Methods for Evaluating Causality in Observational Studies.

Area of Science:

• Medical research methodology
• Epidemiology
• Biostatistics

Background:

• Randomized controlled trials (RCTs) are the gold standard for demonstrating causality in clinical research.
• Ethical and practical limitations often prevent the use of RCTs.
• Observational studies provide an alternative for inferring causality when RCTs are not feasible.

Purpose of the Study:

• To introduce two underutilized methods for assessing causal inferences in observational studies.
• To highlight the potential of novel quasi-experimental designs in medical research.

Main Methods:

• The study describes regression-discontinuity designs and interrupted time series analysis.
• Regression-discontinuity applies when a continuous variable with a threshold determines treatment assignment, approximating randomization near the threshold.
• Interrupted time series, a form of regression-discontinuity, uses time as the assignment variable, often marked by an external event.

Main Results:

• Regression-discontinuity allows causal inference by comparing outcomes for subjects just above and below a threshold.
• Interrupted time series enables the evaluation of intervention effects by analyzing trends before and after an event.
• These methods can establish causal relationships under specific conditions, such as a clear threshold or a distinct external event.

Conclusions:

• Regression-discontinuity and interrupted time series methods can be applied to observational data to derive causal inferences.
• Careful consideration and verification of the prerequisites for applying these methods are essential for valid results.