High-Accuracy Neuro-Navigation with Computer Vision for Frameless Registration and Real-Time Tracking

For the past three decades, neurosurgeons have utilized cranial neuro-navigation systems, bringing millimetric accuracy to operating rooms worldwide. These systems require an operating room team, anesthesia, and, most critically, cranial fixation. As a result, treatments for acute neurosurgical conditions, performed urgently in emergency rooms or intensive care units on awake and non-immobilized patients, have not benefited from traditional neuro-navigation. These emergent procedures are performed freehand, guided only by anatomical landmarks with no navigation, resulting in inaccurate catheter placement and neurological deficits. A rapidly deployable image-guidance technology that offers highly accurate, real-time registration and is capable of tracking awake, moving patients is needed to improve patient safety. The Zeta Cranial Navigation System is currently the only non-fiducial-based, FDA-approved neuro-navigation device that performs real-time registration and continuous patient tracking. To assess this system's performance, we performed registration and tracking of phantoms and human cadaver heads during controlled motions and various adverse surgical test conditions. As a result, we obtained millimetric or sub-millimetric target and surface registration accuracy. This rapid and accurate frameless neuro-navigation system for mobile subjects can enhance bedside procedure safety and expand the range of interventions performed with high levels of accuracy outside of an operating room.

External Ventricular Drains versus Intraparenchymal Pressure Monitors in the Management of Moderate to Severe Traumatic Brain Injury: Experience at Two Academic Centers over a Decade

OBJECTIVE

The choice between external ventricular drain (EVD) and intraparenchymal monitor (IPM) for managing intracranial pressure in moderate-to-severe traumatic brain injury (msTBI) patients remains controversial. This study aimed to investigate factors associated with receiving EVD versus IPM and to compare outcomes and clinical management between EVD and IPM patients.

METHODS

Adult msTBI patients at 2 similar academic institutions were identified. Logistic regression was performed to identify factors associated with receiving EVD versus IPM (model 1) and to compare EVD versus IPM in relation to patient outcomes after controlling for potential confounders (model 2), through odds ratios (ORs) and 95% confidence intervals (CIs).

RESULTS

Of 521 patients, 167 (32.1%) had EVD and 354 (67.9%) had IPM. Mean age, sex, and Injury Severity Score were comparable between groups. Epidural hemorrhage (EDH) (OR 0.43, 95% CI 0.21-0.85), greater midline shift (OR 0.90, 95% CI 0.82-0.98), and the hospital with higher volume (OR 0.14, 95% CI 0.09-0.22) were independently associated with lower odds of receiving an EVD whereas patients needing a craniectomy were more likely to receive an EVD (OR 2.04, 95% CI 1.12-3.73). EVD patients received more intense medical treatment requiring hyperosmolar therapy compared to IPM patients (64.1% vs. 40.1%). No statistically significant differences were found in patient outcomes.

CONCLUSIONS

While EDH, greater midline shift, and hospital with larger patient volume were associated with receiving an IPM, the need for a craniectomy was associated with receiving an EVD. EVD patients received different clinical management than IPM patients with no significant differences in patient outcomes.

External ventricular drains and risk of freehand placement: A systematic review and meta-analysis

OBJECTIVE

External ventricular drains (EVDs) are used to monitor and treat elevated intracranial pressure. EVDs are often placed blindly without the use of imaging guidance, and successful placement with respect to pass attempts and final catheter location may suffer as a result of this freehand technique.

METHODS

A systematic literature search was conducted in PubMed, Embase, Web of Science, and Cochrane databases to identify studies pertaining to freehand EVD placement through March 30, 2022. Studies were included if they reported percentage of EVDs placed successfully on the first pass attempt, or final catheter location as defined by the Kakarla Grading System. Pooled weighted incidence estimates and 95% confidence intervals (95%CI) were calculated using a random effects model.

RESULTS

Of the 2964 results returned from the literature search, 39 studies were included in this meta-analysis. These studies reported on 6313 EVDs placed via freehand technique in 6070 patients with the following respective incidence: successful EVD placement on the first attempt (78%, 95%CI: 67-86%); placement with a Kakarla Grade of 1 (optimal location) (72%, 95%CI: 66-77%); hemorrhage (7%, 95%CI: 6-10%), and infection (5%, 95%CI: 3-8%).

CONCLUSIONS

Only 78% of EVDs in this meta-analysis were placed successfully on the first pass, and only 72% of final placements were deemed optimal. This represents a relatively high rate of suboptimal outcomes with respect to EVD placement, which could potentially be avoided with the use of navigation-assisted placement techniques.

Disparities in Decompressive Cranial Surgery Utilization in Severe Traumatic Brain Injury Patients without a Primary Extra-Axial Hematoma: A U.S. Nationwide Study

OBJECTIVE

Decompressive craniectomy is recommended to reduce mortality in severe traumatic brain injury (TBI). Disparities exist in TBI treatment outcomes; however, data on disparities pertaining to decompressive craniectomy utilization is lacking. We investigated these disparities, focusing on race, insurance, sex, and age.

METHODS

Hospitalizations (2004-2014) were retrospectively extracted from the Nationwide Inpatient Sample. The criteria included are as follows: age ≥18 years and indicators of severe TBI diagnosis. Poor outcomes were defined as discharge to institutional care and death. Multivariable logistic regression models were used to assess the effects of race, insurance, age, and sex, on craniectomy utilization and outcomes.

RESULTS

Of 349,164 hospitalized patients, 6.8% (n = 23,743) underwent craniectomy. White (odds ratio [OR] = 0.50, 95% confidence interval [CI] = 0.44-0.57; P < 0.001) and Black (OR = 0.45, 95% CI = 0.32-0.64; P = 0.003) Medicare beneficiaries were less likely to undergo craniectomy. Medicare (P < 0.0001) and Medicaid beneficiaries (P < 0.0001) of all race categories had poorer outcomes than privately insured White patients. Black (OR = 1.2, 95% CI = 1.08-2.34; P = 0.001) patients with private insurance and Black (OR = 1.39, 95% CI = 1.22-1.58; P < 0.0001) Medicaid beneficiaries had poorer outcomes than privately insured White patients (P < 0.0001). Older patients (OR = 0.74, 95%, CI = 0.71-0.76; P < 0.001) were less likely to undergo craniectomy and were more likely to have poorer outcomes. Females (OR = 0.82, 95% CI = 0.76-0.88; P < 0.001) were less likely to undergo craniectomy.

CONCLUSIONS

There are disparities in race, insurance status, sex, and age in craniectomy utilization and outcome. This data highlights the necessity to appropriately address these disparities, especially race and sex, and actively incorporate these factors in clinical trial design and enrollment.

DDEL-01. ANIMAL STUDIES WITH A NEW CNS DRUG DELIVERY DEVICE TO EFFECTIVELY TREAT LEPTOMENINGEAL CARCINOMATOSIS PATIENTS

Background

Appropriate drug distribution is needed for effective treatment of brain metastases and leptomeningeal carcinomatosis. Lack of proper drug bioavailability in the CNS contributes to the poor outcome in these patients. EnClear Therapies has developed a device delivering intrathecal therapeutics utilizing the active control of CSF flow through an external pump. Herein, we look at intrathecal drug delivery of methotrexate with EnClear’s system and its impact of drug concentration in sheep and non-human primates.

METHODS

The EnClear device utilizes two implantable catheters (intracranial in the ventricular system and lumbar thecal sac) and an extracorporeal pump that controls the speed and direction of CSF flow. Software monitors the output from sensor arrays for pressure, respiration, and heart rate. Gadolinium and MRI were used to determine the CSF flow characteristics. Administration of Methotrexate via the EnClear system was compared to current two standards of delivery, intracerebroventricular or lumbar intrathecal administration. Methotrexate levels were measured with Liquid Chromatography/Mass Spectrometry in CSF, blood, and multiple CNS and peripheral tissues.

RESULTS

There was a two-fold increase in methotrexate levels in multiple regions of the brain including deeper brain structures including the striatum using the EnClear's system, compared to traditional lumbar intrathecal and intracerebroventricular delivery. Higher peak levels of methotrexate were reached in the CSF with the use of EnClear’s system, with simultaneous reduction in the peripheral nerves and systemic tissues – a potential source of toxicity.

Conclusions

In large animal studies, the EnClear device has for the first time, demonstrated increased brain parenchymal drug levels via intrathecal delivery. In addition to the drug parenchyma, drug delivery with this novel device has shown improved distribution throughout CNS and leptomeninges as compared to traditional intrathecal drug delivery and decreased peripheral distribution.

Frameless neuronavigation with computer vision and real-time tracking for bedside external ventricular drain placement: a cadaveric study

OBJECTIVE

A major obstacle to improving bedside neurosurgical procedure safety and accuracy with image guidance technologies is the lack of a rapidly deployable, real-time registration and tracking system for a moving patient. This deficiency explains the persistence of freehand placement of external ventricular drains, which has an inherent risk of inaccurate positioning, multiple passes, tract hemorrhage, and injury to adjacent brain parenchyma. Here, the authors introduce and validate a novel image registration and real-time tracking system for frameless stereotactic neuronavigation and catheter placement in the nonimmobilized patient.

METHODS

Computer vision technology was used to develop an algorithm that performed near-continuous, automatic, and marker-less image registration. The program fuses a subject's preprocedure CT scans to live 3D camera images (Snap-Surface), and patient movement is incorporated by artificial intelligence-driven recalibration (Real-Track). The surface registration error (SRE) and target registration error (TRE) were calculated for 5 cadaveric heads that underwent serial movements (fast and slow velocity roll, pitch, and yaw motions) and several test conditions, such as surgical draping with limited anatomical exposure and differential subject lighting. Six catheters were placed in each cadaveric head (30 total placements) with a simulated sterile technique. Postprocedure CT scans allowed comparison of planned and actual catheter positions for user error calculation.

RESULTS

Registration was successful for all 5 cadaveric specimens, with an overall mean (± standard deviation) SRE of 0.429 ± 0.108 mm for the catheter placements. Accuracy of TRE was maintained under 1.2 mm throughout specimen movements of low and high velocities of roll, pitch, and yaw, with the slowest recalibration time of 0.23 seconds. There were no statistically significant differences in SRE when the specimens were draped or fully undraped (p = 0.336). Performing registration in a bright versus a dimly lit environment had no statistically significant effect on SRE (p = 0.742 and 0.859, respectively). For the catheter placements, mean TRE was 0.862 ± 0.322 mm and mean user error (difference between target and actual catheter tip) was 1.674 ± 1.195 mm.

CONCLUSIONS

This computer vision-based registration system provided real-time tracking of cadaveric heads with a recalibration time of less than one-quarter of a second with submillimetric accuracy and enabled catheter placements with millimetric accuracy. Using this approach to guide bedside ventriculostomy could reduce complications, improve safety, and be extrapolated to other frameless stereotactic applications in awake, nonimmobilized patients.

Seizure Outcomes After Interventional Treatment in Cerebral Arteriovenous Malformation-Associated Epilepsy: A Systematic Review and Meta-Analysis

BACKGROUND

Seizures are the second most common presenting symptom of cerebral arteriovenous malformations (AVMs). Evidence supporting different treatment modalities is continuously evolving and it remains unclear which modality offers better seizure outcomes.

OBJECTIVE

To compare various interventional treatment modalities (i.e., microsurgery, radiosurgery, endovascular embolization, or multimodality treatment), regarding outcomes in AVM-associated epilepsy.

METHODS

PubMed, Embase, and Web of Science were searched on December 31, 2020 for studies that evaluated outcomes in patients with AVM-associated epilepsy after undergoing different treatment modalities. Pooled analysis was performed using a random-effects model and stratified by different modalities.

RESULTS

Forty-nine studies including 2668 patients were included. Interventional management was associated with a 56.0% probability of seizure freedom and a 73.0% probability of seizure improvement. The probability of discontinuing antiepileptic drugs was estimated at 38.0%. The stratified analysis showed that microsurgery was associated with a higher probability of seizure freedom and seizure improvement than was radiosurgery, endovascular, or multimodality treatment. The probability of antiepileptic drug cessation was also higher after microsurgery compared with radiation therapy; however, only clinical but not statistical significance could be inferred because of the lack of comparative analyses.

CONCLUSIONS

Interventional management of AVM-related epilepsy was associated with seizure freedom and seizure improvement in 56% and 73% of cases. Microsurgery seemed to be associated with a higher incidence of seizure freedom and seizure improvement than did other modalities. Future well-designed comparative studies are needed to draw definitive conclusions regarding each modality.

Evaluation of simulation models in neurosurgical training according to face, content, and construct validity: a systematic review

BACKGROUND

Neurosurgical training has been traditionally based on an apprenticeship model. However, restrictions on clinical exposure reduce trainees' operative experience. Simulation models may allow for a more efficient, feasible, and time-effective acquisition of skills. Our objectives were to use face, content, and construct validity to review the use of simulation models in neurosurgical education.

METHODS

PubMed, Web of Science, and Scopus were queried for eligible studies. After excluding duplicates, 1204 studies were screened. Eighteen studies were included in the final review.

RESULTS

Neurosurgical skills assessed included aneurysm clipping (n = 6), craniotomy and burr hole drilling (n = 2), tumour resection (n = 4), and vessel suturing (n = 3). All studies assessed face validity, 11 assessed content, and 6 assessed construct validity. Animal models (n = 5), synthetic models (n = 7), and VR models (n = 6) were assessed. In face validation, all studies rated visual realism favourably, but haptic realism was key limitation. The synthetic models ranked a high median tactile realism (4 out of 5) compared to other models. Assessment of content validity showed positive findings for anatomical and procedural education, but the models provided more benefit to the novice than the experienced group. The cadaver models were perceived to be the most anatomically realistic by study participants. Construct validity showed a statistically significant proficiency increase among the junior group compared to the senior group across all modalities.

CONCLUSION

Our review highlights evidence on the feasibility of implementing simulation models in neurosurgical training. Studies should include predictive validity to assess future skill on an individual on whom the same procedure will be administered. This study shows that future neurosurgical training systems call for surgical simulation and objectively validated models.

The low utility of routine cranial imaging after pediatric shunt revision

OBJECTIVE

Postoperative routine imaging is common after pediatric ventricular shunt revision, but the benefit of scanning in the absence of symptoms is questionable. In this study, the authors aimed to assess how often routine scanning results in a change in clinical management after shunt revision.

METHODS

The records of a large, tertiary pediatric hospital were retrospectively reviewed for all consecutive cases of pediatric shunt revision between July 2013 and July 2018. Postoperative imaging was classified as routine (i.e., in the absence of symptoms, complications, or other direct indications) or nonroutine. Reinterventions within 30 days were assessed in these groups.

RESULTS

Of 387 included shunt revisions performed in 232 patients, postoperative imaging was performed in 297 (77%), which was routine in 244 (63%) and nonroutine in 53 (14%). Ninety revisions (23%) underwent any shunt-related procedure after postoperative imaging, including shunt reprogramming (n = 35, 9%), shunt tap (n = 10, 3%), and a return to the operating room (OR; n = 58, 15%). Of the 244 cases receiving routine imaging, 241 did not undergo a change in clinical management solely based on routine imaging findings. The remaining 3 cases returned to the OR, accounting for 0.8% (95% CI 0.0%-1.7%) of all cases or 1.2% (95% CI 0.0%-2.6%) of cases that received routine imaging. Furthermore, 27 of 244 patients in this group returned to the OR for other reasons, namely complications (n = 12) or recurrent symptoms (n = 15); all arose after initial routine imaging.

CONCLUSIONS

The authors found a low yield to routine imaging after pediatric shunt revision, with only 0.8% of cases undergoing a change in management based on routine imaging findings without corresponding clinical findings. Moreover, routine imaging without abnormal findings was no guarantee of an uneventful postoperative course. Clinical monitoring can be considered as an alternative in asymptomatic, uncomplicated patients.

Incidence and Outcomes of Registry-Based Acute Myocardial Infarction After Aneurysmal Subarachnoid Hemorrhage

BACKGROUND

Acute myocardial infarction (AMI) is the rarest and least studied cardiac complication of aneurysmal subarachnoid hemorrhage (aSAH). Precise estimates of the incidence of AMI after aSAH are unavailable. Our goal was to estimate the incidence of registry-based AMI (rb-AMI) after aSAH and determine its association with clinical outcomes.

METHODS

Adult patients with aSAH in the National Inpatient Samples from 2002 to 2014 were included in the study. We evaluated risk factors for rb-AMI using univariate and multivariate regression models. Clinical outcomes that were assessed included functional status at discharge, in-patient mortality, length of stay, and total hospitalization cost, adjusting for patient demographics and cardiovascular risk factors through an inverse probability weighted analysis. Subgroup analyses were further performed stratified by rb-AMI type (ST-segment elevation myocardial infarction [STEMI] vs. non-STEMI [NSTEMI]).

RESULTS

A total of 139,734 patients with aSAH were identified, 3.6% of whom had rb-AMI. NSTEMI was the most common type of rb-AMI occurring after aSAH (71% vs. 29% for NSTEMI vs. STEMI, respectively). Patient characteristics associated with higher odds of rb-AMI included age, female sex, poor aSAH grade, and various cardiovascular risk factors. Rb-AMI was also associated with poor functional status at discharge, higher in-hospital mortality, and a longer and more costly hospital stay.

CONCLUSIONS

Rb-AMI occurs in 3.6% of patients with aSAH and is associated with poor functional status at discharge, higher in-patient mortality, and a longer and more costly hospitalization. Differentiating between different types of rb-AMI would be important in optimizing the management of patients with aSAH. Our definition of rb-AMI likely includes patients with neurogenic stress cardiomyopathy, which may confound the results.

Ruptured Suprasellar Dermoid Cyst Treated With Lumbar Drain to Prevent Postoperative Hydrocephalus: Case Report and Focused Review of Literature

Background: Ruptured intracranial dermoid cysts are extremely rare. Standard treatment consists of endonasal decompression or craniotomy with evacuation and copious irrigation of subarachnoid spaces to remove any disseminated cystic contents. Disseminated fat particles in the subarachnoid space may be the cause of further sequalae, including the subsequent development of chemical meningitis and hydrocephalus. Here, we present a case of ruptured suprasellar dermoid cyst treated with craniotomy for emergent optic nerve decompression, followed by postoperative hydrocephalus successfully treated with lumbar drain.

Case description: We describe a 30-year-old man with a history of migraines who presented with acute onset of headache, photophobia, nausea, vomiting, and vision loss in the left eye. Head CT and brain MRI demonstrated a ruptured suprasellar dermoid cyst with associated mass effect on the optic nerves and frontal lobes as well as fat attenuation material within the subarachnoid spaces. The patient underwent left frontotemporal craniotomy for cyst resection and developed non-obstructive hydrocephalus on postoperative day 1, refractory to external ventricular drainage. Placement of a lumbar drain cleared the subarachnoid space of debris derived from the ruptured dermoid cyst, and the hydrocephalus resolved. The patient did not require permanent CSF diversion.

Conclusions: Intracranial dermoid cysts are uncommon, and rupture is a rare event. Standard surgical treatment with craniotomy for evacuation may leave disseminated dermoid contents and fat particles throughout the subarachnoid spaces. We highlight a case of ruptured suprasellar dermoid cyst with postoperative communicating hydrocephalus treated with lumbar drain when external ventricular drain (EVD) was ineffective. Review of the current literature reveals inconsistent findings on the effects of remaining fat particles. In cases with clinical evidence of increased intracranial pressure due to non-obstructive hydrocephalus attributable to chemical meningitis, temporary lumbar drainage is an option to be considered before committing the patient to permanent shunting.

The Impact of Age and Severity on Dementia After Traumatic Brain Injury: A Comparison Study

BACKGROUND

Growing evidence associates traumatic brain injury (TBI) with increased risk of dementia, but few studies have evaluated associations in patients younger than 55 yr using non-TBI orthopedic trauma (NTOT) patients as controls to investigate the influence of age and TBI severity, and to identify predictors of dementia after trauma.

OBJECTIVE

To investigate the relationship between TBI and dementia in an institutional group.

METHODS

Retrospective cohort study (2000-2018) of TBI patients aged 45 to 100 yr vs NTOT controls. Primary outcome was dementia after TBI (followed ≤10 yr). Cox proportional hazards models were used to assess risk of dementia; logistic regression models assessed predictors of dementia.

RESULTS

Among 24 846 patients, TBI patients developed dementia (7.5% vs 4.6%) at a younger age (78.6 vs 82.7 yr) and demonstrated higher 10-yr mortality than controls (27% vs 14%; P < .001). Mild TBI patients had higher incidence of dementia (9%) than moderate/severe TBI (5.4%), with lower 10-yr mortality (20% vs 31%; P < .001). Risk of dementia was significant in all mild TBI age groups, even 45 to 54 yr (hazard ratio 4.1, 95% CI 2.7-7.8). A total of 10-yr cumulative incidence was higher in mild TBI (14.4%) than moderate/severe TBI (11.3%) and controls (6.8%) (P < .001). Predictors of dementia include TBI, sex, age, hypertension, hyperlipidemia, stroke, depression, anxiety, and Injury Severity Score.

CONCLUSION

Mild and moderate/severe TBI patients experienced higher incidence of dementia, even in the youngest group (45-54 yr old), than NTOT controls. All TBI patients, especially middle-aged adults with minor injury who are more likely to be overlooked, should be monitored for dementia.

Expandable Versus Static Cages in Minimally Invasive Lumbar Interbody Fusion: A Systematic Review and Meta-Analysis

BACKGROUND

Expandable cages for interbody fusion allow for in situ expansion optimizing fit while mitigating endplate damage. Studies comparing outcomes after using expandable or static cages have been conflicting.

METHODS

This was a meta-analysis A systematic search was performed in accordance with the Preferred Reporting Items for Systemic Reviews and Meta-Analyses (PRISMA) guidelines identifying studies reporting outcomes among patients who underwent minimally invasive lumbar interbody fusion (MIS-LIF).

RESULTS

Fourteen articles with 1129 patients met inclusion criteria. Compared with MIS-LIFs performed with static cages, those with expandable cages had a significantly lower incidence of graft subsidence (expandable: incidence 0.03, I² 22.50%; static: incidence 0.27, I² 51.03%, P interaction <0.001), length of hospital stay (expandable: mean difference [MD] 3.55 days, I² 97%; static: MD 7.1 days, I² 97%, P interaction <0.01), and a greater increase in disc height (expandable: MD -4.41 mm, I² 99.56%; static: MD -0.79 mm, I² 99.17%, P interaction = 0.02). There was no statistically significant difference among Oswestry Disability Index (expandable: MD -22.75, I² 98.17%; static: MD -17.11, I² 95.26%, P interaction = 0.15), fusion rate (expandable: incidence 0.94, I² 0%; static incidence 0.92, I² 0%, P interaction = 0.44), overall change in lumbar lordosis (expandable: MD 3.48 degrees, I² 59.29%; static: MD 3.67 degrees, I² 0.00%, P interaction 0.88), blood loss (expandable: MD 228.9 mL, I² 100%; static: MD 261.1 mL, I² 94%, P interaction = 0.69) and operative time (expandable: MD 184 minutes, I² 95.32%; static: MD 150.4 minutes, I² 91%, P interaction = 0.56).

CONCLUSIONS

Expandable interbody cages in MIS-LIF were associated with a decrease in subsidence rate, operative time and greater in increase in disc height.

Familial Predisposition and Differences in Radiographic Patterns in Spontaneous Nonaneurysmal Subarachnoid Hemorrhage

BACKGROUND

Subarachnoid hemorrhage (SAH) from an intracranial aneurysmal rupture is the most common nontraumatic etiology for SAH, but up to 15% of patients with SAH have no identifiable source.

OBJECTIVE

To assess familial predisposition to spontaneous nonaneurysmal SAH (naSAH) and to evaluate whether family history affects the severity of presentation and prognosis of this condition.

METHODS

We conducted a retrospective analysis of all spontaneous SAH with negative digital subtraction angiography from 2004 to 2018. Patients were divided into 2 groups: patients with first- or second-degree relatives with intracranial aneurysms and patients with no family history. Univariate and multivariate regression analyses were used to study patient presentation, radiographic patterns of hemorrhage, and clinical outcome.

RESULTS

A total of 100 patients met the inclusion criteria. There were no individuals with family history of naSAH. A total of 15 patients (15%) had at least one family member with an intracranial aneurysm, of which 12 (12%) presented as SAH. Patients without family history had a higher percentage of perimesencephalic presentation, whereas those with family history had a higher percentage of nonperimesencephalic SAH presentation (47% vs 13%, odds ratio [OR] 0.17 [95% CI 0.04, 0.81]).

CONCLUSION

We found a high rate of family history of intracranial aneurysms in patients who presented with naSAH. Although there was no difference in clinical outcome in patients with and without family history, there appears to be a higher percentage of nonperimesencephalic radiographic patterns of SAH in those with family history, suggesting possible different etiologies of these hemorrhages.

Automating Clinical Chart Review: An Open-Source Natural Language Processing Pipeline Developed on Free-Text Radiology Reports From Patients With Glioblastoma

PURPOSE

The aim of this study was to develop an open-source natural language processing (NLP) pipeline for text mining of medical information from clinical reports. We also aimed to provide insight into why certain variables or reports are more suitable for clinical text mining than others.

MATERIALS AND METHODS

Various NLP models were developed to extract 15 radiologic characteristics from free-text radiology reports for patients with glioblastoma. Ten-fold cross-validation was used to optimize the hyperparameter settings and estimate model performance. We examined how model performance was associated with quantitative attributes of the radiologic characteristics and reports.

RESULTS

In total, 562 unique brain magnetic resonance imaging reports were retrieved. NLP extracted 15 radiologic characteristics with high to excellent discrimination (area under the curve, 0.82 to 0.98) and accuracy (78.6% to 96.6%). Model performance was correlated with the inter-rater agreement of the manually provided labels (ρ = 0.904; P < .001) but not with the frequency distribution of the variables of interest (ρ = 0.179; P = .52). All variables labeled with a near perfect inter-rater agreement were classified with excellent performance (area under the curve > 0.95). Excellent performance could be achieved for variables with only 50 to 100 observations in the minority group and class imbalances up to a 9:1 ratio. Report-level classification accuracy was not associated with the number of words or the vocabulary size in the distinct text documents.

CONCLUSION

This study provides an open-source NLP pipeline that allows for text mining of narratively written clinical reports. Small sample sizes and class imbalance should not be considered as absolute contraindications for text mining in clinical research. However, future studies should report measures of inter-rater agreement whenever ground truth is based on a consensus label and use this measure to identify clinical variables eligible for text mining.

Extra-Axial Fluid Collections After Decompressive Craniectomy: Management, Outcomes, and Treatment Algorithm

BACKGROUND

Extra-axial fluid collections (EACs) frequently develop after decompressive craniectomy. Management of EACs remains poorly understood, and information on how to predict their clinical course is inadequate. We aimed to better characterize EACs, understand predictors of their resolution, and delineate the best treatment paradigm for patients.

METHODS

We reviewed patients who developed EACs after undergoing decompressive craniectomy for treatment of refractory intracranial pressure elevations. We excluded patients who had an ischemic stroke, as EACs in these patients have a different clinical course. We performed univariate analysis and multiple linear regression to find variables associated with earlier resolution of EACs and stratified our analyses by EAC phenotype (complicated vs. uncomplicated). We conducted a systematic review to compare our findings with the literature.

RESULTS

Of 96 included patients, 73% were male, and median age was 42.5 years. EACs resolved after a median of 60 days. Complicated EACs were common (62.5%) and required multiple drainage methods before cranioplasty. These were not associated with a protracted course or increased risk of death (P > 0.05). Early bone flap restoration with simultaneous drainage was independently associated with earlier resolution of EACs (β = 0.56, P < 0.001). Systematic review confirmed lack of standardized direction with respect to EAC management.

CONCLUSIONS

Our analyses reveal 2 clinically relevant phenotypes of EAC: complicated and uncomplicated. Our proposed treatment algorithm involves replacing the bone flap as soon as it is safe to do so and draining refractory EACs aggressively. Further studies to assess long-term clinical outcomes of EACs are warranted.

Divergence in the epidemiological estimates of traumatic brain injury in the United States: comparison of two national databases

OBJECTIVE

Traumatic brain injury (TBI) is a leading cause of morbidity and mortality in the US, but the true incidence of TBI is unknown.

METHODS

The National Trauma Data Bank National Sample Program (NTDB NSP) was queried for 2007 and 2013, and population-based weighted estimates of TBI-related emergency department (ED) visits, hospitalizations, and deaths were calculated. These data were compared to the 2017 Centers for Disease Control and Prevention (CDC) report on TBI, which used the Healthcare Cost and Utilization Project's National ("Nationwide" before 2012) Inpatient Sample and National Emergency Department Sample.

RESULTS

In the NTDB NSP the incidence of TBI-related ED visits was 59/100,000 in 2007 and 62/100,000 in 2013. However, in the CDC report there were 534/100,000 in 2007 and 787/100,000 in 2013. The CDC estimate for ED visits was 805% higher in 2007 and 1169% higher in 2013. In the NTDB NSP, the incidence of TBI-related deaths was 5/100,000 in 2007 and 4/100,000 in 2013. In the CDC report, the incidence was 18/100,000 in both years. The CDC estimate for deaths was 260% higher in 2007 and 325% higher in 2013.

CONCLUSIONS

The databases disagreed widely in their weighted estimates of TBI incidence: CDC estimates were consistently higher than NTDB NSP estimates, by an average of 448%. Although such a discrepancy may be intuitive, this is the first study to quantify the magnitude of disagreement between these databases. Given that research, funding, and policy decisions are made based on these estimates, there is a need for a more accurate estimate of the true national incidence of TBI.

Survival prediction of glioblastoma patients-are we there yet? A systematic review of prognostic modeling for glioblastoma and its clinical potential

Glioblastoma is associated with a poor prognosis. Even though survival statistics are well-described at the population level, it remains challenging to predict the prognosis of an individual patient despite the increasing number of prognostic models. The aim of this study is to systematically review the literature on prognostic modeling in glioblastoma patients. A systematic literature search was performed to identify all relevant studies that developed a prognostic model for predicting overall survival in glioblastoma patients following the PRISMA guidelines. Participants, type of input, algorithm type, validation, and testing procedures were reviewed per prognostic model. Among 595 citations, 27 studies were included for qualitative review. The included studies developed and evaluated a total of 59 models, of which only seven were externally validated in a different patient cohort. The predictive performance among these studies varied widely according to the AUC (0.58-0.98), accuracy (0.69-0.98), and C-index (0.66-0.70). Three studies deployed their model as an online prediction tool, all of which were based on a statistical algorithm. The increasing performance of survival prediction models will aid personalized clinical decision-making in glioblastoma patients. The scientific realm is gravitating towards the use of machine learning models developed on high-dimensional data, often with promising results. However, none of these models has been implemented into clinical care. To facilitate the clinical implementation of high-performing survival prediction models, future efforts should focus on harmonizing data acquisition methods, improving model interpretability, and externally validating these models in multicentered, prospective fashion.

Decreased Incidence of CSF Leaks after Skull Base Fractures in the 21st Century: An Institutional Report

Objectives Cerebrospinal fluid (CSF) leaks are a possible complication in patients with skull base fractures (SBFs). The widely cited incidence of CSF leaks is 10 to 30% in SBF patients; however, this estimate is based only on a few outdated studies. A recent report found CSF leaks in <2% SBF patients, suggesting the incidence may be lower now. To investigate this, we report here our institutional series.

Design This study is a retrospective chart review.

Setting The study was conducted at two major academic medical centers (2000–2018).

Participants Adult patients with SBF were included in this study.

Main Outcome Measures Variables included age, gender, CSF leak within 90 days, management regimen, meningitis within 90 days, and 1-year mortality.

Results Among 4,944 patients with SBF, 199 (4%) developed a CSF leak. SBF incidence was positively correlated with year of clinical presentation (r-squared 0.78, p < 0.001). Among CSF leaks, 42% were conservatively managed, 52% were treated with lumbar drain, and 7% required surgical repair. Meningitis developed in 28% CSF leak patients. The 1-year mortality for all SBF patients was 11%, for patients with CSF leaks was 12%, and for patients with meningitis was 16%.

Conclusion In the largest institutional review of SBF patients in the 21st century, we found CSF leak incidence to be 4%. This is lower than the widely cited range of 10 to 30%. Nevertheless, morbidity and mortality associated with this complication remains clinically significant, and SBF patients should continue to be monitored for CSF leaks. We provide here our institutional treatment algorithm for these patients that may help to inform the treatment strategy at other institutions.

Natural Language Processing for Automated Quantification of Brain Metastases Reported in Free-Text Radiology Reports

PURPOSE

Although the bulk of patient-generated health data are increasing exponentially, their use is impeded because most data come in unstructured format, namely as free-text clinical reports. A variety of natural language processing (NLP) methods have emerged to automate the processing of free text ranging from statistical to deep learning-based models; however, the optimal approach for medical text analysis remains to be determined. The aim of this study was to provide a head-to-head comparison of novel NLP techniques and inform future studies about their utility for automated medical text analysis.

PATIENTS AND METHODS

Magnetic resonance imaging reports of patients with brain metastases treated in two tertiary centers were retrieved and manually annotated using a binary classification (single metastasis v two or more metastases). Multiple bag-of-words and sequence-based NLP models were developed and compared after randomly splitting the annotated reports into training and test sets in an 80:20 ratio.

RESULTS

A total of 1,479 radiology reports of patients diagnosed with brain metastases were retrieved. The least absolute shrinkage and selection operator (LASSO) regression model demonstrated the best overall performance on the hold-out test set with an area under the receiver operating characteristic curve of 0.92 (95% CI, 0.89 to 0.94), accuracy of 83% (95% CI, 80% to 87%), calibration intercept of -0.06 (95% CI, -0.14 to 0.01), and calibration slope of 1.06 (95% CI, 0.95 to 1.17).

CONCLUSION

Among various NLP techniques, the bag-of-words approach combined with a LASSO regression model demonstrated the best overall performance in extracting binary outcomes from free-text clinical reports. This study provides a framework for the development of machine learning-based NLP models as well as a clinical vignette of patients diagnosed with brain metastases.

Task-Shifting and Task-Sharing in Neurosurgery: An International Survey of Current Practices in Low- and Middle-Income Countries

BACKGROUND

Because nearly 23,000 more neurosurgeons are needed globally to address 5 million essential neurosurgical cases that go untreated each year, there is an increasing interest in task-shifting and task-sharing (TS/S), delegating neurosurgical tasks to nonspecialists, particularly in low- and middle-income countries (LMICs). This global survey aimed to provide a cross-sectional understanding of the prevalence and structure of current neurosurgical TS/S practices in LMICs.

METHODS

The survey was distributed to a convenience sample of individuals providing neurosurgical care in LMICs with a Web-based survey link via electronic mailing lists of continental societies and various neurosurgical groups, conference announcements, e-mailing lists, and social media platforms. Country-level data were analyzed by descriptive statistics.

RESULTS

The survey yielded 127 responses from 47 LMICs; 20 countries (42.6%) reported ongoing TS/S. Most TS/S procedures involved emergency interventions, the top 3 being burr holes, craniotomy for hematoma evacuation, and external ventricular drain. Most (65.0%) believed that their Ministry of Health does not endorse TS/S (24.0% unsure), and only 11% believed that TS/S training was structured. There were few opportunities for TS/S providers to continue medical education (11.6%) or maintenance of certification (9.4%, or receive remuneration (4.2%).

CONCLUSIONS

TS/S is ongoing in many LMICs without substantial structure or oversight, which is concerning for patient safety. These data invite future clinical outcomes studies to assess effectiveness and discussions on policy recommendations such as standardized curricula, certification protocols, specialist oversight, and referral networks to increase the level of TS/S care and to continue to increase the specialist workforce.

Global Perspectives on Task Shifting and Task Sharing in Neurosurgery

BACKGROUND

Neurosurgical task shifting and task sharing (TS/S), delegating clinical care to non-neurosurgeons, is ongoing in many hospital systems in which neurosurgeons are scarce. Although TS/S can increase access to treatment, it remains highly controversial. This survey investigated perceptions of neurosurgical TS/S to elucidate whether it is a permissible temporary solution to the global workforce deficit.

METHODS

The survey was distributed to a convenience sample of individuals providing neurosurgical care. A digital survey link was distributed through electronic mailing lists of continental neurosurgical societies and various collectives, conference announcements, and social media platforms (July 2018-January 2019). Data were analyzed by descriptive statistics and univariate regression of Likert Scale scores.

RESULTS

Survey respondents represented 105 of 194 World Health Organization member countries (54.1%; 391 respondents, 162 from high-income countries and 229 from low- and middle-income countries [LMICs]). The most agreed on statement was that task sharing is preferred to task shifting. There was broad consensus that both task shifting and task sharing should require competency-based evaluation, standardized training endorsed by governing organizations, and maintenance of certification. When perspectives were stratified by income class, LMICs were significantly more likely to agree that task shifting is professionally disruptive to traditional training, task sharing should be a priority where human resources are scarce, and to call for additional TS/S regulation, such as certification and formal consultation with a neurosurgeon (in person or electronic/telemedicine).

CONCLUSIONS

Both LMIC and high-income countries agreed that task sharing should be prioritized over task shifting and that additional recommendations and regulations could enhance care. These data invite future discussions on policy and training programs.

Fibrinolytics and Intraventricular Hemorrhage: A Systematic Review and Meta-analysis

Intraventricular hemorrhage (IVH) is an independent poor prognostic factor in subarachnoid and intra-parenchymal hemorrhage. The use of intraventricular fibrinolytics (IVF) has long been debated, and its exact effects on outcomes are unknown. A systematic review and meta-analysis were performed in accordance with the PRISMA guidelines to assess the impact of IVF after non-traumatic IVH on mortality, functional outcome, intracranial bleeding, ventriculitis, time until clearance of third and fourth ventricles, obstruction of external ventricular drains (EVD), and shunt dependency. Nineteen studies were included in the meta-analysis, totaling 1020 patients. IVF was associated with lower mortality (relative risk [RR] 0.58; 95% confidence interval [CI] 0.47-0.72), fewer EVD obstructions (RR 0.41; 95% CI 0.22-0.74), and a shorter time until clearance of the ventricles (median difference [MD] - 4.05 days; 95% CI - 5.52 to - 2.57). There was no difference in good functional outcome, RR 1.41 (95% CI 0.98-2.03), or shunt dependency, RR 0.93 (95% CI 0.70-1.22). Correction for publication bias predicted an increased risk of intracranial bleeding, RR 1.67 (95% CI 1.01-2.74) and a lower risk of ventriculitis, RR 0.68 (95% CI 0.45-1.03) in IVH patients treated with IVF. IVF was associated with improved survival, faster clearance of blood from the ventricles and fewer drain obstructions, but further research is warranted to elucidate the effects on ventriculitis, long-term functional outcomes, and re-hemorrhage.

An Online Calculator for the Prediction of Survival in Glioblastoma Patients Using Classical Statistics and Machine Learning

BACKGROUND

Although survival statistics in patients with glioblastoma multiforme (GBM) are well-defined at the group level, predicting individual patient survival remains challenging because of significant variation within strata.

OBJECTIVE

To compare statistical and machine learning algorithms in their ability to predict survival in GBM patients and deploy the best performing model as an online survival calculator.

METHODS

Patients undergoing an operation for a histopathologically confirmed GBM were extracted from the Surveillance Epidemiology and End Results (SEER) database (2005-2015) and split into a training and hold-out test set in an 80/20 ratio. Fifteen statistical and machine learning algorithms were trained based on 13 demographic, socioeconomic, clinical, and radiographic features to predict overall survival, 1-yr survival status, and compute personalized survival curves.

RESULTS

In total, 20 821 patients met our inclusion criteria. The accelerated failure time model demonstrated superior performance in terms of discrimination (concordance index = 0.70), calibration, interpretability, predictive applicability, and computational efficiency compared to Cox proportional hazards regression and other machine learning algorithms. This model was deployed through a free, publicly available software interface (https://cnoc-bwh.shinyapps.io/gbmsurvivalpredictor/).

CONCLUSION

The development and deployment of survival prediction tools require a multimodal assessment rather than a single metric comparison. This study provides a framework for the development of prediction tools in cancer patients, as well as an online survival calculator for patients with GBM. Future efforts should improve the interpretability, predictive applicability, and computational efficiency of existing machine learning algorithms, increase the granularity of population-based registries, and externally validate the proposed prediction tool.

Oversight and Ethical Regulation of Conflicts of Interest in Neurosurgery in the United States

BACKGROUND

Developmental incentives are fundamental to surgical progress, yet financial and professional incentives inherently create conflicts of interest (COI). Understanding how to manage COI held by neurosurgeons, industry, hospitals, and journal editors, without thwarting progress and innovation is critical.

OBJECTIVE

To present an overview of COI associated with innovation in neurosurgery, and review ways to manage these in an ethically sound manner.

METHODS

A review of the literature was performed to assess conflicts of interest that affect neurosurgical innovation, and review ways to manage COI of various parties while adhering to ethical standards.

RESULTS

COI are inherent to collaboration and innovation, and are therefore an unavoidable component of neurosurgery. The lack of a clear distinction between clinical practice and innovation, ability to use devices off-label, and unstandardized disclosure requirements create inconsistencies in the way that conflicts of interest are handled. Additionally, lack of requirements to compare innovation to the standard of care and inherent bias that affects study design and interpretation can have profound effects on the medical literature. Conflicts of interest can have both direct and downstream effects on neurosurgical practice, and it is possible to manage them while improving the quality of research and innovation.

CONCLUSION

Conflicts of interest are inherent to surgical innovation, and can be handled in an ethically sound manner. Neurosurgeons, device companies, hospitals, and medical journals can take steps to proactively confront bias and ensure patient autonomy and safety. These steps can preserve public trust and ultimately improve evidence-based neurosurgical practice.

Long-term outcomes among octogenarians with aneurysmal subarachnoid hemorrhage

OBJECTIVE

Data evaluating the long-term outcomes, particularly with regard to treatment modality, of aneurysmal subarachnoid hemorrhage (SAH) in octogenarians are limited. The primary objectives were to evaluate the disposition (living at home vs institutional settings) and analyze the predictors of long-term survival and return to home for octogenarians after SAH.

METHODS

Data pertaining to patients age 80 and older who underwent microsurgical clipping or endovascular coiling for SAH were extracted from 100% nationwide Medicare inpatient claims and linked with the Minimum Data Set (2008-2011). Patient disposition was tracked for 2 years after index SAH admission. Multivariable logistic regression stratified by aneurysm treatment modality, and adjusted for patient factors including SAH severity, evaluated predictors of return to home at 60 and 365 days after SAH. Survival 365 days after SAH was analyzed with a multivariable Cox proportional hazards model.

RESULTS

A total of 1298 cases were included in the analysis. One year following SAH, 56% of the patients had died or were in hospice care, 8% were in an institutional post-acute care setting, and 36% had returned home. Open microsurgical clipping (adjusted hazard ratio [aHR] 0.67, 95% confidence interval [CI] 0.54-0.81), male sex (aHR 0.70, 95% CI 0.57-0.87), tracheostomy (aHR 0.63, 95% CI 0.47-0.85), gastrostomy (aHR 0.60, 95% CI 0.48-0.76), and worse SAH severity (aHR 0.94, 95% CI 0.92-0.97) were associated with reduced likelihood of patients ever returning home. Older age (aHR 1.09, 95% CI 1.05-1.13), tracheostomy (aHR 2.06, 95% CI 1.46-2.91), gastrostomy (aHR 1.55, 95% CI 1.14-2.10), male sex (aHR 1.66, 95% CI 1.20-2.23), and worse SAH severity 1.51 (95% CI 1.04-2.18) were associated with reduced survival.

CONCLUSIONS

In this national analysis, 56% of octogenarians with SAH died, and 36% returned home within 1 year of SAH. Coil embolization predicted returning to home, which may suggest a benefit to endovascular treatment in this patient population.

Palliative Care and Communication Training in Neurosurgery Residency: Results of a Trainee Survey

OBJECTIVE

Neurosurgeons care for critically ill patients near the end of life, yet little is known about how well their training prepares them for this role. We surveyed a random sample of neurosurgery residents to describe the quantity and quality of teaching activities related to serious illness communication and palliative care, and resident attitudes and perceived preparedness to care for seriously ill patients.

METHODS

A previously validated survey instrument was adapted to reflect required communication and palliative care competencies in the 2015 the Accreditation Council for Graduate Medical Education (ACGME) Milestones for Neurological Surgery. The survey was reviewed for content validity by independent faculty neurosurgeons, piloted with graduating neurosurgical residents, and distributed online in August 2016 to neurosurgery residents in the United States using the American Association of Neurological Surgeons (AANS)/Congress of Neurological Surgeons (CNS) Joint Section on Neurotrauma and Critical Care email listserv. Multiple choice and Likert scale responses were analyzed using descriptive statistics.

RESULTS

Sixty-two responses were recorded between August 2016 and October 2016. Most respondents reported no explicit teaching on: explaining risks and benefits of intubation and ventilation (69%), formulating prognoses in neurocritical care (60%), or leading family meetings (69%). Compared to performing craniotomies, respondents had less frequent practice leading discussions about withdrawing life-sustaining treatment (61% vs. 90%, p < 0.01, "weekly or more frequently"), and were less often observed (18% vs. 87%, p < 0.01) and given feedback on their performance (11% vs. 58%, p < 0.01). Nearly all respondents (95%) felt "prepared to discuss withdrawing life-sustaining treatments," however half (48%) reported they "would benefit from more communication training during residency." Most (87%) reported moral distress, agreeing that they "participated in operations and worried whether surgery aligned with patient goals."

CONCLUSIONS

Residents in our sample reported limited formal training, and relatively less observation and feedback, on required ACGME competencies in palliative care and communication. Most reported preparedness in this domain, but many were receptive to more training. Better quality and more consistent palliative care education in neurosurgery residency could improve competency and help ensure that neurosurgical care aligns with patient goals.

Development of machine learning algorithms for prediction of prolonged opioid prescription after surgery for lumbar disc herniation

BACKGROUND CONTEXT

Spine surgery has been identified as a risk factor for prolonged postoperative opioid use. Preoperative prediction of opioid use could improve risk stratification, shared decision-making, and patient counseling before surgery.

PURPOSE

The primary purpose of this study was to develop algorithms for prediction of prolonged opioid prescription after surgery for lumbar disc herniation.

STUDY DESIGN/SETTING

Retrospective, case-control study at five medical centers.

PATIENT SAMPLE

Chart review was conducted for patients undergoing surgery for lumbar disc herniation between January 1, 2000 and March 1, 2018.

OUTCOME MEASURES

The primary outcome of interest was sustained opioid prescription after surgery to at least 90 to 180 days postoperatively.

METHODS

Five models (elastic-net penalized logistic regression, random forest, stochastic gradient boosting, neural network, and support vector machine) were developed to predict prolonged opioid prescription. Explanations of predictions were provided globally (averaged across all patients) and locally (for individual patients).

RESULTS

Overall, 5,413 patients were identified, with sustained postoperative opioid prescription of 416 (7.7%) at 90 to 180 days after surgery. The elastic-net penalized logistic regression model had the best discrimination (c-statistic 0.81) and good calibration and overall performance; the three most important predictors were: instrumentation, duration of preoperative opioid prescription, and comorbidity of depression. The final models were incorporated into an open access web application able to provide predictions as well as patient-specific explanations of the results generated by the algorithms. The application can be found here: https://sorg-apps.shinyapps.io/lumbardiscopioid/ CONCLUSION: Preoperative prediction of prolonged postoperative opioid prescription can help identify candidates for increased surveillance after surgery. Patient-centered explanations of predictions can enhance both shared decision-making and quality of care.

Predictive Score of Adverse Events After Carotid Endarterectomy: The NSQIP Registry Carotid Endarterectomy Scale

Background

The goal of this study was to create a comprehensive, integer‐weighted predictive scale of adverse events after carotid endarterectomy (CEA), which may augment risk stratification and patient counseling.

Methods and Results

The targeted carotid files from the prospective NSQIP (National Surgical Quality Improvement Program) registry (2011–2013) comprised the derivation population. Multivariable logistic regression evaluated predictors of a 30‐day adverse event (stroke, myocardial infarction, or death), the effect estimates of which were used to build a weighted predictive scale that was validated using the 2014 to 2015 NSQIP registry release. A total of 10 766 and 8002 patients were included in the derivation and the validation populations, in whom 4.0% and 3.7% developed an adverse event, respectively. The NSQIP registry CEA scale included 14 variables; the highest points were allocated for insulin‐dependent diabetes mellitus, high‐risk cardiac physiological characteristics, admission source other than home, an emergent operation, American Society of Anesthesiologists’ classification IV to V, modified Rankin Scale score ≥2, and presentation with a stroke. NSQIP registry CEA score was predictive of an adverse event (concordance=0.67), stroke or death (concordance=0.69), mortality (concordance=0.76), an extended hospitalization (concordance=0.73), and a nonroutine discharge (concordance=0.83) in the validation population, as well as among symptomatic and asymptomatic subgroups (P<0.001). In the validation population, patients with an NSQIP registry CEA scale score >8 and 17 had 30‐day stroke or death rates >3% and 6%, the recommended thresholds for asymptomatic and symptomatic patients, respectively.

Conclusions

The NSQIP registry CEA scale predicts adverse outcomes after CEA and can risk stratify patients with both symptomatic and asymptomatic carotid stenosis using different thresholds for each population.

Natural and Artificial Intelligence in Neurosurgery: A Systematic Review

BACKGROUND

Machine learning (ML) is a domain of artificial intelligence that allows computer algorithms to learn from experience without being explicitly programmed.

OBJECTIVE

To summarize neurosurgical applications of ML where it has been compared to clinical expertise, here referred to as "natural intelligence."

METHODS

A systematic search was performed in the PubMed and Embase databases as of August 2016 to review all studies comparing the performance of various ML approaches with that of clinical experts in neurosurgical literature.

RESULTS

Twenty-three studies were identified that used ML algorithms for diagnosis, presurgical planning, or outcome prediction in neurosurgical patients. Compared to clinical experts, ML models demonstrated a median absolute improvement in accuracy and area under the receiver operating curve of 13% (interquartile range 4-21%) and 0.14 (interquartile range 0.07-0.21), respectively. In 29 (58%) of the 50 outcome measures for which a P-value was provided or calculated, ML models outperformed clinical experts (P < .05). In 18 of 50 (36%), no difference was seen between ML and expert performance (P > .05), while in 3 of 50 (6%) clinical experts outperformed ML models (P < .05). All 4 studies that compared clinicians assisted by ML models vs clinicians alone demonstrated a better performance in the first group.

CONCLUSION

We conclude that ML models have the potential to augment the decision-making capacity of clinicians in neurosurgical applications; however, significant hurdles remain associated with creating, validating, and deploying ML models in the clinical setting. Shifting from the preconceptions of a human-vs-machine to a human-and-machine paradigm could be essential to overcome these hurdles.

Task-Shifting and Task-Sharing in Neurosurgery: An International Survey of Current Practices in Low- and Middle-Income Countries

BACKGROUND

Because nearly 23,000 more neurosurgeons are needed globally to address 5 million essential neurosurgical cases that go untreated each year, there is an increasing interest in task-shifting and task-sharing (TS/S), delegating neurosurgical tasks to nonspecialists, particularly in low- and middle-income countries (LMICs). This global survey aimed to provide a cross-sectional understanding of the prevalence and structure of current neurosurgical TS/S practices in LMICs.

METHODS

The survey was distributed to a convenience sample of individuals providing neurosurgical care in LMICs with a Web-based survey link via electronic mailing lists of continental societies and various neurosurgical groups, conference announcements, e-mailing lists, and social media platforms. Country-level data were analyzed by descriptive statistics.

RESULTS

The survey yielded 127 responses from 47 LMICs; 20 countries (42.6%) reported ongoing TS/S. Most TS/S procedures involved emergency interventions, the top 3 being burr holes, craniotomy for hematoma evacuation, and external ventricular drain. Most (65.0%) believed that their Ministry of Health does not endorse TS/S (24.0% unsure), and only 11% believed that TS/S training was structured. There were few opportunities for TS/S providers to continue medical education (11.6%) or maintenance of certification (9.4%, or receive remuneration (4.2%).

CONCLUSIONS

TS/S is ongoing in many LMICs without substantial structure or oversight, which is concerning for patient safety. These data invite future clinical outcomes studies to assess effectiveness and discussions on policy recommendations such as standardized curricula, certification protocols, specialist oversight, and referral networks to increase the level of TS/S care and to continue to increase the specialist workforce.

Task-Sharing for Emergency Neurosurgery: A Retrospective Cohort Study in the Philippines

Objective

The safety and effectiveness of task-sharing (TS) in neurosurgery, delegating clinical roles to non-neurosurgeons, is not well understood. This study evaluated an ongoing TS model in the Philippines, where neurosurgical workforce deficits are compounded with a large neurotrauma burden.

Methods

Medical records from emergency neurosurgical admissions to 2 hospitals were reviewed (January 2015-June 2018): Bicol Medical Center (BMC), a government hospital in which emergency neurosurgery is chiefly performed by general surgery residents (TS providers), and Mother Seton Hospital, an adjacent private hospital where neurosurgery consultants are the primary surgeons. Univariable and multivariable linear and logistic regression compared provider-associated outcomes.

Results

Of 214 emergency neurosurgery operations, TS providers performed 95 and neurosurgeons, 119. TS patients were more often male (88.4% vs. 73.1%; P = 0.007), younger (mean age, 27.6 vs. 50.5 years; P < 0.001), and had experienced road traffic accidents (69.1% vs. 31.4%; P < 0.001). There were no significant differences between Glasgow Coma Scale (GCS) scores on admission. Provider type was not associated with mortality (neurosurgeons, 20.2%; TS, 17.9%; P = 0.68), reoperation, or pneumonia. No significant differences were observed for GCS improvement between admission and discharge or in-hospital GCS improvement, including or excluding inpatient deaths. TS patients had shorter lengths of stay (17.3 days vs. 24.4 days; coefficient, -6.67; 95% confidence interval, -13.01 to -0.34; P < 0.05) and were more likely to undergo tracheostomy (odds ratio, 3.1; 95% confidence interval, 1.30-7.40; P = 0.01).

Conclusions

This study, one of the first to examine outcomes of neurosurgical TS, shows that a strategic TS model for emergency neurosurgery produces comparable outcomes to the local neurosurgeons.

Traumatic Brain Injury as a Risk Factor for Dementia: An Eighteen-Year Two Institution Clinical Experience

INTRODUCTION

The Centers for Disease Control and Prevention (CDC) reports that there were 2.87 million cases of traumatic brain injury (TBI) in the United States in 2014. Some studies suggest a connection between TBI and increased risk of dementia, but it remains unclear whether the risk increases with age and TBI severity. Given our aging population, it is essential to better characterize the link between TBI and dementia.

METHODS

We conducted a retrospective cohort study of 2 major academic medical centers for years 2000 to 2015. We identified all patients with TBI, aged 45 and older. Variables included age, TBI severity, pre-existing dementia, dementia diagnosed after TBI, years to dementia, and follow-up time. TBI severity was determined by head/neck AIS score, using ICD-PIC software. Mild TBI was defined as AIS 0 to 2, and moderate/severe as AIS 3 to 6. Analysis was done in R.v.3.0.1 software.

RESULTS

Overall, there were 14 199 patients with TBI, of which 9938 (70%) were mild and 4261 (30%) were moderate/severe. Mean age was 70.5 (± 14.0). There were 1422 cases (10%) of pre-existing dementia, and 850 (6%) cases of dementia diagnosed after TBI. The mean follow-up time was 1129 (± 1,474) d. The 75 to 84 age group had the highest incidence of TBI (28%). When compared by age group and TBI severity, the proportion of moderate/severe TBI increased with increasing age. The proportion of pre-existing dementia increased with age, as expected. Notably, there is increased incidence of dementia after TBI in patients aged 65 and older (7%-10%, P < .001). There was no observed effect of TBI severity on the risk of dementia after TBI.

CONCLUSION

Our results indicate that TBI is a risk factor for the development of dementia, especially in patients aged 65 and older. This points to the need for public health measures to mitigate the risk of TBI in this patient population.

Predicting nonroutine discharge after elective spine surgery: external validation of machine learning algorithms

OBJECTIVE

Nonroutine discharge after elective spine surgery increases healthcare costs, negatively impacts patient satisfaction, and exposes patients to additional hospital-acquired complications. Therefore, prediction of nonroutine discharge in this population may improve clinical management. The authors previously developed a machine learning algorithm from national data that predicts risk of nonhome discharge for patients undergoing surgery for lumbar disc disorders. In this paper the authors externally validate their algorithm in an independent institutional population of neurosurgical spine patients.

METHODS

Medical records from elective inpatient surgery for lumbar disc herniation or degeneration in the Transitional Care Program at Brigham and Women's Hospital (2013-2015) were retrospectively reviewed. Variables included age, sex, BMI, American Society of Anesthesiologists (ASA) class, preoperative functional status, number of fusion levels, comorbidities, preoperative laboratory values, and discharge disposition. Nonroutine discharge was defined as postoperative discharge to any setting other than home. The discrimination (c-statistic), calibration, and positive and negative predictive values (PPVs and NPVs) of the algorithm were assessed in the institutional sample.

RESULTS

Overall, 144 patients underwent elective inpatient surgery for lumbar disc disorders with a nonroutine discharge rate of 6.9% (n = 10). The median patient age was 50 years and 45.1% of patients were female. Most patients were ASA class II (66.0%), had 1 or 2 levels fused (80.6%), and had no diabetes (91.7%). The median hematocrit level was 41.2%. The neural network algorithm generalized well to the institutional data, with a c-statistic (area under the receiver operating characteristic curve) of 0.89, calibration slope of 1.09, and calibration intercept of -0.08. At a threshold of 0.25, the PPV was 0.50 and the NPV was 0.97.

CONCLUSIONS

This institutional external validation of a previously developed machine learning algorithm suggests a reliable method for identifying patients with lumbar disc disorder at risk for nonroutine discharge. Performance in the institutional cohort was comparable to performance in the derivation cohort and represents an improved predictive value over clinician intuition. This finding substantiates initial use of this algorithm in clinical practice. This tool may be used by multidisciplinary teams of case managers and spine surgeons to strategically invest additional time and resources into postoperative plans for this population.

Age-Dependent Radiographic Vasospasm and Delayed Cerebral Ischemia in Women After Aneurysmal Subarachnoid Hemorrhage

OBJECTIVE

Recent literature suggests there are sex differences in delayed cerebral ischemia (DCI) after aneurysmal subarachnoid hemorrhage (aSAH). Our study serves to compare sex differences in radiographic vasospasm, DCI, and clinical outcome after aSAH, and to determine whether there are age-dependent differences.

METHODS

A total of 328 patients with ruptured cerebral aneurysms were evaluated for radiographic vasospasm, clinical deterioration, cerebral infarction, and modified Rankin Scale-determined clinical outcome at 6 months to 1 year after rupture. Multivariate regression analyses were performed to evaluate the associations between these outcome measures and sex, adjusting for age, hypertension, aneurysm location, admission Hunt and Hess grade, and modified Fisher grade.

RESULTS

After multivariate adjustment, women had higher rates of radiographic vasospasm (β = 0.35; 95% confidence interval [CI], 0.068-0.63; P = 0.015), clinical deterioration (odds ratio [OR], 2.8; 95% CI, 1.3-6.0; P = 0.008) and cerebral infarction (OR, 2.4; 95% CI, 1.0-5.5; P = 0.039), but no difference was observed in follow-up modified Rankin Scale (mRS) outcome score at 6 months to 1 year (P = 0.96). Older women (age >55 years) have a higher rate of clinical deterioration than men in the same age group (OR, 3.5; 95% CI, 1.0-12; P = 0.043). In contrast, younger women (age ≤55 years) had increased radiographic vasospasm (β = 0.55; 95% CI, 0.17-0.93; P = 0.005) and worse mRS outcome score (β = 0.042; 95% CI, -0.021 to 1.1; P = 0.042) compared with men.

CONCLUSIONS

Female sex is associated with a higher risk of radiographic vasospasm, clinical deterioration, and cerebral infarction. Furthermore, this association appears to be age-dependent. This study further supports the unique role of sex, and highlights the need to better understand the possible role of female hormones in the development of complications of subarachnoid hemorrhage.

Adverse Events After Microvascular Decompression: A National Surgical Quality Improvement Program Analysis

BACKGROUND

Although microvascular decompression (MVD) is a durable treatment for medically refractory trigeminal neuralgia, hemifacial spasm, or glossopharyngeal neuralgia attributable to neurovascular conflict, few national studies have analyzed predictors of postoperative complications.

OBJECTIVE

To determine the incidence and risk factors for adverse events after MVD.

METHODS

Patients who underwent MVD were extracted from the prospectively collected National Surgical Quality Improvement Program registry (2006-2017). Multivariable logistic regression identified predictors of 30-day adverse events and unplanned readmission; multivariable linear regression analyzed predictors of a longer hospital stay.

RESULTS

Among the 1005 patients evaluated, the mortality was 0.3%, major neurologic complication rate 0.4%, and 2.8% had a nonroutine hospital discharge. Patient age was not a predictor of any adverse events. Statistically significant independent predictors both of any adverse event (9.2%) and of a longer hospitalization were American Society of Anesthesiologists (ASA) classification III-IV designation and longer operative duration (P ≤ 0.03) The 30-day readmission rate was 6.8%, and the most common reasons were surgical site infections (22.4%) and cerebrospinal fluid leakage (14.3%). Higher ASA classification, diabetes mellitus, and operative time were predictors of readmission (P < 0.04).

CONCLUSIONS

In this National Surgical Quality Improvement Program analysis, postoperative morbidity and mortality after MVD was low. Patient age was not a predictor of postoperative complications, whereas higher ASA classification, diabetes mellitus, and longer operative duration were predictive of any adverse event and readmission. ASA classification provided superior risk stratification compared with the total number of patient comorbidities or laboratory values. These data can assist with preoperative patient counseling and risk stratification.

Predicting discharge placement after elective surgery for lumbar spinal stenosis using machine learning methods

PURPOSE

An excessive amount of total hospitalization is caused by delays due to patients waiting to be placed in a rehabilitation facility or skilled nursing facility (RF/SNF). An accurate preoperative prediction of who would need a RF/SNF place after surgery could reduce costs and allow more efficient organizational planning. We aimed to develop a machine learning algorithm that predicts non-home discharge after elective surgery for lumbar spinal stenosis.

METHODS

We used the American College of Surgeons National Surgical Quality Improvement Program to select patient that underwent elective surgery for lumbar spinal stenosis between 2009 and 2016. The primary outcome measure for the algorithm was non-home discharge. Four machine learning algorithms were developed to predict non-home discharge. Performance of the algorithms was measured with discrimination, calibration, and an overall performance score.

RESULTS

We included 28,600 patients with a median age of 67 (interquartile range 58-74). The non-home discharge rate was 18.2%. Our final model consisted of the following variables: age, sex, body mass index, diabetes, functional status, ASA class, level, fusion, preoperative hematocrit, and preoperative serum creatinine. The neural network was the best model based on discrimination (c-statistic = 0.751), calibration (slope = 0.933; intercept = 0.037), and overall performance (Brier score = 0.131).

CONCLUSIONS

A machine learning algorithm is able to predict discharge placement after surgery for lumbar spinal stenosis with both good discrimination and calibration. Implementing this type of algorithm in clinical practice could avert risks associated with delayed discharge and lower costs. These slides can be retrieved under Electronic Supplementary Material.

Adverse events after clipping of unruptured intracranial aneurysms: the NSQIP unruptured aneurysm scale

OBJECTIVE

The complex decision analysis of unruptured intracranial aneurysms entails weighing the benefits of aneurysm repair against operative risk. The goal of the present analysis was to build and validate a predictive scale that identifies patients with the greatest odds of a postsurgical adverse event.

METHODS

Data on patients who underwent surgical clipping of an unruptured aneurysm were extracted from the prospective National Surgical Quality Improvement Program registry (NSQIP; 2007-2014); NSQIP does not systematically collect data on patients undergoing intracranial endovascular intervention. Multivariable logistic regression evaluated predictors of any 30-day adverse event; variables screened included patient demographics, comorbidities, functional status, preoperative laboratory values, aneurysm location/complexity, and operative time. A predictive scale was constructed based on statistically significant independent predictors, which was validated using both NSQIP (2015-2016) and the Nationwide Inpatient Sample (NIS; 2002-2011).

RESULTS

The NSQIP unruptured aneurysm scale was proposed: 1 point was assigned for a bleeding disorder; 2 points for age 51-60 years, cardiac disease, diabetes mellitus, morbid obesity, anemia (hematocrit < 36%), operative time 240-330 minutes; 3 points for leukocytosis (white blood cell count > 12,000/μL) and operative time > 330 minutes; and 4 points for age > 60 years. An increased score was predictive of postoperative stroke or coma (NSQIP: p = 0.002, C-statistic = 0.70; NIS: p < 0.001, C-statistic = 0.61), a medical complication (NSQIP: p = 0.01, C-statistic = 0.71; NIS: p < 0.001, C-statistic = 0.64), and a nonroutine discharge (NSQIP: p < 0.001, C-statistic = 0.75; NIS: p < 0.001, C-statistic = 0.66) in both validation populations. Greater score was also predictive of increased odds of any adverse event, a major complication, and an extended hospitalization in both validation populations (p ≤ 0.03).

CONCLUSIONS

The NSQIP unruptured aneurysm scale may augment the risk stratification of patients undergoing microsurgical clipping of unruptured cerebral aneurysms.

Non-electrographic Seizures Due to Subdural Hematoma: A Case Series and Review of the Literature

Seizures due to subdural hematoma (SDH) are a common finding, typically diagnosed using electroencephalography (EEG). At times, aggressive management of seizures is necessary to improve neurologic recovery and outcomes. Here, we present three patients who had undergone emergent SDH evacuation and showed postoperative focal deficits without accompanying electrographic epileptiform activity. After infarction and recurrent hemorrhage were ruled out, seizures were suspected despite a negative EEG. Patients were treated aggressively with AEDs and eventually showed clinical improvement. Long-term monitoring with EEG revealed electrographic seizures in a delayed fashion. EEG recordings are an important tool for seizure detection, but should be used as an adjunct to, rather than a replacement for, the clinical examination in the acute setting. At times, aggressive treatment of suspected postoperative seizures is warranted despite lack of corresponding electrographic activity and can improve clinical outcomes.