elderly trauma patients

Anaesthesia and Intensive Care, Vol. 37, No. 5, September 2009

The population demographic of Australia has undergone recent change. Between 1985 to 2005, the population above the age of 65 years increased from 10% to 13% and those above 85 years has increased from less than 1% in 1984 to 1.5% in 20041. As the elderly live longer, they will be exposed to an increase in risk of trauma. This being said, trauma remains an uncommon cause of death in the elderly. According

to the Australian Bureau of Statistics, in 2004 trauma as a cause of death (categorised within external causes of morbidity and mortality) in those aged between 75 to 84 years was ranked eighth and ninth amongst males and females respectively2. However, the impact of trauma upon the elderly is substantial. The recently published National Trauma Registry (Australia and New Zealand) revealed highest mortality in elderly trauma patients: of those aged 65 to 74 years, 79.9% survived to hospital discharge; those aged 75 to 84 years, 71% survived to discharge; and those above 85 years only 64% survived to discharge3.

Reports detailing mortality and functional out-comes of hospitalised elderly trauma patients from the USA, Canada and Europe concluded that age, associated comorbidities and injury severity predict prolonged intensive care unit (ICU) stay and increased mortality in hospital and after discharge4-8. Some data are available with respect to the Australian experience, but none specifically on the cost of

*M.B., B.S., F.J.F.I.C.M. , Senior Registrar in Intensive Care Medicine.†M.B., B.S., F.R.A.C.P., F.J.F.I.C.M., M.D., Associate Professor, Department of Intensive Care Medicine, The Queen Elizabeth Hospital, Woodville.

‡B.N., Grad. Dip. Emerg., M.N.Sc., Grad. Cert. Clinical Teaching, Trauma Co-ordinator, Trauma Service, Royal Adelaide Hospital.

§B. Ap.Sc., Business Manager, Radiology Business Management, The Queen Elizabeth Hospital, Woodville.

** B.Sc., Dip. Math. Stats., Ph.D., Professor of Bioinformatics, School of Mathematical Sciences, University of Adelaide.

Address for reprints: A/Prof J. L. Moran, Department of Intensive Care Medicine, The Queen Elizabeth Hospital, Woodville Road, Woodville, SA 5011.

Accepted for publication on March 4, 2009.

Mortality and cost outcomes of elderly trauma patients admitted to intensive care and the general wards of an Australian tertiary referral hospitalL. Y. L. CHAN*, J. L. MoRAN†, C. CLARkE‡, J. MARTIN§, P. J. SoLoMoN**Department of Intensive Care Medicine, Royal Adelaide Hospital, Adelaide, South Australia, Australia

SUMMARYMortality and cost outcomes of elderly intensive care unit (ICU) trauma patients were characterised in a retrospective cohort study from an Australian tertiary ICU. Trauma patients admitted between January 2000 and December 2005 were grouped into three major age categories: aged ≥65 years admitted into ICU (n=272); aged ≥65 years admitted into general ward (n=610) and aged <65 years admitted into ICU (n=1617). Hospital mortality predictors were characterised as odds ratios (OR) using logistic regression. The impact of predictor variables on (log) total hospital-stay costs was determined using least squares regression. An alternate treatment-effects regression model estimated the mortality cost-effect as an endogenous variable. Mortality predictors (P ≤0.0001, comparator: ICU ≥65 years, ventilated) were: ICU <65 not-ventilated (OR 0.014); ICU <65 ventilated (OR 0.090); ICU age ≥65 not-ventilated (OR 0.061) and ward ≥65 (OR 0.086); increasing injury severity score and increased Charlson comorbidity index of 1 and 2, compared with zero (OR 2.21 [1.40 to 3.48] and OR 2.57 [1.45 to 4.55]). The raw mean daily ICU and hospital costs in A$ 2005 (US$) for age <65 and ≥65 to ICU, and≥65 to the ward were; for year 2000: ICU, $2717 (1462) and $2777 (1494); hospital, $1837 (988) and $1590 (855); ward $933 (502); for year 2005: ICU, $3202 (2393) and $3086 (2307); hospital, $1938 (1449) and $1914 (1431); ward $1180 (882). Cost increments were predicted by age ≥65 and ICU admission, increasing injury severity score, mechanical ventilation, Charlson comorbidity index increments and hospital survival. Mortality cost-effect was estimated at -63% by least squares regression and -82% by treatment-effects regression model. Patient demographic factors, injury severity and its consequences predict both cost and survival in trauma. The cost mortality effect was biased upwards by conventional least squares regression estimation.

key Words: trauma, elderly, intensive care, injury severity score, costs, endogenous variable, treatment-effects model

Anaesth Intensive Care 2009; 37: 773-783

774 L. Y. L. Chan, J. L. Moran et aL


treatment of the ICU elderly trauma cohort. A recent retrospective analysis of data by Lee et al showed that in-hospital mortality rate of the elderly (defined as above 65 years) was twice that of younger patients; the elderly also had significantly higher rates of chest injuries that demanded a longer ICU stay9. A retrospective study on elderly trauma by Day et al found that age was an important factor in survival, but that survivors had good functional recovery10. The purpose of this study was to characterise: 1) the hospital mortality, and its determinants, of elderly trauma patients admitted to an Australian tertiary ICU and high dependency unit (HDU) and 2) the treatment cost of these patients allowing for the heterogeneous costing methodology that has been applied in intensive care setting11,12.

METHoDS

SettingThis was a retrospective cohort study conducted

at a major adult tertiary care trauma centre of South Australia (Royal Adelaide Hospital [RAH]). The hospital provides services in all the major medical and surgical specialties. The ICU is a mixed 24-bed medical and surgical unit, with an additional 10 HDU beds. The study was approved by the RAH Human Research Ethics Committee.

PatientsThe RAH electronic trauma registry was examined

for the period of January 2000 to December 2005. Patients entered into the trauma registry fulfilled the following criteria: all patients admitted to the hospital who had sustained injury(ies) falling within the ICD9-CM code range 800.0 to 959.9 as a result of a traumatic event (ICD10-AM code range S00-T98 excluding poisoning from 1 July 2000) and

died within the emergency department, or •were admitted to ICU/HDU, or •had an injury severity score (ISS) >16, or •patients to whom the trauma team was called, or •patients who were retrieved (by the RAH Medi- •flight Medical Team from South Australia and interstate).Identified trauma patients were grouped into three

major age categories: aged ≥65 years admitted into ICU/HDU; aged ≥65 years admitted into general ward and aged <65 years admitted into ICU/HDU. Elderly patients were defined as age of ≥65 years3,9.

Data collectionThe following data were obtained for all patients:

age, gender, ISS, duration of ventilation (in hours),

length of ICU and hospital stay (days), the Charlson comorbidity index (CCI) and outcome at hospital discharge (dead or alive). The CCI, which measures the underlying severity of comorbid illness, scores from 1 to 6 for 19 specific medical diagnoses, representing increasing levels of illness13. The CCI was computed by examination of the electronic administrative data listings for the ICD-10 codes of each patient and subsequent processing by the Stata™ module “Charlson”14,15. Binary (scored 0,1) categorical variables were generated to reflect patient admission on Saturday or Sunday (“weekend”), admission between 1800 and 0600 hours the following day (“after-hours”) and after-hours not at the weekend (“afterhours-weekday”). Acute physiological and chronic health evaluation scores were unable to be utilised due to incomplete data collection.

Information on the cost of hospital stay (≡total costs) of the patients was gathered from the RAH Finance Department, Case Mix and Clinical Costing Unit. The sum total costs consisted of the drug, medical/surgical, prosthesis, operating theatre and recovery, allied health, pathology, imaging, ICU, medical salaries/wages, nursing salaries/wages, indirect overhead and miscellaneous costs (see Appendix 2 on the online version).

South Australian health consumer price indices (CPI)16 and Australian versus US dollar yearly average exchange rates17 over the time of the study were obtained from the Australian Bureau of Statistics. Yearly health CPIs were (2000 to 2005): 167.6, 169.3, 178.1, 193.3, 204.8 and 214.4 respectively; yearly average exchange rates were (2000 to 2005): 0.5379, 0.5239, 0.5848, 0.7137, 0.7529 and 0.7474 respectively.

Statistical analysisVariables were reported as mean (SD) unless

otherwise indicated. Pearson’s chi-squared was used for the categorical variables and analysis of variance was performed by the kruskal-Wallis non-parametric test; for summarising raw variable time-change, simple non-parametric trend tests were used. Distributions of variables of interest were visualised by trellis plots, based on a formula with the structure y~x a×b, where y is a continuous or factor variable, x is continuous, and a and b are factors18.

Hospital mortality was modelled using logistic regression. ISS and calendar year were considered as continuous variables and candidate categorical predictors were parameterised as simple indicator variables. Non-linear variable effect was demonstrated using fractional polynomials19.

775MortaLitY and Cost outCoMes of eLderLY trauMa patients


Clinically meaningful combinations of variables and their (two-way) interactions were assessed for effect in the logistic model20; model adequacy was gauged by the traditional criteria of discrimination (receiver operating characteristic [RoC] area) and calibration (Hosmer-Lemeshow [H-LĈ] statistic)21. The final model was developed on training (75% of data) and determination sets (25% of data); the random samples being stratified by calendar year. The model chi-squared was calculated for each parameter to adjudge the relative ‘importance’ of the parameter22.

The total cost of care and the mean daily cost of care (over the period of stay in ICU and hospital) were calculated; this was expressed as either costs per current year or adjusted23 by health CPI to 2005 Australian dollars, as indicated in the text. Cost predictors for total hospital costs (as A$ 2005) were determined by ordinary least squares regression analysis using log transformed costs and comparison with a raw-cost regression model was also under-taken (including the concordance of predicted and raw cost24). Final R2, as the squared correlation of predicted versus raw costs, was reported on the back-transformed ‘dollar’ scale, as the expectation was that regression with log transformed dependent variable would yield a higher R2 than using raw costs25. Stability of both the mortality cost model was determined by a split-sample technique (75% determination, 25% validation sub-sets); overall model parsimony was sought.

With respect to the analysis of the impact of outcome (dead or alive) on cost-of-care; mortality outcome, considered as an independent variable in a regression analysis, may have been subject to “endogenous variable bias”26. Hence the cost model was re-estimated using a treatment-effects model27, where specific allowance is made for the correlation of independent predictors and the error terms.

Further details of the statistical approaches are provided in Appendix 1 (see online version). Statistical significance was ascribed at the 0.05 level. Analyses were performed using the Stata™ statistical software (V10.1, 2008: Stata Corporation, College Station TX, USA).

RESULTS

Demographics: patient characteristics and outcomesDuring the study period, January 2000 through

December 2005, there were 1617 patients aged <65 years admitted into the ICU, 272 patients aged ≥65 years admitted into ICU and 610 patients aged ≥65 years admitted to the hospital wards. The ISS of patients admitted to ICU, ≥65 years and

<65 years of age, was not significantly different (P=0.29); but, not surprisingly, the ISS of elderly patients admitted to the ward was significantly lower (P=0.001). Both groups admitted to ICU were ventilated in similar proportions but not for the same time period (P=0.0004). Raw hospital mortalities for the groups created by the interaction of age- category×ventilation status were: ICU age <65 years non-ventilated, 3.1%; ICU, age <65 years ventilated, 17.8%; ICU age ≥65 years non- ventilated, 11.3%; ICU age ≥65 years ventilated, 61.5%; ward: age ≥65 years non-ventilated, 4.1%. A modest decline in the overall proportion of patients aged <65 years occurred over the time of the study, but not for the proportion ventilated (non-parametric trend test; P=0.014 and P=0.25 respectively). A small decrease (1% per year) in mean ISS was also demonstrated over time (least squares regression, P=0.001). Additional patient demographics are shown in Table 1.

tabLe 1Demographic and outcome details of patients in the three

patient categories

Descriptor variable ICU: age <65 years

ICU: age ≥65 years

Ward: age ≥65 years

n 1617 272 610

Age, y: mean (SD) 34.1 (13.6) 75.6 (7.1) 76.7 (7.8)

Gender (% male) 79.4 66.9 55.9

ISS: mean (SD) 21.3 (13.8) 21.7 (14.0) 8.3 (6.3)

Ventilator hours

n 770 122 N/A

mean (SD) 114.4 (170.0) 157.9 (225.6) N/A

median (IQR) 40 (129) 62 (181) N/A

Charlson index score

0 1457 182 458

1 114 58 93

2 46 32 59

Length of stay

ICU (days)

mean (SD) 4.5 (8.2) 4.8 (9.9) N/A

median (IQR) 1.1 (5.3) 1.2 (4.9) N/A

Hospital (days)

mean (SD) 20.8 (28.8) 22.4 (26.6) 11.7 (15.2)

median (IQR) 11 (21) 12.5 (24) 6.5 (12)

Mortality (%) 10.8 33.8 4.1

N/A=not applicable. The ISS score is the injury severity score derived from the sum of the squares of the three highest abbreviated injury scores (range 1-75).



Hospital mortality outcome: logistic regression modelThe data-set was not rich with potential

predictors; the final model included (centred) calendar year as a linear continuous variable, ISS as a 0.5 fractional polynomial (see Appendix 1, “Statistical appendix” and Figure 1, left panel), CCI (scored 0, 1 and 2) and the interacted categorical predictors of [age-category×ventilation status]. The development set (n=1876) had an RoC area of 0.89; with P value for H-LĈ (=13.65) of 0.09. For the validation set (n=623), RoC area was 0.88 and H-LĈ=7.35, P=0.69. The variance inflation factor and condition number for the final model were 2.74 and 12.9 respectively. Reported estimates were therefore generated on the whole data set

(RoC area=0.89; and H-LĈ=17.9, P=0.02) and the parameters of the final model, point estimates (odds ratios [oR]) with P values and 95% confidence interval and variable chi-squared values are displayed in Table 2. As the logistic model was primarily discriminative-explanatory in intention, the relatively poor final calibration status of the whole data-set-model was accepted28.

Mortality oRs were increased with increases in CCI and over calendar years and decreased for all age×ventilation-status combinations compared with ventilated ICU patients, aged ≥65 years. Neither gender nor time(s) of admission demonstrated a mortality effect (P ≥0.2) and no significant interactions were demonstrated between 1) ISS

tabLe 2Parameter estimates of final logistic regression mortality model

Predictors oR P 95% CI: ll 95% CI: ul Chi-squared

Age <65 ICU, non-ventilated 0.014 0.0001 0.008 0.026 102.8

Age <65 ICU, ventilated 0.090 0.0001 0.055 0.148 45.4

Age ≥65 ICU, non-ventilated 0.061 0.0001 0.030 0.125 0.01

Age ≥65 ward, non-ventilated 0.086 0.0001 0.047 0.157 49.2

ISS: F-P 20.212 0.0001 13.318 30.676 329.8

Calendar year 1.127 0.008 1.032 1.230 0.5

CCI score 1 2.206 0.001 1.400 3.480 6.8

CCI score 2 2.567 0.001 1.448 4.549 5.1

CI=confidence interval, ll=lower limit, ul=upper limit, CCI=Charlson comorbidity index, ISS: F-P=parameter estimates for 0.5 fractional polynomial of ISS, oR=odds ratio.Reference categories: age ≥65 years to ICU ventilated, CCI score 0; calendar year, centred.

figure 1: Left panel: Plot of ISS (horizontal axis) modelled as a (0.5) fractional polynomial against log odds ratio mortality (vertical axis) with 95% confidence interval. Right panel: Plot of ISS (horizontal axis) modelled as a (0.5, 3) fractional polynomial against log costs,

2005 Australian dollars (vertical axis) with 95% confidence interval.



fig

ur

e 2

: Tre

llis

(box

-plo

t) d

ispl

ay o

f raw

cos

ts tr

unca

ted

at $

200,

000

to fa

cilit

ate

disp

lay.

Ver

tical

axi

s: c

alen

dar

year

by

cost

pro

file.

Hor

izon

tal a

xis:

age

-cat

egor

y by

ven

tilat

ion

stat

us.



tabLe 3Mean daily (raw) cost per patient

Descriptor Year n A$ costs

ICU current year ICU $ 2005 Hospital current year Hospital $ 2005

iCu <65 Years 2000 255 2124 2717 1436 1837

ICU ≥65 years 2000 53 2171 2777 1243 1590

Ward ≥65 years 2000 82 729 933

ICU <65 years 2001 278 2127 2694 1448 1834

ICU ≥65 years 2001 46 2436 3085 1071 1356

Ward ≥65 years 2001 77 724 917

ICU <65 years 2002 260 2409 2900 1743 2098

ICU ≥65 years 2002 40 2284 2750 1274 1534

Ward >65 years 2002 92 745 897

ICU <65 years 2003 259 2651 2940 1707 1893

ICU ≥65 years 2003 43 2630 2917 1883 2089

Ward ≥65 years 2003 92 773 857

ICU <65 years 2004 243 2747 2876 1771 1854

ICU ≥65 years 2004 38 2831 2964 1697 1777

Ward ≥65 years 2004 116 1020 1068

ICU <65 years 2005 322 3202 3202 1938 1938

ICU ≥65 years 2005 52 3086 3086 1914 1914

Ward ≥65 years 2005 151 1180 1180

Current year, cost as per each year. $ 2005, health CPI adjusted costs to calendar year 2005.

tabLe 4Regression analysis of total costs (A$ 2005): log-transformed fractional polynomial (0.5, 3) model and treatment-effects model

FP model estimates Treatment effect estimates

Predictor variables

Coefficient 95% CI P % change* Coefficient 95% CI (lower, upper)

P % change*

Age <65 to ICU -0.191 -0.323, -0.059 0.0001 -17.6% -0.355 -0.502 ,-0.2070 0.0001 -29.9%

Age ≥65 to ward -0.764 -0.918, -0.610 0.0001 -53.6% -0.856 -1.016, -0.696 0.0001 -57.6%

Gender -0.096 -0.184,-0.008 0.033 -9.2% -0.096 -0.186, -0.007 0.035 -9.3%

Ventilation 0.189 0.097, 0.281 0.0001 20.7% 0.307 0.203, 0.410 0.0001 35.8%

ISS 1.740† 1.617, 1.863 0.0001 1.821† 1.693, 1.950 0.0001

-0.006† -0.007, -0.004 0.0001 -0.004 -0.006, -0.003 0.0001

CCI score 1 0.358 0.231, 0.485 0.0001 42.7% 0.391 0.261, 0.521 0.0001 47.5%

CCI score 2 0.809 0.636, 0.981 0.0001 123.6% 0.857 0.681, 1.034 0.0001 134.8%

outcome -0.988 -1.130, -0.846 0.0001 -62.9% -1.713 -2.020, -1.407 0.0001 -82.2%

Calendar year 0.024 0.002, 0.0462 0.034 2.4% 0.031 0.008, 0.053 0.008 3.1%

Cons 10.311 10.166, 10.456 0.0001

R2‡ 0.42 0.40

R2§ 0.22 0.20

Concordance 0.39 (0.36, 0.42) 0.42 (0.39, 0.45)

CI=confidence interval, ISS=injury severity score, CCI=Charlson comorbidity index, FP=fractional polynomial. Reference categories: age ≥65 years to ICU; female, not-ventilated; CCI=0; alive. * percentage change in dependent variable per unit change in the variable. † parameter estimates for the 0.5,3 fractional polynomial of ISS, % change not reported. R2‡=R2 estimated on the ‘log-dollar’ scale, R2 §=R2 estimated on the ‘dollar’ scale. Concordance: concordance (95% CI)24 between predicted and raw costs (‘dollar’ scale).



and mechanical ventilator status (P=0.30) or CCI (P=0.15) and 2) calendar year and age×ventilation-status combinations (P ≥0.07). Further stratified outcomes are presented in Appendix 1, “Results, Hospital mortality model”.

Cost modelThe cost patient cohort comprised 2499 patients,

with overall mean cost of $28,811 (44,011) (median $13,394, range 263 to 496,850); A$ 2005 costs were: mean $36,331 (53,594), median $16,872 (range 336 to 663,406). A trellis-plot of the total cost distribution (unadjusted for health CPI), by calendar year, over age-category by ventilation status is seen in Figure 2. Ventilated elderly patients sustained increased costs compared with the non-ventilated sub-cohorts and

there were variable cost increments over the time period of the study. Average daily ICU and hospital costs, 2000 to 2005, are seen in Table 3. When adjusted for health CPI (‘A$ 2005’ costs), relatively modest and irregular increments occurred in both ICU and hospital costs.

Covariate cost-effects from the log-cost regression analysis are seen in Table 4. Statistical advantage for modelling the effect of ISS by a fractional polynomial (0.5, 3; Figure 1, right panel) was demonstrated in the determination set (R2=0.42) and the polynomial effect as (0.5, 3) was sustained in the validation set (R2=0.41); final coefficients were reported for the whole data set (R2=0.42; Table 5). A small (2.4%) yearly cost increment was evident; the year effect being modelled as a

tabLe 5Mean total raw and predicted costs (A$ 2005) by calendar year and age category

Year Raw costs

ICU: age <65 years ICU: age ≥65 years Ward: age ≥65 years

Alive Died Alive Died Alive Died

2000 36703 37863 39073 32295 12045 12447

2001 49896 34986 52567 19191 12885 13513

2002 48908 42198 49864 53153 14566 5205

2003 48471 25604 39097 27242 15042 4347

2004 41678 27451 62270 26507 10983 2888

2005 46674 27151 48903 26484 13111 10829

Year FP log cost model



2000 46774 27706 51660 41578 12791 5380

2001 47915 30331 55281 37449 14425 12683

2002 48691 35543 49751 33700 15099 9291

2003 43451 27589 49021 35427 14004 9679

2004 42084 27423 67563 26598 10440 11009

2005 52164 27440 70996 36534 13142 12272

Year Treatment effects model



2000 49339 29129 55705 40169 13231 7208

2001 50147 32102 59908 34606 14894 13457

2002 50859 35250 54056 32066 15621 11021

2003 45444 28591 52307 33048 14430 10584

2004 43958 28980 71334 26459 10856 11454

2005 54232 29152 70548 32922 13678 12601

FP=fractional polynomial model (see Table 4).



linear continuous variable, there being no evidence of non-linearity. No interaction between calendar year and categorical variables, nor between categorical variables was apparent (P ≥0.59). On the ‘dollar’ scale, the final R2 of the log-cost model was 0.22. As expected, the log cost model was superior to the raw cost model in terms of residual analysis and the concordance of predicted and raw costs was superior (0.39 vs 0.32, respectively). Further details of cost model development are provided in Appendix 1, “Results, Cost model”.

Mean overall difference between those alive and dying, for raw and predicted (A$ 2005) costs (fractional polynomial log-cost model), was estimated at $7688 and $9841 respectively. In the treatment-effects model (Table 4), modest changes occurred in the parameter estimates compared with the fractional polynomial model, with the difference in the percentage change for the outcome effect (dead vs alive) being 19.3% (-82.2% vs -62.9%, respectively). The treatment-effects model was well specified with respect to residual analysis and there was minimal evidence of heteroscedasticity (non- constant variance; see also Appendix 1, “Results, Cost model”). Mean overall difference, alive vs dead, was $11,142 (A$ 2005).

Mean cost comparisons (alive vs died, estimated for baseline categories (Table 4), A$ 2005), for fractional polynomial least squares regression versus treatment-selection model and referenced to raw cost, over calendar year and between age categories, are seen in Table 5. Predicted costs for surviving patients generally exceeded cost for those dying and the differential was increased for the treatment -effects model. Cumulative differences, alive vs died, over year and age categories, were: raw costs, $213,382; fractional polynomial model, $257,620; and treatment-effects model, $291,748.

DISCUSSIoN

Mortality outcomeThe adverse impact of age upon mortality-

outcome7,29 was confirmed in the current study of elderly patients admitted into intensive care. The observed mortality rate observed was consistent with the findings of the recent National Trauma Registry, with an estimated 30% elderly mortality at hospital discharge following a major trauma (ISS >15), and with reports published in North America and Europe. Despite higher mortality rates in the elderly compared with the younger age group, these rates were comparable with other disease outcomes such as severe sepsis30.

In agreement with the Major Trauma outcome Study (MToS), ISS increments were associated with increased mortality31, which was not a linear function of the ISS (Figure 1), a finding similar to the modelled effect of the acute physiological and chronic health evaluation III score on mortality outcome32. The injury severity score has proved popular over time despite deficiencies when there is more than one injury per body system. More insightful would have been a detailed breakdown of the injury severity scores to that of individual body systems and analysis of the relationship with overall mortality. It has been shown that head injury in the elderly, compared with younger patients, is associated with decreased surgical intervention and higher mortality33; similarly, rib fractures are also associated with increased risks of death34. Further studies are needed to look at the differential impact of specific organ injuries on the management and survival of the elderly.

The impact of CCI on prognosis in the spectrum of critically ill patients has been variable35. The limited range of the CCI scores observed in the current study were in general agreement with Gabbe et al36 and may have simply represented the characteristics of the cohort of elderly patients; that is, those in better health, being of increased mobility, were more likely to suffer from trauma. of interest, ‘coagulopathy’, which may be important in the trauma setting, does not contribute to the CCI components37. The number of patients prescribed anticoagulants in Australia is increasing at a yearly rate of 9%38. It is possible that over the ensuing years, with the increase use of antiplatelet therapy and anticoagulants, bleeding risks in this cohort of trauma patients will increase significantly and have greater impact on outcomes.

A requirement for artificial ventilation predicted mortality (Table 2) and in the age category ≥65 years, the ventilation hours were also notably longer. Mechanical ventilation is known to be a predictor of mortality in various other disease states, more so in the elderly39. Gender, on the other hand, had no demonstrable effect on survival in our study. While experimental data seem to support the influence of gender on disease processes, with the female gender conferring benefit, clinical evidence in trauma patients has been conflicting40. The reasons for the modest mortality increment over time were not immediately obvious, there being no significant interaction between calendar year and the age×ventilation-status combinations (see “Results, Hospital mortality outcome: logistic regression model” above) and the ISS demonstrated a small decrease with time. The most plausible explanation



would appear to be the decrease in the overall proportion of patients aged <65 years over the study period.

CostsA ‘top-down’ costing of care was used as a broad

assessment of resource utilisation. There were 13 identified cost-buckets that encompassed both cost not directly related to the individual patient care (but necessary for daily maintenance of services) and that which was directly related to cost incurred in care. This method, though not standardised across hospitals in Australia, has been advocated and applied in other studies as a means to uniformly define and compare ICU expenditure12,41. While data on ICU trauma costing is scarce, this study has revealed some interesting comparisons. The total and average daily cost of care for patients in the ICU age categories were comparable with other Australian studies. Rechner et al and Moran et al found that the average daily costs of ICU treatment were $2670 (calendar years 2002 to 2003) and $2395 (calendar year 2002) respectively11,12. The overall (2000 to 2005) mean daily treatment costs in our study were similar (Table 3), albeit the acquisition of treatment cost were calculated differently. The significant cost contributors, for example, drugs, pathology and radiology noted by Rechner et al were not included in the ICU costing (see Appendix 2 on the online version) but were instead factored into the ‘total hospital bill’. Trauma patients would have incurred much higher operating theatre cost, once again captured only in the total hospital cost. In Australia, blood transfusion is funded by the Australian Red Cross, trauma patients being a major consumer, and this may have under-estimated cost when compared with other jurisdictions. Although international comparisons may reflect different treatment-cost acquisition, the above reported costs would appear to be less than the US$ (2004) daily ICU cost of $2575 (≡A$3420) reported by Milbrandt et al23. of interest, the latter authors reported stable ICU daily costs over the period 1994 to 2004, whereas the present study, albeit in trauma patients, found an annual increment (in A$ 2005) of 2.4%, against a background of an annual increase in ICU length of stay of 0.83%.

The factors that increased mortality were also those factors that increased the cost of care, while ‘death’ in itself somewhat paradoxically limited further escalation of cost, although the interaction between ‘outcome’ and ‘Age-category’ was non- significant (global Wald test, P=0.31). This observation differs from that found by Seguin et al, where total expenditure on non-survivors was greater42.

Critique of methodologyThe current study derived its patient cohort and

cost-structure from a single tertiary referral centre, and obviously inferences may not be relevant to other jurisdictions. Although long-term follow-up was not established, the restriction of follow-up to hospital discharge was of advantage in that a correspondence was established between variables predicting (hospital) outcome and costs. Long-term functional outcomes, increasingly recognised as meaningful outcome measures, were also not measured. Intensive care obviously serves the acute-care environment, but it is appropriate that the cost of protracted rehabilitation, any decrement in quality of life and other social-economic costs be considered in the period that follows ‘successful’ intensive care. Age has always been modelled in health economics as having higher impact upon future cost. This supposition has been recently re- assessed with the demonstration that the time-period proximate to death is the period of substantive cost escalation (ICU care being a part thereof)43.

The final cost model had an R2 on the dollar scale of 0.22, which was modest, but comparable with other studies and overviews11,44,45. The mortality effect of outcome on cost in the log-cost least squares model was biased upwards (-62.9% vs -82.2%) with respect to that of the treatment-effects model, which addressed the endogeneity of ‘outcome’ in the least squares regression (identified by the correlation of the error terms of the separate determination of costs and outcome in the treatment-effects regression; see Appendix 1 on the online version). This approach may thus be preferred when binary endogenous ‘outcome’ variables are used as independent regression variables46,47; an analytic direction previously comment upon39. This being said, the overall predicted cost difference, alive versus dead, between the two approaches was modest (Table 5).

CoNCLUSIoNThose factors decreasing survival time were also

associated with cost increments, although mortality per se was associated with an overall cost decrease. Costs increased minimally over the calendar year span of the study. Cost methodology must appropriately address endogenous variable bias.

ACkNoWLEDGEMENTSThe authors are grateful to Ms Dianne Liebelt,

the team from the RAH Finance Department and the Trauma Registry for their contributions.



REFERENCES 1. Australian Bureau of Statistics. Year Book Australia. From

www.abs.gov.au/AUSSTATS/abs@nsf/allprimarymainfeatures/714408F4FEAB4FAECA25726C00772177?opendocument 2006.

2. Australian Bureau of Statistics. Causes of Death, Australia, 2004. From www.abs.gov.au/AUSSTATS/abs@nsf/DetailsPage/3303 02004?openDocument 2004.

3. National Trauma Registry Consortium (Australia and New Zealand). The National Trauma Registry (Australia and New Zealand) Report: 2002. From www.uq.edu.au/ntrc/publica-tions/Report2002.pdf 2004.

4. DeMaria EJ, kenney PR, Merriam MA, Casanova LA, Gann DS. Aggressive trauma care benefits the elderly. J Trauma 1987; 27:1200-1206.

5. Grossman M, Scaff DW, Miller D, Reed J III, Hoey B, Anderson HL III. Functional outcomes in octogenarian trau-ma. J Trauma 2003; 55:26-32.

6. Gubler kD, Davis R, koepsell T, Soderberg R, Maier RV, Rivara FP. Long-term survival of elderly trauma patients. Arch Surg 1997; 132:1010-1014.

7. Nagy kk, Smith RF, Roberts RR, Joseph kT, An GC, Bokhari F et al. Prognosis of penetrating trauma in elderly patients: a comparison with younger patients. J Trauma 2000; 49:190-193.

8. Taylor MD, Tracy Jk, Meyer W, Pasquale M, Napolitano LM. Trauma in the elderly: intensive care unit resource use and outcome. J Trauma 2002; 53:407-414.

9. Lee WY, Cameron PA, Bailey MJ. Road traffic injuries in the elderly. Emerg Med J 2005; 23:42-46.

10. Day RJ, Vinen J, Hewitt-Falls E. Major trauma outcomes in the elderly. Med J Aust 1994; 160:675-678.

11. Moran JL, Peisach AR, Solomon PJ, Martin J. Cost calculation and prediction in adult intensive care: a ground-up utilization study. Anaesth Intensive Care 2005; 32:787-797.

12. Rechner IJ, Lipman J. The costs of caring for patients in a ter-tiary referral Australian intensive care unit. Anaesth Intensive Care 2005; 33:477-482.

13. Charlson ME, Pompei P, Ales kL, Mackenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987; 40:373-383.

14. Quan H, Sundararajan V, Halfon P, Fong A, Burnand B, Luthi J-C et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care 2005; 43:1130-1139.

15. Stagg V. CHARLSoN: Stata module to calculate Charlson index of comorbidity. From econpapers.repec.org/scripts/search asp?ft=charlson 2006 Accessed January 2007.

16. Australian Bureau of Statistics. 6401.0 – Consumer Price Index, Australia, September 2007. From www.abs.gov.au/ AUSSTATS/abs@nsf/second+level+view?ReadForm& prodno=64010&viewtitle=Consumer%20Price%20Index, %20Australia~Sep%202007~Latest~24/10/2007&&tab name=Past%20Future%20Issues&prodno=64010&issue= Sep%202007&num=&view=&2007 Accessed December 2007.

17. Australian Bureau of Statistics. 1350.0 – Australian Economic Indicators, December 2007. From www.abs.gov.au/AUSSTATS/abs@nsf/DetailsPage/13500Dec%202007?openDocument 2007 Accessed December 2007.

18. Mander A. TRELLIS: Stata module to plot explanatory vari-ables vs binary outcome. From econpapers.repec.org/scripts/search asp?ft=trellis 2006.

19. Royston P, Sauerbrei W. Multivariable model-building: A pragmatic approach to regression analysis based on fractional polynomials for modelling continuous variables. John Wiley & Sons Ltd, Chichester, West Sussex 2008.

20. Royston P, Ambler G, Sauerbrei W. The use of fractional poly-nomials to model continuous risk variables in epidemiology. Int J Epidemiol 2000; 28:964-974.

21. Hosmer DW, Lemeshow S. Applied Logistic Regression. 2nd ed. John Wiley & Sons Inc, New York 2000.

22. knaus WA, Wagner DP, Zimmerman JE, Draper EA. Variations in mortality and length of stay in intensive care units. Ann Intern Med 1993; 118:753-761.

23. Milbrandt EB, kersten A, Rahim MT, Dremsizov TT, Clermont G, Cooper LM et al. Growth of intensive care unit resource use and its estimated cost in Medicare. Crit Care Med 2008; 36:2504-2510.

24. Moran JL, Solomon PJ, Peisach AR, Martin J. New models for old questions: generalized linear models for cost prediction. J Eval Clin Pract 2007; 13:381-389.

25. Scott A, Wild C. Transformations and R2. The American Statistician 1991; 45:127-129.

26. Briscoe J, Akin J, Guilkey D. People are not passive acceptors of threats to health: endogeneity and its consequences. Int J Epidemiol 1990; 19:147-153.

27. Wooldridge JM. Limited dependent variable models and sam-ple selection corrections. In: Mason oH, Introductory econo-metrics: A modern approach. Thomson South-Western 2006. p. 582-622.

28. Harrison DA, Brady AR, Parry GJ, Carpenter JR, Rowan k. Recalibration of risk prediction models in a large multicenter cohort of admissions to adult, general critical care units in the United kingdom. Crit Care Med 2006; 34:1378-1388.

29. Richmond TS, kauder D, Strumpf N, Meredith T. Characteristics and outcomes of serious traumatic injury in older adults. J Am Geriatr Soc 2002; 50:215-222.

30. Angus DC, Linde-Zwirble WT, Lidicker J, Clermont G, Carcillo J, Pinsky MR. Epidemiology of severe sepsis in the United States: analysis of incidence, outcome, and associated costs of care. Crit Care Med 2001; 29:1303-1310.

31. Champion HR, Copes WS, Sacco WJ, Lawnick MM, keast SL, Bain LW Jr et al. The Major Trauma outcome Study: establish-ing national norms for trauma care. J Trauma 1990; 30:1356-1365.

32. knaus WA, Wagner DP, Draper EA, Zimmerman JE, Bergner M, Bastos PG et al. The APACHE III prognostic system. Risk prediction of hospital mortality for critically ill hospitalized adults. Chest 1992; 100:1619-1636.

33. Ritchie PD, Cameron PA, Ugoni AM, kaye AH. A study of the functional outcome and mortality in elderly patients with head injuries. J Clin Neurosci 2000; 7:301-304.

34. Bergeron E, Lavoie A, Clas D, Moore L, Ratte S, Tetreault S et al. Elderly trauma patients with rib fractures are at greater risk of death and pneumonia. J Trauma 2003; 54:478-485.

35. Needham DM, Scales DC, Laupacis A, Pronovost PJ. A system-atic review of the Charlson comorbidity index using Canadian administrative databases: a perspective on risk adjustment in critical care research. J Crit Care 2005; 20:12-19.

36. Gabbe BJ, Magtengaard k, Hannaford AP, Cameron PA. Is the Charlson Comorbidity Index useful for predicting trauma outcomes? Acad Emerg Med 2005; 12:318-321.

37. Franko J, kish kJ, o'Connell BG, Subramanian S, Yuschak JV. Advanced age and preinjury warfarin anticoagulation increase the risk of mortality after head trauma. J Trauma 2006; 61:107-110.



38. Australian Pharmaceutical Index. IMS Health, Sydney 2006.39. Moran JL, Bristow P, Solomon PJ, George C, Hart Gk.

Mortality and length-of-stay outcomes, 1993-2003, in the bina-tional Australian and New Zealand intensive care adult patient database. Crit Care Med 2008; 36:46-61.

40. George RL, McGwin G Jr, Metzger J, Chaudry IH, Rue LW 3rd. The association between gender and mortality among trauma patients as modified by age. J Trauma 2003; 54:464-471.

41. Edbrooke DL, Ridley SA, Hibbert CL, Corcoran M. Variations in expenditure between adult general intensive care units in the Uk. Anaesthesia 2001; 56:208-216.

42. Seguin J, Garber BG, Coyle D, Hebert PC. An economic evaluation of trauma care in a Canadian lead trauma hospital. J Trauma 1999; 47:S99-103.

43. Werblow A, Felder S, Zweifel P. Population ageing and health care expenditure: a school of ‘red herrings’? Health Econ 2007; 16:1109-1126.

44. Diehr P, Yanez D, Ash A, Hornbrook M, Lin DY. Methods for analyzing health care utilization and costs. Annu Rev Public Health 1999; 20:125-144.

45. Dudley RA, Harrell FE Jr, Smith LR, Mark DB, Califf RM, Pryor DB et al. Comparison of analytic models for estimat-ing the effect of clinical factors on the cost of coronary artery bypass graft surgery. J Clin Epidemiol 1993; 46:261-271.

46. Angus DC, Linde-Zwirble WT, Sirio CA, Rotondi AJ, Chelluri L, Newbold RC 3rd et al. The effect of managed care on ICU length of stay: implications for medicare. JAMA 1996; 276:1075-1082.

47. Rapoport J, Teres D, Zhao Y, Lemeshow S. Length of stay data as a guide to hospital economic performance for ICU patients. Med Care 2003; 41:386-397.

48. Harrell FE Jr. overview of maximum likelihood estimation. In: Regression modelling strategies: with applications to linear models, logistic regression, and survival analysis. New York: Springer-Verlag 2001. p. 179-212.

49. Goldstein R. Interpretations of dummies in regressions with log dependent variables. Stata Technical Bulletin Reprints 1992; 1:183-184.

50. Greene WH. Heteroscedasticity. In: Econometric Analysis, 4th. Upper Saddle River, NJ: Prentice-Hall Inc. 2000. p. 499-524.

51. Duan N. Smearing estimate: A nonparametric retransforma-tion method. Journal of the American Statistical Association 1983; 78:605-610.

52. Cheung YB. Adjustment for selection bias in cohort studies: an application of a probit model with selectivity to life course epidemiology. J Clin Epidemiol 2001; 54:1238-1243.

53. Maddala GS. Multivariate qualitative variables. In: Limited-Dependent and Qualitative Variables in Econometrics, 1. Cambridgeshire: Cambridge University Press 1983. p. 93-148.

54. Barnow BS, Cian GG, Goldberger AS. Issues in the analysis of selectivity bias. In: Evaluation Studies Review Annual, 1. Beverly Hills: Sage 1981. p. 43-59.

55. Cong R, Drukker DM. Treatment-effects model. Stata Technical Bulletin Reprints 2000; 10:159-169.

elderly trauma patients

Documents