K. Diercks1, C. Johansen1, K. Patel1, D. Sanders1, A. Starr1, C. Park1 1University Of Texas Southwestern Medical Center, Dallas, TX, USA
Introduction: Geriatric trauma patients present a unique challenge given changes in physiology, polypharmacy and frailty. Several scoring systems exist to predict morbidity and mortality, but few are prospective and automated. The Parkland Trauma Index of Mortality (PTIM) is a validated, novel machine learning (ML) algorithm developed to use electronic medical record data to predict mortality within 48 hours during the first 3 days of hospitalization. We hypothesized that PTIM’s predictive capabilities would have the same diagnostic accuracy on patients over 65 and under 65.
Methods: Retrospective single-center study between 2020-2021 at a Level 1 Trauma center. PTIM extracts over 20+ demographic, physiologic and laboratory values and displays these 12 hours into admission, then automatically scores hourly. This embedded ML algorithm went live in December of 2020. Geriatric patients were identified as patients over 65. Level 1 and 2 trauma patients with PTIM scores were included. Demographics were collected and?patient outcomes were analyzed. ROC curves were generated using Stata and used to compare outcomes such as pneumonia and mortality in patients with PTIM scores over 65 with patients under 65.
Results: 171 patients fit the inclusion criteria. 26 patients were > 65 years. Overall rates of mortality were 5.85%, and 24% in patients > 65 years. Overall rates of pneumonia were 10.53%; rates of pneumonia in patients > 65 years was 52%. Age was significantly associated with falling as the mechanism of injury (p<0.001). PTIM had better diagnostic accuracy for mortality in patients over 65 than for patients under 65, with an AUROC of 0.6825 in patients <65 years and 0.9375 for patients > 65. (p=0.07) (Figure 1a). PTIM also demonstrated better predictive capabilities for pneumonia in patients > 65 than patients < 65 (AUROC 0.774 and 0.9583 respectively, p=0.06) (Figure 1b).
Conclusion: We applied an automated machine learning algorithm that predicts mortality in geriatric trauma patients and found PTIM had better diagnostic accuracy for rates of mortality and pneumonia in patients over 65 when compared to patients under 65. Next steps include further investigation of these risk factors in a larger sample size and exploring PTIM’s predictive capabilities in geriatric patients with other centers.
