OBJECTIVES: Efficient allocation of public resources require identification, measurement and quantification of costs and benefits of alternative programs. Patient reported outcomes are now routinely incorporated into economic evaluations of health technologies, but patient experience is often overlooked. The aim of this study is to develop a descriptive system for patient experience that can be valued and used to inform decision making.

METHODS: Analyses were conducted in a patient dataset, the Inpatient survey (2014), which collected information about healthcare delivery from over 62,000 NHS users across England. Statistical approaches were used to identify dimensions and items using data from patients who had an operation or procedure. In the first two approaches, dimensions based on latent construct were derived using exploratory factor analysis (EFA). Item selection for each dimension was conducted using structural equation modelling (SEM) and item response theory (IRT). For comparison logistic regression analyses were applied with respondents’ rating of overall patient experience specified as dependent variable.

RESULTS: EFA identified different factor models for patients with A&E and planned admissions respectively and factors contained 1 to 7 items. Bifactor models were fitted to assess unidimensionality before item selection using SEM and IRT. The two techniques identified different item as most significant variable in each factor. The 11 and 8 items identified for the two group of patients broadly related to trust and dignity, comfort and cleanliness, and clear communication to patients. Regression analyses identified a large number of independent items that were correlated with each other.

CONCLUSIONS: A measure that is amenable to valuation consists of items that are distinct yet related to each other. The measurement model identified from the dataset for patients that underwent an operation or procedure was different for those with planned admission compared to emergency admission. Different methods of item selection yielded different measurement models.