Populasi Penelitian

01Core Definitions and Hierarchy

Understanding study population requires recognizing a three-tier hierarchy. Each tier narrows the group of people involved, and each transition introduces the possibility of bias.

Target population is the entire group to whom the researcher wants to apply the study results. For example, if a researcher studies a new antihypertensive drug, the target population might be "all adults with essential hypertension." This population is theoretical and usually too large to study directly.

Source population (also called the accessible population) is the subset of the target population that the researcher can realistically reach. Using the same example, this might be "adults with essential hypertension who attend outpatient clinics in a particular city." The source population is limited by geography, time, and practical access.

Study sample is the group of individuals actually enrolled in the study. These are the people who met the inclusion criteria, did not meet any exclusion criteria, consented to participate, and were selected through a defined sampling method.

The critical exam concept here is that every time you narrow from one tier to the next, you risk losing representativeness. If the study sample does not accurately reflect the target population, the study's external validity (generalizability) is compromised.

02Sampling Methods

Sampling is the process by which individuals are selected from the source population to form the study sample. The method chosen determines whether the sample is representative and whether the results can be generalized.

Probability sampling methods give every individual in the source population a known, nonzero chance of being selected. These methods reduce selection bias and support generalizability.

Non-probability sampling methods do not guarantee equal opportunity for selection. They are easier and cheaper to perform but carry a higher risk of bias.

A frequently tested distinction is stratified vs. cluster sampling. In stratified sampling, you divide the population into subgroups and sample from each. In cluster sampling, you sample entire groups and study everyone (or a random subset) within them. Stratified sampling increases precision; cluster sampling increases feasibility.

03Inclusion and Exclusion Criteria

Inclusion and exclusion criteria define who is eligible to enter the study. They serve two purposes: they ensure the study sample is appropriate for the research question, and they protect patient safety.

Inclusion criteria describe the characteristics a person must have to be enrolled. These usually involve the diagnosis of interest, age range, and disease severity. For example: "Adults aged 40 to 70 with newly diagnosed heart failure and an ejection fraction below 40%."

Exclusion criteria describe characteristics that disqualify a person from enrollment. These typically involve comorbidities that could confound results, contraindications to the study intervention, or inability to comply with the protocol. For example: "Patients with end-stage renal disease, active malignancy, or pregnancy."

The exam tests this concept in two main ways:

First, overly strict criteria improve internal validity (the study cleanly tests its hypothesis in a homogeneous group) but reduce external validity (the results may not apply to real-world patients who have comorbidities and varied demographics).

Second, overly broad criteria improve generalizability but may introduce confounding variables that make it harder to detect a true treatment effect.

04Bias Related to Study Population

The way a study population is assembled is the single largest source of selection bias. Selection bias occurs when the study sample systematically differs from the target population, producing results that do not reflect the true relationship between exposure and outcome.

Sampling bias (a subtype of selection bias) occurs when the method of selecting participants favors certain individuals over others. A classic example is studying disease prevalence using only hospital patients. Hospital-based populations are sicker than the general community, so prevalence estimates will be artificially high. This phenomenon is called Berkson bias (admission rate bias).

Volunteer bias (self-selection bias) occurs when individuals who volunteer for a study differ systematically from those who do not. Volunteers tend to be healthier, more educated, and more motivated, which can skew results.

Non-response bias occurs when a significant proportion of selected individuals decline to participate, and those who refuse differ from those who agree in ways that affect the outcome.

Loss to follow-up (attrition bias) occurs after enrollment, when participants drop out of a study in a non-random fashion. If patients who experience side effects are more likely to withdraw, the remaining sample will overestimate treatment efficacy.

05Internal Validity vs. External Validity

These two concepts represent a fundamental tension in study design, and exam questions frequently ask you to identify which one is being threatened.

Internal validity asks: "Are the results of this study true for the participants who were actually studied?" A study has high internal validity when it is free of bias and confounding within its own sample. Randomization, blinding, and strict eligibility criteria all protect internal validity.

External validity (generalizability) asks: "Can the results of this study be applied to patients outside the study?" A study has high external validity when its sample closely resembles the target population in demographics, disease severity, comorbidities, and setting.

The tension is straightforward. The more you control for confounders (strict criteria, single-center design, homogeneous sample), the higher your internal validity but the lower your external validity. The more you mirror real-world diversity (multicenter design, broad criteria, pragmatic endpoints), the higher your external validity but the harder it becomes to isolate the true effect.

Efficacy studies (explanatory trials) prioritize internal validity. They test whether an intervention works under ideal, controlled conditions.

Effectiveness studies (pragmatic trials) prioritize external validity. They test whether an intervention works in routine clinical practice.

06Sample Size and Power

Exam questions on study population often involve whether a study enrolled enough participants to detect a meaningful difference.

Statistical power is the probability that a study will correctly reject a false null hypothesis (i.e., detect a real effect if one exists). Power is conventionally set at 80% ($\beta =0.20\backslash beta\; =\; 0.20$) or higher.

The key factors that determine required sample size are:

• Effect size: The smaller the expected difference between groups, the larger the sample needed to detect it.

• Variability: Greater variability in the outcome measurement requires a larger sample.

• Significance level ($\alpha \backslash alpha$): A more stringent threshold (e.g., $\alpha =0.01\backslash alpha\; =\; 0.01$ instead of $0.050.05$) requires a larger sample.

• Power ($1-\beta 1\; -\; \backslash beta$): Higher desired power requires more participants.

• Dropout rate: Anticipated loss to follow-up must be factored in; researchers typically inflate the calculated sample size to compensate.

A Type II error ($\beta \backslash beta$ error) occurs when the study fails to detect a true difference. The most common cause of a Type II error on the exam is an inadequate sample size (underpowered study).

The formula conceptually relates these factors. While you are unlikely to be asked to calculate exact sample sizes, understanding the direction of each relationship is tested:

$$n\propto \frac{(Z_{\alpha }+Z_{\beta }{)}^2\cdot {\sigma }^2}{{\mathrm{\Delta }}^2}n\; \backslash propto\; \backslash frac\{(Z\_\{\backslash alpha\}\; +\; Z\_\{\backslash beta\})^2\; \backslash cdot\; \backslash sigma^2\}\{\backslash Delta^2\}$$

Where $nn$ is sample size per group, $Z_{\alpha }Z\_\{\backslash alpha\}$ and $Z_{\beta }Z\_\{\backslash beta\}$ are Z-scores corresponding to the chosen significance level and power, ${\sigma }^2\backslash sigma^2$ is variance, and $\mathrm{\Delta }\backslash Delta$ is the minimum clinically important difference.

07Traps and High-Yield Pearls

The most common way students miss questions on study population is by confusing the type of validity being compromised. When a vignette describes a study with very narrow inclusion criteria and results that seem clean and convincing, students often assume everything is fine. The trap is that the question is asking about generalizability (external validity), not about whether the results are internally sound.

Another frequent error is confusing stratified sampling with cluster sampling. Remember: in stratified sampling, you divide and then sample individuals from each stratum. In cluster sampling, you select entire clusters and study them. Stratified sampling improves precision; cluster sampling improves feasibility.

Students also conflate selection bias with confounding. Selection bias is a problem with how participants were chosen (it cannot be fixed after the fact with statistical adjustment). Confounding is a problem with a third variable that is associated with both the exposure and outcome (it can be addressed through randomization, matching, restriction, or multivariable analysis).

Finally, watch for questions about studies conducted at a single tertiary referral center. These studies often have high internal validity but poor external validity because their patient population is sicker and more complex than the general population. This is a textbook setup for testing your understanding of the Berkson bias concept and the limits of generalizability.

The core competency being tested is your ability to evaluate the link between how a study population is assembled and whether the study's conclusions are valid and applicable.