Quantitative data analysis

Analysis of Victorian data demonstrates that median length of acute hospital stay for advanced cancer patients in the last 90 days of life is 28 days, and the primary objective is to reduce the acute hospitalisation days in the last 90 days of life by 25% for patients with advanced cancer. This corresponds to approximately 6 days and is the average length of stay of a hospital admission. Length of stay in Victorian data is not normally distributed; transformation to the square root normalises it on inspection of a histogram. Analysis will be conducted at a system level (data described below). Assuming an intracluster correlation (ICC) of 0.01, a two-sided alpha of 0.05, a cell size of 30 participants provides > 99% power to detect a difference of 6 days. Varying the ICC from 0.001–0.1 whilst holding all other assumptions constant does not appreciably change the power. This translates to a total sample in the control and intervention periods of 450.

Description of individual and cell cluster characteristics to assess balance between control and practice change periods will be conducted using means (standard deviations) or median (intraquartile range) for continuous factors and frequencies (percentiles) for categorical variables. To assess the significance of differences in receipt of timely palliative care, a logistic regression model will be fitted with a random effect for cluster and a fixed effect for each step. Further analyses will assess for potential differences in 1) effect size of the practice change periods by site; and 2) temporal decrease in the proportion of participants accessing care during the successive practice change/maintenance periods. For other quantitative outcome variables, a similar analytic approach will be used. Analyses will be undertaken in Stata V15 (Stata Corp, College Station, Tx, USA) and level of significance set at 5%. Analyses will be conducted on an intention-to-treat basis, with reporting consistent with CONSORT guidelines.

Health resource costs will be observed longitudinally (up to death) for patients enrolled during the Care Plus practice change periods compared with patients enrolled during control periods. Costs of Care Plus implementation will be estimated to account for additional resources including: engagement and training of staff; and development, maintenance/ monitoring. Health utilization and the costs of hospital admissions, emergency department presentations, general practitioner and specialist outpatient clinics (including palliative care), the use of prescription medicines, and the use of community-based palliative care and nursing services will be included in the costing summary. Standard parametric techniques will estimate the 95% confidence limits and p-values for the differences in the mean values of the costs.