Mapping SF-36 Onto the EQ-5D Index: How Stable is the Relationship?

Abstract

Rationale: The use of different health status measures in clinical trials has resulted in efforts to map between instruments (preference and non-preference based) in order to use them in economic evaluations. The aim would be to achieve interchangeability between instruments in order to allow comparison between health care programmes. This raises the issue of whether such relationships are stable across factors such as setting, socio-demographics and medical condition.

Objective: The aim of this study is to investigate the relationship between the preference-based EQ-5D index and the non-preference based SF-36 and to determine whether this relationship is affected by age, gender, setting and medical condition. Data and Methods: The study uses data from a cross sectional data repository (known as HODaR) which consists of health data collected from inpatients and outpatients at a large UK NHS Teaching Hospital Trust and links these data to existing routinely collected hospital data. This data set has a sample of 9,081 outpatients and 40,312 inpatients for 2002-2005. A series of models are examined by linear regression. The first model is a simple additive model using eight dimension scores of the SF-36 as explanatory variables with the EQ-5D index as the dependent variable. The specification was extended to examine quadratic and interactions terms. These models were then tested for their stability across inpatients and outpatients settings, age, gender and medical condition (as measured by ICD disease group). The performance of different models was analysed using the goodness of fit between observed and predicted values of Mean Square Error (MSE).

Results: The relationship between the EQ-5D and the SF-36 was affected by age but not gender. The relationship was significantly different for outpatients (MSE= 0.074) and inpatients (MSE=0.109) and varies according to medical condition as measured according to disease group, (for example MSE=0.061 for diseases of the genitourinary system). The specification of the model was improved by the inclusion of quadratic and interaction terms, finding a significant negative relationship for quadratic terms. The results are compared to other mapping approaches.

Conclusion: The relationship between the SF-36 and the EQ-5D is not stable and depends on age, setting and medical condition. These variables need to be accounted for when converting SF-36 data to the EQ-5D index. This work needs to be extended to more variables (such as ethnicity and country) and to examine the relationship between different instruments.

Keywords: Mapping, EQ-5D, SF-36