![[Feedback]](/icons/banner/feedbackACT.gif){kind=icon}
![[For Subscribers]](/icons/banner/subscriberACT.gif){kind=icon}
![[Archive]](/icons/banner/archiveACT.gif){kind=icon}
![[Search]](/icons/banner/searchACT.gif){kind=icon}
![[Contents]](/icons/banner/tocACT.gif){kind=icon}
|
|
|
|
|
||
Roswell Park Cancer Institute, Buffalo, NY 14263
2 To whom correspondence should be addressed. E-mail: james.marshall{at}roswellpark.org.
 |
ABSTRACT |
|---|
 TOP ABSTRACT Measurement as a major...BiomarkersBiomarkers, validity and...What do we need...LITERATURE CITED |
|---|
KEY WORDS: • biomarkers • diet assessment • epidemiology • nutrition • methodology
The goal of nutritional epidemiology is to assess the effects of diet on the risk of disease. An understanding of the effects of diet can then be used to identify aspects of diet that might be altered in efforts to prevent disease. It is necessary, therefore, to measure dietary practice. A common means of measuring diet is to ask subjects to provide summary information. Subjects may be asked to keep written records of their dietary practices for periods ranging from a day to several weeks. They may be asked to recall their dietary intake over a very brief period of time—often a day. They may be asked to describe their usual intake over an extended period, such as a month or a year.
An increasingly attractive option is to obtain a biologic sample from the subject and, characterizing the level of a nutrient, element, compound or metabolite in the sample, use that level or characteristic to describe the person's dietary practice. In this case, we usually attempt to gauge the extent to which an individual tissue is infused with a nutrient, compound, element or metabolite and use that information to reflect on the dietary practices of the people under study. For example, Yoshizawa et al. (1) used selenium in nails to represent long-term selenium intake in a study of prostate cancer. In three other studies of prostate cancer, Gann et al. (2) used serum lycopene to represent lycopene intake, Cook et al. (3) represented food-based ß-carotene intake by serum ß-carotene and Zhang et al. (4) used blood levels of several carotenoids as reflective of carotene intake and as predictive of prostate cancer.
In nutritional epidemiology, the focus may be on nutritional status or on using that status to reflect the intake of a nutrient or some other compound under study. In some instances, the intake reflected by a biomarker is not initially clear. Chan et al. (5), for example, used insulinlike growth factor-1 to predict prostate cancer. As Yu and Rohan (6) recently pointed out, among the factors that affect insulinlike growth factor-1, it appears that both energy and protein intake and possibly alcohol consumption have an impact.
The value of biomarkers in nutritional epidemiology depends, no less than that of any other exposure indicators, on the measurement properties of those biomarkers. Thus this paper considers statistical aspects of biomarker measurement. The first section considers the place of measurement and mismeasurement in nutritional epidemiologic research. The next section compares the limitations of self-reports of dietary and nutritional practice to those of biomarkers, and it outlines repeatability, reliability and validity as they apply to the use of biomarkers in epidemiology. The third section considers critical questions that affect the utility of biomarkers in nutritional epidemiology.
|
Measurement as a major facet of the validity of epidemiologic research |
|---|
|
TOPABSTRACTMeasurement as a major...BiomarkersBiomarkers, validity and...What do we need...LITERATURE CITED |
|---|
Imprecision in the measurement of any exposure has a number of important consequences in epidemiology. Dietary measurement in epidemiology is no exception. Figure 1 depicts this, with exposure treated as a dichotomous variable. For example, subjects might be classified as vegetarian or not, as consumers of a high-fiber diet or not or as consumers of preserved meats or not. A quantitative variable such as fiber, carotene or fat intake, blood cholesterol level or blood carotenoid level may be dichotomized as at the median intake. Imprecision in measurement then leads to misclassification: some exposed people classified as nonexposed, some nonexposed people classified as exposed. Well-behaved imprecision, or measurement error, is most readily understood. It distorts estimates of the effect of exposure, tending in general to lead to underestimation of the impact of exposure (7,8). Table 1 documents this effect. It can be seen that misclassification can cause even a sizeable odds ratio to appear quite unimpressive. What is surprising is that profound mismeasurement and the misclassification that results can cause large and modest relative risks to converge. Thus with 40% misclassification, a true odds ratio of 20 looks remarkably like a true one of 5 or 10; with higher misclassification, a true odds ratio even of 50 resembles one of 3–10.
|
|
|
Also important to epidemiology is that the combination of mismeasurement and confounding can induce biases in the impact of both true risk factors and correlated factors that impart no alteration of risk. These biases are not, in general, eliminated by standard procedures for statistical control. Thus if a single standard-deviation increase in exposure to factor A multiplies risk by a fixed amount, whereas a comparable increase in exposure to a correlate of factor A, factor B, has no effect on risk, well-behaved measurement error can cause the effect of exposure to A to be underestimated and the effect of exposure to B to be overestimated (11,8). Multivariate procedures, which purport to identify distinct effects or associations, do not overcome this problem. Table 3 documents this process. It can be seen that misclassification of a strong confounder that imparts an odds ratio of 10, for example, can lead a completely null factor to be associated with a greater than doubling of the odds ratio. This persists even with adjustment for that strong confounder, because the error in measurement of the confounder leads to less-than-complete control. These biases can be understood only if precise and complete descriptors of mismeasurement can be obtained. Thus if a given dietary exposure, A, confounds estimates of the effect of a different dietary exposure, B, the inability to precisely measure A will confound estimates of the effect of B even if that investigator "adjusts" for the confounding exposure, A (8,12,13). This issue is of greatest concern when A is strongly related to disease.
|
Over- or underreporting of exposure may be related to case status. Despite the wide use of the case-control method in diet and chronic disease studies, there is reason to be concerned that over- or underestimation of dietary exposure could be associated with case status (14). Case individuals may worry and focus on what may have made them sick; they may suspect that their diet was responsible. These conditions may not apply to control individuals. Case individuals may have been sick for an extended period, and their illness or treatment may have affected their eating. In this setting, the case-control method would compare the diets of case individuals made ill by disease or by treatment to those of healthy control individuals and produce a biased assay of the impact of diet on disease risk. The usual means of avoiding this problem has been to ask the case and control individuals to focus and report on their diet during an earlier timespan, when the case individuals would not have been sick. It has been assumed that this method will produce nonbiased results; however the limited literature supporting this assumption does not consistently support it (14). Again, for this kind of error, the investigator has limited possibilities: first, to measure dietary exposures without error; second, to obtain precise estimates of the structure of exposure mismeasurement, including the extent of mismeasurement, correlations among measurement errors and the extent to which they vary among individuals at varying levels of risk, and use that information to adjust estimates for the effects of measurement error or to prospectively collect dietary data as in a cohort study. Another option is to use study designs such as clinical trials so that the study dietary exposure is by design independent of other possible risk-altering exposures. This will enable the investigator to proceed with a minimized possibility of confounding due to inability to assess the impacts of other exposures. However, there are difficulties with regard to long-term dietary interventions.
|
Biomarkers |
|---|
|
TOPABSTRACTMeasurement as a major...BiomarkersBiomarkers, validity and...What do we need...LITERATURE CITED |
|---|
However, repeatability is not the only question that must be posed when considering a biologic exposure marker. The important question is the correspondence of the biologic marker—no matter how repeatable or replicable, or how representative of tissue—with actual dietary exposure. How well does the marker reflect exposure or another variable of interest in terms of understanding the relation of nutrition to intake? Understanding of the nature of the biologic indicator is critical. It might be claimed that the link of actual dietary exposure to biologic status is not as critical as is the ability to use biologic status as a precise, highly replicable variable. Thus one could be concerned with the link of biologic status to a biomarker and less concerned with the link of the biomarker to exposure. It could be asserted that a precise and repeatable marker can be useful if we as investigators have some even vague idea of how it reflects the combination of biologic status and exposure. Clearly, an important question when considering a biomarker is its possible link to biological status. If we wish to use a biologic marker to reflect dietary or nutritional exposure, we need to know what exposure a marker reflects, and how well the marker reflects that exposure. Thus in understanding the research that appears to link insulinlike growth factor to risk of several cancers, it is important to identify the factors that govern blood levels of insulinlike growth factor including potentially nutritional factors (6).
In general, the investigator needs to link a biomarker to a single dietary exposure: he/she wants to evaluate the link of that exposure to disease risk. If the marker reflects more than one dietary exposure, the investigator needs to recognize that the association of the biomarker and link could reflect any of several effects including the combination of effects of several dietary factors. The validity of the measure depends on the dietary exposure the investigator is attempting to ascertain. A highly replicable and very precise measure that cannot be linked to any one single dietary exposure has limited validity for the measurement of that exposure. Calcium in the blood can be readily measured. The repeatability of measurement will be high. Calcium intake may have a number of important effects, so intake may warrant study. But because calcium in the blood has little association with calcium intake, a blood-based measure of calcium status will not be a valuable exposure marker. Urinary calcium appears similarly to be a poor intake marker (17).
Biologic markers offer important opportunities for the measurement of change in clinical trials. What will be important in the future is that we design studies to allow us to carefully evaluate the properties of biologic markers as exposure indicators.
|
Biomarkers, validity and reliability |
|---|
|
TOPABSTRACTMeasurement as a major...BiomarkersBiomarkers, validity and...What do we need...LITERATURE CITED |
|---|
0.7) for blood samples taken on sequential days; it could be much lower (r 0.4) for samples drawn on occasions months or years apart. But repeatability alone does not validate or legitimate a biologic marker. A more restrictive criterion of measurability is reliability. Reliability is the repeatability of a measure net of any repetition of errors of measurement. Returning to the example of the previous paragraph, assume that the repeatability of day-to-day extractions of a blood-based nutrient are high and the repeatability of extractions several months apart are much lower. The high correlation of day-to-day extractions would indicate that the marker under evaluation provides an excellent surrogate for a nutrient on a given day. That the correlation is much lower if the extractions are months apart would indicate that the day-to-day correlation is inflated as a marker of long-term status; it is high because a person whose level is high at one point in time is also likely to have a high level a few days later. The level of the nutrient shifts and cycles over the longer term, and blood drawn at points close together does not reflect this shifting and cycling. Both extractions would provide biased descriptors with respect to the long-term level of the nutrient. The biases or errors of the separate indicators, with respect to long-term intake, would be correlated (19).
A goal of the design of reliability studies is use of the repeatability observed to index reliability. Rendering the repeatability of the measure devoid of extraneous influences depends on what the investigator is trying to measure. A high correlation of two blood-based measures extracted several months apart might suggest high repeatability. However the epidemiologic investigator is usually interested not only in measuring the level of the substance in the blood, but also in using that level as an indicator of intake. The level of the blood could be reflective not just of intake, but also of body size, other exposures such as smoking and individual-to-individual variability in metabolism. The correlation of these levels, even if extracted several months apart, could be inflated as a marker of intake by individual specific characteristics that would remain relatively consistent over time. Biologic sources of variation in biomarkers are discussed by Potischman (15) and Vineis (20). The investigator intending to use a biological marker to describe intake must take care to adjust for biologic sources of variations in that marker.
As the correspondence of two replicate extractions and analyses of a biomarker, reliability is usually considered as the correlation of these separate extractions and analyses. It can be readily shown that reliability in this circumstance refers to the ratio of actual exposure variance to total observed exposure variance (21). A high ratio indicates that most of the variance observed reflects actual true variance in exposure.
Validity is best depicted as a more restrictive yet empirically less directly evaluable characteristic of an exposure biomarker. The ability to identify a highly repeatable characteristic of a biological sample is certainly desirable. This however does not necessarily ensure that the characteristic is reliable or valid. Validity, which is generally defined as the correspondence of a biomarker with the actual exposure the investigator is attempting to ascertain, can be described as the correlation of the measure with actual exposure status (22). It can be readily shown that this correlation is the square root of the reliability coefficient (21). Thus for example, an individual may want to use a blood-based indicator to represent a nutrient exposure. The investigator may draw blood samples at time 1 and at time 2 and analyze those samples under carefully controlled laboratory conditions for levels of the nutrient. If the correlation between the separate levels is high, the individual clearly has a repeatable sample. The assays may be drawn and analyzed so that errors of analyses are not likely to be repeated; the investigator would seem to have a highly reliable measure. But whether these procedures actually establish the validity of the biomarker depends on the nutrient exposure the investigator is trying to measure.
If the investigator is attempting to measure the diet of the person within a short time frame, he/she may be on firm ground to use a brief-period recall. If he/she wants to use a blood sample to represent that intake in a similar time frame, the investigator will want a blood-based indicator that reflects very recent intake. If the investigator wishes to measure long-term diet with respect to the intake of that specific nutrient, this indicator may not be adequate. The investigator will need to know the correspondence of actual long-term intake with the level observed in the blood, net of correlated measurement errors. In attempting to assay exposure to the source of this nutrient, it would be necessary to establish that the blood level reflects intake within the appropriate reference period. If oxidative agents such as smoking or exposure to industrial pollutants could affect status, the measure would have to be adjusted for those other factors. A repeatable biologic marker may not be a valid intake indicator as tempting and attractive as it may appear to be. Establishing the validity of biomarkers poses an important challenge. The investigator who wants to establish the validity of a marker for dietary epidemiology must think very carefully what he/she wishes to represent (23). If the goal is to represent the prior intake of a nutrient, then that investigator will have to juxtapose the biomarker against data on the individual's prior exposure. This may be available through any of a series of self-reports, diaries or other records. It may be possible to use other biologic markers that would reflect this exposure.
It might be asserted that marker validation requires a perfect, gold-standard measure of exposure. Such a measure may simply be impossible to obtain. However, the existence of more than one flawed measure may be adequate if errors in these measures—their imperfections as representations of true exposure—are not correlated (21). Thus an individual's diary of diet intake along with a food frequency may provide an adequate gold standard (19). This requirement can, under carefully evaluated circumstances, be met (19) and utilized in validation studies.
It may, as noted, be necessary for the investigator to hold constant any exposure to oxidative or other agents that might affect this nutrient (17,22,23). Thus the level of the nutrient in the blood would have to be considered net of the level of oxidative agent exposure if the investigator hopes to assay antioxidant exposure. A related example is seen in the assertion by Willett (18) that no nutrient can be evaluated without adjustment for the intake of total energy. Willett has suggested that the measurability of nutrient intake is obscured by total energy intake as well as measurement error, and that the only way to address this confounding and obscuring by energy intake is to consider nutrient intake net of total energy intake. Similarly, the use of any biologic marker requires that the epidemiology of that biomarker be well understood.
|
What do we need to know? |
|---|
|
TOPABSTRACTMeasurement as a major...BiomarkersBiomarkers, validity and...What do we need...LITERATURE CITED |
|---|
We first need to understand the meaning of our biomarkers. It is not enough that we propose to assess whether a biomarker predicts risk. If it predicts risk, we need to understand why it does so. This requires study designs in which we carefully outline measures of long-term nutritional exposure including independent self-reports and independent biologic markers of exposure. We need to understand the half-life of nutritional exposures and so be able to specify the degree to which a biomarker reflects exposure in the past three days, weeks, months or years. For example, selenium in nails can be seen as reflecting exposure that has taken place over a period of several months to several years (18). On the other hand, some blood- and urine-based biomarkers may only reflect exposure within the past few days. The association of the blood level or the excretion of selenium and recent intake may be dependent on bodily saturation (11).
Second, we also need to understand the specificity of biomarkers and their ability to distinguish among different types of nutritional exposures. To date, much of the evaluation of the validity of nutritional exposure reports has focused on what might be called convergent validity or simply the correlation of a given exposure with what is described as a gold-standard measure. For example, Ascherio et al. (24) evaluated the cor relations of vitamin A and E intakes with plasma concentrations of carotenoids and tocopherols among men and women participating in the Nurses Health Study and Health Professionals Followup Study. Forman et al. (25) linked the intake of carotenoid-rich foods to plasma carotenoid concentrations over a brief period and found modest correlations between blood and diet reports for 
-carotene, ß-carotene, ß-cryptoxantin, lutein and lycopene. In a more recent study, Michaud et al. (26) evaluated diet plasma carotenoid con centrations among nonsmoking men and women in the Health Professionals Followup Study and showed that diet and plasma carotenoid associations ranged from 0.2 to 
0.5 for different carotenoid fractions. Pijls et al. (27) used urinary urea ex cretions to evaluate measures of protein intake. The correlations were relatively modest, although the investigators interpreted these correlations to indicate that protein intake and changes in intake could be adequately reflected at a group level. Another important effort was reported by Daures et al. (28). In this study, multiple-day food records and biomarkers were used. For example, the association of ß-carotene was tested against blood and diet records. Daures tentatively claimed that ß-carotene intake could be adequately estimated by a food-frequency questionnaire.
Research that we recently reported (29) indicates that some report-based markers may have very poor discriminant validity. For example, a marker of ß-carotene intake may also mark exposure to the intake of a number of other carotenoids and possibly protective antioxidative vitamins. It may not be specific enough to distinguish between the intake of ß-carotene and those of -carotene, -tocopherol or a number of other nutrients. It is not difficult to envision a situation in which an indicator with a relatively strong relation to a given nutritional variable could also be strongly related to other nutritional variables. The exposures themselves may be correlated, and errors in the measures of the exposures may be correlated (13).
It has been suggested (30) that what protects individuals from chronic diseases is the intake not of specific nutrients but of a wide array of nutrients. Thus it would be the combination of nutrients that is protective, rather than that any specific nutrient protects at all times and against all possible disease-enhancing exposures. Thus what may be needed is a biomarker that reflects the intake of a broad array of diets of fruits and vegetables and the deficit in the intake of certain high-risk factors such as saturated, trans and animal-derived fatty acids. However, whether a single biomarker could reflect this or whether instead a combination of biomarkers is needed is not clear. It is necessary to think very carefully about whether single biomarkers are useful, or whether we need to turn to multiple-biomarker assays for future progress.
Finally, we must consider the degree to which biomarkers might reflect change, i.e., how responsive they might be to changes in dietary practice. An important strategy for evaluating the effects of nutrients involves trials of attempts to change the dietary practices of individuals at elevated disease risk. In the Polyp Prevention Trial (30), the investigators attempted to alter the probability of adenomatous polyp recurrence by changing the diet practice of experimental subjects. Because the study did not succeed in changing this probability (31), it is critical to evaluate whether there was significant change among experimental subjects. This entails critical evaluation of the extent to which the self-reports and biomarkers of change actually reflect change in diet practice. In a similar study that is being conducted among women who have been successfully treated for breast cancer, the goal of the study is to change the risk of breast cancer recurrence by altering dietary practice (32). Whether the study succeeds will not be known for several years. In any case, the ability of self-reports and biological markers to convincingly delineate dietary change will be critical to the interpretation of these trial results. This may require information on the half-life of exposure and changes in exposure. As mentioned earlier, specific biologic characteristics of nutrient exposure, even though they might be reflected in biomarker status, will require more attention than they have received to date.
The goal of this effort has been to identify methodologic and statistical considerations that govern the utility of biomarkers of nutritional exposure and status as we attempt to use these biomarkers in the epidemiology of chronic disease and especially of cancer. It was first noted that measurement imposes an important threat to the validity of epidemiologic research. It bears repeating that if there were indicators of two exposures, both of which imposed a relative risk of 2 at an upper level of exposure, and the first exposure was measured with a greater degree of validity than the second, then the first would appear to be a risk factor and the second would not. The relative validity of the markers would govern the extent to which the first would appear to be a stronger risk factor than the second (8). Clearly the mismeasurement problem is a major challenge to the validity of epidemiologic research.
Biomarkers offer an important opportunity for improving measurement of exposure. It must not be too readily assumed, however, that their greater repeatability ensures that they provide more valid estimates of the association of exposure and disease risk. Their validity as exposure markers depends upon their ability to reflect actual exposure.
We must continue to emphasize that the validity and reliability of biologic markers depend on what the investigator is trying to represent. Repeatability does not establish the validity of a marker. The reliability and the validity of any marker will have to be established if the marker is to be trustworthy as a means of gauging the importance of different exposures.
We addressed the problem of what we need to know. Our use of biomarkers may require that we have some understanding of the epidemiology of status with respect to those biomarkers. This will be important as a means of identifying the link between biomarkers and the exposures we want to consider.
We need to address the degree to which we are willing to do what is necessary to establish biomarkers as a means of progress in nutritional epidemiology. The work we do is painstaking and beset with many impediments. It is expensive and does not readily avail itself of replication. Therefore what we do must be of exquisite quality. Establishing the validity of biomarkers is an expensive and time-consuming operation. We have to decide whether that operation is important enough for us to proceed. We simply cannot be content to work with markers of exposure that have a validity vaguely and euphemistically described as somewhere between 80 and 30%. We need to know much more precisely how good our markers are, and we need to incorporate our understanding of the validity of our biomarkers into our assessments of the impacts of the exposures those biomarkers represent on the risk of disease.
|
FOOTNOTES |
|---|
|
LITERATURE CITED |
|---|
|
TOPABSTRACTMeasurement as a major...BiomarkersBiomarkers, validity and...What do we need...LITERATURE CITED |
|---|
1. Yoshizawa, K., Willett, W. C., Morris, S. J., Stampfer, J. J., Spiegelman, D., Rimm, E. B. & Giovanucci, E. (1998) Study of prediagnostic selenium level in toenails and the risk of advanced prostate cancer. J. Natl. Cancer Inst. 90: 1219–1224.
2. Gann, P. H., Ma, J., Giovannucci, E., Willett, W., Sacks, F. M., Hennekens, C. H. & Stampfer, M. J. (1999) Lower prostate cancer risk in men with elevated plasma lycopene levels: results of a prospective analysis. Cancer Res. 59: 1225–1230.
3. Cook, N. R., Stampfer, M. J., Ma, J., Manson, J. E., Sacks, F. M., Biving, J. E. & Hennekens, C. H. (1999) Beta-carotene supplementation for patients with low baseline levels and decreased risks of total and prostate carcinoma. Cancer 86: 1783–1792.[Medline]
4. Lu, Q., Hung, J.-C., Heber, D., Go, V. L., Reuter, V. E., Cordon-Cardo, C., Scher, H. I., Marshall, J. R. & Zhang, Z. F. (2001) Inverse associations between plasma lycopene and other carotenoids and prostate cancer. Cancer Epidemiol. Biomarkers Prev. 10: 749–756.
5. Chan, J. M., Stampfer, M. J., Giovannucci, E., Gann, P. H., Ma, J., Wilkinson, P., Hennekins, C. H. & Pollack, M. (1998) Plasma insulin-like growth factor-I and prostate cancer risk: a prospective study. Science 279: 563–566.
6. Yu, H. & Rohan, T. (2000) Role of the insulin-like growth factor family in cancer development and progression. J. Natl. Cancer Inst. 92: 1472–1489.
7. Barron, B. A. (1977) The effects of misclassification on the estimation of relative risk. Biometrics 33: 414–418.[Medline]
8. Marshall, J. R. & Hastrup, J. L. (1996) Mismeasurement and the resonance of strong confounders: uncorrelated errors. Am. J. Epidemiol. 143: 1069–1078.
9. Rosner, B., Willett, W. C. & Spiegelman, D. (1989) Correction of logistic regression relative risk estimates and confidence intervals for systematic within-person measurement error. Stat. Med. 8: 1051–1069 (also see discussion, 1071–1073).[Medline]
10. Marshall, J. R. (1989) The use of dual or multiple reports in epidemiologic studies. Stat. Med. 8: 1041–1049 (also see discussion, 1071–1073).[Medline]
11. Greenland, S. (1980) The effect of misclassification in the presence of covariates. Am. J. Epidemiol. 112: 564–569.
12. Rosner, B., Spiegelman, D. & Willett, W. C. (1990) Correction of logistic regression relative risk estimates and confidence intervals for measurement error: the case of multiple covariates measured with error. Am. J. Epidemiol. 132: 734–745.
13. Marshall, J. R., Hastrup, J. L. & Ross, J. S. (1999) Mismeasurement and the resonance of strong confounders: correlated errors. Am. J. Epidemiol. 150: 88–96.
14. Friedenreich, C. M., Howe, G. R. & Miller, A. B. (1993) Recall bias in the association of micronutrient intake and breast cancer. J. Clin. Epidemiol. 46: 1009–1017.[Medline]
15. Potischman, N. (2003) Biologic and methodologic issues for nutritional biomarkers. J. Nutr. 133: 875S–880S.
16. Blanck, H. M., Bowman, B. A., Cooper, G. R., Myers, G. L. & Miller, D. T. (2003) Laboratory issues: use of nutritional biomarkers. J. Nutr. 133: 888S–894S.
17. Byers, T. E., Rosenthal, R. I., Marshall, J. R., Rzepka, T. F., Cummings, K. M. & Graham, S. (1983) Dietary history from the distant past: a methodological study. Nutr. Cancer 5: 69–77.[Medline]
18. Willett, W. C. (1998) Nutritional Epidemiology. Oxford University Press, New York, NY.
19. Spiegelman, D., Schneeweiss, S. & McDermott, A. (1997) Measurement error correction for logistic regression models with an "alloyed gold standard." Am. J. Epidemiol. 145: 184–196.
20. Vineis, P. (1997) Sources of variation in biomarkers. IARC Sci. Publ. 142: 59–71.
21. Carmines, E. & Zeller, R. (1979) Reliability and Validity Assessment. Sage, Beverly Hills, CA.
22. White, E. (1997) Effects of biomarker measurement error on epidemiological studies. IARC Sci. Publ. 142: 73–93.
23. Saracci, R. (1997) Comparing measurements of biomarkers with other measurements of exposure. IARC Sci. Publ. 142: 303–312.
24. Ascherio, A., Stampfer, M. J., Colditz, G. A., Rimm, E. B., Litin, L. & Willett, W. C. (1992) Correlations of vitamin A and E intakes with the plasma concentrations of carotenoids and tocopherols among American men and women. J. Nutr. 122: 1792–1801.
25. Forman, M. R., Lanza, E., Yong, L. C., Holden, J. M., Graubard, B. I., Beecher, G. R., Meltiz, M., Brown, E. D. & Smith, J. C. (1993) The correlation between two dietary assessments of carotenoid intake and plasma carotenoid concentrations: application of a carotenoid food-composition database. Am. J. Clin. Nutr. 58: 519–524.
26. Michaud, D. S., Giovannucci, E. L., Ascherio, A., Rimm, E. B., Forman, M. R., Sampson, L. & Willett, W. C. (1998) Associations of plasma carotenoid concentrations and dietary intake of specific carotenoids in samples of two prospective cohort studies using a new carotenoid database. Cancer Epidemiol. Biomarkers Prev. 7: 283–298.[Abstract]
27. Pijls, L. T., de Vries, H., Donker, A. J. & van Eijk, J. T. (1999) Reproducibility and biomarker-based validity and responsiveness of a food frequency questionnaire to estimate protein intake. Am. J. Epidemiol. 150: 987–995.
28. Daures, J. P., Gerber, M., Scali, J., Astre, C., Bonifacj, C. & Kaaks, R. (2000) Validation of a food-frequency questionnaire using multiple-day records and biochemical markers: application of the triads method. J. Epidemiol. Biostat. 5: 109–115.[Medline]
29. Marshall, J. R., Lanza, E., Bloch, A., Caan, B., Cagguila, A., Quandt, S., Iber, F., Kikendall, W., Slattery, M. & Sowell, A. (1997) Indexes of food and nutrient intakes as predictors of serum concentrations of nutrients: the problem of inadequate discriminant validity. The Polyp Prevention Trial study group. Am. J. Clin. Nutr. 65(suppl. 4): 1269S–1274S.
30. Schatzkin, A., Lanza, E., Freedman, L. S., Tangrea, J., Cooper, M. R., Marshall, J. R., Murphy, P. A., Selby, J. V., Shike, M., Shade, R. R., Burt, R. W., Kikendall, J. W. & Cahill, J. (1996) The Polyp Prevention Trial. I. Rationale, design, recruitment, and baseline participant characteristics. Cancer Epidemiol. Biomarkers Prev. 5: 375–383.
31. Schatzkin, A., Lanza, E., Corle, D., Lance, P., Iber, F., Caan, B., Shike, M., Weissfeld, J., Burt, R., Cooper, M. R., Kikendall, J. W. & Cahill, J. (2000) Lack of effect of a low-fat, high-fiber diet on the recurrence of colorectal adenomas. Polyp Prevention Trial study group. N. Engl. J. Med. 342: 1149–1155.
32. Pierce, J. P., Faerber, S., Wright, F. A., Newman, V., Flatt, S. W., Kealey, S., Rock, C. L., Hryniuk, W. & Greenberg, E. R. (1997) Feasibility of a randomized trial of a high-vegetable diet to prevent breast cancer recurrence. Nutr. Cancer 28: 282–288.[Medline]
This article has been cited by other articles:
|
|
 |
|
  D. C. Greenwood, J. K. Ransley, M. S. Gilthorpe, and J. E. Cade Use of Itemized Till Receipts to Adjust for Correlated Dietary Measurement Error Am. J. Epidemiol., November 15, 2006; 164(10): 1012 - 1018. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. Zamora-Ros, M. Urpi-Sarda, R. M. Lamuela-Raventos, R. Estruch, M. Vazquez-Agell, M. Serrano-Martinez, W. Jaeger, and C. Andres-Lacueva Diagnostic Performance of Urinary Resveratrol Metabolites as a Biomarker of Moderate Wine Consumption Clin. Chem., July 1, 2006; 52(7): 1373 - 1380. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 N. Potischman and J. L. Freudenheim Biomarkers of Nutritional Exposure and Nutritional Status: An Overview J. Nutr., March 1, 2003; 133(3): 873S - 874. [Full Text] [PDF]
|
|
|
|
|
|
H. M. Blanck, B. A. Bowman, G. R. Cooper, G. L. Myers, and D. T. Miller Laboratory Issues: Use of Nutritional Biomarkers J. Nutr., March 1, 2003; 133(3): 888S - 894. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
