PS4-58 USING REAL WORLD DATA TO STRUCTURE AND POPULATE MARKOV MODELS – A CASE STUDY OF TELEMONITORING FOR HEART FAILURE

Wednesday, October 21, 2015
Grand Ballroom EH (Hyatt Regency St. Louis at the Arch)
Poster Board # PS4-58

Praveen Thokala, PhD, Pete Dodd, PhD, Simon Dixon, PhD and Alan Brennan, BSc, MSc, PhD, University of Sheffield, Sheffield, United Kingdom
Purpose: Cost-effectiveness analysis used for health technology assessment frequently make use of Markov models, which characterise the disease progression into specific states based on clinical/biological measures, such as, forced expiratory volume in one second (FEV1) which represents a measure of lung function. However, these measures are not always collected in the evaluation of non-pharmaceutical service developments.  Likewise, detailed measures are not always available to allow locality specific transition probabilities to be generated.  As a result, models based on clinical/biological measures are limited in their applicability to several decision making contexts

Method: We present an alternative approach that uses routinely available hospitalisation data to define the states in the Markov model and estimate the transition probabilities for chronic heart failure patients. Hospitalisation data were accessed through a query run on national Hospital Episodes Statistics (HES) in the UK. Data were obtained for individuals who suffered at least 1 admitted patient care (APC) event for heart failure (ICD10 code I50) between March 2005 and March 2010, including identifiers relating to 152 localities, or Primary Care Trusts (PCTs). We also obtained mortality data from the Office for National Statistics with an anonymised identifier to allow linkage to HES data. These data were analysed using continuous time Markov assumptions to derive estimates of monthly probabilities of transition between states (defined as the number of hospitalisations in the previous year). 

Result: Transition matrices for HF patients in any locality (PCTs) in the UK can be obtained using this approach. A case study of estimating the cost effectiveness of telemonitoring is provided. 

Conclusion: We present an alternative approach to traditional economic modelling, by making use of routinely available data to characterise the disease states based on hospitalisations in the previous year. This approach has a number of advantages over conventional Markov modelling approaches, especially in chronic disease areas where hospitalisation is a useful measure of both effectiveness and disease progression. The use of routine data can also be generalised to other modelling techniques depending on which technique best suits the problem – for example, discrete event simulation can be used for modelling based on time to event data from claims databases. Given the increasing emphasis on using real world evidence, it is likely that these approaches can prove a valuable addition to traditional approaches in cost-effectiveness modelling.