Intern Emerg Med DOI 10.1007/s11739-015-1196-6
IM - ORIGINAL
A prognostic index for 1-year mortality can also predict in-hospital mortality of elderly medical patients Marco Cei • Nicola Mumoli • Jose´ Vitale Francesco Dentali
•
Received: 13 November 2014 / Accepted: 13 January 2015 Ó SIMI 2015
Abstract Elderly patients admitted to the hospital are at increased risk for both in-hospital and post-discharge mortality. Risk assessment models (RAMs) for in-hospital mortality are based mainly on physiological variables and a few laboratory data, whereas RAMs for late mortality usually include other domains such as disability and comorbidities. We aim to evaluate if a previous validated model for 1-year mortality (the Walter Score) would also work well in predicting in-hospital mortality. We retrospectively revised the medical records of patients admitted on our ward, from April to December, 2013. Data regarding gender, activities of daily living (ADLs), comorbidities, and routine laboratory tests were used to calculate a Modified Walter Score (MoWS). The main outcome measure was all cause, in-hospital mortality. The analysis involved 1,004 patients. Of these, 888 were discharged alive, and 116 (11.5 %) died during the hospitalization. The mean MoWS was 4.9 (±3.6) in the whole sample. Stratification into risk classes parallels with in-hospital mortality (Chi square for trend p \ 0.001). When dichotomized, MoWS has a sensitivity of 97.4 % (95 % CI 92.1–99.3), and a specificity of 48.2 % (95 % CI 44.9–51.5) with a good prognostic accuracy (area under the ROC = 0.81; 95 % CI 0.78, 0.84). Subgroup analysis according to different age groups gives similar results. A simple RAM based on
M. Cei N. Mumoli (&) Department of Internal Medicine, Ospedale Civile di Livorno, Viale Alfieri, 36, 57124 Livorno, Italy e-mail: [email protected] J. Vitale F. Dentali Department of Clinical Medicine, Insubria University, Varese, Italy
multiple domains, previously validated for predicting mortality of older adults within 1 year from the index hospitalization, can be useful at the moment of admission to Internal Medicine wards to accurately identify patients at low risk of in-hospital mortality. Keywords Elderly Medical admission Mortality Risk assessment model
Introduction Patients admitted to Internal Medicine units have a real risk for both in-hospital and post-discharge mortality. Several prognostic models have been developed to help clinicians predict adverse outcomes during hospitalization and after a long-term follow-up in this population [1]. In 2001, Walter and coworkers [2] derived and validated a simple score to predict the 1-year mortality of older adults (C70 years old) after hospitalization. After a multivariate analysis, they were able to identify a list of six independent predictors of long-term mortality commonly collected during hospitalization in most Internal Medicine units, making their score easily usable. These include— male gender, number of dependencies in activities of daily living (ADLs), the presence of congestive heart failure (CHF) or cancer; serum creatinine level higher than 3.0 mg/dL, and low albumin levels. To date, no study has evaluated the accuracy of this score in predicting in-hospital mortality in patients admitted to internal medicine units. Furthermore, the score has not been tested in younger patients. Thus, we planned to address this knowledge gap evaluating its accuracy in a large group of consecutive patients admitted to our internal medicine unit.
123
Intern Emerg Med
Methods Patient’s eligibility and data collection Medical records of all patients admitted to the Internal Medicine unit of Livorno Hospital in Italy from 1st April to 31st December 2013 were evaluated. To increase the external validity of the study, no exclusion criteria were planned, and all patients were potentially eligible for the study. However, patients presented more than once during the study period, and patients who were hospitalized for less than 24 h were excluded. The setting was a 43-bed medical unit of a non-teaching Italian hospital that receives unselected medical patients, admitted for acute medical conditions coming mainly from the Emergency Department (ED), either directly or after a brief stay for clinical stabilization. At the time of admission to Internal Medicine, information on patients’ baseline characteristics (age and gender), and on all the other items of the Walter Score (number of dependencies in ADLs, CHF, cancer; serum creatinine and albumin levels), was collected. As a difference from the original score, for the study purpose, the ADLs were considered on the day of admission and not at the time of discharge. This information was collected using a standardized form in all the included patients. Specific thresholds for each item of the Walter score are summarized in Table 1. Thus, a Modified Walter Score (MoWS) was calculated. For each patient, information on the in-hospital mortality was registered. The Institutional Review Board approved the study, which was carried out and is reported according to the Strengthening the Reporting of observational Studies in Epidemiology (STROBE) guidelines for observational studies [3]. Statistical analysis Continuous variables were expressed as mean plus or minus the standard deviation (SD) or as median with Table 1 The Walter score Risk factor
Points
Male gender
1
Dependency in 1–4 ADLs
2
Dependency in all 5 ADLsa
5
Congestive heart failure
2
Solitary or hematologic cancer
3
Metastatic cancer
8
Creatinine [3.0 mg/dL
2
Albumin 3.0–3.4 g/dL
1
Albumin \3.0 g/dL
2
a
ADLs denotes activities of daily living
123
minimum and maximum values when data did not have a normal distribution; categorical data are given as counts and percentages. Initially, according to the results of the original study [2], patients were divided into four risk classes (class I: 0–1 point, class II: 2–3 points, class III: 4–6 points, and class IV: [6 points), and class-specific rate of in-hospital mortality with the corresponding 95 % confidence interval (CI) was calculated. Subsequently, to explore the role of the score in predicting in-hospital mortality, patients were divided into two risk classes (low risk: MoWS \ 4 and high risk: MoWS C 4), and sensitivity, specificity, positive and negative predictive values and likelihood ratios for low vs. high-risk patients were estimated. Finally, the area under the receiver operating characteristic (ROC) curve was measured [4]. Furthermore, we assessed the accuracy of the MoWS in the subgroup of patients younger than 70 years old, and as well in the subgroup of patients of 70 years or older. We planned our sample size to find a difference in the mortality rate in the low and in high-risk groups. Hypothesizing a mean in-hospital mortality rate of about 8–10 %, (5–6 % in the low-risk group vs 11–12 % in the high-risk group), and considering one-third of the whole population at low risk and two-thirds at high risk, at least 850–900 patients should be enrolled to have a power of 90 % to find significant differences (p \ 0.05) between the groups. To be more conservative, the planned sample was incrementally increased to a total of 950–1,000 patients. All analyses were performed using SPSS 19.0 (SPSS Inc, Chicago, IL, USA).
Results A total of 1,055 patients were admitted between April 1 and December 31, 2013. Of these, 51 were excluded from the analysis (41 discharged within 24 h and 10 readmitted for the second time during the study period), leaving 1,004 patients for analysis. Data on the items of the score were complete for the whole sample. The median age was 79.8 ± 12.2 years, ranging from 19 to 102 years; 545 (54.3 %) patients were female. The mean MoWS was 4.9 (± 3.6) in the whole sample. Eight hundred and eightyeight patients were discharged alive, and 116 died during the hospitalization for a mean mortality rate of 11.6 % (95 % CI 9.7, 13.7 %), Table 2. The distribution of patients across the different MoWS risk classes is shown in Table 3 and Fig. 1. Stratification into risk classes significantly correlates with the in-hospital mortality rate (Chi square for trend p \ 0.001). In-hospital mortality is significantly lower in the low-risk group (MoWS \ 4: 0.7 %, 95 % CI 0.2, 2.2 %) compared to the high-risk group (MoWS C 4: 19.7 %, 95 % CI 16.6, 23.3 %) (p \ 0.001). When
1.00
Intern Emerg Med
550 (54.8 %)
Mean age, years (SD)
79.8 (12.2)
Mean duration of hospitalization, days (SD)
9.35 (0.4)
Number of dependencies in ADL (n)
0 (306) 1 (6)
0.25
2 (35)
0.50
1,004
Sensitivity
Patients, n Female gender, n (%)
0.75
Table 2 Demographics of the whole sample
3 (52)
0.00
4 (170) 5 (435) Heart failure, n (%)
214 (21.3)
Solitary or hematologic cancer, n (%)
71 (7.1)
Metastatic cancer
73 (7.3)
Creatinine [3.0 mg/dL, n (%)
37 (3.7)
Albumin 3.0–3.4 g/dL, n (%) Albumin \3.0 g/dL, n (%)
326 (32.4) 405 (40.3)
Discharged, n (%)
888 (88.4)
Deceased, n (%)
116 (11.6)
Median time to death, days (range)
6 (1–47)
Table 3 Distribution of patients across the MoWS categories MoWS
Discharged
0–1
217
2–3
211
2
2.06 (0.18–22.8)
0.98
4–6
188
11
12.7 (1.62–99.25)
\0.002
7–20
272
102
81.38 (11.26–587.9)
\0.001
a
Died
Odd ratio for mortality (95 % CI)a
p
1
As compared to patients with MoWS 0–1
0.00
0.25
0.50
0.75
1.00
1 - Specificity Area under ROC curve = 0.8077
Fig. 2 Receiving operating curve evaluating the discriminatory power of the MoWS to predict in-hospital mortality
97.8, 99.8), and a positive predictive value of 19.7 % (95 % CI 16.6, 23.3) for in-hospital mortality, with a negative likelihood ratio of 0.05 (95 % CI 0.02, 0.16), and a positive likelihood ratio of 1.88 (95 % CI 1.75, 2.01) (Table 3). The discriminatory power of the MoWS to predict in-hospital mortality, expressed as the area under the ROC curve was 0.81 (95 % CI 0.78, 0.84; Fig. 2). One hundred and fifty-seven patients (15.5 %) were younger than 70 years. In this subgroup of patients, the MoWS score has a sensitivity of 100.0 % (95 % CI 65.5, 100), and a specificity of 71.2 % (95 % CI 63.1, 78.3). Eight hundred and forty-eight patients (84.5 %) were C70 years. In this subgroup of patients, the MoWS score has a sensitivity of 97.2 % (95 % CI 91.3, 99.3), and a specificity of 43.7 % (95 % CI 40.1, 47.3).
Discussion
Fig. 1 Distribution of patients across the MoWS categories. Blue, alive; green, dead
dichotomized, the MoWS score has a sensitivity of 97.4 % (95 % CI 92.1, 99.3), and a specificity of 48.2 % (95 % CI 44.9, 51.5), a negative predictive value of 99.3 % (95 % CI
Medical patients have a real risk of death during and after hospital admission. Assessment of this risk at the time of admission should result in numerous benefits. Patients could be better allocated to prevent clinical deterioration, and those at higher risk of death could be carefully evaluated before and after discharge. Walter and coworkers have demonstrated, using a simple score, that 1-year mortality is associated with risk factors belonging to multiple domains such as demographic variables, comorbidities, functional status and laboratory values. With the results of our study, we suggested that a slightly modified version of this score can easily predict the risk of death during hospitalization in a real-world setting. In particular, the score appears to have a very high sensitivity, and more than forty percent of patients may be considered at very low risk of in-hospital mortality. The
123
Intern Emerg Med
validity of these findings is strengthened by the results of subgroup analyses according to different ages. In particular, this score seems to perform even better in younger patients, with no false negative results, and with a higher sensitivity compared to the whole population. However, these latter results should be interpreted with caution due to the limited sample of the subgroup. Many risk assessment models (RAMs) have been developed to assess the short and long-term mortality risk of hospitalized patients. In particular, short-term predictors of mortality are of particular interest for the acute management of patients admitted to Internal Medicine units. While RAMs for inhospital mortality are usually based mainly on physiological variables and a few laboratory data [5–10], prognostic models for late mortality typically include other domains such as disability and the presence of comorbidities [11–17]. RAMs based on physiological measures should be the most suitable to identify an acute deterioration in clinical status of patients, and they are often employed along a call-out cascade [18]. However, we and others have demonstrated that at least the Modified Early Warning Score calculated only a single time at admission, can predict in-hospital mortality [10, 19]. RAMs based on multiple domains, as the MoWS, may be even more appropriate to assess the patients’ hospital outcome [20]. However, to our knowledge, no study has evaluated whether combining these two different approaches could result in an even better prediction model. In a quite large study on elderly patients, Ponzetto and coll. [21] have shown that a multidimensional evaluation at admission can identify patients at risk for early and 5-year mortality. Results of our study suggest that the MoWS (a slightly modified version of the Walter score) may be accurate in predicting also the in-hospital mortality in general patients admitted to Internal Medicine units. In particular, it seems highly accurate in identifying low-risk patients (NPV 99.3 %; negative LR 0.05), whereas it is less accurate in identifying high-risk patients. Furthermore, this score is extremely simple to calculate, and seems to perform even better in younger patients. Thus, it has the potential of being regularly used in clinical practice. Many scores or scales have been derived to predict short-term outcomes in specific clinical conditions (i.e. acute coronary syndrome, alcoholic hepatitis). These scores are probably more accurate in those clinical settings. However, in Internal Medicine units, patients are commonly affected by more than one disease; thus the MoWS may be useful in particular when a single index event is not clearly identified, or when a specific score is not available. However, other studies are necessary to confirm our findings. In particular, since information on the score accuracy in patients \70 years is based on 157 patients only, future studies should include a consistent sample of
123
young patients. Subsequently, randomized, controlled trials should clarify if a strategy of patients’ treatment allocation according to the score results is associated with a reduction of in-hospital mortality before implementing the MoWS in clinical practice. Indeed, whether these risk prediction models will result in better patient outcomes (beyond the demonstration of some improvements in before and after studies), still remains an unanswered question [1]. Our study has some limitations. First, this is an observational retrospective study. Thus, the intrinsic limits of observational research must be considered. However, to avoid misleading results, we paid meticulous attention to data collection, and documentation was complete for the whole population with no missing information. Furthermore, our study has been conducted in a single center non-teaching hospital, and this may impair the external validity of our results. However, the study sample was quite large, and patients were consecutively enrolled with no major exclusion criteria. Thus, we believe that the results of this study can be applied also in other clinical contexts. As already stated, we need to evaluate impairment in ADLs at admission rather than at discharge since only patients’ characteristics assessed at admission can be used when the outcome is in-hospital mortality. Furthermore, the outcomes (in-hospital vs 1-year mortality) and populations (general population admitted to Internal Medicine) of our study are different. Thus, the results of our score should be interpreted with caution, and a validation in a subsequent study is warranted. Last, according to recent literature, several other clinical and laboratory variables (e.g., anemia, pressure ulcers, polypharmacy, or delirium) not considered in the MoWS may affect the short-term prognosis of our population [22]. However, evaluation of other risk factors is beyond the scope of our study, and their potential role should be assessed in properly designed studies.
Conclusions In conclusion, according to the results of our large study, the MoWS, a simple RAM based on multiple domains (demographics, dependency, comorbidities, laboratory values) seems to have good accuracy in predicting the risk of in-hospital mortality in patients admitted to Internal Medicine units. In particular, it seems highly accurate in identifying low-risk patients, but it is less accurate in identifying high-risk patients. Future studies are warranted to confirm our findings, and to clarify whether patients’ treatment allocation according to the score results is associated with a reduction of in-hospital mortality before implementing the MoWS in clinical practice. Conflict of interest
None.
Intern Emerg Med
References 1. Churpek MM, Yuen TC, Edelson DP (2013) Risk stratification of hospitalized patients on the wards. Chest 143(6):1758–1765 2. Walter LC, Brand RJ, Counsell SR, Palmer RM, Landefeld CS, Fortinsky RH, Covinsky KE (2001) Development and validation of a prognostic index for 1-year mortality in older adults after hospitalization. JAMA 285:2987–2994 3. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP (2007) STROBE initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Prev Med 45(4):247–251 4. Cleves MA (2002) From the help desk: comparing areas under receiver operating characteristic curves from two or more probit or logit models. Stata J 2:301–313 5. Knaus WA, Draper EA, Wagner DP, Zimmerman JE (1985) APACHE II: a severity of disease classification system. Crit Care Med 10:818–829 6. Ferreira FL, Bota DP, Bross A, Me´lot C, Vincent JL (2001) Serial evaluation of the SOFA score to predict outcome in critically ill patients. JAMA 14:1754–1758 7. Prytherch DR, Smith GB, Schmidt PE, Featherstone PI (2010) ViEWS—towards a national early warning score for detecting adult inpatient deterioration. Resuscitation 81(8):932–937 8. Smith GB, Prytherch DR, Meredith P, Schmidt PE, Featherstone PI (2013) The ability of the National Early Warning Score (NEWS) to discriminate patients at risk of early cardiac arrest, unanticipated intensive care unit admission, and death. Resuscitation 84(4):465–470 9. Burch VC, Tarr G, Morroni C (2008) Modified Early Warning Score predicts the need for hospital admission and inhospital mortality. Emerg Med J 10:674–678 10. Cei M, Bartolomei C, Mumoli N (2009) In-hospital mortality and morbidity of elderly medical patients can be predicted at admission by the Modified Early Warning Score: a prospective study. Int J Clin Pract 63:591–595 11. Levine SK, Sachs GA, Jin L, Meltzer D (2007) A prognostic model for 1-year mortality in older adults after hospital discharge. Am J Med 120:455–460 12. Inouye SK, Peduzzi PN, Robison JT, Hughes JS, Horwitz RI, Concato J (1998) Importance of functional measures in predicting mortality among older hospitalized patients. 15(279):1187–1193
13. Lee SJ, Lindquist K, Segal MR, Covinsky KE (2006) Development and validation of a prognostic index for 4-year mortality in older adults. JAMA 295:801–808 14. Drame´ M, Novella JL, Lang PO, Somme D, Jovenin N, Lanie`ce I, Couturier P, Heitz D, Gauvain JB, Voisin T, De Wazie`res B, Gonthier R, Ankri J, Jeandel C, Saint-Jean O, Blanchard F, Jolly D (2008) Derivation and validation of a mortality-risk index from a cohort of frail elderly patients hospitalised in medical wards via emergencies: the SAFES study. Eur J Epidemiol 23:783–791 15. Inouye SK, Bogardus ST Jr, Vitagliano G, Desai MM, Williams CS, Grady JN, Scinto JD (2003) Burden of illness score for elderly persons: risk adjustment incorporating the cumulative impact of diseases, physiologic abnormalities, and functional impairments. Med Care 41(1):70–83 16. Charlson ME, Pompei P, Ales KL et al (1987) A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 40:373–383 17. Desai MM, Bogardus ST Jr, Williams CS, Vitagliano G, Inouye SK (2002) Development and validation of a risk-adjustment index for older patients: the high-risk diagnoses for the elderly scale. J Am Geriatr Soc 50(3):474–481 18. Ridley S (2005) The recognition and early management of critical illness. Ann R Coll Surg Engl 87:315–322 19. Groarke JD, Gallagher J, Stack J et al (2008) Use of an admission Early Warning Score to predict patient morbidity and mortality and treatment success—a prospective cohort study. Eur J Intern Med 19:e8 20. Perera YS, Ranasinghe P, Adikari AM, Welivita WD, Perera WM, Wijesundara WM, Karunanayake SA, Constantine GR (2011) The value of the Modified Early Warning Score and biochemical parameters as predictors of patient outcome in acute medical admissions a prospective study. Acute Med 10(3): 126–132 21. Ponzetto M, Maero B, Maina P, Rosato R, Ciccone G, Merletti F, Rubenstein LZ, Fabris F (2003) Risk factors for early and late mortality in hospitalized older patients: the continuing importance of functional status. J Gerontol A Biol Sci Med Sci 58(11):1049–1054 22. Iwata M, Kuzuya M, Kitagawa Y, Suzuki Y, Iguchi A (2006) Underappreciated predictors for postdischarge mortality in acute hospitalized oldest-old patients. Gerontology 52(2):92–98
123
IM - ORIGINAL
A prognostic index for 1-year mortality can also predict in-hospital mortality of elderly medical patients Marco Cei • Nicola Mumoli • Jose´ Vitale Francesco Dentali
•
Received: 13 November 2014 / Accepted: 13 January 2015 Ó SIMI 2015
Abstract Elderly patients admitted to the hospital are at increased risk for both in-hospital and post-discharge mortality. Risk assessment models (RAMs) for in-hospital mortality are based mainly on physiological variables and a few laboratory data, whereas RAMs for late mortality usually include other domains such as disability and comorbidities. We aim to evaluate if a previous validated model for 1-year mortality (the Walter Score) would also work well in predicting in-hospital mortality. We retrospectively revised the medical records of patients admitted on our ward, from April to December, 2013. Data regarding gender, activities of daily living (ADLs), comorbidities, and routine laboratory tests were used to calculate a Modified Walter Score (MoWS). The main outcome measure was all cause, in-hospital mortality. The analysis involved 1,004 patients. Of these, 888 were discharged alive, and 116 (11.5 %) died during the hospitalization. The mean MoWS was 4.9 (±3.6) in the whole sample. Stratification into risk classes parallels with in-hospital mortality (Chi square for trend p \ 0.001). When dichotomized, MoWS has a sensitivity of 97.4 % (95 % CI 92.1–99.3), and a specificity of 48.2 % (95 % CI 44.9–51.5) with a good prognostic accuracy (area under the ROC = 0.81; 95 % CI 0.78, 0.84). Subgroup analysis according to different age groups gives similar results. A simple RAM based on
M. Cei N. Mumoli (&) Department of Internal Medicine, Ospedale Civile di Livorno, Viale Alfieri, 36, 57124 Livorno, Italy e-mail: [email protected] J. Vitale F. Dentali Department of Clinical Medicine, Insubria University, Varese, Italy
multiple domains, previously validated for predicting mortality of older adults within 1 year from the index hospitalization, can be useful at the moment of admission to Internal Medicine wards to accurately identify patients at low risk of in-hospital mortality. Keywords Elderly Medical admission Mortality Risk assessment model
Introduction Patients admitted to Internal Medicine units have a real risk for both in-hospital and post-discharge mortality. Several prognostic models have been developed to help clinicians predict adverse outcomes during hospitalization and after a long-term follow-up in this population [1]. In 2001, Walter and coworkers [2] derived and validated a simple score to predict the 1-year mortality of older adults (C70 years old) after hospitalization. After a multivariate analysis, they were able to identify a list of six independent predictors of long-term mortality commonly collected during hospitalization in most Internal Medicine units, making their score easily usable. These include— male gender, number of dependencies in activities of daily living (ADLs), the presence of congestive heart failure (CHF) or cancer; serum creatinine level higher than 3.0 mg/dL, and low albumin levels. To date, no study has evaluated the accuracy of this score in predicting in-hospital mortality in patients admitted to internal medicine units. Furthermore, the score has not been tested in younger patients. Thus, we planned to address this knowledge gap evaluating its accuracy in a large group of consecutive patients admitted to our internal medicine unit.
123
Intern Emerg Med
Methods Patient’s eligibility and data collection Medical records of all patients admitted to the Internal Medicine unit of Livorno Hospital in Italy from 1st April to 31st December 2013 were evaluated. To increase the external validity of the study, no exclusion criteria were planned, and all patients were potentially eligible for the study. However, patients presented more than once during the study period, and patients who were hospitalized for less than 24 h were excluded. The setting was a 43-bed medical unit of a non-teaching Italian hospital that receives unselected medical patients, admitted for acute medical conditions coming mainly from the Emergency Department (ED), either directly or after a brief stay for clinical stabilization. At the time of admission to Internal Medicine, information on patients’ baseline characteristics (age and gender), and on all the other items of the Walter Score (number of dependencies in ADLs, CHF, cancer; serum creatinine and albumin levels), was collected. As a difference from the original score, for the study purpose, the ADLs were considered on the day of admission and not at the time of discharge. This information was collected using a standardized form in all the included patients. Specific thresholds for each item of the Walter score are summarized in Table 1. Thus, a Modified Walter Score (MoWS) was calculated. For each patient, information on the in-hospital mortality was registered. The Institutional Review Board approved the study, which was carried out and is reported according to the Strengthening the Reporting of observational Studies in Epidemiology (STROBE) guidelines for observational studies [3]. Statistical analysis Continuous variables were expressed as mean plus or minus the standard deviation (SD) or as median with Table 1 The Walter score Risk factor
Points
Male gender
1
Dependency in 1–4 ADLs
2
Dependency in all 5 ADLsa
5
Congestive heart failure
2
Solitary or hematologic cancer
3
Metastatic cancer
8
Creatinine [3.0 mg/dL
2
Albumin 3.0–3.4 g/dL
1
Albumin \3.0 g/dL
2
a
ADLs denotes activities of daily living
123
minimum and maximum values when data did not have a normal distribution; categorical data are given as counts and percentages. Initially, according to the results of the original study [2], patients were divided into four risk classes (class I: 0–1 point, class II: 2–3 points, class III: 4–6 points, and class IV: [6 points), and class-specific rate of in-hospital mortality with the corresponding 95 % confidence interval (CI) was calculated. Subsequently, to explore the role of the score in predicting in-hospital mortality, patients were divided into two risk classes (low risk: MoWS \ 4 and high risk: MoWS C 4), and sensitivity, specificity, positive and negative predictive values and likelihood ratios for low vs. high-risk patients were estimated. Finally, the area under the receiver operating characteristic (ROC) curve was measured [4]. Furthermore, we assessed the accuracy of the MoWS in the subgroup of patients younger than 70 years old, and as well in the subgroup of patients of 70 years or older. We planned our sample size to find a difference in the mortality rate in the low and in high-risk groups. Hypothesizing a mean in-hospital mortality rate of about 8–10 %, (5–6 % in the low-risk group vs 11–12 % in the high-risk group), and considering one-third of the whole population at low risk and two-thirds at high risk, at least 850–900 patients should be enrolled to have a power of 90 % to find significant differences (p \ 0.05) between the groups. To be more conservative, the planned sample was incrementally increased to a total of 950–1,000 patients. All analyses were performed using SPSS 19.0 (SPSS Inc, Chicago, IL, USA).
Results A total of 1,055 patients were admitted between April 1 and December 31, 2013. Of these, 51 were excluded from the analysis (41 discharged within 24 h and 10 readmitted for the second time during the study period), leaving 1,004 patients for analysis. Data on the items of the score were complete for the whole sample. The median age was 79.8 ± 12.2 years, ranging from 19 to 102 years; 545 (54.3 %) patients were female. The mean MoWS was 4.9 (± 3.6) in the whole sample. Eight hundred and eightyeight patients were discharged alive, and 116 died during the hospitalization for a mean mortality rate of 11.6 % (95 % CI 9.7, 13.7 %), Table 2. The distribution of patients across the different MoWS risk classes is shown in Table 3 and Fig. 1. Stratification into risk classes significantly correlates with the in-hospital mortality rate (Chi square for trend p \ 0.001). In-hospital mortality is significantly lower in the low-risk group (MoWS \ 4: 0.7 %, 95 % CI 0.2, 2.2 %) compared to the high-risk group (MoWS C 4: 19.7 %, 95 % CI 16.6, 23.3 %) (p \ 0.001). When
1.00
Intern Emerg Med
550 (54.8 %)
Mean age, years (SD)
79.8 (12.2)
Mean duration of hospitalization, days (SD)
9.35 (0.4)
Number of dependencies in ADL (n)
0 (306) 1 (6)
0.25
2 (35)
0.50
1,004
Sensitivity
Patients, n Female gender, n (%)
0.75
Table 2 Demographics of the whole sample
3 (52)
0.00
4 (170) 5 (435) Heart failure, n (%)
214 (21.3)
Solitary or hematologic cancer, n (%)
71 (7.1)
Metastatic cancer
73 (7.3)
Creatinine [3.0 mg/dL, n (%)
37 (3.7)
Albumin 3.0–3.4 g/dL, n (%) Albumin \3.0 g/dL, n (%)
326 (32.4) 405 (40.3)
Discharged, n (%)
888 (88.4)
Deceased, n (%)
116 (11.6)
Median time to death, days (range)
6 (1–47)
Table 3 Distribution of patients across the MoWS categories MoWS
Discharged
0–1
217
2–3
211
2
2.06 (0.18–22.8)
0.98
4–6
188
11
12.7 (1.62–99.25)
\0.002
7–20
272
102
81.38 (11.26–587.9)
\0.001
a
Died
Odd ratio for mortality (95 % CI)a
p
1
As compared to patients with MoWS 0–1
0.00
0.25
0.50
0.75
1.00
1 - Specificity Area under ROC curve = 0.8077
Fig. 2 Receiving operating curve evaluating the discriminatory power of the MoWS to predict in-hospital mortality
97.8, 99.8), and a positive predictive value of 19.7 % (95 % CI 16.6, 23.3) for in-hospital mortality, with a negative likelihood ratio of 0.05 (95 % CI 0.02, 0.16), and a positive likelihood ratio of 1.88 (95 % CI 1.75, 2.01) (Table 3). The discriminatory power of the MoWS to predict in-hospital mortality, expressed as the area under the ROC curve was 0.81 (95 % CI 0.78, 0.84; Fig. 2). One hundred and fifty-seven patients (15.5 %) were younger than 70 years. In this subgroup of patients, the MoWS score has a sensitivity of 100.0 % (95 % CI 65.5, 100), and a specificity of 71.2 % (95 % CI 63.1, 78.3). Eight hundred and forty-eight patients (84.5 %) were C70 years. In this subgroup of patients, the MoWS score has a sensitivity of 97.2 % (95 % CI 91.3, 99.3), and a specificity of 43.7 % (95 % CI 40.1, 47.3).
Discussion
Fig. 1 Distribution of patients across the MoWS categories. Blue, alive; green, dead
dichotomized, the MoWS score has a sensitivity of 97.4 % (95 % CI 92.1, 99.3), and a specificity of 48.2 % (95 % CI 44.9, 51.5), a negative predictive value of 99.3 % (95 % CI
Medical patients have a real risk of death during and after hospital admission. Assessment of this risk at the time of admission should result in numerous benefits. Patients could be better allocated to prevent clinical deterioration, and those at higher risk of death could be carefully evaluated before and after discharge. Walter and coworkers have demonstrated, using a simple score, that 1-year mortality is associated with risk factors belonging to multiple domains such as demographic variables, comorbidities, functional status and laboratory values. With the results of our study, we suggested that a slightly modified version of this score can easily predict the risk of death during hospitalization in a real-world setting. In particular, the score appears to have a very high sensitivity, and more than forty percent of patients may be considered at very low risk of in-hospital mortality. The
123
Intern Emerg Med
validity of these findings is strengthened by the results of subgroup analyses according to different ages. In particular, this score seems to perform even better in younger patients, with no false negative results, and with a higher sensitivity compared to the whole population. However, these latter results should be interpreted with caution due to the limited sample of the subgroup. Many risk assessment models (RAMs) have been developed to assess the short and long-term mortality risk of hospitalized patients. In particular, short-term predictors of mortality are of particular interest for the acute management of patients admitted to Internal Medicine units. While RAMs for inhospital mortality are usually based mainly on physiological variables and a few laboratory data [5–10], prognostic models for late mortality typically include other domains such as disability and the presence of comorbidities [11–17]. RAMs based on physiological measures should be the most suitable to identify an acute deterioration in clinical status of patients, and they are often employed along a call-out cascade [18]. However, we and others have demonstrated that at least the Modified Early Warning Score calculated only a single time at admission, can predict in-hospital mortality [10, 19]. RAMs based on multiple domains, as the MoWS, may be even more appropriate to assess the patients’ hospital outcome [20]. However, to our knowledge, no study has evaluated whether combining these two different approaches could result in an even better prediction model. In a quite large study on elderly patients, Ponzetto and coll. [21] have shown that a multidimensional evaluation at admission can identify patients at risk for early and 5-year mortality. Results of our study suggest that the MoWS (a slightly modified version of the Walter score) may be accurate in predicting also the in-hospital mortality in general patients admitted to Internal Medicine units. In particular, it seems highly accurate in identifying low-risk patients (NPV 99.3 %; negative LR 0.05), whereas it is less accurate in identifying high-risk patients. Furthermore, this score is extremely simple to calculate, and seems to perform even better in younger patients. Thus, it has the potential of being regularly used in clinical practice. Many scores or scales have been derived to predict short-term outcomes in specific clinical conditions (i.e. acute coronary syndrome, alcoholic hepatitis). These scores are probably more accurate in those clinical settings. However, in Internal Medicine units, patients are commonly affected by more than one disease; thus the MoWS may be useful in particular when a single index event is not clearly identified, or when a specific score is not available. However, other studies are necessary to confirm our findings. In particular, since information on the score accuracy in patients \70 years is based on 157 patients only, future studies should include a consistent sample of
123
young patients. Subsequently, randomized, controlled trials should clarify if a strategy of patients’ treatment allocation according to the score results is associated with a reduction of in-hospital mortality before implementing the MoWS in clinical practice. Indeed, whether these risk prediction models will result in better patient outcomes (beyond the demonstration of some improvements in before and after studies), still remains an unanswered question [1]. Our study has some limitations. First, this is an observational retrospective study. Thus, the intrinsic limits of observational research must be considered. However, to avoid misleading results, we paid meticulous attention to data collection, and documentation was complete for the whole population with no missing information. Furthermore, our study has been conducted in a single center non-teaching hospital, and this may impair the external validity of our results. However, the study sample was quite large, and patients were consecutively enrolled with no major exclusion criteria. Thus, we believe that the results of this study can be applied also in other clinical contexts. As already stated, we need to evaluate impairment in ADLs at admission rather than at discharge since only patients’ characteristics assessed at admission can be used when the outcome is in-hospital mortality. Furthermore, the outcomes (in-hospital vs 1-year mortality) and populations (general population admitted to Internal Medicine) of our study are different. Thus, the results of our score should be interpreted with caution, and a validation in a subsequent study is warranted. Last, according to recent literature, several other clinical and laboratory variables (e.g., anemia, pressure ulcers, polypharmacy, or delirium) not considered in the MoWS may affect the short-term prognosis of our population [22]. However, evaluation of other risk factors is beyond the scope of our study, and their potential role should be assessed in properly designed studies.
Conclusions In conclusion, according to the results of our large study, the MoWS, a simple RAM based on multiple domains (demographics, dependency, comorbidities, laboratory values) seems to have good accuracy in predicting the risk of in-hospital mortality in patients admitted to Internal Medicine units. In particular, it seems highly accurate in identifying low-risk patients, but it is less accurate in identifying high-risk patients. Future studies are warranted to confirm our findings, and to clarify whether patients’ treatment allocation according to the score results is associated with a reduction of in-hospital mortality before implementing the MoWS in clinical practice. Conflict of interest
None.
Intern Emerg Med
References 1. Churpek MM, Yuen TC, Edelson DP (2013) Risk stratification of hospitalized patients on the wards. Chest 143(6):1758–1765 2. Walter LC, Brand RJ, Counsell SR, Palmer RM, Landefeld CS, Fortinsky RH, Covinsky KE (2001) Development and validation of a prognostic index for 1-year mortality in older adults after hospitalization. JAMA 285:2987–2994 3. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP (2007) STROBE initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Prev Med 45(4):247–251 4. Cleves MA (2002) From the help desk: comparing areas under receiver operating characteristic curves from two or more probit or logit models. Stata J 2:301–313 5. Knaus WA, Draper EA, Wagner DP, Zimmerman JE (1985) APACHE II: a severity of disease classification system. Crit Care Med 10:818–829 6. Ferreira FL, Bota DP, Bross A, Me´lot C, Vincent JL (2001) Serial evaluation of the SOFA score to predict outcome in critically ill patients. JAMA 14:1754–1758 7. Prytherch DR, Smith GB, Schmidt PE, Featherstone PI (2010) ViEWS—towards a national early warning score for detecting adult inpatient deterioration. Resuscitation 81(8):932–937 8. Smith GB, Prytherch DR, Meredith P, Schmidt PE, Featherstone PI (2013) The ability of the National Early Warning Score (NEWS) to discriminate patients at risk of early cardiac arrest, unanticipated intensive care unit admission, and death. Resuscitation 84(4):465–470 9. Burch VC, Tarr G, Morroni C (2008) Modified Early Warning Score predicts the need for hospital admission and inhospital mortality. Emerg Med J 10:674–678 10. Cei M, Bartolomei C, Mumoli N (2009) In-hospital mortality and morbidity of elderly medical patients can be predicted at admission by the Modified Early Warning Score: a prospective study. Int J Clin Pract 63:591–595 11. Levine SK, Sachs GA, Jin L, Meltzer D (2007) A prognostic model for 1-year mortality in older adults after hospital discharge. Am J Med 120:455–460 12. Inouye SK, Peduzzi PN, Robison JT, Hughes JS, Horwitz RI, Concato J (1998) Importance of functional measures in predicting mortality among older hospitalized patients. 15(279):1187–1193
13. Lee SJ, Lindquist K, Segal MR, Covinsky KE (2006) Development and validation of a prognostic index for 4-year mortality in older adults. JAMA 295:801–808 14. Drame´ M, Novella JL, Lang PO, Somme D, Jovenin N, Lanie`ce I, Couturier P, Heitz D, Gauvain JB, Voisin T, De Wazie`res B, Gonthier R, Ankri J, Jeandel C, Saint-Jean O, Blanchard F, Jolly D (2008) Derivation and validation of a mortality-risk index from a cohort of frail elderly patients hospitalised in medical wards via emergencies: the SAFES study. Eur J Epidemiol 23:783–791 15. Inouye SK, Bogardus ST Jr, Vitagliano G, Desai MM, Williams CS, Grady JN, Scinto JD (2003) Burden of illness score for elderly persons: risk adjustment incorporating the cumulative impact of diseases, physiologic abnormalities, and functional impairments. Med Care 41(1):70–83 16. Charlson ME, Pompei P, Ales KL et al (1987) A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 40:373–383 17. Desai MM, Bogardus ST Jr, Williams CS, Vitagliano G, Inouye SK (2002) Development and validation of a risk-adjustment index for older patients: the high-risk diagnoses for the elderly scale. J Am Geriatr Soc 50(3):474–481 18. Ridley S (2005) The recognition and early management of critical illness. Ann R Coll Surg Engl 87:315–322 19. Groarke JD, Gallagher J, Stack J et al (2008) Use of an admission Early Warning Score to predict patient morbidity and mortality and treatment success—a prospective cohort study. Eur J Intern Med 19:e8 20. Perera YS, Ranasinghe P, Adikari AM, Welivita WD, Perera WM, Wijesundara WM, Karunanayake SA, Constantine GR (2011) The value of the Modified Early Warning Score and biochemical parameters as predictors of patient outcome in acute medical admissions a prospective study. Acute Med 10(3): 126–132 21. Ponzetto M, Maero B, Maina P, Rosato R, Ciccone G, Merletti F, Rubenstein LZ, Fabris F (2003) Risk factors for early and late mortality in hospitalized older patients: the continuing importance of functional status. J Gerontol A Biol Sci Med Sci 58(11):1049–1054 22. Iwata M, Kuzuya M, Kitagawa Y, Suzuki Y, Iguchi A (2006) Underappreciated predictors for postdischarge mortality in acute hospitalized oldest-old patients. Gerontology 52(2):92–98
123
