OUP user menu

Prediction of mortality 1 year after hospital admission

J. Kellett, S. Rasool, B. McLoughlin
DOI: http://dx.doi.org/10.1093/qjmed/hcs099 847-853 First published online: 11 June 2012

Abstract

Objective: Hospital admission, especially for the elderly, can be a seminal event as many patients die within a year. This study reports the prediction of death within a year of admission to hospital of the Simple Clinical Score (SCS) and ECG dispersion mapping (ECG-DM). ECG-DM is a novel technique that analyzes low-amplitude ECG oscillations and reports them as the myocardial micro-alternation index (MMI).

Methods: a convenient sample of 430 acutely ill medical patients (mean age 67.9 ± 16.6 years) was followed up for 1 year after their last admission to hospital.

Results: Seventy-four (16.3%) patients died within a year—all but seven had a SCS ≥5 and 40% of those with an MMI ≥50% died. Only six of variables were found by logistic regression to be independent predictors of mortality (i.e. age, MMI, SCS, a discharge diagnosis of cancer, hemoglobin <11 gm% and prior illness that required the patient to spend >50% of daytime in bed). The SCS and MMI plus age were comparable predictors of 1-year mortality: SCS ≥12 predicted 1-year mortality with the highest odds (16.1, chi square 79.09, p < 0.0001) and a score of age plus MMI >104 had an odds ratio of 9.4 (chi square 73.50, p < 0.0001), identified 69% of deaths, and 43% of the 119 patients who exceeded this score were dead within a year.

Conclusion: SCS and ECG-DM plus age are clinically useful for long-term prognostication. ECG-DM is inexpensive, only takes a few seconds to perform and requires no skill to interpret.

Introduction

Hospital admission, especially for the elderly, can be a seminal event. It has been reported that more than a quarter of patients >65 years of age are dead within a year of hospital admission.1 In 2006, we reported the Simple Clinical Score (SCS), a reliable and objective predictor of in-hospital mortality that has been independently validated in several different clinical settings.26 Its ability to predict long-term prognosis, however, has never been reported.

The five most important components of the SCS are hypotension, hypoxia, hypothermia, inability to stand unaided, and ECG abnormality—only 4 out of 4177 patients with a normal ECG died within 24 h.7 Performing and interpreting an ECG is time consuming, requires training and skill, and can, therefore, be costly and impractical. ECG dispersion mapping (ECG-DM) is a recently developed technology that provides non-invasive assessment of myocardial pathology by analyzing low-amplitude oscillations of conventional ECG signals. This method examines not only the tiny fluctuations of the T-wave (T wave alternans) but also micro-fluctuations of the P wave and QRS complex. All these micro-deviations are combined into one overall myocardial micro-alternation index (MMI), which is expressed as a percentage8–10—if pathological micro-deviations are present throughout the entire myocardium, the MMI is 100%, whereas an MMI of 0% indicates a total absence of abnormal micro-deviations.

This study reports the 1-year follow-up of a previously reported cohort of acutely ill medical patient11–12 admitted to our hospital. It examines the ability of the SCS and ECG-DM to predict 1-year mortality compared with the standard 12 lead ECG and other commonly used mortality predictors.

Methods

All patients were recruited from the unselected acutely ill medical patients admitted to Nenagh Hospital between 30 July 2009 and 31 March 2010. Of the 1680 patients admitted during this period, 455 (27.1%) had the SCS measured and ECG-DM performed on them within 10–20 min of hospital admission. Since 25 of these patients were readmitted during the study period, this study followed up 430 individual patients [mean age (SD) 67.9 (16.6) years, range 16–99 years] for 365 days after the time of their last admission.

Nenagh Hospital has recently undergone substantial reconfiguration of its service. At the time of this study, Nenagh Hospital was small general hospital in rural Ireland serving a population of 60 000. It had a 36 bed acute medical unit with 2800 admissions per year almost all of which were unplanned emergencies.13 It was served by four consultant physicians each assisted by a team of three physicians in training—each team was on-call every fourth day. The hospital had a five-bed ICU capable of cardiac monitoring, external and temporary transvenous pacing, non-invasive and invasive ventilation. Renal dialysis, hematology, and oncology units were available at Limerick Regional Hospital, 25 miles away.

The SCS can be quickly performed at the bedside and requires no additional information or investigation other than a 12 lead ECG; ECG-DM (HeartVue system, Medical Computer Systems Ltd, Moscow) is a non-invasive procedure that takes 30 s to perform and only requires the placement of the four standard ECG limb leads. All data were corrected for errors and then entered into an Epi-Info version 6.0 database (Center for Disease Control and Prevention, USA). The ICD10 coded diagnoses recorded at discharge or death were also entered into this database. In particular, we examined the 10 commonest of 32 co-morbid diagnoses that we have previously reported13 to have an increased risk of in-hospital mortality (i.e. heart failure, atrial fibrillation, diabetes, hyponatremia, chronic obstructive lung disease, anemia, altered mental status, pneumonia, neoplasia and acute myocardial infarction).

Follow-up was performed by contacting the patient directly and/or their family doctor and/or the hospitals patient administration system for those patients still attending out-patient services. For those patients who had died, the exact date of death was obtained from the North Tipperary Registrar of Births Marriages and Deaths and/or the website rip.ie. No patient was lost to follow-up.

For univariate analysis, the continuous variables of age, MMI etc., were converted into categorical variables by determining the odds ratio with the highest chi-squared number for predicting in-hospital mortality. This was done by a process of trial and error. Continuous variables were compared by Student’s t-test and categorical variables by chi-square analysis that applied Yates continuity correction—all calculations were performed using Epi-Info version 6.0 (Center for Disease Control and Prevention, USA), and statistical significance was set at a p value <0.05. The independence of variables was tested by logistic regression using Logistic Version 3.11E software (G.E. Dallal, Andover MA).14 The area under the receiver operating characteristic curves (AUROC) was calculated according to the method of Hanley and McNeil.15

Ethical approval of the study was obtained from the Mid-Western Regional Hospital Complex Scientific Research Ethics Committee, which granted exemption for patient consent.

Results

Sixteen patients died during their hospital admission [mean length of hospital stay 10.4 (SD) (7.7) days], and only one patient who died in hospital was under 68 years of age. One year after admission, a total of 74 (16.3%) patients had died. The majority of these patients were elderly—1 year after admission, 69 (23.8%) of the 244 patients >68 years of age were dead. All but seven of the 74 patients who died during the 1-year follow-up period had a SCS ≥5 (Figure 1). Of the seven out of 162 patients with a SCS ≤4 who died, four had cancer, one liver failure and two heart failure with MMI values of 40% and 50%. All of the patients with an MMI <10% and 90% of those with an MMI between 10 and 20% were still alive 1 year later, while 40% of those with an MMI ≥50% died (Figure 2).

Figure 1.

Mortality at 365 days of patients according to their Simple Clinical Score (SCS). Only 7 out of 162 patients with a SCS ≤4 died: four had cancer, one liver failure and two heart failure with MMI values of 40% and 50%.

Figure 2.

Mortality at 365 days of patients according to their myocardial micro-alternation index (MMI).

Apart from breathing rate and oxygen saturation, there were no differences in the vital signs on admission of the patients who died. Although patients who died had a higher white cell count and urea and lower hemoglobin levels, their admission sodium and potassium levels were similar to those of survivors. However, numerous categorical variables identified patients who more likely to die within a year (Table 1), those with highest chi-square number were: SCS ≥12, age combined with MMI >104, prior illness that required the patient to spend >50% of daytime in bed, age >68 years, being unable to stand without help, a discharge diagnosis of cancer, and an MMI ≥24%.

View this table:
Table 1

Continuous and categorical variables considered as possible predictors of death within 1 year of admission

AdmittedAliveDeadOdds ratioChi SquareP
Number of patients43035674
Age, mean (SD) (years)67.9 (16.6)65.4 (16.5)79.9 (10.4)0.000000
Temperature, mean (SD) (°C)36.6 (0.7)36.6 (0.7)36.6 (0.9)0.52
Systolic blood pressure, mean (SD) (mmHg)142 (26)142 (26)138 (29)0.24
Diastolic blood pressure, mean (SD) (mmHg)79 (16)80 (15)76 (18)0.08
Heart rate per minute, mean (SD)84 (21)83 (21)87 (22)0.18
Oxygen saturation, mean (SD) (%)95.5 (3.7)95.9 (2.9)93.5 (5.8)0.000001
Breathing rate per minute, mean (SD)21 (5)20 (4)23 (7)0.000000
Hemoglobin, mean (SD) (g%)13.0 (2.2)13.3 (2.1)11.7 (2.2)0.000000
White cell count, mean (SD) (mm3)10.5 (5.0)10.0 (4.5)12.6 (6.4)0.00006
Serum sodium, mean (SD) (mmol/l)140.0 (4.6)139.1 (4.0)138.5 (6.8)0.39
Serum potassium, mean (SD) (mmol/l)4.4 (0.6)4.4 (0.7)4.4 (0.6)0.94
Urea, mean (SD) (mmol/l)8.9 (7.9)7.9 (7.4)13.2 (8.8)0.000000
Simple Clinical Score, mean (SD)6.5 (3.9)5.6 (3.3)10.5 (4.2)0.000000
MMI, mean (SD) (%)23.0 (13.5)21.5 (12.5)30.2 (15.6)0.000000
Length of hospital stay, mean (SD) (days)6.6 (6.9)5.5 (5.2)11.7 (10.7)0.000000
Male, n (%)240 (55.8)204 (57.3)36 (48.6)1.4 (0.8–2.4)1.530.22
Chest pain, n (%)65 (15.1)63 (17.7)2 (2.7)0.1 (0.0–0.6)9.600.002
Breathlessness, n (%)148 (34.4)112 (31.5)36 (48.6)2.1 (1.2–3.5)7.180.007
Nursing home resident, n (%)33 (7.7)19 (5.3)14 (18.9)4.1 (1.8–9.3)14.090.0002
Spends >50% of daytime in bed, n (%)63 (14.7)30 (8.4)33 (44.6)8.7 (4.6–16.6)61.00<0.00001
Unable to stand without help, n (%)79 (18.4)46 (12.9)33 (44.6)5.4 (3.0–9.8)38.72<0.00001
Altered mental status, n (%)52 (12.1)32 (9.0)20 (27.03.8 (1.9–7.4)17.090.00004
Stroke, n (%)14 (3.3)6 (1.7)8 (10.8)7.1 (2.1–24.1)13.430.0002
Heart failure, n (%)81 (18.8)58 (16.3)23 (31.1)2.3 (1.3–4.3)7.820.005
Acute myocardial infarction, n (%)19 (4.4)15 (4.2)4 (5.4)1.3 (0.4–4.4)0.020.89
Chronic obstructive pulmonary disease, n (%)63 (14.7)53 (14.9)10 (13.5)0.9 (0.4–2.0)0.020.90
Pneumonia, n (%)57 (13.3)42 (11.8)15 (20.3)1.9 (0.9–3.8)3.120.08
Cancer, n (%)16 (3.7)12 (3.4)4 (5.4)17.0 (4.8–65.6)34.86<0.00001
Diabetes, n (%)62 (14.4)49 (13.8)13 (17.6)1.3 (0.6–2.7)0.440.51
Abnormal ECG, n (%)305 (70.9241 (67.7)64 (86.5)3.1 (1.4–6.6)9.600.002
Atrial fibrillation, n (%)48 (11.2)35 (9.8)13 (17.6)2.0 (0.9–4.1)2.960.09
Hemoglobin <11, n (%) g%52 (12.1)32 (9.0)20 (27.0)3.8 (1.9–7.4)17.090.00004
Serum sodium <134, n (%) mmol/l45 (10.5)32 (9.0)13 (17.6)2.2 (1.0–4.6)3.940.05
Temperature <35, n (%)4 (0.9)3 (0.8)1 (1.4)1.6 (0.0–17.9)0.060.80
Temperature ≥39, n (%)2 (0.5)2 (0.6)0 (0.0)0.0 (0.0–20.2)0.090.77
Simple Clinical Score ≥12, n (%)41 (9.5)13 (3.7)28 (37.8)16.1 (7.3–35.7)79.09<0.00001
MMI ≥24%, n (%)140 (32.6)96 (27.0)44 (59.5)4.0 (2.3–6.9)28.00<0.00001
Age >68 years, n (%)244 (56.7)175 (49.2)69 (93.2)14.3 (5.3–41.6)46.73<0.00001
Age + MMI >104, n (%)119 (27.7)68 (19.1)51 (68.9)9.4 (5.2–17.2)73.50<0.00001
  • MMI = myocardial micro-alternation index

Apart from MMI, there were no statistical significant differences in any of the continuous or categorical variables between those patients who died with an MMI above and below 24%. Only six of variables were found by logistic regression to be independent predictors of mortality (i.e. age, MMI, SCS, a discharge diagnosis of cancer, a hemoglobin <11 gm%, and prior illness that required the patient to spend >50% of daytime in bed) (Table 2).

View this table:
Table 2

Logistic regression identified only six independent predictors of mortality at 1 year

CoefficientSEOdds ratiop
Constant−8.12481.2415<0.001
Age0.04550.01651.04650.0060
MMI0.03460.01001.03520.0005
Cancer2.78530.729816.20430.0001
Hemoglobin <11 gm%0.95870.41492.60830.0209
Simple Clinical Score0.21290.05821.23730.0003
Spends more than 50% of daytime in bed0.93060.41252.53610.0241
  • MMI = myocardial micro-alternation index.

SCS ≥12 predicted 1-year mortality with the highest odds (16.1, chi square 79.09, p < 0.0001). However, the AUROC of age combined with MMI as a predictor of annual mortality is 0.79 (95% CI 0.72–0.85) and is comparable to the AUROC of the SCS (0.82, 95% CI 0.76–0.88). A score of age plus MMI >104 identified 69% of deaths, and 43% the 119 patients who exceeded this score were dead within a year (odds ration 9.4, chi square 73.50, p < 0.0001). Quartiles of this simple score distinguished patients with a very low (3%) annual risk of death (i.e. age plus MMI <74), a low (7%) risk (i.e. age plus MMI ≥74 to <91), an intermediate (15%) risk (i.e. age plus MMI ≥91 to <106) from those with a very high (41%) risk (i.e. age plus MMI ≥106) (Figure 3). A more precise estimate of survival at 1 year can be made from the formula 1/1 + ez, where z is −8.2476 + age * 0.0806 + MMI * 0.0275.

Figure 3.

Kaplan–Meir survival curves for 365 days of patients according to quartiles of their age combined with their myocardial micro-alternation index (MMI). One quarter of patients had a very low (3%) annual risk of death (i.e. age plus MMI <74), one quarter a low (7%) risk (i.e. age plus MMI ≥74 to <91), one quarter an intermediate (15%) risk (i.e. age plus MMI ≥91 to <106) and one quarter a very high (41%) risk (i.e. age plus MMI ≥106).

Discussion

A predictive score’s performance is measured by determining its discrimination, calibration and universal validity. Compared with other predictive mortality scores, the SCS provides accurate predictions, excellent discrimination between survivors and non-survivors,16 and has been validated in diverse clinical settings.36 This study demonstrates that the SCS retains its discrimination for up to 1 year after hospital admission. It also suggests that ECG-DM is a significant independent predictor of annual mortality, which when combined with age has comparable discrimination to the SCS. Although the standard 12 lead ECG is a highly sensitive means of ruling out the risk of in-hospital mortality in acutely ill medical patients,7 the influence of slight ECG abnormalities on mortality risk has not been adequately quantified. MMI, on the other hand, is reported as a continuous variable that increases as the risk of death rises.

This small study of a convenient sample of patients it obviously limited. Doubtless a larger study would have detected more than six independent predictors of mortality and would have yielded more accurate comparisons of their relative predictive power. This study was performed just before services in the hospital underwent reconfiguration part of which was the diversion after midnight of seriously ill patients directly to Limerick Regional Hospital. The overall mortality rate of 3.5%, therefore, was comparable to the in-hospital mortality observed in both hospitals over the last 10–15 years. The patients studied represent those patients that could be seen promptly after their arrival in hospital. Since the authors worked a one-in-four 24 hour roster (i.e. 25% of the time) and 27% of all admissions were studied the patients represent most of the patients the authors cared for. It is possible, of course, that ECGDM would not be able to predict mortality so accurately in a different patient population. However, we are currently following a cohort of 99 cardiac out-patients with a lower mean MMI than our study population (17% SD 9% versus 23.0% SD 13.5%, p < 0.00001)—after 1 year of follow-up, the only patient that has died was the one with the highest age + MMI score of 124 (i.e. age 89 + MMI 35%).

The MMI is more than likely a proxy for cardiac disease. The MMI reflects abnormalities in the myocardium at a metabolic level, which may include ischemia and other causes. It is important to stress that it should not be relied upon for diagnostic purposes. Of 19 patients who had acute myocardial infarctions, six had normal MMI values (i.e. <20%) early in their presentation—two of these patients had obvious classical acute ischemic changes on 12 lead ECG at the time ECG-DM was performed.11 An ECG-DM, therefore, is not diagnostic but a non-specific indicator of myocardial health. Measurement is carried out not only directly on an ECG-signal but also indirectly by examining the asymmetry of the ECG signal between the right and left ventricles, which allows signal averaging to be obtained within 30 s. From the six traditional ECG leads of I, II, III, aVR, aVL and aVF patented formulae determine asymmetry of the ECG signal between the ventricles in very a short time.8,9 This signal is highly accurate and correlates with the amplitude of micro-alternations not only of the T wave but of the entire PQRS complex. This method is much more sensitive than direct “beat-to-beat” measurement of ECG micro-alternations, and this allows measurements to be performed at rest. Although other techniques might diagnose cardiac disease better than ECG-DM, 1 year mortality is most likely related to the severity of disease rather than diagnoses per se. Unlike clinical opinion and other diagnostic techniques, ECG-DM is objective, quick, easy and cheap to perform and does not require any skill or training to interpret.

Prognostication has been recognized as an important but often neglected part of patient management.17 Acute hospitalization, especially of the elderly, is an opportunity to recognize patients at risk of imminent decline and mortality and make appropriate arrangements in keeping with their wishes.18 As much as a quarter of the costs of health care are spent in the last year of life.19 If this money is to be spent wisely, care should be planned in advance with attention to symptom management, caregiver support, so that the resources available are efficiently directed to the needs and desires of the patient.20,21

Other prognostic indices are available to predict mortality in hospitalized older adults, but their use has not been incorporated into routine medical practice.1 Some existing models are applicable only to specific patient populations and disease states22–24 or require subjective assessments of risk by clinicians.25 Others require use of lengthy formulas26 or knowledge of certain laboratory data and functional status,27,28 which are not always available in a patient’s chart.29 This study suggests that the SCS has a comparable performance to other more sophisticated tools.22,26 Even though already contained in the SCS, time in bed was also an independent predictor of mortality at 1 year. This indicates that the weighting given to time in bed by the SCS requires readjustment to more accurately predict 1-year mortality.

The finding that MMI is predicts 1-year mortality independently of age indicates that it is not just a proxy for age. Moreover, a combination of age and ECGDM quickly and objectively identifies within seconds of admission to hospital a substantial cohort of acutely ill medical patients who have a >40% risk of being dead within a year, which translates into a life expectancy of 1.5–1.7 years. This information should prompt clinicians to initiate discussions on advance care planning and goals of care in their patients.

Conclusion

  1. The SCS retains its discrimination (AUROC 82%) for mortality prediction of acutely ill medical patients up to 1 year after admission to hospital.

  2. ECG-DM combined with age has comparable discrimination for 1-year mortality prediction to the SCS but is quicker, cheaper and easier to perform. It also requires no skill to interpret.

Clinical significance

  • The Simple Clinical Score (SCS) is a reliable independently validated predictor of in-hospital mortality

  • ECG-DM is a novel technique that reports abnormal low-amplitude ECG oscillations as the myocardial micro-alternation index (MMI).

  • The SCS and MMI plus age were comparable predictors of 1-year mortality

  • SCS and ECG-DM are clinically useful for prognostication. ECG-DM is cheap, quick and easy to interpret.

Acknowledgements

The authors would like to acknowledge the help of Ms Margaret Hayes and her staff, Tipperary Registry of Births, Marriages and Deaths, Nenagh.

Conflict of interest: None declared.

References

View Abstract