These studies observe outcomes without manipulating exposure or treatment.
Describes a small group of patients with a shared condition or treatment.
Useful for:
Rare diseases
Unusual complications
Early therapeutic observations
Assesses individuals at a single point in time.
Evaluates associations between risk factors and outcomes.
Measures:
Prevalence (burden of disease)
Not suitable for determining causality
Compares individuals with a disease (cases) to those without (controls).
Retrospective design: looks back to identify risk factors.
Key features:
Controls should be from the same population as cases.
Ideal for studying rare diseases.
Vulnerable to recall bias and selection bias.
Participants grouped by exposure status (e.g., antibiotic use).
Can be:
Prospective: cohort followed over time to observe outcomes.
Good for rare exposures
Time-consuming and costly
Retrospective: both exposure and outcome have already occurred.
Limited by existing data
Measures incidence and can establish temporal relationships.
These involve active intervention and randomization.
Participants randomly assigned to treatment or control groups.
Controls for known and unknown confounders.
Groups:
Experimental group: receives intervention
Control group: receives placebo, standard therapy, or alternative
Reduces bias in outcome reporting and interpretation.
Types:
Single-blind: participants unaware of group assignment
Double-blind: both participants and investigators unaware
Intention-to-Treat (ITT):
Analyzes participants based on original assignment
Preserves randomization
Reflects real-world effectiveness
Per Protocol:
Analyzes only those who adhered to the assigned treatment
Measures ideal efficacy but risks bias
Participants switch between treatment and control during follow-up.
Each subject serves as their own control.
Requires washout periods to avoid carryover effects.
No blinding; all parties know treatment assignment.
Higher risk of bias.
Exploratory analysis conducted after study completion.
Not pre-specified; may identify unexpected patterns.
Risk of false positives due to multiple comparisons.
Examines specific patient subsets (e.g., age, comorbidities).
Can inform personalized care.
Risk of false positives/negatives if not pre-specified or adequately powered.
Synthesizes evidence from multiple studies addressing a specific question.
Follows a predefined protocol for study selection and evaluation.
Minimizes bias through transparency and reproducibility.
Combines data from multiple studies to produce a summary effect size.
Often visualized using a Forest plot.
Requires:
Homogeneity testing (e.g., Iยฒ statistic)
Careful pooling of studies with similar populations and outcomes
Cost-Benefit Analysis:
Both costs and benefits expressed in monetary terms
Cost-Effectiveness Analysis:
Compares cost per unit of health benefit (e.g., cost per life-year saved)
Benefit not expressed in dollars
Systematic error that distorts study findings.
Errors in data collection or measurement
Example: Recall bias in case-control studies
Systematic differences in how participants are chosen
Example: Healthy controls differ from cases in unmeasured ways
Occurs when a third variable influences both the exposure and outcome.
Not due to poor design, but natural mixing of effects
A confounder must:
Be associated with both exposure and outcome
Not lie on the causal pathway
Example: Hot weather correlates with drowning, but the true confounder is increased swimming pool use
Just because two variables are associated does not mean one causes the other.
Consistency: Multiple studies show similar results.
Strength: Strong association between exposure and outcome.
Temporality: Exposure precedes outcome.
Dose-response relationship: Greater exposure leads to greater effect.
Biologic plausibility: Mechanism makes sense biologically.
Consistency across populations: Findings hold across diverse groups.
Randomized, double-blinded, controlled trials minimize bias and confounding, making them the most reliable for establishing causation.
Null hypothesis (Hโ): No association between exposure and outcome.
Alternative hypothesis (Hโ): There is an association.
Reject Hโ if p-value < 0.05 (i.e., <5% chance result is due to random variation).
One-tailed: Assumes effect in one direction (based on prior knowledge).
Two-tailed: Tests for effect in either direction.
A 95% confidence interval (CI) gives a range for the estimate (e.g., OR or RR).
If the CI does not include the null value (e.g., OR = 1), the result is statistically significant.
Accuracy of the association within the study population.
Reflects how well the study was conducted.
Applicability of results to other populations or settings.
Consistency of a measurement across repeated trials.
Reflects precision and reproducibility.
Quantitative; infinite possible values.
Examples: blood pressure, glucose level
Qualitative; grouped into categories.
Categories without inherent order.
Example: blood types (A, B, AB, O)
Categories with a logical order, but intervals not equal.
Example: income levels (low, medium, high)
Bell-shaped curve; symmetric around the mean.
Asymmetric; tail on one side.
Mean: Average; sensitive to outliers.
Median: Middle value; less affected by extremes.
Mode: Most frequent value.
Extreme values tend to move closer to the average on repeat measurement.
Standard Deviation (SD): Variation around the mean.
Standard Error of the Mean (SEM): Variation of sample mean from population mean.
Formula: SEM = SD / โn
New cases in a population over time.
Total cases (new + existing) at a single point in time.
Ability to correctly identify those with the disease.
Formula: A / (A + C)
1 โ sensitivity = false negative rate
Ability to correctly identify those without the disease.
Formula: D / (B + D)
1 โ specificity = false positive rate
Positive Predictive Value (PPV): A / (A + B)
Negative Predictive Value (NPV): D / (C + D)
PPV and NPV depend on prevalence; sensitivity and specificity do not.
Positive LR: Sensitivity / (1 โ Specificity)
Negative LR: (1 โ Sensitivity) / Specificity
Graph of sensitivity (TP rate) vs 1 โ specificity (FP rate)
Visualizes trade-off between sensitivity and specificity
X-axis: FP rate (1 โ specificity)
Y-axis: TP rate (sensitivity)
Used when both the risk factor and outcome are categorical
Assesses whether observed proportions differ from expected proportions by chance
Compares means of continuous data between two groups
Used for related groups or matched pairs
Greater statistical power when pairs are similar
Helps reduce confounding
Used for independent groups (e.g., treatment vs control)
Compares means across more than two groups
Accounts for:
Group differences (e.g., placebo vs treatment)
Random variability (sampling error, individual differences)
Probability of observing a result as extreme as the one obtained, assuming the null hypothesis is true
p โค 0.05 is commonly used to reject the null hypothesis
Range around a sample estimate that reflects precision
95% CI means we are 95% confident the true value lies within the interval
Narrower CI = more precision (larger sample size helps)
If CI excludes the null value, the result is statistically significant
Incorrectly rejecting the null hypothesis when it is true
False positive; typically set at 0.05
Failing to reject the null hypothesis when it is false
False negative; commonly set at 0.2
Power = 1 โ ฮฒ: ability to detect a true difference
Larger sample size โ higher power
Ratio of disease probability in exposed vs unexposed
Formula:
Interpretation:
RR = 1 โ no association
RR > 1 โ increased risk
RR < 1 โ decreased risk (protective)
Used in cohort studies and clinical trials
Difference in incidence between exposed and unexposed
Reflects the absolute impact of exposure
Ratio of odds of exposure in cases vs controls
Formula:
Used in case-control studies
Interpretation:
OR = 1 โ no association
OR > 1 โ exposure more likely in cases
OR < 1 โ exposure less likely in cases (protective)
Compares instantaneous risk of an event between two groups
HR = 3.0 โ one group has 3ร the event rate of the other
Common in survival analysis
Average number of patients needed to treat to prevent one adverse outcome
Formula:
Lower NNT = more effective treatment
Example: NNT = 100 โ treat 100 patients to benefit 1
Provides structured, transparent ratings of evidence quality
Supports evidence-based recommendations
Study design (RCTs > observational)
Risk of bias
Consistency of results
Directness of evidence
Precision of estimates
Evidence is rated (high, moderate, low, very low)
Recommendations are made based on strength of evidence and clinical relevance