Biomedical Engineering Reference
In-Depth Information
begins in infancy or early childhood
18
; most children that suer the disease
have their rst asthmatic episode before their third birthday
9
. Pre-school
children hospitalized for asthma reportedly account for a disproportion-
ately large fraction of total acute pediatric asthma admissions and tend to
present the most dicult long-term disease management problems
9
.
Several previous studies have examined the relationship between neona-
tal conditions and asthma. For example, Schaubel et. al.
21
conducted a
retrospective cohort study that examined the eect of a wide variety of
suspected risk factors for pre-school asthma incidence using a large health
administrative database. Of current interest are risk factors for asthma-
attributable hospitalizations and days hospitalized, since both quantities
reect disease severity as well as health care costs. Since the correlation
among within-subject hospitalizations was not of direct interest, a marginal
model was chosen for our analysis.
Lawless and Nadeau
13
proposed a general class of marginal means mod-
els for recurrent event data. The class can accommodate time-dependent
effects. However, to the best of our knowledge, the appropriateness of
asymptotic results in nite samples in the presence of time-dependent ef-
fects has not been evaluated.
In this investigation, we examine the nite-sample properties of the
non-proportional rates model through simulation. We then assess the
eect of birth characteristics on hospitalizations and days hospitalized for
pre-school asthma using semi-parametric marginal rates models with time-
dependent covariates. The remainder of this article is organized as follows.
In Section 2, we describe the data sets used in this study. In Section 3, the
marginal rates model is described, with comparisons made to alternative ap-
proaches. A simulation study to evaluate the non-proportional rates model
is presented in Section 4. An analysis of the asthma data set is presented
in Section 5. The article concludes with a discussion in Section 6.
2. Data Sources
Data were obtained from Manitoba Health, a provincial health administra-
tion organization in Canada. Health care in Canada is publicly funded and
hence is, in theory, universal. For example, residents do not pay when they
go to the hospital. Manitoba was among the rst provinces in Canada to
assign a unique identier to residents at birth; the identier is known as the
Personal Health Identication Number (PHIN). Through the PHIN, it is
possible to track utilization of various health services (e.g., physician oce
Search WWH ::
Custom Search