TY - JOUR

T1 - The Clinical Course of Cirrhosis

T2 - The Importance of Multistate Models and Competing Risks Analysis

AU - Jepsen, Peter

AU - Vilstrup, Hendrik

AU - Kragh Andersen, Per

PY - 2015/7/1

Y1 - 2015/7/1

N2 - Multistate models are models of disease progression that, for a patient group, define multiple outcome events, each of which may affect the time to develop another outcome event. Multistate models are highly relevant for studies of patients with cirrhosis; both the classical perception of cirrhosis as either compensated or decompensated and the recent, more complex models of cirrhosis progression are multistate models. Therefore, researchers who conduct clinical studies of patients with cirrhosis must realize that most of their research questions assume a multistate disease model. Failure to do so can result in severely biased results and bad clinical decisions. The analyses that can be used to study disease progression in a multistate disease model may be called competing risks analysis, named after the competing risks disease model, which is the simplest multistate disease model. In this review article, we introduce multistate disease models and competing risks analysis and explain why the standard armamentarium of Kaplan-Meier survival estimates and Cox regression sometimes gives bad answers to good questions. We also use real data to answer typical research questions about the course of cirrhosis and illustrate biases resulting from inadequate methods. Finally, we suggest statistical software packages that are helpful and accessible to the clinician-researcher.

AB - Multistate models are models of disease progression that, for a patient group, define multiple outcome events, each of which may affect the time to develop another outcome event. Multistate models are highly relevant for studies of patients with cirrhosis; both the classical perception of cirrhosis as either compensated or decompensated and the recent, more complex models of cirrhosis progression are multistate models. Therefore, researchers who conduct clinical studies of patients with cirrhosis must realize that most of their research questions assume a multistate disease model. Failure to do so can result in severely biased results and bad clinical decisions. The analyses that can be used to study disease progression in a multistate disease model may be called competing risks analysis, named after the competing risks disease model, which is the simplest multistate disease model. In this review article, we introduce multistate disease models and competing risks analysis and explain why the standard armamentarium of Kaplan-Meier survival estimates and Cox regression sometimes gives bad answers to good questions. We also use real data to answer typical research questions about the course of cirrhosis and illustrate biases resulting from inadequate methods. Finally, we suggest statistical software packages that are helpful and accessible to the clinician-researcher.

UR - http://www.scopus.com/inward/record.url?scp=84933182808&partnerID=8YFLogxK

U2 - 10.1002/hep.27598

DO - 10.1002/hep.27598

M3 - Review

C2 - 25376655

AN - SCOPUS:84933182808

VL - 62

SP - 292

EP - 302

JO - Hepatology

JF - Hepatology

SN - 0270-9139

IS - 1

ER -