Modeling trajectories of routine blood tests as dynamic biomarkers for outcome in spinal cord injury

Abstract Importance Early outcome prediction after acute traumatic spinal cord injury (SCI) is challenging due to pathological complexities and population heterogeneity. Routinely collected data during standard medical practice, such as laboratory analytics, can be a surrogate of underlying pathophysiological processes and used as a biomarker. We hypothesized that distinct temporal trends of blood analytics could be modeled after SCI and that those would predict distinct outcome parameters. Objective To test the hypothesis and develop machine learning models for predicting SCI outcomes. Design We developed and validated the models using retrospective data from the MIMIC-III and MIMIC-IV datasets and the prospective TRACK-SCI study, covering the period from 2001 to 2020. Setting Multi-center, involving data obtained from intensive care units across several different hospital settings in the United States. Participants Patients 15 years and older with traumatic SCI or vertebral fractures, admitted to emergency facilities, were included, resulting in a final cohort of 2,615 patients for modeling. Exposure(s) NA Main Outcome(s) and Measure(s) Primary outcomes included in-hospital mortality, occurrence of SCI and vertebral fracture in spine trauma patients, and SCI severity measured by the ASIA Impairment Scale. Blood biomarker level trajectory memberships served as predictors. Results Our study analyzed 2,752 patients, comprising 2,615 from the MIMIC dataset and 137 from the TRACK-SCI study. We identified multiple trajectory classes for 20 common blood markers that serve as dynamic predictors in machine learning classifiers. The in-hospital mortality model achieved an area under the Precision-Recall curve (PR-AUC) of 0.92 in the training set by leveraging trajectory data and baseline covariates from as early as day one post-injury. For SCI severity, the models distinguished between complete and incomplete motor outcomes with a PR-AUC of 0.78. The trajectory-based models showed significant improvement over traditional severity scores, such as Simplified Acute Physiology Score (SAPS) II, especially when combined with demographic information. Conclusions and Relevance Real-world routinely obtained blood test data can be used to model dynamic changes after SCI with prediction validity for patient outcomes. This work establishes the basis for further development of dynamic biomarker data for outcome prediction in neurotrauma and other neurological conditions. Key Points Question Can dynamic changes of routinely collected acute blood test data serve as biomarkers to predict outcomes in patients with traumatic spinal cord injury (SCI)? Findings In this study using data from the MIMIC and TRACK-SCI datasets, we developed machine learning models that categorize patients into distinct groups based on the temporal and non-linear dynamics of blood biomarkers. These models effectively predicted in-hospital mortality and SCI severity, indicating significant predictive utility from as early as the first day of hospitalization. Meaning The application of dynamic machine learning models to blood test data has potential to significantly predict the prognosis and enhance management of traumatic spinal cord injury in clinical settings.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.