Trends in Injury Outcomes Across Canadian Trauma Systems

Key Points

Question  How have mortality, hospital length of stay, and unplanned readmission evolved in Canadian trauma systems between 2006 and 2012?

Findings  This national cohort study found a statistically significant decrease in risk-adjusted mortality incidence (from 12.1% to 9.9%) and mean hospital length of stay (from 11.6 to 10.6 days), but no change in incidence of unplanned readmissions. Trends varied across provinces.

Meaning  Improvements in injury outcomes were observed in some Canadian provinces between 2006 and 2012, indicating that the configuration of trauma systems may be influential.

Importance  In response to the burden of injury, the structure of injury care has changed considerably across Canada in the past decade. However, little is known about how patient outcomes have evolved.

Objective  To evaluate trends in mortality, hospital length of stay, and unplanned readmission in Canadian trauma systems between 2006 and 2012.

Design, Setting, and Participants  A pan-Canadian retrospective cohort study was conducted among adults admitted for major injury to a Canadian level I or II trauma center between April 1, 2006, and March 31, 2012. Data analysis was conducted from April 15 to December 3, 2015.

Exposures  Trauma centers and systems.

Main Outcomes and Measures  Multilevel generalized linear models were used to evaluate trends in the risk-adjusted incidence of mortality and readmission and risk-adjusted mean length of stay. Trend analyses were conducted globally and by province.

Results  Among 78 807 patients (mean [SD] age, 50.7 [22.0] years; 22 540 women and 56 267 men) admitted for major injury during the study period, risk-adjusted mortality decreased from 12.1% (95% CI, 9%-16.1%) to 9.9% (95% CI, 7.4%-13.3%; P < .001) and mean length of hospital stay decreased from 11.6 (95% CI, 9.9-13.6) to 10.6 (95% CI, 9.1-12.5) days (P < .001). Statistically significant reductions in mortality were observed for Ontario (12% [95% CI, 10.7%-13.6%] to 8% [95% CI, 6.9%-9.2%]; P < .001), Alberta (12% [95% CI, 10%-14.3%] to 9.1% [95% CI, 7.7%-10.8%]; P = .02), and Manitoba (13% [95% CI, 9.1%-18.4%] to 11.1% [95% CI, 8.3%-14.7%]; P = .04). Risk-adjusted hospital stay decreased significantly in Québec (11.6 [95% CI, 11.1-12] to 9.1 [95% CI, 8.9-9.5] days; P < .001), British Columbia (12.5 [95% CI, 12-13.1] to 11.4 [10.9-11.9] days; P < .001), and Ontario (10.1 [95% CI, 9.8-10.4] to 9.8 [95% CI, 9.5-10.1] days; P < .001). No change in the incidence of readmission was observed.

Conclusions and Relevance  We observed an 18.2% relative decrease in risk-adjusted mortality in Canadian trauma centers during the study period, representing 248 additional lives saved in 2012 vs 2006. Risk-adjusted mean hospital stay decreased by 8.6%, representing nearly 10 000 hospital days saved. A better understanding of the structures and processes behind observed improvements is needed to further reduce the burden of injury in Canada.

Injury represents a loss of human potential and a burden to society because, compared with many individuals who have chronic diseases, patients who have sustained an injury are relatively young and previously in good health.¹ The introduction of trauma systems, representing an organized, multidisciplinary response to injury across the care continuum, has improved injury outcomes in many high-income countries.² The concept of organized trauma care was founded in the United States in the 1970s and was introduced in most major Canadian cities in the 1980s.³ Trauma Association of Canada accreditation guidelines were then published in 1993, accreditation activities were initiated in 1995, and accreditation guidelines were revised in 2003, 2007, and 2011.⁴ Major characteristics of Canadian trauma systems are in the eTable in the Supplement. The structure of Canadian trauma care has changed considerably during the past decade, and the rate and extent of change has varied significantly across provinces.⁵ However, little is known about the evolution of the outcomes of injury in Canadian trauma centers. In this study, we aimed to assess trends in mortality, hospital length of stay (LOS), and unplanned readmission in Canadian trauma centers between 2006 and 2012 overall and by province.

We conducted a retrospective, multicenter cohort study based on admissions with injury to level I and II trauma centers across Canada between April 1, 2006 (implementation of the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision), and March 31, 2012 (the last year of available data). Patient-level data were extracted from the National Trauma Registry, which contains data on patients admitted for major trauma to designated level I and II trauma centers across Canada, and abstracted from patient files by trained data coders using a standardized data dictionary. Anatomical injury was coded with the Abbreviated Injury Scale (AIS).⁶ The AIS score is a measure of anatomical injury severity attributed by expert consensus to each injury code in the AIS lexicon. The score ranges from 1 (least severe) to 6 (most severe). Until 2013, the registry was centralized at the Canadian Institutes of Health Information, where it was subject to data quality control procedures. Each injury admission in the National Trauma Registry was then linked to the same admission in the Discharge Abstract Database using patients’ unique health insurance number, hospital to which the patient was admitted, and admission date. We were then able to obtain information on admissions and associated diagnoses up to 12 months before and 12 months after the index admission for injury.

We included all adults (≥16 years) with major trauma, defined as an Injury Severity Score (ISS) greater than 12, except patients who were coded as dead on arrival and those who arrived with no vital signs and died within 30 minutes. In-hospital mortality was evaluated for the entire study population. Hospital LOS was defined as the number of days of acute care in the index hospital and was evaluated among patients discharged alive. Unplanned readmission was defined as an unplanned return to any acute care hospital within 30 days of discharge for the index injury admission. Admissions after transfer from the trauma center and elective admissions for planned interventions were not considered to be unplanned readmissions. Readmission was evaluated in patients discharged alive from a trauma center when their injury admission in the trauma registry could be linked to hospital discharge data. Ethics approval for this study was obtained from the Centre Hospitalier Universitaire de Québec Research Ethics Board. No patient consent was required as this was a clinical registry that was denominalized before the data were transferred for analyses.

Data analysis was conducted from April 15 to December 3, 2015. We evaluated trends in injury outcomes using a multilevel generalized linear model. A logit link function was used to model the incidence of mortality and readmission, and an identity link function was used to model log-hospital stay. Trauma center and province were entered as random effects to control for clustering. Risk adjustment was based on literature review and expert opinion. Mortality was adjusted for age, mechanism of injury, anatomical injury severity (AIS score of the 2 most severe injuries), body region of the most severe injury, physiological derangement (Glasgow Coma Scale [GCS] score, respiratory rate, and systolic blood pressure on arrival at the emergency department), and direct vs indirect transport to the facility where definitive care was provided. Hospital stay was adjusted for age, sex, mechanism of injury, anatomical injury severity (maximum AIS score in each body region), the GCS score on admission, and direct vs indirect transport to the facility where definitive care was provided. Readmission was adjusted for age, sex, comorbidities (as described by Elixhauser et al⁷), anatomical injury severity (ISS), body region of the most severe injury, and the number of admissions in the 12 months before injury. The GCS score measures level of consciousness and varies from 3 (deep unconsciousness) to 15 (fully alert).⁸ Quantitative variables thought to have a nonlinear association with the outcome under evaluation were modeled using fractional polynomials.⁹ Trends over time were evaluated by entering year as a linear term in multilevel regression models. We adjusted for cohort effects by adding a term for year of birth to each model. Analyses were performed for the entire study population and for each province.

The GCS score, respiratory rate, and systolic blood pressure on arrival were missing for 15 761 [19.6%], 16 786 [20.9%], and 3152 [3.9%] of 80 353 data observations, respectively. Missing data on these variables were simulated with multiple imputation. We conducted quantitative and qualitative evaluations of missing data mechanisms to evaluate the plausibility that variables were missing at random given other data available in the National Trauma Registry. Because data were missing mostly for patients with extracranial trauma and for those who were sedated or intubated on arrival, and were correlated with data on head injury, injury severity, age, prehospital interventions, and transfer status, the missing at random postulate was considered plausible. Imputation models included all independent and dependent variables used in respective analysis models.¹⁰

We conducted sensitivity analyses to evaluate the robustness of our findings to analytical assumptions. We thus repeated trend analyses excluding the following patients: those with missing physiological data, those 85 years or older, and those transferred from another acute care hospital. We also repeated trend analyses for mortality and hospital stay with adjustment for comorbidity in observations that could be matched with hospital discharge data. We then evaluated the correlation between annual estimates in the original analysis and in each of the sensitivity analyses. Finally, we repeated LOS analyses, including deaths, in a competitive risks framework to assess the potential effect of survival bias. A Cox proportional hazards regression model was used to model the hazard of discharge with deaths censored at the maximum observed LOS.¹¹

All analyses were performed using SAS software, version 9.4 for Windows (SAS Institute Inc). Statistical tests were 2-sided with P < .05 considered significant.

The National Trauma Registry comprehensive data set comprised 80 353 adults admitted for major trauma during the study period. We excluded 968 patients (1.2%) who were dead on arrival. Data from New Brunswick (2 level I trauma centers and 1 level II trauma center) and Saskatchewan (1 level I trauma center and 1 level II trauma center) could not be used, as anatomical injury coding was based on an updated version of the AIS that was incompatible with the National Trauma Registry framework. An additional 578 patients (0.7%) were excluded because they had missing information on injury severity. A total of 78 807 patients (98.1% of all patients admitted for major trauma) were included in the final study sample and used for analyses on in-hospital mortality. Of these patients, 70 425 (87.6% of all patients admitted for major trauma) were discharged alive and were used for analyses on hospital stay. Of these patients, 59 228 (73.7% of all patients admitted for major trauma) could be linked with the Discharge Abstract Database and were used for analyses on unplanned readmission. Hospital admissions that could not be linked with discharge data were non-Canadian residents (362 [0.5%]) and all observations from Manitoba and Newfoundland and Labrador. Compared with patients included in analyses on readmission, those who were excluded because they could not be linked with discharge data were younger (mean age, 45 vs 50 years) but had similar injury severity (mean ISS, 23 for both groups).

The proportion of patients 65 years or older increased from 25.7% in 2006 to 33.9% in 2012 (Table 1). The proportion of patients with a critical injury (ISS≥25) and treated in a level II facility remained fairly stable from 2006 to 2012 (critical injury, 45% vs 42.9%; treatment in a level II facility, 23.3% vs 22.3%). The proportion of patients in a coma on arrival (GCS score ≤8) and who were transferred from another hospital was higher in 2006 than in 2012 (GCS score ≤8, 12.7% vs 11.8%; transferred from another hospital, 57.3% vs 46.6%). Crude mortality appeared to change little from 2006 to 2012 (11% vs 10.8%), whereas the proportion readmitted within 30 days increased slightly (8.5% vs 9.2%) and hospital stay decreased slightly (mean, 11.5 vs 10.7 days).

Risk-adjusted mortality in the entire population decreased from 12.1% (95% CI, 9%-16.1%) to 9.9% (95% CI, 7.4%-13.3%) during the study period (Table 2). Decreases were observed between 2006 and 2007 and between 2008 and 2011, with the biggest decrease seen between 2006 and 2007. The trend remained statistically significant between 2007 and 2012. During the entire study period, risk-adjusted mortality was lowest in Québec (7%; 95% CI, 6.5%-7.6%) and highest in Manitoba (14.2%; 95% CI, 12.3%-16.4%). Decreases in mortality were observed in all provinces from 2006 to 2012 but were statistically significant for only Ontario (12% [95% CI, 10.7%-13.6%] to 8% [95% CI, 6.9%-9.2%]; P < .001), Alberta (12% [95% CI, 10%-14.3%] to 9.1% [95% CI, 7.7%-10.8%]; P = .02), and Manitoba (13% [95% CI, 9.1%-18.4%] to 11.1% [95% CI, 8.3%-14.7%]; P = .04).

Risk-adjusted hospital LOS in all provinces decreased from a mean of 11.6 days (95% CI, 9.9-13.6) in 2006 to 10.6 (95% CI, 9.1-12.5) in 2012 (Table 3). Decreases were observed from 2006 to 2008 and from 2009 to 2010. During the entire study period, mean hospital LOS was lowest in Ontario (9.6 days; 95% CI, 9.5-9.7) and highest in Newfoundland and Labrador (15.5 days; 95% CI, 14.4-16.6). The decrease in mean hospital LOS from 2006 to 2012 was statistically significant for Québec (11.6 [95% CI, 11.1-12] vs 9.1 [95% CI, 8.9-9.5] days; P < .001), British Columbia (12.5 [95% CI, 12-13.1] vs 11.4 [95% CI, 10.9-11.9] days; P < .001), and Ontario (10.1 [95% CI, 9.8-10.4] vs 9.8 [95% CI, 9.5-10.1] days; P < .001).

The 30-day risk-adjusted incidence of unplanned readmission did not change between 2006 and 2012 globally or in any province (Table 4). Risk-adjusted incidence of unplanned readmission in the entire population was 8.9% (95% CI, 8.1%-9.7%) in 2006 and 9% (95% CI, 8.3%-9.9%) in 2012 (P = .80 for trend).

Correlations between annual risk-adjusted estimates of mortality, readmission, and mean hospital LOS in the original analysis were strong when excluding observations with missing data (r = 0.85, 0.84, and 0.89, respectively) and very strong when excluding patients 85 years or older (r = 0.98, 0.95, and 0.96, respectively), patients presenting more than 48 hours after injury (r = 0.98, 0.97, and 0.99, respectively), and patients transferred from another hospital (r = 0.99, 0.98, and 0.97, respectively). Very strong correlation with original estimates was also observed when comorbidities were included in models of mortality and LOS for patients who could be linked with hospital discharge data (r = 0.99, 0.98, and 0.99, respectively). When deaths were included in a competitive risks framework for LOS analyses, the national decrease in LOS remained statistically significant (coefficient estimate for trend, 0.009; 95% CI, 0.005-0.013; P < .001). Provincial decreases remained significant for Quebec (coefficient estimate for trend, 0.0252; 95% CI, 0.0170-0.0334; P < .001) and Ontario (coefficient estimate for trend, 0.007; 95% CI, 0.001-0.014; P = .049), but not for Manitoba (coefficient estimate for trend, –0.009; 95% CI, –0.030 to 0.013; P = .40).

In this retrospective, multicenter cohort study, we observed an 18.2% relative decrease in risk-adjusted mortality in Canadian trauma centers during the study period, representing 248 additional lives saved in 2012 vs 2006. Risk-adjusted mean hospital stay decreased by 8.6%, representing nearly 10 000 hospital days saved. No change was observed in the incidence of unplanned readmission. In analyses by province, significant relative decreases in mortality were observed for Ontario (33.3%), Alberta (24.2%), and Manitoba (14.6%), whereas relative decreases in hospital LOS were significant in Québec (21.6%), British Columbia (8.8%), and Ontario (3%).

National evaluations of trends in injury outcomes are scarce, but our results are consistent with those of studies observing improvements in survival up to 10 years after the implementation of trauma systems.^2,12-14 Changes in primary and secondary prevention policies have probably played a role in reducing the burden of injuries, but the reductions in mortality observed in our study could only be explained by improvements in tertiary prevention strategies, as our denominator was patients hospitalized for injury and we adjusted for injury severity. Observed improvements could be owing to changes in trauma care organization over time and/or improvements in treatment strategies and adherence to evidence-based standards of care. The interprovincial differences in time trends may be owing to national variations in trauma system structure (eTable in the Supplement). For example, the implementation of an inclusive trauma system in Alberta in 2008 led to changes in triage patterns that were associated with improved hospital mortality.¹⁵

Several studies have reported higher use of resources in trauma centers than in hospitals that are not designated as trauma centers, but care in trauma centers is more cost-effective when additional lives saved are considered.^16,17 Other studies have reported decreases in costs and LOS after implementation of a trauma system¹⁸ or after the evolution of an exclusive system (level I and II centers in large urban areas only) to an inclusive system (network of level I to IV trauma centers covering a geographical territory).¹⁵ The reduction in hospital LOS observed in Canadian trauma centers may be owing to resource constraints, improvements in the efficiency of acute trauma care over time, or the reduction in risk factors of prolonged hospital stay, such as complications. The fact that readmission rates have not increased over time indicates that inappropriately early discharge is probably not a factor.

Trauma centers have been reported to have a lower incidence of readmissions than hospitals not designated as trauma centers.¹⁹ However, previous studies have not evaluated the influence of evolving trauma systems on readmission. The absence of change in readmission rates in Canadian trauma centers could be owing to a lack of quality interventions targeting unplanned readmission.²⁰ The lack of improvement in readmission rates may also be owing to the increase in patients surviving critical injuries. However, similar increases in LOS would be expected if this were the case.

This study is based on a nationally representative sample and detailed clinical data subject to standardized data collection procedures and data quality control mechanisms. The Canadian National Trauma Registry is not population-based but is estimated to represent 90% of reportable Canadian major trauma admissions.²¹ Finally, trend analyses included robust risk adjustment, respected the hierarchical nature of data, and included adjustment for cohort effects to account for changes in life expectancy over birth cohorts.

Some limitations may influence the interpretation of our study results. First, our study was designed to assess the effect of trauma center care and thus did not include prehospital deaths, representing approximately 50% of all injury fatalities. Studies evaluating the global burden of injury across Canada should include prehospital deaths, but detailed clinical data are currently unavailable for these cases, precluding robust risk adjustment. Second, selection bias could have occurred if access to level I or II trauma centers changed significantly during the study period. This is likely in provinces that have implemented significant changes in triage criteria or trauma system configuration, such as Alberta.¹⁵ Limiting analyses to major trauma and using robust adjustment for changes in patient case mix should have minimized the influence of selection bias. However, differential changes in trauma system structure across provinces may partly explain the difference in outcome trends. Third, changes in data collection procedures over time, in particular an improvement in the completeness of injury coding, may explain improvements in risk-adjusted outcomes. However, as data collection is mandatory and based on uniform collection criteria that have not changed over time (no change in coding versions of the AIS or International Statistical Classification of Diseases and Related Health Problems, Tenth Revision), the risk of significant modification in data capture during the study period is minimal. Fourth, data on physiological reaction to injury were simulated for 20% of study observations. However, multiple imputation of physiological data in trauma registries has been shown to lead to valid effect estimates,²² the proportion of missing data was never systematically associated with outcomes and calendar year, and results from analyses restricted to complete observations did not change study conclusions. Fifth, the observed decrease in hospital LOS could be owing to an increase in interhospital transfers to complete the acute phase of care and reductions in mortality may be owing to an increase in deaths after discharge or changes regarding end-of-life decisions. However, we did not observe any trend in discharge to acute care during the study period (difference of <1% between 2006 and 2012), and previous research suggests that there is a near-perfect correlation between in-hospital mortality and 30-day mortality, including deaths after discharge.²³ Finally, to truly monitor progress in injury outcomes, we require national data on short- and long-term functional outcomes and on quality of life after injury.

We observed a reduction in mortality and hospital LOS in Canadian trauma centers between 2006 and 2012. Results indicate that efforts should be made to reduce unplanned readmissions. Future research should aim to improve our understanding of the structures and procedures of care that drive optimal injury outcomes.

Accepted for Publication: July 20, 2016.

Corresponding Author: Lynne Moore, PhD, Population Health and Optimal Health Practices Research Unit, Trauma, Emergency, and Critical Care Medicine, Centre de Recherche du Centre Hospitalier Universitaire de Québec Hôpital de l’Enfant-Jésus, Université Laval, 1401, 18e rue, local H-012a, Québec City, QC G1J 1Z4, Canada (lynne.moore@fmed.ulaval.ca).

Published Online: November 9, 2016. doi:10.1001/jamasurg.2016.4212

Author Contributions: Dr Moore had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis

Study concept and design: Moore, Hameed, Kortbeek, Turgeon.

Acquisition, analysis, or interpretation of data: Moore, Stelfox, Evans, Hameed, Yanchar, Simons, Bourgeois, Clément, Turgeon, Lauzier.

Drafting of the manuscript: Moore, Evans, Kortbeek.

Critical revision of the manuscript for important intellectual content: Moore, Stelfox, Evans, Hameed, Yanchar, Simons, Bourgeois, Clément, Turgeon, Lauzier.

Statistical analysis: Moore.

Administrative, technical, or material support: Moore, Yanchar, Bourgeois, Lauzier.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was supported by a Canadian Institutes of Health Research New Investigator Award (Drs Moore, Stelfox, and Turgeon), a Fonds de Recherche du Québec–Santé New Investigator Award (Dr Lauzier), and research grant 110996 from the Canadian Institutes of Health Research (Dr Moore).

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.