Investigating Treatment Variations in Secondary Care Treatment: A Hierarchical Count Modelling Approach.

Presenter: Grace Lordan, University of Queensland

Abstract

Rationale: The literature to date has citied many causes as to why a patient may receive differential treatment to another, above and beyond their own illness characteristics. Examples include health insurance status, socioeconomic status, gender, age and other types of patient heterogeneity. Many of these papers fail to control adequately for the heterogeneity associated with patient illness type. The latter raises the question as to whether such results are spurious, with the significant patient characteristics arising only because of omitted illness heterogeneity effects.

Objectives: This paper analyses whether patients receive similar treatment to other patients who present to secondary care facilities with heart related problems. The data considered relates to hospitals in the Republic of Ireland and represents the micro level data of those who received treatment for such problems over a three year period. The primary objective of this work is to examine whether, even after controlling for illness heterogeneity, there still exists differences/inequities in the treatment received by patients in Ireland. These results are compared to an identical model that ignores illness heterogeneity.

Methodology: This work argues that once illness type, patient characteristics, hospital characteristics and the number of procedures received by each patient are controlled for, the length of stay of each patient should be relatively comparable. ICD-10 codes are used to control for illness heterogeneity. A hierarchical count model is argued to be the most appropriate for the analysis. The two levels in the data represent patients clustered with particular ICD 10 codes. The shape of the dependant variable lends itself to a count modeling approach, given that once admitted a patient must spend at least one night in hospital. To our knowledge this is the first application of a hierarchical count modeling approach using classical estimation techniques. The results of the hierarchical model are compared to non hierarchical modeling approach.

Results: The results from this analysis show that once patient illness is adequately controlled for patient characteristics that were once significant are now insignificant. However, the analysis is able to identify some remaining inequities that relate to patient characteristics. Even so, it seems that studies that fail to control adequately for patient illness will always run a danger of overestimating the treatment differential effects associated with the type of care that they are considering.

Conclusions: The treatment differential inequities identified here will be of interest to policy makers. The main conclusion from this work is that researchers should proceed with caution when interpreting any analysis that fails to control for patient level illness, but identifies treatment differentials in a particular health care type.