We are in an era of evidence-based medicine, with an exponentially increasing information base arising from descriptive and interventional studies. Clinicians are required to use this evidence to guide their clinical strategies for diagnosis and management of patients. This chapter discusses the basic principles of epidemiology and medical biostatistics.
TYPES OF EPIDEMIOLOGIC STUDIES
Epidemiologic studies are broadly divided into observational studies and experimental studies.1-6
Observational studies are based on naturally occurring phenomena and do not involve an intervention by the study team. There are several types:
Case reports and case series: These are descriptive studies that report the characteristics of the disease and frequency of events. They tend to report sentinel events that are useful in generating a hypothesis but cannot test them. They are useful for describing very rare diseases. In the hierarchy of the evidence pyramid, laboratory research and animal studies are at the very bottom followed by editorials, case reports, and case series.
Cross-sectional studies: These are analytic studies that evaluate a specified population for a given time. They may be used to determine the number of cases of a disease in the population at a particular point in time (ie, prevalence of a disease) and hence are also called prevalence studies. They are not suitable to evaluate incidence rates or to evaluate rare diseases or exposures. These are the weakest form of evidence among analytical observational studies, falling just above case series in the evidence pyramid.
Case-control studies: These are a form of analytic observational studies in which patients with a disease are selected for the case arm, and appropriate matched participants are selected for the control arm. These studies start with the disease and attempt to retrospectively identify causal associations. The advantages include that they are fast and inexpensive to conduct. They are well suited to study uncommon diseases and allow evaluation for multiple causal exposures. They are, however, subject to significant bias such as recall bias. Appropriate matching of control participants may be difficult. Case-control studies rank above cross-sectional studies but below cohort studies in the hierarchy of evidence.
Cohort studies: These are observational studies that follow a group of participants or a cohort with a known exposure as well as a cohort without the exposure, until development of the disease, to look for a difference in the incidence of the disease in the two groups. These studies may be prospective, retrospective, or mixed. They are suitable to estimate incidence, evaluate rare exposures, and evaluate for multiple outcomes. They can, however, be expensive and require long periods of follow-up.
Experimental studies involve an active controlled intervention such as an exposure or treatment.
Randomized controlled trials (RCTs): These are the most well-known experimental design. These are the best design to prove the relationship between exposure and event. Randomization and blinding are usually performed to minimize bias. They may be expensive and time consuming and are of limited value in evaluating exposures that are known to have harmful outcomes. When reviewing these studies, one must question the internal validity as well as external validity of the study. RCTs rank higher than observational studies in the evidence pyramid, with systematic reviews and meta-analyses at the top.
BIAS IN EPIDEMIOLOGIC STUDIES
Bias in a study is an error in the design or execution of a study that consistently encourages one outcome or answer over others.1 The sources and types of bias vary depending on the study design (eg, observational studies such as case-control studies or experimental studies such as RCTs).
Common types of bias include:
Bias due to confounders: A confounding factor or covariate is a factor that is associated with the exposure or risk factor and disease being studied and is distributed unequally between the study groups. Bias due to confounders can be minimized in observational studies by matching cases and control participants to ensure equal distribution of known confounding variables. The effect of confounders can also be minimized by analytical strategies such as multivariable analysis.
Selection bias: This occurs when the method of selecting participants in the study results in a difference between the study population and the target population. There is a distortion in the magnitude of the association between the exposure and disease in the study versus the target population.
Exclusion bias: This can occur when the study is designed to exclude people with certain conditions to prevent confounding. This occurs more often when the exclusion criteria are different for cases versus control participants. An example is if a study were designed to evaluate the association between chronic obstructive pulmonary disease (COPD) and lung cancer and smokers were excluded from the control arm to prevent undiagnosed cases of COPD but smokers were included in the cases arm. A significant association would be noted between COPD and lung cancer; however, the differential exclusion of smokers would be a source of bias.
Nonresponse bias: This occurs because people who volunteer to enroll in studies or respond to calls to enroll in studies tend to behave differently from those who do not and hence may not fully represent the target population.
Berkson bias: This is a form of selection bias or hospital admission bias that occurs in hospital-based case-control studies. This can occur if the exposure increases the chance of admission, leading to a higher number of participants with that exposure.
Recall bias: This occurs predominantly because of one of two reasons. One tends to recall more recent events compared with distant events, as well as more severe or traumatic events compared with less severe events.
Reporting bias: Participants tend to report and emphasize events or exposures that they believe to be more important or relevant.
Hawthorne effect: The behavior or response of a participant can change if he knows he is being observed.
Placebo effect: This can occur when patients subjected to the exposure or intervention report feeling better as a result of knowing that they are receiving treatment.
Observer bias and ascertainment bias: These can occur when knowing the arm of the study that the participant is in subconsciously influences the person determining the outcome.
The design of RCTs may incorporate a number of methods to reduce bias:
Randomization: Randomization helps prevent selection bias, preventing the allotting of a certain cohort of patients to one arm of the trial versus the other. Although matching allows for equitable distribution of known confounding factors, it has little effect of unknown confounders. Randomization aids in the equitable distribution of unknown confounders.
Stratification: Stratification during randomization assists with equal distribution of known confounding factors such as stratification by age or by center in multicenter trials.
Single blinding or blinding of patients: This involves use of a placebo or sham procedure to minimize the placebo effect.
Double blinding: This involves blinding of the patient and the physician as well as outcomes assessor. This minimizes observer or ascertainment bias, as well as deviations in care based on the treating physicians bias toward the intervention.
Triple blinding: In addition to double blinding, the person analyzing the data and the monitoring committee is also blinded. This can minimize bias in analysis and evaluation. Triple-blinded studies are rarely done because of the need to monitor safety of the intervention.