Complications, burden and in-hospital death among hospital treated injury patients in Victoria, Australia: a data linkage study

Morbidity data was extracted from the Victorian Admitted Episodes Dataset (VAED), which contains unit records of all public and private hospital admissions in Victoria. Among the information stored in the VAED are patient demographics, morbidity, external causes of morbidity and patient discharge status. Up to forty diagnosis codes can be recorded per patient episode; this includes external cause information. Morbidity information in the VAED is coded according to the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modifications (ICD-10-AM) [41]. Treatment costs at hospitals for each episode were extracted from the Victorian Cost Data Collection (VCDC), these represent the total actual costs of the episodes as reported by the hospital.

VAED hospital admissions data were internally linked within the VAED and linked with VCDC data using variables such as campus code, unique record number and department of health key. Linkage within the VAED was enhanced with the Better Patient Data dataset (which is a separate collection from clinical data, containing identifying information such as patients’ names, addresses, sex, date of birth etc.) in order to create across-service patient records. CVDL used deterministic data linkage with some fuzzy matching to allow for slight variation in the linkage variables, such as incorrect names and dates. No clerical review of linkages was carried out but a number of general quality checks were performed by CVDL.

Injury cases were selected using a protocol established for previous national reporting in Australia [1]: VAED records with an injury diagnosis (ICD-10-AM diagnosis code in the range “S00” to “T75” or “T79” in the principal diagnosis code). The selected cases were Victorian residents with their index injury hospital admission during the study period. ‘Index injury’ is the first injury record in the VAED for a patient during that period. Records with an admission source indicating a transfer from another hospital or statistical separation (change in care type within the same hospital) in consecutive records were all considered to be part of the index injury hospitalisation.

Patients with an index injury admission between 01 July 2012 and 30 June 2014 in the VAED were selected. Using internally linked data within the VAED (i.e. using a patient-specific identifier), patients were followed up for all hospital admissions subsequent to their index injury admission. The follow-up times were varied by outcome: one year for in-hospital death and cost, two years for LOS, discharge destination, intensive care unit (ICU) stay and mechanical ventilator (MV) use, and thirty days for readmissions, from the index injury date.

Patients were classified into six age groups, closely aligned to the groups used by the Australian Bureau of Statistics [45]; 0–14, 15–24, 25–44, 45–64, 65–84, 85 years and over. Body region groups were created by regrouping the “blocks” contained in the ICD-10-AM Chapter 19 [41] and injury groups according to the injury “type” classification at the three-character level in the same chapter; both based on the principal diagnosis code. Injury severity was calculated using the ICD-based Injury Severity Score (ICISS) [46]. Computation of the ICISS was as follows. First, each injury diagnosis in every record was allocated a survival risk ratio (SRR) (proportion of survivors among all patients with that particular ICD-10-AM diagnosis); SRRs were sourced from the National Injury Surveillance Unit [47]. Next, using the worst injury method [48], the lowest SRR was assigned as the ICISS. Finally, a serious injury was considered to be one with an ICISS less than or equal to 0.941 (survival probability of 94.1%) [49]. Socio economic status was classified as per the Socio-Economic Indexes for Areas (SEIFA) [50]. The specific SEIFA used in this study was the Index of Relative Socio-Economic Advantage and Disadvantage (IRSAD); with state deciles based on Statistical Local Areas (SLAs). A low score indicates relatively greater disadvantage. Comorbidities were identified using a combination of the Charlson Comorbidity Index (CCI) [51] and Elixhauser Comorbidity Measure (ECM) [52] comorbidity groups. The ICD-10 codes for these conditions were extracted mainly from Quan et al. (2005) [53] and Sundararajan et al. [54]. This included all 17 conditions from the CCI list and all 30 conditions from the ECM list. Thirty-four comorbidity groups were identified in total by combining these two lists, of which three were eventually excluded. The excluded groups were weight loss and fluid-electrolyte disorders, which were more likely to be complications or symptoms, rather than comorbidities, and the group “other neurological disorders” which were a diverse selection of codes that did not narrow down to a specific condition. Additional codes from the Sundararajan list included the following: B23 (HIV/AIDS), C80 (any malignancy), R02 (peripheral vascular disease), N01, N072-N074 (renal disease).

Descriptive analysis was undertaken to determine the distribution of outcomes across selected demographic and patient characteristics of the study population; these were captured as frequencies and proportions for binary outcomes, means with 95% confidence intervals (CIs) and medians with inter-quartile ranges (IQRs) for continuous variables.

Sample sizes were large in this study, therefore effect sizes were more meaningful than p values to establish associations between outcomes and population factors [55]. Univariate modelling was carried out using: (i) logistic regression for discharge destination, in-hospital death, CCS use and readmissions, (ii) negative binomial regression for LOS and number of complications and (iii) linear regression with a natural log transformation for costs. The aim of this study was to describe outcomes and their association with factors, therefore establishing associations between outcomes and base factors using univariate analysis was adequate. Additional analysis for associations between outcomes were also carried out using univariate modelling. Once again using simple tests for associations such as the chi squares were not possible with this large dataset, because all results are highly significant and therefore does not provide further insights. Association between the dependent and independent variables were assessed based on effect sizes (using odds ratios (ORs) for logistic regressions, incident rate ratios (IRRs) for negative binomial regressions and beta coefficients for linear regressions, and their confidence intervals); with an effect size of ±30% considered as notable. The normative (logical) or largest categories were used as reference groups for regression analysis. SAS software, Version 9.4 of the SAS System for Windows [56] and Stata 14.0 (StataCorp) [57] were used to analyse the data.

In-hospital death increased with age. Compared to patients in the 65–84-year age group, those who were 85 years and over had over two and a half times the odds of death in hospital while death was far less likely among all other age groups. In-hospital death was not notably different for females and males. Likelihood of death increased with the number of comorbidities; patients with more than three comorbidities had thirty-six times the odds of an in-hospital death than those with no comorbidities. Serious injury compared to non-serious injury had a greater impact on in-hospital death (nearly 17 times the odds). Among patients with serious injury, 5.0% died in hospital. Patients with hip & thigh injuries had around 3.4 times the odds of death in hospital than those with head injuries. Apart from hip & thigh, injury to any other body region had a lesser likelihood of death or was not notably different from the likelihood of death from head injuries. Compared to intracranial injury, most other types of injury had more than 30% lower odds of death. There was no notable difference in the odds of in-hospital death among those in more disadvantaged socio-economic areas against more advantaged areas (Tables 2 and 3).

A total of 49,440 complications were identified in this study population. The top four classes of complications were cardiovascular (13.7%), gastrointestinal (12.3%), metabolic disorders (12.1%) and genitourinary (10. 7%) complications. The most common subclasses were electrolyte disorders without dehydration (6.4%), hypotension (5.8%), alterations to mental state (5.4%), constipation (5.1%) and cardiac arrhythmias, conduction disturbances and abnormal heart beat (4.8%) (Table 4).

The average number, and odds of occurrence of complications increased with increasing age and with increasing number of comorbidities. The odds of complications were higher for patients with serious injury. The likelihood of complications varied by body region, with injuries to neck, thorax, shoulder and upper arm and some lower extremities incurring higher likelihood of complications than head injuries. Complications were more likely to occur in those with injury to internal organs (64% higher odds) compared to those with fractures (Tables 2 and 3). There was no notable difference in likelihood of complications between socioeconomic areas.

More than one-fifth of patients required further support services after discharge from hospital. This was observed in all age groups, with higher odds in children and those 85 years and over compared to the reference group (65–84 years). The impact of gender did not stand out (females 8% higher odds than males) while more severe injuries and increasing number of comorbidities increased the odds of needing support services. Patients with upper limb and hip and thigh injuries were more likely to require further care compared to head injuries. Patients with eye injuries were more likely to require further care compared to fractures. The likelihood of being discharged to care with support services decreased with increase in socioeconomic advantage (Tables 2 and 3).

The cost of treating patients increased with increasing patient age, number of comorbidities, injury severity and female gender. Severe injuries and increasing number of comorbidities increased the likelihood of higher costs. The highest likelihood of increased costs were for knee and lower leg and hip and thigh injuries compared to head injuries, while injuries to most other body parts were also likely to increase costs compared to head injuries. Compared to fractures, injuries to internal organs were more likely to increase costs while all other injury types were likely to result in lower costs or did not produce a notable difference. There was no notable difference in likelihood of incurring higher costs between socioeconomic areas (Tables 2 and 3).

The median LOS (in days) was greater for those aged 65–84 (3, IQR 1–13) and 85 years and over (6, IQR 1–21) as opposed to the younger age groups (1, IQR 1–3). Relative to the 65–84 years age group, those 85 years and above were 36% more likely to have a longer LOS and those in the younger age groups were around 60–87% less likely. The mean LOS was 5.4 days (not shown in tables). Although the gender difference in median LOS was not notable, the mean was 4.1 days for males and 7.1 days for females (not shown in tables). Severe injuries and more comorbidities were associated with greater LOS. Injuries to the neck, trunk, shoulder and upper arm and lower extremities compared to the head had higher incident rates. Fractures had higher incident rates compared to most other types of injuries. There was no notable difference in incident rates between socioeconomic areas (Tables 2 and 3).

ICU stay and MV use proportions and the likelihood of using these services was highest among those aged 25–44 years and 45–64-year age group, while CCU use was greater among the 65–84-year and 86 years and over age groups. Females were more likely to stay in the CCU but gender was not statistically significantly with ICU and MV use. Severe injuries and comorbidities were associated with increased likelihood of CCS use. The body region injured and the type of injury affected the likely use of CCSs. The likelihood of ICU-stay and MV use were less as the socioeconomic advantage increased while it was the opposite for CCU stay (Tables 2 and 3).

The proportions of cases with 30-day readmissions increased with increasing age, injury severity and number of comorbidities. Potentially-avoidable readmissions within 30 days were equally common among males and females than the other two types of readmissions, which were more common among females. The likelihood of readmission was notably lower in the 0–14 and 15–24-year-old age groups compared to the 65–84-year (reference) age group for all three types of readmissions, while the 25–44- and 45–64-year age groups also had a notably lower likelihood of all-cause and non-planned readmissions compared to the 65–84-year (reference) age group. The effect of gender was not notable among all three types of readmissions while severe injuries were more than 70% more likely to result in any type of readmission. Increasing number of comorbidities increased the likelihood of all three types of readmissions. Injuries to the trunk and hip and thigh and burns to some body regions were more likely to result in any type of readmissions compared to the head, while shoulder and upper arm injuries were more likely to result in all-cause and potentially avoidable readmissions. Knee and lower leg and ankle and foot injuries were also more likely to result in potentially avoidable readmissions. Compared to fractures, eye injuries were more likely to result in all-cause readmissions and non-planned readmissions, while injury to internal organs were more likely to result in non-planned readmissions. Burns were more likely to result in both non-planned and potentially avoidable readmissions compared to fractures. There was no notable difference in likelihood of 30-day readmissions between socioeconomic areas (Tables 2 and 3).

Many of the outcomes were associated with each other. The results and a discussion of this analysis is provided as an appendix with supporting Additional file 1: Tables S1 and S2, as the main focus of this study was to establish the associations between base factors (mostly those present at point of hospital admission) and outcomes.

Our finding that increasing age, injury severity and comorbidities are associated with mortality after injury has been previously documented [5]. The proportion of patients who died in hospital (0.9% of index admissions) in this study is not dissimilar to a reported national proportion of 0.97% for 2005–06, though the latter was not limited to injury and contained more exclusions than ours (excluded palliative care and neonates) [59]. The proportion of 5.0% in-hospital deaths among serious injury patients in this study is close to the 4.4% reported for a severe trauma population for Victoria using data from 1996 to 2001 [21]. These two findings imply that injury patients have a higher rate of in-hospital death over general hospital admitted patients (includes injury and non-injury) and greater injury severity increases the risk. The proportion of deaths-in-hospital for the 85 years and over group (4.4%) is lower than previously published research of 9.5% for a cohort of patients admitted to an academic trauma centre in USA between 2006 and 2010 [17]. The latter cohort consisted of patients 90 years and over, the difference in the age cut-off most likely to contribute to the difference.

Intracranial injury was more likely to result in in-hospital death compared to other injury types, which echoes findings from previous work [17]. This study found that comorbidities had a notable effect on the outcome. There is varied findings on the effect of comorbidity on mortality among injury patients. Some studies also argued that the effect of increasing age and serious injury dominate the outcome over the presence of comorbidities [5, 11, 14, 20, 21, 23, 35, 60]; in other words, these studies [11, 14, 21, 35, 60] concluded that once age and injury severity were adjusted for, the effect of comorbidity was not significant, while others [5, 20, 23] concluded that the effect of comorbidity was significant among patients with minor or moderately severe injuries. The majority of injuries, however, are not severe, and therefore assessing the effects of comorbidity among mild to moderately injured patients is still valuable.

The findings from this study that age, comorbidities and injury severity increase the likelihood of complications confirms findings of previous research [10]. This study also found associations between body region and injury type with complications. The finding here is novel in revealing information on the effect of patient characteristics on hospital acquired complications for injury patients: increasing age, injury type and severity, and comorbidities all increased the likelihood of hospital-acquired complications.

This study confirms the finding that injury severity and comorbidities are likely to influence discharge destination [9, 10, 12, 15], but found that gender had no effect; contradicting a finding from a previous study [36]. The difference can be attributed to the fact that the study by Brown et al. (2012) [36] was based on a group of traumatic brain injury patients aged 65 years and over and they categorised patients discharged home with support services as not needing long term care, contrary to our categorisation. They also adjusted for marital status and ethnicity. We found older females to have a higher proportion of being discharged with support services over their male counterparts which could partly explain the findings in the former study. The exclusion of patients who originally came from long-term care institutions has reduced the number of patients needing further care. However, the proportion of such patients was less than 2%; therefore, the impact on the results would have been relatively minor.

The finding that hospital admission costs for injury patients are associated with injury severity has been reported in previous studies [14, 18]. These studies also found age to be associated with increased costs. The association of costs with comorbidity has also been established previously [14]. Findings from this study that costs can be higher for patients with lower extremity injuries and injury to blood vessels and internal organs have not been documented in past research. Injuries to lower extremities limit mobility and therefore delay discharge from hospital. This can be observed in the extended LOS for patients with hip and thigh injuries. The extended LOS and possibly major surgeries could be attributable to higher costs for this group of patients. The higher costs associated with injury to blood vessels and internal organs could be attributable to the fact that some of the patients with such injuries were the most likely to have an ICU stay and use the mechanical ventilator.

Age, injury severity and number of comorbidities were found to be associated with LOS in this study and previous research [14, 35]. Injuries to the lower extremities and trunk were likely to increase LOS, which was confirmed in a former study [14]. The mean hospital LOS of 5.4 days in this Victorian population was higher than the national mean of 3.6 days as reported by Pointer (2018) [2] for injury patients in 2014–15, and the current study does corroborate the finding from Pointer (2018) that females are likely to stay longer in hospital. The longer length of stay in the current study was due to the capture of days in hospital for all subsequent statistical separations and inward transfers due to complications within the index admission. The national report on the other hand picks up all records with an injury in the principal diagnosis but does not identify subsequent statistical separations and transfers as part of one episode. These extra records were captured by the use of data linkage in this study. This reveals that data linkage can provide more accurate estimates of burden. The median LOS of 3–6 days among those aged 65 year and over in the present study aligns well with another study in NSW [18] which reported 5–7 days, the higher LOS in the latter could be explained by the fact that it was solely based on level I trauma centre data.

Our findings that age, injury severity, comorbidities and body region are associated with ICU and ventilator use was seen in previous work [10, 11, 61]. We did not find any resources with details on CCU use of injury patients to compare, and the low proportion (0.4%) could be the reason that most studies did not focus on this outcome. Factors associated with CCS use are quite different to other outcome measures such as death, cost, LOS and 30-day readmissions. For example, ICU stay or the use of the MV is relatively more common in the 15–64-year age groups whereas the other outcomes are high among the 65 and over age groups. The effect of certain injury types (i.e., intracranial injuries and injuries to internal organs, blood vessels and nerves and spinal cord) is greater on ICU stay and MV use than on other outcomes.

This study’s finding that age, injury severity and comorbidity are associated with 30-day readmissions is aligned with previous research. The non-planned 30-day readmission proportion of 7.9% in this population is a little higher compared to previous findings of 5.9% for trauma patients aged 16 years and above from a Canadian Provincial Trauma system between 1998 and 2009 [19]. For potentially-avoidable 30-day readmissions, the SQLape methodology was utilised in this study. Potentially-avoidable 30-day readmissions are defined as unforeseen readmissions due to a previously known affection under this methodology. These generally consists of (i) iatrogenic complications, (ii) conditions generally resulting from pre-existing conditions and (iii) preventable diseases like deep vein thrombosis, pulmonary embolism and decubitus ulcer [43]. Since pre-existing conditions (comorbidities) were a factor for consideration in this study, patients with a 30-day readmission outcome with a primary diagnosis indicating it was a complication due to a pre-existing condition were not included. This could have contributed to the drop in this outcome (it has been shown that this quotient of patients account for nearly 80% of potentially-avoidable readmissions [43]). Therefore, this study’s use of a limited component of SQLape methodology may not be sufficient to capture the outcome. Finally, even if this component was included, the results may not have been accurate, as per the discussion by Walraven et al., (2011) [62]. They undertook a systematic review of literature on potentially-avoidable readmissions which included the SQLape methodology and stated that though many methods exist for identifying these readmissions, the methods had deficits and the proportions computed using the various methods varied significantly. They concluded that the true proportions of these potentially avoidable readmissions are still unclear.

There is evidence to suggest inter-relationships between outcomes exist; especially the effect of complications on readmissions [19], complications on LOS [33], ventilator use on mortality and cost and LOS [14] and ICU use on readmissions [63]. This study established all the associations mentioned except the association between complications and readmissions which were not significant in this instance. Apart from those noted here, many other associations were also established. Establishing inter-relationships between eleven outcomes is a study strength and this information will be useful in tertiary prevention. For e.g., complications are related to MV use, therefore precautions can be taken to mitigate complications for patients on the MV. They are also useful in research: one outcome may be used as proxy for another. e.g. LOS as proxy for cost, when cost cannot be measured.

Some limitations are noted in this study. We were unable to censor out-of-hospital deaths when selecting 30-day readmission cases, which may have resulted in an under-estimate in 30-day readmissions. We were also unable to estimate the hospital treatment costs for all the patients in the cohort; data on treatment costs were unavailable for private hospitals patients and missing for some of the public hospital patients (around one-fifth in total). Since the discussion of costs in this study is around averages as opposed to totals, we expect that the missing public hospital data would only have a minor impact on the overall conclusions.

Analysis of hospital admissions data does not provide information on all prevalent comorbidities. Coding has relevance for hospital reimbursement, and mostly, only conditions that had an impact on the patient management of the present episode are coded [39]. Validation studies have found that more serious and life-threatening conditions (e.g., cancer) are more likely to be captured in the codes than clinically non-specific and symptomatic conditions (e.g., rheumatologic) [22]; furthermore, coding can be insensitive towards certain conditions such as dementia [64]. Conversely, minor conditions are more likely to get coded among healthier patients as there are no serious ailments to record. Apart from that, in the case of fatalities or long hospital stay, more comorbid conditions are recorded, resulting in higher ORs. Prevalence of comorbidities have been found to be lower in administrative data than in medical records [65] and their clinical severity not recorded [22]. We acknowledge that the use of the U78-U88 codes (supplementary codes for chronic disease identification) incorporated into the VAED since 2015/16 will add value to this work and we propose using these in further research.

The number of complications identified in this study is not a perfect estimate of the total due to: (1) limitations in using hospital administrative databases such as the VAED and (2) limitations of CHADx. Regarding (1): a previous study on the VAED [66] revealed that only 76.2% of admissions were correctly allocated a complication in the ‘condition onset’ flag, which means that this study could be failing to capture about a quarter of the complications. Regarding (2), the drawback of the CHADx: although it aims to minimise double counting of complications, it has been shown to be less than perfect [67], due to the linear representation of conditions in the diagnosis codes, leaving the possibility for some overestimation of CHADx conditions [67]. For the purpose of providing an overview of injury outcomes in terms of complications, however, we considered the CHADx to be adequate.

Further information on the cause of readmission in the VAED would enable better classification of 30-day readmissions. A former study found that most of the readmissions were due to comorbidity than the index primary diagnosis [26], which states that paying attention to the acute condition of the index disease during the post-discharge period can mar the observation /treatment of comorbidities, resulting in readmissions.

Injury outcomes studies can benefit from including other factors such as marital status [68], ethnicity [16, 69] and insurance status [17]. However, we considered these variables to be non-essential for this descriptive study and therefore they were excluded due to space limitations. Further information on the available facilities and capacities, and the location of the individual hospitals (travel time to get access) can also add value to the prediction of outcomes as these features are likely to contribute. Apart from these, a major limitation of predefined databases is that not all the variables required for a study may be readily available.