A multilevel analysis of three randomised controlled trials of the Australian Medical Sheepskin in the prevention of sacral pressure ulcers

Pressure ulcers (PUs) are a highly prevalent problem.1,2 They have a large impact on quality of life3,4 and high costs are involved in treatment.5-7 There is a lack of knowledge about the effectiveness and cost-effectiveness of many of the devices that help prevent PUs.8,9 One such preventive device is the sheepskin. Sheepskins were commonly used in the past, but due to poor product specification and inferior manufacturing practices, combined with the influx of synthetic, imitation products, the use of genuine sheepskins declined. Poorly manufactured natural sheepskins were difficult to launder and did not meet infection-control standards; synthetic sheepskins were found to have little pressure-relieving capacity.10,11

The Commonwealth Scientific and Industrial Research Organisation (CSIRO) renewed interest in the natural product and in 1998 introduced the Australian Medical Sheepskin, specified by the Australian Standard AS4480.1-1998.12 This sheepskin is capable of withstanding repeated washing at 80°C, satisfying infection-control requirements.13 Its dense, 25 mm natural wool pile reduces pressure, shear and friction forces, and has the capacity to absorb moisture away from the patient’s skin. Its effectiveness has been studied in two randomised controlled trials (RCTs) in hospital patients14,15 (in 2000 and 2004) and in a recent RCT in nursing homes (in 2009).16

We aimed to estimate the overall effectiveness of the Australian Medical Sheepskin in preventing sacral PUs across these three RCTs, using two methods: first, conventional meta-analysis of the pooled published effect sizes; and second, multilevel analysis of the combined individual patient data, which should give a more reliable and informative estimate of the effect size.17,18

Selection of the three trials used for the combined analysis was based on the search results of two systematic reviews,9,19 complemented by an additional recent trial known to us.16 The characteristics of the three trials are described in Box 1.

To enable us to pool the results, we identified the common interventions and outcomes. In all three trials, patients were randomly allocated to either use of an Australian Medical Sheepskin overlay in their beds, or usual care with a standard mattress. While both Australian trials used the incidence of any PUs, the Dutch trial used the incidence of sacral PUs only. We therefore analysed the effect of the sheepskin only on the incidence of sacral PUs; we considered the three trials to be sufficiently comparable to allow this analysis. Severity of ulcers was categorised in the three trials according to comparable four-grade systems. The length of follow-up in the hospital trials was until discharge from hospital, and in the nursing home trial, to 30 days after admission or until discharge (whichever came first).

Differences in patient characteristics (PU risk, age, and hospital or nursing ward) between the trials were taken into account in the analyses.

We contacted the study authors and asked for permission to use their individual patient data and, if they consented, to send the original (de-identified) data of all patients for whom a completed record was obtained. Datasets were standardised by giving common variables the same names, and basic frequency analyses were performed and checked against the published results. The standardised datasets were sent to the study authors for verification that the process had been applied correctly, before being merged into one dataset of only those variables used in all three trials. The combined data were again sent to the study authors for a final check that the merging process had been performed correctly.

A conventional meta-analysis was performed on the pooled published effect sizes from each trial using the Mantel–Haenszel method with a random effects model, computed using Review Manager, version 5.0, 2008 (Nordic Cochrane Centre, Cochrane Collaboration, Copenhagen, Denmark).

Multilevel binary logistic regression analysis was performed on the individual patient data using MLwiN, version 2.02, 2005 (Centre for Multilevel Modelling, University of Bristol, Bristol, UK), with three nested levels: patients within wards within institutions. The independent variable was the allocated group; the dependent variable was the incidence of sacral PUs during the period of observation. We controlled for age, sex, number of days of observation and the PU risk at admission (measured with the Braden scale in all three RCTs20). Multilevel analyses were done for the three trials separately and for the combined dataset. The results are expressed as odds ratios with 95% confidence intervals.

Risk of bias in the studies was assessed on five aspects as published in the two systematic reviews9,19 for the two Australian trials, and on our own judgement for the Dutch trial (Box 2).

Ethics approval had been obtained for each of the trials. Permission to pool the data was obtained from the primary authors and their managers. All individual patient data were de-identified.

The total population consisted of 1281 patients from 11 institutions and involved 45 nursing wards. The three combined RCTs covered more than 20 000 observation days. The mean observation length was about 17 days, varying from about 7 and 11 in the hospital trials to 28 in the nursing home trial. Intervention and control groups were comparable in the separate trials and in the combined dataset with respect to age, sex, PU risk and number of observation days. The main characteristics of the patients in the individual studies and in the total group are shown in Box 3.

The overall incidence of sacral PUs in the three trials was 5.4% (35 PUs) in the intervention group versus 12.2% (78 PUs) in the control group (bivariate χ2, P < 0.001). The relative risk reduction was 56% (95% CI, 35%–70%). Of the 113 newly acquired sacral PUs, 92 (81%) were grade 1 (least severe), 19 (17%) were grade 2, and 2 (2%) were grade 3.

The conventional meta-analysis showed an odds ratio for developing a grade 1–4 sacral PU during the period of observation while using a sheepskin of 0.37 (95% CI, 0.17–0.77; heterogeneity, I2 = 59%), compared with no sheepskin. The multilevel logistic regression analysis showed an odds ratio of 0.35 (95% CI, 0.23–0.55). The results of both analyses are shown in Box 4 and Box 5.

Conventional meta-analysis of effect sizes for the three trials demonstrated that the Australian Medical Sheepskin was effective in preventing sacral PUs, with an odds ratio of 0.37. The multilevel analysis on the combined individual patient data resulted in an odds ratio of 0.35, confirming that the Australian Medical Sheepskin is effective. Using individual patient data gave a confidence interval almost 50% smaller than that from the conventional meta-analysis, and hence a more precise effect estimate. Moreover, multilevel analysis of individual patient data is preferred because it can take patient and ward characteristics into account.

Although both approaches give an overall effect estimate, both remain liable to risk of bias in the trials, for instance due to inadequate allocation concealment or suboptimal outcome assessment. Furthermore, although the risk of bias of the underlying studies was assessed as much as possible on judgements from published systematic reviews, there might still be a bias in the pooled analysis because we were all also co-authors of the pooled studies.

In line with earlier studies, our results have shown that the Australian Medical Sheepskin is an effective aid in preventing PUs.9,19 However, further research about the effectiveness of the Australian Medical Sheepskin is still needed. All three trials involved short observation periods, and this can be a critical stage, as it is known that many PUs develop when patients are relatively immobile,21 and most PUs in nursing homes develop within the first 3 weeks after admission.22 How good the Australian Medical Sheepskin is in preventing PUs in patients confined to bed for longer terms is still unknown. Research involving larger sample sizes is also needed to study its effect in the prevention of higher grade PUs. In practice, the Australian Medical Sheepskin is an effective option for use in the prevention of pressure ulcers, and our findings suggest that guidelines need no longer discourage its use.

5 Conventional and multilevel analyses of pooled data from three randomised controlled trials of the Australian Medical Sheepskin

* Pooled published effect sizes. † Pooled individual patient data.