Observational Studies
Observational studies are used to observe and measure outcomes in a cohort with no control over risk factors or variables. They are often retrospective. Types of observational studies include cross-sectional studies, case-control studies, and cohort studies.
Cross-sectional studies
- Collect data about an entire population at a single point in time (cross-section of time)
- Participants are not followed over time.
- Typically used to measure prevalence (i.e., how many people have the disease in a population)
- Also called prevalence studies
- Advantages:
- Inexpensive, if making use of routinely collected data
- Easy to perform
- Disadvantages:
- Unable to determine causality because exposures and outcomes are measured simultaneously
- Unable to include data on confounding variables
Case-control studies
- Start from an outcome and work backward (retrospectively) in time to see who had an exposure
- 2 groups are compared based on an outcome or the presence of disease:
- Cases: those who have or have had the outcome or disease
- Control: those who lack the outcome or disease
- The proportions of those with an exposure in the case and control groups are compared to see if there is an association.
- Matching: selecting controls so that they share similar characteristics with cases that are identified as possible confounding factors (e.g., sex, age, smoking status) in an effort to remove the confounding effect
- Individual matching: matching an individual control to an individual case
- Group or frequency matching: equal percentages of controls and cases with a characteristics group with characteristics (e.g., 50% of cases and controls are men)
- Advantages:
- Inexpensive
- Can be carried out by small groups of investigators
- Shorter in duration
- Disadvantages:
- Cannot measure the incidence
- Cannot reliably determine a subject’s exposure status over time (subject to observation bias)
- Identifying a sample of controls can be difficult and subject to selection bias.
Cohort studies
- An observational study that chooses a cohort that shares a common exposure and observes this group over time (prospectively) to see who develops the outcome of interest
- A longitudinal study in which the temporal relationship between exposure and outcome is established
- Cohorts can also be “retrospective” (i.e., looking into medical records and seeing if the outcome developed itself prospectively in the past), but this restricts the ability of the investigator to control for confounding and bias.
- Typically used for establishing possible cause and effect before randomized control trials (RCTs) are undertaken
| Advantages | Disadvantages |
|---|---|
| Direct calculation of incidence rates | Time-consuming |
| May yield information on the incidence of disease | Often requires a large sample size |
| Clear temporal relationship between exposure and disease | Expensive |
| Particularly efficient for the study of rare exposures | Not efficient for the study of rare diseases |
| Can yield information on multiple exposures | Losses to follow-up may diminish their validity |
| Can yield information on multiple outcomes of a particular exposure | Changes over time in diagnostic methods may lead to biased results |
| Minimizes bias |
| Exposure | Outcome | Study type |
|---|---|---|
| Uncommon | Common | Cohort |
| Common | Uncommon | Case-control |
| Scenario | Study type | |
| Short-term study of current, uncommon exposures and common outcomes | Prospective cohort | |
| Short- or long-term study of historic, uncommon exposures and common outcome | Retrospective cohort | |
| Outbreak investigation | Case-control | |
| Instantaneous survey, common outcomes and no change over time | Cross-sectional | |
| Is there an association between climbing Mount Everest and getting diabetes? | Cohort | |
| Is there an association between left-handedness and getting mad cow disease? | Case-control | |
| Is there an association between left-handedness and gender? | Cross-control | |
| What factor was likely responsible for the salmonella outbreak at the office holiday party? | Case-control | |
| Was there an association between working on the nuclear bomb project in World War II and developing cancer 5 years later? | Retrospective cohort | |
A flowchart of the different types of research (observational vs. experimental) and their representative studies. Observational studies passively look at a cohort with a disease or condition over a period of time (often retrospectively); experimental studies prospectively look at the effect of a particular intervention or exposure in a group as compared to controls.
Image by Lecturio.Interventional Studies
Interventional studies are prospective experimental trials in which investigators compare the effects of an intervention on subjects with a control group. The most compelling type of interventional study is the RCT.
Randomized controlled trial
- Also called randomized clinical trials
- Gold standard for evidence:
- Provides the strongest evidence for establishing causation
- Used to:
- Compare the efficacy of new therapy regimens and drugs against current therapies
- Evaluate new screening and prevention strategies
- Come up with new ways to organize and deliver healthcare
- Can be performed in both clinical and community settings
- Study groups: also called study arms
- Intervention/experimental group: Participants are exposed or receive the intervention.
- Control group: Participants receive the current standard of care, a placebo, or nothing at all.
- Groups are chosen so that the 2 groups have almost identical conditions, except for the exposure of interest.
- Groups must be balanced: equivalent in size
- Principle of intention to treat:
- If subjects change groups, the results should be analyzed according to the initial group allocation of subjects.
- Randomization:
- Random allocation of subjects to either intervention or control group
- Participants are assigned to either group by chance.
- Reduces the chances of selection bias because the chance of being in either group is 50/50
- Blinding:
- Subjects and/or investigators don’t know the groups to which subjects are assigned.
- Single-binding: Subjects do not know if they are getting the intervention or a placebo.
- Double-blinding: Both participants and investigators do not know to which groups subjects are assigned.
- Triple-blinding: blinding of the meta-analyst
- Blinding reduces the chances of the Hawthorne and Rosenthal effects
- Hawthorne effect: the tendency of subjects to behave differently if they know they are being studied
- Rosenthal effect: changes in the behavior of subjects based on the researcher’s expectations
- Placebo:
- A medically ineffective intervention that mimics a real intervention so that a subject will not know to which study group they are assigned
- Placebo effect: Any treatment, even if ineffective, results in improvement because the recipient believes it will.
| Advantages | Disadvantages |
|---|---|
| Confounding variables are well-controlled | Cost |
| Temporal relationship is well-established | Study groups do not necessarily represent the real world. |
| If blinded, provides strong evidence for causation | Ethically problematic |
Flowchart showing how a defined population is distributed into different study groups through randomization when setting up an RCT. Randomization is carried out by the investigators so participants don’t choose which group they’ll be a part of.
Image by Lecturio.| RCTs | Case-control studies |
|---|---|
| Prospective | Retrospective/prospective |
| Experimental study: An intervention (exposure) (e.g., drug, screening test) is applied to a group and effects are compared to a control group. | Observational study: A group is chosen with a common exposure and followed longitudinally to track to development of an outcome (e.g., disease, condition). |
| Subjects are randomized by investigators before exposure occurs. | The subjects, or medical records, report their exposure. |
Quasi-experiment
- A study that lacks the randomization and often the blinding component
- The investigator decides who will be assigned to which group.
- Lacks internal validity but has great external validity
Conceptual map showing the distribution of the subjects in studies that employ a quasi-experimental design. Note the lack of randomization.
Image by Lecturio.Natural experiments
- Resembles an RCT but the allocation is decided by an external force (e.g., natural disaster)
Conceptual map showing the distribution of the study population in natural experiments. Randomization is not carried out by the investigators but by an external force and is subject to bias.
Image by Lecturio.Ecological studies
- Populations are the unit of analysis instead of individuals.
- Ecological fallacy: Associations for populations cannot be applied to individuals.
References
- Hammill, B. G. (2013). Chapter 12: Observational study designs. In R. D. Lopes, & R. A. Harrington (Eds.), Understanding clinical research. New York, NY: The McGraw-Hill Companies. accessmedicine.mhmedical.com/content.aspx?aid=57836443
- Dawson, B., & Trapp, R. G. (2004). Chapter 2: Study designs in medical research. Basic & clinical biostatistics, 4th ed. New York, NY: The McGraw-Hill Companies. accessmedicine.mhmedical.com/content.aspx?aid=2046062
- Schnipper, J. L. (2017). Research in the hospital. In S. C. McKean, J. J. Ross, D. D. Dressler & D. B. Scheurer (Eds.), Principles and practice of hospital medicine, 2nd ed. New York, NY: McGraw-Hill Education. accessmedicine.mhmedical.com/content.aspx?aid=1137607526
