Attribution of patient reported outcomes to the effects of care providers

Lead Research Organisation: University of Manchester
Department Name: Medical and Human Sciences


Until this year, the NHS has not collected information routinely on the health of patients receiving different types of treatments from different health care providers. Consequently, it has been limited in its ability to perform three important functions: (1) spot well-performing and poorly-performing hospitals and general practices; (2) identify where its resources are best spent; and (3) offer useful information to patients when they have a choice of where to go for treatment or of the types of treatment they could receive. The collection of information on health and well-being is being rapidly introduced for all patients across the country. This is likely to show substantial variations between providers and treatments. Variations like these may reflect differences in the effectiveness and quality of different providers. However, there is a substantial amount of variation in levels of health experienced by different individuals that has little or nothing to do with the care they have received. To compare providers and interventions on a ?like-for-like? basis, and be sure we are really seeing variations in effect, we need to examine how the outcomes for the same (or very similar) patients change when they receive treatment from different providers. This is complicated by the fact that patients with long-term health problems often receive treatment from many different providers. We propose to use methods that economists have recently developed to separate out the effects of care providers from the large variation we observe between individual patients. We will apply these to the data that the NHS is currently collecting on patient outcomes and five datasets that have been collected in other recent research projects. By comparing the results across these six datasets, we will be able to make recommendations on the amount of data that needs to be collected from patients and how variations between care providers should be interpreted. Our team involves individuals with expertise in different types of analysis and medicine. These include two senior NHS managers who have responsibility for promoting the effective use of resources. They will ensure we communicate our findings effectively to local and national decision-makers.

Technical Summary

Patient-reported outcomes are being collected routinely for four elective surgical procedures and will be extended to more patient groups and services over the next few years. Publication of their mean levels for care providers and interventions are intended to enable better patient and commissioner choice and increase allocative efficiency. These scores may be misleading if they do not accurately reflect the true effects of providers and interventions. A significant contribution to the variation in these scores will be heterogeneity between individual patients. Only a small proportion of this heterogeneity can be captured by measured patient characteristics. We have assembled a team containing health economic, econometric, statistical, clinical and health services research expertise and senior NHS roles. We propose to apply and develop techniques recently adopted by labour econometricians for estimating and separating effects attributable to employees and employers in matched datasets. These fixed-effect models allow flexibly for correlation between individual and group heterogeneity and we can use them to retrieve estimates of the extent to which there is selection or assortative matching of patients and providers. Thus, we can identify when variations in patient outcomes can be reliably attributed to provider and/or intervention ?effects?. We propose to apply these techniques to six datasets containing patient-reported outcomes. The datasets have widely-varying design properties, including different extents of longitudinal data on individuals, focuses on acute and long-term conditions, and numbers and types of providers. They are therefore a good representation of the types of outcomes datasets that will exist. We will compare the performance of the estimation techniques that can be applied to datasets with these design properties in terms of their theoretical appeal, the empirical distribution of provider effects that they produce and the results of applying placebo tests where pseudo-provider effects are implausible. Through effective presentation and dissemination, and the close involvement of two senior NHS economists, we will ensure that our results feed into national and local processes for analysing and interpreting this valuable new source of information on patient outcomes.


10 25 50