School absenteeism as a proxy outcome for infection rates: A simulation-based power analysis to inform the design of...

Evaluating the impact of school-based non-pharmaceutical interventions (NPIs), also referred to as public health and social measures (PHSM), on respiratory infections is crucial but challenging. Effect sizes are typically modest, while direct measurement of infections at scale in schools is complicated by individual informed consent requirements and substantial resource demands. We recently demonstrated the value of routinely registered daily school absenteeism data as a proxy for acute respiratory illness rates among students. To support the design of future intervention studies, we provide practical guidance for using school absenteeism as an outcome measure and examine how key parameters influence statistical power.

We simulated school-based intervention trials evaluating the effect of an intervention on school absenteeism. The outcome was the number of absent student days. In the primary cluster-randomized design, schools were randomized to intervention or control and absenteeism was analysed at the school level; within-school randomization and school-level cross-over designs were explored as alternatives. We evaluated approximately 2800 trial scenarios, with 250 simulation runs per scenario, varying the overall intervention effect, between-school variation in intervention effect, allocation ratio, number of participating classes per school, study duration, and baseline absenteeism levels. Power curves were used to determine the minimum number of schools needed to achieve 80% power. Random forest models were used to identify the parameters most strongly influencing sample size requirements.

The base case scenario assumed a 20% intervention effect in reducing absenteeism, moderate between-school variation, 1:1 allocation, participation of all classes per school, a 26-week study period (autumn to May break), and standard baseline absenteeism. Under these assumptions, 70 schools (35 intervention, 35 control) were required to achieve 80% power. Based on ∼2800 simulated trial scenarios, the impact of varying key design parameters, including average intervention effect, number of participating classes per school, and allocation ratio, on required sample size was quantified. Sample size requirements were lower for class-level randomization and cross-over designs than for the cluster-randomized design.

This study provides practical guidance for the design of well-powered school-based intervention trials using school absenteeism as a proxy outcome measure for infectious diseases and demonstrates how key design assumptions influence sample size requirements and feasibility.