External validation of mortality prognostic indices after hospital discharge in older adults (original) (raw)

2017, European journal of internal medicine

Prognostic indices of mortality in the elderly population may be useful to the clinician: they provide a mortality objective estimation that complements clinical judgment when giving information to patients; they can help select populations at high risk of poor outcomes related to adjustments in therapeutic interventions and, finally, they may also be useful for comparing outcomes among units, hospitals and care systems [1]. In addition, as regards clinical decision-making for older adults, prognostic indices offer clinical information beyond arbitrary cutoff points based on age as well as contribute to reduce potential ageism. To use them effectively the indices must incorporate specific variables related with multi-morbidity as well as functional, mental and nutritional status that, in very old patients, present a greater prognostic ability than specific markers of disease [2-4]. Indices with a stronger and more consistent predictive accuracy in old patients usually include comorbidity and functional status. Several prognostic indices that collect these specific predictive variables in the old population have been published with a discriminatory accuracy between moderate and good [5]. However, the external validity of these indices has been scarcely proven [6]. Therefore, the aim of this study is to develop an external validation of the published mortality index in elderly patients admitted due to acute medical illness. For this purpose, we included patients aged 75 and older admitted to an acute geriatric unit (AGU) in 2009. We only selected patients admitted for the first time in the unit. Patients who died during admission were excluded. The baseline variables were collected from the hospital's medical records's discharge summary and coding service: sociodemographic variables (age, sex, social support network), clinical variables (diagnoses at admission, comorbidity measured by the Charlson Index [7]), analytical parameters (serum albumin, creatinine and creatinine clearance at admission). Barthel Index (BI) [8] two weeks prior to admission and on discharge was the functional variable collected and cognitive situation was assessed using the Mental Scale of the Red Cross (MRC). The date of death after hospital discharge was extracted from the National Death Index of the Spanish Ministry of Health. (http://msssi.gob.es/estadEstudios/estadisticas/ etministerio/IND_tipodifusion.htlm). We selected prognostic indices to test if they were developed in older patients cohorts admitted due to acute medical illness [5]. Four indices that fit our selection criteria and have sufficient data for application or for adaptation were: Walter index, Levine Index, Burden of Illness Score for Elderly Persons (BISEP), and "Sujet Agé Fragile: Evaluation et Suivi Frail Elderly Subject" (SAFES). Finally, we also assessed comorbidity (Charlson Index) [7] and functional scores (Barthel Index) [8]. Receiver operating characteristic (ROC) curves were calculated to estimate the areas under the curves (AUC) and the C statistic was used to assess the predictive accuracy in relation to death at 6, 12 and 36 months. To compare C statistics for different prognostic indices, standard errors and Zscores were calculated based on methods described by Hanley and McNeil [9]. During the study 1256 patients aged 75 years old or older were admitted in the unit. The 166 patients readmitted during the year and the 113 that died during admission were excluded, leaving 977 patients. The characteristics of these patients according to their survival one year after discharge are presented in Table 1. Patients who died were older, had greater comorbidity, worse functional and mental status previous to admission, and poorer nutritional status. They also had worse health outcomes with longer hospital stays and greater institutionalization at discharge. The predictive accuracy of the four scales studied at 6, 12 and 36 months is shown in Table 1. The scale that proved the best prognostic accuracy was the Walter index, being very similar to the Barthel Index at discharge. Charlson's index showed the worst predictive accuracy, even with the lowest AUC index (p < 0.01 when compared with the BISEP Index). The predictive ability of the different indices evaluated was similar when compared with each other, and could be classified as moderate-good, the predictive accuracy was lower than that found in the original studies [5]. When we compare the mortality indices with the predictive capacity of the functional status measured by the Barthel Index, the results are similar.