The use of aggregate data in the evaluation of hospital interventions: The case of payment by results for maternity care in England

It is common to use individual level data (e.g. patient, practice or hospital admission) for the evaluation of health policies. Individual level data allow health care researchers to keep track of patients over time. However, the authorization of the use of patient level data from public bodies is usually a long process and can be costly. Whilst aggregate data, for example at hospital or trust level, are generally considered to provide less information, we discuss here under what conditions aggregate and individual level data can provide the same level of information for estimation of treatment effects. We compare use of the Hospital Episode Statistics individual spell level data with the National Health Services (NHS) trust level data at the English NHS setting to evaluate the impact of Payment by Results (PbR) in maternity care. We exploit a short panel of two years from the Hospital Episode Statistics database and our identification strategy is based on the differences in the roll out of the policy across NHS trusts. We investigate under what circumstances trust level and individual spell level analyses provide identical results once appropriate weights are used at trust level. We examine when the estimated impact of PbR is identical comparing ordinary least square (OLS) and fixed effects models and how the use of fixed effects models contributes to our identification strategy by correcting for unobservable time invariant factors at hospital level. We show how to switch from individual level data to aggregate level data analysis within a difference in differences setting for OLS estimation with and without fixed effects. © Peter Lang GmbH Internationaler Verlag der Wissenschaften Berlin 2020. All rights reserved.