DHS 652 - Research Seminar Module 4 Analytic Research Designs What is the difference between descriptive and analytic studies? y Descriptive: Analytic: Approaches in Analytic Studies Experimental Observational 2 Types of Research Designs Observational Cohort Prospective Historical (Retrospective) Case-Control Longitudinal Cross-Sectional Experimental Randomized Clinical Trial 3 Selecting a Research Design 4 Discussion Question The following diagram describes which of the following study designs: a) cross-sectional b) cohort c) case-control d) longitudinal e) experimental 5 Discussion Question The following diagram describes which of the following study designs: a) cross-sectional b) cohort c) case-control d) longitudinal e) experimental 6 Sampling & Populations 7 Types of Sampling 8 Stratified Sampling (Proportional) 9 Sample Size Estimation 1. Significance level (type I error [] )? Usually set to .05 y 2. Statistical power Equivalent to 1- (type 2 error) Standard is 80% power 3. Estimated effect size (d) Estimate f E ti t from previous studies i t di Cohen: small=.1, med=.3, large=.5 4. 4 Degree of variability 10 Discussion Questions 1. Which research design would you recommend to y study risk factors of a rare birth defect? 2. Which sampling method would you use if population density varies greatly within a region? 3. True or False: By convention, the type 2 error for sampling is usually set to .2 4. True or False: A homogeneous population requires a 4 T F l h l ti i larger sample size than a heterogeneous population. 11 Discussion Questions A researcher is studying the effect of a prophylactic drug on malaria. The researcher found that rate of use of prophylactic drugs was 70% among cases and 40% among controls. What is the ff t i ? Wh t i th effect size? Would Cohen consider this a small, medium, or large effect? 12 Sample Size Calculator p Dichotomous Outcome: Unmatched Design Website: http://biostat.mc.vanderbilt.edu/wiki/Main/PowerSampleSize 13 Sample Size Calculator p Dichotomous Outcome: Matched Case Control Website: http://biostat.mc.vanderbilt.edu/wiki/Main/PowerSampleSize 14 Data Analysis Methods for Dichotomous Dependent Variables 15 Summary Descriptive vs. Analytic Designs Types Types of Research Designs Types of Sampling Sample Size Data Analysis Methods 16 Module 4 - Background COMMON RESEARCH DESIGNS USED IN EPIDEMIOLOGY Required Reading Aschengrau A, Seage GR (2003). Chapter 6: Overview of Epidemiologic Study Designs. Essentials of Epidemiology in Public Health, Boston: Jones & Bartlett Publishers. Retrieved May 28, 2012, at: http://publichealth.jbpub.com/aschengrau/Aschengrau06.pdf Beasley, J. M., Coronado, G. D., Livaudais, J., Angeles-llerenas, A., Ortega-olvera, C., Romieu, I., . . . Torres-meja, G. (2010). Alcohol and risk of breast cancer in mexican women. Cancer Causes & Control, 21(6), 863-70. doi: http://dx.doi.org/10.1007/s10552-0109513-. Retrieved from Proquest Central. Cohen J (1992). A Power Primer. Psychological Bulletin, 112(1), 155-159. Retrieved May 28, 2012, at: http://web.vu.lt/fsf/d.noreika/files/2011/10/Cohen-J-1992-A-powerprimer-kokio-reikia-imties-dyd%C5%BEio.pdf Delucchi KL (2004). Sample Size Estimation in Research With Dependent Measures and Dichotomous Outcomes. American Journal of Public Health, 94(3), 372-377. Retrieved May 28, 2012, at: http://www.ajph.org/cgi/content/full/94/3/372 Larsson HJ, Eaton WW, Madsen KM, Vestergaard M, Olesen AV, Agerbo E, Schendel D, Thorsen P, & Mortensen PB. (2005). Risk Factors for Autism: Perinatal Factors, Parental Psychiatric History, and Socioeconomic Status. American Journal of Epidemiology, 161(10), 916-25; discussion 926-8. Retrieved May 28, 2012, at: http://aje.oxfordjournals.org/content/161/10/916.full.pdf+html Mann CJ (2003). Observational research methods. Research design II: cohort, cross sectional, and case-control studies, 20:54-60. Retrieved May 28, 2012, at: http://emj.bmj.com/cgi/content/full/20/1/54 UCLA. Statistical Computing Seminars. Introduction to Power Analysis. Retrieved May 28, 2012, at: http://www.ats.ucla.edu/stat/seminars/Intro_power/default.htm Additional Reading John Concato, Nirav Shah, & Ralph I Horwitz. (2000). Randomized, controlled trials, observational studies, and the hierarchy of research designs. The New England Journal of Medicine, 342(25), 1887-92. Retrieved May 28, 2012, at: http://www.nejm.org/doi/full/10.1056/NEJM200006223422507 UCSF. Department of Epidemiology and Biostatistics (2006). Power and Sample Size Programs. Retrieved May 28, 2012, at: http://www.epibiostat.ucsf.edu/biostat/sampsize.html Vassar Stats. Calculator for Phi Coefficient of Association. Retrieved November 4, 2013, at: http://faculty.vassar.edu/lowry/tab2x2.html Module 4 - Home COMMON RESEARCH DESIGNS USED IN EPIDEMIOLOGY Modular Learning Outcomes Upon successful completion of this module, the student will be able to satisfy the following outcomes: Case o Critically review and evaluate research methodology in analytic studies. SLP o Propose methodology for a research study. Discussion o Identify sources of health data. for research. Module Overview Functions of Research Design The basic function of most epidemiologic research designs is to allow a fair, unbiased comparison to be made between a group with and a group without a risk factor or intervention. A good design should perform the following functions: (1) enable a comparison of a variable (such as disease frequency) between two or more groups at one point in time or, in some cases, between one group before and after receiving an intervention or being exposed to a risk factor; (2) allow the comparison to be quantified either in absolute terms (as with a risk difference or rate difference) or in relative terms (as with a relative risk or odds ratio); (3) permit the investigators to determine when the risk factor and the disease occurred, in order to determine the temporal sequence; and (4) minimize biases, confounding, and other problems that would complicate interpretation of the data. Analytic Research Designs Analytic research designs are used to test hypotheses (as opposed to descriptive research designs which are useful for generating hypotheses, but not to test them). Below are the common analytic research designs used in studies. Analytic studies may be one of the following: 1) observational (in which researchers make observations without interventions), or 2) experimental (in which researchers conduct interventions and examine the effects. Observational studies include case-control, cohort, and crosssectional designs. Cohort Studies In a cohort study, a group of people who share a common experience within a defined time period (cohort) are categorized based upon their exposure status and then they are followed to determine whether or not they develop a specific disease or health outcome. Cohort studies have well-defined populations (i.e., the total number of exposed is known), and therefore a relative risk is used to determine whether an association exists between an exposure and a disease. Relative risk is defined as ratio of the incidence rate among exposed individuals to the incidence rate among unexposed individuals. There are two types of cohort studies: a prospective cohort and retrospective (historical) cohort. 1) Prospective Cohort Studies In a prospective cohort study, the investigator assembles the study groups in the present time, collects baseline data on the cohort, and continues to collect data for a period that can last anywhere from hours to many years. There are many advantages of performing prospective studies. First, the investigator is able to control the data collection as the study progresses and can check the outcome events (e.g., diseases and death) carefully when they occur, thereby making sure that they are correctly classified. The second advantage is that the estimates of risk obtained from prospective cohort studies are true (absolute) risks for the groups studied. Third, many different disease outcomes can be studied, including some that were not anticipated at the beginning of the study. Disadvantages of cohort studies are that they tend to be more expensive, involve more time, and are limited to those risk factors measured at the beginning of the study. 2) Retrospective (Historical) Cohort Studies In this approach, the investigator goes back into history to define a cohort and follows the group members up to the present to identify what outcomes have occurred. If data relating to the research question has already been collected, this would be an ideal study design since cohort studies are one of the most robust types of study designs. Additionally, retrospective cohort studies are inexpensive and involve fewer resources. Case-Control Studies In a case-control study, the cases and controls are selected on the basis of the outcome. The researcher identifies a group of people who meet the case definition and a group of people who do not have the illness (controls). Cases and controls are assembled and then they are questioned (or their relatives or medical records are consulted) regarding past exposure to risk factors. The objective is to determine if the two groups differ in the rate of exposure to a specific factor or factors. In contrast to a cohort study, the total number of people exposed in a case-control study is unknown. Therefore, relative risk cannot be used. Instead, an odds ratio or risk ratio is used. An odds ratio measures the odds that an exposed individual will develop a disease in comparison to an unexposed individual. A case-control study differs from a cross-sectional in that the sample for the case-control study is chosen specifically from groups with and without the disease of interest. Case-control studies are useful when a study must be done quickly and inexpensively or when the disease being studied is rare. Although in these types of studies only one outcome (one disease) can be considered per study, many risk factors may be considered, and this makes case-control studies useful for generating hypotheses concerning the causes of a disease. A major disadvantage of case-control studies is the potential for recall bias. In addition, it is difficult to pick the appropriate control group for the cases. The controls are usually matched individually to cases on the basis of age, sex, and sometimes race. If possible, the investigator obtains controls from the same setting in which the cases were found, in order to avoid potential bias. Cross-Sectional Surveys A cross-sectional survey is a survey of a population at a single point in time. They have the advantage of being fairly quick and easy to perform. They are useful for determining the prevalence of risk factors and the frequency of prevalent cases of disease for a defined population. They are also useful for measuring current health status and planning for some health services, including setting priorities for disease control. For example, many surveys have been undertaken to determine the knowledge, attitudes, and health practices for various populations regarding human immunodeficiency virus (HIV) and acquired immunodeficiency syndrome (AIDS). A disadvantage of using cross-sectional surveys however, is that data about both the exposure to risk factors and the presence or absence of disease are collected simultaneously. Determining the temporal relationship of a cause and effect is then difficult. Another disadvantage is that a cross-sectional survey selects for longer-lasting diseases. Randomized Clinical Trial (Experimental Design) Patients are enrolled in a study and then randomly assigned to one of the following groups: (1) the intervention group, which will receive the experimental treatment, or (2) the control group, which will receive the non-experimental treatment, consisting either of a placebo or of a standard method of treatment. Randomized controlled clinical trials are considered the "gold standard" for studying interventions, because of their ability to minimize bias in the information obtained from the study subjects. To be enrolled in a randomized controlled clinical trial, the patients must agree to participate without knowing whether they will be given the experimental or non-experimental treatment. If possible, the observers who collect the data are also prevented from knowing which type of treatment each patient is given. When this is done the trial is said to be a double-blind study. These studies are timeconsuming and usually costly; may have problems related to therapy changes and dropouts and may be limited in generalizability. Sampling Sampling involves selecting units from a target population so that results from the sample can be generalized to the study population. Probability sampling is most preferable, as it involves random sampling (ensuring that the sample is representative of the population). Some common probability sampling methods are: 1) simple random sampling, 2) systematic random sampling, 3) stratified random sampling, and 4 ) multi-stage cluster sampling. Please review live conference slides and background reading for more information on sampling methods. Sample sizes are determined based on: Significance level (usually set to .05) Statistical power (usually 0.8) Estimated effect size Degree of variability Click Sample Size Example to view a step-by-step example on how to determine sample size using the article by Beasley et al (listed in Background reading). Live Conference #4 Research Seminar for Module 4 PowerPoint presentation. This course is no longer a "Live Course with an EL Conference" as indicated in the PPT for each Module. The archived live conference may be viewed at http://econference.trident.edu/play_recording.html? recordingId=1261515188760_1346191582553 Module 4 - Sample Size Example COMMON RESEARCH DESIGNS USED IN EPIDEMIOLOGY Sample Question In the study conducted by Beasley et al (2010), did the researchers have sufficient statistical power to detect an association between alcohol use during one's lifetime and breast cancer? Provide support for your answer by including alpha and effect sizes (i.e., phi and measure of association). Also reference the statistical calculator used. How to determine if sufficient statistical power 1. Select a sample size calculator that is appropriate for the type of dependent variable in the study (refer to Background reading for sample size calculators). In this study, the dependent variable is dichotomous (i.e., presence/absence of breast cancer). PS Power and Sample Size Calculations was used in this example as it can be used with dichotomous dependent variables. 2. Under dichotomous tab in PS Power and Sample Size Calculations, enter the output, design, and input variables. The following variables were entered as shown below: OutputSelect Power as the question is asking about whether there is sufficient statistical power. DesignAs the study by Beasley et al (2010) is a matched case control design, matched and case control was entered This refers to the Type I error probability for a two sided test. This is the probability that we will falsely reject the null hypothesis. This is generally set to .05 (i.e. the p value) nFor case-control studies n is the number of case patients. For prospective studies n is the number of patients receiving the experimental treatment. In this study by Beasley et al (2010), the number of case patients is 1000. p0 For case-control studies, p0 is the probability of exposure in controls. In prospective studies, p0 is the probability of the outcome for a control patient. In the study by Beasley et al (2010), the rate of alcohol use (i.e., ever drinking alcohol) among controls was 473/ (473+567) = .4548 (as obtained in Table 2). Privacy Policy | Contact