OBJECTIVES

: Quality of life (QOL) measurements are extremely important in outcome research as they reflect the effect of illness and treatment as perceived by the patients. The availability of QOL data, however, becomes a major barrier for including the patient perspective in big data healthcare studies. This research is trying to identify key predictors that can estimate QOL values within reasonable accuracy so they can be used in studies in conjunction with claims or EHR data.

METHODS

: National Health and Wellness Survey is self-administered online survey of adults 18 years and over. In China, the survey was administered to the urban population. Type 2 Diabetes (T2D) patient data from 2013 to 2017 were extracted for this analysis. 795 patients were included in the final working data set with 869 variables including QOL measure using EQ5D instruments. An iterative process of data transformation, recoding and LASSO regression was used to identify a subset of 14 variables that descriptively covers the clinical, social and economic aspects of T2D patients. A two-layer feed-forward artificial neural network (ANN) model was fitted to the data set with five-fold cross-validation.

RESULTS

: The ANN model demonstrated a good fit for the training data set with R-squared value of 0.81. The model also performed very well in the cross-validation with R-squared value of 0.86. Residual analysis supported that a good fit was achieved.

CONCLUSIONS

: QOL outcomes can be reasonably accurately approximated with a small set of predictive variables that are commonly available in large real-world databases. The combination of optimal predictive variables and robust models may enable QOL outcomes to be included in more real-world data studies.