Perioperative Mortality Rate in Indian Trauma Patients

Introduction

Trauma can be defined as damage inflicted on the body as the result of an external force, and the body's associated response. It is one of the leading causes of death world-wide, causing almost five million deaths yearly, making up for close to nine per cent of global deaths. Of these, road traffic injury alone causes 1.3 million, making it the ninth leading cause of death, with a predicted rise to seventh leading cause by 2030. Overall, however, the world has experienced an increase in life-expectancy over the past 30 years, as well as a decrease in infant and maternal mortality. With evidence suggesting that surgical care results in at least one million deaths per year, twice the number of maternal deaths, it has been established that new health metrics are needed for assessing health system performance.

The WHO Safe Surgery Saves Lives initiative proposes two metrics: day-of-surgery mortality, and 30-day postoperative in-hospital mortality. The subsequently introduced perioperative mortality rate (POMR) comprises these two. To date, no standard timespan for day-of-surgery measurement prevails - the same day as surgery has been suggested for convenience, while some institutions measure at 24, and others at 48 hours after surgery. The outcomes required to calculate POMR (death and number of surgeries) are easily defined. Additionally, there is an option to end follow-up at discharge, which makes the metric suitable for use in low-resource settings, where 30-day follow-up is rare. These are qualities that make POMR the choice of index for tracking the quality and safety of surgery in the world.

Analyses of databanks of surgical patients establish some factors that are more intimately linked to perioperative mortality (POM) than others. In trauma, especially, thorough risk adjustment is highly stressed to enable comparisons at hospital and patient level. However, the data yielding adjustment models largely stems from high-income countries, which is relevant considering that not all risk factors associated with POM in a high-resource setting are applicable in a low-resource setting. This owes to the scarcity of data on surgical programs in low-resource settings, not least in India. As the country faces increasing motorisation, and with it an increase in trauma patients for which factors associated with POM is poorly understood, bridging this knowledge gap is important.

Therefore, this study will be estimating the POMR and assessing the association between type of acute surgical intervention and POM among adult trauma patients in an urban Indian setting.

Objectives/aims

To estimate the POMR and asses the association between type of acute surgery and POM in adult trauma patients in a low resource setting. Based on subject matter knowledge and clinical experience the authors of this article hypothesise that the distribution of the multinomial variable type of acute surgery will differ significantly between survivors and non-survivors, and that this difference will remain significant after risk adjustment.

Study design

This is a retrospective analysis of the cohort study Towards Improved Trauma Care Outcomes in India (TITCO).

Setting

The data used is available in the TITCO database, which includes a total of 16,047 trauma patients enrolled prospectively from four public university hospitals between July 2013 and December 2015. The units are all located in megacities (Kolkata, Mumbai and Delhi) and classified as free-to-public. Data was collected through an on-site trained and supervised observer who systematically noted routine data on arrival through direct observation in the emergency room and through extraction from medical records. The observer worked rotating 8-hour shifts covering morning, day and night. Data from patients admitted outside the observer's shifts was retrospectively extracted from patient records. Patients were followed until discharge from hospitals, or for 30 days, or until death.

Source and method of participant selection

The on-site observer included patients presenting to participating centres, either by direct observation or by retrospective data extracted from records.

Explanatory variable

The main explanatory variable will be type of acute surgery, categorised using the Nordic Centre for Classifications in Health Care (NOMESCO) classification of surgical procedures (NCSP), which is based on functional-anatomic body system groups.

Other covariates

The association between type of acute surgery and POM will be adjusted for mechanism of injury, patient age in years, sex, systolic blood pressure, heart rate, transfer status and Glasgow coma scale on arrival. A health centre identifier will be included to account for centre level nesting in the data.

Data sources/measurements

The POMR will be calculated for adult trauma patients who underwent acute surgical intervention within 24 hours of arrival by dividing the number of surgical patients who died within 48 hours of arrival, and the number of patients who died within 30 days of arrival, with the total number patients who underwent surgery. The timespan of 48 hours for day-of-surgery mortality measurement is deemed most accurate due to the database format, in which it is stated whether a patient is taken to the operating room (OR) within 24 hours or not. Only procedures performed in an OR under regional or general anaesthesia qualify. The resulting POMR will be reported as a percentage.

Bias

Observers collecting the data were holders of a health science master's degree, and were trained by project management. They were also cross-checked by project management on two occasions, upon each a random selection of enrolled patient records were compared with those on file with the hospital. No major discrepancies were identified during these quality control sessions.

Inclusion of injury severity score (ISS) in the risk adjustment model, as recommended for predicting trauma mortality by the American College of Surgeons (ACS), will not be done in this study due to ISS performing poorly in predicting early mortality in low-resource settings. A physiological injury score, in the form of systolic blood pressure, heart rate and GCS score, is however included to adjust for physiological injury severity. ASA physical score and the need for ventilator use will not be included as this information is not available in the database.

Limitations in the database do not allow taking into account whether a patient has had any secondary, subacute surgery outside of the first 24 hours, within 30 days. Thus, subacute surgical interventions made after the first 24 hours are not counted, and so are not included in the denominator for POMR. Effectively, instead of the denominator being defined, as is recommended, as the total number of procedures, this redefines it as the total number of patients. The authors acknowledge that this is not ideal as it confers to the potential risk of overestimating POMR in the case where a patient receiving additional surgery expires perioperatively.

Study size

Simulation studies of logistic regression models indicate a need for ten events, i.e. patients with the outcome, per parameter used in the model. For the logistic regression model in this study, the type of acute surgery will account for 14 parameters, i.e. coefficients to be estimated. Mechanism of injury will account for four parameters. The four different centres will account for three. Because age, heart rate and systolic blood pressure will be modelled using splines with four degrees of freedom these variables will account for three parameters each. Sex, Glasgow coma scale and transfer status for one respectively, coming to a total of 33 parameters. This analysis will thus require at least 330 non-survivors to allow for logistic regression analysis of both POMR48H and POMR30D. Assuming a POMR48H of 10%, the study size (the number of adults who underwent surgery within 24 hours) needs to amount to least n = 3300. All adult patients in the TITCO database who received surgical intervention within 24 hours of arrival will be included, and is expected to provide a sufficient study size by a wide margin.

Quantitative variables

All quantitative variables, i.e. age, systolic blood pressure, heart rate and Glasgow coma scale will be analysed as continuous. Age, systolic blood pressure, and heart rate are assumed to be non-linearly associated with mortality why these will be modelled using spline smoothing terms with four degrees of freedom, resulting in coefficients for three basis functions to be estimated for each of these. Glasgow coma scale is assumed to display a linear association with mortality.

Statistical methods and analyses

R, a language and environment for statistical computing, will be used for all statistical analyses. The main analysis will be a complete case analysis, in which observations with missing values in any of the covariate variables will be excluded. In the event of missing data being too extensive to allow for analysis, multiple imputation using chained equations will be applied. Sample characteristics will be reported using medians and values at the 25th and 75th percentiles as well as minimum and maximum values for quantitative variables and proportions for qualitative variables. The association between type of surgery and POMR will be assessed using multilevel generalised additive models, in two steps. First, the crude associations between type of surgery and POM will be estimated using a model that does not include additional covariates. Second, the adjusted associations will be assessed using a model that does include the covariates listed above. 95% confidence intervals and a 5% significance level will be used.

Sensitivity analyses

One sensitivity analysis will be conducted by using a generalised additive model without accounting for centre level nesting.