Triage staff prioritize emergency department (ED) Predictive Validity of a Computerized Emergency Triage Tool. Abstract

Predictive Validity of a Computerized Emergency Triage Tool Sandy L. Dong, MD, MSc, Michael J. Bullard, MD, David P. Meurer, BScN, Sandra Blitz, MSc, ...
0 downloads 1 Views 168KB Size
Predictive Validity of a Computerized Emergency Triage Tool Sandy L. Dong, MD, MSc, Michael J. Bullard, MD, David P. Meurer, BScN, Sandra Blitz, MSc, Edward Akhmetshin, PhD, Arto Ohinmaa, PhD, Brian R. Holroyd, MD, Brian H. Rowe, MD, MSc

Abstract Background: Emergency department (ED) triage prioritizes patients on the basis of the urgency of need for care. eTRIAGE is a Web-based triage decision support tool that is based on the Canadian Triage and Acuity Scale (CTAS), a five level triage system (CTAS 1 = resuscitation, CTAS 5 = nonurgent). Objectives: To examine the validity of eTRIAGE on the basis of resource utilization and cost as measures of acuity. Methods: Scores on the CTAS, specialist consultations, computed-tomography use, ED length of stay, ED disposition, and estimated ED and hospital costs (if the patient was subsequently admitted to hospital) were collected for each patient over a six month period. These data were queried from a database that captures all regional ED visits. Correlations between CTAS score and each outcome were measured by using logistic regression models (categorical variables), univariate analysis of variance (continuous variables), and the Kruskal-Wallis analysis of variance (costs). A multivariate regression model that used cost as the outcome was used to identify interaction between the variables presented. Results: Over the six month study, 29,524 patients were triaged by using eTRIAGE. When compared with CTAS level 3, the odds ratios for consultation, CT scan, and admission were significantly higher in CTAS 1 and 2 and were significantly lower in CTAS 4 and 5 (p < 0.001). When compared with CTAS levels 2–5 combined, the odds ratio for death in CTAS 1 was 664.18 (p < 0.001). The length of stay also demonstrated significant correlation with CTAS score (p < 0.001). Costs to the ED and hospital also correlated significantly with increasing acuity (median costs for CTAS levels in Canadian dollars: CTAS 1 = $2,690, CTAS 2 = $433, CTAS 3 = $288, CTAS 4 = $164, CTAS 5 = $139, and p < 0.001). Significant interactions between the data collected were found in a multivariate regression model, although CTAS score remained highly associated with costs. Conclusions: Acuity measured by eTRIAGE demonstrates excellent predictive validity for resource utilization and ED and hospital costs. Future research should focus on specific presenting complaints and targeted resources to more accurately assess eTRIAGE validity. ACADEMIC EMERGENCY MEDICINE 2007; 14:16–22 ª 2007 by the Society for Academic Emergency Medicine Keywords: emergency department, triage, information technology, computerized decision support, patient acuity, cost

T

riage staff prioritize emergency department (ED) patients according to urgency of need for medical assessment, based on a brief initial clinical assessment. In Canada, the Canadian Triage and Acuity Scale (CTAS), a five level acuity scale (1 = resuscitation, 2 =

From the Departments of Emergency Medicine (SLD, MJB, DPM, SB, EA, BRH, BHR) and Public Health Sciences (AO, BRH, BHR), Faculty of Medicine and Dentistry, University of Alberta; and Capital Health (SLD, MJB, BRH, BHR), Edmonton, Alberta, Canada. Received May 17, 2006; revision received July 25, 2006; accepted August 23, 2006. Contact for correspondence and reprints: Michael J. Bullard, MD; e-mail: [email protected].

16

ISSN 1069-6563 PII ISSN 1069-6563583

emergent, 3 = urgent, 4 = semi-urgent, and 5 = nonurgent), is the nationally recognized ED triage standard.1–3 Recent research has reported the ability of CTAS to predict ED resource utilization as a measure of validity.4,5 A Web application (eTRIAGE; University of Alberta, Edmonton, Canada) that provides point of care triagedecision support on the basis of CTAS has been developed in Canada and is now used in a number of EDs.6 The eTRIAGE user selects from a standardized complaint list, which brings up a complaint specific CTAS-based template that displays all relevant discriminators, to assist the nurse in assigning the appropriate triage score. The triage nurse is able to override the computer-assisted triage score if his or her clinical impression disagrees; however, the reason for the override must be recorded and can be

ª 2007 by the Society for Academic Emergency Medicine doi: 10.1197/j.aem.2006.08.021

ACAD EMERG MED

www.aemj.org

17

used as a quality assurance measure. Previous research demonstrated that eTRIAGE was easy to learn, even for novice computer users; did not increase triage nurse assessment time; and was widely accepted by triage nurses.7 Triage nurses using eTRIAGE also demonstrated good independent agreement as well as higher agreement with a consensus standard than did the case of triage nurses applying CTAS on the basis of memory alone.8,9 The objective of this study was to access an administrative database to evaluate the validity of eTRIAGE to predict severity, resource utilization, and incurred ED and inpatient costs in an urban tertiary care ED over a six month period.

who left before physician assessment and who left before completion of care), admitted to hospital, transferred to another facility, or dead in the ED (includes dead on arrival and in ED deaths). Because this was a tertiary care hospital, the vast majority of patients who were transferred to another facility were transferred for admission because of bed shortages. This study a priori considered all admissions, transfers, and deaths to be equivalent to hospital admission as a measure of the severity of the patient’s condition. Costs included operating costs of the ED and inpatient costs if the patient was subsequently admitted to hospital. Costs included all direct department costs and costs incurred by other departments (e.g., food service, housekeeping, pharmacy, administration, etc.). All costs were expressed in 2003 and 2004 Canadian dollars (CDN). ED costs were based on the average provincial unit cost per type of visit. There are 425 visit types, according to the grouper methodology used by the ACCS. Inpatient costs were estimated by multiplying the resource intensity weight (RIW) of each admission by the average provincial inpatient cost per RIW. The RIW is a relative measure of the resource cost of inpatient cases, for which the provincial average of all cases equals one. One RIW in Alberta in 2003 and 2004 was valued at CDN $4,459. The RIW methodology was developed by the Canadian Institute for Health Information.11,12 Physician fees were excluded from the cost analysis. In Alberta, physicians are remunerated from a separate budget, the provincial health care insurance plan, and not from the hospital budget.



January 2007, Vol. 14, No. 1



METHODS Study Design This study examined data to assess the validity of eTRIAGE on the basis of resource utilization and cost as measures of acuity. The study was approved by the Health Research Ethics Board at the University of Alberta. Study Setting and Population The data for this study were obtained from the Ambulatory Care Classification System (ACCS). Each hospital in the province of Alberta is mandated to track all emergency services provided, and all hospitals contribute to this central provincial registry. Because the Canadian health care system is a public, single payer system, the ACCS database captures and stores every provincial emergency service except those provided to nonresidents. Each record in the ACCS includes the personal health number (a unique identifier for each person in the province), demographic information, the service provided, the date and time of the visit, the diagnosis, and the disposition status. Patient visit data are entered into the ACCS by trained medical records nosologists within each hospital. The data were extracted from the ACCS for a large Canadian urban tertiary care teaching hospital ED that has 52 ED patient care beds and an annual volume of approximately 67,000 visits. CTAS was introduced at the Royal Alexandra Hospital ED in 1997 after nursing education, and it remained a memory based process until 2003. After its development and initial reliability testing, eTRIAGE was implemented in July 2003. Data from all adult patients (R17 yr of age) presenting to the ED between January 1, 2004, and June 30, 2004, were used for this study. Patients under 17 years of age are triaged by using a different eTRIAGE tool that is based on Pediatric CTAS,10 and such patients were not included in this study. Study Protocol The following data were extracted for each individual patient visit and were used in this study for analysis: each patient’s age and gender and each patient’s triage score (CTAS score), as assigned by triage nurses using eTRIAGE. We used specialist consultation, use of computed tomography (CT), and ED length of stay (LOS) as markers of resource utilization. Disposition is coded in the ACCS into five categories: discharged, left against medical advice (includes those

Data Analysis The primary null hypothesis used in this study was no correlation between CTAS score and the measured outcome. Correlation between each categorical outcome (consultation, CT, death, and admission) and CTAS score was tested by using a univariate logistic regression model for each outcome. A priori, the CTAS score with the greatest number of patients was used as the reference risk CTAS score for logistic regression models. Odds ratios generated were compared with this CTAS level. Correlation between the LOS and the CTAS score was tested by using univariate analysis of variance (ANOVA) after normalizing the data with a square root transformation. The raw data for LOS were reported as median and interquartile range (IQR; 0.25, 0.75). Correlation between cost and CTAS score was tested by using the KruskalWallis ANOVA. The raw data for costs were reported as median and IQR (0.25, 0.75). A multivariate regression model was developed to determine significant interactions between the outcomes reported. In this model, the total cost, normalized with a logarithm function, was the dependent variable. The independent variables were the following: age, gender, CTAS score, CT scan, specialist consultation, admission, LOS, interaction terms with CTAS score and specialist consultation, CTAS score and LOS, specialist consultation and admission, specialist consultation and LOS, and admission and LOS. Statistical calculations were conducted by using SPSS (SPSS, Chicago, IL) and SAS (SAS, Cary, NC) software packages.

18

Dong et al.



VALIDITY OF COMPUTERIZED TRIAGE

RESULTS There were 29,524 adult patient visits to the ED between January 1 and June 30, 2004. Complete patient acuity and resource utilization data were available for 29,346 (99.4%) patients, and complete cost data were available for all patients. The mean patient age (SD) was 46.7 (SD  20.0) years, and 52.2% were male. Figure 1 shows the CTAS score distribution. CTAS 3 was the most common CTAS score and was used as the reference risk CTAS score in logistic regression models. Resource Utilization There was a significant correlation between CTAS score and the odds of specialist consultation and CT scan (Figure 2). Compared with patients in CTAS 3, the odds ratios (ORs) for consultation were 3.54 (95% confidence interval [CI] = 2.75 to 4.57), 2.40 (95% CI = 2.21 to 2.62), 0.42 (95% CI = 0.39 to 0.46), and 0.14 (95% CI = 0.10 to 0.17) for patients in CTAS 1, 2, 4, and 5, respectively. Compared with patients in CTAS 3, the OR for a CT scan was 3.70 (95% CI = 2.82 to 4.86), 1.97 (95% CI = 1.77 to 2.18), 0.52 (95% CI = 0.47 to 0.57), and 0.13 (95% CI = 0.09 to 0.18) for patients in CTAS 1, 2, 4, and 5, respectively. The overall median LOS was 253 minutes (IQR: 144, 428). The median LOS for patients in CTAS 1 was 197 minutes (IQR: 70, 358), in CTAS 2 it was 351 (IQR: 215, 351), in CTAS 3 it was 309 (IQR: 183, 491), in CTAS 4 it was 206 (IQR: 120, 338), and in CTAS 5 it was 130 (64, 213). An ANOVA conducted on normalized data showed a significant correlation between the LOS and CTAS score (p < 0.001). Patient Severity There were 70 patient deaths in the ED (0.2%). Fifty-seven deaths occurred in patients in CTAS 1, seven were in CTAS 2, four were in CTAS 3, and two were in CTAS 4. There were no deaths among patients with a CTAS 5. The small number of patient deaths in the CTAS 2–5 categories prevented a meaningful logistic regression model that used all five CTAS levels. Therefore, a logistic regression model was developed that used CTAS 1 as one category, compared with CTAS 2–5, used collectively as a second category. In this model, the OR of an in ED death

Figure 2. Odds ratios for consultation, CT scan, and admission by Canadian Triage and Acuity Scale (CTAS) score. Each plot represents the odds ratio and 95% confidence interval compared with CTAS 3.

for patients in CTAS 1 compared with in CTAS 2–5 was 664.18 (95% CI = 357.69 to 1,233.30). The overall proportion of patients admitted to hospital, transferred to other facilities for admission, or admitted to the morgue was 19.3%. Compared with patients in CTAS 3, the ORs for admission were 4.45 (95% CI = 3.45 to 5.73), 2.22 (95% CI = 2.04 to 2.41), 0.36 (95% CI = 0.33 to 0.39), and 0.16 (95% CI = 0.13 to 0.20) when compared with patients in CTAS 1, 2, 4, and 5, respectively (Figure 2). Cost Figure 3 demonstrates median, interquartile range, and total range of the ED and hospital costs associated with each CTAS level for all patients; patients who were seen only in the ED (patients discharged or left against medical advice); and patients who were subsequently admitted to hospital, transferred for admission, or admitted to the morgue. Although there was a wide range of costs associated with each CTAS level (i.e., outliers), CTAS 1 patients incurred the highest cost overall, with a significant decline in lower acuity patients. These differences were statistically significant (p < 0.001) within the disposition categories when analyzed by using the KruskalWallis ANOVA. The multivariate regression model found that gender was the only variable that was not a significant predictor of the total cost (p = 0.53). All other variables and interaction terms in the model were significant at the 0.05 threshold. The adjusted R2 was 0.67 in this model. DISCUSSION

Figure 1. Distribution of Canadian Triage and Acuity Scale (CTAS) score.

Triage is a complex process in the emergency setting. Traditionally in Canada, triage nurses use memory based knowledge to assign a triage score. Memory supports, such as a computerized triage decision support tool, have been developed; eTRIAGE is one such tool. This decision support tool has been shown to be easy to use7; it is reliable and more valid than memory based approaches.8,9 The present study was designed to examine

ACAD EMERG MED



January 2007, Vol. 14, No. 1



www.aemj.org

19

Figure 3. Box and whisker plot of emergency department and hospital costs by Canadian Triage and Acuity Scale (CTAS) score and disposition. Bars represent median cost, boxes represent the interquartile range, and whiskers represent the range.

this tool’s validity by using a variety of clinically meaningful outcomes for patients presenting to a large, urban, tertiary care ED. Overall, this study demonstrated excellent predictive validity using eTRIAGE. We believe that the differences in the outcomes when stratified by CTAS level are clinically meaningful. A valid triage system is necessary to identify patients who are in greatest need of medical attention, to minimize delays in patient care, and to define a department’s acuity.3 Previous work by our group demonstrated that nurses using eTRIAGE had better agreement with an expert panel’s consensus over nurses with no decision support, when compared with the expert panel.9 We also demonstrated elsewhere good interrater agreement that was not affected by different degrees of ED overcrowding.8 With a valid system, administrators can better define resource needs, compare sites and regions, and perform benchmarking comparisons. The Canadian Association of Emergency Physicians and the National Emergency Nurses Affiliation identify consistent triage and prospective data collection as important steps in defining and managing ED overcrowding.13 Other triage systems have demonstrated the ability to predict patient outcomes. CTAS, without electronic decision support, was demonstrated by Spence et al.4 and Stenstrom et al.14 to have good predictive validity in terms of hospital admission, use of imaging, use of a complete blood count, and LOS in the ED. Cooke and Jinks evaluated the ability of the Manchester triage system to detect patients requiring critical intervention on arrival to the ED.15 Eitel et al. and Tanabe et al. have demonstrated the Emergency Severity Index (ESI), in two different versions, to be able to predict hospital admission, intensive care admission, and resource consumption.16–18 These results are important. To our knowledge, this is the first large study to evaluate the validity of an electronic triage decision support system.

Although there are widely accepted measures of interrater reliability for ED triage, there is currently no reference standard for validity. An individual patient’s urgency of need for medical attention does not necessarily correlate with the measures of resource utilization or patient acuity that are used in this or other studies. For example, a patient with severe anaphylaxis may meet criteria for CTAS 1 because of the need for immediate physician assessment and intervention; however, with medical therapy and observation, this patient may not need to be admitted to hospital, nor require any laboratory investigation or imaging. However, an elder patient may present after a fall and be unable to ambulate because of hip pain but have nondiagnostic plain radiographs. This patient may require a prolonged LOS and more detailed imaging (including CT scan, bone scan, or magnetic resonance imaging) and eventually may be admitted to hospital, despite a lower triage acuity score. There are numerous clinical scenarios such as this in which the need for urgent medical assessment may not correlate with resource utilization or hospital admission. Another way of looking at this apparent disconnect is to consider the different approaches of various triage data users. Researchers have been interested in measuring the reliability and validity of the overall triage scoring system. Researchers have asked the question, ‘‘how well does the triage system identify patients in urgent need of medical assessment in different patient populations, ED volumes, and staff experience?’’ Administrators are interested in costs and in establishing a means of identifying resource needs, benchmarking standards, and departmental efficiency. Administrators ask, ‘‘how well can we predict the resources required to effectively serve our patients and community?’’ Clinicians (nurses and physicians) are interested in an efficient system that accurately prioritizes patients on the basis of acuity. Clinicians ask, ‘‘how well does triage perform in the clinical setting and

20

does it safely assign urgency on the basis of patient presentation?’’ The researchers’ agenda may be better served by also using outcome measures in addition to admission rates and diagnostic imaging usage. Moreover, these comparisons may provide a more useful analysis if groups of patients with specific presenting complaints are selected. For example, comparing head CT use with CTAS score may be more informative for headache or headinjury complaints than for all patients. Triage acuity alone should not be considered a suitable measuring stick to use for benchmarking, patient casemix acuity, or departmental workload. Other factors, such as patient type and complexity, patient volume, rate and surges of patient presentations, and overcrowding also affect workload. In this study, the mean LOS of patients in the CTAS 1 cohort was less than that of patients in the CTAS 2 or 3 cohorts. Patients who are dead on arrival or who are not successfully resuscitated can be transported to the hospital morgue relatively quickly. Critically ill and injured patients represent relatively easy disposition decisions and are usually transferred to the intensive care unit or operating room more expeditiously than are patients awaiting a ward bed. This is important knowledge if one is planning to use ED LOS as a measure of resource utilization. If the critically ill or dead patients are quickly removed from the ED, then despite the high patient acuity, these patients will only expend high intensity ED resources for a relatively short period. There was a statistically significant correlation between costs and CTAS score overall. Hospitalized patients appeared to incur similar median costs, regardless of CTAS level (Figure 3). There was still a statistically significant correlation in the costs of hospitalized patients and CTAS score. Despite this statistical significance, the clinical significance for the differences in cost may not be as meaningful. We hypothesize that hospitalization costs are so large relatively that they eclipse the ED costs. This study did not examine length of hospitalization. Tanabe et al. reported the inability of another triage system (ESI) to predict hospital LOS.18 This study found significant interaction between some of the predictors in a regression model of costs. Despite this, the CTAS score remained a significant contributor to the overall model. We hypothesize that this is a result of the complexity of cost models in any ED; not all variables can be included in any model, no matter how extensive. Despite this complexity, this multivariate regression model accounted for 67% of the variance in costs. LIMITATIONS This study examined data from more than 29,000 patients. This large sample size may reduce the study’s efficiency. This study was conducted in a tertiary care urban teaching hospital. The triage nurses had been using the five level CTAS triage system for approximately seven years. The results of the study may not be applicable to other centers that have different volumes and CTAS experience. The eTRIAGE application was introduced six months before the study date. Some of the triage nurses may still have been learning the application during the

Dong et al.



VALIDITY OF COMPUTERIZED TRIAGE

study period. It is uncertain at what point an eTRIAGE user achieves maximal competency or expertise. The rate of mistriage with eTRIAGE is not known. Similarly, the extent of miscoding of the ACCS database is not known. Miscoding would cause data contamination and could result in an underestimation or overestimation of the outcome differences, contributing to the probability of type II error. Audits and observed data entry may provide useful information on the rate of error. The cost data did not include physician fees. Emergency physicians, consultants, and hospital physicians are remunerated from the provincial health plan on a fee for service basis, independently of the hospital budget. CONCLUSIONS This study demonstrated excellent correlation among eTRIAGE CTAS scores and patient severity (deaths, admission rate), resource utilization (consultation, use of CT scanning, and LOS), and health care costs. There is a need for further research that focuses on common patient presentations to develop appropriate criterion standards for measuring CTAS validity. Dr. Rowe is supported by a Canada Research Chair from the Canadian Institute of Health Research (Ottawa, Ontario, Canada). Ms. Blitz and Dr. Akhmetshin are supported by the Emergency Medicine Research Group (EMeRG) in the Department of Emergency Medicine, University of Alberta. The authors thank the Royal Alexandra Hospital Foundation, the University of Alberta Hospital Foundation, and the Kingsway Emergency Agency for funding this research. Drs. Bullard, Holroyd, and Rowe were supported with grant funding to develop eTRIAGE. David Meurer programs and maintains the application.

References 1. Murray MJ, Bullard MJ, Grafstein E. Revisions to the Canadian Emergency Department Triage and Acuity Scale Implementation Guidelines. Can J Emerg Med. 2004; 6:421–7. 2. Beveridge R. CAEP issues. The Canadian Triage and Acuity Scale: a new and critical element in health care reform. Canadian Association of Emergency Physicians. J Emerg Med. 1998; 16:507–11. 3. Beveridge R, Clarke B, Janes L, et al. Canadian Emergency Department and Triage Scale: implementation guidelines. Can J Emerg Med. 1999; 1(3 suppl):S1–24. 4. Spence JM, Beaton DE, Murray MJ, Morrison LJ. Does the Canadian emergency department triage and acuity scale correlate with admission to the hospital from the emergency department? [abstract]. Can J Emerg Med. 2004; 6:180. 5. Murray MJ, Levis G. Does triage level (Canadian Triage and Acuity Scale) correlate with resource utilization for emergency department visits? [abstract]. Can J Emerg Med. 2004; 6:180. 6. Smith R. eTRIAGE brings order to emergency departments. Express News, University of Alberta. 2005; April. Available at: http://www.expressnews.ualberta. ca/article.cfm?id=6514. Accessed Nov 16, 2006. 7. Bullard MJ, Meurer D, Pratt S, Colman I, Holroyd BR, Rowe BH. Evaluation of triage nurse satisfaction with

ACAD EMERG MED

www.aemj.org

21

training and use of an electronic triage tool. Can J Emerg Med. 2003; 5:183–4. 8. Dong SL, Bullard MJ, Meurer DP, et al. Reliability of computerized emergency triage. Acad Emerg Med. 2006; 13:269–75. 9. Dong SL, Bullard MJ, Meurer DP, et al. Emergency triage: comparing a novel computer triage program with standard triage. Acad Emerg Med. 2005; 12:502–7. 10. Warren D, Jarvis A, Leblanc L, et al. Canadian Pediatric Triage and Acuity Scale: implementation guidelines for emergency departments. Can J Emerg Med. 2001; 3(4 suppl):S1–27. 11. Canadian Institute for Health Information. DAD resource intensity weights and expected length of stay. Ottawa, Ontario, Canada: Canadian Institute for Health Information, 2001, p v. 12. Canadian Institute for Health Information. Resource intensity weights: summary of methodology 1996/ 97. Ottawa, Ontario, Canada: Canadian Institute for Health Information, 1995.

13. Canadian Association of Emergency Physicians and National Emergency Nurses. A joint position statement on emergency department overcrowding. Can J Emerg Med. 2001; 3:82–8. 14. Stenstrom R, Grafstein GE, Innes G, Christenson J. The predictive validity of the Canadian Triage and Acuity Scale (CTAS) [abstract]. Can J Emerg Med. 2003; 5:184. 15. Cooke MW, Jinks S. Does the Manchester triage system detect the critically ill? J Accid Emerg Med. 1999; 16:179–81. 16. Eitel DR, Travers DA, Rosenau AM, et al. The emergency severity index triage algorithm version 2 is reliable and valid. Acad Emerg Med. 2003; 10:1070–80. 17. Tanabe P, Gimbel R, Yarnold PR, et al. Reliability and validity of scores on the Emergency Severity Index version 3. Acad Emerg Med. 2004; 11:59–65. 18. Tanabe P, Gimbel R, Yarnold PR, et al. The Emergency Severity Index (version 3) 5-level triage system scores predict ED resource consumption. J Emerg Nurs. 2004; 30:22–9.



January 2007, Vol. 14, No. 1



There are things that no one tells you There are things that no one tells you when you’re a resident. We’re raised on the ‘‘see one, do one, teach one’’ model, which is okay for central lines and intubations, but it doesn’t prepare you to sit down with a daughter and a wife and tell them that her father, her husband has died. In lectures, I was told to be direct, to use clear language. No euphemisms. I remember to take off the bloody gloves, to sit down. Where are the words when I grasp for them? Do you say ‘‘He’s dead’’ or ‘‘He died’’? Does it matter? Do they really hear me at all? I want to cry, and in my nervousness, I want also to laugh. My hands still shake from the energy of the resuscitation, their own memory of the room, the power of the moments just before. Notification came in of an elder man with head trauma after a fall. I never know what to make of notifications. As a resident they thrilled me, the red trauma phone, like something straight out of television. I have since been humbled; the calls are either false alarms or a harbinger of someone else’s nightmare. Mr. W came in with all the attendant chaos and energy of one severely injured. That electricity jumps quickly from out-of-hospital to hospital staff. It was clear from the beginning that he was injured critically; massive head trauma was noted on the chart. These are images that, even after one has seen hundreds, are difficult to depict. Words seem cold and cannot get to the essence of what it’s like to have a person dying in front of you. I can record bilateral raccoon eyes, a septal fracture with blood from the nares and from the mouth, but I don’t know that I can describe what it feels like to stand back and take in the last moments of someone’s life, to know that just an hour before, Mr. W was on his way upstairs for a nap when his left foot missed the step and the whole world came crashing down. He came to us still breathing, gurgling blood. Eyes so purple and swollen shut that they looked colored on. We went through our protocols, ABCs, by rote. Tubes went in, monitors were attached. Maybe someone spoke to the gravity of the situation, or maybe no one said anything until after his heart stopped and we had shocked it back into activity once, twice, and a third time; I cannot recall. We bat around phrases like medical futility and fatal insult. We got him to computed tomography long enough to see all the blood in his brain and rushed him back to resuscitate him again. I stood there, watching the tubes and the compressions and the medications and electricity, and I wondered, when is it enough? So I pull off my bloody gloves and sit down with his wife and daughter and tell them of the blood in his head and of his heart stopping. I say that we had shocked him and shocked him again, and they ask whether there is any chance of recovery. And I think, ‘‘how can I say no, do I really know that there is no chance?’’ So definite, the terms of life and death. And they say, ‘‘No more. No more electricity and pain, if he can feel pain. No more tubes or machines. Please just make him comfortable.’’ No one told me in residency what it was like to pull an endotracheal tube, to watch blood trickle from the corner of someone’s mouth or that it turns nearly black as time passes. No one told me that his pacemaker would