Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

A Correlated Random Effects Hurdle Model for Excess Zeros with Clustered Data Based on BLUP (REMQL) Estimation.

Kim, Sung Hee (2011) A Correlated Random Effects Hurdle Model for Excess Zeros with Clustered Data Based on BLUP (REMQL) Estimation. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

[img]
Preview
PDF
Primary Text

Download (814kB) | Preview

Abstract

Community-acquired pneumonia (CAP) is a common, costly, and fatal illness; more than four million episodes occur in the United States each year. Providing quality and cost-effective care for CAP has an important implication in public health. Since inpatient treatment costs 20 times as much as outpatient treatment, and the costs of hospitalization drive inpatient costs, reducing length of stay (LOS) for patients with CAP may substantially reduce medical care costs and improve the effectiveness of health utilization. A potentially useful metric of efficiency is bed days, defined as zero for outpatients and LOS for inpatients, where LOS is the difference between discharge and admission dates. A surrogate for hospitalization costs, bed days, has problematic statistical properties (i.e., excess zeros and possible overdispersion). In multi-site studies, we also need to account for possible clustering by site. Researchers used finite mixture (FM) models or zero-inflated (ZI) models for bed days, assuming that valid zeros occur in both component distributions. However, the hurdle (H) model presumes all zero bed days are from outpatients. The H model has been extended to include correlated random effects in a generalized linear mixed model (GLMM) framework previously. Maximum Likelihood (ML) estimation is one of the most common approaches for estimating GLMMs. To avoid the intensive computing, convergence problems, and biased estimates of variance components associated with ML, we implemented best linear unbiased prediction (BLUP)-type estimation with restricted maximum quasi-likelihood (REMQL) of variance components in the correlated random effects H model. Several simulation studies validate this approach. We also applied the proposed random effects H model to the Emergency Department Community Acquired Pneumonia (EDCAP) study, a 32-site cluster-randomized trial to assess the effect of implementing medical practice guidelines on two aspects of care, e.g., admission and LOS. This allowed us to investigate whether the distribution of bed days varies by intervention arm (site-level) and the risk status (patient-level) among low risk patients with pneumonia. This appropriate modeling of bed days may facilitate identification of predictors of costly hospitalizations.


Share

Citation/Export:
Social Networking:
Share |

Details

Item Type: University of Pittsburgh ETD
Status: Unpublished
Creators/Authors:
CreatorsEmailPitt UsernameORCID
Kim, Sung Heesunghee0701@gmail.com
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairStone, Roslyn Aroslyn@pitt.eduROSLYN
Committee MemberChang, Chung-Chou Hchangj@pitt.eduCHANGJ
Committee MemberKim, Kevin Hkhkim@pitt.eduKHKIM
Committee MemberFine, Michael JMichael.Fine@va.gov
Committee MemberMazumdar, Satimaz1@pitt.eduMAZ1
Date: 29 June 2011
Date Type: Completion
Defense Date: 1 April 2011
Approval Date: 29 June 2011
Submission Date: 5 April 2011
Access Restriction: No restriction; Release the ETD for access worldwide immediately.
Institution: University of Pittsburgh
Schools and Programs: School of Public Health > Biostatistics
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: Best linear unbiased prediction (BLUP)-type estima; Excess zeros; Generalized linear mixed model; Maximum likelihood; Overdispersion; Restricted maximum quasi-likelihood
Other ID: http://etd.library.pitt.edu/ETD/available/etd-04052011-164650/, etd-04052011-164650
Date Deposited: 10 Nov 2011 19:34
Last Modified: 15 Nov 2016 13:38
URI: http://d-scholarship.pitt.edu/id/eprint/6797

Metrics

Monthly Views for the past 3 years

Plum Analytics


Actions (login required)

View Item View Item