Estimating the effect of health service delivery interventions on patient length of stay: a Bayesian survival analysis approach

Research output: Contribution to journalArticlepeer-review

Colleges, School and Institutes

External organisations

  • Queen Elizabeth Hospital Birmingham

Abstract

Health service delivery interventions include a range of hospital ‘quality improvement’ initiatives and broader health system policies. These interventions act through multiple causal pathways to affect patient outcomes and they present distinct challenges for evaluation. In this article, we propose an empirical approach to estimating the effect of service delivery interventions on patient length of stay considering three principle issues: (i) informative censoring of discharge times due to mortality; (ii) post-treatment selection bias if the intervention affects patient admission probabilities; and (iii) decomposition into direct and indirect pathways mediated by quality. We propose a Bayesian structural survival model framework in which results from a subsample in which required assumptions hold, including conditional independence of the intervention, can be applied to the whole sample. We evaluate a policy of increasing specialist intensity in hospitals at the weekend in England and Wales to inform a cost-minimisation analysis. Using data on adverse events from a case note review, we compare various specifications of a structural model that allows for observations of hospital quality. We find that the policy was not implemented as intended but would have likely been cost saving, that this conclusion is sensitive to model specification, and that the direct effect accounts for almost all of the total effect rather than any improvement in hospital quality.

Bibliographic note

Publisher Copyright: © 2021 The Authors. Journal of the Royal Statistical Society: Series C (Applied Statistics) published by John Wiley & Sons Ltd on behalf of Royal Statistical Society

Details

Original languageEnglish
JournalJournal of the Royal Statistical Society Series C (Applied Statistics)
Early online date10 Jun 2021
Publication statusE-pub ahead of print - 10 Jun 2021

Keywords

  • Bayesian, causal model, direct and indirect effects, health services research, survival analysis