Fair comparisons of hospitals
The Authority investigated methods to support fairer comparisons between hospitals.
Initially two different approaches to hospital-level risk-adjustment were undertaken and the results of both were similar. Results of univariate analyses identified risk factors for healthcare-associated S. aureus bloodstream infection that had some face validity. Multivariate analyses could not be completed as both approaches to modelling resulted in unacceptable models (see Future development).
As a result, the Authority developed the peer group classification system described in this section as an interim measure until reliable risk adjustment can be supported by the data.
The peer group methodology selected is a two-tiered system. Tier 1 is based on the established AIHW hospital peer classification system of 15 categories according to hospital size and hospital type. The Authority collapsed these 15 groups into six appropriate groups: major (A1), large (B1, B2), medium (C1, C2), small (D1, D2, D3), other (E2, E4, E5, E9, F, G) and specialist women’s and children (A2). These groups are shown in Table 1.
The group called ‘other’ hospitals is not a homogenous group and will not feature in comparison graphics in the Authority’s report. The specialist women’s and children’s hospitals (including Mercy Hospital for Women) are not compared because the demographics of patients admitted to these hospitals are very different from those admitted to major, large, medium and small hospital peer groups covered in this report. Results (cases and crude rates) for these hospitals will still be available on the MyHospitals website.
Tier 2 divides the modified AIHW peer groups into groups based on lower or higher percentages of vulnerable patient bed days. This division is applied because it is known that some patients are more vulnerable to S. aureus bloodstream infections or receive care that provides increased opportunity for bacterial infection of the bloodstream.
Table 1: Allocation of AIHW peer groups to modified AIHW peer groups (Teir 1)
|D1, D2, D3
|E2, E4, E5, E9, F, G
||Specialist women’s and children
Using information from literature review, clinical advisory groups and the above-mentioned univariate analyses, the Authority identified 10 hospital-level risk factors in two categories:
- Percentage of patient bed days of hospital care for immunosuppressed patients, including those admitted for:
- Bone marrow transplant
- Disorders of the immune system
- Oncology, including haematological malignancy
- Percentage of patient bed days for patients having treatments or procedures that increase the opportunity for infection, including those patients with:
- Acute renal failure
- Acute spinal injury
- Surgery (including cardiac surgery)
- Venous catheterisation.
All risk factors are statistically significant predictors of hospital-level healthcare-associated S. aureus bloodstream infection rates (p < 0.005) in the univariate analysis.
Based on the risk factor investigation, the Authority has defined an episode as ‘vulnerable’ if any of the 10 risk factors appear in an episode of care. The percentage of vulnerable patient days is calculated by dividing the total patient bed days of all vulnerable patient episodes in a hospital by the total all patient bed days at that hospital.
The split point used to identify hospitals with more or fewer vulnerable patients (tier 2) used histograms of the percentage of vulnerable patients (rounded to whole percentages) by the number of hospitals. The data were then investigated to one decimal place to assess the exact split point, arriving at the following divisions:
- Major hospitals ≥35.0%
- Large hospitals ≥30.0%
- Medium hospitals ≥22.5%
- Small hospitals ≥17.0%
Two hospitals in the large hospitals, more vulnerable patients peer group had especially high percentages of patient bed days attributable to vulnerable patients: Peter MacCallum Cancer Institute with 95% and the Royal Victorian Eye and Ear Hospital with 81%. Because these facilities are clear outliers with regard to patient bed days attributable to vulnerable patients, we advise caution when comparing these results to others in the peer group.
Table 2 presents all eight peer groups reported on and the average rate for each group. The table shows the average S. aureus bloodstream infections rate for the high percentage group are higher than the low percentage group within each tier 1 group. Goodness-of-fit statistics show that introducing tier 2 peer grouping increases the R-square from 0.29 to 0.37.