Testing independence between two Poisson-generated multinomial variables in case-series and cohort studies

Mounia Hocine (1) , Didier Guillemot (2) , Pascale Tubert-Bitter (1) , Thierry Moreau (1)

1 Epidémiologie et Biostatistique
2 CeRBEP - Centre de Ressources en Biostatistiques, Epidémiologie et Pharmaco-épidémiologie appliquées aux Maladies Infectieuses

Abstract In case-series or cohort studies, we propose a test of independence between the occurrences of two types of recurrent events (such as two repeated infections) related to an intermittent exposure (such as an antibiotic treatment). The test relies upon an extension of a recent method for analysing case-series data, in the presence of one type of recurrent event. The test statistic is derived from a bivariate Poisson generated-multinomial distribution. Simulations for checking the validity of the test concerning the type I error and the power properties are presented. The test is illustrated using data from a cohort on antibiotics bacterial resistance in schoolchildren.

Mots clés: conditional likelihood; bivariate multinomial; recurrent events; case‐series; cohort; bacterial resistance

Domaines: Méthodologie [stat.ME]; Santé publique et épidémiologie

Mounia Hocine, Didier Guillemot, Pascale Tubert-Bitter, Thierry Moreau. Testing independence between two Poisson-generated multinomial variables in case-series and cohort studies. Statistics in Medicine, 2005, 24 (24), pp.4035-4044. ⟨10.1002/sim.2413⟩. ⟨hal-02497777⟩