Penelitian Observasional

01Foundational Concept

Not every clinical question can be answered with a randomized controlled trial. It would be unethical to randomize patients to smoke cigarettes for 20 years to study lung cancer. When an exposure is harmful, rare, or impractical to assign, researchers instead observe naturally occurring groups and compare outcomes.

The fundamental trade-off is this: observational studies sacrifice the ability to prove causation (because there is no randomization to eliminate confounders) in exchange for the ability to study questions that cannot be ethically or practically tested with an experiment. Exam questions exploit this trade-off repeatedly. When a vignette describes a study without randomization, you must immediately recognize it as observational and understand that confounding is the primary threat to its internal validity.

The hierarchy of evidence places observational studies below randomized controlled trials but above expert opinion and case reports. Within observational designs, cohort studies generally provide stronger evidence than case-control studies, which in turn provide stronger evidence than cross-sectional studies.

02Classification of Observational Study Designs

03Cohort Studies

3A. Design and Logic

A cohort study begins by identifying a group of individuals based on their exposure status. One group has the exposure (e.g., smokers), another does not (e.g., non-smokers). Both groups are then followed over time, and the investigator records who develops the outcome of interest (e.g., lung cancer).

The word "cohort" literally means a group marching together through time. This forward-looking logic (exposure → outcome) is the defining feature. Even in a retrospective cohort, the investigator still starts with exposure and looks forward to outcome; the difference is only that all the data already exist in medical records, so no real-time follow-up is needed.

Because you start with a known number of exposed and unexposed individuals, you can directly calculate the incidence of disease in each group. This allows you to compute a Relative Risk (RR).

$$RR=\frac{\text{Incidence in exposed}}{\text{Incidence in unexposed}}RR\; =\; \backslash frac\{\backslash text\{Incidence\; in\; exposed\}\}\{\backslash text\{Incidence\; in\; unexposed\}\}$$

• $RR=1RR\; =\; 1$: no association

• $RR>1RR\; >\; 1$: exposure is associated with increased risk

• : exposure is associated with decreased risk (protective)

3B. Strengths and Weaknesses

3C. Exam-Relevant Biases

Loss to follow-up (attrition bias) is the hallmark vulnerability of prospective cohort studies. If the people who drop out differ systematically from those who remain, the results become distorted. A vignette might describe a study where 40% of participants in the exposed group moved away before the study ended. That is your cue to identify attrition bias.

Confounding occurs when a third variable is associated with both the exposure and the outcome, creating a false or distorted association. For example, a cohort study finds that coffee drinkers have higher rates of lung cancer. But coffee drinkers are also more likely to smoke. Smoking is the confounder. Exam questions will test whether you can identify the confounding variable and know that stratification, multivariate analysis, or matching are used to control for it.

04Case-Control Studies

4A. Design and Logic

A case-control study starts from the opposite end. The investigator identifies individuals who already have the disease (cases) and individuals who do not have the disease (controls). Both groups are then asked about or assessed for prior exposures.

Because you start by selecting based on disease status, you do not know the total number of exposed and unexposed people in the source population. Therefore, you cannot calculate incidence, and you cannot calculate relative risk. Instead, you calculate the Odds Ratio (OR).

$$OR=\frac{a\times d}{b\times c}OR\; =\; \backslash frac\{a\; \backslash times\; d\}\{b\; \backslash times\; c\}$$

Where, using the standard 2x2 table:

The odds ratio approximates the relative risk when the disease is rare (prevalence < 10%). This is known as the rare disease assumption, and it is a commonly tested concept.

4B. Strengths and Weaknesses

4C. Exam-Relevant Biases

Recall bias is the defining vulnerability of case-control studies. Patients who are sick (cases) tend to remember and report exposures more thoroughly than healthy controls. A mother whose child has a birth defect will recall every medication she took during pregnancy in vivid detail; a mother with a healthy child may not. This asymmetry in recall distorts the odds ratio.

Selection bias arises when the controls are not drawn from the same source population as the cases. If your cases come from a tertiary referral center but your controls come from a community clinic, the two groups may differ in ways beyond just the disease status, invalidating the comparison.

A classic exam vignette: "Researchers identified 200 patients with pancreatic cancer and 200 patients without. They interviewed all subjects about prior dietary habits." This is a case-control study. The measure of association is the odds ratio. The primary bias threat is recall bias.

05Cross-Sectional Studies

5A. Design and Logic

A cross-sectional study measures both exposure and outcome at the same point in time. Think of it as a snapshot or a photograph of a population at one moment. It answers the question: "What is happening right now?"

Because both exposure and disease are assessed simultaneously, you cannot determine which came first. This means cross-sectional studies cannot establish temporal sequence and therefore cannot demonstrate causation or even a clear direction of association.

The primary measure is prevalence:

$$\text{Prevalence}=\frac{\text{Number of existing cases at a point in time}}{\text{Total population at that point in time}}\backslash text\{Prevalence\}\; =\; \backslash frac\{\backslash text\{Number\; of\; existing\; cases\; at\; a\; point\; in\; time\}\}\{\backslash text\{Total\; population\; at\; that\; point\; in\; time\}\}$$

Cross-sectional studies are often used for needs assessments, public health planning, and hypothesis generation. National health surveys (e.g., NHANES) are classic examples.

5B. Strengths and Weaknesses

5C. Exam-Relevant Bias

Prevalence-incidence bias (Neyman bias) is the key pitfall. Because you only capture people alive and present at the time of the survey, rapidly fatal conditions are underrepresented. If a disease kills patients quickly, a cross-sectional study will systematically miss those cases and underestimate the true burden of disease.

The classic exam trap: a cross-sectional study finds a low prevalence of a disease known to have high incidence. The explanation is that the disease is quickly fatal, so affected individuals die before they can be captured in the snapshot.

06Ecological Studies

6A. Design and Logic

Ecological studies analyze data at the population or group level, not the individual level. For example, comparing countries' per-capita alcohol consumption with their rates of liver cirrhosis. The unit of analysis is the population (country, state, region), not the person.

6B. The Ecological Fallacy

This is the single most tested concept related to ecological studies. The ecological fallacy is the error of assuming that associations observed at the group level also apply to individuals within those groups.

Example: A study finds that countries with higher fat consumption have higher rates of breast cancer. It would be an ecological fallacy to conclude that individual women who eat more fat are at higher risk for breast cancer. It may be that within those countries, the women who actually develop breast cancer are not the ones consuming the most fat.

Whenever you see population-level data being used to draw individual-level conclusions, identify the ecological fallacy.

07Case Reports and Case Series

These are purely descriptive. They cannot test hypotheses, calculate associations, or establish causation. They sit at the bottom of the evidence hierarchy. Their value lies in being the first signal that something new exists (e.g., the first reports of Pneumocystis pneumonia in previously healthy young men that heralded the AIDS epidemic).

08Measures of Association: Summary

09Bias and Confounding: High-Yield Review

Methods to control confounding:

• At the design stage: randomization (only in trials), restriction, matching

• At the analysis stage: stratification, multivariate regression, standardization

10Distractors

11Traps and High-Yield Pearls

The single most common way students lose points on observational study questions is by confusing case-control with retrospective cohort studies. Both look at past data. The distinction hinges entirely on the starting point: if the researcher begins by grouping subjects by disease status (has the disease vs. does not), it is case-control and the measure is the odds ratio. If the researcher begins by grouping subjects by exposure status (exposed vs. unexposed) and then determines who developed disease, it is a cohort study and the measure is relative risk, even if all the data were collected retrospectively from old charts.

Another frequent trap: a question will describe a cohort study and then ask you to calculate an odds ratio, or describe a case-control study and ask for relative risk. Remember that relative risk can only be directly calculated from cohort studies because you need incidence data. Case-control studies use the odds ratio because the incidence is unknown by design.

Watch for the phrase "at a single point in time" or "surveyed." That is the hallmark of a cross-sectional study. If you see "followed over time" or "followed for X years," it is a cohort study.

Finally, always remember that no observational study can independently prove causation. Even a well-designed cohort study can only demonstrate association. If a question asks what would best establish a causal relationship, the answer is almost always a randomized controlled trial, not an observational study. The exception is when ethical constraints make a trial impossible, in which case the answer shifts to a well-controlled prospective cohort study as the next best option.

The core competency being tested is your ability to rapidly identify a study design from a clinical scenario, select the correct measure of association, and recognize the bias that most threatens that design's validity.