Deployment of Clinical Prediction Models: A Practical Guide to Nomograms and Online Calculators.

The use of predictive models within neurosurgery is increasing and many models described in published journal articles are made available to readers in formats such as nomograms and online calculators. The present chapter details a step-by-step methodology with accompanying R code that may be used to implement models both in the form of traditional nomograms and as open-access, online calculators through RStudio's Shinyapps. The chapter assumes a basic understanding of predictive modeling in R and utilizes open-access files created by the Machine Intelligence in Clinical Neuroscience (MICN) Lab (Department of Neurosurgery and the Clinical Neuroscience Center of the University Hospital Zurich). When implemented correctly, tools such as nomograms and predictive calculators have the potential to improve user understanding of the underlying statistical models, facilitate broader adoption, and to streamline the eventual use of such models in clinical settings.

Overview of Algorithms for Natural Language Processing and Time Series Analyses.

A host of machine learning algorithms have been used to perform several different tasks in NLP and TSA. Prior to implementing these algorithms, some degree of data preprocessing is required. Deep learning approaches utilizing multilayer perceptrons, recurrent neural networks (RNNs), and convolutional neural networks (CNNs) represent commonly used techniques. In supervised learning applications, all these models map inputs into a predicted output and then model the discrepancy between predicted values and the real output according to a loss function. The parameters of the mapping function are then optimized through the process of gradient descent and backward propagation in order to minimize this loss. This is the main premise behind many supervised learning algorithms. As experience with these algorithms grows, increased applications in the fields of medicine and neuroscience are anticipated.

A Brief History of Machine Learning in Neurosurgery.

The history of machine learning in neurosurgery spans three decades and continues to develop at a rapid pace. The earliest applications of machine learning within neurosurgery were first published in the 1990s as researchers began developing artificial neural networks to analyze structured datasets and supervised tasks. By the turn of the millennium, machine learning had evolved beyond proof-of-concept; algorithms had success detecting tumors in unstructured clinical imaging, and unsupervised learning showed promise for tumor segmentation. Throughout the 2000s, the role of machine learning in neurosurgery was further refined. Well-trained models began to consistently best expert clinicians at brain tumor diagnosis. Additionally, the digitization of the healthcare industry provided ample data for analysis, both structured and unstructured. By the 2010s, the use of machine learning within neurosurgery had exploded. The rapid deployment of an exciting new toolset also led to the growing realization that it may offer marginal benefit at best over conventional logistical regression models for analyzing tabular datasets. Additionally, the widespread adoption of machine learning in neurosurgical clinical practice continues to lag until additional validation can ensure generalizability. Many exciting contemporary applications nonetheless continue to demonstrate the unprecedented potential of machine learning to revolutionize neurosurgery when applied to appropriate clinical challenges.

Plastic surgery wound closure following resection of spinal metastases.

OBJECTIVE: Surgical site infection and dehiscence are devastating complications of surgery for spinal metastases. Wound closure involving plastic surgeons has been proposed as a strategy to lower post-operative complications. Here we investigated whether plastic surgery closure is associated with lower rates of wound complications, wound infection, and wound reoperation compared to simple closure by spine surgeons. METHODS: Patients surgically treated for metastatic tumors at a single comprehensive cancer center between April 2013-2020 were retrospectively identified. Primary pathology, demographic information, clinical characteristics, pre-operative laboratory values, tumor location, operative characteristics, and post-operative outcomes were collected. Univariable analyses used student t-tests for continuous variables and χ2 tests for categorical variables. Multivariable regressions were performed to control for confounders. RESULTS: We included 317 patients, of which 56 underwent closure by plastic surgeons and 291 by neurosurgeons. Patients in the plastic surgery cohort were more likely to have received prior radiation to the surgical site, more often on long-term corticosteroid therapy, and more likely to have sacrococcygeal tumors. Operations involving plastic surgeons were more likely to be revision surgeries, corpectomies, and to involve a staged approach. Additionally, patients in the plastic surgery cohort had longer incision lengths, longer surgeries, greater intraoperative blood loss (IOBL), were more likely to receive transfusions, and had longer hospitalizations. Local paraspinous advancement flaps were the most common complex wound closure technique. Plastic surgery closure was not significantly associated with a difference in rates of post-operative wound complications, wound infection, or wound-related reoperations compared to simple wound closure. CONCLUSION: We identified that patients undergoing plastic surgery wound closure had worse baseline risk, longer surgeries, greater IOBL, and longer hospitalizations compared to patients receiving simple closure. Despite their increased risk, complex wound closure did not significantly alter the rates of post-operative wound complications, wound infection, or wound-related reoperations. Consideration may be given to plastic surgery closure in patients at high risk of wound complications or with extensive wound defects.

Interrater and Intrarater Reliability of the Vertebral Bone Quality Score.

BACKGROUND: Vertebral bone quality had a significant impact on postoperative outcomes in spinal fusion surgery. New magnetic resonance imaging-based measures, such as the Vertebral Bone Quality (VBQ) score, may allow for bone quality assessment without the radiation associated with conventional testing. In the present study, we sought to assess the intrarater and interrater reliability of VBQ scores calculated by medical professionals and trainees. METHODS: Thirteen reviewers of various specialties and levels of training were recruited and asked to calculate VBQ scores for 30 patients at 2 time points separated by 2 months. Scored volumes were acquired from patients treated for both degenerative and oncologic indications. Intrarater and interrater agreement, quantified by intraclass correlation coefficient (ICC), was assessed using 2-way random effects modeling. Square-weight Cohen κ and Kendall Tau-b were used to determine whether raters assigned similar scores during both evaluations. RESULTS: All raters showed moderate to excellent reliability for VBQ score (ICC 0.667-0.957; κ0.648-0.921) and excellent reliability for all constituent components used to calculate VBQ score (ICC all ≥0.97). Interrater reliability was also found to be good for VBQ score on both the first (ICC = 0.818) and second (ICC = 0.800) rounds of assessment; scores for the constituent component all had ICC values ≥0.97 for the constituent components. CONCLUSIONS: The VBQ score appears to have both good intrarater and interrater reliability. In addition, there appeared to be no correlation between score reliability and level of training. External validation and further investigations of its ability to accurately model bone biomechanical properties are necessary.

Risk Factors for Wound-Related Complications After Surgery for Primary and Metastatic Spine Tumors: A Systematic Review and Meta-Analysis.

OBJECTIVE: We systematically reviewed the literature to compare risk factors for postoperative complications at the surgical wound site in primary and metastatic tumor operations. METHODS: We screened English-language publications on the outcomes of primary and metastatic spinal tumor operations. Pooled analyses and meta-analyses with random-effects modeling were performed comparing patients with and without wound complications, which were defined as surgical site infection or sterile wound dehiscence. RESULTS: Our search identified 5471 unique citations, from which we included 23 studies describing 5104 patients. A total of 1936 patients underwent surgery for primary tumors, with a wound complication rate of 8.1%. Subgroup analysis of benign and malignant primary tumors yielded significantly different wound complication rates of 7.8% and 26.9%, respectively. The metastatic tumor cohort included 168 patients and a complication rate of 6.6%. In a pooled analysis of primary tumors, higher wound complication rates were associated with sacral operations and the use of instrumentation. In the metastatic tumor cohort, higher complication rates were associated with female sex, smoking history, preoperative chemotherapy, preoperative radiotherapy, corticosteroid use, and previous spine surgery. Instrumentation remained a statistically significant risk factor for primary tumors with the addition of random-effects meta-analysis. CONCLUSIONS: Risk factors for wound complications after primary tumor operations were related to tumor histology and the spinal location of the operation. Risk factors for metastatic tumors may be related to several systemic preoperative treatments and baseline comorbidities. Random-effects meta-analysis showed the limited generalizability of these findings because of the small heterogenous primary literature.