Errors in the Calculation of the Population Attributable Fraction.

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Population attributable fraction (PAF) is a key metric in public health for estimating the impact of exposures.
• Common errors in PAF calculation can lead to inaccurate public health interventions.
• The Levin formula is frequently used for PAF estimation, but its application requires careful consideration of risk ratios.

Purpose of the Study:

• To identify and visually demonstrate common errors in population attributable fraction (PAF) calculations.
• To analyze the impact of using adjusted risk ratios in the Levin formula.
• To illustrate the bias introduced by varying standardized mortality ratios (SMR) and associational risk ratios (aRR).

Main Methods:

• Visual analysis using wireframes to depict errors in PAF calculation.
• Systematic variation of standardized mortality ratio (SMR) and associational risk ratio (aRR) with fixed exposure prevalence.
• Examination of absolute and relative bias under different SMR and aRR conditions.

Main Results:

• Positive absolute bias occurs when SMR > 1 and SMR > aRR, increasing with the SMR-aRR difference.
• Negative absolute bias, with relatively small magnitude, is observed when aRR > SMR > 1.
• Relative bias exceeds one when SMR > aRR and is less than one when SMR < aRR.

Conclusions:

• The choice of risk ratio (adjusted vs. unadjusted) significantly impacts PAF calculation accuracy.
• Understanding the bias introduced by SMR and aRR is crucial for correct PAF interpretation.
• The target of causation for PAF is the exposed group, not the entire population.