Study population and study design
The Australian PPS data used in this study was collected in 2018 in a sample of adult patients in 19 public, large acute care hospitals. The surveillance methodology was based on the European Centre for Disease Prevention and Control (ECDC) PPS protocol [13]. The types of HAIs that were selected for this study were as described in Cassini et. al [4]. HAIs were defined as per the ECDC protocol [13], with data collected by two research assistants, and entered into a secure online web-based survey tool.
A total of 2767 patients were sampled. Results from this PPS have previously been reported in detail [11, 14]. The median age of patients was 67 (IQR 49–79, range 18–104). Of these, 52.9% (1465) were male, 46.6% (1289) female and 0.5% [13] unknown/other. A majority (85.7%) of patients were from major city hospitals, with the remaining 14.3% from regional services.
Outcome measures
As well as the number of cases, we estimate deaths and DALYs for each condition. DALYs are a composite measure of years lived with disability (YLDs) and years life lost (YLLs), accounting for incidence, severity, and mortality of disease simultaneously. They also provide a way to compare the impact of disease across conditions, as opposed to simply ranking by incidence of prevalence.
Estimation methodology
The same approach as used for the estimation of the burden of HAIs in Germany was applied to this PPS [3], except for the choices of age strata. As the Australian PPS was only collected in adults, and involved a smaller sample size, strata were chosen to be 18–24, 25–34, 35–44, …, > 75. As the probability of death following an SSI is dependent on age (and thus on strata), the BHAI R package was modified to be compatible with these strata (Personal communication, B. Zacher). The disease outcome trees, transition probabilities and disability weights were otherwise the same as though used by Cassini et al. [4]. For full details of the outcome trees, see the supplement of Cassini et al. or the ECDC BCoDE toolkit [10].
The process of estimation can be summarised into three steps. The first step is to use the PPS data to estimate the hospital prevalence, which is estimated as
$$\begin{aligned} P = & {\text{Beta}}\left( {n_{obs} ,N - n_{obs} + 1} \right) \\ & + \left( {1 - r} \right){\text{Beta}}\left( {n_{obs} + 1,N - n_{obs} } \right) \\ \end{aligned}$$
(1)
\(n_{obs}\) is the number of patients observed with a HAI and \(N\) is the total number of patients in the PPS. This formula extrapolates from a zero-inflated binomial sample (which is seen here due to the relatively low prevalence of HAIs) to a population level estimate using a mixture of two Beta distributions. Next, this estimate is converted to hospital incidence,
$$I = P\frac{LA}{{LOI}}$$
where \(P\) is the hospital prevalence from Eq. (1), \(LA\) is the mean length of stay and \(LOI\) is the mean length of infection. For this study, the mean length of stay, \(LA\), was set to 5.3 days, from the AIHW 2018 statistics on all public hospitals (excluding same-day separations). Following the methodology of Zacher et. al, the mean length of infection was estimated using the censored length of infection from the survey and the Grenander estimator.
The final step in the estimation is the population incidence, which is calculated as
$$I_{pop} = I \times N_{discharges}$$
The survey used in this study was in acute public hospitals for patients over 18 years of age, which accounts for approximately 60% of separations in public hospitals for patients over 18 years of age, giving \(N_{discharges} = 3,713,513\).
To enable comparison between the European and German burden estimates, both datasets were re-aggregated to match the wider stratification used for estimation in the Australian setting. It is noted that the data for these surveys is aggregated into five-year age bands, and so the lowest age category for these studies is 15–24 (as opposed to 18–24). However, the burden in those aged between 15 and 18 is relatively low, so is expected to have little impact on the results.
As the Australian PPS used the ‘light’ survey design as specified by the ECDC, McCabe scores are not recorded. We applied the McCabe score distribution of the ECDC PPS to Australia, assuming that the McCabe score distribution in Australia would be similar to that observed in the EU. It is noted that there is little evidence of the applicability or lack thereof of these estimates to the Australian population.