Early Hospital Mortality among Adult Trauma Patients Significantly Declined between 1998-2011: Three Single-Centre Cohorts from Mumbai, India

Background Traumatic injury causes more than five million deaths each year of which about 90% occur in low- and middle-income countries (LMIC). Hospital trauma mortality has been significantly reduced in high-income countries, but to what extent similar results have been achieved in LMIC has not been studied in detail. Here, we assessed if early hospital mortality in patients with trauma has changed over time in an urban lower middle-income setting. Methods We conducted a retrospective study of patients admitted due to trauma in 1998, 2002, and 2011 to a large public hospital in Mumbai, India. Our outcome measure was early hospital mortality, defined as death between admission and 24-hours. We used multivariate logistic regression to assess the association between time and early hospital mortality, adjusting for patient case-mix. Injury severity was quantified using International Classification of Diseases-derived Injury Severity Score (ICISS). Major trauma was defined as ICISS<0.90. Results We analysed data on 4189 patients out of which 86.5% were males. A majority of patients were between 15 and 55 years old and 36.5% had major trauma. Overall early hospital mortality was 8.9% in 1998, 6.0% in 2002, and 8.1% in 2011. Among major trauma patients, early hospital mortality was 13.4%, in 1998, 11.3% in 2002, and 10.9% in 2011. Compared to trauma patients admitted in 1998, those admitted in 2011 had lower odds for early hospital mortality (OR = 0.56, 95% CI = 0.41–0.76) including those with major trauma (OR = 0.57, 95% CI = 0.41–0.78). Conclusions We observed a significant reduction in early hospital mortality among patients with major trauma between 1998 and 2011. Improved survival was evident only after we adjusted for patient case-mix. This finding highlights the importance of risk-adjustment when studying longitudinal mortality trends.


Background
Traumatic injury is a major threat to global public health [1,2].More people die annually from traumatic injuries than from HIV/ AIDS, tuberculosis, malaria, and obstetric conditions combined [1].Over 90% of the five million annual deaths from traumatic injuries occur in low-and middle-income countries (LMIC) [3,4].A recent study estimated that almost two million lives could be saved each year, if hospital care for the injured, i.e. trauma care in LMIC can be improved and reach the same level as in highincome countries (HIC) [5].However, it is currently unclear how such reduction can be achieved.
Although a large part of trauma mortality occurs at the injurysite and during pre-hospital transportation, around 30-50% of trauma mortality occurs in hospital [6,7].In HIC, hospital trauma mortality has been significantly reduced in recent years [8,9].Among the most important explanations for this reduction are implementation of trauma care systems and improved medical and surgical treatment [10,11], including management of traumatic brain injury [12], haemorrhage control [13], and musculoskeletal injuries [14,15].These improvements in clinical care have been driven, in large part, by research progress, from the basic sciences to the systems and policy level.
Hospital trauma registers have played a key role in the advancement of patient-based research and trauma care in HIC [16,17].Trauma registers offer a unique opportunity to document patient characteristics and audit outcomes [18], thereby creating a platform for innovative clinical research.Moreover, trauma registers in HIC have been used to study longitudinal trends in hospital mortality [9].Such studies are important as they form the basis for further research on factors underlying changes in hospital mortality.In contrast, few trauma registers exist in LMIC [19,20], which limit the potential for studies of longitudinal trends in hospital mortality in these countries.
Studies from HIC have stressed the importance of riskadjustment when comparing trauma hospital mortality rates between hospitals or over time [21,22].Risk-adjustment is commonly done by incorporating measures of patient case-mix including injury severity or mortality risk in the analysis [23].Several different methods for assigning mortality risk exist but the evidence for which one to use is inconclusive [24].Recent research from HIC has used the data-driven score International Classification of Disease (ICD) based Injury Severity Score (ICISS) to adjust for mortality risk [25], but there is no consensus regarding the appropriate approach in studies from LMIC.We adopted an explorative approach to risk-adjustment using ICISS, with the aim to assess if early hospital mortality in patients with trauma has changed over time in an urban lower middle-income setting.

Study design
We conducted a retrospective study of patients presenting to Lokmanya Tilak Municipal General Hospital (LTMGH), Mumbai, India (Figure 1).

Ethics statement
The institutional ethics committee of LTMGH approved the collation of the 2011 database, reference number IEC/22/10, and approved new analyses of all three dataset in an amendment to IEC/22/10.The ethics committee granted a waiver of consent.

Setting
India is the world's second most populous country, classified as a lower middle-income country by the World Bank [26].In 2011, it was estimated that more than 500,000 people died from traumatic injuries in India [27].In 2015, around 200,000 Indians are expected to die from road traffic injuries (RTI) [28].Furthermore, in India, more than 50% of mortality from RTI occur in hospitals [29].Mumbai is the most populous city in India, with more than18 million inhabitants [30].Lokmanya Tilak Municipal General Hospital (Figure 2) is one of the four biggest public hospitals in Mumbai.

Data
We retrospectively analysed three datasets of patient cohorts admitted to LTMGH trauma ward during 1998, 2002, and 2011 (Table 1).All three datasets were collated for previous research.At least one of the authors (NR, VK, or SD) of this study supervised collection of all three datasets.The three datasets were merged and duplicate entries were detected and removed using the automated approach suggested in STATA's FAQ [31].

Eligibility criteria
All non-duplicate observations were considered eligible.Observations falling outside of accepted value ranges were set to missing.

Variables
Our outcome of interest was early hospital mortality, defined as death between admission and 24 hours.According to Haider et al. [23], risk adjusted outcome studies should account for sex, age, mechanism of injury, and physiological-and anatomical injury severity.We were able to include all except physiological injury severity as covariates.Age was defined as a categorical variable, with the conventionally used cut-offs ,15, 15-55, and .55years.Mechanism of injury was defined as a nominal variable, categorised as fall, railway injury (RI), road traffic injury (RTI), assault, other, and unknown.

Anatomical injury severity
We used ICISS to quantify anatomical injury severity [32].Our rationale was that ICISS has been shown to perform well compared to the more conventional Injury Severity Score (ISS) and other established injury severity measures such as Trauma and Injury Severity Score (TRISS) [33,34], while also being easily computed using ICD-codes.To calculate ICISS, each ICD-code is assigned a survival risk ratio (SRR).Each SRR is equal to the proportion of patients who survived with a specific ICD-code in a reference population.In studies with large samples, the study sample is often used as the reference population [35,36].Smaller studies may use published SRRs from a larger, similar, reference population [37].In literature searches, we found no published SRRs from a LMIC trauma population and therefore we decided to use our own sample as the reference population.Because our sample was small, this analysis should be considered explorative and we are well aware of the same-sample bias that potentially is introduced using this approach.
MG coded all patients' injuries using ICD-10: 2010 edition, down to the fourth level.We calculated SRRs by dividing the number of survivors with a specific ICD-code with the total number of patients with the same ICD-code.This was done separately for each cohort to account for potential differences in factors such as transport time and care that might have influenced outcomes in the three time periods.We used the ICISS1 version of ICISS, which means that for overall scoring of patients' injury severity we counted only the worst injury (lowest SRR).We chose ICISS1 because this version has been showed to have a higher predictive value compared to versions that take into account all of a patient's injuries [38].We defined major trauma as ICISS,0.9 [39].

Data collection methods
The surgical registrar of each unit collected data on patients admitted during his or her unit's duty.
The surgical registrar of each unit collected data on patients admitted during his or her unit's duty.
One dedicated data collector performed all data collection.Her timings in the trauma ward were randomized, covering eight hours each day, five days a week.She retrieved data from patients admitted during her off shift periods from intake forms.

Categories of variables collected
Demographics

Statistical methods
We used STATA (STATA 12, StataCorp, Texas) for statistical analyses.Where applicable, a significance level of 5% and a confidence level of 95% were used.Sample characteristics are reported for each cohort.Continuous variables are reported using their median, inter-quartile range, and range because they were found to be non-normally distributed.Categorical variables are reported as proportions.
We assessed the univariate associations between covariates and early hospital mortality using logistic regression and report their unadjusted odd-ratios (OR).We used multivariate logistic regression to evaluate the association between time and early hospital mortality.Dummy variables were created for the three cohorts and the 2002 and 2011 cohorts were included in the logistic regression model and compared early hospital mortality with the early hospital mortality in 1998.Age, sex, mechanism of injury, and ICISS were included as covariates [23].We conducted 13 subgroup analyses and assessed early hospital mortality trends separately in patients with and without major trauma, males, females, each age category, and in patients with history of fall, railway injury, and road traffic injury.
We used multiple imputation using chained equations to handle missing data.For each variable with missing data, we assessed the association between the probability of missing data and early hospital mortality using logistic regression.We found no significant associations, i.e.P-value.0.05 in all analyses, and hence deemed  multiple imputation as appropriate.Each cohort was imputed separately.We specified our imputation model to impute sex using logistic regression, and age and mechanism of injury using multinominal logistic regression.We performed 20 imputations to achieve stable estimates.

Results
We analysed data on 4189 patients (Table 2), out of which 86.5% were male.Road traffic injury was the most common mechanism of injury across cohorts, followed by railway injury and falls.Out of all patients, 81.5% were between 15 and 55 years old.Early hospital mortality among all patients was 7.9% (Figure 3A), the median ICISS was 0.93, and 36.5% had major trauma (ICISS,0.9).The early hospital mortality rate among major trauma patients was 12.2% overall and 13.4%, 11.3%, and 10.9% in 1998, 2002, and 2011 respectively.

Discussion
Our study showed that early hospital mortality in trauma patients decreased over a 13-year period in an urban lower middle-income setting.This trend was not observed in unadjusted analysis and our study therefore highlights the importance of riskadjustment when comparing mortality rates over time.However, we used an explorative approach to compute ICISS as a measure of anatomical injury severity and this should be kept in mind when interpreting our results.The observed reduction in early hospital mortality was most pronounced in patients with major trauma, while no significant changes could be detected in patients with minor trauma.Similar trends as the one shown here have been observed for other major conditions in the same geographical area.For example, the infant mortality rate and under-five mortality rate in Maharashtra dropped by 25% and 34% respectively between 1992 and 2006 [40].Similarly, Maharashtra's maternal mortality rate dropped by 23% between 1998 and 2009 [41,42].It is important to note that these figures are not riskadjusted.
The implementation of trauma systems has been widely claimed as a major reason for reduced trauma mortality over time in HIC [11,22,43,44].The trauma system concept originates from the designation of dedicated trauma centres in the US some 40 years ago [45].Today such systems ideally include streamlined preventive, prehospital, hospital, and rehabilitative measures [8].Interestingly, the reduction in early hospital mortality observed in our study has occurred in the absence of a comprehensive trauma system.On the hospital level, LTMGH does employ a trauma team model, and one hypothetical explanation for the observed trend might be a maturation of trauma care efforts at institutional level, with improved equipment and enforced education of residents according to advanced trauma life support principles.
A functioning prehospital system is generally considered a vital component of a modern trauma system and improvements in prehospital triage and care may influence overall mortality rates.A recent systematic review found that the implementation of prehospital systems in LMIC has led to reduced overall mortality rates [46].However, there is no organised prehospital system in Mumbai and most trauma patients arrive to hospital in taxis or are brought by the police [47].Ambulances are used almost exclusively for inter-hospital transfers and are largely used as transport vehicles without any resuscitation equipment [47].Therefore, the reduced odds of early hospital mortality can most likely not be attributed to improvements in prehospital care, but rather improved hospital trauma care.
A high proportion of the patients in this study were males.This finding is consistent with other research from LMIC [48][49][50].Interestingly, the proportion of females seems to be higher in HIC trauma than in our study population [9,51].This discrepancy might be partly due to more women involved in work outside of the home in HIC compared to LMIC.Lastly, we found that RTI, falls, and RI were the most common mechanisms of injury.Although RTI and falls are major causes of trauma around the world, RI appears to be rather unique for Mumbai, and thus interesting to explore further as a subset of the Mumbai trauma population.

Methodological considerations and limitations
First, the difference in inclusion criteria between the three cohorts is a major limitation.To adjust for this limitation and to minimize bias, we adjusted our analyses for case-mix and also conducted several subgroup analyses.Second, we used ICISS to adjust for anatomical injury severity and used our own data to calculate SRRs and generate finals scores.As we highlighted in the methodology section this approach might have introduced a samesample bias that in part can explain the good fit of ICISS.Our approach should therefore be considered explorative in nature.However, we argue that this was a more sensible approach compared to using SRRs derived from a completely different trauma population.Third, we analysed available datasets and were therefore not able to choose our time-points.Unfortunately, retrospective inquires into patient files spanning more than ten years back constitute a more or less impossible undertaking as files are paper based and stored under variable conditions over time.Finally, the data did not allow us to explore outcome measures beyond mortality between admission and 24 hours.We do acknowledge that later mortality and functional outcomes are important and should be the focus of prospective studies in the same setting.

Conclusions
In our study, we observed a significant reduction in early trauma mortality among patients with major trauma between 1998 and 2011.Improved survival was evident after adjusting for patient case-mix and this finding therefore highlights the importance of risk-adjustment in studies of mortality over time.Furthermore, as one of few studies from an urban lower-middle income setting spanning more than ten years, we show that survival has improved despite the absence of a trauma system and functioning prehospital organisation.

Figure 3 .
Figure 3. Unadjusted and adjusted comparison of early hospital mortality across cohorts.A. Unadjusted early hospital mortality rates across cohorts.B. Odds ratios for the 2002 and 2011 cohorts when treated as indicator variables in a multivariate logistic regression analysis with early hospital mortality as the dependent variable.The model was adjusted for sex, age, mechanism of injury and ICISS.Abbreviations: ICD International Classification of Disease, ICISS ICD-derived Injury Severity Score.doi:10.1371/journal.pone.0090064.g003

Figure 4 .
Figure 4. Comparison of early hospital mortality odds ratios in subgroup analyses.Odds ratios are early hospital mortality odds ratios.A. Major trauma was defined as ICISS,0.9.The model was adjusted for sex, age, mechanism of injury, and ICISS.B. The model was adjusted for age, mechanism of injury, and ICISS.C. The model was adjusted for sex, mechanism of injury, and ICISS.D. The model was adjusted for sex, age, and ICISS.Abbreviations: ICD International Classification of Disease, ICISS ICD-derived Injury Severity Score.*P-value,0.05,compared to the reference category.doi:10.1371/journal.pone.0090064.g004

Table 3 .
Results from univariate logistic regression assessing associations between patient characteristics and early hospital mortality.