Author(s)
Borg S1, Gerdtham UG2, Rydén T3, Munkholm P4, Odes S5, Langholz E6, Moum B7, Annese V8, Bagnoli S9, Beltrami M10, Clofent J11, Friger M12, Milla M13, Mouzas I14, O'Morain C15, Politi P16, Riis L4, Stockbrugger R17, Tsianos E18, Vardi H12, Lindgren S191The Swedish Institute for Health Economics, Lund, Sweden, and Faculty of Medicine, Lund University, Lund, Sweden, Lund, Sweden, 2Department of Economics, Lund University, Lund, Sweden, and Faculty of Medicine, Lund University, Lund, Sweden, 3Royal Institute of Technology, Stockholm, Sweden, 4Herlev Hospital and University of Copenhagen, Copenhagen, Denmark, 5Soroka Medical Center, Beer-Sheva, Israel, 6Gentofte Hospital and University of Copenhagen, Hellerup, Denmark, 7Oslo University Hospital & University Oslo, Oslo, Norway, 8University Hospital Careggi, Florence, Italy, 9Firenze, Florence, Italy, 10Reggio Emilia, Rome, Italy, 11Hospital Meixoeiro, Vigo, Spain, 12Ben Gurion University of the Negev, Beer-Sheva, Israel, 13University Hospital Carreggi, Florence, Italy, 14Crete University, Heraklion, Greece, 15Trinity Centre for Health Sciences, Dublin, Ireland, 16Istituti Ospitalieri di Cremona, Cremona, Italy, 17University Hospital Maastricht, Maastricht, Netherlands, 18University of Ioannina, Ioannina, Greece, 19Lund University, Malmö, Sweden
OBJECTIVES: Crohn's disease is a chronic relapsing-remitting inflammatory bowel disease with heterogeneous disease course, requiring life-long treatment. Phenotypes explaining disease heterogeneity is of interest in optimizing allocation of healthcare resources, e.g. to avoid expensive maintenance treatment to prolong remission in patients who seldom relapse. To develop economic models for evaluation of treatments, our objective was to estimate parameters of a Markov chain from data on disease activity and resource consumption and to improve model fit by allowing different phenotypes. METHODS: We had individual data on relapse and remission, surgery, use of medicines and other resources, aggregated over three month periods, from inflammatory bowel disease patients from 1991 and ten year onwards. Data from Crohn's disease patients were extracted. An exact maximum likelihood estimator using observations aggregated over time was used to estimate monthly transition probabilities. This estimator was adjusted to allow different disease phenotypes using an Expectation-Maximization method which identifies the phenotypes that best describe patient heterogeneity. The estimated parameters were used to derive the mean durations of a relapse and a period of remission to describe the phenotypes. RESULTS: At least two distinct phenotypes were found in each country, seldom-relapsing (once/3 years). The best fit was with four phenotypes in Denmark, three phenotypes in the Netherlands and in Italy, and two in Norway, Israel, Ireland, Spain, and Greece. In Denmark and Italy there was a single seldom-relapsing phenotype and more than one often-relapsing phenotype. In Netherlands there was two seldom-relapsing phenotypes. Denmark, Netherlands, Israel, Ireland and Italy have roughly as many seldom-relapsing as often-relapsing patients. Norway, Spain and Greece have a majority of seldom-relapsing patients. CONCLUSIONS: Allowing for different phenotypes improves model fit. Healthcare resource allocation can be optimized using phenotypes. Using data aggregated over time appears to remain a challenge.
Conference/Value in Health Info
2012-11, ISPOR Europe 2012, Berlin, Germany
Value in Health, Vol. 15, No. 7 (November 2012)
Code
PRM38
Topic
Real World Data & Information Systems
Topic Subcategory
Reproducibility & Replicability
Disease
Gastrointestinal Disorders, Systemic Disorders/Conditions
