Multivariate binary logistic regression
A multivariate binary logistic regression analysis was undertaken to determine the significance of contributors to the BOOST score and LACE index (Table 4 ). Model 1 considered each of the eight BOOST factors along with their sub-components where variables were not dichotomous. This model correctly predicted 71.8% of cases with a pseudo-R2 value of 0.308. The only value in this model which significantly contributed to readmission was recent hospitalisation (p = 0.005) which had an odds ratio of 4.6 (95% CI 1.6 – 13.6). Model 2 was composed of the eight BOOST score components alone. This model accurately predicted the highest number of cases with 74.5% of correct cases identified, and a pseudo-R2value of 0.197. Again, the only significant component was recent hospitalisation (p = 0.002) which had an odds ratio of 4.4 (95% CI 1.7 – 11.3). Model 3 was of the LACE index components. In this model emergency admissions were excluded as 100% of cases were classified as an emergency fo both readmisison and no readmisison. This model accurately predicted 72.7% of cases with a pseudo-R2value of 0.165. Recent attendance was the only significant contributor (p = 0.001) with an odds ratio of 2.0 (95% CI 1.3 – 3.0). Taken together all three models were able to predict roughly three quarters of cases, and in each the only significant predictor of readmissions was related to the number of recent hospitalisations. The BOOST score found a higher odds ratio associated with recent attendance compared to the LACE index which is likely due to the different nature of scoring this section; the BOOST score is binary for any attendance within past six months, and the LACE index provides a higher score for more attendances