Colloque / Séminaire

SEMINAIRE LABO - Guy LACROIX, Université Laval, Québec

Investigating adverse outcomes in a Neonatal Intensive Care Unit: Machine Learning and Bayesian Semiparametric Panel Data Models


Marc Beltempoa, Georges Bressonb, Guy Lacroixc,

aDepartment of Pediatrics, McGill University Health Centre, Montreal, QC, Canada
bDepartment of Economics, Université Paris II, Paris, France
cDepartment of Economics, Université Laval, Québec, QC, Canada



Neonatal intensive care units (NICUs) must contend with ever changing caseloads, patient mix and unplanned admissions (Tucker et al. (1999)). Workforce management is thus challenging and nursing overtime is often used to meet required nurse-to-patient ratios (Berney and Needleman (2005); Beltempo et al. (2016)). The increasing use of overtime hours as a labor management strategy has become an important issue across NICUs in Canada and elsewhere (Griffiths et al. (2014)). Indeed, nursing overtime has been found by some to be deleterious to adult patients’ health (Bae (2013); Lin (2014); Cimiotti et al. (2012)), although others have concluded otherwise (e.g. (Cook et al. (2012); Duffield et al. (2011)).)


The lack of clear evidence linking overtime and patient health may be due to methodological factors (Bae and Favry (2013); Weinstein et al. (2008)). Indeed, most studies use cross-sectional data and contrast health outcomes stemming from heterogeneous units and/or hospitals. Such analyses are likely to omit important unobserved patient characteristics and unit-specific work arrangements. As for NICUs, given that the mix of neonatologists, fellows, residents, nurse practitioners, etc. varies greatly across hospitals, singling out the contribution of overtime on the health outcomes of neonates is clearly a difficult task. This difficulty is compounded by the fact that many covariates may interact in a highly non-linear fashion, something traditional multivariate regressions techniques are incapable of unearthing.


The presentation will focus on the CHU de Quebec NICU, a tertiary/quaternary referral center with a 51-bed capacity that tends to a population of 1.7 million over a territory of 452,600 km2 in Canada. Focusing on a single unit removes some of the aforementioned variations in specialty mix across NICUs. We study the occurrence of health care-associated infections and medical incidents among all neonates admitted to the NICU between April 2008 and March 2013. Daily exposure to overtime and regular hours of work, as well as numerous individual and NICU-specific covariates are used to analyze the onset of the latter two outcomes.




Liste des horaires :