CHAPTER 1
Research Process: Overview
To do successful research, you don't need to know everything;you just need to know of one thing that isn't known.
—Arthur Schawlow
Highlights: notion of scientific research, research questions and hypotheses; purpose of researchand research objectives; outcome and independent variables; clinical versus population medicine/health; epidemiologic methods
INTRODUCTION
Scientific research is a systematic, controlled, empirical, and critical investigation of naturalphenomena guided by theories and hypotheses about the presumed relations among them.Consequently, research implies an organized and systematic way of finding solutions toquestions. This attempt requires that solutions to the postulated questions be approachedwhen it is feasible, interesting, novel, ethical, and relevant. Clinical research thus involvesthree basic elements:
1. conceptualization,
2. design process, and
3. statistical inference.
As will be discussed further in this text, fundamental thinking in clinical research involvesthese elements:
a) biologic relevance,
b) clinical importance, and
c) statistical stability.
The statistical inference allows conclusions to be drawn from data because the entire populationis never studied. The combination of this tripartite approach to evidence discovery from thedata remains the cornerstone of valid and reliable results from clinical studies and theircautious interpretation in the attempt to improve patient care.
Conducting research involves planning, and that requires the measurement and quantification ofthe variables in the study, careful administration of well-designed instruments, data collection,appropriate analysis, and the interpretation of results. Depending on the research question andthe type of design, this process could be very time-consuming and complex. For example, inclinical trials, an elaborate protocol for participants' enrollment, randomization, and treatmentadministration are used to ensure appropriate documentation of events during the trial.
The materials in this chapter will enable readers to understand research conceptualizationand the distinction between clinical, population medicine and epidemiology as well as theirimplications in epidemiologic/population-based, clinical and biomedical research. Thischapter, as a brief overview of research processes, focuses on the description of terms appliedin research conceptualization and provides practical examples—process of research conduct,research objectives, purpose of research, research questions, hypotheses, and the descriptionof clinical and population medicine within a research context.
STRUCTURE AND FUNCTION OF RESEARCH
Conducting research involves several processes:
(a) A well-defined problem statement, research questions, purposes, and potential benefits ofthe study to science, society, and humanity
For example, one may formulate a research question around the effectiveness of treatmentin improving some pathologic conditions—does cervical spine surgery stabilize the spine inchildren with skeletal dysplasia (SKD)? Clinicians and biomedical researchers should realizethat research questions are not study topics, since such topics are generally broad and researchquestions must be very specific. Specifically, research questions should be formulated in sucha way that they can be answered by observable evidence. Therefore, unless the research forthe question is feasible, it cannot answer the question posed in a measurable manner.
(b) Identification of the theory and assumptions and a search of background literature toaddress the magnitude of the problem
Such a literature review is aimed at identifying what has been done in terms of previousresearch, the gap in literature, and what the proposed study intends to add or contribute to theexisting body of knowledge in the field. It is a common mistake made by novice researchers toavoid the intensive literature review of the subject of their interest until the research questionand study designs are formulated. We caution against such an approach since there is apossibility that such questions have already been answered. It may lead to a study that is notnovel and hence will have very little to offer to science in medicine.
(c) Measurement of variables and statement of researchable hypotheses
Research is basically an exercise with measurement, which is always subject to errors. Goodresearch aims therefore at minimizing measurement error, thus reducing variability in whatis being measured.
(d) Identification of the appropriate study design and methodology to fit the research questionor hypothesis
For example, in the previously mentioned illustration with cervical spine surgery and SKD,the selected design may be the one that involves the administration of surgery to all patientswith SKD, which makes it a single sample design with repeated measure. The design couldbe prospective or retrospective (these terms will be defined and clarified later in the text).
(f) Development of an instrument for data gathering and identifying an appropriate samplingtechnique
Since the generalization of findings requires probability sampling of study subjects prior tobaseline or preoperative data collection, the sample must be drawn from the population ofpatients with the same condition. Practically, every sample in the study must have a nonzeroprobability of being selected for the study.
(g) Selection of data analysis tools or statistical techniques to test the hypothesis or answerthe research question and presentation of results and interpretation
For example, in the illustration with SKD, the null hypothesis may be that cervical spinesurgery does not stabilize the spine in children with SKD. In testing this hypothesis, one canuse a paired t-test or repeated measure analysis of variance ( ANOVA) if two or more than twomeasurement times were used and the outcome was measured on a continuous scale. Thesetests will be explained later in the text.
(g) Offering conclusions that are based on the data as well as what the data suggests andrecommendations
OBJECTIVE OF STUDY/RESEARCH PURPOSE
Research conceptualization begins with the purpose of the study, which is a statement thatdescribes the intent and the direction of the study. The purpose statement or purposeof a study is the rationale behind the study. Although the purpose of the study is oftenused interchangeably with the research question or the problem statement, they are distinctentities and should be very clearly differentiated in the proposal development and manuscriptpreparation phases of the study. Simply, the purpose of the study describes the objectives,intent, and aims of the study. The research question therefore remains a specific statementthat needs to be answered, following the hypothesis testing in quantitative research.
In clinical, epidemiologic, or biomedical research, the purpose of a study is the statement of theoverall objective of the study. This statement identifies the independent and response variablesas well as how the variables will be measured and the design to be used to achieve the expectedrelationship, correlation, or association. In published articles, we often use words and phrasessuch as relationship, mean comparison, effectiveness, efficacy, and association to express thenexus or link between the response, outcome, or dependent variable and the independent,predictor, explanatory, or antecedent variable. For example, in the previous illustration, thepurpose of the study may be stated as, "To examine the effectiveness of cervical spine surgeryin stabilizing the spine in children with SKD."
The objective of a study is a concise statement describing the intent of the research. Suchobjectives can be evaluated from several standpoints or dimensions. For example, if theobjective of a study is to assess the role of electrocautary in the development of deep woundinfection in pediatric patients with neuromuscular scoliosis and electrocautary cannot bemeasured directly but is dependent on an "indirect measure termed proxy," then it is not a goodstudy objective. Therefore, in clinical, experimental, and nonexperimental studies, a goodstudy objective is that which is measurable. For example, if the objective of a study was toexamine the effectiveness and safety of posterior spine fusion in correcting curve deformitiesand maintaining correction in children with leukodystrophy (degenerative disease of thewhite matter of the brain), the measurement of the effectiveness would be the postoperativereduction in the major curve angles (thoracic and thoraco-lumbar curves), while safety willbe measured by less complications (psuedoathrosis, instrument failure, and screw pull out)following surgery.
In practice, while the purpose of the study may not be easily differentiated from the objective,the research question remains a concise statement about the study objective. A researchquestion's purpose is to imply what issue the research will address. Similarly, the purposeof the study is a broad scope of what the study intends to accomplish in terms of benefits tosociety, medicine, and science.
RESEARCH QUESTIONS AND STUDY HYPOTHESES
Research Questions
Fundamental to clinical research is a clear statement of the question proposed to be answered.This question should be carefully selected and clearly defined prior to beginning to conductresearch. For example, a research question may be framed in the context of trying to determinewhether a new treatment, compared with the current or standard treatment, reduces the riskof coronary heart disease (CHD). Research questions could be primary or secondary. Theprimary question refers to the main outcome of the study. For example, "Does angiotensinconverting enzyme (ACE) inhibitor reduce the incidence of CHD?" A secondary researchquestion aimed at a secondary outcome may assess physical activities or look at racial/ethnicvariation in the response to ACE inhibitor. For example, the secondary research questionmay be stated as, "Are there racial/ethnic differences in CHD incidence reduction followingACE inhibitor use?" Yet another example of a primary research question may be, "DoesADT prolong the survival of elderly men diagnosed with prostate cancer ?" and the secondaryquestion may be, "Is there a difference in CaP survival in elderly men diagnosed with loco-regionalversus metastatic disease?"
Depending on the research question, participants may be randomly assigned to differenttreatments or procedures. This approach, which is called a human experiment or clinical trial,will be discussed at some length in the upcoming section on design. If a clinical trial is notfeasible (ethical considerations), the researcher may collect data on participants (independent,dependent, or response variables and confounding) and conduct a nonexperimental study.The information on confounding variables is collected in order to statistically control for theinfluence of the confounding on the response, dependent, or outcome variable (nonexperimentalstudies' results may be influenced by selection biases and confounding if not minimized andcontrolled for respectively). If the intent is to generalize from the research participants to alarger population, as is often the case in epidemiologic (nonexperimental) and experimentalstudies, the investigator will utilize probability sampling to select participants (samplingtechnique). This approach allows every study subject the same probability of being selectedfrom the general population for inclusion in the study sample and the variables to be studiedto be termed "random variables" (suitable for hypothesis testing). In such a situation,statistical inference can then be applied to the study, implying the generalization of thefindings beyond the sample studied to the larger population.
Study Hypothesis
Hypothesis testing is essential in quantitative study designs. The hypothesis of a study isa statement that provides the basis for the examination of the significance of the findings.Whenever an investigator wishes to make a statement beyond the sample data (descriptivestatistics) or simply draw an inference from the data, hypothesis testing is required. A hypothesisis a statement about the population or simply the universe/world that is testable. An example ofa hypothesis statement could be, "Selenium in combination with vitamin D decreases the riskof prostate cancer." To assess this association, the investigator needs to examine the data forevidence against the null hypothesis. If the data provide evidence against the null hypothesis(no association), the null hypothesis is rejected; in contrast, if data fail to provide evidenceagainst the null hypothesis, the null hypothesis is not rejected, thus negating any inclinationto the alternative hypothesis of association between the response and independent variable.While hypotheses will be described more in this text's companion, the biostatistics, it servesto note here that hypothesis testing includes the following:
a) generation of the study hypothesis and the definition of the null hypothesis,
b) determination of the level below which results are considered statistically significant,implying ? level 0.05, and
c) identification and selection of the appropriate statistical test for determining whetherto accept or reject the null hypothesis.
In determining whether or not an association exists, the investigator must first assume thatthe null hypothesis is true and then determine how likely the data collected from seleniumwith vitamin D are. Generally, if the data from selenium and Vitamin D and prostate cancerare extremely unlikely should the null hypothesis be true, then there is evidence against thenull hypothesis; in contrast, if the data in question are not unlikely if the null hypothesis weretrue, then there is no evidence against the null hypothesis. While hypotheses in medicaland clinical sciences are often stated as alternative hypotheses, the testing of a hypothesisassumes the null—no difference. For example, there is no difference in prostate cancer riskamong men who take selenium combined with vitamin D compared to those who do not. Inclinical sciences, hypothesis testing involves single-group as well as two- and multiple-groupcomparison. Examples of such tests are single-group means using a single or one-samplet-test, population proportion using z statistics (tests difference between the sample proportionand hypostasized population proportion), differences between population proportions, anddifferences between population means. The appropriate statistics and their assumptionsare an essential part of these materials and are presented in detail in subsequent chapters.
PRIMARY VERSUS SECONDARY OUTCOMES
The primary outcomes represent what the investigators intend to measure to answer theirquestions. This outcome is measured by the primary response variable. A primary responsevariable may be the incidence of a disease, disease severity, a biochemical end point (such asprostate specific antigen (PSA) level), or a biomarker (such as for prostate cancer diagnosis).As an example, in a diabetes mellitus study, if the primary outcome is disease incidence, theresponse variable may be the blood-glucose ascertainment in a follow-up study of overweightand normal-weight adults until the diagnosis of DM by well-defined clinical and laboratorymeasures. If the outcome is DM severity, the response variable may be measured by retinopathy,the need for lower extremity amputation, erectile dysfunction in men, vascular stenosis, and/or the occurrence of comorbidities during the study period. With mortality as the primaryoutcome, death from all causes or from a given disease remains a clear, feasible, and reliablemeasure of the outcome. For example, the effectiveness of androgen-deprivation therapy inthe treatment of locoregional prostate cancer may be measured by the number of deaths due toprostate cancer or cause-nonspecific mortality occurring in the treatment arm (ADT) versuscontrol arm (non-ADT). These variables (response and independent) may be measured on acontinuous, nominal, or discrete scale, as will be illustrated in detail in the upcoming chapters.A study may also be conducted to assess other outcomes, which are not primary to the researchquestion, termed secondary outcomes and measured by secondary variables. In this case,care should be taken to consider secondary outcomes that are more relevant to the primaryresearch questions, since the more secondary outcome measures selected, the more likelihoodfor one to encounter a nonsignificant result, as well as inconsistent and conflicting findings.A secondary response variable may be a biochemical failure, in which the primary responseis death from the tumor. For example, if a study is conducted to examine the effectivenessof Tamoxifen in prolonging survival of older women with nonoperable breast cancer, tumorgrade/size may represent a secondary outcome measure, while mortality remains the measureof survival as the primary outcome.
