionnaire, and the study subjects were not aware of the study objective. NOWAC is a population-based cohort with good external validity (17), and both cases and controls came from this population. In typical case-control studies, the problem of recall bias may be more serious, and risk estimates may also be affected by recruitment bias. A large European case-control study of melanoma found no association with sunbed use or some established risk factors (26). The authors point to indications of recall bias and recruitment bias as possible explanations (7).

A comparison of odds ratios ignores the paired nature of the observations, as pointed out by others (12), and captures not only recall bias but the overall effect of differential and nondifferential measurement errors. Many factors that affect recall apply equally to cases and controls, for example, time interval since exposure and the degree of detail required. Therefore, data from case-control studies will generally also have nondifferential measurement errors. In the present study, we observed a shift in both cases and controls toward reporting a darker skin color after chronic sun exposure. A possible explanation is that the responses were influenced by recent tanning, as the retrospective questionnaire was mailed shortly after Easter, a time of recreation and the combined effects of sun and snow.

In the present study, only questions with identical wording were compared, and the historical questions referred to the same time periods. However, for sunbathing vacations at the age of 40–49 years, the shift toward more exposure in the second questionnaire may not indicate misclassification. Many women could not report exposure for the entire age interval in the first questionnaire, as the analysis included subjects aged <49 years at enrollment. Further, some pigmentation factors could change with age, for example, hair color and the number of nevi. Thus, shifts may not be attributed to misclassification alone. Whereas changes in hair color should occur equally in cases and controls, nevi can be precursors to melanoma. Thus, the larger shift among cases toward more nevi in the second questionnaire may in part be biologic.

Cases were analyzed after a maximum follow-up time of 13 years, but we did not find that time since diagnosis affected recall in any systematic manner. Inclusion of only live cases may cause some selection bias, but we think this is of less importance to recall bias. Among all women excluded because of death before the present study (n = 2,138), 1% had melanoma as their cause of death.

Recall bias was assessed by measures of agreement (PA and {kappa}w) and shifts. The kappa statistic depends on the marginal frequencies of the 2-way table, as previously discussed (15). Marginal frequencies explain that {kappa}w can be low or even negative for high values of PA, as seen for asymmetric nevi and solarium use at age <10 (controls) and age 10–19 years (cases and controls). This typically occurs for high values of PA when the data cluster in few categories. Only 1 case in the prospective questionnaire and 6 cases in the retrospective questionnaire reported some solarium use at age 10–19 years. The shift proportion gives useful information in addition to measures of agreement, as high PA or {kappa}w does not imply an absence of systematic shifts. Because the calculation of shifts was based on only the misclassified responses, a high PA implies a small sample size for the shift proportion and the corresponding confidence interval, in particular for cases. This must be kept in mind as there were fewer significant shifts among cases with a wide confidence interval for the shift in hair color.

The statistical challenges of correcting disease risk estimates for measurement error in categorical exposure variables with more than 2 categories have previously been discussed (15). Dichotomization is not a solution as collapsing categories, which was necessary for some variables in our analyses, may in itself cause differential misclassification (27).

In conclusion, the limited body of literature at present indicates that retrospective measures of melanoma risk factors are susceptible to recall bias, but the results are not consistent for the different exposures. We observed indications of recall bias in a study that was designed to minimize the focus on melanoma risk factors and the potential for selection bias. The explanation may be that recall bias in this population was generated by general public awareness about melanoma risk factors, which is likely after increasing publicity about melanoma and campaigns to reduce risk. This complicates retrospective measurements of melanoma risk factors and future case-control studies in well-informed populations.


   ACKNOWLEDGMENTS
 
Author affiliations: Institute of Basic Medical Sciences, Department of Biostatistics, University of Oslo, Oslo, Norway (Christine L. Parr, Petter Laake, Marit B. Veierød); Cancer Registry of Norway, Institute of Population-based Cancer Research, Oslo, Norway (Anette Hjartåker); and Institute of Community Medicine, University of Tromsø, Tromsø, Norway (Eiliv Lund).

The work of C. L. P. was supported by grant 2002/2/0015 from the Norwegian Foundation for Health and Rehabilitation via the Norwegian Cancer Society.

Conflict of interest: none declared.


   NOTES
 
Editor's note: An invited commentary on this article appears on page 267, and the authors’ response is published on page 271.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 

  1. Gandini S, Sera F, Cattaruzza MS, et al. Meta-analysis of risk factors for cutaneous melanoma. I. Common and atypical naevi. Eur J Cancer (2005) 41(1):28–44.[CrossRef][Web of Science][Medline]
  2. Gandini S, Sera F, Cattaruzza MS, et al. Meta-analysis of risk factors for cutaneous melanoma. II. Sun exposure. Eur J Cancer (2005) 41(1):45–60.[CrossRef][Web of Science][Medline]
  3. Gandini S, Sera F, Cattaruzza MS, et al. Meta-analysis of risk factors for cutaneous melanoma. III. Family history, actinic damage and phenotypic factors. Eur J Cancer (2005) 41(14):2040–2059.[CrossRef][Web of Science][Medline]
  4. Han J, Colditz GA, Hunter DJ. Risk factors for skin cancers: a nested case-control study within the Nurses’ Health Study. Int J Epidemiol (2006) 35(6):1514–1521.[Abstract/Free Full Text]
  5. Veierød MB, Weiderpass E, Thorn M, et al. A prospective study of pigmentation, sun exposure, and risk of cutaneous malignant melanoma in women. J Natl Cancer Inst (2003) 95(20):1530–1538.[Abstract/Free Full Text]
  6. Cho E, Rosner BA, Feskanich D, et al. Risk factors and individual probabilities of melanoma for whites. J Clin Oncol (2005) 23(12):2669–2675.[Abstract/Free Full Text]
  7. de Vries E, Boniol M, Severi G, et al. Public awareness about risk factors could pose problems for case-control studies: the example of sunbed use and cutaneous melanoma. Eur J Cancer (2005) 41(14):2150–2154.[CrossRef][Web of Science][Medline]
  8. Coughlin SS. Recall bias in epidemiologic studies. J Clin Epidemiol (1990) 43(1):87–91.[CrossRef][Web of Science][Medline]
  9. Gustafson P. Measurement Error and Misclassification in Statistics and Epidemiology: Impacts and Bayesian Adjustments (2004) Boca Raton, FL: Chapman & Hall/CRC.
  10. Walter SD, Marrett LD, From L, et al. The association of cutaneous malignant melanoma with the use of sunbeds and sunlamps. Am J Epidemiol (1990) 131(2):232–243.[Abstract/Free Full Text]
  11. Walter SD, Marrett LD, Shannon HS, et al. The association of cutaneous malignant melanoma and fluorescent light exposure. Am J Epidemiol (1992) 135(7):749–762.[Abstract/Free Full Text]
  12. Weinstock MA, Colditz GA, Willett WC, et al. Recall (report) bias and reliability in the retrospective assessment of melanoma risk. Am J Epidemiol (1991) 133(3):240–245.[Abstract/Free Full Text]
  13. Cockburn M, Hamilton A, Mack T. Recall bias in self-reported melanoma risk factors. Am J Epidemiol (2001) 153(10):1021–1026.[Abstract/Free Full Text]
  14. Relova AS, Marrett LD, Klar N, et al. Predictors of self-reported confidence ratings for adult recall of early life sun exposure. Am J Epidemiol (2005) 162(2):183–192.[Abstract/Free Full Text]
  15. Veierød MB, Parr CL, Lund E, et al. Reproducibility of self-reported melanoma risk factors in a large cohort study of Norwegian women. Melanoma Res. (2008) 18(1):1–9.
  16. Gefeller O, Brenner H. Re: "Reliability of reported sunburn history in a case-control study of cutaneous malignant melanoma" [letter]. Am J Epidemiol (1996) 144(7):707–708.[Free Full Text]
  17. Lund E, Kumle M, Braaten T, et al. External validity in a population-based national prospective study—the Norwegian Women and Cancer Study (NOWAC). Cancer Causes Control (2003) 14(10):1001–1008.[CrossRef][Web of Science][Medline]
  18. Lund E, Dumeaux V, Braaten T, et al. Cohort profile: the Norwegian Women and Cancer Study—NOWAC—Kvinner og kreft. Int J Epidemiol (2008) 37(1):36–41.[Free Full Text]
  19. English JS, Swerdlow AJ, Mackie RM, et al. Site-specific melanocytic naevus counts as predictors of whole body naevi. Br J Dermatol (1988) 118(5):641–644.[CrossRef][Web of Science][Medline]
  20. Fariñas-Alvarez C, Ródenas JM, Herranz MT, et al. The naevus count on the arms as a predictor of the number of melanocytic naevi on the whole body. Br J Dermatol (1999) 140(3):457–462.[CrossRef][Web of Science][Medline]
  21. Geigy scientific tables. Units of Measurement, Body Fluids, Composition of the Body, Nutrition—Lentner C, ed. (1981) Vol 1, 8th ed. Basel, Switzerland: Ciba-Geigy.
  22. Brown LD, Cai TT, DasGupta A, et al. Interval estimation for a binomial proportion. Stat Sci. (2001) 16(2):101–133. (http://www-stat.wharton.upenn.edu/~tcai/paper/Binomial-StatSci.pdf).[Web of Science]
  23. SAS Institute, Inc. SAS Online Doc 9.1.3 (2007) Cary, NC: SAS Institute, Inc. (http://support.sas.com/onlinedoc/913/docMainpage.jsp).
  24. Altman DG, Machin D, Bryant TN, et al. Statistics With Confidence (2000) 2nd ed. London, United Kingdom: BMJ Publishing Group.
  25. Carroll RJ, Ruppert D, Stefanski LA, et al. Measurement Error in Nonlinear Models: A Modern Perspective (2006) 2nd ed. Boca Raton, FL: Chapman & Hall.
  26. Bataille V, Boniol M, De Vries E, et al. A multicentre epidemiological study on sunbed use and cutaneous melanoma in Europe. Eur J Cancer (2005) 41(14):2141–2149.[CrossRef][Web of Science][Medline]
  27. Wacholder S, Dosemeci M, Lubin JH. Blind assignment of exposure does not always prevent differential misclassification. Am J Epidemiol (1991) 134(4):433–437.[Abstract/Free Full Text]

Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?

Related articles in Am. J. Epidemiol.:

Invited Commentary: Recall Bias in Melanoma—Much Ado About Almost Nothing?
Olaf Gefeller
Am. J. Epidemiol. 2009 169: 267-270. [Abstract] [FREE Full Text]  

Parr et al. Respond to "Recall Bias in Melanoma"
Christine L. Parr, Anette Hjartåker, Petter Laake, Eiliv Lund, and Marit B. Veierød
Am. J. Epidemiol. 2009 169: 271-272. [Extract] [FREE Full Text]  



This article has been cited by other articles:


Home page
 Am J EpidemiolHome page
A. E. Cust, H. Schmid, J. A. Maskiell, J. Jetann, M. Ferguson, E. A. Holland, C. Agha-Hamilton, M. A. Jenkins, J. Kelly, R. F. Kefford, et al.
Population-based, Case-Control-Family Design to Investigate Genetic and Environmental Influences on Melanoma Risk: Australian Melanoma Family Study
Am. J. Epidemiol., December 15, 2009; 170(12): 1541 - 1554.
[Abstract] [Full Text] [PDF]

Home page
 Am J EpidemiolHome page
C. L. Parr, A. Hjartaker, P. Laake, E. Lund, and M. B. Veierod
Parr et al. Respond to "Recall Bias in Melanoma"
Am. J. Epidemiol., February 1, 2009; 169(3): 271 - 272.
[Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
169/3/257    most recent
kwn363v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Related articles in Am. J. Epidemiol.
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (3)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Parr, C. L.
Right arrow Articles by Veierød, M. B.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Parr, C. L.
Right arrow Articles by Veierød, M. B.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?