ORIGINAL PAPER
Predicting mortality of elderly patients acutely admitted to the Department of Internal Medicine B. Smolin,1 Y. Levy,1 E. Sabbach-Cohen,1 L. Levi,2 T. Mashiach3
1
SUMMARY
What’s known
Aims: This study addresses the common practice of providing aggressive treatments of limited clinical benefit and cost-effectiveness to seriously ill and frail elderly. We have created a statistical model of 6-month mortality risk prediction following acute hospitalisation admission, and identified a subset of patients with poorest prognosis that requires comfort-focused care. Methods: We have studied electronic medical records of 26,937 patients age 65 years or older, hospitalised in the internal medicine departments of one tertiary-care teaching medical center in Northern Israel from January 1, 2008 through December 31, 2011 and mortality data from the Israeli Internal Ministry Registry. Norton score records were employed for the performance status evaluation. Multivariate logistic regression analysis was used to predict the risk of 6-month mortality. Results: Variables associated with an increased risk of 6-month mortality included: metastatic cancer, age above 85 years, decreased values of blood albumin and haemoglobin, increased blood urea nitrogen and decreased physical/mental status and activity. The receiver operating characteristic area for the predicted probability of death was 0.845 and 0.847 in external validation cohort. Using predictive values of the logistic regression analysis, the study cohort was stratified into six groups with various predictive mortality risks. Conclusion: The majority of deaths that have occurred within 6 months following the acute hospitalisation could be predicted on patient admission based on a few simple and easily obtained parameters. Earlier recognition of patients nearing the end of their lives may lead to better care and more efficient use of available resource.
Introduction Acute care hospitals in Israel admit many elderly adults who are seriously ill and frail. For many of them, an acute medical event or any exacerbation of advanced chronic disease may be followed by a progressive decline and death (1). Given that the focus of modern hospital care is diagnosis, treatment and cure, enormous resources are allocated for continuing standard problem and cure oriented care for terminally ill patients. Often, dying patients are exposed to aggressive treatment modalities and assessments of limited clinical benefit and cost-effectiveness. Evidence suggests that physicians perform poorly in predicting when patients will die (2,3). Consequently, the absence of methods of mortality prediction in the elderly people and frail is a major impediment to appropriate care (4) as early recognition of patients at the end of their lives allows for ª 2014 John Wiley & Sons Ltd Int J Clin Pract doi: 10.1111/ijcp.12564
Physicians perform poorly in prediction of grave prognosis of seriously ill and frail patients admitted to departments of internal medicine. Consequently, inappropriate care is commonly provided to patients who may require comfort-focused care.
What’s new This study demonstrates an easily obtained 6-month mortality prediction model based on short history, Norton Scoring System and routine biochemical tests. Data are collected on the very first day of admission. By incorporation of this model to admission records, improved decision making, efficient use of available resources and personalised care are anticipated.
Department of Internal Medicince D, Technion Faculty of Medicine, Rambam Health Care Campus, Haifa, Israel 2 Hospital Management, Technion Faculty of Medicine, Rambam Health Care Campus, Haifa, Israel 3 Biostatistics Unit, Technion Faculty of Medicine, Rambam Health Care Campus, Haifa, Israel Correspondence to: Yishai Levy, MD, Department of Internal Medicine D, Rambam Health Care Campus, Haifa, Bat Galim 31096, Israel Tel.: + 972-4-8542263 Fax: + 972-4-8543286 Email: ys_levy@rambam. health.gov.il
Disclosures None.
better planning of care and more efficient use of limited resources. Despite the extent of the problem, only a few studies have been dedicated to predictors of mortality of hospitalised elderly patients (5). Most of these studies have investigated diagnosis-related outcomes or focused on intensive care unit patients (6–8). Other selected populations such as those in nursing homes have been also studied. While physical competence and cognitive status were consistently found to correlate highly with mortality in studies of older medical patients (9–12), internists and general practitioners do not generally convert the physical and mental performance status of their patients into measurable variables. Thus, further standardisation of their estimates is impossible. Therefore, methods incorporating key descriptors of function and cognition into routine clinical practice should be actively pursued (13).
1
2
Predicting mortality of elderly patients
In this report, we suggest that the Norton scale score is a good and simple indicator of physical and mental performance status of elderly patients admitted to internal medicine departments. At our medical center, the Norton scoring system (NSS) has been routinely used by the nursing staff since 2008 for pressure sore risk evaluation. The NSS was created in England in 1962 for pressure sore risk evaluation and its ease of use makes it still widely used today. NSS includes five domains: physical, mental, activity level, mobility and continence. Rating of each domain varies from 1 to 4 points, indicating good, fair, poor and very bad estimates, respectively. Recently, data from another Israeli hospital has demonstrated that admission Norton scores correlated with postoperative complications of hip arthroplasty and all-cause rehabilitation outcomes in elderly patients (14–16). Hence, admission Norton score together with other more traditional mortality predictors from medical and laboratory records may be useful in stratifying older hospitalised patients into appropriate risk categories. Such a prognostic model could help to provide improved personalised ‘tailor made’ care for elderly patients with varied outcomes and prognoses.
Methods Data source Demographical variables (age and gender), clinical parameters such as Norton score and Charlson index and various laboratory values on admission or during the first 3 days of hospitalisation were collected from Rambam Medical Center electronic registry (BO). Mortality records were obtained from the Israeli Ministry of Interior registry.
Study population We identified 26,939 admissions of patients aged 65 or older, in five departments of internal medicine at Rambam Medical Center between January 1, 2008 and December 31, 2011. For patients with multiple admissions, only the first hospitalisation for each 182-day period was included, therefore 7682 (27%) recurrent admissions were excluded. Finally, we excluded 533 patients with missing Norton score data (2%). Thus, the final sample comprised 19,257 patients 65 years or older admitted directly from the emergency room into departments of internal medicine.
Statistical analysis Bivariate logistic regression was used for the calculation of the odds ratios (OR) with 95% confidence
intervals (CI) and p values for assessment of the association between patient characteristics and 6month mortality risk. Variables were selected as candidates for the multivariate analysis on the basis of the level of significance of the bivariate association with 6-month mortality (p < 0.05). Multivariable Forward Stepwise Logistic Regression analysis was performed. The Hosmer–Lemeshow goodness-of-fit statistic was calculated. The area under the receiver operating characteristic (ROC) curve was used as a measure of model discrimination. The statistical model was internally validated using the technique of bootstrap resampling. Resampling methods (using 200 samples) were used to calculate beta coefficient and confidence intervals of estimates. External validation was performed based on admissions of patients aged 65 years or older into departments of internal medicine in Rambam Medical Center between January 1, 2012 and December 31, 2012. The patients were selected by the same criteria as the derivation set. Using predictive values of the logistic regression analysis, the study cohort was stratified into six groups with various mortality risks. Two-tailed p values of 0.05 or less were considered statistically significant. All statistical analyses were performed using SPSS (Statistics Products Solutions Services “SPSS” 18 for Windows, “SPSS” Inc., Chicago, Illinois, USA) 18.0 for Windows.
Results Patient characteristics Fifty-two per cent of the patients were females, 23% were 85 years and older, but only 2% of patients were older than 95 years. Seventy eight per cent of the admissions included in the study were first time hospitalisations of patients during the period of 2008–2011; 12% were the first hospitalisation in the second set of admissions for the same patient and only 11% of included admissions represented the third or more hospitalisations during the 4-year period studied. While 76% of the patients were mentally intact on admission, only 21% were completely physically preserved. According to Charlson’s comorbidity index, 16.8% of patients included in the study had a very high comorbidity burden of diseases (index > 5+), while 9.7% had no chronic illnesses on admission. All patients received standard care according to the medical working diagnosis. The overall in-hospital, 90-day and 6-month mortality rates were 9.5%, 20%, 25.9%, respectively and did not change significantly during the 2008–2011 study period.
ª 2014 John Wiley & Sons Ltd Int J Clin Pract
Predicting mortality of elderly patients
Table 1 Bivariable associations of risk factors with 6-month mortality in the derivation cohort
Risk factor
Gender Age
Physical
Mental
Activity
Mobility
Incontinence
BUN BUN BUN BUN BUN CREA
HGB
GLB
ALB
Total F M 65–74 75–85 85–95 ≥ 95 4 3 2 1 4 3 2 1 4 3 2 1 4 3 2 1 4 3 2 1 ≤ 30 30.01–40 40.01–60 > 60 Missing 0.5–1.29 0.1–0.5 1.3–1.49 1.5–1.9 2–2.99 ≥3 Missing 12+ 10–11 9–10
Predicting mortality of elderly patients acutely admitted to the Department of Internal Medicine B. Smolin,1 Y. Levy,1 E. Sabbach-Cohen,1 L. Levi,2 T. Mashiach3
1
SUMMARY
What’s known
Aims: This study addresses the common practice of providing aggressive treatments of limited clinical benefit and cost-effectiveness to seriously ill and frail elderly. We have created a statistical model of 6-month mortality risk prediction following acute hospitalisation admission, and identified a subset of patients with poorest prognosis that requires comfort-focused care. Methods: We have studied electronic medical records of 26,937 patients age 65 years or older, hospitalised in the internal medicine departments of one tertiary-care teaching medical center in Northern Israel from January 1, 2008 through December 31, 2011 and mortality data from the Israeli Internal Ministry Registry. Norton score records were employed for the performance status evaluation. Multivariate logistic regression analysis was used to predict the risk of 6-month mortality. Results: Variables associated with an increased risk of 6-month mortality included: metastatic cancer, age above 85 years, decreased values of blood albumin and haemoglobin, increased blood urea nitrogen and decreased physical/mental status and activity. The receiver operating characteristic area for the predicted probability of death was 0.845 and 0.847 in external validation cohort. Using predictive values of the logistic regression analysis, the study cohort was stratified into six groups with various predictive mortality risks. Conclusion: The majority of deaths that have occurred within 6 months following the acute hospitalisation could be predicted on patient admission based on a few simple and easily obtained parameters. Earlier recognition of patients nearing the end of their lives may lead to better care and more efficient use of available resource.
Introduction Acute care hospitals in Israel admit many elderly adults who are seriously ill and frail. For many of them, an acute medical event or any exacerbation of advanced chronic disease may be followed by a progressive decline and death (1). Given that the focus of modern hospital care is diagnosis, treatment and cure, enormous resources are allocated for continuing standard problem and cure oriented care for terminally ill patients. Often, dying patients are exposed to aggressive treatment modalities and assessments of limited clinical benefit and cost-effectiveness. Evidence suggests that physicians perform poorly in predicting when patients will die (2,3). Consequently, the absence of methods of mortality prediction in the elderly people and frail is a major impediment to appropriate care (4) as early recognition of patients at the end of their lives allows for ª 2014 John Wiley & Sons Ltd Int J Clin Pract doi: 10.1111/ijcp.12564
Physicians perform poorly in prediction of grave prognosis of seriously ill and frail patients admitted to departments of internal medicine. Consequently, inappropriate care is commonly provided to patients who may require comfort-focused care.
What’s new This study demonstrates an easily obtained 6-month mortality prediction model based on short history, Norton Scoring System and routine biochemical tests. Data are collected on the very first day of admission. By incorporation of this model to admission records, improved decision making, efficient use of available resources and personalised care are anticipated.
Department of Internal Medicince D, Technion Faculty of Medicine, Rambam Health Care Campus, Haifa, Israel 2 Hospital Management, Technion Faculty of Medicine, Rambam Health Care Campus, Haifa, Israel 3 Biostatistics Unit, Technion Faculty of Medicine, Rambam Health Care Campus, Haifa, Israel Correspondence to: Yishai Levy, MD, Department of Internal Medicine D, Rambam Health Care Campus, Haifa, Bat Galim 31096, Israel Tel.: + 972-4-8542263 Fax: + 972-4-8543286 Email: ys_levy@rambam. health.gov.il
Disclosures None.
better planning of care and more efficient use of limited resources. Despite the extent of the problem, only a few studies have been dedicated to predictors of mortality of hospitalised elderly patients (5). Most of these studies have investigated diagnosis-related outcomes or focused on intensive care unit patients (6–8). Other selected populations such as those in nursing homes have been also studied. While physical competence and cognitive status were consistently found to correlate highly with mortality in studies of older medical patients (9–12), internists and general practitioners do not generally convert the physical and mental performance status of their patients into measurable variables. Thus, further standardisation of their estimates is impossible. Therefore, methods incorporating key descriptors of function and cognition into routine clinical practice should be actively pursued (13).
1
2
Predicting mortality of elderly patients
In this report, we suggest that the Norton scale score is a good and simple indicator of physical and mental performance status of elderly patients admitted to internal medicine departments. At our medical center, the Norton scoring system (NSS) has been routinely used by the nursing staff since 2008 for pressure sore risk evaluation. The NSS was created in England in 1962 for pressure sore risk evaluation and its ease of use makes it still widely used today. NSS includes five domains: physical, mental, activity level, mobility and continence. Rating of each domain varies from 1 to 4 points, indicating good, fair, poor and very bad estimates, respectively. Recently, data from another Israeli hospital has demonstrated that admission Norton scores correlated with postoperative complications of hip arthroplasty and all-cause rehabilitation outcomes in elderly patients (14–16). Hence, admission Norton score together with other more traditional mortality predictors from medical and laboratory records may be useful in stratifying older hospitalised patients into appropriate risk categories. Such a prognostic model could help to provide improved personalised ‘tailor made’ care for elderly patients with varied outcomes and prognoses.
Methods Data source Demographical variables (age and gender), clinical parameters such as Norton score and Charlson index and various laboratory values on admission or during the first 3 days of hospitalisation were collected from Rambam Medical Center electronic registry (BO). Mortality records were obtained from the Israeli Ministry of Interior registry.
Study population We identified 26,939 admissions of patients aged 65 or older, in five departments of internal medicine at Rambam Medical Center between January 1, 2008 and December 31, 2011. For patients with multiple admissions, only the first hospitalisation for each 182-day period was included, therefore 7682 (27%) recurrent admissions were excluded. Finally, we excluded 533 patients with missing Norton score data (2%). Thus, the final sample comprised 19,257 patients 65 years or older admitted directly from the emergency room into departments of internal medicine.
Statistical analysis Bivariate logistic regression was used for the calculation of the odds ratios (OR) with 95% confidence
intervals (CI) and p values for assessment of the association between patient characteristics and 6month mortality risk. Variables were selected as candidates for the multivariate analysis on the basis of the level of significance of the bivariate association with 6-month mortality (p < 0.05). Multivariable Forward Stepwise Logistic Regression analysis was performed. The Hosmer–Lemeshow goodness-of-fit statistic was calculated. The area under the receiver operating characteristic (ROC) curve was used as a measure of model discrimination. The statistical model was internally validated using the technique of bootstrap resampling. Resampling methods (using 200 samples) were used to calculate beta coefficient and confidence intervals of estimates. External validation was performed based on admissions of patients aged 65 years or older into departments of internal medicine in Rambam Medical Center between January 1, 2012 and December 31, 2012. The patients were selected by the same criteria as the derivation set. Using predictive values of the logistic regression analysis, the study cohort was stratified into six groups with various mortality risks. Two-tailed p values of 0.05 or less were considered statistically significant. All statistical analyses were performed using SPSS (Statistics Products Solutions Services “SPSS” 18 for Windows, “SPSS” Inc., Chicago, Illinois, USA) 18.0 for Windows.
Results Patient characteristics Fifty-two per cent of the patients were females, 23% were 85 years and older, but only 2% of patients were older than 95 years. Seventy eight per cent of the admissions included in the study were first time hospitalisations of patients during the period of 2008–2011; 12% were the first hospitalisation in the second set of admissions for the same patient and only 11% of included admissions represented the third or more hospitalisations during the 4-year period studied. While 76% of the patients were mentally intact on admission, only 21% were completely physically preserved. According to Charlson’s comorbidity index, 16.8% of patients included in the study had a very high comorbidity burden of diseases (index > 5+), while 9.7% had no chronic illnesses on admission. All patients received standard care according to the medical working diagnosis. The overall in-hospital, 90-day and 6-month mortality rates were 9.5%, 20%, 25.9%, respectively and did not change significantly during the 2008–2011 study period.
ª 2014 John Wiley & Sons Ltd Int J Clin Pract
Predicting mortality of elderly patients
Table 1 Bivariable associations of risk factors with 6-month mortality in the derivation cohort
Risk factor
Gender Age
Physical
Mental
Activity
Mobility
Incontinence
BUN BUN BUN BUN BUN CREA
HGB
GLB
ALB
Total F M 65–74 75–85 85–95 ≥ 95 4 3 2 1 4 3 2 1 4 3 2 1 4 3 2 1 4 3 2 1 ≤ 30 30.01–40 40.01–60 > 60 Missing 0.5–1.29 0.1–0.5 1.3–1.49 1.5–1.9 2–2.99 ≥3 Missing 12+ 10–11 9–10
