Expert-augmented automated machine learning optimizes hemodynamic predictors of spinal cord injury outcome.

Artificial intelligence and machine learning (AI/ML) is becoming increasingly more accessible to biomedical researchers with significant potential to transform biomedicine through optimization of highly-accurate predictive models and enabling better understanding of disease biology. Automated machine learning (AutoML) in particular is positioned to democratize artificial intelligence (AI) by reducing the amount of human input and ML expertise needed. However, successful translation of AI/ML in biomedicine requires moving beyond optimizing only for prediction accuracy and towards establishing reproducible clinical and biological inferences. This is especially challenging for clinical studies on rare disorders where the smaller patient cohorts and corresponding sample size is an obstacle for reproducible modeling results. Here, we present a model-agnostic framework to reinforce AutoML using strategies and tools of explainable and reproducible AI, including novel metrics to assess model reproducibility. The framework enables clinicians to interpret AutoML-generated models for clinical and biological verifiability and consequently integrate domain expertise during model development. We applied the framework towards spinal cord injury prognostication to optimize the intraoperative hemodynamic range during injury-related surgery and additionally identified a strong detrimental relationship between intraoperative hypertension and patient outcome. Furthermore, our analysis captured how evolving clinical practices such as faster time-to-surgery and blood pressure management affect clinical model development. Altogether, we illustrate how expert-augmented AutoML improves inferential reproducibility for biomedical discovery and can ultimately build trust in AI processes towards effective clinical integration.

Topological network analysis of patient similarity for precision management of acute blood pressure in spinal cord injury.

Background: Predicting neurological recovery after spinal cord injury (SCI) is challenging. Using topological data analysis, we have previously shown that mean arterial pressure (MAP) during SCI surgery predicts long-term functional recovery in rodent models, motivating the present multicenter study in patients. Methods: Intra-operative monitoring records and neurological outcome data were extracted (n = 118 patients). We built a similarity network of patients from a low-dimensional space embedded using a non-linear algorithm, Isomap, and ensured topological extraction using persistent homology metrics. Confirmatory analysis was conducted through regression methods. Results: Network analysis suggested that time outside of an optimum MAP range (hypotension or hypertension) during surgery was associated with lower likelihood of neurological recovery at hospital discharge. Logistic and LASSO (least absolute shrinkage and selection operator) regression confirmed these findings, revealing an optimal MAP range of 76-[104-117] mmHg associated with neurological recovery. Conclusions: We show that deviation from this optimal MAP range during SCI surgery predicts lower probability of neurological recovery and suggest new targets for therapeutic intervention. Funding: NIH/NINDS: R01NS088475 (ARF); R01NS122888 (ARF); UH3NS106899 (ARF); Department of Veterans Affairs: 1I01RX002245 (ARF), I01RX002787 (ARF); Wings for Life Foundation (ATE, ARF); Craig H. Neilsen Foundation (ARF); and DOD: SC150198 (MSB); SC190233 (MSB).

Spinal cord injury is a devastating condition that involves damage to the nerve fibers connecting the brain with the spinal cord, often leading to permanent changes in strength, sensation and body functions, and in severe cases paralysis. Scientists around the world work hard to find ways to treat or even repair spinal cord injuries but few patients with complete immediate paralysis recover fully. Immediate paralysis is caused by direct damage to neurons and their extension in the spinal cord. Previous research has shown that blood pressure regulation may be key in saving these damaged neurons, as spinal cord injuries can break the communication between nerves that is involved in controlling blood pressure. This can lead to a vicious cycle of dysregulation of blood pressure and limit the supply of blood and oxygen to the damaged spinal cord tissue, exacerbating the death of spinal neurons. Management of blood pressure is therefore a key target for spinal cord injury care, but so far, the precise thresholds to enable neurons to recover are poorly understood. To find out more, Torres-Espin, Haefeli et al. used machine learning software to analyze previously recorded blood pressure and heart rate data obtained from 118 patients that underwent spinal cord surgery after acute spinal cord injury. The analyses revealed that patients who suffered from either low or high blood pressure during surgery had poorer prospects of recovery. Statistical models confirming these findings showed that the optimal blood pressure range to ensure recovery lies between 76 to 104-117 mmHg. Any deviation from this narrow window would dramatically worsen the ability to recover. These findings suggests that dysregulated blood pressure during surgery affects to odds of recovery in patients with a spinal cord injury. Torres-Espin, Haefeli et al. provide specific information that could improve current clinical practice in trauma centers. In the future, such machine learning tools and models could help develop real-time models that could predict the likelihood of a patient's recovery following spinal cord injury and related neurological conditions.

Dexmedetomidine Exerts an Anti-inflammatory Effect via α2 Adrenoceptors to Prevent Lipopolysaccharide-induced Cognitive Decline in Mice.

BACKGROUND: Clinical studies have shown that dexmedetomidine ameliorates cognitive decline in both the postoperative and critical care settings. This study determined the mechanism(s) for the benefit provided by dexmedetomidine in a medical illness in mice induced by lipopolysaccharide. METHODS: Cognitive decline, peripheral and hippocampal inflammation, blood-brain barrier permeability, and inflammation resolution were assessed in male mice. Dexmedetomidine was administered in the presence of lipopolysaccharide and in combination with blockers. Cultured macrophages (RAW 264.7; BV-2) were exposed to lipopolysaccharide ± dexmedetomidine ± yohimbine; tumor necrosis factor α release into the medium and monocyte NFκB activity was determined. RESULTS: In vivo, lipopolysaccharide-induced cognitive decline and inflammation (mean ± SD) were reversed by dexmedetomidine (freezing time, 55.68 ± 12.31 vs. 35.40 ± 17.66%, P = 0.0286, n = 14; plasma interleukin [IL]-1ß: 30.53 ± 9.53 vs. 75.68 ± 11.04 pg/ml, P < 0.0001; hippocampal IL-1ß: 3.66 ± 1.88 vs. 28.73 ± 5.20 pg/mg, P < 0.0001; n = 8), which was prevented by α2 adrenoceptor antagonists. Similar results were found in 12-month-old mice. Lipopolysaccharide also increased blood-brain barrier leakage, inflammation-resolution orchestrator, and proresolving and proinflammatory mediators; each lipopolysaccharide effect was attenuated by dexmedetomidine, and yohimbine prevented dexmedetomidine's attenuating effect. In vitro, lipopolysaccharide-induced tumor necrosis factor α release (RAW 264.7: 6,308.00 ± 213.60 vs. 7,767.00 ± 358.10 pg/ml, P < 0.0001; BV-2: 1,075.00 ± 40.41 vs. 1,280.00 ± 100.30 pg/ml, P = 0.0003) and NFκB-p65 activity (nuclear translocation [RAW 264.7: 1.23 ± 0.31 vs. 2.36 ± 0.23, P = 0.0031; BV-2: 1.08 ± 0.26 vs. 1.78 ± 0.14, P = 0.0116]; phosphorylation [RAW 264.7: 1.22 ± 0.40 vs. 1.94 ± 0.23, P = 0.0493; BV-2: 1.04 ± 0.36 vs. 2.04 ± 0.17, P = 0.0025]) were reversed by dexmedetomidine, which was prevented by yohimbine. CONCLUSIONS: Preclinical studies suggest that the cognitive benefit provided by dexmedetomidine in mice administered lipopolysaccharide is mediated through α2 adrenoceptor-mediated anti-inflammatory pathways.

Clinical Implementation of Novel Spinal Cord Perfusion Pressure Protocol in Acute Traumatic Spinal Cord Injury at U.S. Level I Trauma Center: TRACK-SCI Study.

OBJECTIVE: We sought to report the safety of implementation of a novel standard of care protocol using spinal cord perfusion pressure (SCPP) maintenance for managing traumatic spinal cord injury (SCI) in lieu of mean arterial pressure goals at a U.S. Level I trauma center. METHODS: Starting in December 2017, blunt SCI patients presenting <24 hours after injury with admission American Spinal Injury Association Impairment Scale (AIS) A-C (or AIS D at neurosurgeon discretion) received lumbar subarachnoid drain (LSAD) placement for SCPP monitoring in the intensive care unit and were included in the TRACK-SCI (Transforming Research and Clinical Knowledge in Spinal Cord Injury) data registry. This SCPP protocol comprises standard care at our institution. SCPPs were monitored for 5 days (goal ≥65 mm Hg) achieved through intravenous fluids and vasopressor support. AISs were assessed at admission and day 7. RESULTS: Fifteen patients enrolled to date were aged 60.5 ± 17 years. Injury levels were 93.3% (cervical) and 6.7% (thoracic). Admission AIS was 20.0%/20.0%/26.7%/33.3% for A/B/C/D. All patients maintained mean SCPP ≥65 mm Hg during monitoring. Fourteen of 15 cases required surgical decompression and stabilization with time to surgery 8.8 ± 7.1 hours (71.4% <12 hours). At day 7, 33.3% overall and 50% of initial AIS A-C had an improved AIS. Length of stay was 14.7 ± 8.3 days. None had LSAD-related complications. There were 7 respiratory complications. One patient expired after transfer to comfort care. CONCLUSIONS: In our initial experience of 15 patients with acute SCI, standardized SCPP goal-directed care based on LSAD monitoring for 5 days was feasible. There were no SCPP-related complications. This is the first report of SCPP implementation as clinical standard of care in acute SCI.

Dexmedetomidine modulates neuroinflammation and improves outcome via alpha2-adrenergic receptor signaling after rat spinal cord injury.

BACKGROUND: Spinal cord injury induces inflammatory responses that include the release of cytokines and the recruitment and activation of macrophages and microglia. Neuroinflammation at the lesion site contributes to secondary tissue injury and permanent locomotor dysfunction. Dexmedetomidine (DEX), a highly selective α2-adrenergic receptor agonist, is anti-inflammatory and neuroprotective in both preclinical and clinical trials. We investigated the effect of DEX on the microglial response, and histological and neurological outcomes in a rat model of cervical spinal cord injury. METHODS: Anaesthetised rats underwent unilateral (right) C5 spinal cord contusion (75 kdyne) using an impactor device. The locomotor function, injury size, and inflammatory responses were assessed. The effect of DEX was also studied in a microglial cell culture model. RESULTS: DEX significantly improved the ipsilateral upper-limb motor dysfunction (grooming and paw placement; P<0.0001 and P=0.0012), decreased the injury size (P<0.05), spared white matter (P<0.05), and reduced the number of activated macrophages (P<0.05) at the injury site 4 weeks post-SCI. In DEX-treated rats after injury, tissue RNA expression indicated a significant downregulation of pro-inflammatory markers (e.g. interleukin [IL]-1ß, tumour necrosis factor-α, interleukin (IL)-6, and CD11b) and an upregulation of anti-inflammatory and pro-resolving M2 responses (e.g. IL-4, arginase-1, and CD206) (P<0.05). In lipopolysaccharide-stimulated cultured microglia, DEX produced a similar inflammation-modulatory effect as was seen in spinal cord injury. The benefits of DEX on these outcomes were mostly reversed by an α2-adrenergic receptor antagonist. CONCLUSIONS: DEX significantly improves neurological outcomes and decreases tissue damage after spinal cord injury, which is associated with modulation of neuroinflammation and is partially mediated via α2-adrenergic receptor signaling.

Ultra-Early (<12 Hours) Surgery Correlates With Higher Rate of American Spinal Injury Association Impairment Scale Conversion After Cervical Spinal Cord Injury.

BACKGROUND: Cervical spinal cord injury (SCI) is a devastating condition with very few treatment options. It remains unclear if early surgery correlated with conversion of American Spinal Injury Association Impairment Scale (AIS) grade A injuries to higher grades. OBJECTIVE: To determine the optimal time to surgery after cervical SCI through retrospective analysis. METHODS: We collected data from 48 patients with cervical SCI. Based on the time from Emergency Department (ED) presentation to surgical decompression, we grouped patients into ultra-early (decompression within 12 h of presentation), early (within 12-24 h), and late groups (>24 h). We compared the improvement in AIS grade from admission to discharge, controlling for confounding factors such as AIS grade on admission, injury severity, and age. The mean time from injury to ED for this group of patients was 17 min. RESULTS: Patients who received surgery within 12 h after presentation had a relative improvement in AIS grade from admission to discharge: the ultra-early group improved on average 1.3. AIS grades compared to 0.5 in the early group (P = .02). In addition, 88.8% of patients with an AIS grade A converted to a higher grade (AIS B or better) in the ultra-early group, compared to 38.4% in the early and late groups (P = .054). CONCLUSION: These data suggest that surgical decompression after SCI that takes place within 12 h may lead to a relative improved neurological recovery compared to surgery that takes place after 12 h.

Motor Evoked Potentials Correlate With Magnetic Resonance Imaging and Early Recovery After Acute Spinal Cord Injury.

BACKGROUND: While the utilization of neurophysiologic intraoperative monitoring with motor evoked potentials (MEPs) has become widespread in surgery for traumatic spine fractures and spinal cord injury (SCI), clinical validation of its diagnostic and therapeutic benefit has been limited. OBJECTIVE: To describe the use of intraoperative MEP at a large level I trauma center and assess the prognostic capability of this technology. METHODS: The SCI REDCap database at our institution, a level I trauma center, was queried for acute cervical SCI patients who underwent surgery with intraoperative monitoring between 2005 and 2011, yielding 32 patients. Of these, 23 patients had severe SCI (association impairment scale [AIS] A, B, C). We assessed preoperative and postoperative SCI severity (AIS grade), surgical data, use of steroids, and early magnetic resonance imaging (MRI) findings (preoperatively in 27 patients), including axial T2 MRI grade (Brain and Spinal Injury Center score). RESULTS: The presence of MEPs significantly predicted AIS at discharge (P< .001). In the group of severe SCI (ie, AIS A, B, C) patients with elicitable MEPs, AIS improved by an average of 1.5 grades (median = 1), as compared to the patients without elicitable MEP who improved on average 0.5 grades (median = 0, P< .05). In addition, axial MRI grade significantly correlated with MEP status. Patients without MEPs had a significantly higher axial MRI grade in comparison to the patients with MEPs (P< .001). CONCLUSION: In patients with severe SCI, MEPs predicted neurological improvement and correlated with axial MRI grade. These significant findings warrant future prospective studies of MEPs as a prognostic tool in SCI.

Complications and outcomes of vasopressor usage in acute traumatic central cord syndrome.

OBJECT The optimal mean arterial pressure (MAP) for spinal cord perfusion after trauma remains unclear. Although there are published data on MAP goals after spinal cord injury (SCI), the specific blood pressure management for acute traumatic central cord syndrome (ATCCS) and the implications of these interventions have yet to be elucidated. Additionally, the complications of specific vasopressors have not been fully explored in this injury condition. METHODS The present study is a retrospective cohort analysis of 34 patients with ATCCS who received any vasopressor to maintain blood pressure above predetermined MAP goals at a single Level 1 trauma center. The collected variables were American Spinal Injury Association (ASIA) grades at admission and discharge, administered vasopressor and associated complications, other interventions and complications, and timing of surgery. The relationship between the 2 most common vasopressors-dopamine and phenylephrine-and complications within the cohort as a whole were explored, and again after stratification by age. RESULTS The mean age of the ATCCS patients was 62 years. Dopamine was the most commonly used primary vasopressor (91% of patients), followed by phenylephrine (65%). Vasopressors were administered to maintain MAP goals fora mean of 101 hours. Neurological status improved by a median of 1 ASIA grade in all patients, regardless of the choice of vasopressor. Sixty-four percent of surgical patients underwent decompression within 24 hours. There was no observed relationship between the timing of surgical intervention and the complication rate. Cardiogenic complications associated with vasopressor usage were notable in 68% of patients who received dopamine and 46% of patients who received phenylephrine. These differences were not statistically significant (OR with dopamine 2.50 [95% CI 0.82-7.78], p = 0.105). However, in the subgroup of patients > 55 years, dopamine produced statistically significant increases in the complication rates when compared with phenylephrine (83% vs 50% for dopamine and phenylephrine, respectively; OR with dopamine 5.0 [95% CI 0.99-25.34], p = 0.044). CONCLUSIONS Vasopressor usage in ATCCS patients is associated with complication rates that are similar to the reported literature for SCI. Dopamine was associated with a higher risk of complications in patients > 55 years. Given the increased incidence of ATCCS in older populations, determination of MAP goals and vasopressor administration should be carefully considered in these patients. While a randomized control trial on this topic may not be practical, a multiinstitutional prospective study for SCI that includes ATCCS patients as a subpopulation would be useful for examining MAP goals in this population.

Cytokine activity contributes to induction of inflammatory cytokine mRNAs in spinal cord following contusion.

Injury of the spinal cord leads to an inflammatory tissue response, probably mediated in part by cytokines. Because a common therapy for acute spinal cord injury is the use of an antiinflammatory synthetic glucocorticoid (methylprednisolone), we sought to determine mechanisms contributing to inflammation shortly after acute injury. Cytokine mRNAs [interleukin (IL)-1alpha, IL-1beta, tumor necrosis factor (TNF)-alpha, and IL-6] were increased during the first 2 hr following weight-drop compression injury by RNase protection assay, prior to the reported appearance of circulating lymphocytes. This immediate pattern of cytokine mRNA induction could be replicated in cultured, explanted spinal cord slices but not in whole blood of injured animals, which is consistent with a tissue source of cytokine mRNAs. Western blotting detected IL-1beta-like immunoreactivity released into culture medium following explantation and pro-IL-1beta-like immunoreactivity in freshly dissected spinal cord tissue. Pharmacologically blocking IL-1 and TNF-alpha receptors significantly reduced expression of IL-1alpha, IL-1beta, and TNF-alpha mRNAs. Finally, mice lacking both IL-1 and TNF-alpha receptors exhibited diminished induction of TNF-alpha, IL-6, and IL-1ra mRNAs following injury. Therefore, we conclude that contusion injury induces an immediate release of cytokines, which then contributes to the induction of cytokine mRNAs.