A Real-Time Automated Machine Learning Algorithm for Predicting Mortality in Trauma Patients: Survey Says it's Ready for Prime-Time

BACKGROUND

Though artificial intelligence ("AI") has been increasingly applied to patient care, many of these predictive models are retrospective and not readily available for real-time decision-making. This survey-based study aims to evaluate implementation of a new, validated mortality risk calculator (Parkland Trauma Index of Mortality, "PTIM") embedded in our electronic healthrecord ("EHR") that calculates hourly predictions of mortality with high sensitivity and specificity.

METHODS

This is a prospective, survey-based study performed at a level 1 trauma center. An anonymous survey was sent to surgical providers and regarding PTIM implementation. The PTIM score evaluates 23 variables including Glasgow Coma Score (GCS), vital signs, and laboratory data.

RESULTS

Of the 40 completed surveys, 35 reported using PTIM in decision-making. Prior to reviewing PTIM, providers identified perceived top 3 predictors of mortality, including GCS (22/38, 58%), age (18/35, 47%), and maximum heart rate (17/35, 45%). Most providers reported the PTIM assisted their treatment decisions (27/35, 77%) and timing of operative intervention (23/35, 66%). Many providers agreed that PTIM integrated into rounds and patient assessment (22/36, 61%) and that it improved efficiency in assessing patients' potential mortality (21/36, 58%).

CONCLUSIONS

Artificial intelligence algorithms are mostly retrospective and lag in real-time prediction of mortality. To our knowledge, this is the first real-time, automated algorithm predicting mortality in trauma patients. In this small survey-based study, we found PTIM assists in decision-making, timing of intervention, and improves accuracy in assessing mortality. Next steps include evaluating the short- and long-term impact on patient outcomes.

Geospatial Assessment to Improve Time to Treatment (GAITT)

INTRODUCTION

Geographic information systems (GIS) can optimize trauma systems by identifying ways to reduce time to treatment. Using GIS, this study analyzed a system in Maryland served by Johns Hopkins Suburban Hospital and the University of Maryland Capital Region Medical Center. It was hypothesized that including Walter Reed National Military Medical Center (WRNMMC) in the Maryland trauma system in an access simulation would provide increased timely access for a portion of the local population.

MATERIALS AND METHODS

Using ArcGIS Online, catchment areas with and without WRNMMC were built. Catchment areas captured Johns Hopkins Suburban Hospital, University of Maryland Capital Region Medical Center, and WRNMMC at 5-, 10-, 15-, 20-, 25-, 30-, 45-, and 60-min. Various time conditions were simulated (12 am, 8 am, 12 pm, and 5 pm) on a weekday and weekend day. Data was enriched with 19 variables addressing population size, socioeconomic status, and diversity.

RESULTS

All catchment areas benefited on at least one time-day simulation, but the largest increases in mean population coverage were in the 0-5 (10.5%), 5-10 (12.3%), and 10-15 min (5.7%) catchment areas. These areas benefited regardless of time-day simulation. The lowest increase in mean population coverage was seen in the 20-25-min catchment area (0.1%). Subgroup analysis revealed that all socioeconomic status and diversity groups gained coverage.

CONCLUSIONS

This study suggests that incorporating WRNMMC into the Maryland trauma system might yield increased population coverage for timely trauma access. If incorporated, WRNMMC may provide nonstop or flexible coverage, possibly in different traffic scenarios or while civilian centers are on diversion status.

Association of Time to Definitive Hemostasis With Mortality in Patients With Solid Organ Injuries

Introduction The Golden Hour is a term used in the trauma setting to refer to the first 60 minutes after injury. Traditionally, definitive care within this period was believed to dramatically increase a patient's survival. Though the period of 60 minutes is unlikely to represent a point of distinct inflection in survival, the effect of time to definitive care on survival remains incompletely understood. This study aims to measure the association of time to definitive hemostasis with mortality in patients with solid organ injuries as well as the effect of survival bias and a form of selection bias known as indication by severity on the relationship between time to treatment and survival. Methodology This is a retrospective cohort study using data obtained from the American College of Surgeons National Trauma Data Bank (NTDB) from the years 2017 through 2019 selecting patients treated for blunt liver, spleen, or kidney injury who required angioembolization or surgical hemostasis within six hours. A Cox proportional hazards regression was used to analyze time to death. The association of probability of death with time was examined with a multivariate logistic regression initially treating the relationship as linear and subsequently transforming time to hemostasis with restricted cubic splines to model a non-linear association with the outcome. To model survival and indication by severity bias, we created a computer-generated data set and used LOESS regressions to display curves of the simulated data. Results The multivariate Cox proportional hazards analysis shows a coefficient of negative 0.004 for minutes to hemostasis with an adjusted hazard ratio of 0.9959 showing the adjusted hazard of death slightly diminishes with each increasing minute to hemostasis. The likelihood ratio chi-square difference between the model with time to hemostasis included as a linear term versus the model with the restricted cubic spline transformation is 97.46 (p<0.0001) showing the model with restricted cubic splines is a better fit for the data. The computer-generated data simulating treatment of solid organ injury with no programmed bias displays an almost linear association of mortality with increased treatment delay. When indications by severity bias and survival bias are introduced, the risk of death decreases with time to hemostasis as in the real-world data. Conclusion Decreasing mortality with increasing delay to hemostasis in trauma patients with solid organ injury is likely due to confounding due to indication by severity and survival bias. After taking these biases into account, the association of delayed hemostasis with better survival is not likely due to the benefit of delay but rather the delay sorts patients by severity of injury with those more likely to die being treated first. These biases are extremely difficult to eliminate which limits the ability to measure the true effect of delay with retrospective data. The findings may however be of value as a predictive model to anticipate the acuity of a patient after an interval of unavoidable delay such as with a long transfer time.

Equity of geographical access to public health facilities in Nepal

Introduction

Geographical accessibility is important against health equity, particularly for less developed countries as Nepal. It is important to identify the disparities in geographical accessibility to the three levels of public health facilities across Nepal, which has not been available.

Methods

Based on the up-to-date dataset of Nepal formal public health facilities in 2021, we measured the geographical accessibility by calculating the travel time to the nearest public health facility of three levels (ie, primary, secondary and tertiary) across Nepal at 1×1 km² resolution under two travel modes: walking and motorised. Gini and Theil L index were used to assess the inequality. Potential locations of new facilities were identified for best improvement of geographical efficiency or equality.

Results

Both geographical accessibility and its equality were better under the motorised mode compared with the walking mode. If motorised transportation is available to everyone, the population coverage within 5 min to any public health facilities would be improved by 62.13%. The population-weighted average travel time was 17.91 min, 39.88 min and 69.23 min and the Gini coefficients 0.03, 0.18 and 0.42 to the nearest primary, secondary and tertiary facilities, respectively, under motorised mode. For primary facilities, low accessibility was found in the northern mountain belt; for secondary facilities, the accessibility decreased with increased distance from the district centres; and for tertiary facilities, low accessibility was found in most areas except the developed areas like zonal centres. The potential locations of new facilities differed for the three levels of facilities. Besides, the majority of inequalities of geographical accessibility were from within-province.

Conclusion

The high-resolution geographical accessibility maps and the assessment of inequality provide valuable information for health resource allocation and health-related planning in Nepal.

Prehospital Intervals and In-Hospital Trauma Mortality: A Retrospective Study from a Level I Trauma Center

BACKGROUND

The increase in mortality and total prehospital time (TPT) seen in Qatar appear to be realistic. However, existing reports on the influence of TPT on mortality in trauma patients are conflicting. This study aimed to explore the impact of prehospital time on the in-hospital outcomes.

METHODS

A retrospective analysis of data on patients transferred alive by Emergency Medical Services (EMS) and admitted to Hamad Trauma Center (HTC) of Hamad General Hospital (HGH; Doha, Qatar) from June 2017 through May 2018 was conducted. This study was centered on the National Trauma Registry database. Patients were categorized based on the trauma triage activation and prehospital intervals, and comparative analysis was performed.

RESULTS

A total of 1,455 patients were included, of which nearly one-quarter of patients required urgent and life-saving care at a trauma center (T1 activations). The overall TPT was 70 minutes and the on-scene time (OST) was 24 minutes. When compared to T2 activations, T1 patients were more likely to have been involved in road traffic injuries (RTIs); experienced head and chest injuries; presented with higher Injury Severity Score (ISS: median = 22); and had prolonged OST (27 minutes) and reduced TPT (65 minutes; P = .001). Prolonged OST was found to be associated with higher mortality in T1 patients, whereas TPT was not associated.

CONCLUSIONS

In-hospital mortality was independent of TPT but associated with longer OST in severely injured patients. The survival benefit may extend beyond the golden hour and may depend on the injury characteristics, prehospital, and in-hospital settings.

MEWS++: Enhancing the Prediction of Clinical Deterioration in Admitted Patients through a Machine Learning Model

Early detection of patients at risk for clinical deterioration is crucial for timely intervention. Traditional detection systems rely on a limited set of variables and are unable to predict the time of decline. We describe a machine learning model called MEWS++ that enables the identification of patients at risk of escalation of care or death six hours prior to the event. A retrospective single-center cohort study was conducted from July 2011 to July 2017 of adult (age > 18) inpatients excluding psychiatric, parturient, and hospice patients. Three machine learning models were trained and tested: random forest (RF), linear support vector machine, and logistic regression. We compared the models’ performance to the traditional Modified Early Warning Score (MEWS) using sensitivity, specificity, and Area Under the Curve for Receiver Operating Characteristic (AUC-ROC) and Precision-Recall curves (AUC-PR). The primary outcome was escalation of care from a floor bed to an intensive care or step-down unit, or death, within 6 h. A total of 96,645 patients with 157,984 hospital encounters and 244,343 bed movements were included. Overall rate of escalation or death was 3.4%. The RF model had the best performance with sensitivity 81.6%, specificity 75.5%, AUC-ROC of 0.85, and AUC-PR of 0.37. Compared to traditional MEWS, sensitivity increased 37%, specificity increased 11%, and AUC-ROC increased 14%. This study found that using machine learning and readily available clinical data, clinical deterioration or death can be predicted 6 h prior to the event. The model we developed can warn of patient deterioration hours before the event, thus helping make timely clinical decisions.

Assessing Spatial Accessibility to Medical Resources at the Community Level in Shenzhen, China

Spatial accessibility to medical resources is an integral component of universal health coverage. However, research evaluating the spatial accessibility of healthcare services at the community level in China remains limited. We assessed the community-level spatial access to beds, doctors, and nurses at general hospitals and identified the shortage areas in Shenzhen, one of the fastest growing cities in China. Based on hospital and population data from 2016, spatial accessibility was analyzed using several methods: shortest path analysis, Gini coefficient, and enhanced 2-step floating catchment area (E2SFCA). The study found that 99.9% of the residents in Shenzhen could get to the nearest general hospital within 30 min. Healthcare supply was much more equitable between populations than across communities in the city. E2SFCA scores showed that the communities with the best and worst hospital accessibility were found in the southwest and southeast of the city, respectively. State-owned public hospitals still dominated the medical resources supply market and there was a clear spatial accessibility disparity between private and public healthcare resources. The E2SFCA scores supplement more details about resource disparity over space than do crude provider-to-population ratios (PPR) and can help improve the efficiency of the distribution of medical resources.