International Journal for Quality in Health Care Advance Access originally published online on July 3, 2008
International Journal for Quality in Health Care 2008 20(5):331-338; doi:10.1093/intqhc/mzn027
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Interpreting process indicators in trauma care: Construct validity versus confounding by indication
1 Centre of Research Excellence in Patient Safety
2 Department of Epidemiology and Preventive Medicine, School of Public Health and Preventive Medicine, Faculty of Medicine, Nursing and Health Sciences, Monash University
3 National Trauma Research Institute
4 Emergency and Trauma Centre, The Alfred, Commercial Road, Melbourne, Victoria 3004, Australia
Address reprint requests to: Peter A. Cameron, Department of Epidemiology and Preventive Medicine, School of Public Health and Preventive Medicine, Faculty of Medicine, Nursing and Health Sciences, Monash University, The Alfred, Commercial Road, Melbourne, Victoria 3004, Australia. Tel: +61-3-9903-0581; Fax: +61-3-9903-0576; E-mail: peter.cameron{at}med.monash.edu.au
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionConclusionsFundingAcknowledgementsReferences |
|---|
Objective. Quality indicators (QIs) are routinely used in health systems, often on the assumption that they provide a valid reflection of the outcome of care. This study investigated the construct validity of 14 trauma QIs through their ability to identify patients at risk of poor outcomes, including increased mortality, longer lengths of stay and greater use of the intensive care unit (ICU).
Methods. Data were analysed from the Victorian State Trauma Registry from January 2001 to March 2006. Patients included blunt trauma, injury severity score >15 and aged >16 years. Univariate analyses and logistic regression modeling were used to adjust for significant covariates.
Results. The study included 5104 cases. Three QIs were associated with increased mortality (abdominal surgery >24 h after arrival, blunt compound tibial fracture treatment >8 h after arrival and non-fixation of femoral diaphyseal fracture) and three with increased lengths of stay (cranial or abdominal surgery >24 h after arrival and patients developing deep vein thromboses, pulmonary emboli or decubitus ulcers, the latter also associated with increased ICU use). All remaining QIs exhibited reduced risks of poor outcomes or no significant associations.
Conclusion. The investigated QIs generally demonstrated poor construct validity and limited usefulness in predicting outcomes. Although QIs associated with poor patient outcomes may represent an avenue for further refinement, additional investigation of QIs in comparative trauma systems could provide insight into the utility of these measures at the system level.
Keywords: indicators, injury, trauma, quality
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsFundingAcknowledgementsReferences |
|---|
Despite the widespread use of quality indicators (QIs) to assess health services, there is a paucity of evidence concerning their ability to provide generalizable, interpretable and useable information for clinicians or health service managers. To improve quality, indicators need reliability, feasibility, sensitivity, specificity, the ability to provide actionable results, be relevant to users and have an established link with outcomes [1–4]. Construct validity is largely an umbrella term for explaining multiple processes which encompass individual test validation [5]. Validity is high when these procedures establish a relationship between the measure and the construct. Generating meaningful and useful indicators is made difficult by the complexity of the ‘quality’ construct. This paper examines the challenges of construct validation through exploring the process-outcome link in indicators developed for trauma patients, highlighting the problems facing developers and users of QIs.
Trauma QIs were originally promulgated in 1987 by the American College of Surgeons Committee on Trauma (ACSCOT) to identify patients at risk of poor outcomes. Multiple US-based studies have reported reduced risks of poor outcomes in patients flagged by several of the ACSCOT indicators [6–9]. We have previously discussed the limitations of this literature, particularly the generalised lack of risk adjustment between process measures and outcomes [10].
The Victorian State Trauma System employs four in-hospital QIs based on those proposed by ACSCOT. This study aimed to investigate the association of the Victorian State Trauma System and ACSCOT indicators with the quality of care construct, by evaluating the relationship between these indicators and poor patient outcomes, including increased mortality, longer lengths of hospital stay and greater use of the intensive care unit (ICU).
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsFundingAcknowledgementsReferences |
|---|
Data were collected from the Victorian State Trauma Registry (VSTR) and included cases from January 2001 to March 2006. Criteria for inclusion on the VSTR have been published previously [11, 12]. The registry captures a range of items relating to injured patients, provides formal training to data collectors and data verification before data entry [11]. The QIs were abstracted using variables routinely collected on the registry, placing no additional requirements on data collectors. The registry includes cases meeting one of the following criteria: death after injury, Injury Severity Score greater than 15, admittance to the ICU for more than 24 hours and requiring mechanical ventilation, or those requiring urgent surgery. This study included patients classified as blunt trauma, aged 16 years or more and with an Injury Severity Score of greater than 15. As ICU stay of more than 24 hours was an outcome of interest, cases admitted to the registry solely on the basis of ICU stay more than 24 hours and requiring mechanical ventilation were excluded. The Victorian State Trauma System serves a population of approximately five million and comprises two major adult trauma centres, one paediatric trauma centre and 134 additional trauma services located in metropolitan and regional areas. Each service provides a staged level of care to definitively manage patients in accordance with triage guidelines at an appropriate service. A comprehensive prehospital network supports the trauma system along with an integrated patient transfer service.
Table 1 lists the indicators investigated in the current study. Additional indicators proposed by ACSCOT were unobtainable from VSTR or yielded insufficient cases for analysis and were excluded. Differences in mortality, length of stay and ICU use were compared between cases which were flagged by the indicator (indicating substandard care) and cases that were not flagged by the indicator (indicating good quality care). Analysis of each indicator was limited to cases meeting or not meeting the criteria of the indicator and excluded irrelevant cases to the indicator, eg patients not undergoing head CT scanning were not included in analysis of time to head CT scans. As length of stay can be influenced by deaths soon after admission, only those cases who survived to discharge were included for length of stay analyses.
|
Statistical Analysis
Continuous variables were categorised and analysed using chi-square for differences between flagged and not flagged cases for each indicator. The relationships between the indicator and selected outcomes were explored using logistic regression models adjusting for significant covariate factors identified through univariate analysis (Table 2). Covariates included physiologic measures (systolic blood pressure and pulse rate), Injury Severity Score, mechanism of injury, age, gender, Glasgow Coma Scale and co-morbidities. The Injury Severity Score category of greater than 41 was merged into Injury Severity Score category 25–75 as only 10% of cases recorded a value of more than 41. Prolonged length of stay was analysed as a dichotomous outcome (less than or greater than 29 days), based on the 90th percentile. A P-value of <0.05 was considered statistically significant. Statistical analyses were conducted using STATA Version 9.2 (StataCorp, TX, USA, 2006).
|
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsFundingAcknowledgementsReferences |
|---|
The characteristics of the 5104 cases included for analysis are outlined in Table 3. Median age was 41 years (inter-quartile range: 26.0–61.5) and 72.8% of patients were male. The Injury Severity Score ranged from 16 to 75 with 62.7% scoring 16–25. Car/motorcycle occupants accounted for 48.2% of patients, low and high falls 15.6 and 11.5%, respectively, and pedestrians 9.6% of injuries. The most commonly flagged indicator related to laparotomy performed 2 h after arrival (67.4% of applicable cases), while the least commonly raised indicator concerned deep vein thrombosis, pulmonary embolus and decubitus ulcer development (2.6%) (Table 4).
|
|
For nine of the 14 indicators, statistically significant differences existed between Injury Severity Score for flagged and not flagged cases (indicators 1, 2, 3, 5, 6, 7, 10, 12 and 14), with six of the not flagged groups recording higher proportions of those with Injury Severity Score >25 than those flagged by the indicator (indicating a less injured patient group) (indicators 2, 3, 5, 6, 7 and 10). The lowest mortality rate occurred in patients receiving femoral fracture fixation (2.3%) and the highest in those not receiving femoral fracture fixation (63.2%, P < 0.001, indicator 14). Indicators 1, 3 and 7–10 had significantly higher mortality rates in patients not flagged by the indicator compared with those flagged by the indicator. In contrast, indicators 13 and 14, which relate to lower extremity fractures, found significantly greater mortality in flagged cases.
In several instances, cases flagged by the indicator demonstrated reduced risks for adverse outcomes (Table 5). Patients not receiving trauma team activation (indicator 1) demonstrated reduced risk of lengths of stay greater than 29 days (odds ratio 0.3, P < 0.001) and reduced the risk of ICU stay more than 24 h (odds ratio 0.3, P < 0.001) in comparison to patients receiving team activation. Non-intubated patients with Glasgow Coma Scale <9 (indicator 2) demonstrated a significantly reduced risk of ICU stay more than 24 h (odds ratio 0.1, P < 0.001) which was mirrored by the ACSCOT indicator relating to comatose patients leaving the emergency department (ED) before a mechanical airway was established (odds ratio 0.1, P < 0.001). While the Victorian State Trauma System indicator relating to head CT scans performed more than 2 h after arrival (indicator 3) failed to reach significance, the ACSCOT version restricted to cases with a Glasgow Coma Scale <13 (indicator 5) demonstrated reduced risk of ICU stay more than 24 h compared with those receiving a scan within 2 h (odds ratio 0.7, P = 0.047). Laparotomy delayed by more than 2 h (indicator 7) demonstrated reduced mortality risk (odds ratio 0.3, P = 0.037) as well as reduced risk of hospital stay more than 29 days (odds ratio 0.4, P = 0.026).
|
In contrast, a number of indicators demonstrated associations with adverse outcomes. Patients undergoing abdominal surgery more than 24 h after arrival (indicator 11) were found to have 5.6 times increased odds for the risk of death (P = 0.010) and 4.2 times increased odds for the risk of extended stay (P < 0.001), compared with abdominal surgery within 24 h. The greatest mortality risk occurred in patients flagged by indicators 13 and 14, in which blunt compound tibial fracture treatment occurred more than 8 h after admission (odds ratio 7.8, P = 0.005) and patients not receiving fixation of femoral diaphyseal fractures (odds ratio 55.4, P = 0.002). Patients developing deep vein thromboses, pulmonary emboli or decubitus ulcers were more likely to have extended hospital and ICU stays compared with patients not developing such complications.
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsFundingAcknowledgementsReferences |
|---|
The indicators investigated in this study were designed as intermediate markers of poor quality and considered by experts as associated with poor outcomes. In the absence of an observable measure of ‘quality’, this study has used poor patient outcomes as auxiliary constructs. ‘Good’ indicators will possess additional properties including reliability and sensitivity; however, the link between processes and outcomes remains important and largely depends on who is using the indicators and for what purpose [2]. Clinicians are likely to support process measures linked with patient outcomes that reflect care provided by clinicians [13]. The challenge is to identify indicators capable of monitoring provider performance and which maximize clinician engagement through established process-outcome links.
This study has furthered previous investigations by employing logistic regression models to assess the process-outcome link while adjusting for variables including Injury Severity Score, age, mechanism of injury and physiological measures. This study found that two process indicators were associated with lower mortality (indicators 7 and 8), two with reduced lengths of stay (indicators 1 and 7) and four with reduced risk of ICU admission of more than 24 h (indicators 1, 2, 5 and 6). In contrast, several indicators were associated with increased mortality (indicators 11, 13 and 14), longer lengths of stay (indicators 9, 11 and 12) and increased risk of ICU stay of more than 24 h (indicator 12). These results support previous investigations of similar indicators in US populations [6–9, 14, 15]. The poor construct validity may be due to multiple influences, including the impact of unmeasured confounding variables, problems with indicator or construct measurement, unclear indicator specification or flaws in the theory that indicators can be legitimate measures of the construct [16]. The difficulties in uncovering the sources which compound the process-outcome association, and which compound indicator construct validation, add to tensions between those endorsing indicators in health care and those under scrutiny. In response, the 2006 version of ‘Optimal Resources for the Injured Patient’ published by ACSCOT does not specify indicators be monitored by trauma-care providers [17]. Despite the difficulties of construct validation, trauma indicators have persisted as measurement tools and are widely reported within hospitals and to governments, funding agencies and public interest groups. The use of indicators found to be inversely associated with patient outcomes may favour providers with more severely injured patients, leading to biased conclusions regarding quality of care. The current focus on financial incentives linked with performance measures highlights the need for valid QIs.
In their current format, the ability of these indicators to identify patients at risk of poor outcomes appears limited. Deficiencies in the measurement of indicators and outcomes coupled with indicator specification uncertainty each influence the construct validity of indicators [16]. For example, patients receiving the same interventions at different time points may do so as a function of injury severity not quality of care. Those receiving rapid surgical interventions may have more severe injuries compared with those receiving delayed surgery. Consequently, this selection bias may cause these indicators to identify less injured patients with reduced risks of adverse outcomes, while reflecting appropriate clinical decisions which prioritize treatment [9].
Poor indicator specification may also result in a selection bias which selects patients with different injury profiles as well as different injury severities. Prioritizing treatments for multi-trauma patients will result in immediate management of life-threatening injuries, with delays to less time critical surgery. Strategic delays to tibial fracture management to allow treatment of additional life-threatening injuries may be linked to increased mortality attributable to these other injuries. Although some have reported reduced risks of non-fatal outcomes such as Acute Respiratory Distress Syndrome in patients undergoing immediate fracture fixation [18], others have failed to find associations between surgical timing and wound infection rates, suggesting these indicators function better as treatment guidelines than fixed indicators [19–22]. Refining the indicators to specify the appropriate patient, injury and treatment is a task beyond the scope of this investigation, however, may assist in reducing selection bias and produce more clinically meaningful tools.
The application of general indicators across trauma populations needs to consider the variability of trauma patients and differences in case mix [2]. Criteria restricting the patients for which an indicator is appropriate may improve construct validity. For example, patients receiving abdominal surgery beyond 24 h from admission were at an increased risk of mortality compared with those treated within 24 h. While this indicator appears capable of identifying at-risk patients, subgroups such as small bowel injuries may be susceptible to even shorter delays. However, the indicator must continue to identify adequate cases to allow overall quality assessment [23]. Similarly, VSTR indicator 2 flags patients with Glasgow Coma Scales less than 9 who were not intubated in hospital. Cases with unknown Glasgow Coma Scale and/or intubation status and those already intubated were removed from analysis, significantly reducing the number of eligible cases. Indicators which yield small patient numbers need to be considered in the context of the purpose of the indicator program which may be for local quality improvement, provider decision-making or system monitoring [13]. Refined indicator criteria may reduce the volume of identified cases and, in turn, the ability of the measure to provide an accurate picture of quality.
Measurement of complex concepts frequently use ‘auxiliary’ constructs to quantify relationships [16]. The process-mortality link as an auxiliary construct of "quality" has been complemented in the present study through the exploration of the relationship between indicators and length of stay, which often acts as a target for quality improvement [24]. Non-intubation in patients with Glasgow Coma Scale less than 9 and head CT scans performed beyond 2 h in those with Glasgow Coma Scale less than 13 were both found to have reduced risks for ICU stay, perhaps due to decreased need for ventilation and post-surgical management [25]. An abnormal Glasgow Coma Scale resulting from intoxication likely eliminates the need for ventilation, radiological imaging or ICU admission. Therefore, the indicator may be more reflective of clinical decision-making than quality of care.
Underlying deficiencies in the theoretical concept linking processes with patient-level outcomes may also explain poor construct validation results [16]. Measures of ICU use and length of stay often act as surrogates of resource use and are of interest to hospital administrators, financiers and clinical staff [26]. However, viewed as a process itself, the relationship between length of stay and indicators focused on complications may be difficult to establish. For example, it is difficult to determine a causal relationship for patients developing deep vein thromboses, pulmonary emboli or decubitus ulcers and increased length of stay. Similarly, patients undergoing abdominal or cranial surgery more than 24 h following admission had longer hospitalizations and greater ICU use. Surgical delays in patients with small bowel injuries have previously been reported to increase lengths of stay (9.1 days compared with 25.7 days, P < 0.001) [27] possibly as a consequence of greater post-operative complications [28]. Further, Nakayama et al. [9] reported four paediatric patients identified by the indicator as specified by ACSCOT (which combines cranial, abdominal, thoracic and vascular surgeries) received inappropriate care, prompting conclusions that the indicator was a useful criterion.
Limitations
This study had a number of limitations. While use of trauma registries for measuring quality has been proposed in the literature [11], only routinely collected variables could be included in analyses. In addition, the accuracy of the Injury Severity Score to accurately capture the true extent of injury has also been questioned [29]. As variation attributable to measurement error greatly influences construct validity, the results from this investigation also reflect registry data limitations, its effect upon statistical modeling and the implications this has for case-mix adjustment.
Indicators of complex constructs do not provide definitive evidence of poor quality, rather they suggest areas, processes or systems which may benefit from review. As part of the construct validation process, indicators need regular re-definition to reflect new findings, theories and clinical practices that challenge and clarify existing concepts [5, 7]. The VSTR indicators performed in a manner similar to those specified by ACSCOT. These results have supported reviews of the VSTR indicators, such as activation of the trauma team at major trauma services, which is being revised to better reflect an appropriate patient group for whom the trauma team is warranted. This study has offered insight into the use of registries to analyse indicator data; however, any re-definition of indicator criteria needs to reflect current clinical opinions and standards of care best gained from prospective trials [8]. Developing new indicators from rigorous clinical investigations rather than solely on expert opinion is likely to produce more informative measures with greater ability to influence quality of care.
All VSTR data collectors have formal clinical training and comprehensive data coding and abstraction training by VSTR staff and external bodies. In addition, the registry incorporates manual and automated data quality procedures. Despite these efforts, data accuracy remains largely dependent on clinical record documentation which can vary between patient and hospital. In addition, exclusion criteria restricted this investigation to those with blunt trauma, an Injury Severity Score greater than 15 and those aged 16 or more, limiting the generalizability of these findings to other trauma populations.
Finally, the theory that indicators relate to patient level outcomes, such as individual mortality, may account for the limited construct validity seen in this study. Indicators are influenced by the level at which they measure performance, be it practitioner, unit, hospital or system. The move to measure quality beyond individual performance recognizes good quality as not solely reliant upon the knowledge and skills of clinical staff. Clinicians require additional resources, equipment and personnel delivered through systems to provide optimal patient outcomes. An organized management approach through a structured trauma system has been recognized as an essential component in achieving best quality care [30]. Therefore, indicator construct validity may be better measured through associations with provider-level quality rather than practitioner performance or individual patient outcomes. Indicators not associated with individual patient outcomes may still prove useful in measuring hospital and system quality. Additional analyses of these indicators, singly and in combination across providers and systems, would establish the ability of the indicators to assess quality of care at the system level.
|
Conclusions |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsFundingAcknowledgementsReferences |
|---|
Indicators without established accuracy and efficacy may incur considerable economic costs without yielding an improvement in quality of care. This study has demonstrated the challenges of indicator construct validation through the difficulties in identifying links between processes and outcomes. Future investigations which shift the focus from the performance of individual indicators for individual patients to that of a suite of indicators for comparing trauma providers may prove more useful.
|
Funding |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsFundingAcknowledgementsReferences |
|---|
The Centre of Research Excellence in Patient Safety is funded by the Australian Commission on Safety and Quality in Health Care and is designated as a National Health and Medical Research Council (NHMRC) Centre of Research Excellence. The Victorian State Trauma Registry is funded by the Department of Human Services, Victoria, and the Victorian Trauma Foundation. Mr. Willis is supported by an NHMRC Postgraduate Public Health Scholarship.
|
Acknowledgements |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsFundingAcknowledgementsReferences |
|---|
We would also like to acknowledge the assistance of Mr Andrew Hannaford and Ms Pam Simpson, as well as all trauma services supporting the Victorian State Trauma Registry.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsFundingAcknowledgementsReferences |
|---|
- Freeman T. Using performance indicators to improve health care quality in the public sector: A review of the literature. Health Serv Manage Res (2002) 15:126–37.
[Abstract/Free Full Text] - Mainz J. Defining and classifying clinical indicators for quality improvement. Int J Qual Health Care (2003) 15:523–30.
[Abstract/Free Full Text] - Rubin HR, Pronovost P, Diette GB. From a process of care to a measure: the development and testing of a quality indicator. Int J Qual Health Care (2001) 13:489–96.
[Abstract/Free Full Text] - Turpin RS, Darcy LA, Koss R, et al. A model to assess the usefulness of performance indicators. Int J Qual Health Care (1996) 8:321–9.
[Abstract/Free Full Text] - Smith GT. On construct validity: issues of method and measurement. Psychol Assess (2005) 17:396–408.[CrossRef][Web of Science][Medline]
- Copes WS, Staz CF, Konvolinka CW, et al. American College of Surgeons audit filters: associations with patient outcome and resource utilization. J Trauma (1995) 38:432–8.[Web of Science][Medline]
- Cryer HG, Hiatt JR, Fleming AW, et al. Continuous use of standard process audit filters has limited value in an established trauma system. J Trauma (1996) 41:389–94.[Web of Science][Medline]
- Nayduch D, Moylan J, Snyder BL, et al. American College of Surgeons trauma quality indicators: an analysis of outcome in a statewide trauma registry. J Trauma (1994) 37:565–73.[Web of Science][Medline]
- Nakayama DK, Gardner MJ, Waggoner T, et al. Audit filters in quality assurance in pediatric trauma care. J Pediatr Surg (1993) 28:19–25.[CrossRef][Web of Science][Medline]
- Willis CD, Gabbe BJ, Cameron PA. Measuring quality in trauma care. Injury (2007) 38:527–37.[CrossRef][Web of Science][Medline]
- Cameron PA, Gabbe BJ, McNeil JJ, et al. The trauma registry as a statewide quality improvement tool. J Trauma (2005) 59:1469–76.[Web of Science][Medline]
- Cameron PA, Finch CF, Gabbe BJ, et al. Developing Australia's first statewide trauma registry: what are the lessons? ANZ J Surg (2004) 74:424–8.[CrossRef][Web of Science][Medline]
- Rubin HR, Pronovost P, Diette GB. The advantages and disadvantages of process-based measures of health care quality. Int J Qual Health Care (2001) 13:469–74.
[Abstract/Free Full Text] - Kissoon N, Tepas JJ III, Peterson RJ, et al. The evaluation of pediatric trauma care using audit filters. Pediatr Emerg Care (1996) 12:272–6.[Web of Science][Medline]
- Rhodes M, Sacco W, Smith S, et al. Cost effectiveness of trauma quality assurance audit filters. J Trauma (1990) 30:724–7.[Web of Science][Medline]
- Westen D, Rosenthal R. Improving construct validity: Cronbach, Meehl, and Neurath's ship. Psychol Assess (2005) 17:409–12.[CrossRef][Web of Science][Medline]
- American College of Surgeons Committee on Trauma. Resources for Optimal Care of the Injured Patient (2006) Chicago: American College of Surgeons.
- Bone LB, Johnson KD, Weigelt J, et al. Early versus delayed stabilization of femoral fractures: a prospective randomized study. 1989. Clin Orthop (2004) 422:11–16.[CrossRef][Medline]
- Charalambous CP, Siddique I, Zenios M, et al. Early versus delayed surgical treatment of open tibial fractures: effect on the rates of infection and need of secondary surgical procedures to promote bone union. Injury (2005) 36:656–61.[CrossRef][Web of Science][Medline]
- Harley BJ, Beaupre LA, Jones CA, et al. The effect of time to definitive treatment on the rate of nonunion and infection in open fractures. J Orthop Trauma (2002) 16:484–90.[CrossRef][Web of Science][Medline]
- Khatod M, Botte MJ, Hoyt DB, et al. Outcomes in open tibia fractures: relationship between delay in treatment and infection. J Trauma (2003) 55:949–54.[Web of Science][Medline]
- Bednar DA, Parikh J. Effect of time delay from injury to primary management on the incidence of deep infection after open fractures of the lower extremities caused by blunt trauma in adults. J Orthop Trauma (1993) 7:532–5.[CrossRef][Web of Science][Medline]
- Mainz J. Developing evidence-based clinical indicators: A state of the art methods primer. Int J Qual Health Care (2003) 15:i5–11.
[Abstract/Free Full Text] - O'Keefe GE, Jurkovich GJ, Maier RV. Defining excess resource utilization and identifying associated factors for trauma victims. J Trauma (1999) 46:473–8.[Web of Science][Medline]
- Freeman VA, Walsh J, Rudolf M, et al. Intensive care in critical access hospitals. J Rural Health (2007) 23:116–23.[CrossRef][Web of Science][Medline]
- Thomas JW, Guire KE, Horvat GG. Is patient length of stay related to quality of care? Hosp Health Serv Adm (1997) 42:489–507.[Web of Science][Medline]
- Fang JF, Chen RJ, Lin BC, et al. Small bowel perforation: is urgent surgery necessary? J Trauma (1999) 47:515–20.[Web of Science][Medline]
- Fakhry SM, Brownstein M, Watts DD, et al. Relatively short diagnostic delays (<8 hours) produce morbidity and mortality in blunt small bowel injury: an analysis of time to operative intervention in 198 patients from a multicenter experience. J Trauma (2000) 48:408–14.[Web of Science][Medline]
- Rutledge R, Hoyt DB, Eastman AB, et al. Comparison of the Injury Severity Score and ICD-9 diagnosis codes as predictors of outcome in injury: analysis of 44,032 patients. J Trauma (1997) 42:477–87. discussion 487–9.[Web of Science][Medline]
- Atkin C, Freedman I, Rosenfeld JV, et al. The evolution of an integrated State Trauma System in Victoria, Australia. Injury (2005) 36:1277–87.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||