Background Identifying the risk factors in patients that are more susceptible to drug related problems (DRPs) promotes closer pharmacotherapy monitoring that prevents morbidity–mortality in these patients.
Purpose To develop a predictive model of hospital mortality risk in older patients.
Material and methods We included patients >64 years admitted to a geriatric centre with 233 beds in a university hospital from January to September 2016. We determined the relationship between mortality and number of DRPs detected during admission, adjusted to these variables: age, sex, admission unit (acute geriatric unit (AGU), convalescence, psychogeriatric), Barthel Index and Pfeiffer test before admission, length of stay, number of chronic drugs/patient, DRP type (indication, efficacy, safety, other) and number of potentially inappropriate prescriptions (PIP, according to STOPP-START 2015, Beers 2015 and Priscus criteria) with pharmacist intervention. We used a predictive model of multivariate logistic regression, including significant variables in the bivariate analysis by using the χ2 test for binary qualitative data, the Kruskal–Wallis test for >2 categories and the Mann–Whitney U test for quantitative data. In the bivariate model, p≤0.1 was considered statistically significant and in multivariate analysis, p<0.05 was considered statistically significant. Statistical analysis was performed with Stata13.
Results 523 patients were included. Admission unit: AGU 359 (68.6%) patients; convalescence 103 (19.7%); and psychogeriatrics 61 (11.6%). Median age 86 (82–89) years. Women 292 (55.8%). Discharged 488 (93.3%). Died 102 (19.5%). Of 13 potential predictors, 8 were statistically significant in the bivariate analysis and 3 in the multivariate analysis. Protective factors: Barthel Index (OR=0.99; 95% CI 0.98–1.00); length of stay (OR=0.97; 95% CI 0.95–0.99); number of drugs (OR=0.97, 95% CI 0.91–1.04); intervention of PIPs (OR=0.91; 95% CI 0.69–1.20); and PRM security (OR=0.33, 95% CI 0.08–1.47). Risk factors: age (OR=1.04; 95% CI 1.00–1.09); Pfeiffer test (OR=1.02; 95% CI 0.93–1.13); and psychogeriatrics (OR=2.58; 95% CI 1.19–5.58). The model likelihood ratio test was significant (χ2=37.46, df=10, p<0.001). Regarding the goodness of fit test, the model explained 13.0% of data uncertainty (Nagelkerke index). It correctly classified mortality in 82.21% of patients. Sensitivity: 8.33%; specificity: 99.4%; positive predictive value: 77.78%; and negative predictive value: 82.3%. The AUC of the ROC curve for the mortality and mortality predicted variable was 0.69 (95% CI 0.653–0.741).
Conclusion The results indicate that this logistic model acceptably classifies patients with an increased risk of mortality, and helps us to identify which patients should undergo pharmacotherapy monitoring.
References and/or acknowledgements PMID: 27194906.
No conflict of interest