Nested partially latent class models for dependent binary data; Estimating disease etiology

Research output: Contribution to journalArticle

Abstract

The Pneumonia Etiology Research for Child Health (PERCH) study seeks to use modern measurement technology to infer the causes of pneumonia for which gold-standard evidence is unavailable. Based on case-control data, the article describes a latent variable model designed to infer the etiology distribution for the population of cases, and for an individual case given her measurements. We assume each observation is drawn from a mixture model for which each component represents one disease class. The model conisidered here addresses a major limitation of the traditional latent class approach by taking account of residual dependence among multivariate binary outcomes given disease class, hence reducing estimation bias, retaining efficiency and offering more valid inference. Such "local dependence" on each subject is induced in the model by nesting latent subclasses within each disease class. Measurement precision and covariation can be estimated using the control sample for whom the class is known. In a Bayesian framework, we use stick-breaking priors on the subclass indicators for model-averaged inference across different numbers of subclasses.Assessment of model fit and individual diagnosis are done using posterior samples drawn by Gibbs sampling. We demonstrate the utility of the method on simulated and on the motivating PERCH data.

Original languageEnglish (US)
Pages (from-to)200-213
Number of pages14
JournalBiostatistics
Volume18
Issue number2
DOIs
StatePublished - Apr 1 2017

Fingerprint

Latent Class Model
Binary Data
Dependent Data
Pneumonia
Health
Case-control Data
Research
Latent Class
Gold
Binary Outcomes
Latent Variable Models
Gibbs Sampling
Mixture Model
Observation
Demography
Model
Technology
Valid
Class
Latent class model

Keywords

  • Bayesian methods
  • Case-control studies
  • Etiology
  • Latent class model
  • Local dependence

ASJC Scopus subject areas

  • Statistics and Probability
  • Medicine(all)
  • Statistics, Probability and Uncertainty

Cite this

Nested partially latent class models for dependent binary data; Estimating disease etiology. / Wu, Zhenke; Knoll, Maria Deloria; Zeger, Scott.

In: Biostatistics, Vol. 18, No. 2, 01.04.2017, p. 200-213.

Research output: Contribution to journalArticle

@article{f799794e924f4a13855f69fd7a63329e,
title = "Nested partially latent class models for dependent binary data; Estimating disease etiology",
abstract = "The Pneumonia Etiology Research for Child Health (PERCH) study seeks to use modern measurement technology to infer the causes of pneumonia for which gold-standard evidence is unavailable. Based on case-control data, the article describes a latent variable model designed to infer the etiology distribution for the population of cases, and for an individual case given her measurements. We assume each observation is drawn from a mixture model for which each component represents one disease class. The model conisidered here addresses a major limitation of the traditional latent class approach by taking account of residual dependence among multivariate binary outcomes given disease class, hence reducing estimation bias, retaining efficiency and offering more valid inference. Such {"}local dependence{"} on each subject is induced in the model by nesting latent subclasses within each disease class. Measurement precision and covariation can be estimated using the control sample for whom the class is known. In a Bayesian framework, we use stick-breaking priors on the subclass indicators for model-averaged inference across different numbers of subclasses.Assessment of model fit and individual diagnosis are done using posterior samples drawn by Gibbs sampling. We demonstrate the utility of the method on simulated and on the motivating PERCH data.",
keywords = "Bayesian methods, Case-control studies, Etiology, Latent class model, Local dependence",
author = "Zhenke Wu and Knoll, {Maria Deloria} and Scott Zeger",
year = "2017",
month = "4",
day = "1",
doi = "10.1093/biostatistics/kxw037",
language = "English (US)",
volume = "18",
pages = "200--213",
journal = "Biostatistics",
issn = "1465-4644",
publisher = "Oxford University Press",
number = "2",

}

TY - JOUR

T1 - Nested partially latent class models for dependent binary data; Estimating disease etiology

AU - Wu, Zhenke

AU - Knoll, Maria Deloria

AU - Zeger, Scott

PY - 2017/4/1

Y1 - 2017/4/1

N2 - The Pneumonia Etiology Research for Child Health (PERCH) study seeks to use modern measurement technology to infer the causes of pneumonia for which gold-standard evidence is unavailable. Based on case-control data, the article describes a latent variable model designed to infer the etiology distribution for the population of cases, and for an individual case given her measurements. We assume each observation is drawn from a mixture model for which each component represents one disease class. The model conisidered here addresses a major limitation of the traditional latent class approach by taking account of residual dependence among multivariate binary outcomes given disease class, hence reducing estimation bias, retaining efficiency and offering more valid inference. Such "local dependence" on each subject is induced in the model by nesting latent subclasses within each disease class. Measurement precision and covariation can be estimated using the control sample for whom the class is known. In a Bayesian framework, we use stick-breaking priors on the subclass indicators for model-averaged inference across different numbers of subclasses.Assessment of model fit and individual diagnosis are done using posterior samples drawn by Gibbs sampling. We demonstrate the utility of the method on simulated and on the motivating PERCH data.

AB - The Pneumonia Etiology Research for Child Health (PERCH) study seeks to use modern measurement technology to infer the causes of pneumonia for which gold-standard evidence is unavailable. Based on case-control data, the article describes a latent variable model designed to infer the etiology distribution for the population of cases, and for an individual case given her measurements. We assume each observation is drawn from a mixture model for which each component represents one disease class. The model conisidered here addresses a major limitation of the traditional latent class approach by taking account of residual dependence among multivariate binary outcomes given disease class, hence reducing estimation bias, retaining efficiency and offering more valid inference. Such "local dependence" on each subject is induced in the model by nesting latent subclasses within each disease class. Measurement precision and covariation can be estimated using the control sample for whom the class is known. In a Bayesian framework, we use stick-breaking priors on the subclass indicators for model-averaged inference across different numbers of subclasses.Assessment of model fit and individual diagnosis are done using posterior samples drawn by Gibbs sampling. We demonstrate the utility of the method on simulated and on the motivating PERCH data.

KW - Bayesian methods

KW - Case-control studies

KW - Etiology

KW - Latent class model

KW - Local dependence

UR - http://www.scopus.com/inward/record.url?scp=85020223448&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85020223448&partnerID=8YFLogxK

U2 - 10.1093/biostatistics/kxw037

DO - 10.1093/biostatistics/kxw037

M3 - Article

C2 - 27549120

AN - SCOPUS:85020223448

VL - 18

SP - 200

EP - 213

JO - Biostatistics

JF - Biostatistics

SN - 1465-4644

IS - 2

ER -