CHAPTER 1
Basics of Biomedical and Clinical Research
Natural phenomenon while important in shaping clinical decisions or biomedical thinking remains subjective and non- generalizable knowledge. Therefore, to improve the care of future patients we must apply valid research methodology to natural phenomenon observation in order to obtain reliable inference that will guide practice beyond the sample of patients we studied.
—L. Holmes, Jr., Santa Monica, California, March 2013
Highlights: Clinical and biomedical research rationale, research question and hypothesis, sampling, sample size and power estimations, generalizability, screening test, NNT, NNH
INTRODUCTION
Biomedical and clinical research remain tools to understanding disease pathways, treatment modalities, and outcomes of care. While knowledge of biomedical sciences and clinical medicine is significant for advances in this field, the generation of such knowledge requires solid and reliable design processes as well as adequate statistical techniques. Biomedical and clinical research is conducted primarily to enhance therapeutics, implying the in- tent to improve patients' care. The application of this concept in biomedical sciences, public health, and clinical medicine signaled a departure from nihilism, which claimed that disease improved without therapy. The scientific medical discoveries on pellagra, diabetes mellitus, and antibiotics like penicillin and sulfonamide provide reliable data on medicinal benefits in therapeutics. Today, with biomedical and clinical research, clinical investigators applying reliable and valid research methodologies can demonstrate the efficacy and effectiveness of agents and devices, competing therapies, combination treatments, comparative effectiveness, and diagnostic and screening criteria for most diseases.
In claiming the advantage of therapeutics in medicine (complementary versus traditional), there is a need to understand the biological theories and the complexities of disease among clinical investigators, who may be expert physicians, as well as other health-care providers and those who are indeed researchers. While understanding the biological and clinical importance of a disease is essential in formulating the research question, the clinician is also expected to acquire statistical reasoning. The combination of these two models enhances the analysis and the interpretation of the data from clinical research. Therefore, in clinical research, there is an investigator (clinical) who examines the formal hypothesis or establishes the biology based on work in the clinical settings (experience, observation, and data), as well as another investigator (biostatistician/epidemiologist) whose contribution is to generalize observations from sample to target population, as well as combine empirical (observation and data) and theory-based knowledge (probability and determinism) with the understanding of the results of the study. Despite these distinctions, effective clinical and biomedical research involves the understanding of these two models of thinking or reasoning by the investigators, clinicians, and epidemiologists/biostatisticians. Without this integration, our effort toward the design and interpretation of research findings is limited, since making reasonable, accurate, and reliable inferences from data in the presence of uncertainty remains the cornerstone of clinical research results utilization for improving the health care of future patients. In stressing the essence of this integration, one is not claiming the relevance of statistical reasoning over biological and clinical importance, since clinical research thinking is fundamentally biologic, clinical, and statistical.
The approach to biomedical and clinical research involves research conceptualization, the design process, and statistical inference. In biomedical sciences, for example, the research conceptualization may involve therapeutics in mice or rats with cancer, and because of the similarities to human malignancies, these findings would be translational, and hence generalization (biologic) to human malignancies without a formal statistical model can be made. The design process may involve treated and untreated mice, with a follow-up time to determine the survival difference in the two groups. The statistical inference, given that adequate numbers (sample size and power estimations) of mice were studied, involves the use of Kaplan-Meir's survival estimates, as well as the log rank test for the equality of survival in these two groups. Finally, statistical stability is examined in ruling out random variation using p value (significance level) and 95 percent confidence interval (precision). A similar approach is utilized in clinical research that involves human subjects or patients in clinical settings. The research conceptualization in this context involves the clinical investigator or clinician utilizing his or her experience in the management of patients with malignancy (leukemia, for example), observation, and data to formulate hypotheses regarding therapeutics. A case-comparison/control design could be applied here in which the treated-group (cases) are placed on the new drug X, while comparison (control) are placed on a standard care drug Y and both are followed for the assessment of outcome (death or biochemical failure). The statistical inference and the interpretation of the results are similar to the example with mice and malignancy therapeutics. There are excellent books in study designs, including one by the authors of this book.
Historically, central to clinical research and therapeutics is the concept of disease screening and diagnostic testing. We can view the disease diagnosis as well as the diagnostic test as key elements in the ascertainment of subjects for clinical research. Inappropriate patient ascertainment may result in selection, information, and misclassification biases (discussed in subsequent chapters). This historical concept remains valid in research conduct and is the main material elaborated in this chapter. The sensitivity, specificity, predictive values, and likelihood ratios are described with examples. Thus, the validity of the results obtained in clinical research depend on how adequately the subjects were identified and assigned to treatment (experimental design or clinical trial) or followed after exposure (nonexperimental design).
WHY CONDUCT CLINICAL RESEARCH?
Conducting research in biomedical, clinical, and public health involves a response to a health or health-related issues. Thus, research conducted commences with question formulation, which is followed by design plans to answer the question, data collection and analysis, the drawing of conclusions from the results or findings, and then information sharing through publication or dissemination. For example, a reason to conduct research may be to understand the natural history of a disease, such as unicameral bone cyst, which is a benign tumor of the bone. Another example of the natural history of a disease is the Swedish prospective cohort study of men diagnosed with prostate cancer (CaP) and followed for ten years without specific treatment (watchful waiting or observational management) for CaP. In the latter example, with the primary end point being cause-specific mortality (prostate cancer dead), the experience of this group was compared and a ten-year relative survival of 87.0 percent was reported. The natural history of disease refers loosely to the collectivity of the actual sequence of events since this phenomenon (actual sequence of events) can vary widely among patients. In more concrete terms, we normally refer to the natural history of a disease as the assessment of the actual sequence of events for many patients in order to obtain some estimates of these events. In this respect, the natural history of a disease can be characterized using measures of disease occurrence, such as case fatality, mortality rate, median survival time, etc. Research may also be conducted to relate laboratory data or information with screening, diagnosis, treatment, and prognosis. A researcher/investigator may wish to use the laboratory value for blood glucose level, for example, to screen, diagnose, and determine the prognosis of diabetes mellitus. Finally, a natural history of a disease may be studied in a randomized, placebo-controlled clinical trial, where treatment is allocated to one group, while the other group (control) is given the placebo. The result in the control group without the treatment represents the natural history of the disease studied.
The primary reason to conduct research in clinical medicine is to address questions pertaining to screening, diagnosis, treatment, and outcome of care (prognosis), with the ultimate goal being the improvement of patients' care. This effort involves protocol development and management/coordination, recruitment and data collection/entry, data management and analysis, and making sense of the data through interpretation and inference.
The biomedical or clinical researcher should have a clear idea of the concept to be measured. This step allows the investigator to clearly address the research questions. The statement of such questions must reflect the scale of measurement of the variable—such as nominal, ordinal, interval, or ratio. The reliability of the variables to be measured (if questionnaires are used to collect information) has to be examined on the basis of the stability of the response to the question over time—a sort of test and retest reliability. As is often seen in clinical research involving radiographic measures, reliability could be measured by examining the agreement between two observers or surgeons (interrater reliability). Very important is the validity or accuracy of the measure, which is simply the extent to which the empirically observed association between measures of the concepts agree with the testable hypothesis about the association between variables assessed.
STUDY SUBJECTS
Biomedical research may involve animals as well as human subjects. For example, biomedical researchers may be interested in finding out if a certain drug (n34) enhances programmed cell death (apoptosis) in transgenic mice induced with neuroblastoma. A case-control nonexperimental epidemiologic design may be proposed for this investigation. The issues to be addressed include study subject selection since the control (mice not administered n34) must be comparable to the treatment mice (n34 mice). Since these mice are genetically homogenous, such selection is feasible, minimizing selection bias, sampling and generalization errors.
SUBJECT SELECTION
Investigators must select samples that are representative of the patient population, implying performing the investigation on number of patients large enough to minimize random error (increased sample representativeness) in the generalization of the study findings to the targeted population of patients. One must stress the importance of sampling design since the generalizability of the results of a study is dependent on the accuracy of the sampling design. Additionally, inference remains invalid if drawn from an erroneous sampling design. An example of a study sample would be children with adolescent idiopathic scoliosis (AIS) who have undergone posterior spinal fusion for curve deformities correction between 2000 and 2011. Well-structured inclusion and exclusion criteria are essential in appropriate subject selection. The purpose of this is to ensure that the study findings are reasonably generalized. Therefore, criteria should be selected in such a manner that the generalization of the study findings to the targeted population is feasible.
SAMPLING
Biomedical and clinical researchers must determine a priori who should be in the sample. Reason behind sampling, among others (research complexity and efficiency—limited resources) include increasing precision (minimize sampling variability) and ensuring accuracy of the estimates, such as the mean or proportion. Commonly used probability sampling techniques include the simple random sample, systematic random sample, and stratified sample, as well as the cluster sample. It is however important to note that while appropriate probability sampling technique is essential in lessening sampling variability, completely eliminating it is impossible, and there remains the possibility of random variability, hence the need to quantify such errors (random) by probability value (p).
The utility of findings from an inferential study depends on appropriate sampling. Sampling as a desirable approach in research is based on the rationale of appropriate samples representing the target population. While study size may be influenced by the available resources, the study sample must reflect the characteristics of the targeted population. A sample is described as a subset of the targeted population, which is always desirable, given the impossibility of studying the entire population. In clinical research, the study sample rarely meets the requirement for probability sampling. In this context, convenient (subjects who meet criteria and are accessible to investigators) and consecutive (the entire patient population over a long period of time) samples are often used.
Inferential studies that quantify random error require probability sampling. This approach ensures that the study sample represents the targeted population and that data derived from such sampling techniques reflect the true experience in the population that the study sample was drawn from. These techniques include the simple random sample, stratified sample, systematic sample, and cluster sample.
GENERALIZATION
When studying physical phenomena, we can apply the findings easily without determining how to reasonably apply the findings to geographic locations in which the phenomena were not observed. However, biologic or biomedical studies differ because of the heterogeneity of species and the changing environmental conditions. Can we generalize the findings of a study based on a consecutive sample? This question requires the investigators' determination as to whether the sample is comparable to the probability sample to justify generalization. For example, if the investigator estimates that the consecutive sample was large enough to minimize random error (representative sample), then such a study finding could be generalized to the target population with the assumption that the sampling technique used is similar to the probability sampling technique. In contrast, if the consecutive sample is judged not to minimize random error, such a finding should not be generalized. These studies' results should be presented with descriptive statistics, without any attempt to quantify random error.
SAMPLE SIZE AND POWER ESTIMATIONS
While this issue is discussed in detail in subsequent chapters, the importance of understanding the essence of the study size needed for a research project needs to be mentioned early. Sample size estimation is important at the beginning of the design of a research project. In inferential or analytic studies, the findings require generalization. This process involves a clear statement of the null and alternative hypotheses, indicating the direction of the test, selecting the test statistic based on the scale of measurement of the outcome and independent or predictor variables, clinically reasonable effect size, variability, and the statement of type I error tolerance and type II error. Given the importance of power, adequate sample size is necessary in order to avoid missing a real difference and concluding that there is no difference. Subsequent chapters will provide details in specific statistical test settings (t-test, chi-square, correlation coefficient, logistic regression, survival analysis) for how to estimate the study size. However, researchers must plan to address loss to follow-up by compensating for the attrition rate, as well as to increase the study size while utilizing multivariable statistical modeling to adjust for confounding at the analysis phase of the research.
SCREENING (DETECTION) & DIAGNOSTIC (CONFIRMATION) TESTS
Clinical or medical research also involves designs that examine the effect of testing on outcome. Simply, a screening or diagnostic test is beneficial if survival is prolonged among those screened for the disease compared to those who are not screened. Remarkably, the outcome of diagnostic tests is not the mere diagnosis or disease stage or grade of tumor, as in the Gleason score used in prostate cancer clinical assessment, but involves the mortality or morbidity that could be prevented among those who tested positive for the disease.
