Introduction

34 C.M. Schmidt

mixed methods. An optimal study should be adequately powered, free of bias, and be able to be conducted in a reasonable timeframe with resources available to the investigator.

Reviewing the Literature

A thorough review of the literature is essential on your topic of interest. Prior to reviewing the literature, however, one must pick a topic of interest. Choice of your topic of interest may directly facilitate a thorough review of the literature. The topic of interest should optimally be as narrow as possible. This will facilitate a thorough review of the literature on the topic and subsequent mastery of the subject matter. A broad topic, conversely, will be cumbersome to review and may not result in true mastery of the subject matter.

A proper review of the literature uses multiple sources of information on the topic of interest. These resources include original peer-reviewed papers, textbooks, chapters, reviews, editorials, but also online and personal resources, e.g., experts in the field. Most important of these, however, are original, peer-reviewed papers. Textbooks, chapters, reviews, editorials, and online resources, while helpful in identifying primary sources of information on your topic and organizing large amounts of information, should not be weighed heavily in your literature review. These resources notoriously contain significant bias on the subject matter.

In order to perform a thorough review of original peerreviewed papers, a Medline search of all relevant works is an excellent first step. Investigators should have a low threshold to use the library staff at their institution/university to insure proper Subject and MeSH terms and headings are used. The quality of the original work should be reviewed in a systematic fashion. Questions to address when reviewing an original work in the literature (or designing a study) are:

1.Internally valid?

2.Externally valid?

3.Appropriate conclusions?

An internally valid study is adequately powered and free from selection, time, and information bias. It is also free of misclassification errors and confounding variables. Adequate power is particularly important when no significant effect is demonstrated in a study under review. For negative studies, it is critical to appreciate whether there was an adequate sample size to determine a significant difference in the parameter being assessed. Otherwise, “no effect” is an invalid conclusion. Such an error is called a beta error. This is related to power by the formula, Power = 1 − beta, where beta is the chance that one fails to detect a difference when one exists. Selection bias is when subjects in treatment groups are not selected randomly, e.g., the investigator chooses which patients to enroll on study. Time bias is when subjects in one treatment group have advantage or disadvantage of time. An example of lead time bias is when patients who have colon cancers detected on screening colonoscopy live longer than patients who actually developed symptoms prior to colon cancer detection. Colonoscopy does not result in longer life from colon cancer, rather when screening examinations reveal cancer, it simply detects the cancer sooner. Information bias is when the data collection is not equivalent between groups (e.g., derived from different resources, individuals, or with variable rigor). Misclassification errors are when treatments or groups are incorrectly assigned due to suboptimal classification, e.g., if the study relies on subjective recall of past events. Finally, confounding variables may invalidate a study. Confounding variables are variables in the treatment groups that result in the “effect” or the “non-effect” being studied. For example, patients undergoing abdominal CT scan compared to patients who have not undergone CT scan have a higher incidence of gastrointestinal disorders. CT does not cause the gastrointestinal disorder, but rather is associated with gastrointestinal disorders because it is commonly employed to diagnose them.

An externally valid study is one where the results actually have meaning beyond the numbers and may be generalizable to other like populations. The external validity largely depends upon the inclusion and exclusion criteria, but also depends upon

36 C.M. Schmidt

the study population. If the study population is unique, the results may not be broadly applicable to other populations even if the patients meet the inclusion and exclusion criteria. Clinical relevance may also affect the external validity of a study. Even though the results may be statistically significant, the effect may be so small that it is not actually meaningful in the real world. Finally, sometimes the results may be statistically significant, but not actually real. Finding a difference when no difference exists is termed an alpha error, e.g., p-value (alpha)=0.05. This indicates a 5% chance that the difference detected between two or more groups is due to random chance rather than a true effect.

Finally, after reviewing the internal validity and external validity of a study, one must make sure that the authors of the study have drawn appropriate conclusions. Randomized controlled trials may actually determine cause and effect relationships, whereas other studies may only determine association.

Once you have researched the topic with all available resources, consulting with experts in the field is essential. Individuals who have studied or practiced in the field of interest will have assimilated the relevant information on multiple occasions and may have a perspective one cannot glean from reviewing the literature on one occasion in isolation.

With a proper and thorough review of the literature, one will become an expert in the subject matter. With this expertise, an original and relevant hypothesis is most likely to be developed.

Developing a Hypothesis

With a thorough review of the literature accomplished, one may reasonably develop original and relevant (i.e., meaningful) hypotheses. The hypothesis should be original so there is no unnecessary duplication of research. Repeating a study may indeed have merit, e.g., if control population was suboptimal in the original study or if there was inadequate power and the original study failed to demonstrate a difference in the parameters under study. Simply repeating a study for the sake of repeating it, however, is not optimal in terms of resource utilization and really needs to be justified based on concerns of

improving the original design. In addition to original, the hypothesis should be relevant.A relevant hypothesis is one that once proven true or false should result in a change in behavior for the betterment of mankind. Consider the following hypothesis: individuals who are left-handed are able to perform righthanded tasks better than right-handed individuals perform left-handed tasks. It is unclear how such a hypothesis is relevant and will result in any change in behavior for the betterment of mankind. A hypothesis may be relevant, but not feasible (i.e., able to be tested). A hypothesis may not be testable due to statistical or ethical considerations. Consider the following hypothesis: Individuals with a rare condition (that affects only two individuals in the world) have significant symptom resolution with treatment A vs. B vs. C. Such a study is not feasible since there are only two individuals (N =2) that may be tested. Consider another hypothesis: Routine catheterization of the carotid artery in humans results in more accurate blood pressure determination than radial artery catheterization.This study is not testable due to being prohibited on ethical grounds.

With these caveats in mind, the steps to creating a hypothesis are as follows:

1.Define the question.

2.Define the population.

3.Define the intervention.

4.Define the results.

5.Define the next question.

In developing a hypothesis, a principal or primary question needs to be identified. Once identified, the principal or primary question needs to be defined precisely. A precise definition will involve:

1.Description of intervention(s) and treatment group(s), e.g., resection vs. observation

2.Response variables, e.g., quality of life, survival, diseasefree survival

3.Measurement methods for response variables and group comparisons