4.4 AI made

Case Report/Series

Describes one or few patient observations. Useful for rare or novel findings.

Retrospective Observational Studies

Uses past data (e.g., cohort or cross-sectional). Observes associations, not causality.

Prospective Observational Studies

Monitors outcomes forward in time.

Non-Randomized Clinical Trials

Intervention applied prospectively, but without randomization or control group.

Randomized Controlled Trials (RCTs)

Gold standard. Random allocation of participants to intervention vs control.

Meta-Analysis of RCTs

Pooling data from multiple RCTs for aggregate analysis.

Phase I

Healthy volunteers; Safety, tolerability, pharmacokinetics; Typical Size

Phase II

Patients with the target disease; Preliminary efficacy + safety; Typical Size Hundreds.

Phase III

Large patient populations; Efficacy + safety → FDA approval; Typical Size Thousands.

Phase IV

Post-marketing surveillance; Long-term safety, rare adverse effects; Typical Size Population-scale.

Power (1-β)

Probability of detecting a true effect. Typical power target: 0.8-0.9 (10-20% chance of Type II error).

Type II error (β)

Failing to detect a difference that actually exists.

Underpowered studies

Lead to inconclusive results.

Sample size

Depends on expected event frequency, effect size (magnitude of difference expected), desired power and α (commonly 0.05).

Traditional RCT

Highly controlled, rigid protocol.

Pragmatic Trial

More flexible, real-world clinical environment.

Superiority Trial

Determine if intervention is better than control.

Non-Inferiority Trial

Determine if intervention is 'no worse than' comparator.

2x2 Factorial Design

Tests two interventions simultaneously; Efficient but complex and risk of confounding.

Randomized controlled trials

The gold standard in cardiovascular research because randomization minimizes bias and allows causal inference by ensuring groups differ only by the intervention.

Limitation of an underpowered clinical trial

It cannot distinguish between a true lack of difference and a failure to detect an existing difference (Type II error).

Primary Outcome

Central endpoint the trial is powered to detect. Determines sample size and trial success.

Secondary Outcomes

Exploratory or supportive analyses (e.g., subgroup results, mechanistic effects).

Hard Clinical Outcomes

Events directly relevant to patients and clinicians.

Composite Outcomes

Combines several events to increase event rate.

Surrogate Outcomes

Indirect markers correlated with clinical outcomes.

Hazard Ratio (HR)

Event rate ratio over time. Used for time-to-event outcomes (death, MI).

Odds Ratio (OR)

Ratio of event odds between groups. Used for binary outcomes.

Mean/Median Change

Difference in continuous variables BP, LDL-C.

Relative Risk Reduction (RRR)

% reduction in event risk. Often reported but may overstate benefit.

Absolute Risk Reduction (ARR)

Actual difference in event rates. Clinically more meaningful.

Statistical Significance

Unlikely due to chance. Suggests reliability of findings.

Clinical Significance

Magnitude and real-world importance. Determines if change matters for patients.

Subgroup Analyses

Explore disease severity, comorbidities, and demographics to personalize clinical application.

Population Studied

Who was included/excluded in the trial.

Intervention(s)

What treatment was tested in the trial.

Primary Outcome(s)

What endpoints were measured in the trial.

Main Result(s)

Numerical findings and statistical measures from the trial.

Conclusions

Clinical relevance and applicability of the trial findings.

Example of HR = 0.97, p < 0.001

Statistically significant but clinically trivial (3% reduction).

Example of HR = 0.76, p = 0.049

Clinically meaningful (24% reduction) but borderline statistically.

Why are composite outcomes used?

They increase event rates, improve feasibility, and capture multiple clinically relevant events.

Difference between hard clinical outcomes and surrogate outcomes

Hard outcomes measure real patient events like death or MI, whereas surrogate outcomes use biomarkers like LDL or BP that predict but don't guarantee clinical benefit.

Components differing between groups

Drug, delivery, setting

Causality inference

Only if all else is equal can causality be inferred

Well-designed RCT

Ensures identical groups except for intervention

Strength of evidence

Assessed by study design and sample size

Study limitations

Includes funding, bias, and power

Journal impact

Quality of peer-review

Transparency of funding

Clarity in funding and data analysis

Balanced randomization

Ensures equal distribution of characteristics

Adequate power

Necessary for the primary outcome

Practical clinical utility

Relevance of study findings to clinical practice

Underpowered trial

Risks Type II error when finding 'no difference'

Hard clinical outcome

Example: Stroke

Hazard ratio of 0.75

Implies borderline statistical but meaningful clinical significance

Surrogate endpoint

Example: Change in serum LDL-C

Pragmatic trials

Assess real-world effectiveness

Trial applicability

Most critical factor: Similarity of patient population

Required sample size

Most affected by expected event rate