Randomisation by cluster and the problem of social class bias

Area of Science:

• Epidemiology
• Clinical Trials
• Public Health

Background:

• Individual randomisation is standard in population trials.
• Cluster randomisation uses groups (e.g., general practices) as the unit of randomisation.
• This design is increasingly discussed in statistical and epidemiological literature.

Purpose of the Study:

• To report on the experience of the Edinburgh randomised trial of breast cancer screening, which used general practices as clusters.
• To examine mortality outcomes (all causes, cardiovascular disease, lung cancer) over a 5-year follow-up period.
• To investigate methods for adjusting analyses for socio-economic status.

Main Methods:

• Utilised cluster randomisation, with general practices as the randomisation unit.
• Analysed mortality data from the first 5 years of follow-up.
• Examined methods for adjusting for socio-economic status in the analysis.

Main Results:

• Spurious mortality differences were observed when socio-economic status was not accounted for in the analysis.
• Methods for adjusting for socio-economic status at the analysis stage were examined.
• The study concluded that adjustments for socio-economic status are feasible and intended for future analyses.

Conclusions:

• Cluster randomisation requires careful consideration of potential biases, particularly related to socio-economic status.
• Adjustments for socio-economic status can be incorporated into the analysis phase of cluster-randomised trials.
• For future studies where outcomes are linked to socio-economic status, stratification by these criteria during randomisation is recommended if feasible.