SECTION 5

PROPOSAL DEVELOPMENT

[Formulating a hypothesis] [Identifying the key variables] [Developing the research design] [Sampling techniques and sample size] [Diagnostic tests] [Research with human subjects] [Writing a proposal]

5.1 Formulating a hypothesis

Following a thorough literature review, the next step is to refine a hypothesis for the study. A well-defined hypothesis crystallizes the research question and influences the statistical tests that will be used in analyzing the results.

There are two types of hypotheses. The null hypothesis predicts no difference between comparison groups or association among tested variables. The alternative hypothesis predicts either a simple difference (two-tailed hypothesis) or a difference in a particular direction (one-tailed hypothesis).

Example:

• Null hypothesis
  -There is no association between saturated fat intake and coronary heart disease.

• Two-tailed hypothesis
  -There is an association between saturated fat intake and coronary heart disease.

• One-tailed hypothesis
  -There is a positive association between saturated fat intake and coronary heart disease.

5.2 Identifying the key variables

A variable is defined as a characteristic that can be manipulated or observed, and can take on different values, either quantitatively or qualitatively (e.g., family income, age, gender, heart disease, blood pressure, etc.). Variables can be classified into the following three categories:

Independent variable This variable, also called a predictor variable, is independent of the outcome itself. It is presumed to cause, affect, or influence the outcome. An example of an independent variable is unsaturated fat intake in the hypotheses above.

Dependent variable This variable, also called an outcome variable, is dependent on predictor variables; the outcome presumably depends on how the independent variable is managed or manipulated. An example of a dependent variable is coronary heart disease in the hypotheses above.

Control variable This is a variable which must be controlled (i.e., held constant or randomized) so that its effects are neutralized, canceled out, or equated for all conditions. Typically, control variables include such factors as age, sex, socioeconomic status, educational level, etc. It is often possible to redefine these particular examples as either independent or dependent variables according to the intent of the research.

5.3 Developing the research design

The research design refers to how the study is conducted, including:

• How the independent variable is manipulated and introduced (intervention)

• How the group differences or outcomes are measured (dependent variables)

• How many groups of subjects will be tested

• How subjects are selected and assigned to groups

• What is the temporal (time) sequence of interventions and measurements

There are five basic types of research design: cross-sectional; cohort; case control; experimental; and quasi-experimental. The first three are also called observational studies. A general rule for selecting a particular research design is to define the purpose of the study that narrows the options of research design (e.g., will cause-effect relationship between variables be examined?).

An experimental study is used to establish a cause-effect relationship. A true experimental study requires a high degree of control, where subjects must be randomly assigned to experimental and control groups. A quasi-experimental study describes a situation where subjects cannot be randomly assigned to groups, or control groups cannot be used.

Among the three observational studies, a case control study is used to examine the occurrence of a rare disease; a cross-sectional study to examine the prevalence of a disease; and a cohort study to examine incidence of a disease. (Prevalence is defined as the proportion of people in a population having the disease, and incidence is defined as the proportion of people acquiring the disease over a period of time.) A more detailed discussion of types of research design is included in Appendix M.

5.4 Sampling techniques and sample size

Sampling techniques
For all types of clinical research, a study group is formed by selecting individuals from a population. This group is called a sample, and each individual in the group is called a subject. Population refers to a complete set of individuals with a certain characteristics. For example, in studying the effects of various treatments on diabetes, the population of interest would be all people in the world who have diabetes. Obviously, it is unrealistic to work with this entire population! Instead, a group of subjects is selected from this population, assuming those subjects to be representative of the population so that the results can be generalized.

Sampling techniques refer to the different ways by which subjects are selected from the population. There are two types of sampling techniques: probability and non-probability sampling. The former assures that each subject in the population has a known chance of being included in the sample, whereas the latter does not.

There are four types of probability sampling:

• Simple random sampling uses flips of a coin, rolls of dice, or use of a table of random numbers, to randomly select subjects from a population (e.g., 100 of the 1000 diabetic patients a hospital treated over the preceding year).

• Stratified random sampling uses subcategories (e.g., gender) to randomly select subjects (e.g., 50 male and 50 female diabetics).

• Systematic sampling selects every nth case from a population (e.g., every tenth patient with diabetes).

• Cluster sampling, or multistage sampling, involves successive random sampling of a series of units within the population. For example, to study all diabetic patients in the country, it would be impractical to generate a complete list of all patients from which the sample can be randomly selected. Instead, it might be possible to recruit representative hospitals throughout the country, and then select a 100-patient random sample from each participating hospital.

There are three types of non-probability sampling:

• Convenience sampling, or accidental sample, selects subjects based on availability (e.g., all diabetic patients seen in a clinic over the past five years).

• Quota sampling is similar to stratified random sampling in that a certain proportion of subjects is predetermined for selection; however, the selection process is nonrandom.

• Purposive sampling selects subjects on the basis of specific criteria. For instance, in a study determining whether the presence of a certain disease is associated with a particular agent, some “typical” individuals exposed to the agent are selected.

Simple random sampling is a basic technique and is incorporated into all the more complex probability sampling designs. Despite the popular use of random samplings in research, non-probability samples are sometimes preferred because of practicality and cost effectiveness. Non-probability is often used in pilot studies or a pre-testing situation. Sometimes a combination of probability and non-probability is useful, such as a random selection of some hospitals within certain regions.

Sample Size

In clinical research, it is often neither practical nor desirable to recruit a large number of subjects. However, in order to carry out a meaningful statistical test, a certain minimum number of subjects is required to reduce the error of the results to an acceptable level. Therefore, it is necessary to determine the minimum number of subjects needed for a study.

The minimum number of subjects required in a study depends on several key factors, such as:

• Significance level (P) is a probability measure used in mistakenly rejecting the null hypothesis (see section 5.1). A P value of 0.05, 0.01, or less is generally considered as the acceptable level for rejecting the null hypothesis. For instance, the null hypothesis for a study comparing the national board scores of male and female physicians predicts that there will be no difference between men and women. The results of the study show that there is a difference between men and women at a P value of 0.05. This means that there is a gender difference (rejecting the null hypothesis) 95 out of 100 times. Only five times out of 100 can the gender difference be attributed to chance alone. In reporting results, a P value of less than 0.05 usually is regarded as significant and less than 0.01 as highly significant. A more stringent significance level requires a larger sample size (i.e., more subjects are needed if the P value is set at the 0.01 level than at the 0.05 level).

• Effect size is the observed difference prior to and after invention, or the difference between the control and experimental groups. A larger effect size requires fewer subjects for valid testing. For instance, a study tests the effectiveness of an exercise program on patients with heart disease. Patients are randomly assigned into two groups – an experimental group that will participate in the program and a control group which will not. All patients in the study are taking a drug which is known to cause improvement in 20 percent of patients in the absence of exercise. With the introduction of an exercise program, the alternative hypothesis predicts that this level of improvement will increase to 60 percent. Twenty-two subjects are needed in each group, which will give enough statistical power to determine whether there is an effect of the exercise program. Alternatively, if the exercise program is only expected to increase the level of improvement to 30 percent in the experimental group, a total of 292 subjects is needed in each group.

• Variance is a measure of the variation within a set of data. As the variance decreases, a smaller number of subjects is required. For example, a drug is expected to lower the diastolic blood pressure from 90 mm Hg to 80 mm Hg. If the standard deviation of the blood pressure measurement is 10, then 17 subjects are needed in each group to give enough statistical power to state that there is an effect of the drug. However, if the standard deviation is 30, then 142 subjects are needed in each group to demonstrate an effect of the drug. See section 8.2 for a more discussion of variance.

• Power is the probability of rejecting the null hypothesis when the alternative hypothesis is true. It ranges from 0 to 1. Usually the power of 0.80 or higher is considered as an acceptable level for rejecting the null hypothesis when it is false. As a sample size increases so does the power. In other words, more subjects are needed if the power is set at the 0.90 level than at the 0.80 level.

In general, the use of continuous variables also permits more options for statistical tests than the use of categorical variables. Blood pressure, for instance, can be expressed either as millimeters of mercury (continuous), or as the presence of absence of hypertension (category). Another example is national board scores, which can be expressed either as actually scores (continuous) or pass/fail (category).

Confidence intervals give the range above and below the observed result that correspond to the specific likelihood that the result is correct. For instance, a survey is conducted to assess medical care of poor patients in an underserved area. It is estimated that about 20 percent of these patients are treated by primary care physicians. Based on Table 1, at least 200 subjects must be included in the survey to be able to state, with 95 percent confidence, that the actual proportion of the poor patients treated by the total population of primary care physicians is between 14 and 26 percent (20 + 6 percent). The more subjects surveyed, the higher the confidence level, all the other factors being constant. More subjects result in a narrower confidence interval, which is more precise than a wider one.

Table 1. Confidence Intervals for Samples with Different Percentages of a Study Characteristics and Different Sample Sizes*

* Assumes a simple random sample, a large population, and being correct in 95 of 100 samples. (Bauman, 1980).

In the early planning stages, it is helpful to make a rough estimate of the minimum number of subjects needed for a clinical study. If the sample size estimate is beyond realistic limits, it may be possible to redesign the study by controlling variability in the sample or increasing the effect size. Alternatively, a decision should be made to forgo the study if the likelihood of obtaining significant results is small.

5.5 Diagnostic tests

Evaluations of diagnostic tests are frequently reported in medical literature. The common question raised in diagnostic tests is: Among the patients with a disease, is a clinical test useful for diagnosis of a related disease?

Frequently, the test result serves as the predictor variable and the disease as the outcome variable. Consider a test for cancer in women with solitary breast masses. Of 100 women with breast cancer, 65 have a positive test and 35 have a negative test; of 100 women without breast cancer, 70 have a negative test, and 30 have a positive test.

Now we have four situations:

• True-Positive (TP): When the test is positive and the patient has the disease;
  

• False-Positive (FP): When the test is positive and the patient does not have the disease;
  

• False-Negative (FN): When the test is negative and the patient does not have the disease;
  

• True-Negative (TN): When the test is negative and the patient has the disease.

Sensitivity

-The proportion of subjects with the disease who have a positive test: TP/(TP+FN)
-The breast cancer test in the example had a sensitivity of 65%: Out of 100 women with breast cancer; 65 have positive tests.

Specificity

-The proportion of subjects without the disease who have a negative test: TN(TN+FP)
-The breast cancer test had a specificity of 70: Out of 100 women with benign nodules, 70 had a negative result on the test.

In general, diagnostic tests are evaluated by calculating their sensitivity and specificity.

5.6 Research with human subjects

Any research proposal involving OMC patients and medical records must first be reviewed by the Human Subject Subcommittee of OU-COM’s Research and Scholarly Affairs Committee (see policy in Appendix J). The proposal will then be submitted to OU’s Institutional Review Board for Review of Research Involving Human Subjects (IRB). Any research project conducted at Ohio University involving human subjects requires prior approval by the IRB (See OU Policy and Procedure #19.052 in Appendix H.) Appendix K contains the forms and procedures for obtaining approval from the IRB. The Office of Compliance (740-593-0664) is available to discuss a proposed study, particularly concerning how human subjects are to be recruited, benefits and risks involved, and appropriate measures to be taken if harmful side effects occur.

• Institutional Review Board (IRB) The purpose of an Institutional Review Board is to review and approve any research investigation involving human subjects, or declare it exempt from review by the board. Ohio University’s IRB is composed of 12 members from various campus departments and the community, and their approval of all research involving human subjects is required before enrollment can begin. The review and approval process is to ensure the protection of subjects from potential harm, whether physical or breach of confidentiality. Studies that continue over an extended period of time require ongoing IRB endorsement consisting of at least an annual review and approval of the protocol. Some research may be exempt from IRB review, but that determination is also made by the board. Researchers submitting protocols to the IRB are notified in writing of acceptance or rejection within two to four weeks.

Whenever human subjects are involved, the IRB will review the study for ethical problems. Never assume that a study is so inconsequential that ethical review is unnecessary.

PLEASE TAKE NOTE! Prior approval by the IRB is mandatory before beginning a project (even a survey) involving human subjects!!

• Informed Consent All subjects must be provided with information regarding the research project, and documentation of this process is required by a signed consent form. The IRB Guidelines in Appendix K explain the process and format for informed consent, and a sample consent form can be found in Appendix L.

5.7 Writing a proposal

The formal structure and length of a research proposal vary widely, depending upon the specific group or agency for which it is targeted. Before beginning to write the proposal, it is essential to obtain the guidelines or instructions from the potential sponsor and follow them to the letter. Many agencies follow the guidelines established for proposals submitted to the Public Health Service (e.g., all National Institutes of Health [NIH] grants, AOA grants, etc.). These research proposals must contain the following sections:

• Abstract
  The abstract is a short synopsis of the project. It should clearly state the broad long-term objectives and specific aims of the project, and describe the research design and methods for achieving these goals. A good abstract serves as a succinct and accurate description of the proposed work even when separated from the application. The actual length of the abstract varies depending upon the rules of specific agencies, with typical limits of 100 to 1000 words.

• Specific aims
  This section outlines the broad, long-term objectives and provides a concise and realistic description of the research project, with emphasis on what will be accomplished and what hypotheses will be tested. This section is usually about one page in length.

• Background and significance
  This section reviews the background to the proposal, critically evaluating existing knowledge, and specifically identifying the gaps that the project is intended to fill. A concise statement of the importance of the research should identify how the specific aims relate to the improvement of medical knowledge and health care delivery. This section is usually about one to three pages in length.

• Preliminary studies
  This section summarizes the results from preliminary studies that are pertinent to the project. It is important here to demonstrate that the investigator has the necessary experience and competence to pursue the project. Usually, copies of papers and abstracts published by the applicant that are related to the project are included in an appendix. If there is no publication record in the field, a good case must be constructed by the applicant to convince the reviewers that the project is feasible. Often, this can be accomplished through collaborations with experienced investigators. AOA grants limit this section to one page, while NIH grants recommend a limit of six to eight pages.

• Research design and methods
  This section outlines the experimental design and the procedures to be used for testing the hypotheses of the project, including the means by which data will be collected, analyzed, and interpreted. Describe the advantages of any proposed new methodologies. Discuss any potential difficulties or limitations of the proposed procedures, along with alternative approaches that might be necessary to achieve the desired goals. A tentative sequence or timetable for the project should be included. Finally, identify any procedures, situations, or materials that might be hazardous to subjects or personnel and explain appropriate precautions that will be exercised in the conduct of the project. The length of this section usually varies from five to fifteen pages.

• Human subjects and vertebrate animals
  Projects involving humans and/or animals usually require specific data and justification regarding protocols and risk factors. Ohio University’s Institutional Review Board (IRB) must approve all projects involving human subjects (see section 5.6). In addition, protocols in projects involving the use of vertebrate animals (refer to OU policy #19.049) must be approved by the Institutional Animal Care and Use Committee (593-0372 or 593-2857).

• Hazardous materials
  The use of hazardous materials in a research project must be identified when the proposal is submitted. Hazardous materials are regulated by the Department of Environmental and Health Safety (EHS)at Ohio University (593-1663). The use of radioactive materials requires the approval of the Radiation Safety Committee at EHS (593-1667). The use of biohazardous materials requires the approval of the Institutional Biosafety Committee at EHS (593-1662).

• Literature cited (bibliography)
  Finally, a list of relevant and current (not necessarily exhaustive) literature citations must be included. Each citation usually includes the title, names of all authors, book or journal, volume number, page numbers, and year of publication. It is important to follow the instructions regarding citations to the letter!

Section 6

Return to Section 4

Return to Table of Contents