A certain amount of tension, much of it unhealthy in my view, exists between laboratory-based (“basic”) medical scientists and epidemiologists. Some mistrust and misunderstanding may stem from the fact that both types of researchers often find themselves competing for the same limited pool of research dollars, although I believe the true source of the problem is much deeper. Regardless of its source, the mutual mistrust betrays an unnecessarily narrow vision. Moreover, such mistrust hinders the kind of collaboration that could speed scientific advances and benefit the society that we aim to serve.

In this article, I will attempt to trace some of the fundamental reasons behind the divisions that exist in these two different, but ultimately compatible, approaches to medical research. Being a pediatrician and pediatric investigator, I will focus my comments on issues related to health and disease in children. I strongly believe, however, that many of these issues are generalizable to other areas of medical research. Also, because my training and background in epidemiology clearly places me on one side of this divide, I cannot claim to be without bias in my views. Nonetheless, I believe that good scientists, regardless of their training and experience, are more persuaded by evidence than by anecdote. I have worked with laboratory-based scientists on numerous grant review committees, expert panels, and task forces. This experience convinces me that bridging the gap between the two approaches is not only feasible, but highly desirable.

ROOTS OF THE SCHISM

To the general public, the word “science” conjures up images of test tubes, high-tech gadgetry, and white laboratory coats. From the earliest ages at which children are exposed to science in school, the emphasis is on chemistry sets, microscopes, and “how things work.” As taught in primary school, high school, and even at university, biology tends to be largely descriptive and physiologic, showing what living organisms look like on the gross and microscopic levels and demonstrating how they function physiologically. The study of chemistry and physics consists of physical measurements and demonstrations of chemical reactions and the laws of mechanics, thermodynamics, optics, magnetism, and electricity.

Although these aspects of education are certainly relevant to the world of science and probably succeed in attracting many children and adolescents to its wonders and excitement, they are also defective in some respects. Little or no emphasis is placed on logic and, specifically, the roles of induction and deduction in evaluating scientific evidence. Principles of causal reasoning are absent from the science curriculum of most primary schools and universities. Development of skills in hypothesis formulation and testing is usually delayed until graduate study, and even then emerges only in the context of a thesis research project, rather than in course work. Finally, the concepts of systematic error (bias) and random error (variability, uncertainty, probability, and chance) receive scant attention in science education at any level. These principles, skills, and concepts are the essence of epidemiology. Little wonder, then, that pediatric and other medical researchers with traditional science education have little understanding of or appreciation for epidemiologic approaches to medical research. Conversely, most epidemiologists lack advanced training in biology and other basic sciences, deficiencies that can occasionally result in biologically implausible hypotheses and inferences that lead basic scientists to mistrust not only those hypotheses and inferences, but even the discipline and methods that gave rise to them.

I have been careful in both the title and text of this commentary to dichotomize medical research as basic versus epidemiologic. For the purposes of this article, I use the term “basic” to denote medical research that attempts to understand the biologic and molecular processes underlying health and disease. By “epidemiologic,” I mean research that assesses health and disease states (outcomes) in groups of human subjects, exposure to factors that may increase or reduce the likelihood of health or disease, and the causal relationship between these outcomes and exposures. [Many basic scientists may be unaware of a schism within epidemiology itself-that between “classical”(population-based) and “clinical” (patient-and clinical intervention-based) epidemiologists. This latter schism, which shows some encouraging recent signs of healing, is beyond the scope of my essay.]

The dichotomy more commonly used in clinical departments is basic versus clinical research, but I have no idea what people mean when they use the term “clinical research”(1). The majority of articles published in the Journal of Clinical Investigation, for example, are based on laboratory experiments in animals, tissues, cells, or subcellular components. Some investigators define“clinical research” as any investigation in which the unit of study is one or more intact human beings. Thus, cloning a gene from a blood sample in a single child would qualify as clinical research under this definition. I prefer the term epidemiologic to “clinical” because it covers an identifiable and coherent set of principles and techniques that clearly distinguish it from the principles and techniques of basic biology.

I also prefer “epidemiologic” to “clinical” because the latter term has become rather pejorative in recent years. One of the reasons that “clinical research” has a bad name among laboratory-based scientists is historical. In many departments of pediatrics(and presumably other clinical specialties), faculty members with no more training than residency and clinical fellowship are expected not only to see patients and teach, but also to carry out research. Although it is widely accepted that good laboratory research requires formal research training,“clinical research” has too often been regarded as an activity that any well trained clinician can and should engage in “on the side.” Acquiring the methodologic skills in research design and statistical analysis required to carry out fundable, hypothesis-driven research usually requires at least 2 years of advanced training, including both formal course work and one or more research projects carried out under the supervision of an experienced epidemiologic investigator. In my view, a single year (such as provided in many clinical fellowships and in typical Masters of Public Health degree programs) does not provide sufficient course work or research supervision to enable a clinician to succeed as an independent investigator in today's highly competitive research“market.” As in basic science, methods matter. And as in basic science, learning these methods during a formal training period of adequate duration is usually more effective than a “do-it-yourself” approach.

Although methodologic training is as necessary in epidemiologic research as it is in laboratory research, in neither approach are good methods sufficient to produce good science. Good science requires focus, depth, and above all, a good question. Some notable epidemiologists have played major roles in advancing the methodologic aspects of the discipline, and a few have made landmark contributions in several substantive areas, but basic scientists may be justified in criticizing epidemiologists who peripatetically wander from one project to another (“Have Methods Will Travel”) and, by so doing, fail to make an important impact in any domain of inquiry. Industry-supported clinical trials for marketing “me-too” pharmaceutical agents is another, perhaps more insidious, example of good epidemiologic methods applied to unimportant questions. Rigorous methodology should be coupled with substantive expertise to ensure that the hypothesis tested is a useful one. Nor does being an expert gene cloner guarantee that the cloned genes will add important knowledge about health and disease, lead to improved diagnosis or treatment, or pave the way to new scientific breakthroughs. Unfortunately, trivial science can and does take place on both sides of the schism.

One of the problems with epidemiologic research is that its“message” is usually comprehensible to most well educated lay persons without formal scientific training. By design, the exposures and outcomes studied by epidemiologists are often those with large potential impacts on public health. Exposures such as diet, environmental tobacco smoke, and chemical contaminants are never far from public scrutiny. Similarly, outcomes such as cancer, heart disease, and birth defects are so common, or commonly feared, that any studies bearing on their cause or prevention is almost guaranteed high visibility. The two-by-two table published in this week's New England Journal of Medicine becomes next week's feature on 60 Minutes.

But the superficial comprehensibility and high visibility of epidemiology are deceptive. Epidemiology is a lot like baseball; the rules are simple and any kid can play. But just as baseball on a major-league level is not child's play, neither is rigorous epidemiologic science within the reach of any clinician or public health worker with access to a database and a personal computer.

Good epidemiologic science is time-consuming and often quite expensive, especially when it requires long-term follow-up. Unfortunately, however, long-term studies are associated with another problem not usually confronted by basic scientists: their results may already be out of date by the time they are published and thus of little relevance to current clinical or public health decision-making. For example, follow-up studies of extremely preterm infant survivors born 10 or 20 years ago may not be generalizable to those surviving today.

THE TWO APPROACHES CONTRASTED

Focus. Most laboratory-based basic research asks the following question: “How does it work?” The main emphasis is on mechanism and experimental manipulation, i.e. assessment of cause at the“micro” level. Epidemiology, on the other hand, usually focuses on the questions “Does it really work?” and “If so, how well?” The primary aim is the assessment of cause at the“macro” level, i.e. inferring whether, and to what extent, exposure to a suspected causal agent, behavior, or treatment causes the outcome under study. Epidemiologists make use of both observational methods (in which they observe outcomes in subjects exposed versus those not exposed to the suspected cause) and human experiments (i.e. controlled clinical trials of prevention or treatment). Although basic scientists also carry out controlled experiments of preventive or therapeutic interventions (usually in laboratory animals or human cells or cellular extracts), they are often not satisfied by establishing or quantifying the effects of the interventions and may be more interested in discovering the underlying physiologic or molecular mechanisms, independent of the effects.

Epidemiologists are far more comfortable than basic scientists with knowledge of cause at the “macro” level and may urge clinical or public action based on that knowledge, even if the basic underlying mechanism is unknown. For example, John Snow(2), an acknowledged“father” of epidemiology, discovered that contaminated water was the cause of cholera epidemics in London in the 1850s, several decades before acceptance of the germ theory of disease and discovery of Vibrio cholerae. Basic scientists often require a fuller understanding of molecular mechanisms before making confident etiologic inferences.

Grant applications. Several important differences between laboratory and epidemiologic approaches to medical research can be gleaned by comparing the content of research proposals, as commonly submitted in grant applications. To be sure, laboratory and epidemiologic scientists share a common set of values that pervade both the preparation and peer review of grant applications. These include emphasis on originality of hypotheses, rigorous methods to test those hypotheses, and well organized arguments presented in a clear writing style.

Perhaps the most important difference in the two approaches is the background (literature review) section of the grant application. It is in this section that the applicant attempts to “build” the case for his or her research by reviewing previous work in the field and explaining how the proposed study either advances previous knowledge or overcomes deficiencies in earlier work. In basic science, the primary desideratum is often an argument for biologic plausibility. In other words, “Can the proposed experiment work?” The applicant attempts to convince the reviewers and funders that the research hypothesis can be defended based on the available evidence. Citation of the previous literature is usually selective, although one-sided(biased) selection often will be detected and criticized by knowledgeable reviewers.

In contrast, reviewers of epidemiologic research proposals increasingly expect a systematic review (even in the absence of formal meta-analysis) of the evidence bearing on the hypothesis under study(3, 4). Systematic review requires that the reviewer attempt to locate all previous studies that have addressed the question/hypothesis under study and to document the search methods used. The applicant must then explain why (e.g. bias, insufficient sample size) the question addressed remains unresolved by previous studies, and how the proposed research will provide a clear answer. I have seen many an epidemiologic project rejected primarily because of selective literature citation. In reviewing the available evidence, the emphasis is not on biologic plausibility, i.e. “Can it work?” but rather,“Does it work?” This contrast thus reflects the difference in focus (causal inference versus mechanism) that I believe characterizes epidemiologic versus basic research.

Finally, laboratory-based and epidemiologic grant proposals differ with respect to funding agencies' requirements for “relevance” or“significance” arguments. With some exceptions (which may become more common in the current era of fiscal restraint), basic scientists are more at ease with serendipity than are epidemiologists. Basic scientists emphasize rigorous science, supported by an abiding faith that eventually, somewhere, sometime, new knowledge will help improve human health. A well known reference on writing grant applications recommends: “I am not saying that you should not mention (author's italics) a possible distant impact of your research; but you should not dwell on it”(5). To be sure, so-called “blue sky” research into fundamental biologic mechanisms can lead to unforeseen payoffs in preventive and therapeutic advances(6). Any comprehensive research strategy would be remiss if it ignored the potential long-term return of fundamental, untargeted investigation. By contrast, an epidemiologic project proposal that does not clearly indicate how the results will be used to guide clinical decision making or health care policy has two strikes against it, irrespective of the quality of its science. This requirement for relevance in epidemiologic research grant applications should not be confused with research contracts; as with basic science proposals, both the hypothesis to be tested and the methods to be used originate with the investigator, not with the funding agency.

Project funding. Another important contrast concerns different funding procedures for laboratory versus epidemiologic operating grants. In both the United States and Canada, many basic scientists are funded for their laboratory, with budgets covering the salary of one or more technicians, equipment, supplies, and so forth. The investigator proposes a series of experiments that require these resources, and, if her previous research has yielded new and useful findings (as evidenced by publications in the peer-reviewed literature and presentations at scientific meetings), her previous performance is taken as a reasonable guarantee of success during the next funding cycle. Once funded, however, the investigator is free to pursue leads suggested by her initial results. The results of initial experiments are used to generate new hypotheses for testing in the next set of experiments. Here is another passage from the above-noted reference on writing grant applications: “Remember that you will be able to support lots of research that is not described in the proposal, provided that it bears some (author's italics) relationship to the proposal”(5). This freedom and flexibility are essential for laboratory scientists. Unforeseen (and unforeseeable) problems may arise with the planned experiments, initial findings may suggest new hypotheses and experiments, and the knowledge base may change rapidly in high profile areas with intense activity and competition.

Although similar flexibility may characterize program grants, funding of epidemiologic projects is almost exclusively project-specific. The time spent by research assistants and other personnel whose salaries are paid for by epidemiologic operating grants is budgeted according to the tasks required for that project. The design and implementation of the project tend to be far more predictable than the series of laboratory experiments proposed in a basic science application. This is true not only for clinical trials, but for observational studies as well. Little opportunity exists for altering the study plan in the light of early results, because the results are not usually available until the end of the funding cycle. In fact, multiple peeks at the data along the way are frowned upon by most statisticians, because they provide opportunities for observing “statistical significance” that exceed the nominal threshold (e.g. p = 0.05) for testing the null hypothesis. Thus the epidemiologic project is a sink-or-swim enterprise. To be sure, if the project is successful (as in the case of the basic scientist, based on publications and presentations), the investigator will build a track record that enhances the chances of obtaining funding for the next project. But reviewers are usually far more interested in the details of the project and are reluctant to approve funding based on a promissory note for ongoing studies, regardless of the investigator's track record.

Laboratory versus human experiments. In almost all basic science research, the experimental intervention that is assigned to a group of animals, cells, or cellular extracts is the same as the intervention received, except perhaps for inadvertent error on the part of the investigator or technician performing the experiment. The analysis of the results of the experiment is therefore the same for the assigned and received intervention, and the comparison is a “clean” contrast of the results under the experimental versus control study conditions.

The human analogy to the laboratory experiment is the randomized controlled trial (RCT). Unfortunately, however, even double-blind RCTs can rarely be as clean as laboratory experiments, because they are subject to the vicissitudes and foibles of human behavior. Unlike laboratory animals, cells, or cellular extracts, intact human beings and their care givers do not always adhere to the treatment to which they are assigned. Patients may not continue treatment if undesirable side effects occur. Switching to the other study treatment may occur if the initial (assigned) treatment does not result in hoped-for clinical benefits in the short term. The treatment assigned in RCTs is therefore not necessarily the same as the treatment actually received by the study subjects.

To benefit from the scientific rigor of randomization, analysis must proceed according to the treatment assigned (by randomization). Were the results of the trial to be analyzed according to the treatment actually received rather than the treatment assigned, the design would change from an RCT toward a mere observational study of treatment. This would lead to confounding (i.e. bias) of the treatment comparison, because the factors associated with poor compliance, crossover, and suboptimal results are likely to affect the study outcome.

For example, in the multicenter Coronary Drug Project from the 1970s, patients randomized to receive clofibrate had no increased or decreased risk of the main study endpoints(7). Those who actually received clofibrate had much better outcomes; so too, however, did those who complied with placebo treatment. In many situations, good compliers tend to have better study outcomes than poor compliers. Good compliance may be a marker for psychologic or other characteristics that have beneficial effects on outcome. Good compliance may even be a consequence, rather than a cause, of favorable outcomes, because patients who are doing well on a treatment are more likely to continue it. Any analysis that depends on good compliance with treatment is therefore likely to be confounded. It also ignores the methodologic advantage of randomization, because the compared groups are no longer entirely determined by chance.

Thus the comparison under study in most randomized trials is a comparison of treatment policies or treatment decisions (whence the term “intention to treat”), rather than of treatments received. Such comparisons are closer to the real-world situation of clinical treatments of individual patients by individual physicians, each of whom exhibits behaviors and makes decisions affected by characteristics unique to the human condition. But they are very different from the clean, either-or treatment comparisons that characterize laboratory experiments.

PAST CONTRIBUTIONS TO CHILD HEALTH

An understanding of molecular, biochemical, and physiologic mechanisms does not automatically indicate whether exposure to a hypothesized agent or intervention will cause a given outcome in intact, free-living human beings. In other words, knowledge of fundamental mechanisms is usually not sufficient to infer causes of human health outcomes. For example, the molecular defect in sickle cell anemia has been understood down to the level of nucleotide substitution for over a quarter of a century. Although such knowledge has recently created the opportunity for prenatal diagnosis, it has not led to any substantial advances in treatment to improve the mortality, morbidity, or quality of life of children affected with this disease. In other words, such knowledge has not been sufficient thus far to develop improved treatment. While awaiting the future potential benefits of gene therapy, other major advances have substantially improved the lot of affected children and adults. By and large, these have not been the result of basic research into the molecular mechanism of red cell sickling. Rather, they have come about as a result of newborn screening(8), penicillin prophylaxis(9), the development and administration of polyvalent pneumococcal(10) and conjugate Haemophilus influenzae type b(11, 12) vaccines, and the improved use of potent analgesics to treat painful crises(13).

Biologic systems are extraordinarily complex, and biologic knowledge is therefore almost always incomplete. The unfortunate consequences of killed(inactivated) measles vaccine(14) and the early respiratory syncytial virus vaccine(15) are but two examples of how such incomplete understanding can lead not only to a failure to achieve expected benefits but also cause unforeseen harm. The biologic explanations for such mishaps may seem “clear” in retrospect but cannot necessarily be anticipated a priori.

If it is granted that knowledge of fundamental mechanisms may not be sufficient for improving treatment, other examples can be used to show that such knowledge may not be necessary, either. The gene for cystic fibrosis was discovered only several years ago(16), yet the life expectancy and quality of life of children and adults with cystic fibrosis have been improving for several decades. The major advances in treating this disease have occurred largely through clinical trial-and-error and accumulated experience with improved therapies for pancreatic insufficiency and lung disease(17).

Our success in preventing Reye's syndrome by avoiding aspirin use in children with influenza and other viruses has occurred in the complete absence of an understanding of the underlying molecular or biochemical mechanisms. In the past, many children with influenza and other viral infections were exposed to aspirin; yet Reye's syndrome was always a relatively rare disease. Children who developed the disease presumably had an underlying genetic defect, which, in combination with a viral infection and exposure to aspirin, led to the microvesicular hepatic fat deposition and encephalopathy characteristic of that disease. Yet knowledge of the strong epidemiologic association with aspirin exposure(18, 19) led to widespread clinical and public health interventions to avoid aspirin use and consequently to a dramatic reduction in incidence despite complete ignorance of the genetic defect(20, 21).

Table 1 summarizes a number of recent important advances in child health, and categorizes those contributions according to the types of research (if any) that appear most responsible for the improvement:1) clinical trial-and-error, 2) basic science, and 3) epidemiology. Although neither basic scientists nor epidemiologists may like to admit it, much of the progress in pediatric care and child health has come about almost exclusively as a result of trial-and-error and the common sense and accumulated wisdom of astute clinicians. Such important improvements as the treatment of cystic fibrosis(17), mechanical ventilation of preterm infants(22–24), and surgical repair of complex cardiac malformations(25–27) appear to fall within this category. As its name implies, however, clinical trial-and-error can occasionally lead to serious errors. Weight loss caused by withholding of solid foods in infants and children with diarrhea due to gastroenteritis(80, 81) and iatrogenic (oxygen-induced) retinopathy of prematurity(82) are but two painful examples of the fallibility of this approach.

Table 1 Roles of clinical trial-and-error, basic science, and epidemiology in recent advances in child health
Full size table

Other important breakthroughs have occurred through the joint efforts of basic science and clinical trial-and-error, again requiring no apparent contribution from the epidemiologic community. These include the prevention of Rh hemolytic disease of the newborn(28, 29), the development of parenteral nutrition(30, 31), organ transplantation(32, 33), and prenatal diagnosis and induced abortion for congenital anomalies(34–36). Basic science alone has been responsible for such advances as the prevention of iron deficiency (through supplementation of infant formula and solid foods)(42–44), the development of biosynthetic(human) growth hormone(45–47), and screening and treatment for phenylketonuria and other inborn errors of metabolism(48–50).

By the same token, several notable preventive breakthroughs have come about purely as a result of epidemiologic studies. Besides the above-mentioned example of preventing Reye's syndrome by avoiding aspirin use(18–21), these include the supine sleeping position to prevent the sudden infant death syndrome(65, 66), reduction in maternal smoking during pregnancy to prevent fetal growth retardation(67–70), periconceptional folate intake to prevent neural tube defects(71, 72), and vitamin A supplementation to reduce mortality from measles and other infectious illnesses in children from developing countries(73, 74). In the childhood injury field, major epidemiologic contributions include the use of seat belts and other restraints to prevent death and injury from motor vehicle accidents(75–77) and of bicycle helmets to prevent severe head injuries(77–79).

In many other areas, however, evidence from both basic science and epidemiology have combined to achieve an important improvement in child health. Vaccine development provides an obvious and recurrent illustration. Fundamental knowledge of viral structure and replication and of the host's immunologic reactions are used to develop inactivated viral antigens or attenuated live viruses that elicit protective antibody and/or lymphocyte responses in the recipient child. The immunogenicity, protective efficacy, and safety of these vaccines must then be demonstrated in rigorously designed randomized trials. Recent examples of the success of such a combined approach include the development of conjugate H. influenzae type b(58–60) and acellular pertussis(61, 62) vaccines. Similar successes may be possible for a conjugate pneumococcal vaccine in the near future.

Knowledge of phospholipid biochemistry and pulmonary physiology led to the purification of natural surfactant and to the synthesis of artificial surfactant(22, 24, 54). Demonstration of the efficacy and safety of surfactant in the treatment of respiratory distress syndrome was then accomplished by large, multicenter randomized trials(55). The resurgence of breast feeding over the last 25 years can be attributed to the joint contributions of basic scientists who have identified immunologically and hormonally active components in human milk, “observational epidemiologists” who have demonstrated health benefits in the recipient infant, and clinical trialists who have assessed the effectiveness of breast feeding promotion interventions(63, 64). Other examples of joint contributions from basic and epidemiologic science include antenatal steroids in the prevention of respiratory distress syndrome(51–53) and the development of oral rehydration solution(56, 57).

Finally, several pediatric advances have benefited from clinical trial-and-error as well as basic and epidemiologic research. These include extracorporeal membrane oxygenation for neonatal respiratory failure(37, 38) and improved treatments for acute lymphoblastic leukemia(39, 40) and hyperactivity/attention deficit(41).

FUTURE OPPORTUNITIES

There is little doubt that laboratory-based discoveries of new preventive or therapeutic interventions will continue to require demonstration of efficacy and safety in randomized trials. Examples include the potential neuroprotective effects of prostaglandin inhibitors, magnesium, and antioxidants and the development and testing of new strategies to reduce vertical transmission of the human immunodeficiency virus. As far as I can tell, such cross-fertilization has been considerably less common in the opposite direction, i.e. the investigation of basic mechanisms underlying novel epidemiologic associations. One good recent example of cross-fertilization from epidemiology to basic science is the discovery that a genetic defect in 5,10-methylene tetrahydrofolate reductase appears to explain the association between low periconceptional folate intake and neural tube defects(83). Two other avenues worthy of future attention are the racial and socioeconomic disparities in risk of preterm birth(84–86) and the“programming” hypothesis for causation of coronary heart disease and other chronic diseases in adults(87, 88).

What are the endocrinologic and cellular biologic changes underlying the preterm onset of labor and preterm prelabor rupture of membranes? Why do these changes occur more commonly among black women and among the socially disadvantaged? How are they aided and abetted by psychosocial factors such as stress and depression? Recent studies suggest that genital tract infections and their effects on secretion of cytokines and prostaglandin precursors may provide one important causal pathway. Placentally derived corticotropin-releasing hormone is now being investigated as a possible hormonal mediator of stress and other psychosocial factors(89, 90). Clearly, much more fundamental research is required in this area.

Barker and his colleagues at the University of Southampton (UK) have reported a large series of studies suggesting that variations in fetal growth are linked to long-term changes in blood pressure, insulin secretion and sensitivity, lipid levels, and coronary heart disease(87, 88). These investigations have led to the hypothesis that nutritional influences operating during fetal life and infancy permanently “program” the hormonal and metabolic milieu in harmful ways that manifest themselves as hypertension, diabetes, and coronary heart disease many decades later. Although some epidemiologists remain skeptical of the results and interpretation of these studies, the studies should stimulate the development of animal models and investigation of the molecular and physiologic mechanisms that may underlie programming effects(91, 92). Long-term epidemiologic cohort(follow-up) studies are plagued by unavoidable problems of losses to follow-up and the inability to adequately measure and control for confounding differences other than fetal or infant growth that could explain the observed associations with long-term outcome. Although more and better epidemiologic studies are needed, so too are laboratory investigations that can confirm or undermine the associations observed in human populations and explain the biochemical and cellular processes underlying them.

Although cross-fertilization between basic scientists and epidemiologists should be encouraged, I believe that future advances in pediatric research will increasingly depend on their collaboration. The use of molecular and other biologic markers not only can provide more valid and precise measurements of potentially causal exposures and disease outcomes but can also be used to assess causal mechanisms and pathways. Investigations of gene-environment interactions are becoming increasingly important in understanding disease etiology. Such investigations, which are often large, expensive, and time-consuming, are virtually impossible to mount nowadays without the joint contributions of basic and epidemiologic scientists. Funding agencies seem increasingly likely to favor grant applications that combine both research approaches. The March of Dimes' new Perinatal Epidemiologic Research Initiative is a good example; the Request for Proposals specifically seeks “innovative approaches that integrate robust epidemiologic investigations and at least one biologic or biochemical measurement.” In addition to their application to large-scale studies of disease etiology, such“innovative approaches” could also be used to advantage in clinical trials and in the development and evaluation of new diagnostic tests.

CONCLUSION

I have tried to summarize what I believe to be the principal sources of division between the epidemiologic and basic science approaches to pediatric research. Being an epidemiologist myself, my goal has been to explain the epidemiologic approach to the basic scientists who constitute the majority of the readership of Pediatric Research. My main aim, however, has been to argue for the ultimate complementarity of the two research approaches, to illustrate their independent and joint contributions to recent advances in child health, and to urge greater collaboration between the scientists who apply them. Such collaboration should help provide rapid and useful answers to the important pediatric questions of today and tomorrow.