Journal of Gerontology: MEDICAL SCIENCES 1992, Vol. 47, No. 2, M35-39
Copyright 1992 by The Gerontological Society of America
Predicting Mortality and Length of Stay of Geriatric Patients in an Acute Care General Hospital Raffaele Antonelli Incalzi, Antonella Gemma, Oliviero Capparella, Luciano Terranova, Pasquala Porcedda, Emilio Tresalti, and Pierugo Carbonin Department of Geriatrics, The Catholic University of the Sacred Heart, Rome.
T
HE prognosis of hospitalized geriatric patients has been the object of several studies aiming to identify predictors of death and/or long stay (1-9). Medical conditions (4,7-9), age in itself (1-7), and functional and mental status have been shown to have a major predictive role (1,3-7,9). These studies have been conducted in different clinical settings (1-9), so that results are comparable only to some extent. The most recent study has involved almost exclusively male patients admitted to a Veterans Administration teaching hospital having special facilities for geriatric patients (9). Thus, the studied population was poorly representative of the average geriatric population, and the type of care differed from that generally provided to the elderly in acute care hospitals. The present study adds to the knowledge of the role of medical and functional status, as reflected by easily measurable indicators, in predicting adverse outcomes of acute care hospitalization of a broad population of geriatric patients. It demonstrates that mortality is independently predicted by polypharmacy and abnormal functional and mental status, and extended stay by polypharmacy and polypathology.
Sociodemographic data: sex, age, living location, and living arrangement prior to admission. The patient's living arrangement was classified as living alone or with someone, because living with someone aside from a spouse has recently been shown to predict the receipt of informal support (10,11). Medical data: the primary diagnosis, i.e., the admitting diagnosis, the number of additional diagnoses resulting from the problem list 24-36 hours after admission, and the number of prescribed medications were collected from the hospital chart. Functional data: activities of daily living (ADLs) (12) were expressed by a simplified and previously validated score (13). Briefly, scores 1 to 3 identify, respectively, patients completely independent, dependent in 1 to 5 ADLs, and dependent in all ADLs. The cognitive status was explored by Mini-Mental State test (MMS) (14); patients untested because of impaired consciousness were considered to have abnormal MMS scores. The affective status was assessed by the Geriatric Depression Scale (15); three patients could not be tested because of impaired consciousness, and four because of advanced dementia.
METHODS
It must be observed that changes in functional/cognitive status prior to admission might be stronger prognostic indicators than the absolute functional/cognitive status at admission because they reflect the severity of the current illness. However, retrospectively assessing these changes might affect data reliability and completeness. Furthermore, any attempt to detect and quantify functional/cognitive status changes during the stay would require a continuous or, at least, a twostep assessment program that could not be recommended for routine use. These considerations led us to focus on the absolute functional/cognitive status at admission. The date of death or discharge was recorded, and discharged patients were followed up for one year; long-term
All patients over age 70 admitted to the University Hospital "A. Gemelli" from May 15 to June 14, 1988 were seen by a geriatrician and a geriatric nurse within 24 hours of admission. The eligibility to a given ward in this 1800-bed teaching hospital relied on the judgment of the emergency room head physician. After exclusion of patients admitted to Pediatrics, Obstetrics, and Intensive Care units, patients over 70 accounted for 18% of all the admissions in the study period. The remaining patients had a mean age of 43 ± 17 years. Each ward was headed by a senior specialist with one house officer for every 8-10 beds. A multidimensional assessment was performed by collecting the following data:
M35
Downloaded from http://geronj.oxfordjournals.org/ at Stockholms Universitet on August 31, 2015
Three-hundred-eight geriatric patients (mean age = 76.7 yr, range = 70-94 yr) consecutively admitted to an acute care general hospital were followed up to identify the predictors of in-hospital mortality and long stay. Sociodemographic, medical, and functional data were collected within 24 hours from admission and their correlation with the outcomes assessed by logistic regression analysis. The following variables were shown to be independent predictors of death: use of more than 6 drugs (odds ratio = 3.04, confidence limits = 1.05-8.76); abnormal Mini-Mental State score (o.r. = 1.72, c.I. = 1.05-1.83); low ADL score (o.r. = 2.4, c.l. = 1.07-5.56). Extended stay was significantly and independently predicted by polypharmacy (o.r. = 1.94, c.l. = 1.18-3.2) and comorbidity (o.r. = 2.06, c.l. — 1.24-3.38). The mortality rates of patients with cognitive impairment and polypharmacy with or without functional impairment were 40% and 22%, respectively. The proposed method allows identification of high-risk geriatric inpatients by a simple medical and functional assessment on admission.
M36
INCALZIETAL.
RESULTS
The characteristics of the 308 patients studied are reported in Tables 1 and 2. Females were 53.3% vs 62% of the Italian geriatric population (19) and vs 57% of the hospitalized fraction of that population (20). Most of the patients were living at home at time of admission, and 48% were widowed. Only 191 patients were completely independent, whereas abnormal cognitive status and depressed mood were diagnosed in 25% and 27% of the patients, respectively. Only 2.6% of the patients were admitted to the geriatrics ward; even if followed by an attending physician certified in Geriatrics, however, they did not benefit from special geriatric programs. Neoplastic and cardiovascular diseases together accounted for 38.3% of the primary diagnoses. Ophthalmic Table 1. Sociodemographic and Functional Characteristics of the Studied Population Number Males, % Age, years, mean ± SD (range) Widowed, % Living location prior to admission, % Home, with spouse and/or children Home, alone Nursing home Performance on Mini-Mental State examination, % Normal (3= 24) Abnormal (< 24) Performance on Geriatric Depression Scale, % Normal (5= 18) Abnormal (< 18) ADL score*, %
308 46.7 76.5 ± 5 (70-94) 48 76 21 3 75 25 71 29
1
62
2
26
3
12
*1 = independent in all ADLs; 2 = dependent in 1 to 5 ADLs; 3 = dependent in all ADLs.
diseases were the fifth most common primary diagnosis, given the availability of a 42-bed ward of Ophthalmology and the high prevalence of these diseases in the elderly. Diabetes was an uncommon primary diagnosis but the second most common additional disease, as most of the diabetic patients presented to the emergency ward because of acute cerebrovascular, cardiovascular, and renal complications. The mortality rate was 4.8% in specialty wards, 3.8% in surgical wards, and 9.7% in medical wards (Table 3). The last finding confirms results by Narain et al. (9). The average stay was 21.2 ± 18.7 days, whereas nongeriatric patients in the same period were hospitalized for 17.2 ± 16.3 days. Ninety percent of the patients were discharged to their homes, 6.8% to nursing homes, and 3.2% to geriatric rehabilitation hospitals. It must be observed that in Italy only 12 nursing home beds for 1000 subjects over age 65 are available, and there are virtually no community-based care programs (21). This could to some extent account for the relatively long stay, at least if compared to that of Medicare patients in the United States (22). Also, a recent survey of 46 acute care Italian hospitals regarding 3,435 geriatric patients demonstrates an average stay of 18 ± 13 days (20). Patients admitted to the geriatrics ward had a mortality rate of 12.5% and length of stay of 22 ± 5 days vs 9.5% (n.s.) and 19.9 ± 16 days (n.s.), respectively, in the remaining medical wards. The relationship of the baseline variables to the outcomes, as assessed by univariate analysis, is reported in Table 3: mortality was significantly correlated with age, cognitive impairment, physical disability, heavy drug use, and depression and was not predicted by other sociodemographic variables nor by the diagnostic categories. The relationship between neoplastic diseases and mortality is probably underestimated, since some of the neoplastic patients in the
Table 2. Clinical Characteristics of the Studied Population
Neoplastic Cardiovascular Surgical Neurological Ophthalmic Respiratory Orthopaedic Urologic Diabetes Other Number of additional diagnoses (mean ± SD) Number of prescribed medications (mean ± SD)
Admitting Diagnosis
Additional Diagnosis*
21 17.5 13.6 11 9.4 8.8 5.5 1.9
4 35 3 .8
5.9 3 .2 7 .2 3 .8 5 .1 16.6 15 .4
1.6 9.7
3 .4 :t 1.2 5 .9 :t 3.7
% admission to: Medical wards Surgical wards Specialty ward (Dermatology, Gynecology, Neurology, Ophthalmology) "Resulting from the problem list 24-36 hours after admission.
46 .7
26 27.3
Downloaded from http://geronj.oxfordjournals.org/ at Stockholms Universitet on August 31, 2015
prognostic indicators are the object of a previous study (13). Statistical analysis was performed by SPSS and BMDP statistical software. Data were primarily analyzed by descriptive statistics. The relationships between the collected baseline variables and the outcomes were singly assessed by univariate analysis. Each variable proven to have a significant correlation with the outcomes at the univariate analysis was entered into a logistic regression analysis (16) to correct its predictive role for interactions with the remaining variables. Odds ratio and 95% confidence intervals were calculated. When " 1 " was not included in the confidence intervals, the variable was considered significantly correlated with the outcome. The goodness of fit of the logistic model was assessed by the C. C. Brown goodness-of-fit test (17). This test was preferred to the goodness-of-fit chi square and to the "Hosmer" goodness-of-fit test because, in the present models, cell frequencies are small and most of the covariates are discrete. This limits reliability of both the former and the latter test (17,18). Once the logistic regression analysis had identified variables predicting mortality, eight groups of patients were assembled according to various combinations of such variables, and their respective mortality rates were computed.
M37
OUTCOMES OF GERIATRIC INPATIENTS
terminal stage of their illness chose to die at home and were thus discharged. Extended stay was directly related to polypharmacy and comorbidity. The difference in length of stay among medical diagnostic categories was almost significant (Kruskal-Wallis H = 11.5, p < .08). The respiratory diagnostic category had the shortest length of stay because of the substantial proportion of respiratory patients with acute bronchopneumonia (30%) promptly responsive to antibiotic therapy.
Table 3. Relation of the Baseline Variables to the Outcome Events, as Assessed by Univariate Analysis
Variables Age 3
1.94 2.06
1.18-3.2 1.24-3.38
C. C. Brown goodness-of-fitt 0.024 p = .988 Model A Model B 0.638 p = .727 *2 = dependent in 1 to 5 ADLs; 3 = dependent in all ADLs. ti.e., length of stay 3= 26 days. tN.d.R.: The larger the p value the higher is the possibility that the logistic model is the best fitting model (reference 17).
Downloaded from http://geronj.oxfordjournals.org/ at Stockholms Universitet on August 31, 2015
Mortality,
Results of logistic regression analysis are shown in Table 4: polypharmacy emerged as the main independent predictor of mortality preceding abnormal physical and mental status, while age and depressed mood lacked predictivity. It is likely that depression was, on average, reactive to the disease state and to physical dependence, which would explain its association with mortality on univariate testing (Table 3), but no more after correction for the interactions among variables (Table 4). The significant correlation between depression and physical dependence described elsewhere (23) supports this hypothesis. Similarly, the association between age and mortality in the univariate but not the multivariate model (Tables 3 and 4) is probably due to the slightly lower mean age of patients without risk factors (Figure 1); correcting this association for interactions among age and the remaining independent variables resulted in loss of predictivity by age (Table 4). Extended stay was independently predicted by polypharmacy and comorbidity (Table 4), which confirms results by univariate analysis (Table 3). The C. C. Brown goodness-of-fit test shows that both statistical models fit the data adequately (Table 4). Eight groups of patients were identified according to various combinations of risk factors for death (Figure 1). The highest mortality rate (40%) was found, as expected, in patients with mental and functional impairment who were consuming seven different drugs or more. The group characterized by polypharmacy and cognitive impairment had 23.1% mortality rate. Then, cognitive impairment, although lacking predictivity if isolated, helped define the two groups with the highest mortality rates. Only 4 out of 21 deaths occurred in the groups of patients with one risk factor. These data should be interpreted with some caution because of the uneven size of the groups. Accordingly, the mortality risk of patients with cognitive impairment, either isolated or coexisting with impaired physical function, is probably underestimated, given the small size of these groups. However, a well-defined profile of risk emerged for the extreme groups with clustering of 81% of death in the two highest risk
M38
INCALZIETAL.
No.
AGE y. ADL
80
75+4
I
MMS DRUGS T >24
24
Copyright 1992 by The Gerontological Society of America
Predicting Mortality and Length of Stay of Geriatric Patients in an Acute Care General Hospital Raffaele Antonelli Incalzi, Antonella Gemma, Oliviero Capparella, Luciano Terranova, Pasquala Porcedda, Emilio Tresalti, and Pierugo Carbonin Department of Geriatrics, The Catholic University of the Sacred Heart, Rome.
T
HE prognosis of hospitalized geriatric patients has been the object of several studies aiming to identify predictors of death and/or long stay (1-9). Medical conditions (4,7-9), age in itself (1-7), and functional and mental status have been shown to have a major predictive role (1,3-7,9). These studies have been conducted in different clinical settings (1-9), so that results are comparable only to some extent. The most recent study has involved almost exclusively male patients admitted to a Veterans Administration teaching hospital having special facilities for geriatric patients (9). Thus, the studied population was poorly representative of the average geriatric population, and the type of care differed from that generally provided to the elderly in acute care hospitals. The present study adds to the knowledge of the role of medical and functional status, as reflected by easily measurable indicators, in predicting adverse outcomes of acute care hospitalization of a broad population of geriatric patients. It demonstrates that mortality is independently predicted by polypharmacy and abnormal functional and mental status, and extended stay by polypharmacy and polypathology.
Sociodemographic data: sex, age, living location, and living arrangement prior to admission. The patient's living arrangement was classified as living alone or with someone, because living with someone aside from a spouse has recently been shown to predict the receipt of informal support (10,11). Medical data: the primary diagnosis, i.e., the admitting diagnosis, the number of additional diagnoses resulting from the problem list 24-36 hours after admission, and the number of prescribed medications were collected from the hospital chart. Functional data: activities of daily living (ADLs) (12) were expressed by a simplified and previously validated score (13). Briefly, scores 1 to 3 identify, respectively, patients completely independent, dependent in 1 to 5 ADLs, and dependent in all ADLs. The cognitive status was explored by Mini-Mental State test (MMS) (14); patients untested because of impaired consciousness were considered to have abnormal MMS scores. The affective status was assessed by the Geriatric Depression Scale (15); three patients could not be tested because of impaired consciousness, and four because of advanced dementia.
METHODS
It must be observed that changes in functional/cognitive status prior to admission might be stronger prognostic indicators than the absolute functional/cognitive status at admission because they reflect the severity of the current illness. However, retrospectively assessing these changes might affect data reliability and completeness. Furthermore, any attempt to detect and quantify functional/cognitive status changes during the stay would require a continuous or, at least, a twostep assessment program that could not be recommended for routine use. These considerations led us to focus on the absolute functional/cognitive status at admission. The date of death or discharge was recorded, and discharged patients were followed up for one year; long-term
All patients over age 70 admitted to the University Hospital "A. Gemelli" from May 15 to June 14, 1988 were seen by a geriatrician and a geriatric nurse within 24 hours of admission. The eligibility to a given ward in this 1800-bed teaching hospital relied on the judgment of the emergency room head physician. After exclusion of patients admitted to Pediatrics, Obstetrics, and Intensive Care units, patients over 70 accounted for 18% of all the admissions in the study period. The remaining patients had a mean age of 43 ± 17 years. Each ward was headed by a senior specialist with one house officer for every 8-10 beds. A multidimensional assessment was performed by collecting the following data:
M35
Downloaded from http://geronj.oxfordjournals.org/ at Stockholms Universitet on August 31, 2015
Three-hundred-eight geriatric patients (mean age = 76.7 yr, range = 70-94 yr) consecutively admitted to an acute care general hospital were followed up to identify the predictors of in-hospital mortality and long stay. Sociodemographic, medical, and functional data were collected within 24 hours from admission and their correlation with the outcomes assessed by logistic regression analysis. The following variables were shown to be independent predictors of death: use of more than 6 drugs (odds ratio = 3.04, confidence limits = 1.05-8.76); abnormal Mini-Mental State score (o.r. = 1.72, c.I. = 1.05-1.83); low ADL score (o.r. = 2.4, c.l. = 1.07-5.56). Extended stay was significantly and independently predicted by polypharmacy (o.r. = 1.94, c.l. = 1.18-3.2) and comorbidity (o.r. = 2.06, c.l. — 1.24-3.38). The mortality rates of patients with cognitive impairment and polypharmacy with or without functional impairment were 40% and 22%, respectively. The proposed method allows identification of high-risk geriatric inpatients by a simple medical and functional assessment on admission.
M36
INCALZIETAL.
RESULTS
The characteristics of the 308 patients studied are reported in Tables 1 and 2. Females were 53.3% vs 62% of the Italian geriatric population (19) and vs 57% of the hospitalized fraction of that population (20). Most of the patients were living at home at time of admission, and 48% were widowed. Only 191 patients were completely independent, whereas abnormal cognitive status and depressed mood were diagnosed in 25% and 27% of the patients, respectively. Only 2.6% of the patients were admitted to the geriatrics ward; even if followed by an attending physician certified in Geriatrics, however, they did not benefit from special geriatric programs. Neoplastic and cardiovascular diseases together accounted for 38.3% of the primary diagnoses. Ophthalmic Table 1. Sociodemographic and Functional Characteristics of the Studied Population Number Males, % Age, years, mean ± SD (range) Widowed, % Living location prior to admission, % Home, with spouse and/or children Home, alone Nursing home Performance on Mini-Mental State examination, % Normal (3= 24) Abnormal (< 24) Performance on Geriatric Depression Scale, % Normal (5= 18) Abnormal (< 18) ADL score*, %
308 46.7 76.5 ± 5 (70-94) 48 76 21 3 75 25 71 29
1
62
2
26
3
12
*1 = independent in all ADLs; 2 = dependent in 1 to 5 ADLs; 3 = dependent in all ADLs.
diseases were the fifth most common primary diagnosis, given the availability of a 42-bed ward of Ophthalmology and the high prevalence of these diseases in the elderly. Diabetes was an uncommon primary diagnosis but the second most common additional disease, as most of the diabetic patients presented to the emergency ward because of acute cerebrovascular, cardiovascular, and renal complications. The mortality rate was 4.8% in specialty wards, 3.8% in surgical wards, and 9.7% in medical wards (Table 3). The last finding confirms results by Narain et al. (9). The average stay was 21.2 ± 18.7 days, whereas nongeriatric patients in the same period were hospitalized for 17.2 ± 16.3 days. Ninety percent of the patients were discharged to their homes, 6.8% to nursing homes, and 3.2% to geriatric rehabilitation hospitals. It must be observed that in Italy only 12 nursing home beds for 1000 subjects over age 65 are available, and there are virtually no community-based care programs (21). This could to some extent account for the relatively long stay, at least if compared to that of Medicare patients in the United States (22). Also, a recent survey of 46 acute care Italian hospitals regarding 3,435 geriatric patients demonstrates an average stay of 18 ± 13 days (20). Patients admitted to the geriatrics ward had a mortality rate of 12.5% and length of stay of 22 ± 5 days vs 9.5% (n.s.) and 19.9 ± 16 days (n.s.), respectively, in the remaining medical wards. The relationship of the baseline variables to the outcomes, as assessed by univariate analysis, is reported in Table 3: mortality was significantly correlated with age, cognitive impairment, physical disability, heavy drug use, and depression and was not predicted by other sociodemographic variables nor by the diagnostic categories. The relationship between neoplastic diseases and mortality is probably underestimated, since some of the neoplastic patients in the
Table 2. Clinical Characteristics of the Studied Population
Neoplastic Cardiovascular Surgical Neurological Ophthalmic Respiratory Orthopaedic Urologic Diabetes Other Number of additional diagnoses (mean ± SD) Number of prescribed medications (mean ± SD)
Admitting Diagnosis
Additional Diagnosis*
21 17.5 13.6 11 9.4 8.8 5.5 1.9
4 35 3 .8
5.9 3 .2 7 .2 3 .8 5 .1 16.6 15 .4
1.6 9.7
3 .4 :t 1.2 5 .9 :t 3.7
% admission to: Medical wards Surgical wards Specialty ward (Dermatology, Gynecology, Neurology, Ophthalmology) "Resulting from the problem list 24-36 hours after admission.
46 .7
26 27.3
Downloaded from http://geronj.oxfordjournals.org/ at Stockholms Universitet on August 31, 2015
prognostic indicators are the object of a previous study (13). Statistical analysis was performed by SPSS and BMDP statistical software. Data were primarily analyzed by descriptive statistics. The relationships between the collected baseline variables and the outcomes were singly assessed by univariate analysis. Each variable proven to have a significant correlation with the outcomes at the univariate analysis was entered into a logistic regression analysis (16) to correct its predictive role for interactions with the remaining variables. Odds ratio and 95% confidence intervals were calculated. When " 1 " was not included in the confidence intervals, the variable was considered significantly correlated with the outcome. The goodness of fit of the logistic model was assessed by the C. C. Brown goodness-of-fit test (17). This test was preferred to the goodness-of-fit chi square and to the "Hosmer" goodness-of-fit test because, in the present models, cell frequencies are small and most of the covariates are discrete. This limits reliability of both the former and the latter test (17,18). Once the logistic regression analysis had identified variables predicting mortality, eight groups of patients were assembled according to various combinations of such variables, and their respective mortality rates were computed.
M37
OUTCOMES OF GERIATRIC INPATIENTS
terminal stage of their illness chose to die at home and were thus discharged. Extended stay was directly related to polypharmacy and comorbidity. The difference in length of stay among medical diagnostic categories was almost significant (Kruskal-Wallis H = 11.5, p < .08). The respiratory diagnostic category had the shortest length of stay because of the substantial proportion of respiratory patients with acute bronchopneumonia (30%) promptly responsive to antibiotic therapy.
Table 3. Relation of the Baseline Variables to the Outcome Events, as Assessed by Univariate Analysis
Variables Age 3
1.94 2.06
1.18-3.2 1.24-3.38
C. C. Brown goodness-of-fitt 0.024 p = .988 Model A Model B 0.638 p = .727 *2 = dependent in 1 to 5 ADLs; 3 = dependent in all ADLs. ti.e., length of stay 3= 26 days. tN.d.R.: The larger the p value the higher is the possibility that the logistic model is the best fitting model (reference 17).
Downloaded from http://geronj.oxfordjournals.org/ at Stockholms Universitet on August 31, 2015
Mortality,
Results of logistic regression analysis are shown in Table 4: polypharmacy emerged as the main independent predictor of mortality preceding abnormal physical and mental status, while age and depressed mood lacked predictivity. It is likely that depression was, on average, reactive to the disease state and to physical dependence, which would explain its association with mortality on univariate testing (Table 3), but no more after correction for the interactions among variables (Table 4). The significant correlation between depression and physical dependence described elsewhere (23) supports this hypothesis. Similarly, the association between age and mortality in the univariate but not the multivariate model (Tables 3 and 4) is probably due to the slightly lower mean age of patients without risk factors (Figure 1); correcting this association for interactions among age and the remaining independent variables resulted in loss of predictivity by age (Table 4). Extended stay was independently predicted by polypharmacy and comorbidity (Table 4), which confirms results by univariate analysis (Table 3). The C. C. Brown goodness-of-fit test shows that both statistical models fit the data adequately (Table 4). Eight groups of patients were identified according to various combinations of risk factors for death (Figure 1). The highest mortality rate (40%) was found, as expected, in patients with mental and functional impairment who were consuming seven different drugs or more. The group characterized by polypharmacy and cognitive impairment had 23.1% mortality rate. Then, cognitive impairment, although lacking predictivity if isolated, helped define the two groups with the highest mortality rates. Only 4 out of 21 deaths occurred in the groups of patients with one risk factor. These data should be interpreted with some caution because of the uneven size of the groups. Accordingly, the mortality risk of patients with cognitive impairment, either isolated or coexisting with impaired physical function, is probably underestimated, given the small size of these groups. However, a well-defined profile of risk emerged for the extreme groups with clustering of 81% of death in the two highest risk
M38
INCALZIETAL.
No.
AGE y. ADL
80
75+4
I
MMS DRUGS T >24
24
