Cross-calibration of Stroke disability measures: Bayesian analysis of longitudinal ordinal categorical data using negative dependence

Giovanni Parmigiani, Heidi W. Ashih, Gregory P. Samsa, Pamela W. Duncan, Sue Min Lai, David B. Matchar

Research output: Contribution to journalArticle

Abstract

It is common to assess disability of stroke patients using standardized scales, such as the Rankin Stroke Outcome Scale (RS) and the Barthel Index (BI). The RS, which was designed for applications to stroke, is based on assessing directly the global conditions of a patient. The BI, which was designed for more general applications, is based on a series of questions about the patient's ability to carry out 10 basic activities of daily living. Because both scales are commonly used, but few studies use both, translating between scales is important in gaining an overall understanding of the efficacy of alternative treatments, and in developing prognostic models that combine several datasets. The objective of our analysis is to provide a tool for translating between BI and RS. Specifically, we estimate the conditional probability distributions of each given the other. Subjects consisted of 459 individuals who sustained a stroke and who were recruited for the Kansas City Stroke Study from 1995 to 1998. We assessed patients with BI and RS measures 1, 3, and 6 months after stroke. In addition, we included data from the Framingham study, in the form of a table cross-classifying patients by RS and coarsely aggregated BI. Our statistical estimation approach is motivated by several goals: (a) overcoming the difficulty presented by the fact that our two sources report data at different resolutions; (b) smoothing the empirical counts to provide estimates of probabilities in regions of the table that are sparsely populated; (c) avoiding estimates that would conflict with medical knowledge about the relationship between the two measures; and (d) estimating the relationship between RS and BI at three months after the stroke, while borrowing strength from measurements made at 1 month and 6 months. We address these issues via a Bayesian analysis combining data augmentation and constrained semiparametric inference. Our results provide the basis for comparing and integrating the results of clinical trials using different disability measures, and integrating clinical trials results into a comprehensive decision model for the assessment of long-term implications and cost-effectiveness of stroke prevention and acute treatment interventions. In addition, our results indicate that the degree of agreement between the two measures is less strong than commonly reported, and emphasize the importance of trial designs that include multiple assessments of outcome.

Original languageEnglish (US)
Pages (from-to)273-281
Number of pages9
JournalJournal of the American Statistical Association
Volume98
Issue number462
DOIs
StatePublished - Jun 2003

Fingerprint

Negative Dependence
Ordinal Data
Data Dependence
Nominal or categorical data
Disability
Bayesian Analysis
Stroke
Calibration
Clinical Trials
Table
Semiparametric Inference
Estimate
Data Augmentation
Statistical Estimation
Cost-effectiveness
Bayesian analysis
Categorical data
Decision Model
Conditional probability
Conditional Distribution

Keywords

  • Constrained Markov chain Monte Carlo
  • Crebrovascular disease
  • Local odds ratio
  • Longitudinal contingency tables
  • Outcome measures
  • Stroke
  • Totally positive dependence

ASJC Scopus subject areas

  • Mathematics(all)
  • Statistics and Probability

Cite this

Cross-calibration of Stroke disability measures : Bayesian analysis of longitudinal ordinal categorical data using negative dependence. / Parmigiani, Giovanni; Ashih, Heidi W.; Samsa, Gregory P.; Duncan, Pamela W.; Lai, Sue Min; Matchar, David B.

In: Journal of the American Statistical Association, Vol. 98, No. 462, 06.2003, p. 273-281.

Research output: Contribution to journalArticle

Parmigiani, Giovanni ; Ashih, Heidi W. ; Samsa, Gregory P. ; Duncan, Pamela W. ; Lai, Sue Min ; Matchar, David B. / Cross-calibration of Stroke disability measures : Bayesian analysis of longitudinal ordinal categorical data using negative dependence. In: Journal of the American Statistical Association. 2003 ; Vol. 98, No. 462. pp. 273-281.
@article{763af2eea3c54c84af85672c4439bb3a,
title = "Cross-calibration of Stroke disability measures: Bayesian analysis of longitudinal ordinal categorical data using negative dependence",
abstract = "It is common to assess disability of stroke patients using standardized scales, such as the Rankin Stroke Outcome Scale (RS) and the Barthel Index (BI). The RS, which was designed for applications to stroke, is based on assessing directly the global conditions of a patient. The BI, which was designed for more general applications, is based on a series of questions about the patient's ability to carry out 10 basic activities of daily living. Because both scales are commonly used, but few studies use both, translating between scales is important in gaining an overall understanding of the efficacy of alternative treatments, and in developing prognostic models that combine several datasets. The objective of our analysis is to provide a tool for translating between BI and RS. Specifically, we estimate the conditional probability distributions of each given the other. Subjects consisted of 459 individuals who sustained a stroke and who were recruited for the Kansas City Stroke Study from 1995 to 1998. We assessed patients with BI and RS measures 1, 3, and 6 months after stroke. In addition, we included data from the Framingham study, in the form of a table cross-classifying patients by RS and coarsely aggregated BI. Our statistical estimation approach is motivated by several goals: (a) overcoming the difficulty presented by the fact that our two sources report data at different resolutions; (b) smoothing the empirical counts to provide estimates of probabilities in regions of the table that are sparsely populated; (c) avoiding estimates that would conflict with medical knowledge about the relationship between the two measures; and (d) estimating the relationship between RS and BI at three months after the stroke, while borrowing strength from measurements made at 1 month and 6 months. We address these issues via a Bayesian analysis combining data augmentation and constrained semiparametric inference. Our results provide the basis for comparing and integrating the results of clinical trials using different disability measures, and integrating clinical trials results into a comprehensive decision model for the assessment of long-term implications and cost-effectiveness of stroke prevention and acute treatment interventions. In addition, our results indicate that the degree of agreement between the two measures is less strong than commonly reported, and emphasize the importance of trial designs that include multiple assessments of outcome.",
keywords = "Constrained Markov chain Monte Carlo, Crebrovascular disease, Local odds ratio, Longitudinal contingency tables, Outcome measures, Stroke, Totally positive dependence",
author = "Giovanni Parmigiani and Ashih, {Heidi W.} and Samsa, {Gregory P.} and Duncan, {Pamela W.} and Lai, {Sue Min} and Matchar, {David B.}",
year = "2003",
month = "6",
doi = "10.1198/016214503000044",
language = "English (US)",
volume = "98",
pages = "273--281",
journal = "Journal of the American Statistical Association",
issn = "0162-1459",
publisher = "Taylor and Francis Ltd.",
number = "462",

}

TY - JOUR

T1 - Cross-calibration of Stroke disability measures

T2 - Bayesian analysis of longitudinal ordinal categorical data using negative dependence

AU - Parmigiani, Giovanni

AU - Ashih, Heidi W.

AU - Samsa, Gregory P.

AU - Duncan, Pamela W.

AU - Lai, Sue Min

AU - Matchar, David B.

PY - 2003/6

Y1 - 2003/6

N2 - It is common to assess disability of stroke patients using standardized scales, such as the Rankin Stroke Outcome Scale (RS) and the Barthel Index (BI). The RS, which was designed for applications to stroke, is based on assessing directly the global conditions of a patient. The BI, which was designed for more general applications, is based on a series of questions about the patient's ability to carry out 10 basic activities of daily living. Because both scales are commonly used, but few studies use both, translating between scales is important in gaining an overall understanding of the efficacy of alternative treatments, and in developing prognostic models that combine several datasets. The objective of our analysis is to provide a tool for translating between BI and RS. Specifically, we estimate the conditional probability distributions of each given the other. Subjects consisted of 459 individuals who sustained a stroke and who were recruited for the Kansas City Stroke Study from 1995 to 1998. We assessed patients with BI and RS measures 1, 3, and 6 months after stroke. In addition, we included data from the Framingham study, in the form of a table cross-classifying patients by RS and coarsely aggregated BI. Our statistical estimation approach is motivated by several goals: (a) overcoming the difficulty presented by the fact that our two sources report data at different resolutions; (b) smoothing the empirical counts to provide estimates of probabilities in regions of the table that are sparsely populated; (c) avoiding estimates that would conflict with medical knowledge about the relationship between the two measures; and (d) estimating the relationship between RS and BI at three months after the stroke, while borrowing strength from measurements made at 1 month and 6 months. We address these issues via a Bayesian analysis combining data augmentation and constrained semiparametric inference. Our results provide the basis for comparing and integrating the results of clinical trials using different disability measures, and integrating clinical trials results into a comprehensive decision model for the assessment of long-term implications and cost-effectiveness of stroke prevention and acute treatment interventions. In addition, our results indicate that the degree of agreement between the two measures is less strong than commonly reported, and emphasize the importance of trial designs that include multiple assessments of outcome.

AB - It is common to assess disability of stroke patients using standardized scales, such as the Rankin Stroke Outcome Scale (RS) and the Barthel Index (BI). The RS, which was designed for applications to stroke, is based on assessing directly the global conditions of a patient. The BI, which was designed for more general applications, is based on a series of questions about the patient's ability to carry out 10 basic activities of daily living. Because both scales are commonly used, but few studies use both, translating between scales is important in gaining an overall understanding of the efficacy of alternative treatments, and in developing prognostic models that combine several datasets. The objective of our analysis is to provide a tool for translating between BI and RS. Specifically, we estimate the conditional probability distributions of each given the other. Subjects consisted of 459 individuals who sustained a stroke and who were recruited for the Kansas City Stroke Study from 1995 to 1998. We assessed patients with BI and RS measures 1, 3, and 6 months after stroke. In addition, we included data from the Framingham study, in the form of a table cross-classifying patients by RS and coarsely aggregated BI. Our statistical estimation approach is motivated by several goals: (a) overcoming the difficulty presented by the fact that our two sources report data at different resolutions; (b) smoothing the empirical counts to provide estimates of probabilities in regions of the table that are sparsely populated; (c) avoiding estimates that would conflict with medical knowledge about the relationship between the two measures; and (d) estimating the relationship between RS and BI at three months after the stroke, while borrowing strength from measurements made at 1 month and 6 months. We address these issues via a Bayesian analysis combining data augmentation and constrained semiparametric inference. Our results provide the basis for comparing and integrating the results of clinical trials using different disability measures, and integrating clinical trials results into a comprehensive decision model for the assessment of long-term implications and cost-effectiveness of stroke prevention and acute treatment interventions. In addition, our results indicate that the degree of agreement between the two measures is less strong than commonly reported, and emphasize the importance of trial designs that include multiple assessments of outcome.

KW - Constrained Markov chain Monte Carlo

KW - Crebrovascular disease

KW - Local odds ratio

KW - Longitudinal contingency tables

KW - Outcome measures

KW - Stroke

KW - Totally positive dependence

UR - http://www.scopus.com/inward/record.url?scp=0042243515&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0042243515&partnerID=8YFLogxK

U2 - 10.1198/016214503000044

DO - 10.1198/016214503000044

M3 - Article

AN - SCOPUS:0042243515

VL - 98

SP - 273

EP - 281

JO - Journal of the American Statistical Association

JF - Journal of the American Statistical Association

SN - 0162-1459

IS - 462

ER -