Concise Biostatistical Principles and Concepts: Guidelines to Study Analysis, Interpretation and Application
เกี่ยวกับ eBook เล่มนี้
Clinical medicine or surgery continues to make advances through evidence that is judged to be objectively drawn from the care of individual patients. The natural observation of individuals remains the basis for our researchable questions’ formulation and the subsequent hypothesis testing. Evidence-based medicine or surgery depends on how critical we are in evaluating evidence in order to inform our practice. These evaluations no matter how objective are never absolute but probabilistic, as we will never know with absolute certainty how to treat future patients who were not a part of our study. Despite the obstacles facing us today in an attempt to provide an objective evaluation of our patients, since all our decisions are based on a judgment of some evidence, we have progressed from expert opinion to the body of evidence from randomized controlled clinical trials, as well as cohort investigations, prospective and retrospective.
The conduct of clinical trials though termed the “gold standard”, which yields more reliable and valid evidence from the data relative to non-experimental or observational designs, depends on how well it is designed and conducted prior to outcomes data collection, analysis, results, interpretation, and dissemination. The designs and the techniques used to draw statistical inferences are often beyond the average clinician’s understanding. A text that brings hypothesis formulation, analysis, and how to interpret the results of the findings is long overdue and highly anticipated. Statistical modeling which is fundamentally a journey from sample to the application of findings is essential to evidence discovery.
The four past decades have experienced modern advances in statistical modeling and evidence discovery in biomedical, clinical, and population-based research. With these advances come the challenges in accurate model stipulation and application of models in scientific evidence discovery. While the application of novel statistical techniques to our data is necessary and fundamental to research, the selection of a sample and sampling method that reflects the representativeness of that sample to the targeted population is even more important. Since one of the rationale behind research conduct is to generate new knowledge and apply it to improve life situations including the improvement of patient and population health, sampling, sample size, and power estimations remain the basis for such inference. With the essential relevance of sample and sampling technique to how we come to make sense of data, the design of the study transcends statistical technique, since no statistical tool no matter how sophisticated can correct the errors of sampling.
This text is written to highlight the importance of appropriate design prior to analysis by placing emphasis on subject selection and probability sample, randomization process when applicable prior to the selection of the analytic tool. In addition, it stresses the importance of biological and clinical significance in the interpretation of study findings. The basis for statistical inference, implying the quantification of random error is a random sample. When studies are conducted without random samples as often encountered in clinical and biomedical research, it is meaningless to report the findings with p value. However, in the absence of a random sample, the p-value can be applied to designs that utilize consecutive samples, and disease registries, since these samples reflect the population of interest, and hence representative sample, justifying inference and generalization.
Essential to the selection of test statistics is the understanding of the scale of the measurement of the variables, especially the response, outcome or dependent variable, type of sample (independent or correlated), hypothesis, and normality assumption. In terms of the selection of statistical tests, this text is based on the scale of measurement (binary), type of sample (single, independent), and relationship (linear). For example, if the scale of measurement of the outcome variable is binary, repeated measure, and normality is not assumed, the repeated measure logistic regression model remains a feasible model for evidence discovery in using the independent variables to predict the repeated outcome.
This book presents a simplified approach to evidence discovery by recommending the graphic illustration of data and normality test for continuous (ratio/interval scale) data prior to statistical test selection. Unlike current text in biostatistics, the approach taken to present these materials is very simple. First, this text uses applied statistics by illustrating what, when, where, and why a test is appropriate. Where a text violates the normality assumption, readers are presented with a non-parametric alternative. The rationale for the test is explained with a limited mathematical formula and is intended in order to stress the applied nature of biostatistics.
Attempts have been made in this book to present the most commonly used statistical model in biomedical or clinical research. We believe since no book is complete to have covered the basics that will facilitate the understanding of scientific evidence discovery. We hope this book remains a useful guide, which is our intention in bridging the gap between theoretical statistical models and reality in the statistical modeling of biomedical and clinical research data. As researchers we all make mistakes and we believe we have learned from our mistakes during the past three decades hence the need to examine flaws and apply reality in the statistical modeling of biomedical and research data. We hope this text results in increased reliability in the conduct, analysis,
เกี่ยวกับผู้แต่ง
Dr. Laurens Holmes, Jr., educated at the Catholic University of Rome, Italy, University of the Health Sciences, Antigua, School of Medicine, University of Amsterdam, Faculty of Medicine, and the University of Texas, Texas Medical Center, School of Public Health, Laurens (Larry) Holmes, Jr., is a former Chief Epidemiologist (Orthopedic Department), former head of the Epidemiology Laboratory at the Nemours Center for Childhood Cancer Research, former Senior Principal Research Scientist at the Nemours / A.I. duPont Children’s Hospital, Office of Health Equity & Inclusion. He is also an Affiliate Professor of clinical trials and molecular epidemiology in the Department of Biological Sciences, University of Delaware, Newark, Delaware. He is recognized for his work on epidemiology and control of prostate cancer, and pediatric acute lymphoblastic leukemia (ALL), but has also published scientific papers on other aspects of hormone-related malignancies, cardiovascular and chronic disease epidemiology, utilizing various statistical models, including log binomial family, exact logistic model, probability estimation from logistic model by margin and Signal Amplification and Risk Specific Stratification model (SARSSmodel), single proponent of this model, SARSS.
Dr. Holmes is a strong proponent of reality in the statistical modeling of cancer and experimental as well as non-experimental research data, where he presents the rationale for tabular analysis in most non-experimental research data, which are often not randomly sampled (probability sampling) rendering statistical inference application meaningless to such data, as sampling variability, and not the point estimate (effect size), as a clinically and biologically meaningful difference. Since controlling for known confounders of variabilities in subgroup health and healthcare outcomes often fails to remove or explain these imbalances, a feasible alternative is to consider subgroup biologic/cellular events/molecular level variances or differences. One of the Biostatistical approaches to address valid inference in population-based, biomedical and clinical research is a model that amplifies signals in the data prior to risk stratification, implying the role of biostatistics as a tool in scientific evidence discovery, had been proposed by Dr. Holmes – “Signal Amplification and Risk Specific Stratification, SARSSm” Professor Holmes is currently a Distinguished Professor at FAMU, Institute of Public Health, Tallahassee, FL., USA.
