Biostatistics is the application of statistics to biology and medicine. It is concerned with the assessment of observed variation in living organisms, particularly human beings. It seeks better insight into the life process, with focus on the cause, treatment, and prevention of disease. It uses the theories and methodology of statistics, but has created specialized methods of its own.
The development of statistical inference in the late-nineteenth and early-twentieth centuries was motivated by problems in biology, and its growth stimulated by the subsequent explosion of research in science and technology and the advent of the electronic computer. Responding to challenges posed by large-scale biomedical research programs, biostatistics emerged as a vigorous distinct discipline. Its scope includes data collection and analysis pertaining to virtually all facets of the vast healthcare system. The study of health factors affecting populations, with emphasis on public health issues, is the realm of epidemiology, a closely related field using the theories and methods of biostatistics.
Experimentation on human subjects in clinical research involves both biostatistics and ethics, including ethical aspects of clinical trials. But the two fields also intersect on broader concerns related to medical uncertainty and complexity: poor understanding on the part of the public, conflicts of interest, manipulation by the market, and questions of responsibility. Greater awareness of these issues is needed to help address critical problems facing contemporary medicine.
Concepts and Methods of Biostatistics
In the field of descriptive statistics, biostatistics contributes to the preparation of official records characterizing the health of the nation. As participant in the biomedical research process, it provides study design based on theories of statistical inference, primarily the classical Neyman-Pearson theory of hypothesis testing. Applying a wide range of standard techniques, it considers the two types of error in testing, determines required sample size for desired power, and assesses the statistical significance of results. It estimates outcomes of interest with associated confidence intervals. Its best-known specialized technique is the randomized clinical trial (RCT) for controlled experiments. For observational research the chief methods are cohort and case-control studies.
HEALTH STATISTICS. An illustration of data provided by the National Center for Health Statistics is given in Figures 1 and 2, showing cancer death rates in the United States from 1930 to 2000 for the major sites, for males and females. Such records of health statistics are an important resource for public health policies and biomedical research, in this case for studies of the etiology, treatment, and prevention of cancer. For example, although lung cancer remains the leading cause of cancer death, the decreasing rate for males in the last decade reflects the decrease in the prevalence of smoking, with a plateau in the death rate seen thus far for women.
EXPERIMENTAL RESEARCH: THE RANDOMIZED CLINICAL TRIAL. A clinical trial is an experiment in which a selected group of patients is given a particular treatment (intervention), typically a drug, and followed over time to observe the outcome. In a randomized clinical trial, also called randomized controlled trial (both referred to as RCT), patients are assigned at random to one of two or more treatments to assess relative effectiveness. Individual differences among patients that may affect their response are assumed to be balanced out by the random assignment. Ethical mandates include clinical equipoise (lack of medical consensus on the superiority of any of the treatments) and informed consent (willing participation of fully informed patients). The research protocol describing the proposed trial must be approved by the local Institutional Review Board (IRB).
The study may conclude before an outcome is observed for each patient (for example, the patient is still alive when the outcome is death). Such patients are said to be still at risk, and have a censored observation. The graphic summary of results is the so-called survival curve, which shows the proportion of patients alive (or disease-free if the outcome is recurrence) at each point in time along the period of observation. It is based on the life-table or actuarial method, with time 0 representing the entry point of each patient into the trial. Showing two or more arms of a study on the same graph offers a visual comparison of treatment outcomes. Special techniques of survival analysis can compare groups with inclusion of censored observations. There are methods to test the hypothesis that there is no difference between treatments, including adjustment for observed patient characteristics that may affect outcome.
Figure 3 presents five-year results of a three-arm RCT comparing disease-free survival of breast cancer patients treated with total mastectomy, segmental mastectomy (lumpectomy), and segmental mastectomy with radiation therapy. All patients with positive axillary lymph nodes received adjuvant chemotherapy. The first graph shows lumpectomy to be just as effective as mastectomy; the other two indicate lumpectomy with radiation therapy to be significantly better than either surgical procedure alone.
OBSERVATIONAL RESEARCH: COHORT AND CASE-CONTROL STUDIES. The two main approaches to addressing questions for which experimentation is ethically not feasible or otherwise not practicable are the observational designs of cohort and case-control studies, the basic tools of epidemiology. They aim to discover or confirm an association between some exposure or risk factor and a disease, using specific criteria of statistical theory and methodology.
Cohort Study. This is usually a prospective study that identifies a large group (cohort) of individuals without the disease, but with information about the presence or absence of the risk factor under study. The cohort is then followed over time to observe for the occurrence of the disease. Smoking is a risk factor that cannot be studied in RCTs. In the hypothetical example shown in Table 1, a cohort of 2,000 adult males is followed to observe for a diagnosis of lung cancer; 500 of the men are smokers and 1,500 nonsmokers at the beginning of the study. As a possible outcome after twenty years, 24 percent of smokers and 2 percent of nonsmokers have contracted lung cancer. The measure of association used is the relative risk (RR) or risk ratio, 24/2 = 12.0.
Case-Control Study. This is a retrospective design, which identifies a group of people who have the disease (cases), selects a group as similar as possible to the cases except that they do not have the disease (controls), and then determines how many in each group were exposed to the risk factor. An actual example is shown in Table 2, in a study of the association between stroke in young adults and drug abuse, with 214 cases and 214 controls. It was found that seventy-three of the stroke victims had a history of drug abuse, compared with eighteen in the control group. The odds of drug abuse given the stroke are 73/214 to 141/214, and given no stroke, 18/214 to 196/214. The measure of association is the odds ratio (OR) as the estimate of relative risk, in this case .5177/.0918 = 5.64.
Comparison of Research Designs. The relationship between cohort and case-control studies is shown symbolically in Table 3. For both measures of association, RR and OR, a value of 1.0 indicates no association. There are statistical methods to test the hypothesis of no association and to provide a confidence interval for RR or OR. Confidence intervals that do not include 1.0 reflect a significant association; values less than 1.0 denote a protective effect of the factor being studied. The examples above (RR = 12, OR = 5.64) show strong associations. Media reports for a study claiming a 20 percent increase in relative risk, for example, would correspond to RR = 1.2, a weak association even if statistically significant.
Cohort studies permit careful selection of the study population and recording of the risk factor, and rates of disease can be calculated for both the exposed and unexposed group. But long observation of a large number of subjects is required, many may be lost to followup or their exposure may change, and the studies tend to be expensive. Case-control studies require fewer subjects, cost less, and can be completed in a relatively short time period. Instead of the risk of disease given the exposure, they estimate the odds of being exposed given the disease. But case-control studies rely on recall of past exposures that may be impossible to confirm and the selection of an appropriate control group is extremely difficult. A different group of controls, or just a change of a few, could completely alter the outcome. These are some reasons that so many conflicting results are reported in the medical literature. Others include small, improperly done clinical trials and those with short follow-up. But in any case, claims can only be valid for an association between exposure (or intervention) and disease. The assessment of causation is a lengthy, tentative process, with general guidelines to aid the research community (Hill 1967).
DIAGNOSIS AND SCREENING. Further uncertainties exist in the diagnosis of disease, and biostatistics provides methods to evaluate tests used in diagnostic and screening procedures. Most tests have an overlapping range of values for a healthy population and patients with the disease, so that in setting a cutoff point to distinguish positive from negative test results, two types of error may be made. The four possible outcomes are shown in Table 4, with the standard performance characteristics of diagnostic tests. Sensitivity is the ability of a test to detect disease when present, and specificity its ability to indicate nondisease when none is present. The numeric example shows a test for fetal malformation with ultrasonography, which has reported 56 percent sensitivity and 99.5 percent specificity. The prior probability that a woman with poorly controlled diabetes has a malformed fetus is given as P(D+) = .20. Using these numbers, one can apply a formula for conditional probabilities known as Bayes' Theorem to estimate the predictive value of the test, the posterior probability of malformation given a positive or negative test result. In this example a positive ultrasound yields a 96.6 percent probability that the fetus is malformed, and a negative result a 90 percent probability that it is normal.
Inanyonecaseaseriesoftestsmaybeusedtoestablish diagnosis, with the sensitivity and specificity of common tests established in previous studies. But there is inherent variation in the laboratory and imaging process itself, as well as the reliability of human raters. In addition, promising new markers for disease may present new uncertainties concerning cutoff points and criteria for treatment.
Decision to Treat: Prostate Cancer. For example, wide use of the test for prostate-specific antigen (PSA) has resulted in earlier diagnosis and decrease in the death rate from prostate cancer since the early 1990s (Figure 1). The test measures the blood level of PSA, a protein made by the prostate. It was defined as positive at 4.0 ng/ml, although higher levels also often indicate benign conditions. But a 2004 study reported prostate cancer on biopsy in 15 percent of 2,950 men with seven years of normal PSA levels and negative digital examinations. The prevalence of cancer was positively correlated with increasing PSA level from less than 0.5 to 4.0 ng/ml. Most of these cancers will not progress to life-threatening disease, and the question is whether to try to diagnose and treat these previously missed early cases.
Decision to Treat: Breast Cancer. About 50,000 cases of ductal carcinoma in situ (DCIS) are diagnosed in the United States each year, 20 percent of all breast cancers. These are cancers within the duct, not palpable, found on biopsy of suspicious regions identified by increasingly sensitive mammography. After lumpectomy an estimated 10 to 15 percent of DCIS will recur as invasive breast cancer. Prognosis is uncertain in individual cases, and variations of further treatment tend to be the recommended procedure. There has been a downward trend in breast cancer mortality (Figure 2), but breast cancer remains the second leading cause of cancer death for women, and a DCIS diagnosis creates vexing uncertainties for affected women.
NUMBER NEEDED TO TREAT (NNT): AN ESSENTIAL CONCEPT. Evaluating the results of a clinical trial, factors to consider include the type of patients studied, the length of follow-up, and the safety and effectiveness of treatment. The latter is especially important in prevention trials, when observed advances may involve long-term treatment of large populations. It is more informative to present NNT, the number of patients that have to be treated to prevent a single adverse event, than the usually reported relative percent reduction by the experimental treatment. For example, the anticoagulant warfarin was reported to achieve a 69 percent reduction in the annual relative risk of stroke in patients with atrial fibrillation. As shown in Table 5, the absolute reduction was 3.1 percent, from 4.5 to 1.4 percent, with its reciprocal as the NNT of thirty-two. This means that for every patient who benefits from the treatment, thirty-two on average have to be treated, with all thirty-two subject to side effects. For some low-risk patients the NNT is 145. People are far more critical in accepting treatment when results are expressed as NNT, rather than the large relative percent reductions heralded in promotions and the media.
Highlights of History
"One must attend in medical practice not primarily to plausible theories, but to experience combined with reason" (Hippocrates 1923, p. 313).This maxim appears in the Hippocratic Corpus, the writings collected under the name of the Greek physician Hippocrates (c. 460–c. 377 b.c.e.) that became the foundation of Western medicine. Nevertheless until a gradual change beginning around the mid-nineteenth century, medical practice was nearly always based on tradition and authority. Milestones in this transformation were discoveries made by two astute physicians who brought mathematics to medical investigation. One challenged the value of bloodletting, a common treatment dating back to antiquity. The other established the cause of childbed fever, a deadly disease of young mothers that was in fact an infection transmitted by physicians. The work of both met with hostility from the medical community.
PIERRE C. A. LOUIS AND THE NUMERICAL METHOD. By the early-nineteenth century there were large public hospitals in the major cities of Europe, and Paris was leading in the development of pathological anatomy, the use of autopsies to explore changes in the body caused by disease. The French physician Pierre Charles Alexandre Louis (1787–1872) spent years collecting and analyzing data on hospital patients, including the results of autopsies on fatal cases. He called his approach the Numerical Method, which involved tabulating data for groups of patients according to diagnosis and treatment received, and comparing their course of illness and survival patterns. In his major work on bleeding, published in 1835, he studied the effects of bloodletting in series of patients with different diagnoses and found essentially no difference in death rate or duration and severity of symptoms between patients bled and not bled and those bled at different stages of their disease. His findings completely contradicted the teachings of the day and met with sharp criticism, such as the argument that patients could not be compared in groups, because they differed in many respects. Louis reasoned that comparison was being made of essential features, abstracted from the general variability of other factors. The result was a systematic record of what was observed, not the anecdotal evidence of individual physicians who tended to remember the favorable cases. He developed guidelines for designing studies to evaluate different modes of treatment in his Essay on Clinical Instruction (1834).
Louis had great influence on the development of scientific medicine in the United States, because many young Americans were then studying medicine in Paris. One of these was Oliver Wendell Holmes (1809–1894), who in later recollections of Louis described the impact of the change he had observed: "The history of practical medicine had been like the story of the Danaides. 'Experience' had been, from time immemorial, pouring its flowing treasures into buckets full of holes. At the existing rate of supply and leakage they would never be filled; nothing would ever be settled in medicine. But cases thoroughly recorded and mathematically analyzed would always be available for future use, and when accumulated in sufficient number would lead to results which would be trustworthy, and belong to science" (Holmes 1883, p. 432).
IGNAZ SEMMELWEIS: A MEDICAL THEORY BASED ON MATHEMATICS. In July 1846 the young Hungarian physician Ignaz Semmelweis (1818–1865), trained at the medical school of Vienna, then the leading center of medicine in Europe, began work in the maternity clinic of its General Hospital. Confronted with the high death rates from childbed (puerperal) fever that would strike young women and often their babies shortly after childbirth, he undertook with passion to find the real cause of the disease. Occurring in hospitals throughout Europe and the United States, childbed fever was believed to have many different and vague causes, like cosmic-telluric-atmospheric influences and miasmas. In Vienna the first division of the maternity clinic, used for the training of medical students, had much higher mortality rates than the second division staffed by student midwives. Between January and June 1846 the death rate had ranged from 10 to 19 percent, compared with under 3 percent for the midwives, and it remained high as Semmelweis pursued his intense study of patient conditions and autopsies.
Two observations would fuse to spark the flash of insight in May 1847: (1) The staff of the first division, himself included, came to the maternity clinic directly from the dissection room where they had performed autopsies on the diseased patients (unlike the midwives); and (2) A colleague who had died of a wound sustained during a dissection revealed the same lesions on autopsy as the victims of childbed fever. Semmelweis's discovery entailed the recognition that the doctor and the women had died of the same cause, and the infectious material had been transmitted to the patients by the contaminated hands of the examining physicians. Semmelweis ordered all staff to wash their hands in chlorine of lime after autopsies, and immediately the death rate fell. When one woman with an ulcerating cancer of the uterus and another with an ulcerating knee injury gave birth, and in each case most of the patients nearby died of childbed fever, Semmelweis realized that the infectious material could also come from live tissue and be transmitted in the air, so that special precautions were needed for such cases. By 1848 the death rates were 1.27 percent in the first division and 1.33 percent in the second.
In his book The Etiology, Concept, and Prophylaxis of Childbed Fever, published in German in 1861, Semmelweis gave a detailed exposition of his theory, documented with extensive tables. For nearly forty years after the founding of Vienna's General Hospital, from 1784 to 1822, the death rate in the maternity clinic had averaged 1.27 percent. Between 1823 and 1840, after pathological anatomy studies were introduced, the rate rose to 5.9 percent. Then the clinic was split into two divisions, and between 1841 and 1846, the rate was 9.92 percent in the first division and 3.38 percent in the second, the pattern strongly implicating autopsies. Along these same lines, using careful observation, statistical evidence, and clear arguments, Semmelweis systematically eliminated the many other causes that had been proposed for childbed fever over the years.
Semmelweis held that invisible particles in decaying animal-organic matter were the universal necessary cause of childbed fever. Contrary to what had been claimed by others, childbed fever was a transmissible but not a contagious disease, like smallpox. Smallpox always caused smallpox, and every case of smallpox was caused by smallpox. Childbed fever was caused by resorption of decaying animal-organic matter of any source, and the latter could cause infection of any wound surface. Childbed fever was not a distinct disease, but a wound infection. The theory had complete explanatory power; it accounted for every case of the disease and its prevention. It established the etiologic approach to defining disease, the foundation of scientific medicine.
The Semmelweis theory was validated by the French chemist Louis Pasteur (1822–1895), founder of microbiology, who in 1879 identified streptococci as the chief microorganism causing childbed fever, and the English physician Joseph Lister (1827–1912), who introduced antiseptic methods in surgery. The germ theory of disease would follow. If invisible particles in decaying animal-organic matter is replaced by a current phrase containing bacteria, the Semmelweis theory remains valid and it has become a textbook case study in the philosophy of science (Hempel 1966).
Childbed fever is a tragic chapter in the history of medicine, not primarily because of the sad fate of Ignaz Semmelweis. (Suffering some sort of mental breakdown, he died abandoned, under suspicious circumstances, shortly after being committed against his will to a Viennese insane asylum.) Known from antiquity, childbed fever assumed serious proportions when childbirth became a hospital procedure, with doctors replacing midwives. Coupled with the rise of medical research in the autopsy room, progress cost the lives of hundreds of thousands of healthy young women who came to the charity hospitals to deliver. And the real tragedy was how long it took for the old theories to fade after the evidence was in, how long the debate went on about the causes of childbed fever as mothers went on dying. The problem of childbed fever was not definitively solved until the late 1930s, with the introduction of the sulfonamide drugs and then penicillin.
"Quels faits! Quelle logique!" was Pierre C. A. Louis's exasperated response as his critics proclaimed the merits of bloodletting. "Oh Logik!! Oh Logik!!" echoed Semmelweis in the closing paragraph of his great work, urging enrollment in a few semesters of logic before answering the noble call to argue the etiology of disease.
MODERN STATISTICAL INFERENCE. During the nineteenth century probability theory came to be used in the analysis of variation in astronomy, the social sciences, physics, and biology. The intense study of heredity, stimulated by the theory of evolution, spawned the birth of modern statistics around the turn of the twentieth century, associated with the names of Sir Francis Galton (1822–1911), Karl Pearson (1857–1936), and Sir Ronald Fisher (1890–1962). Formal statistical inference, with methods of hypothesis testing and estimation, was gradually introduced across a wide range of disciplines, including medicine.
In his work on the design of experiments in agriculture, Fisher proposed the idea of randomization, to make the experimental plots as similar as possible except for the treatment being tested. Applied to medicine, the approach led to the randomized clinical trial. The first strictly controlled clinical trial using random assignment of patients was set up by the British Medical Research Council in 1946 to evaluate streptomycin in the treatment of pulmonary tuberculosis. The trial was designed by the statistician Sir Austin Bradford Hill (1897–1991), who played a key role in bringing modern statistical concepts to medicine. In the United States randomized clinical trials were introduced in the mid-1950s when Congress authorized the National Cancer Institute to establish the Cancer Chemotherapy National Service Center to coordinate the testing of new compounds as possible anticancer agents. This launched the formation of national cooperative groups that became the mechanism for large-scale clinical trials, with funding provided for related research in statistical methodology.
Biostatistics is a strong academic discipline, with its professionals engaged in teaching and research, and working as consultants and collaborators throughout the healthcare field. The range of developments in theory and methodology—there is now a six-volume encyclopedia—as well as the increasing complexity of biomedical science and technology make the biostatistician an essential member of the research team.
In planning quality studies to assess risk factors of disease or the effectiveness of treatments, questions pertaining to research design, proposed controls, sample size, type of data to collect, length of study, and methods of analysis need to be guided by statistical considerations. Historical, rather than concurrent controls, may be appropriate for new treatment of a rare, usually fatal disease. In a randomized clinical trial, stratified randomization may be used, where patients are assigned at random within subgroups known to affect prognosis (for example, menopausal status in breast cancer). There are methods to assess the effect of multiple risk factors on outcome, such as Cox regression, logistic regression, and loglinear analysis. The essential means of modern analysis is provided by electronic database management and statistical software systems.
In approaches to statistical inference there is lively interest in Bayesian methods and decision theory. Within medicine there are the movements of outcomes research, to explore the effectiveness of medical interventions in the general population, and evidence-based medicine, to make more effective use of the medical literature in everyday practice. A related area is meta-analysis, which seeks to combine the results of published studies to obtain the best possible assessment of risk factors and treatments. Evaluating alternative medicine has become a pressing issue. The broader field of health services research also studies the cost-effectiveness of medical procedures.
Biostatistics and Ethics
The Hippocratic maxim, "Help or at least do no harm," has for 2500 years been the basis of medical ethics. How this can be done is explained by the Hippocratic precept cited earlier. To this end, medical practice should be based on experience combined with reason, namely, carefully collected observations (experience) analyzed with the tools of scientific methodology (reason). Biostatistics has assumed this function, and played a significant role in the great achievements of medical science and technology. Since the closing decades of the twentieth century, it has been faced with a crisis in U.S. (and Western) medicine, as the costs of health care spiral out of control.
Important advances include antibiotics and immunization, control of diabetes and hypertension, treatments for heart disease, cancer, and psychiatric disorders, diagnostic imaging, neonatal and trauma medicine, biomechanics, and organ transplants, with research continuing unabated on every front. But past successes have led many to unrealistic expectations of perpetual progress, putting them at risk for exploitation by a profit-driven healthcare industry. Medical technology tends to be oversold by the market, and an often poorly informed, vulnerable public is buying. Promotion in the media focuses on conditions that affect large segments of the population, such as chronic pain, which requires safe and effective individualized treatment for adequate control.
DEBATE OF MARKET VS. SCIENCE. In September 2004 the arthritis pain medication Vioxx, with sales of $2.5 billion in 2003, was withdrawn from the market by its manufacturer Merck because of findings of an increased risk of heart attacks and strokes. This triggered charges that the company had ignored earlier warnings, and the rival drugs Celebrex and Bextra, made by Pfizer, also came under scrutiny. Although helpful for many, these Cox-2 inhibitor agents did not claim greater effectiveness, only fewer gastrointestinal side effects than older alternatives like aspirin, ibuprofen, and naproxen. In the absence of adequate comprehensive studies, controversy continued concerning the relative risks and benefits of the various agents and the indications for their use. The larger debated issue is that of postmarketing surveillance (safety monitoring of drugs after release on the market), and the role of the Food and Drug Administration (FDA). The high cost of new drugs like Vioxx, challenged by medical critics, raises a further ethical concern. It is not only the physical harm done to so many, but the emotional and financial harm to all those struggling on limited means.
The individual must be more assertive in asking questions: Is this drug treatment necessary? What is the effectiveness (NNT) of the drug for a patient with the given characteristics? What are the side effects for this class of patient and how long is the follow-up of observation? Is there a less expensive, better-evaluated alternative? What are the interactions of the drugs the patient is taking? All drugs have side effects, and harmful effects of legally prescribed drugs are estimated to cause over 100,000 deaths in the United States each year. Ultimately it is up to the public to demand answers.
THE ETHICS OF EVIDENCE. An approach has been proposed for dealing with medical uncertainty, called the Ethics of Evidence. (Miké 1999, 2003). It can be expressed in two simple rules or imperatives: The first calls for the creation, dissemination, and use of the best possible scientific evidence as a basis for every phase of medical decision making. Complementing it, the second focuses on the need to increase awareness of, and come to terms with, the extent and ultimately irreducible nature of uncertainty.
There is a need for greater insight and closer involvement on the part of the public. Biostatistics can help to discern what is necessary, safe, and effective treatment, and should be fully utilized to produce the best available evidence. But even when it is properly used, uncertainties remain that are intrinsic to the techniques themselves and the limitations of medical knowledge. Most major diseases do not have a single cause, but result from the complex interplay of genetic and environmental factors. Systematic study of individual risk factors and their interactions must continue, in the search for better prevention and control. When Semmelweis made his great discovery, the numeric results were so dramatic that no formal statistical procedures were needed (and they did not yet exist). In the early twenty-first century it is a slow, incremental process to find and confirm small improvements. The real promise for medicine in the near future points to changes in lifestyle.
A study released in July 2004 estimates that 195,000 Americans die each year as a result of preventable medical error, and the data pertain only to hospitals. More open and direct participation of patients in their own treatment would help reduce error rates, keep in the forefront questions about the safety and effectiveness of proposed interventions, and curb the reflexive urge for malpractice litigation. An alert, educated public has a realistic view of medicine and does not expect it to solve all of life's problems. But it insists on well-funded biomedical research and its careful assessment, with effective government policies in place to ensure the best possible healthcare for all.
American Cancer Society. (2004). Cancer Facts and Figures 2004. Atlanta, GA: American Cancer Society. Annual report issued by the American Cancer Society, also available online at http://www.cancer.org/docroot/STT/content/STT_1x_Cancer_Facts__Figures_2004.asp.
Angell, Marcia. (2004). The Truth about the Drug Companies: How They Deceive Us and What to Do About It. New York: Random House. Documented critique of the pharmaceutical industry by a former editor of the New England Journal of Medicine.
Armitage, Peter, and Theodore Colton, eds. (1998). Encyclopedia of Biostatistics, 6 vols. New York: John Wiley & Sons.
Carter, H. Ballentine. (2004). "Prostate Cancer in Men with Low PSA Levels—Must We Find Them?" New England Journal of Medicine 350: 2292–2294.
Carter, K. Codell, and Barbara L. Carter. (1994). Childbed Fever: A Scientific Biography of Ignaz Semmelweis. Westport, CT: Greenwood Press.
Dawson-Saunders, Beth, and Robert G. Trapp. (1994). Basic and Clinical Biostatistics, 2nd edition. New York: McGraw-Hill/Appleton & Lange.
Evidence-Based Medicine Working Group. (1992). "Evidence-Based Medicine: A New Approach to Teaching the Practice of Medicine." Journal of the American Medical Association 268: 2420–2425.
Gehan, Edmund A., and Noreen A. Lemak. (1994). Statistics in Medical Research: Developments in Clinical Trials. New York: Plenum Medical Book. History of clinical trials, with emphasis on statistical aspects.
Hempel, Carl G. (1966). Philosophy of Natural Science. Englewood Cliffs, NJ: Prentice Hall. Includes case study of Semmelweis discovery.
Hill, Austin Bradford. (1967). Principles of Medical Statistics, 9th edition. New York: Oxford University Press. A classic text of biostatistics by one of its pioneers. Includes discussion of criteria for establishing causality in observational studies.
Hippocrates. (1923). "Precepts." In Works, Vol. 1, trans. W. H. S. Jones. London: Heinemann.
Holmes, Oliver Wendell. (1883). "Some of My Early Teachers: A Farewell Address to the Medical School at Harvard University, November 28, 1882." In Medical Essays: 1842–1882. Boston: Houghton, Mifflin.
Kohn, Linda T.; Janet M. Corrigan; and Molla S. Donaldson, eds. (2000). To Err Is Human: Building a Safer Health System. Washington, DC: National Academy Press. Includes discussion of large studies of medical errors in hospitals.
Louis, Pierre C. A. (1834). An Essay on Clinical Instruction, trans. Peter Martin. London: S. Highley. Original in French.
Miké, Valerie. (1988). "Modern Medicine: Perspectives on Harnessing Its Vast Emerging Power." Technology in Society 10: 327–338. Analysis of the need for better assessment of medical technology and related issues.
Miké, Valerie. (1999). "Outcomes Research and the Quality of Health Care: The Beacon of an Ethics of Evidence." Evaluation & the Health Professions 22: 3–32. Commentary by Edmund D. Pellegrino, "The Ethical Use of Evidence in Biomedicine," is included in this issue.
Miké, Valerie. (2003). "Evidence and the Future of Medicine." Evaluation & the Health Professions 26: 127–152. Further development of the Ethics of Evidence.
Miké, Valerie, and Robert A. Good. (1977). "Old Problems, New Challenges." Science 198: 677–678. Introduction to Birnbaum Memorial Symposium "Medical Research: Statistics and Ethics." International multidisciplinary conference with focus on ethical issues in biostatistics. Proceedings published in Science.
Miké, Valerie, and Ralph C. Marcove. (1978). "Osteogenic Sarcoma under the Age of 21: Experience at Memorial Sloan-Kettering Cancer Center." In Immunotherapy of Cancer: Present Status of Trials in Man, eds. William D. Terry, and Dorothy Windhorst. New York: Raven Press. Example of a clinical trial where use of a historical control was the appropriate procedure.
Miké, Valerie, and Kenneth E. Stanley, eds. (1982). Statistics in Medical Research: Methods and Issues, with Applications in Cancer Research. New York: John Wiley & Sons. Includes multidisciplinary discussion of clinical trials, including ethical aspects.
Redmond, Carol, and Theodore Colton, eds. (2001). Biostatistics in Clinical Trials. New York: John Wiley & Sons. Includes articles on Bayesian methods and NNT.
Shryock, Richard Harrison. (1969). The Development of Modern Medicine: An Interpretation of the Social and Scientific Factors Involved. New York: Hafner Publishing.
Semmelweis, Ignaz. (1983). The Etiology, Concept, and Prophylaxis of Childbed Fever, trans. K. Codell Carter. Madison: University of Wisconsin Press. Originally published in German as Die Aetiologie, der Begriff, und die Prophylaxis des Kindbettfiebers (1861).
Topol, Eric J. (2004). "Good Riddance to a Bad Drug." New York Times, October 2, p. A15. Op-Ed article by chairman of department of cardiovascular medicine at Cleveland Clinic, who in 1991 published results on cardiac risks of Vioxx.
Health Grades Inc. "Patient Safety in American Hospitals." Health Grades. Available from http://www.healthgrades.com/PressRoom/index.cfm?fuseaction=PressReleases. July 2004 study.
Biostatistics in the public health context consists primarily of developing descriptive statistics describing the overall health and well being of a population. These statistics include such measures as birth, death, and infant death rates; disease incidence and prevalence; and trends of this data over time. Proper adjustment of these rates so as to allow for correct interpretation of and comparison among populations also falls within the purview of the public health biostatistician. The biostatistician works closely with other public health disciplines to develop outcome measures to ascertain the effectiveness of programmatic activities and to develop the means to collect such measures, which may include surveys, lab reports, and hospital discharge data.
(see also: Birth Certificates; Data Sources and Collection Methods; Statistics for Public Health; Vital Statistics )