Commentary Beyond Black Box Epidemiology Douglas L. Weed, MD, PhD "Black box epidemiology" is the most recent addition to a long list of disciplinary subgroups although few practitioners would likely trumpet their allegiance to it. Epidemiology's many subdisciplines represent the wide applicability of a growing and dynamic field and, paradoxically, an element of professional incohesiveness. Nowhere is this paradox more evident than in a provocative discussion in the recent literature; I call it the "black box" debate. Its provocativeness stems from strong claims and counterclaims regarding a colorful metaphor. Its importance lies in its potential to unite epidemiology in all its disciplinary complexity. My purpose is to briefly describe the historical threads of this discussion, weaving in the black box concepts of systems theory. What arises is a foundation for building conceptual bridges within epidemiology. Two problems also emerge, the solutions of which may frame future discussions. The first involves weaknesses inherent in systems theory. The second concems the divisive forces creating conceptual rifts among epidemiologists, including contributors to the black box debate. Historical Background The history of epidemiology can be seen both as discrete eras or paradigms' and as gradually evolving concepts.2 Some consider history a fabric woven of many threads.3 The black box discussion, occurring at the junction of two eras and reflecting evolving paradigms, comprises at least two such threads: the first in papers by Peto,4 Vandenbroucke,5 Savitz,6 and Skrabanek,7 and the second in papers by Loomis and Wing,8 Krieger,9 and Susser and Susser.10 The First Thread In 1984, Peto described two complementary approaches to cancer epidemiology and prevention.4 The first he called a "mechanistic" approach; it emphasized the biology of carcinogenesis. The second he dubbed the "black box strategy" because it ignored biology in favor of behavioral risk correlates. Peto noted its "low scientific repute." Vandenbroucke echoed this sentiment when he argued that epidemiology must integrate molecular biology with its traditional black box strategy or suffer academic disrepute.5 Recently, Savitz6 defended the tradition of black box epidemiology, arguing that it allows for disease prevention in the absence of a clear understanding of mechanism. Skrabanek,7 on the other hand, marked black box strategies as futile exercises in non-science and an "embarrassing liability" to those who dismantle the black box in their search for "universal laws." The Second Thread For Loomis and Wing,8 neither black box strategies nor molecular-based strategies are adequate. These researchers suggest an integration of biology, behavior, and sociopolitical forces. Krieger9 also calls for a broader conceptualization, as have Susser and Susser, who proclaim the advent of an expansive era called "eco-epidemiology," stretching from societal dynamics to intracellular dynamics." 10 These second-thread The author is with the Preventive Oncology Branch, Division of Cancer Prevention and Control, National Cancer Institute, Bethesda, Md. Requests for reprints should be sent to Douglas L. Weed, MD, PhD, Preventive Oncology Branch, Division of Cancer Prevention and Control, National Cancer Institute, EPS T-41, 9000 Rockville Pike, Bethesda, MD 20892. This paper was accepted April 25, 1997. January 1998, Vol. 88, No. 1 Commentary authors assume that black box and mechanistic strategies can be integrated. But given the strongly negative opinion of black box strategies found in the first historical thread,45'7 it is not clear how that integration can occur. An examination of the nature of black boxes, black box thinking, and their parent concept, systems theory, provides a clue. Black Box Epidemiology One first needs to take a black box to be a metaphor" for the individual organism and take black box tinking to be a methodology that ignores rather than explores the insides ofthe box. One then places the black box near the middle of a structure of scientific knowledge that has recently been labeled as a set of Chinese boxes-each one nested inside another.'0 There, the environment, interpersonal dynamics, and social forces such as race, ethnicity, economics, and politics lie "above" the level of the individual, while organ systems, cells, genes, proteins, atoms, and quarks lie "below" that same level.'2"23 Black box thinking labels the methodologic approach that ignores biology and thus treats all levels of the structure below that of the individual as one large opaque box not to be opened. For some,7 black box hinking is indefensible-hence, the pejorative connotation. A serious problem with such an opinion, however, is that the same negative labeling can apply to scientists who ignore other parts of the structure. After all, unopened black boxes lie above and below every level. Thus, black box thinking labels molecular biologists who fail to explain DNA repair in terms of quantum forces or other entities below molecules. It also characterizes molecular epidemiologists who fail to examine the behavioral implications of genetic knowledge. An unfortunate legacy of the first historical thread, therefore, is a pejorative label for epidemiologists who constrain their conceptualization of the proper domain of investigation. To reach the more expansive view found in the second historical thread, a constructive change in the way we use the ideas of black boxes and black box thinking is necessary. This change represents more than a new set of labels; it is a change in the way epidemiologists view what is important in disease etiology and prevention, a change in conceptual framework. It is called a general systems approach and ironically, a "black box" is neither a limiting construct nor a derogatory label. Rather, it is a central precept. January 1998, Vol. 88, No. I A simple system is composed of inputs, outputs, and mathematical model(s) in between.'4 A 2-by-2 table, the essence of epidemiological analysis, is a simple system, with the input to the black box being the counts of diseased and nondiseased classified by exposure status, the output being the relative risk estimate, and the model being a formula for the odds ratio. Employment selection and its impact on health status-the "healthy worker effect"-is another example,'5 as are infectious disease transmission processes'6 and population screening programs.'7 Engineers have developed systems analysis most extensively, and they teach that a complete understanding of the inner components of a black box at any level of inquiry is impossible.'8 All mathematical models are imperfect representations of reality, although a good model of the inner components is preferred if controlling output is the goal. A central premise of systems theory is that the knowledge of the box's interior and control of its input-output relationships are closely linked. A wide variety of models, including nonlinear, linear, static, or dynamic, are possible; the best model is one that best represents inputoutput measurements. More complex systems arise when the size and scope of the mathematical model are increased (e.g., through parameterization) and black boxes are linked togetherthat is, when the mathematical models developed within the boxes are linked together. The input to one box can represent the output from another. Indeed, when faced with a problem as complex as that represented by the broad structure of scientific knowledge described earlier, a systems analyst will use a "divide and conquer" strategy in which subsystems-the black boxes at each level of the structure-are investigated separately and independently, followed by a more detailed characterization of their connections (K. Lilly Jablokow, personal communication). Even this incomplete presentation of the principles and practice of systems theory warrants the conclusion that the approach has merit. I believe epidemiologists should embrace rather than denigrate the idea of black boxes. Beyond Black Boxes Despite its many strengtis, a systems approach also has weaknesses, especially when a good mathematical model is not available at a particular level of explana- tion.'9 Systems theory is not very effective for solving qualitative problems of morality, law, politics, knowledge, attitudes, beliefs, and social forces, to name a few. Social forces-values-within the scientific community of epidemiology deserve consideration. Values have an undeniable influence on the way epidemiologists thinkl20 and therefore on the way we understand the relationship between the evergrowing list of subdisciplines and the level(s) of the structure of knowledge they examine. Molecular and behavioral epidemiology provide an apt illustration; not everyone is equally enthusiastic about a balanced effort between the two. Although it is reasonable to conceptualize connections from the "top" to the "bottom" of the structure of scientific knowledge,21 epidemiologists may not be willing to make connections between different levels nor suffer others to do the same. The sharp debate between proponents6 and opponents7 of black box epidemiology reveals epidemiologists' vastly different worldviews about the conduct and interpretation of research involving behavior and biology. Philosophers call such stark differences "incommensurabilities,"22 and there is something to be said for examining this problem in terms of the philosophy of science.23 There it is claimed that social forces can foster divisiveness24 such as that seen in epidemiology, with its relatively new subdisciplines of public health25 and clinical epidemiology. Add to this trend toward subspecialization26 a derisive voice7 and sentiments such as the one that nonmedically trained epidemiologists lack sophisticated biological knowledge,27 and the result is a social environment within epidemiology ripe for fractionation and replete with partially incommensurable methodologic paradigms22 providing fodder for the black box discussion. Beyond appeals to reasonableness28 and disclosure,29 countering such divisive forces requires a reassessment of some basic questions: What is the nature of epidemiology? What is our professional telos-that is, the goals intemal to the practice of epidemiology?30 Do we share a common vision and a common purpose? Answers to these questions may not come easily. As shown above, there is diversity in what counts as a legitimate scientific approach. And even if we could agree on a coherent scientific paradigm, we need something more than science to ensure our commitment to public health.0l We need a common set of moral values,23 yet we do not even have a clear consensus on something as basic as our obligation to public health.3' American Journal of Public Health 13 Commentary Recommendations In closing, two recommendations emerge. First, there is a need to explicate and to agree on the basic values of the discipline in this era of professional incohesiveness. Perhaps the questions posed above will help frame such a discussion. Second, epidemiologists should get beyond the pejorative connotation of black box thinking by embracing a systems theory approach while remaining aware of its weaknesses. In so doing, they will secure access to the broad scope of scientific knowledge with the behavior ofpopulations near one extreme and the behavior of molecules near the other. Toward that goal, and in the spirit of professional reunification consistent with the historical and metaphorical thrusts of this paper, I offer one more way to make black boxes at all levels of scientific knowledge a little less opaque. In Anglo-Saxon times, the word black meant "pale," as in a pale-cheeked maiden or a pale light. The words bleak and bleach come from the same root. How the meaning changed so drastically over time is not clear. Perhaps because a pale complexion takes on a bluish tint, the designation was passed on to the darker colors of the spectrum, and finally in modem English it came to mean the total absence of color.32 Whatever the case, history teaches that a black box need not be an opaque box. Rather, it is a pale window through which we peer, contemplating the complexities hidden inside ourselves and catching the pale reflections from the boxes surrounding us. D Acknowledgments The comments and suggestions on an earlier draft, by Drs Steven S. Coughlin, Karen Gerlach, Nancy Krieger, Kathryn Lilly Jablokow, and Diana Pettiti, were greatly appreciated. 14 American Journal of Public Health References 1. Susser M, Susser E. Choosing a future for epidemiology: I. eras and paradigms. Am JPublic Health. 1996;86:668-673. 2. Winkelstein W Jr. Editorial: eras, paradigms, and the future of epidemiology. Am J Public Health. 1996;86:621-622. 3. Lilienfeld AM, Lilienfeld DE. The epidemiologic fabric: I, weaving the threads. Int JEpidemiol. 1980;9:199-206. 4. Peto R. The need for ignorance in cancer research. In: Duncan R, Weston-Smith M, eds. The Encyclopedia ofMedical Ignorance. Oxford, England: Pergamon Press; 1984: 129-133. 5. Vandenbroucke JP. Is "The Causes of Cancer" a miasma theory for the end of the twentieth century?Int&JEpidemiol. 1988;17:708-709. 6. Savitz DA. In defense of black box epidemiology. Epidemiol. 1994;5:550-552. 7. Skrabanek P. The emptiness of the black box. Epidemiol. 1994;5:553-555. 8. Loomis D, Wing S. Is molecular epidemiology a germ theory for the end of the twentieth century?IntJEpidemiol. 1990;19:1-3. 9. Krieger N. Epidemiology and the web of causation: has anyone seen the spider? Soc Sci Med. 1994;39:887-903. 10. Susser M, Susser E. Choosing a future for epidemiology: II. from black box to Chinese boxes and eco-epidemiology. Am J Public Health. 1996;86:674-677. 11. Lakoff G, Johnson M. Metaphors We Live By. Chicago, Ill: Chicago University Press; 1980. 12. Susser M. Systems and levels of organization. In: Causal Thinking in the Health Sciences: Concepts and Strategies of Epidemiology. New York, NY: Oxford University Press; 1973:48-63. 13. Weed DL. Epistemology and ethics in epidemiology. In: Coughlin SS, Beauchamp TL, eds. Ethics and Epidemiology. New York, NY: Oxford University Press; 1996;76-94. 14. von Bertalanffy L. General System Theory. New York, NY: George Braziller; 1968. 15. Weed DL. An epidemiologic application of Popper's method. J Epidemiol Community Health. 1985;39:277-285. 16. Koopman JS. Comment: emerging objectives and methods in epidemiology. Am J Public Health. 1996;86:630-632. 17. van Oortmarssen GJ, Habbema JDF, van der Maas PJ, et al. A model for breast cancer screening. Cancer. 1990;66:1601-1612. 18. Doebelin EO. System Modeling and Response. New York, NY: John Wiley and Sons; 1980. 19. Nijhuis HGJ, Van der Maesen LJG. The philosophical foundation of public health (editorial). J Epidemiol Community Health. 1994;48:1-3. 20. Susser M, Stein Z, Kline J. Ethics in epidemiology. Reprinted in Coughlin SS, ed. Ethics in Epidemiology and Clinical Research. Newton, Mass: ERI; 1995:159-174. 21. Schulte PA. A conceptual and historical framework for molecular epidemiology. In: Schulte PA, Perera FP, eds. Molecular Epidemiology: Principles and Practices. San Diego, Calif: Academic Press; 1993:3-44. 22. Veatch RM, Stempsey WE. Incommensurability: its implications for the patient/physician relationship. J Med Philos. 1995; 20:253-269. 23. Weed DL. Epidemiology, the humanities, and public health. Am J Public Health. 1995; 85:914-918. 24. Longino HE. Science as Social Knowledge. Princeton, NJ: Princeton University Press; 1990. 25. Mackenbach JP. Public health epidemiology. J Epidemiol Community Health. 1995; 49:333-334. 26. Pettiti gives SER address. Epidemiol Monitor. 1995;16:1-2. 27. Holland WW. Editorial: the hazards of epidemiology. Am J Public Health. 1995; 85:616-617. 28. McMichael AJ. Invited commentary: "molecular epidemiology": new pathway or new travelling companion? Am J Epidemiol. 1994; 140:1-11. 29. Bayley C. Our worlds may be incommensurable: now what? J Med Philos. 1995; 20:271-284. 30. Pellegrino ED. Toward a virtue-based normative ethics for the health professions. Kennedy Inst Ethics J. 1995;5:253-277. 31. Weed DL. Science, ethics guidelines, and advocacy in epidemiology. Ann Epidemiol. 1994;4:166-171. 32. Edwards E. Words, Facts, and Phrases. Philadelphia, Pa: JB Lippincott; 1881. January 1998, Vol. 88, No. 1 ORIGINAL ARTICLE The Value of Risk-Factor (\"Black-Box\") Epidemiology Sander Greenland,* Manuela Gago-Dominguez, and Jose Esteban Castelao Abstract: Risk-factor epidemiology has been denigrated by some as an empty search for associations, unguided by underlying theory. It has been defended for occasionally identifying useful (if poorly understood) potential interventions. We further defend risk-factor epidemiology as a valuable source of seemingly unrelated facts that await coherent explanation by novel theories and that provide empiric tests of theories. We illustrate these points with a theory that invokes lipid peroxidation as an explanation of an apparently incoherent accumulation of facts about renal-cell carcinoma.1 The example illustrates the value of viewing epidemiologic, laboratory, and clinical observations as a body of facts demanding explanation by proposed causal theories, whether or not those observations were collected with any hypothesis in mind. (Epidemiology 2004;15: 529 -535) M ost epidemiologic articles are built around a standard format in which the design, execution, and data from the study are described in some detail, along with statistical analyses to identify \"risk factors\" (antecedents of adverse health outcomes that remain associated with the outcomes after adjustments for measured potential confounders). With rare exception, these analyses account only for the possible role of random error (\"chance\") in producing the results; possible roles for biases and causal mechanisms are dealt with by informal discussion. This standard format has been criticized for its failure to account adequately for biases and their interactions (apart from adjustments for measured confounders); the alternative Editors' note: Commentaries on this article appear on pages 519, 521, 523, and 525, and a response from the author on page 527. Submitted 2 April 2003; nal version accepted 21 May 2004. From the *Departments of Epidemiology and Statistics, University of California Los Angeles, Los Angeles, California; and the Department of Preventive Medicine, USC/Norris Comprehensive Cancer Center, Keck School of Medicine of the University of Southern California, Los Angeles, California. Correspondence: Sander Greenland, Departments of Epidemiology and Statistics, University of California Los Angeles, Los Angeles, CA 90095-1772. E-mail: lesdomes@ucla.edu Copyright 2004 by Lippincott Williams & Wilkins ISSN: 1044-3983/04/1505-0529 DOI: 10.1097/01.ede.0000134867.12896.23 Epidemiology Volume 15, Number 5, September 2004 offered is sensitivity analysis and its Monte-Carlo and Bayesian extensions.2- 8 The standard format has also been attacked for its failure to adequately justify the study within a context of biomedical or social theory.9 -11 Some of the latter criticisms have gone so far as to imply that much of epidemiology is pointless and wasteful \"black-box\" research that yields too many nonsense associations and false alarms.12 Among defenses of \"black-box\" epidemiology are that it has identied useful interventions and that the absence of a known causal mechanism has no bearing on the validity of the study results.13 Skrabanek,12 however, responded that the latter point was never in dispute; he asserted instead that \"there is no logical link between black box epidemiologic studies and science\" and that risk-factor epidemiology \"cannot contradict pertinent scientic data.\" We argue here that such claims are logically erroneous and counterproductive. One genuine problem of risk-factor epidemiology is overinterpretation of observed associations as causal. We believe that this problem stems from pressure on researchers to demonstrate scientic or policy relevance of their results, or at least to explain their results as arising from something other than error or bias. We argue that it is a mistake to impose such an explanatory burden on empiric research reports, and that the problem of overinterpretation could be addressed by encouraging a descriptive approach to riskfactor studies. We then illustrate our arguments with a case study on the epidemiology of renal cell carcinoma.1 THE VALUE OF DESCRIPTIVE OBSERVATION REPORTS A theory could be said to explain an observation (or fact) when the observation can be deduced from or is predicted by the theory. According to one popular philosophy of science, the primary role of scientic observation is to supply the facts that require explanation by theories. From this viewpoint, assertions that risk-factor studies could \"provide testable hypotheses of causality\"12 or are \"hypothesis-generating\" are logically backward, even though they could reect some of the psychology of theory creation.14 Furthermore, purely descriptive (atheoretical) approaches to the reporting of single epidemiologic studies (and even metaanalyses) are not only scientic, but benecial; the publication of mere observations is useful, because such observations supply data 529 Greenland et al for the scientic community to use in tests of theories, including theories not even conceived at the time of publication. This means that the incoherence or implausibility of observations in light of current theory should never be a deterrent to publication; a eld that appears to be nothing more than an incoherent jumble of haphazard or implausible observations could be rapidly transformed by the introduction of a new and unifying theory. One consequence of this philosophy is that (contrary to some teachings) precision and replication are not secondary in importance to validity and novelty in study design. Precision and replication establish the facts that need to be explained by theories. A study that is perfectly valid but so imprecise as to be incapable of determining the direction and size of an association is of little or no use in testing predictions about the association. Precise null results are much more informative than so-called \"null\" or \"nonsignicant\" ndings that are compatible with a wide range of possibilities (and hence are of little use for testing theories). Thus, precision, not statistical signicance, measures the statistical importance of results.15 Although many imprecise studies could be combined to provide enough data to precisely establish an association's size, this phenomenon involves some form of study replication. Without sufciently precise data on variation in disease occurrence, there is little that epidemiology can contribute to choosing among proposed social or biologic theories of the disease. With enough description, however, researchers have on hand immediate tests of predictions from proposed theories. A large bank of observations (epidemiologic, laboratory, and clinical) allows rapid winnowing down of theories to promising candidates. The observations can be used regardless of whether they were gathered for the purpose of testing a given theory, although observations collected for that purpose might yield more powerful tests of that theory than would other data sources. Selecting or formulating theories based on the proportion of observations explained raises the specter of overtting, that is, proposing a theory so exible that it can explain not only all that has been observed, but also almost anything likely to be observed. Parsimony (keeping the theory simple) is commonly offered as a safeguard against overtting.14 Nonetheless, it has been argued that uncritical devotion to parsimony can be harmful when modeling epidemiologic data-generating processes and methodologic problems, because parsimony in statistical models often leads to overcondent epidemiologic inferences.8,16 A more direct safeguard against overtting is to check whether a proposed theory excludes some plausible possibilities for data that were not used in formulating the theory (such as observation not yet made). This safeguard is the well-known falsiability or testability criterion for preferring theories. 530 Epidemiology Volume 15, Number 5, September 2004 A theory formulated to explain laboratory results could have a large body of epidemiologic associations available for testing purposes, provided epidemiologists have been thorough in examining and reporting their data. More generally, a promising biologic theory will provide a link among observations from various research specialties (eg, epidemiologic, experimental, and clinical), as well as links among different risk factors; such a theory will also make predictions that are testable with further feasible observations. THE EPIDEMIOLOGY OF RENAL CELL CARCINOMA As an example of these points, the epidemiology of renal cell carcinoma might seem an incoherent body of facts, but for a theory that explains both epidemiologic and laboratory results. That theory posits lipid peroxidation as an intermediate step that leads to a nal common pathway shared by numerous observed risk factors such as obesity, hypertension, smoking, oophorectomy, hysterectomy, parity, diabetes, and dietary antioxidants.1 The theory thus explains the epidemiology of renal cell carcinoma; as we discuss, it also explains numerous experimental and clinical ndings, and has implications for disease prevention. We begin by summarizing the literature on risk factors for renal cell carcinoma. This summary is analogous to the data-description portion of a study report; it is a dry description of our perception of published data. For brevity, we rely heavily on reviews of renal cell carcinoma, along with some original reports, and provide only rough summaries of previous ndings. Like with data descriptions in study reports, the reviews and our summary are subject to reporting errors and biases (eg, publication bias). Unlike primary data descriptions, however, the reader can check the following summaries against the literature: 1. Age: Renal cell carcinoma rates rise rapidly with age.17 2. Sex: Men have age-specic renal-cell carcinoma rates about double the rates in women.17 The remaining summaries refer to associations adjusted for age, sex, and (in some analyses) other variables: 3. Ethnicity: Renal cell carcinoma rates from U.S. cancer registry data are higher in blacks than in whites.18 4. Obesity: Several case-control and cohort studies have reported obesity associated with renal cell carcinoma.17 For example, body mass index (BMI, dened as kg/m2) over 30 appears associated with 4 times the renal cell carcinoma risk of BMI under 22.19 5. Hypertension: Several case-control and cohort studies have reported hypertension associated with renal cell carcinoma.17 A history of hypertension appears associated with a doubling of renal cell carcinoma risk.19 2004 Lippincott Williams & Wilkins Epidemiology Volume 15, Number 5, September 2004 6. Smoking: Several case-control studies have reported cigarette use associated with renal cell carcinoma.17 For example, smokers appear to have a 40% higher risk of renal cell carcinoma than nonsmokers.20 7. Oophorectomy and hysterectomy: Although earlier casecontrol studies had reported no association between hysterectomy or oophorectomy and renal cell carcinoma, more recent investigations have observed these procedures associated with a doubling of renal cell carcinoma risk.21-24 8. Parity: Although previous case-control studies have reported no clear relation between parity and renal-cell carcinoma risk, more recent studies have reported positive associations.21,23-26 For example, women with 5 or more births appear to have twice the risk among women with one or 2 births.23-25 9. Antioxidants: In several case-control studies, high measured dietary intake of antioxidants was associated with a 20% to 50% decrease in the risk of renal cell carcinoma relative to low intakes.27,28 10. Diabetes: Although some case-control studies had reported no association between diabetes and renal cell carcinoma, more recent investigations have consistently found positive associations.21 Several large case-control and cohort studies reported diabetes associated with a 30% to 70% elevation in renal cell carcinoma risk.29 -31 11. Preeclampsia: There have been case reports of renal cell carcinoma in women with severe preeclampsia or hypertension during pregnancy.32,33 12. Analgesics: Although results are not entirely consistent, several case-control studies have reported chronic use of analgesics (aspirin, other nonsteroidal antiinammatory drugs, acetaminophen, and phenacetin) associated with an increased renal cell carcinoma risk.21,34 EXPERIMENTAL AND CLINICAL STUDIES OF LIPID PEROXIDATION What sense can be made of a dozen risk factors ranging from hypertension to hysterectomy? This diversity might suggest that the associations arise from a variety of biologic mechanisms and methodologic problems; the associations might instead be dismissed as the sort of scrap heap that unchecked risk-factor epidemiology produces. A third alternative, however, is that there is a terminal pathway in renal cell carcinoma development affected by all the factors. Rodent experiments point to lipid peroxidation of the proximal renal tubules as a principal mechanism in renal carcinogenesis.1 Renal tumors induced by ferric nitrilotriacetate in rodents appear to be the counterpart of human renal cell carcinoma.35 Similar to human renal cell carcinoma, rodent renal cell carcinoma incidence is higher in males than females, with histopathologic features marked by clear or granular-type cells.35,36 On the clinical side, lipid peroxida 2004 Lippincott Williams & Wilkins The Value of Risk-Factor Epidemiology tion markers are elevated in human renal cell carcinoma tissue.37,38 Thus, our next step is to summarize experimental and clinical observations that link the dozen epidemiologic risk factors discussed here to the hypothesized shared intermediate steps (lipid peroxidation) and the outcome (renal cell carcinoma): 1. Lipid peroxidation in serum increases with age in humans39 and in renal tissue increases with age in animals.40 2. Men have higher serum and plasma levels of lipid peroxidation than women.41- 44 Testosterone treatment or oophorectomy increase ferric nitrilotriacetate-induced lipid peroxidation and renal cell carcinoma incidence in rodents, whereas estradiol treatment or castration decrease ferric nitrilotriacetate-induced lipid peroxidation and renal cell carcinoma incidence.36,45 3. Among patients with diabetes, African-Caribbeans appear to have higher lipid peroxidation levels than whites46 (we are unaware of relevant data for nondiabetics). Black have a higher prevalence of several conditions associated with elevated lipid peroxidation such as hypertension (in both men and women) and obesity (in women).47 4. Obesity/hypertension induces lipid peroxidation and the development of renal damage in rats.48 Obesity has been associated with elevated lipid peroxidation among human subjects, and the elevation appears to be removable by weight reduction.49 -54 5. Hypertension has been associated with elevated lipid peroxidation among human subjects, and the elevation appears to be removable by antihypertensive therapy.55-57 6. Oophorectomy in rodents increases the extent of ferric nitrilotriacetate-induced lipid peroxidation of the renal tubules45 and the incidence of renal cell carcinoma.36 Female mice lacking both ovaries show increased lipid peroxide in serum.41 Human oophorectomy is also associated with elevated serum lipid peroxidation41; because human hysterectomy is usually accompanied by oophorectomy, we would expect hysterectomy to be related to lipid peroxidation as well. 7. Dietary antioxidants and antilipidemic agents prevent ferric nitrilotriacetate-induced lipid peroxidation, nephrotoxicity, or renal cancer development in rats,58 - 60 and they appear to inhibit lipid peroxidation in humans.61- 64 8. Relevant to parity, pregnant rats have higher serum lipid peroxidation than nonpregnant rats,65 and pregnant women have higher lipid peroxidation than nonpregnant women.66 9. Tobacco smoking is associated with lipid peroxidation in humans.67,68 531 Greenland et al 10. Diabetes has been associated with elevated lipid peroxidation.69 11. Preeclampsia is associated with increased lipid peroxidation in women.66,70,71 12. Aspirin, naproxen, and diclofenac appear to increase lipid peroxidation in humans and rodents.72-74 Diclofenac-induced nephrotoxicity in mice has been linked to its ability to increase lipid peroxidation.72 Regarding genetic factors, somatic mutations of the VHL gene seem to occur at high frequency in sporadic cases of renal cell carcinoma. It has been shown that renal carcinoma cells lacking a functional VHL gene express hypoxiainducible genes such as VEGF75-77; furthermore, reintroduction of the wild-type VHL gene into these cells is sufcient to repress VEGF gene expression under normoxic conditions and to restore its normal regulation by hypoxia. It appears that hypoxia can induce lipid peroxidation,78 and also that lipid peroxidation can increase the expression of hypoxiainducible genes.79 Not every study supports the lipid peroxidation-renal cell carcinoma hypothesis. Isolated normal renal tubular tissues have displayed higher levels of malondialdehyde (MDA; a lipid peroxidation marker) than have renal cell carcinoma tissues.80 4-hydroxy-2-nonenal (HNE) (another lipid peroxidation marker) protein adducts occur in kidneys in both normal and tumor cells, and immunomorphologic analyses suggest less HNE protein adducts in tumor than in normal cells.81 Although most studies have found increased lipid peroxidation in men relative to women,41- 44 one study did not.82 In apparent conict with point 5 above, some studies have suggested an association of diuretics or other antihypertensives with renal cell carcinoma21; others have reported no association.19,83 These results reect uncertainties in separating the effect of hypertension from that of its medical treatment. One explanation is that antihypertensives increase risk directly while indirectly lowering risk through hypertension reduction. Indeed, one study observed blood pressure reduction associated with reduced renal cell carcinoma risk.84 Many of these associations (such as those with sex, smoking, and antioxidants) are explainable through other mechanisms or are not specic to kidney cancer. Nonetheless, these observations do not conict with the lipid peroxidation hypothesis because the hypothesis does not preclude other mechanisms, and lipid peroxidation could play a role in other cancers. If and when a truly competing hypothesis is proposed, the challenge for its proponents will be to explain as large a proportion of the available evidence as does the lipid peroxidation hypothesis. A promising theory will not only make predictions testable with further observations; it will also point to new observations that will make past observations useful in testing the theory. For example, antilipidemic agents appear to pre- 532 Epidemiology Volume 15, Number 5, September 2004 vent lipid peroxidation in humans.62- 64 In isolation, this observation says nothing about the lipid peroxidation hypothesis, but it can form part of another test of the theory when combined with data on the relation of antilipidemic agents to renal cell carcinoma. The limited available data support an inverse association of statins with cancers (including renal cell carcinoma),85 and further data could be easily collected in case-control studies of renal cell carcinoma. Lipid-lowering agents could exert benecial effects in patients with different types of renal disease86 such as end-stage renal disease,87,88 a condition itself associated with an increased renal cell carcinoma risk89; thus, we would expect that further epidemiologic renal cell carcinoma data will support the lipid peroxidation hypothesis. In summary, it appears that most available evidence can be explained by the lipid peroxidation hypothesis, and the amount of contradicting evidence is no more than what one might expect given the many sources of error and bias in typical studies. Therefore, we think that the lipid peroxidation hypothesis is well-supported at this time. DISCUSSION Research articles can be useful even if they only report how the study was done and what associations were observed without attempting to interpret or explain these observations in terms of methodologic, biologic, or social theories. Readers can weave the observations into their own explanatory theory, one that incorporates their knowledge of related literature. This is not to discourage the study authors from proposing their own explanations. In particular, the study authors are usually most familiar with the study methods and therefore often best equipped to offer methodologic explanations of their results (ie, explanations of those results based on the particulars of the study design and conduct rather than in terms of the biologic or social mechanisms of ultimate interest). In particular, they could be well-positioned to propose realistic models for sources of bias in their study. These bias models can then be used to examine the sensitivity of the results to plausible bias combinations, and (more ambitiously) can be used to quantify uncertainty about the observed associations in a manner less misleading than ordinary statistical methods.2- 8 We suggest that removing pressure for study reports to delve into general theory could be constructive in allowing the writers more time and the report more space to describe the methods, analyses, and data. The saved space can help ensure that \"null\" associations are reported in as much detail (with their interval estimates) as other associations, and so reduce publication bias in the general literature. The added detail can facilitate formulation of explanations by readers. A descriptive orientation could also encourage deferral of general explanations to more thorough and comprehensive reviews. When forced to cram a minireview into a study report, 2004 Lippincott Williams & Wilkins Epidemiology Volume 15, Number 5, September 2004 study authors can end up biased toward overweighting the evidence that they know well (eg, their own studies) and must inevitably sacrice thoroughness. Our suggestion is analogous to suggestions that authors should be under no pressure to offer public health implications of their results. With regard to the latter, it has been argued that policy implications should be reserved for separate articles that synthesize evidence in a balanced fashion, along with costs and benets of proposed actions.90 Whether considering scientic explanations or policy formulation, rarely if ever does a single study have any implications in isolation from the broader interdisciplinary whole. Yet, for most topics, even an outline of that whole would be too lengthy to include in every research report, and such inclusion would be wastefully redundant. Our suggestion would specialize the use of contextual theory (biology, sociology, and so on) in epidemiologic papers into 3 major categories. In study reports, contextual theory could be used to justify the study and the choice of analytic models (including bias models), but the theory need not be reviewed in detail; the primary mission would be accurate presentation of methods and data. Reviews would then bear the task of evaluating contextual theories against all lines of evidence and against each other. Finally, policy analyses would integrate review information about likely effect sizes, along with cost considerations, to compare the likely costs and benets of various actions (including inaction). Some epidemiologists could nd such article specialization in conict with their conception of public health research. Most researchers accept, however, that personal specialization has become necessary, because one \"renaissance scientist\" can no longer master all the elds that contribute to understanding and control of complex diseases like cancer (pathology, toxicology, genetics, animal experimentation, epidemiology, sociology, statistics, risk assessment, policy analysis, and so on). Article specialization is an analogous necessity when description, analysis, and interpretation of every relevant study cannot be handled by one \"renaissance paper.\" REFERENCES 1. Gago-Dominguez M, Castelao JE, Yuan J-M, et al. Lipid peroxidation: a novel and unifying concept of the etiology of renal cell carcinoma (United States). Cancer Causes Control. 2002;13:287-293. 2. Eddy DM, Hasselblad V, Schachter R. Meta-analyisis by the Condence Prole Method. New York: Academic Press; 1992. 3. Greenland S. Basic methods for sensitivity analysis of biases. Int J Epidemiol. 1996;25:1107-1116. 4. Lash TL, Silliman RA. A sensitivity analysis to separate bias due to confounding from bias due to predicting misclassication by a variable that does both. Epidemiology. 2000;11:544 -549. 5. Greenland S. Sensitivity analysis, Monte Carlo risk analysis, and Bayesian uncertainty assessment. Risk Analysis. 2001;21:579 -583. 6. Lash TL, Fink AK. Semi-automated sensitivity analysis to assess systematic errors in observational epidemiologic data. Epidemiology. 2003; 14:451- 458. 2004 Lippincott Williams & Wilkins The Value of Risk-Factor Epidemiology 7. Phillips CV. Quantifying and reporting uncertainty from systematic errors. Epidemiology. 2003;14:459 - 466. 8. Greenland S. The impact of prior distributions for uncontrolled confounding and response bias. J Am Stat Assoc. 2003;98:47-54. 9. Skrabanek P. The poverty of epidemiology. Perspect Biol Med. 1992; 35:182-185. 10. Skrabanek P. The epidemiology of errors. Lancet. 1993;342:1502. 11. Skrabanek P. Has risk-factor epidemiology outlived its usefulness? Am J Epidemiol. 1993;138:1016 -1017. 12. Skrabanek P. The emptiness of the black box. Epidemiology. 1994;5: 553-555. 13. Savitz DA. In defense of black box epidemiology. Epidemiology. 1994; 5:550 -552. 14. Popper KR. The Logic of Scientic Discovery, 2nd ed. New York: Harper; 1968. 15. Poole C. Low P-values or narrow condence intervals: which are more durable? Epidemiology. 2001;12:291-294. 16. Greenland S. When should epidemiologic regressions use random coefcients? Biometrics. 2000;56:915-921. 17. Yu MC, Yuan J-M, Ross RK. Epidemiology of renal cell carcinoma. In: Petrovich Z, Baert L, Brady LW, eds. Carcinoma of the Kidney and Testis, and Rare Urologic Malignancies. Innovations in Management. Berlin: Springer-Verlag; 1999:3-13. 18. Chow WH, Devesa SS, Warren JL, et al. Rising incidence of renal cell cancer in the United States. JAMA. 1999;281:1628 -1631. 19. Yuan J-M, Castelao JE, Gago-Dominguez M, et al. Hypertension, obesity and their medications in relation to renal cell carcinoma. Br J Cancer. 1998;77:1508 -1513. 20. Yuan J-M, Castelao JE, Gago-Dominguez M, et al. Tobacco use in relation to renal cell carcinoma. Cancer Epidemiol Biomarker Prev. 1998;7:429 - 433. 21. Van Poppel H, Nilsson S, Algaba F, et al. Precancerous lesions in the kidney. Scand J Urol Nephrol Suppl. 2000;205:136 -165. 22. Gago-Dominguez M, Castelao JE, Yuan JM, et al. Increased risk of renal cell carcinoma subsequent to hysterectomy. Cancer Epidemiol Biomarker Prev. 1999;8:999 -1003. 23. Lindblad P, Mellemgaard A, Schlehofer B, et al. International renal-cell cancer study. V. Reproductive factors, gynecologic operations and exogenous hormones. Int J Cancer. 1995;61:192-198. 24. Chow W-H, McLaughlin JK, Mandel JS, et al. Reproductive factors and the risk of renal cell cancer among women. Int J Cancer. 1995;60:321- 324. 25. Kreiger N, Marrett LD, Dodds L, et al. Risk factors for renal cell carcinoma: results of a population-based case-control study. Cancer Cause Control. 1993;4:101-110. 26. Lambe M, Lindblad P, Wuu J, et al. Pregnancy and risk of renal cell cancer: a population-based study in Sweden. Br J Cancer. 2002;86: 1425-1429. 27. Yuan J-M, Gago-Dominguez M, Castelao JE, et al. Cruciferous vegetables in relation to renal cell carcinoma. Int J Cancer. 1998;77:211- 216. 28. Wolk A, Gridley G, Niwa S, et al. International renal cell cancer study. VII. Role of diet. Int J Cancer. 1996;65:67-73. 29. Lindblad P, Chow WH, Chan J, et al. The role of diabetes mellitus in the aetiology of renal cell cancer. Diabetologia. 1999;42:107-112. 30. Schlehofer B, Pommer W, Mellemgaard A, et al. International renalcell-cancer study. VI. The role of medical and family history. Int J Cancer. 1996;66:723-726. 31. Wideroff L, Gridley G, Mellemkjaer L, et al. Cancer incidence in a population-based cohort of patients hospitalized with diabetes mellitus in Denmark. J Natl Cancer Inst. 1997;89:1360 -1365. 32. Tydings A, Weiss RR, Lin JH, et al. Renal-cell carcinoma and mesangiocapillary glomerulonephritis presenting as severe pre-eclampsia. N Y State J Med. 1978;78:1950 -1954. 33. Morrison JJ, Robinson RE. Renal cell carcinoma: an unusual cause of second trimester hypertension. Br J Obstet Gynaecol. 1992;99:775-776. 34. Gago-Dominguez M, Yuan JM, Castelao JE, et al. Regular use of analgesics is a risk factor for renal cell carcinoma. Br J Cancer. 1999;81:542-548. 35. Li JL, Okada S, Hamazaki S, et al. Subacute nephrotoxicity and 533 Epidemiology Volume 15, Number 5, September 2004 Greenland et al 36. 37. 38. 39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52. 53. 54. 55. 56. 57. 58. 59. induction of renal cell carcinoma in mice treated with ferric nitrilotriacetate. Cancer Res. 1987;47:1867-1869. Deguchi J, Miyamoto M, Okada S. Sex hormone-dependent renal cell carcinogenesis induced by ferric nitrilotriacetate in Wistar rats. Jpn J Cancer Res. 1995;86:1068 -1071. Okamoto K, Toyokuni S, Uchida K, et al. Formation of 8-hydroxy-2deoxyguanosine and 4-hydroxy-2-nonenal-modied proteins in human renal-cell carcinoma. Int J Cancer. 1994;58:825- 829. Biri H, Cimen MY, Kacmaz M, et al. Antioxidant potential of cancerous human kidney tissues. Cancer Biochem Biophys. 1998;16:265-272. Balkan J, Kanbagli O, Mehmetcik G, et al. Increased lipid peroxidation in serum and low-density lipoproteins associated with aging in humans. Int J Vitam Nutr Res. 2002;72:315-320. Reckelhoff JF, Kanji V, Racusen LC, et al. Vitamin E ameliorates enhanced renal lipid peroxidation and accumulation of F2-isoprostanes in aging kidneys. Am J Physiol. 1998;274:R767-774. Yagi K. Female hormones as natural antioxidantsa survey of our research. Acta Biochim Pol. 1997;44:701-709. Ide T, Tsutsui H, Ohashi N, et al. Greater oxidative stress in healthy young men compared with premenopausal women. Arterioscler Thromb Vasc Biol. 2002;22:438 - 442. Knight JA, Smith SE, Kinder VE, et al. Reference intervals for plasma lipoperoxides: age-, sex-, and specimen-related variations. Clin Chem. 1987;33:2289 -2291. Dincer Y, Ozen E, Kadioglu P, et al. Effect of sex hormones on lipid peroxidation in women with polycystic ovary syndrome, healthy women, and men. Endocr Res. 2001;27:309 -316. Toyokuni S, Okada S, Hamazaki S, et al. Combined histochemical and biochemical analysis of sex hormone dependence of ferric nitrilotriacetate-induced renal lipid peroxidation in ddY mice. Cancer Res. 1990; 50:5574 -5580. Mehrotra S, Ling KL, Bekele Y, et al. Lipid hydroperoxide and markers of renal disease susceptibility in African-Caribbean and Caucasian patients with Type 2 diabetes mellitus. Diabetic Med. 2001;18:109 -115. Health, United States. Hyattsville, MD: National Center for Health Statistics; 1998. Dobrian AD, Davies MJ, Prewitt RL, et al. Development of hypertension in a rat model of diet-induced obesity. Hypertension. 2000;35:1009 - 1015. Van Gaal LF, Vertommen J, De Leeuw IH. The in vitro oxidizability of lipoprotein particles in obese and non-obese subjects. Atherosclerosis. 1998;137:S39 -S44. Kuno T, Hozumi M, Morinobu T, et al. Antioxidant vitamin levels in plasma and low density lipoprotein of obese girls. Free Radic Res. 1998;28:81- 86. Vasankari T, Fogelholm M, Oja P, et al. Effect of weight reduction on LDL oxidation. Med Sci Sports Exerc. 1997;29:S130. Dietrich M, Block G, Hudes M, et al. Antioxidant supplementation decreases lipid peroxidation biomarker F(2)-isoprostanes in plasma of smokers. Cancer Epidemiol Biomarker Prev. 2002;11:7-13. Olusi SO. Obesity is an independent risk factor for plasma lipid peroxidation and depletion of erythrocyte cytoprotective enzymes in humans. Int J Obesity. 2002;26:1159 -1164. Vasankari TJ, Kujala UM, Vasankari TM, et al. Reduced oxidized LDL levels after a 10-month exercise program. Med Sci Sports Exerc. 1998; 30:1496 -1501. Usyal M, Bulur H, Sener D, et al. Lipid peroxidation in patients with essential hypertension. Int J Clin Pharmacol Ther Toxicol. 1986;24: 474 - 476. Kumar KV, Das UN. Are free radicals involved in the pathobiology of human essential hypertension? Free Radic Res Commun. 1993;19:59 - 66. Keidar S, Kaplan M, Shapira C, et al. Low density lipoprotein isolated from patients with essential hypertension exhibits increased propensity for oxidation and enhanced uptake by macrophages: a possible role for angiotensin II. Atherosclerosis. 1994;107:71- 84. Zhang D, Okada S, Yu Y, et al. Vitamin E inhibits apoptosis, DNA modication, and cancer incidence induced by iron-mediated peroxidation in Wistar rat kidney. Cancer Res. 1997;57:2410 -2414. Hashimoto S. Study on the effects of the hypocholesteremic agents, 534 60. 61. 62. 63. 64. 65. 66. 67. 68. 69. 70. 71. 72. 73. 74. 75. 76. 77. 78. 79. cholestyramine and compactin on the induction of renal tumors in rats by N-ethyl-N-hydroxyethyl nitrosamine. Jpn J Urol. 1992;83:75- 84. Qin X, Zhang S, Zarkovic M, et al. Inhibitory effect of probucol on nephrotoxicity induced by ferric nitrilotriacetate (Fe-NTA) in rats. Carcinogenesis. 1995;16:2549 -2552. Brude IR, Drevon CA, Hjermann I, et al. Peroxidation of LDL from combined-hyperlipidemic male smokers supplied with omega-3 fatty acids and antioxidants. Arterioscler Thromb Vasc Biol. 1997;17:2576 - 2588. Salonen R, Nyssonen K, Porkkala-Sarataho E, et al. The Kuopio Atherosclerosis Prevention Study (KAPS): effect of pravastatin treatment on lipids, oxidation resistance of lipoproteins, and atherosclerotic progression. Am J Cardiol. 1995;76:34C-39C. De Caterina R, Cipollone F, Filardo FP, et al. Low-density lipoprotein level reduction by the 3-hydroxy-3-methylglutaryl coenzyme-A inhibitor simvastatin is accompanied by a related reduction of F2-isoprostane formation in hypercholesterolemic subjects: no further effect of vitamin E. Circulation. 2002;106:2543-2549. Girona J, La Ville AE, Sola R, et al. Simvastatin decreases aldehyde production derived from lipoprotein oxidation. Am J Cardiol. 1999;83: 846 - 851. Yoshioka T, Motoyama H, Yamasaki F, et al. Lipid peroxidation and vitamin E levels during pregnancy in rats. Biol Neonate. 1987;52:223- 231. Morris JM, Gopaul NK, Endresen MJ, et al. Circulating markers of oxidative stress are raised in normal pregnancy and pre-eclampsia. Br J Obstet Gynaecol. 1998;105:1195-1199. Miller ER III, Appel LJ, Jiang L, et al. Association between cigarette smoking and lipid peroxidation in a controlled feeding study. Circulation. 1997;96:1097-1101. al Senaidy AM, al Zahrany YA, al Faqeeh MB. Effects of smoking on serum levels of lipid peroxides and essential fat-soluble antioxidants. Nutr Health. 1997;12:55- 65. Jain SK, McVie R, Jaramillo JJ, et al. The effect of modest vitamin E supplementation on lipid peroxidation products and other cardiovascular risk factors in diabetic patients. Lipids. 1996;31:S87-S90. Hubel CA, Roberts JM, Taylor RN, et al. Lipid peroxidation in pregnancy: new perspectives on preeclampsia. Am J Obstet Gynecol. 1989; 161:1025-1034. Wang YP, Walsh SW, Guo J, et al. The imbalance between thromboxane and prostacyclin in preeclampsia is associated with an imbalance between lipid peroxides and vitamin E in maternal blood. Am J Obstet Gynecol. 1991;165:1695-1700. Hickey EJ, Raje RR, Reid VE, et al. Diclofenac induced in vivo nephrotoxicity may involve oxidative stress-mediated massive genomic DNA fragmentation and apoptotic cell death. Free Radic Biol Med. 2001;31:139 -152. Durak I, Karaayvaz M, Cimen MY, et al. Aspirin impairs antioxidant system and causes peroxidation in human erythrocytes and guinea pig myocardial tissue. Human Exp Toxicol. 2001;20:34 -37. Ji B, Masubuchi Y, Horie T. A possible mechanism of naproxen-induced lipid peroxidation in rat liver microsomes. Pharmacol Toxicol. 2001;89: 43- 48. Gnarra JR, Zhou S, Merrill MJ, et al. Post-transcriptional regulation of vascular endothelial growth factor mRNA by the product of the VHL tumor suppressor gene. Proc Natl Acad Sci U S A. 1996;93:10589 - 10594. Iliopoulos O, Levy AP, Jiang C, et al. Negative regulation of hypoxiainducible genes by the von Hippel-Lindau protein. Proc Natl Acad Sci U S A. 1996;93:10595-10599. Siemeister G, Weindel K, Mohrs K, et al. Reversion of deregulated expression of vascular endothelial growth factor in human renal carcinoma cells by von Hippel-Lindau tumor suppressor protein. Cancer Res. 1996;56:2299 -2301. Sarada SK, Dipti P, Anju B, et al. Antioxidant effect of beta-carotene on hypoxia induced oxidative stress in male albino rats. J Ethnopharmacol. 2002;79:149 -153. Ruef J, Hu ZY, Yin LY, et al. Induction of vascular endothelial growth factor in balloon-injured baboon arteries. A novel role for reactive oxygen species in atherosclerosis. Circulation Res. 1997;81:24 -33. 2004 Lippincott Williams & Wilkins Epidemiology Volume 15, Number 5, September 2004 80. Yang AH, Oberley TD, Oberley LW, et al. In vitro modulation of antioxidant enzymes in normal and malignant renal epithelium. In Vitro Cell Dev Biol. 1987;23:546 -558. 81. Oberley TD, Toyokuni S, Szweda LI. Localization of hydroxynonenal protein adducts in normal human kidney and selected human kidney cancers. Free Radic Biol Med. 1999;27:695-703. 82. Block G, Dietrich M, Norkus EP, et al. Factors associated with oxidative stress in human populations. Am J Epidemiol. 2002;156:274 -285. 83. McLaughlin JK, Chow WH, Mandel JS, et al. International renal-cell cancer study. VIII. Role of diuretics, other anti-hypertensive medications and hypertension. Int J Cancer. 1995;63:216 -221. 84. Chow WH, Gridley G, Fraumeni JFJr, et al. Obesity, hypertension, and the risk of kidney cancer in men. N Engl J Med. 2000;343:1305-1311. 85. Graaf MR, Beiderbeck AB, Egberts ACG, et al. The risk of cancer in users of statins. Proc Am Soc Clin Oncol 2003;22:846, 39th Annual ON MIASMAS The Value of Risk-Factor Epidemiology Meeting, Chicago, May 31-June 3; Abstract 3400 and oral presentation. 86. Elisaf M, Mikhailidis DP. Statins and renal function. Angiology. 2002; 53:493-502. 87. McKenney JM. Potential nontraditional applications of statins. Ann Pharmacother. 2003;37:1063-1071. 88. van den Akker JM, Bredie SJ, Diepenveen SH, et al. Atorvastatin and simvastatin in patients on hemodialysis: effects on lipoproteins, C-reactive protein and in vivo oxidized LDL. J Nephrol. 2003;16: 238 -244. 89. Stewart JH, Buccianti G, Agodoa L, et al. Cancers of the kidney and urinary tract in patients on dialysis for end-stage renal disease: analysis of data from the United States, Europe, and Australia and New Zealand. J Am Soc Nephrol. 2003;14:197-207. 90. Rothman KJ. Policy recommendations in epidemiologic research papers. Epidemiology. 1993;4:94 -95. AND ASTROLOGY \"For want of knowing any other cause, epidemics were attributed by the ancients to the atmosphere, without any evidence; just as political and social events were believed to be occasioned by the stars.\" JOHN SNOW 2004 Lippincott Williams & Wilkins 535 Critique regarding the value of black box epidemiology Abstract Since the World War II, the epidemiologists have largely focused on the risk factors of various diseases, using an approach called as \"black box \"epidemiology. The \"black box\" is rather used as metaphor and thinking it as a methodology that ignores rather than explores the insides of the box which means that the methodologic approach that ignores biology and thus treats all levels of the structure below that of the individual (organ systems, cells, genes, proteins, atoms, and quarks) as one large opaque box not to be opened. Main body The black box strategy is a non-randomized studies of disease causation conducted among individuals within a given population has been questioned since a long time. Taking into consideration that there is no underlying hypotheses for this type of research beyond a simple instinct that diseases of civilization are caused by civilization. There are many proponents and opponents over the black box epidemiology. Among which Savitz defended the tradition of black box epidemiology, arguing that it allows for disease prevention in the absence of a clear understanding of mechanism. Skrabanek, On the other hand, marked black box strategies as futile exercises in non-science and an "embarrassing liability" to those who dismantle the black box in their search for "universal laws." While there can be several problems associated with this strategy, for example, the scientists who ignore major parts of the system would be negatively labelled. Also black box thinking labels molecular biologists who fail to explain DNA repair in terms of quantum forces or other entities below molecules. It also characterizes molecular epidemiologists who fail to examine the behavioral implications of genetic knowledge. The weakness related to the risk factor epidemiology is that it is not as effective as other methodologies because a good mathematical model is not available at a particular level of explanation. But, there are associations that can be supported with the value of risk factor epidemiology, for example, as commented by Feinstein that even in the absence of a compelling hypothesis, carefully made observations in epidemiologic studies can play a role in the development of theories of etiologic pathways leading to the occurrence of diseases, and so such observations should be welcomed, not shunned. Researchers must understand that a complete understanding of the inner components of a black box at any level of inquiry is impossible. So, many of the authors assumed that black box and mechanistic strategies (as stated by Peto in 1984) can be integrated. But because of the negative opinion on the black box epidemiology, it is still not clear how it has to be done. Risk factor epidemiology is an ancillary methodology which , if governed by sufficient scientific approach, then it can surely provide testable hypotheses of causality. In order to ensure the raveling of causation of diseases with more accuracy, the epidemiologists must get beyond the pejorative connotation of black box thinking by embracing a systems theory approach while remaining aware of its weaknesses. In doing so, they will secure access to the broad scope of scientific knowledge with the behavior of populations near one extreme and the behavior of molecules, near the other. Therefore, the black box which was before considered as one large opaque box not to be opened, must be offered different approaches such that it give ways to make black boxes at all levels of scientific knowledge a little less opaque. The \"black-box\"epidemiology has identied useful interventions and that the absence of a known causal mechanism has no bearing on the validity of the study results. While the research had merit, possible efforts must be taken to improve its effectivity such that when the epidemiologists draw broader etiologic conclusions or suggest policy implications, it should be based on all relevant research, not just on the findings presented by the investigators in a particular report. References 1. Sander Greenland,* Manuela Gago-Dominguez, and Jose Esteban Castelao (2004) The Value of Risk-Factor (\"Black-Box\") Epidemiology Epidemiology 2004;15: 529-535. 2. Weiss, Noel S. Presents Can Come in Black Boxes, too Epidemiology September 2004 - Volume 15 - Issue 5 - pp 525-526 doi: 10.1097/01.ede.0000135175.11460.27 Commentary 3. Petr Skrabanek (1994) The emptiness of black box Epidemiology Resources Inc. 4. Douglas L. Weed, MD, PhD (January 1998) Beyond Black Box Epidemiology American Journal of Public Health, Vol. 88, No. I. Commentary