Injury, Int. J. Care Injured (2007) 38, 318—328

www.elsevier.com/locate/injury

A mathematical program to predict survival and tosupport initial therapeutic decisions for traumapatients with long-bone and pelvic fractures

Kevin Lu, William C. Shoemaker *, Charles C.J. Wo,Jackson Lee, Demetrios Demetriades

Departments of Surgery and Orthopedics, LAC+USC Medical Center, Keck School of Medicine,University of Southern California, 1200 N State Street, Los Angeles, CA 90033, USA

Accepted 20 June 2006

KEYWORDSStochastic analysis andcontrol program;Outcome prediction;Therapeutic decisionsupport system;Noninvasivehaemodynamicmonitoring;Thoracic bioimpedanceestimation of cardiacoutput;Pulse oximetry;Transcutaneous oxygenand carbon dioxidetensions;Tissue perfusion

Summary

Aim: To test a mathematical program to monitor early haemodynamic patterns ofpatients with fractures, predict survival and support initial therapeutic decisions.Methods: A mathematical search and display program based on non-invasive hae-modynamic monitoring was used to study 430 consecutively monitored patients withfractures during the first 48 h after admission to the emergency department of aninner city public hospital. We studied four types of fractures: simple extremityfractures, long-bone fractures, pelvic fractures and fractures incidental to severetrauma. The program continuously displayed haemodynamic patterns and predictedsurvival probability (SP), which was evaluated by the actual outcome at hospitaldischarge. The program also assessed the effectiveness of therapies according tohaemodynamic responses.Results: The cardiac index, heart rate, mean arterial pressure, arterial saturationand transcutaneous oxygen tensions at the initial resuscitation were significantlyhigher in survivors than in non-survivors. After the first 48 h, the haemodynamicpatterns were more influenced by fever, sepsis, complications and organ failures. Thecalculated survival probability averaged 81% � 18% in the first 48 h for survivors and72% � 20% for non-survivors.Conclusion: Early continuous non-invasive haemodynamic monitoring using the pro-posed information system is helpful in predicting outcome and guiding therapy forpatients with fractures.# 2006 Elsevier Ltd. All rights reserved.

* Corresponding author. Tel.: +1 323 226 7784; fax: +1 323 226 7784.E-mail address: wcshoemaker00@hotmail.com (W.C. Shoemaker).

0020–1383/$ — see front matter # 2006 Elsevier Ltd. All rights reserved.doi:10.1016/j.injury.2006.06.117

Therapeutic decisions for trauma patients with long-bone and pelvic fractures 319

Introduction

Invasive pulmonary artery (PA) thermodilution(Swan-Ganz) catheters provide the maximum cir-culatory data at the bedside, but require intensivecare unit (ICU) conditions. There was no advan-tage found for the PA catheter with goal-directedtherapy when this invasive monitoring was startedlate in the course of illness, i.e. >24 h afteradmission to the Emergency Department (ED), orafter onset of an organ failure.9,14,11,26,1 Delays incorrecting circulatory deficiencies have resulted

Figure 1 A 40-year-old woman was struck by an automobsection, cardiac index (CI); second section, heart rate; thiroximetry; fifth section, transcutaneous oxygen tension indexeFiO2); lowest section, survival probability. The woman was rcells. CI rose from 2.5 l/min/m2 shortly after admission to 5 l/normal and SP was maintained at 0.80 or above. The woman

in organ failures and death.7,21,8,33,15 However,badly injured patients who arrive in severe shockand older patients have survival benefit whenmanaged with a PA catheter.13 Although manyarticles have pointed out the hazards and harmfuleffects of associated trauma on the patients withpelvic and long-bone injury,19,18,31,12 we wereunable to find data in the literature on evaluationand decision-making in orthopaedic trauma casesusing early haemodynamic monitoring. Since timeis crucial in the resuscitation and management ofcritically ill emergency cases, early non-invasive

ile and sustained pelvic and tibial/fibular fractures. Topd section, mean arterial pressure; fourth section, pulsed to the fractional inspired oxygen concentration (PtcO2/esuscitated with 2.5 l crystalloids and 7 units packed redmin/m2 in the first hour, PtcO2/FiO2 gradually returned torecovered.

320 K. Lu et al.

monitoring with an outcome predictor is useful inidentifying and correcting haemodynamic andother changes as early as possi-ble.7,21,8,33,15,13,20,5,17,10,24

It is often difficult to separate the physiologicaleffects of fracture per se from the effects of asso-ciated trauma. As an attempt to probe this question,we evaluated and compared the survivors and non-survivors among 430 acute emergency patients withsingle extremity fractures, long-bone fractures, pel-vic fractures and fractures associated with othersevere trauma.

Figure 2 A 47-year-old woman sustained fractures of the fefrom a motor vehicle accident. Top section, cardiac index (CIpressure; fourth section, pulse oximetry; fifth section, transcuoxygen concentration (PtcO2/FiO2); lowest section, survival prFiO2 at normal values and SP at 0.75 to 0.80 or above. The w

Bayard et al.4,22 recently developed and tested amathematical probability analysis that used a largedatabase of non-invasively monitored haemody-namic variables to predict outcome early and toprovide therapeutic decision support for acuteemergency cases. The purposes of the presentreport are, first, to describe the early haemody-namic patterns of surviving and non-surviving ortho-paedic trauma cases and, second, to prospectivelytest the survival probability (SP) analysis madeshortly after resuscitation against the actual survi-val at the time of hospital discharge.

moral neck, ulna, humerus and multiple ribs on the right); second section, heart rate; third section, mean arterialtaneous oxygen tension indexed to the fractional inspiredobability. CI was maintained close to 4.0 l/min/m2, PtcO2/oman survived.

Therapeutic decisions for trauma patients with long-bone and pelvic fractures 321

Materials and methods

Study design

This was an observational study of 430 consecutivelymonitored orthopaedic trauma patients using con-tinuous non-invasive haemodynamic monitoring andan information system consisting of a mathematicalsearch and display analysis. Retrospectively, patientswere classified as having fractures of the extremity(excluding the femur), long bones (including thefemur), pelvis or fractures associated with othersevere trauma. Non-invasive monitoring was usuallybegun shortly after ED admission and the patient wasfollowed to the Radiology Department, theatre andthe ICU (Figs. 1 and 2). Most of these individuals,

Table 1 Clinical features of the series

Extremity fractures Long bone fractures

S (N = 28) S (N = 28) S (N = 100) NS (N = 15)

Age, years;mean � S.D.

36 � 18 48 � 21 50 � 21

Injury: fall,N (%)

4(14%) 8(8%) 1 (7%)

Gunshot wound,N (%)

7(25%) 7(7%) 1 (7%)

Blunt trauma,N (%)

15(54%) 85 (85%) 13 (86%)

Stab wound,N (%)

2(7%) 0 (0%) 0 (0%)

Bodily injury***

Head,N (%)

8 (44%) 20 (33%) 8 (67%)

Spinal cord,N (%)

0 (0%) 1 (2%) 0 (0%)

Chest,N (%)

5 (28%) 16 (26%) 1 (8%)

Abdomen, N (%) 5 (28%) 24 (39%) 3 (25%)

Organ failuresARDS 2 (7%) 22 (22%) 7 (47%)Hepaticfailure

2 (7%) 19 (19%) 4 (27%)

Renalfailure

0 (0%) 4(4%) 3 (20%)

Sepsis 3 (11%) 17 (17%) 5 (33%)Estimated

blood loss, ML1788 � 2319 1597 � 1683 * 2728 � 3274

Injury severityscore

14.4 � 9.7 19.4 � 10.5 ** 30.0 � 10.4

Glascow comascore

12.1 � 4.2 12.8 � 3.8 * 10.6 � 4.6

* P-value < 0.05.** P-value < 0.01.*** Not mutually exclusive.

except for thosewith simpleextremity fractures, hadother associated trauma of variable magnitude(Table 1). Patients with brain death or severe headinjuries and a Glasgow Coma Score (GCS) <8 wereexcluded because they have a different haemody-namic pattern22; the disruption of the central auto-nomic vasoconstriction centres after severe headinjury and brain death produces hyperdynamic hae-modynamic responses with high flow and augmentedtissue perfusion from unopposed peripheral vasodi-latory mechanisms.22 Invasive pulmonary arterycatheterisation was instituted to validate the non-invasive cardiac output estimations when clinicallyindicated after the patient arrived in the ICU. Table 1lists the salient clinical features. The Institution’sreview board approved the protocol.

Pelvic fractures Fractures with other asso-ciated severe trauma

S (N = 55) NS (N = 11) S (N = 218) NS (N = 38)

47 � 23 49 � 20 38 � 18 41 � 22

9(16%) 1 (9%) 20(9%) 3 (8%)

1(2%) 0 (0%) 47 (22%) 6 (16%)

45(82%) 10 (91%) 138 (63%) 27 (71%)

0 (0%) 0 (0%) 13 (6%) 2 (5%)

14 (22%) 4 (40%) 66 (26%) 21 (41%)

2 (3%) 0 (0%) 12 (5%) 1 (5%)

11 (17%) 1 (10%) 98 (38%) 14 (27%)

37 (58%) 5 (50%) 80 (31%) 14 (27%)

14 (25%) 7 (64%) 39 (18%) 20 (53%)14 (25%) 3 (27%) 30 (14%) 9 (24%)

1 (2%) 2 (18%) 4 (2%) 11 (29%)

9 (16%) 3 (27%) 44 (20%) 14 (37%)1118 � 1376 ** 2769 � 1781 1899 � 1765 ** 5365 � 8639

23.8 � 12.5 * 33.6 � 14.7 21.5 � 11.7 ** 36.8 � 12.6

12.6 � 3.8 11.3 � 4.6 12.2 � 4.0 ** 8.1 � 5.3

322 K. Lu et al.

Protocol for non-invasive haemodynamicmonitoring

Continuous non-invasive haemodynamic monitoringprovided on-line, real-time visual displays of cardiac,pulmonary and tissue perfusion patterns in acutely illemergency cases.24,4,22 These haemodynamic datawere continuously acquired, downloaded at 30-sintervals, averaged over 5-min intervals and enteredinto a computerised spreadsheet. When stable, the

Figure 3 Survivors’ (solid line) and non-survivors’ (dasheadmission. Mean values � S.E.M. are shown for cardiac indtranscutaneous oxygen tension indexed to the fractional inssurvivors’ values were generally higher than those of the non

haemodynamic data and SP were averaged over per-iods of 2—6 h for purposes of presentation (Fig. 3).

Cardiac output

An improved thoracic bioelectric impedance devicefor cardiac output measurements (IQ Model 101,Non-invasive Medical Technologies, Las Vegas, NV,or Physio-Flow, VasoCOM, Bristol, PA) was appliedshortly after arrival in the ED. The non-invasive

d line) temporal patterns for the first 48 h after theirex, heart rate, mean arterial pressure, pulse oximetry,pired oxygen concentration and survival probability. The-survivors.

Therapeutic decisions for trauma patients with long-bone and pelvic fractures 323

disposable pre-wired hydrogen electrodes werepositioned on the skin, and three ECG leads wereplaced across the praecordium and shoulders.23,32 A100-kHz, 4-mA alternating current was passedthrough the patient’s thorax by the outer pairs ofelectrodes and the voltage was sensed by the innerpairs of electrodes which captured the baselineimpedance (Zo), the first derivative of the impe-dance waveform (dZ/dt) and the ECG. The ECG andbioimpedance signals were filtered with an all-inte-ger-coefficient technology to decrease computa-tions and signal processing time. The signal-processing algorithm used a time-frequency distri-bution analysis that increased signal-to-noiseratios.32 Pulmonary artery catheters were placedwhen indicated by appropriate clinical criteria tovalidate the non-invasive cardiac output measure-ments. Previous studies have shown satisfactoryagreement between thermodilution and the IQbioimpedance cardiac output values for traumapatients in the ED, theatre and ICU.24,23

Pulse oximetry

Routine pulse oximetry (Nellcor, Pleasanton, CA)was used to assess arterial oxygen saturation(SapO2) continuously. Values were observed andrecorded at the time of cardiac index measure-ments. Sudden changes in these values were notedand confirmed by in vitro arterial oxygen saturationobtained by standard blood gas analysis.23

Transcutaneous oxygen tension

Conventional transcutaneous oxygen tension mea-surements (PtcO2) were indexed to the fractionalinspired oxygen concentration (PtcO2/FiO2) andcontinuously monitored throughout the observationperiod. This technology uses the Clark polarographicoxygen electrode routinely involved in standardblood gas measurements.16,25,6,27—30 Oxygen ten-sions were measured in a representative area ofthe skin surface heated to 44 8C to increase diffusionof oxygen across the stratum corneum and to avoidvasoconstriction in the area of skin being mea-sured.27 Previous studies have demonstrated thecapacity of PtcO2 to reflect tissue oxygen ten-sion,16,27—29 reflecting the delivery of oxygen tothe local area of skin; it also parallels the mixedvenous oxygen tension, except under terminal con-ditions where peripheral shunting leads to highmixed venous haemoglobin saturation (SvO2)values.30 Oxygen tension of a segment of the skinmay not reflect the state of oxygenation of alltissues and organs, but it has the advantage of beingthe most sensitive early warning tissue for the

adrenomedullary stress response. Cutaneous vaso-constriction is an early stress response to hypovolae-mia, shock and trauma syndromes.16,27—30

Transcutaneous oxygen tension was indexed to theFiO2 to give a PtcO2/FiO2 ratio, because changes ofthe inspired oxygen produce marked PtcO2 changes.The electrodemust bemoved to a nearby site and re-calibrated at 4-h intervals to avoid skin erythema.

The mathematical search and displayprogram

A search and display probability analysis programwas developed to estimate the individual patient’sSP from a database of patients with similar clinicaland haemodynamic states, defined in terms of theprimary diagnosis, covariates and haemodynamicvariables.22 Such a similar group of patients,referred to statistically as nearest neighbours, havea diagnosis, covariates and haemodynamic patternsvery close to those of the newly admitted patientunder study. The mathematical model is a searchand display approach for nearest neighbours who arethen used as surrogates for the study patient. Thestochastic analysis was developed from methods ofmachine learning25,6 and methods of dynamic pro-gramming.2,3 The program describes: the trajectoryof the patient’s course in three-dimension vectorsplotted over time; the first derivative of the initialvector projecting the patient’s course if there are noinherent changes or external influences; the secondderivative tracking change in the patient’s coursefrom either internal compensations, further dete-rioration, spontaneous improvement or externalinfluences such as changes in therapy; and theintegral summing the total influences.

Control input definition

The control input or mode of therapy is chosen froma finite set of therapies that are appropriate to thepatient.

Probability of survival

A patient’s SP was calculated by first extracting the40 (or more) nearest neighbour states; it is thencalculated as the fraction of these nearest neigh-bours who survived. The SP serves as a survivalpredictor as well as an objective digitised measureof the severity of illness.4

Therapeutic decision support system

The SP analysis of haemodynamic patterns was alsoused for therapeutic decision making based on data

324 K. Lu et al.

Figure 4 The upper half of the window, left side, shows data of a patient at 2.8 h after admission, the calculatedsurvival probability andmost recent therapy. Next section, first column, current haemodynamic values (CI, cardiac index;HR, heart rate; MAP, mean arterial pressure; SapO2, oxygen saturation; PtcO2/FiO2, PtcCO2, transcutaneous oxygentension indexed to the fractional inspired oxygen concentration), haematocrit value; second column, net cumulativeexcess (+) or deficit (�) of each variable up to this point in time. Lower half of the window, first column, number ofnearest neighbours who had received each therapy; second column, average SP of these nearest neighbours beforetherapy; third column, number of nearest neighbours who had received each of the specified therapies in columns 6—12;column 4, SP of these nearest neighbours after administration of therapy specified in columns 6—12. WEB/PRBC, wholeblood or packed red cells; COLL, colloids (albumin or starch); XTALS, crystalloids; FFP, fresh frozen plasma; DOB,dobutamine; DOP, dopamine; CRYO, cryoprecipitate. According to these data, FFP, COLL andWB/PRBCmay be consideredlikely to elicit an appropriate response.

observed in nearest neighbours selected by theprobability program from the database. These near-est neighbours are then surrogates for the newlyadmitted study patient. Where possible, therapieswere administered one at a time, continuous mea-surements being made before, during and after eachtherapy. Fig. 4 is an example of a patient’s nearestneighbours’ responses to various therapeutic inter-ventions, measured in terms of improved SP.

Statistical analyses

Mean values and S.D. were calculated for each vari-able at comparable times after ED admission. Whenthere were multiple measurements from the samepatient, these were averaged over each time period,and the average value was used for further computa-tions. Evaluation of data sets obtained under com-parable temporal conditions involved the two-tailed

Student’s t-test and the Mann—Whitney test. Forcategorical variables, differences in proportionsbetween survivors and non-suvivors were evaluatedby the chi-squared test and Fisher’s exact test. Bon-ferroni’s correction factor was applied to multiplecomparison procedures. We used GraphPad Prismsoftware for these calculations. Differences wereconsidered significant at probability values <0.05.

Results

Clinical series

We studied 313men (72.8%) and 117 women (17.2%).The mean age was 41 � 20 years for all patients;40 � 20 years for the survivors (372); and 44 � 21years for the non-survivors (58). The overall mor-tality was 13.5%.

Therapeutic decisions for trauma patients with long-bone and pelvic fractures 325

Table 2 Non-invasive haemodynamic values (mean � S.D.) during the first 2 days after ED admission

Variable(units)

Extremity fractures Long bone fractures Pelvic fractures Fractures with otherassociated severe trauma

S (N = 28) NS (N = 0) S (N = 100) NS (N = 15) S (N = 55) NS (N = 11) S (N = 218) NS (N = 38)

CI(l/min/m2)

4.07 � 1.05 3.74 � 1.02 3.19 � 0.81 3.50 � 1.13 3.84 � 0.64 3.93 � 1.25 3.57 � 1.15

MAP(mmHg)

88 � 16 88 � 16* 79 � 19 85 � 17 83 � 23 86 � 17 80 � 22

HR(beat/min)

101 � 21 104 � 19 113 � 21 102 � 18 110 � 21 102 � 22 ** 115 � 24

SapO2

(%)98 � 3 98 � 3 97 � 4 98 � 3 98 � 3 99 � 3** 97 � 7

PtcCO2

(torr)53 � 22 46 � 16** 60 � 23 42 � 12* 53 � 31 46 � 16 48 � 25

PtcO2/FiO2

(torr)219 � 115 218 � 133 ** 71 � 69 216 � 148* 103 � 95 212 � 116 ** 144 � 117

SP (%) 83 � 17 83 � 17** 67 � 16 79 � 19 73 � 19 82 � 18 ** 72 � 20* P-value < 0.05.** P-value < 0.01.

There were 28 patients with extremity fractures,including 6 ulnar, 8 radial, 2 elbow, 3 carpal, 3 knee,3 patellar, 4 ankle and 2 tarsal fractures; all thesepatients survived. There were 115 patients withlong-bone fractures, including 68 femoral, 42 tibialor fibular and 11 humeral fractures; the femoralfractures carried a mortality of 16.2%, whereas thenon-femoral long-bone fractures carried a mortalityof 10.3%. There were 66 pelvic fractures (mortality16.7%) and 256 fractures with other severe truncaltrauma (mortality 15%). The demographic data ofeach of these groups are listed in Table 1.

Haemodynamic values and the probabilityof survival

Table 2 lists the mean values � S.E.M. for survivorsand non-survivors during the first 48 h after EDadmission for cardiac index (CI), heart rate (HR),mean arterial pressure (MAP), SapO2 by pulse oxi-metry, transcutaneous oxygen tension indexed tothe FiO2 (PtcO2/FiO2) and the SP. The CI, MAP,SapO2, PtcO2/FiO2, and SP values of the survivorswere significantly higher than the corresponding

Table 3 Summary of classifications excluding severe head

Predicted to die Pre

N (Row%) N

Actual outcomeDied 43 79.6 11Lived 14 4.2 322

Total (%) 57 14.6 333

Misclassifications:25/390 = 6.4%.

values of those who died, whereas the HR was higherin the non-survivors during the first 48 h after theirED admission. Fig. 3 illustrates the course during thefirst 2 days of continuously monitored survivor andnon-survivor haemodynamic values.

Initial calculation of survival probabilitycompared with actual hospital outcome

Table 3 summarises the classifications of the seriesexcluding severe head injury with GCS <8. Of the390 trauma patients without severe head injury, 336survived and 54 died; the mortality was 13.8%.There were 43 deaths (75.4%) among the 57 patientswho were predicted to die; 14 (24.6%) of the 57survived with continuing resuscitative therapy.There were 322 (96.7%) survivors among the 333patients who were predicted to survive; 11 (3.3%) ofthe 333 died of late medical complications or organfailures.

In the series as a whole there were 65/430 (15.1%)misclassifications, but only 25/390 (6.4%) misclassi-fications in patients without severe head injury andbrain death (Table 3).

injury (N = 390)

dicted to live Total

(Row%) N (Col%)

20.4 54 13.895.8 336 86.2

85.4 390 100.0

326 K. Lu et al.

Table 4 Incidence of organ failures in survivors and non-survivors

Survivors (N = 372) Non-survivors (N = 58) P-value (Fisher’sexact test)Number Percent Number Percent

ARDS 77 20.7 34 58.6 0.01Hepatic failure 65 17.5 15 27.6 0.22Renal failure 9 2.4 16 27.6 0.01Sepsis 73 19.6 22 37.9 0.03

Organ failures and sepsis

In the series as a whole, there were 111 patientswith adult respiratory distress syndrome (ARDS) orrespiratory failure, 80 patients with hepatic failure,25 patients with acute renal failure and 95 patientswith sepsis. There were higher percentages of organfailures among the non-survivors compared withsurvivors (Table 4).

Discussion

The proposed approach was designed to provide anearly ED prediction of survival, based on the clinicaldiagnosis, covariates and non-invasive haemody-namic patterns. This mathematical approach usescontinuous on-line displays of non-invasively mon-itored values of blood flow, blood pressure, haemo-globin oxygen saturation and tissue oxygenation.The major finding in the present study was thatearly continuous non-invasive haemodynamic mon-itoring of patients with fractures can identify cir-culatory abnormalities which, if not corrected early,may lead to shock, organ failure and death. The non-invasive monitoring satisfactorily predicted survi-val, was safe, less expensive, of equivalent accuracyto invasive monitoring and could be used anywherein the hospital or prehospital area.22—24

This haemodynamic analysis of patterns also pro-vided a therapeutic decision support system. Theknown responses of nearest neighbours in the data-base provide bedside attendants with an array oftherapies, together with their likely effects on thenewly admitted study patient. The SP and thera-peutic decision support program afforded objectiveevidence to support therapeutic management ofcritically ill patients with fractures.

The limitations in this study include difficulty inseparating the circulatory effects of the fracturesfrom those of associated soft-tissue injury, becausetrauma is the precipitating event for both. Similarly,it may be difficult to differentiate the clinicalaspects of these two. However, as a first step toresolve this problem, we studied the survivor andnon-survivor patterns of patients with fractures of

increasing levels of severity and mortality, fromsimple uncomplicated extremity fractures to long-bone fractures, pelvic fractures and fractures asso-ciated with severe injury.

The survivors of each of the fracture groups hadbetter flow and tissue oxygenation than did thecorresponding non-survivors. All those with simpleuncomplicated extremity fractures survived withonly slightly increased CI and HR. In the otherfracture groups, the survivors’ CI, MAP, SapO2,PtcO2/FiO2, and SP values were higher than thecorresponding values of those who died, whereasthe HR and PtcO2 were higher in the non-survivorsthan in the survivors during the first 48 h after theirED admission. There was a remarkable similarity inthe haemodynamic patterns of present orthopaedicpatients compared with those of previouslyreported trauma patients.22—24 Moreover, the hae-modynamic responses of the trauma survivors aswell as those of the surviving orthopaedic patientswere greater than those of the corresponding non-survivors (Table 2). The haemodynamic patterns ofpatients with femoral fractures and pelvic fractureswere similar to those of patients with severe asso-ciated trauma and severely traumatised patientswithout fractures.22—24 Furthermore, the mortalityincreased with increasing severity in the fracturegroups and with the severity of their associatedinjuries. Femoral fractures, pelvic fractures andfractures with severe associated injuries had>15% mortalities. There were greater haemody-namic abnormalities among the non-survivors thanamong those of each group who survived.

The haemodynamic patterns were consistentwith the concept that fractures produced a signifi-cant adrenal medullary stress response, which wasapparent from the haemodynamic responses of eachfracture group. The stress response is known toincrease HR and to maintain arterial pressure inthe face of inadequate blood flow by increasingvascular resistance.24,23 With the intense stressexperienced by acutely ill or dying patients, theincreased vasomotion may become uneven or mal-distributed at the microcirculatory level. This leadsto increasingly uneven meta-arteriolar flow, withpoor tissue perfusion and local areas of tissue

Therapeutic decisions for trauma patients with long-bone and pelvic fractures 327

hypoxia that become more pronounced as evi-denced by decreased PtcO2/FiO2 and increasedPtcCO2. Progressively worsening tissue perfusionleads to global tissue hypoxia, organ failures andincreased mortality.24,22

The proposed mathematical analysis of the cir-culatory status defines the patient’s state by specificdiagnostic categories, clinical covariates, haemody-namic variables, their first and second derivativesand their integrals. The accuracy and reliability ofthis approach depends on the size and comparabilityof the database needed to provide an adequategroup of nearest neighbours. This approach is similarto that of experienced clinicians who, on seeing anunusual patient, reach back in their memory torecall a similar patient who responded well to aspecific therapy. This computerised probability pro-gram searches the database in an analogous manner,to find very similar patients in order to predict theprobability of survival from analysis of early clinicalhaemodynamic patterns. Moreover, the severity ofillness may be inferred from the probability ofsurvival; that is, a person with a high likelihood ofdeath is severely ill, whereas a person with a highestimated survival may not be very ill at all. More-over, the search and display probability program isable to quantitatively evaluate the relative effec-tiveness of the therapies that may be given. Inessence, this information system emulates the pro-cess of good clinical judgment, using a database ofsimilar patients with a computerised program toobviate memory lapses.

The probability analysis and control programcompares data from a very similar homogeneoussubset of the database population with a newpatient’s demographic characteristics, clinical diag-nosis and haemodynamic patterns, to compute theprobability of survival for the new patient in realtime. This approach was tested in the extenuatingand sometimes chaotic circumstances of severelytraumatised emergency patients in a large inner citypublic hospital. Under these circumstances, thesurvival probability was found to track clinicaland haemodynamic changes and changes after ther-apy throughout the course of recovery. Excludinghead injuries, which have a very different physio-logical pattern, it correctly predicted outcome inover 90% of the patients, beginning with the initialmonitored resuscitation.

The usefulness of this study may be the doc-umentation of haemodynamic stability for practi-cal clinical decisions such as when to take patientsback to theatre for fracture immobilisation. Hae-modynamic stability should not be assumed fromthe vital signs, which are often rather lateresponses to shock. Alternatively, the present

study demonstrates the early use of non-invasivelymonitored haemodynamic variables, including CIand PtcO2, as outcome predictors. The patientswho promptly achieved and maintained optimalhaemodynamic goals survived; those who failedto maintain optimal haemodynamic values hadsignificantly higher mortality (Table 2, Figs. 1and 2).

Acknowledgement

This study was supported in part by DOD BAA99-1.

References

1. Artegas A, Bernard GR, Brigham RL, et al. The American-European Consensus Conference on ARDS. Parts 1 and 2. Am JRespir Crit Care Med 1994;149:818—24;Artegas A, Bernard GR, Brigham RL, et al. The American-European Consensus Conference on ARDS. Parts 1 and 2. Am JRespir Crit Care Med 1998;157:1332—47.

2. Bayard DS. A forwardmethod for optimal stochastic nonlinearand adaptive control. IEEE Trans Automatic Control 1991;36:1046—53.

3. Bayard DS. Reduced complexity dynamic programming basedon policy iteration. J Math Anal Appl 1992;170:75—103.

4. Bayard DS, Botnen A, Shoemaker WC, Jelliffe RW. Stochasticanalysis of therapeutic modalities using a database of patientresponses. IEEE Symp Comp Based Med Sys 2001;11:439—44.

5. Berlauk JF, Abrams JH, Gilmour IJ, et al. Pre—operativeoptimization of cardiovascular hemodynamics improves out-come in peripheral vascular surgery. Ann Surg 1991;214:189.

6. Bertsekas DP, Tsitsiklis JN. Neuro-dynamic programming Bel-mont, MA: Athena Scientific; 1996.

7. Bishop MW, Shoemaker WC, Appel PL, et al. Relationshipbetween supranormal values, time delays and outcome inseverely traumatized patients. Crit Care Med 1993;21:56.

8. Boyd O, Grounds M, Bennett D. Preoperative increase ofoxygen delivery reduces mortality in high-risk surgicalpatients. JAMA 1993;270:2699.

9. Conners Jr AF, Speroff T, Dawson NV, et al. The effectivenessof right heart catheterization in the initial care of critically illpatients. JAMA 1996;276:899—997.

10. Creamer JE, Edwards JD, Nightingale P. Hemodynamicand oxygen transport variables in cardiogenic shock sec-ondary to acute myocardial infarction. Am J Cardiol 1990;65:1287.

11. Dalen JE, Bone RC. Is it time to pull the pulmonary arterycatheter? JAMA 1996;276:916—8.

12. Ferrera PC, Hill DA. Good outcomes of open pelvic fractures.Injury 1999;30:187—90.

13. Friese RS, Shafi S, Gentilello LM. Pulmonary artery catheteruse is associated with reduced mortality in severely injuredpatients: A National Trauma Data Bank analysis of 53, 312patients. Crit Care Med 2007;34:1597—601.

14. Guyatt G. Ontario Intensive Care Group. A randomized con-trol trial of right heart catheterization in critically illpatients. J Intensive Care Med 1991;6:91—5.

15. Kern JW, Shoemaker WC. Meta-analysis of hemodynamicoptimization in high-risk patients. Crit Care Med 2002;30:1686—92.

328 K. Lu et al.

16. Lubbers DW. Theoretical basis of transcutaneous blood gasmeasurements. Crit Care Med 1981;9:721—30.

17. Moore FA,Haemel JB,MooreEE, et al. Incommensurateoxygenconsumption in response to maximal oxygen availability pre-dicts postinjury oxygen failure. J Trauma 1992;33:58.

18. Prince RA, Hoffman CJ, Scanlon CJ, Mayberry JA. The distinctand secondary harmful effects of pelvic and extremity injuryon the outcome of laparotomy for trauma. J Surg Res2005;124:3—8.

19. Richmond J, Egol KA, Koval KJ. Management of orthopaedicinjuries in polytrauma patients. Bull Hosp Joint Dis 2002;60:162—7.

20. Scalea TM, Simon HM, Duncan AO, et al. Geriatric bluntmultiple trauma: improved survival with early invasive mon-itoring. J Trauma 1990;30:129—36.

21. Shoemaker WC, Appel PL, Kram HB, et al. Prospective trial ofsupranormal values of survivors as therapeutic goals in high-risk surgical patients. Chest 1988;94:1176.

22. Shoemaker WC, Bayard DS, Botnan A, et al. Mathematicalprogram for outcome prediction and therapeutic supportbeginning within 1 hr of admission. Crit Care Med 2005;33:1499—506.

23. Shoemaker WC, Belzberg H, Wo CCJ, et al. Multicenter studyof non-invasivemonitoring systems as alternatives to invasivemonitoring of acutely ill emergency patients. Chest 1998;114:1643—52.

24. Shoemaker WC,Wo CCJ, Chan L, et al. Outcome prediction ofemergency patients by non-invasive hemodynamic monitor-ing. Chest 2001;120:528—37.

25. Sutton RS, Barto AG. Reinforced learning Cambridge, UK: MITPress; 1998.

26. Taylor RW, and the Pulmonary Artery Catheter ConsensusConference participants. Pulmonary Artery Catheter Consen-sus Conference. Crit Care Med 1997;25:910—25.

27. Tremper KK, Huxtable RF. Dermal heat transport analysis fortranscutaneous O2 measurements. Acta Anaesthiol Scand1978;68:4—9.

28. Tremper KK, Shoemaker WC. Transcutaneous oxygen mon-itoring of critically ill adults with and without low flow shock.Crit Care Med 1981;9:706—9.

29. Tremper KK, Waxman K, Bowman R, Shoemaker WC. Con-tinuous transcutaneous oxygen monitoring during respiratoryfailure, cardiac decompensation, cardiac arrest, and CPR.Crit Care Med 1980;8:337.

30. Tremper KK, Waxman K, Shoemaker WC. Effects of hypoxiaand shock on transcutaneous PO2 values in dogs. Crit CareMed 1979;7:526.

31. Turen CH, Dube MA, LeCroy MC. Approach to the polytrau-matized patient with musculoskeletal injuries. J Am AcadOrthop Surg 1999;7:154—65.

32. Wang X, Van De Water JM, Sun H, et al. Hemodynamicmonitoring by impedance cardiography with an improvedsignal processing technique. IEEE Eng Med Biol Mag 1993;15:699—700.

33. Wilson J, Woods I, Faucett J, et al. Reducing the risk ofmajor elective surgery: randomized control trial of pre-operative optimization of oxygen delivery. BMJ 1999;318:1099—103.