A panel data model of length of stay in hospitals for hip replacements

Abstract

Inequality between private and public patients in Australia has been an ongoing concern due to its two tiered insurance system. This article investigates the variations in hospital length of stay for hip replacements using the Victorian Admitted Episodes Dataset from 2003/2004 to 2014/2015, employing a Bayesian hierarchical random coefficients model with trend. We find systematic differences in the length of stay between public and private hospitals, after observable patient complexity is controlled. This suggests shorter stays in public hospitals due to pressure from the Activity-based funding scheme, and longer stays in private system due to potential moral hazard. Our counterfactual analysis shows that public patients stay 1.8 days shorter than private patients in 2014, which leads to the “quicker but sicker” concern that is commonly voiced by the public. We also identify widespread variations among individual hospitals. Sources for such variation warrant closer investigation by policy makers.

Keywords:

• Gibbs sampler
• hierarchical random coefficients
• hospital ranking
• length of stay; public versus private care

JEL CLASSIFICATION:

• C11
• H51
• I14

Acknowledgments

We extend our sincere thanks to the editors and reviewers for their insightful comments, which helped improve the article.

Funding

Gao, Zhang and Zhao would like to thank the Australian Research Council's Discovery Projects funding scheme under grant numbers [DP170104421, DP130104229 and DP140102345].

Notes

¹A patient entitled under laws that are or were in force in Victoria, other than Veterans’ Affairs legislation, to the payment of, or who has been paid compensation for, damages or other benefits in respect of the injury, illness or disease. This category includes workers compensation, transport accident, criminal injury and common law cases, members of the Defence Forces and seamen with personnel entitlements.

² An eligible person whose charges for this episode of care are met by the DVA (e.g., a gold card holder).

³ The AR-DRG classification is partly hierarchical, with 23 Major Diagnostic Categories (MDCs) into which the 661 AR-DRGs can be grouped. I03A and I03B are for THRs with high and low severity respectively, where I is an MDC standing for diseases and disorders of the musculoskeletal system and connective tissue; A indicates the highest consumption of resources and B the second highest.

⁴ To confirm our hypothesis, we run separate regressions by hospitals and compare the slope coefficients of each variable for all hospitals. We find substantial variation among the slope coefficients for severity, but less variation for other regressors.

⁵ We suspect that the significantly short LOS in p₇ is caused by patients from surrounding rural areas which leads to high pressure on hospitals to free up beds. Data shows patients in this hospital exhibit similar demographic characteristics and severity levels. They stay 1–3 days fewer than those in other hospitals.