Ultra-Massive Transfusion: Can a Machine Learning Model Predict Outcomes and Survivability in Adult Trauma Patients? Open Access
Meyer, Courtney (Summer 2022)
Published
Abstract
Background: Despite the widespread use of ultra-massive transfusion (UMT) in the resuscitation of trauma patients, mortality remain high. There is scarce evidence determining the clinical and physiologic parameters in which this intervention is most effective. Simultaneously, the US faces a critical blood product shortage and appropriate allocation of resources remains an important public health issue. Therefore, this study sought to investigate the efficacy of UMT for trauma patients at a single institution and utilize machine learning modeling to predict outcomes and survivability.
Methods: A retrospective cohort study of adult trauma patients undergoing UMT (defined as ≥20 units of red cell products within 24 hours) was conducted at a Level I trauma center from May 2018-Nov 2021. Data was triangulated from the blood bank, electronic medical record, and institutional trauma registry. The outcome of interest was mortality at 24 hours and discharge. Demographics, injury characteristics, clinical presentation, and total products transfused were compared between those who survived and those who died. A statistical analysis and hour-by-hour time series analysis were conducted and machine learning (ML) predictive models were generated and validated using R (version 4.1.1).
Results: There were 1,164 patients with MTP activations and 193 (16.6%) were adult trauma patients meeting criteria for UMT. The in-hospital mortality rate was 38.8% at 24 hours and 54% at discharge. Those who died were more hemodynamically unstable and in a more advanced state of shock at the time of presentation. The deceased cohort received more total blood products at each time interval studied, with significantly higher rates of packed red blood cell and fresh frozen plasma transfusion. Ten distinct ML models were generated successfully identified clinical and physiologic parameters most strongly associated with mortality.
Conclusions: This study demonstrates that mortality rates for UMT remain high and increased blood product transfusion is not associated with improved outcomes. Analysis of physiologic and clinical parameters further supports that early hemorrhage control and achievement of hemodynamic stability are critical to survivability. With blood as a limited resource, it is imperative to continue research in this field in order to identify which patients will benefit most from this aggressive therapy.
Table of Contents
1. Chapter 1: Introduction
1.1. Overview & Significance …………………………………………………….……... 1-2
1.2. Purpose Statement …………………………………………………...………….…….. 2
1.3. Specific Aims …………………………………………………………………….….... 2
1.4. Definition of Terms …………………………………………………...………………. 3
1.5. Abbreviations ………………………………………………………....………………. 4
2. Chapter 2: Comprehensive Review of the Literature
2.1. History of trauma resuscitation ………………………………………….………...…. 5-7
2.2. Blood products and principles of balanced transfusion ………………….……..….… 7-8
2.3. Principles of massive transfusion and ultra-massive transfusion ……….………...….. 8-9
2.4. Current state of ultra-massive transfusion in trauma ………………………………... 9-10
2.5. Current state of national blood shortage ………………………………………………. 10
3. Chapter 3: Manuscript
3.1. Title page ……………………………………………………………………...………. 11
3.2. Abstract …………………………………………………………………….……… 12-13
3.3. Introduction ………………………………………………………………………... 13-14
3.4. Methods ……………………………………………………………………..……... 14-16
3.5. Results ………………………………………………………………………..……. 16-19
3.6. Discussion ………………………………………………………………….……... 19- 22
4. Chapter 4: Extended Methodology and Results
4.1. Extended methodology
4.1.1. Introduction ……………………………………………………………..……… 23
4.1.2. Machine learning predictive modeling overview ……………………...…… 23-24
4.1.3. Machine learning predictive modeling methods …………………..……….. 24-26
4.1.4. Hour-by-hour time series analysis methods ………………….…………..… 26-27
4.2. Extended results
4.2.1. Machine learning predictive modeling results …………..………………….. 28-29
4.2.2. Intra-operative hour by hour time series analysis results …………..………. 29-31
4.2.2.1. Vital signs and laboratory values
4.2.2.2. Blood products
5. Chapter 5: Limitations, Conclusions, Public Health and Ethical Implications
5.1. Limitations ………………………………………………………………….…………. 32
5.2. Conclusions, public health and ethical implications ………………………………. 32-34
6. References
6.1. Thesis References ……………………………………………………………….…. 35-38
6.2. Manuscript References ………………………………………………………….…. 39-40
7. Appendices
7.1. Manuscript tables …………………………………………………………………. 41- 43
7.2. Manuscript figures …………………………………………………………………. 44-46
7.3. Extended results
7.3.1. Machine learning predictive modeling tables ………………….………………. 47
7.3.2. Hour-by-hour time series analysis tables ……………………………...……. 47-51
7.3.3. Machine learning predictive modeling figures …………………...………… 52-54
7.3.4. Hour-by-hour hour time series analysis figures ……………….…………… 55-58
About this Master's Thesis
| School | |
|---|---|
| Department | |
| Degree | |
| Submission | |
| Language |
|
| Research Field | |
| Keyword | |
| Committee Chair / Thesis Advisor | |
| Committee Members |
Primary PDF
| Thumbnail | Title | Date Uploaded | Actions |
|---|---|---|---|
|
|
Ultra-Massive Transfusion: Can a Machine Learning Model Predict Outcomes and Survivability in Adult Trauma Patients? () | 2022-08-01 10:10:00 -0400 |
|
Supplemental Files
| Thumbnail | Title | Date Uploaded | Actions |
|---|