|
Introduction The frequency, expense and possible quality of care aspects of hospital readmissions make it important to identify patients who are at risk of unplanned return to hospital.1 Initial work on the development of a risk model for identifying elderly patients at risk of multiple unplanned hospital admissions was performed by the pharmacy practice research group as part of an overall investigation into drug-related problems in the elderly.2
Method A total of 487 elderly patients (aged 76.0±7.3 years) with non-elective emergency admissions was identified. Each hospitalised patient was interviewed and medical, demographic and socio-economic data obtained from their medical charts. These patients were followed up for one-year post-discharge with all unplanned readmissions during this time period recorded. The data were dichotomised and entered into SPSS for analysis. Univariate analysis (chi-squared) was used to identify variables that had a minimal association with one or more readmissions 12 months post-discharge (P<0.25). These were entered into backward stepwise elimination logistic regression analysis (model entry = 0.05). Examination of the significance of the log-likelihood ratio test for each variable determined its contribution to the model. The refined model was then validated using similar data retrospectively collected from medical charts regarding 732 elderly patients (aged 75.8±7.1 years). |
Focal points
|
Results Backward logistic regression analysis yielded a nine-variable model (Table 1), which contains both predictive and protective variables for one or more readmissions 12 months post-discharge from hospital.
|
This model has a specificity of 79.3 per cent, sensitivity of 60.0 per cent and overall accuracy of 71.2 per cent (cut-off point = 0.5). Receiver operating characteristic curves yielded an AUC of 0.77, thus indicating accurate discrimination of true and false positives by the refined model across all thresholds of risk.
When the model was applied to the validation population, a specificity of 68.7 per cent, sensitivity of 52.5 per cent and accuracy of 63.7 per cent was obtained (cut-off point = 0.3). Application of Hosmer and Lemeshow's goodness-of-fit test yielded a chi-squared value of 3.62 with eight degrees of freedom (P=0.89), therefore the refined model had acceptable fit to the validation population data.
Discussion This refined and validated hospital readmissions predictive model could be used by different health care professionals to identify vulnerable patients upon admission to hospital who require comprehensive discharge planning. Pharmacists can use these screening criteria to enable effective discharge counselling to take place with those patients who need it most.
School of pharmacy, Queen's University of Belfast, 97 Lisburn Road, Belfast, BT9 7BL; *academic practice unit, Antrim Area hospital, 45 Bush Road, Antrim
| 1. Waite K, Oddone E, Weinberger M, Samsa G, Foy M, Henderson W. Lack of association between patients' measured burden of disease and risk for hospital readmission. J Clin Epidemiol 1994;47:1229-36. |
| 2. McElnay JC, McCallion CR, Al-Deagi F, Scott MG. Hospital readmissions of elderly patients: a predictive model. Proc ESCP Drug Information Conference 1997:38-39. |