Development of a preoperative predictive model for major complications following adult spinal deformity surgery

A narrative review of machine learning as promising revolution in clinical practice of scoliosis

Machine learning (ML), as an advanced domain of artificial intelligence (AI), is progressively changing our view of the world. By implementing its algorithms, our ability to detect previously undiscoverable patterns in data has the potential to revolutionize predictive analytics. Scoliosis, as a relatively specialized branch in the spine field, mainly covers the pediatric, adult and the elderly populations, and its diagnosis and treatment remain difficult. With recent efforts and interdisciplinary cooperation, ML has been widely applied to investigate issues related to scoliosis, and surprisingly augment a surgeon's ability in clinical practice related to scoliosis. Meanwhile, ML models penetrate in every stage of the clinical practice procedure of scoliosis. In this review, we first present a brief description of the application of ML in the clinical practice procedures regarding scoliosis, including screening, diagnosis and classification, surgical decision making, intraoperative manipulation, complication prediction, prognosis prediction and rehabilitation. Meanwhile, the ML models and specific applications adopted are presented. Additionally, current limitations and future directions are briefly discussed regarding its use in the field of scoliosis. We believe that the implementation of ML is a promising revolution to assist surgeons in all aspects of clinical practice related to scoliosis in the near future.

Does the Global Alignment and Proportion score overestimate mechanical complications after adult spinal deformity correction?

OBJECTIVE

The goal of this study was to validate the Global Alignment and Proportion (GAP) score in a cohort of patients undergoing adult spinal deformity (ASD) surgery. The GAP score is a novel measure that uses sagittal parameters relative to each patient's lumbosacral anatomy to predict mechanical complications after ASD surgery. External validation is required.

METHODS

Adult ASD patients undergoing > 4 levels of posterior fusion with a minimum 2-year follow-up were included. Six-week postoperative standing radiographs were used to calculate the GAP score, classified into a spinopelvic state as proportioned (P), moderately disproportioned (MD), or severely disproportioned (SD). A chi-square analysis, receiver operating characteristic curve, and Cochran-Armitage analysis were performed to assess the relationship between the GAP score and mechanical complications.

RESULTS

Sixty-seven patients with a mean age of 52.5 years (range 18-75 years) and a mean follow-up of 2.04 years were included. Patients with < 2 years of follow-up were included only if they had an early mechanical complication. Twenty of 67 patients (29.8%) had a mechanical complication. The spinopelvic state breakdown was as follows: P group, 21/67 (31.3%); MD group, 23/67 (34.3%); and SD group, 23/67 (34.3%). Mechanical complication rates were not significantly different among all groups: P group, 19.0%; MD group, 30.3%; and SD group, 39.1% (χ2 = 1.70, p = 0.19). The rates of mechanical complications between the MD and SD groups (30.4% and 39.1%) were less than those observed in the original GAP study (MD group 36.4%-57.1% and SD group 72.7%-100%). Within the P group, the rates in this study were higher than in the original study (19.0% vs 4.0%, respectively).

CONCLUSIONS

The authors found no statistically significant difference in the rate of mechanical complications between the P, MD, and SD groups. The current validation study revealed poor generalizability toward the authors' patient population.

State-of-the-art reviews predictive modeling in adult spinal deformity: applications of advanced analytics

Adult spinal deformity (ASD) is a complex and heterogeneous disease that can severely impact patients' lives. While it is clear that surgical correction can achieve significant improvement of spinopelvic parameters and quality of life measures in adults with spinal deformity, there remains a high risk of complication associated with surgical approaches to adult deformity. Over the past decade, utilization of surgical correction for ASD has increased dramatically as deformity correction techniques have become more refined and widely adopted. Along with this increase in surgical utilization, there has been a massive undertaking by spine surgeons to develop more robust models to predict postoperative outcomes in an effort to mitigate the relatively high complication rates. A large part of this revolution within spine surgery has been the gradual adoption of predictive analytics harnessing artificial intelligence through the use of machine learning algorithms. The development of predictive models to accurately prognosticate patient outcomes following ASD surgery represents a dramatic improvement over prior statistical models which are better suited for finding associations between variables than for their predictive utility. Machine learning models, which offer the ability to make more accurate and reproducible predictions, provide surgeons with a wide array of practical applications from augmenting clinical decision making to more wide-spread public health implications. The inclusion of these advanced computational techniques in spine practices will be paramount for improving the care of patients, by empowering both patients and surgeons to more specifically tailor clinical decisions to address individual health profiles and needs.

Systematic review and evaluation of predictive modeling algorithms in spinal surgeries

In order to better educate patients, predictive models have been implemented to stratify surgical risk, thereby instituting greater uniformity across surgical practices and prioritizing the safety and outcomes of patients. The purpose of this study is to conduct a systematic review summarizing the major predictive models used to evaluate patients as candidates for spinal surgery. A search was conducted for articles related to predictive modeling in spinal surgeries using PubMed, MEDLINE, and Scopus databases. Papers with area under the receiver operating curve (AUROC) scores reported were included in the analysis. Models not relevant to spinal procedures were excluded. Comparison between models was only attainable for those that reported AUROCs for individual procedures. Based on a combination of AUROC scores and demonstrated applicability to spinal procedures, the models by Scheer et al. (0.89), Ratliff et al. (0.70), the Seattle Spine Score (0.712), Risk Assessment Tool (0.67-0.7), and the Spine Sage calculator (0.81-0.85) were determined to be ideal for predictive modeling in spinal surgeries and were subsequently broken down into their individual inputs and outputs to determine what elements a theoretical model should assimilate. Alongside the model by Scheer et al., the Spine Sage calculator, Seattle Spine Score, Risk Assessment Tool, and a model by Ratliff et al. showed the most promise for patients undergoing spinal procedures. Using the first model as a springboard, new spinal predictive models can be optimized through use of larger prospective databases, with longer follow-up times, and greater inclusion of reliable high impact variables.

Identification of Risk Factors for Readmission in Patients Undergoing Anterior Cervical Discectomy Fusion: A Predictive Risk Scale

STUDY DESIGN

This was a retrospective cohort study.

OBJECTIVE

The objective of this study was to analyze readmission rates among patients undergoing anterior cervical discectomy and fusion (ACDF), determine which factors were associated with higher readmission rates, and develop a scale for utilization during surgical planning.

SUMMARY OF BACKGROUND DATA

ACDF is the most common surgical treatment for many cervical disk pathologies. With the Centers for Medicare and Medicaid Services selecting readmissions as a measure of health care quality, there has been an increased focus on reducing readmissions.

MATERIALS AND METHODS

There were 114,174 recorded ACDF surgeries in the derivation cohort, the State Inpatient Database (SID) of New York and California between 2006 and 2014. There were 115,829 ACDF surgeries recorded in the validation cohort, the SID from Florida and Washington over the same time period. After identification of risk factors using univariate and multivariate analysis of the derivation cohort, a predictive scale was generated and tested utilizing the validation cohort.

RESULTS

Overall, readmission rates within 30 days of discharge were 5.87% and 5.52% in the derivation and validation cohorts, respectively. On multivariate analysis of the derivation cohort, age older than 80 years [odds ratio (OR), 1.67] male sex (OR, 1.16), Medicaid insurance (OR, 1.90), Medicare insurance (OR, 1.64), revision ACDF (OR, 1.43), anemia (OR, 1.45), chronic lung disease (OR, 1.23), coagulopathy (OR, 1.42), congestive heart failure (OR, 1.31), diabetes (OR, 1.23), fluid and electrolyte disorder (OR, 1.56), liver disease (OR, 1.37), renal failure (OR, 1.59), and myelopathy (OR, 1.19) were found to be statistically significant predictors for readmission. These factors were incorporated into a numeric scale that, that when tested on the validation cohort, could explain 97.1% of the variability in readmission rate.

CONCLUSIONS

Overall, 30-day readmission following ACDF surgery was 5%-6%. A novel risk scale based on factors associated with increased readmission rates may be helpful in identifying patients who require additional optimization to reduce perioperative morbidity.

LEVEL OF EVIDENCE

Level III-prognostic.

Artificial intelligence facilitates decision-making in the treatment of lumbar disc herniations

PURPOSE

Apart from patients with severe neurological deficits, it is not clear whether surgical or conservative treatment of lumbar disc herniations is superior for the individual patient. We investigated whether deep learning techniques can predict the outcome of patients with lumbar disc herniation after 6 months of treatment.

METHODS

The data of 60 patients were used to train and test a deep learning algorithm with the aim to achieve an accurate prediction of the ODI 6 months after surgery or the start of conservative therapy. We developed an algorithm that predicts the ODI of 6 randomly selected test patients in tenfold cross-validation.

RESULTS

A 100% accurate prediction of an ODI range could be achieved by dividing the ODI scale into 12% sections. A maximum absolute difference of only 3.4% between individually predicted and actual ODI after 6 months of a given therapy was achieved with our most powerful model. The application of artificial intelligence as shown in this work also allowed to compare the actual patient values after 6 months with the prediction for the alternative therapy, showing deviations up to 18.8%.

CONCLUSION

Deep learning in the supervised form applied here can identify patients at an early stage who would benefit from conservative therapy, and on the contrary avoid painful and unnecessary delays for patients who would profit from surgical therapy. In addition, this approach can be used in many other areas of medicine as an effective tool for decision-making when choosing between opposing treatment options, despite small patient groups.

The SIFK score: a validated predictive model for arthroplasty progression after subchondral insufficiency fractures of the knee

PURPOSE

The purpose of this study was to create a predictive model utilizing baseline demographic and radiographic characteristics for the likelihood that a patient with subchondral insufficiency fracture of the knee will progress to knee arthroplasty with emphasis on clinical interpretability and usability.

METHODS

A retrospective review of baseline and final radiographs in addition to MRIs were reviewed for evaluation of insufficiency fractures and associated injuries. Patient and radiographic factors were used in building predictive models for progression to arthroplasty with Train: Validation: Test subsets. Multiple models were compared with emphasis on clinical utility.

RESULTS

Total of 249 patients with a mean age of 64.6 (SD 10.5) years were included. Knee arthroplasty rate was 27% at mean of 4 years of follow-up. Lasso Regression was non-inferior to other models and was chosen for ease of interpretability. In order of importance, predictors for progression to arthroplasty included lateral meniscus extrusion, Kellgren-Lawrence Grade 4, SIFK on MFC, lateral meniscus root tear, and medial meniscus extrusion. The final SIFK Score stratified patients into low-, medium-, and high-risk categories with arthroplasty rates of 8.8%, 40.4%, and 78.9% (p < 0.001) and an area under the curve of 82.5%.

CONCLUSION

In this validated model, lateral meniscus extrusion, K-L Grade 4, SIFK on MFC, lateral meniscus root tear, and medial meniscus extrusion were the most important factors in predicting progression to arthroplasty (in that order). This model assists in patient treatment and counseling in providing prognostic information based on patient-specific risk factors by classifying them into a low-, medium-, and high-risk categories. This model can be used both by medical professionals treating musculoskeletal injuries in guiding patient decision making.

LEVEL OF EVIDENCE

Level III.

The effect of anterior lumbar interbody fusion staging order on perioperative complications in circumferential lumbar fusions performed within the same hospital admission

OBJECTIVE

Anterior lumbar interbody fusion (ALIF) is a powerful technique that provides wide access to the disc space and allows for large lordotic grafts. When used with posterior spinal fusion (PSF), the procedures are often staged within the same hospital admission. There are limited data on the perioperative risk profile of ALIF-first versus PSF-first circumferential fusions performed within the same hospital admission. In an effort to understand whether these procedures are associated with different perioperative complication profiles, the authors performed a retrospective review of their institutional experience in adult patients who had undergone circumferential lumbar fusions.

METHODS

The electronic medicals records of patients who had undergone ALIF and PSF on separate days within the same hospital admission at a single academic center were retrospectively analyzed. Patients carrying a diagnosis of tumor, infection, or traumatic fracture were excluded. Demographics, surgical characteristics, and perioperative complications were collected and assessed.

RESULTS

A total of 373 patients, 217 of them women (58.2%), met the inclusion criteria. The mean age of the study cohort was 60 years. Surgical indications were as follows: degenerative disease or spondylolisthesis, 171 (45.8%); adult deformity, 168 (45.0%); and pseudarthrosis, 34 (9.1%). The majority of patients underwent ALIF first (321 [86.1%]) with a mean time of 2.5 days between stages. The mean number of levels fused was 2.1 for ALIF and 6.8 for PSF. In a comparison of ALIF-first to PSF-first cases, there were no major differences in demographics or surgical characteristics. Rates of intraoperative complications including venous injury were not significantly different between the two groups. The rates of postoperative ileus (11.8% vs 5.8%, p = 0.194) and ALIF-related wound complications (9.0% vs 3.8%, p = 0.283) were slightly higher in the ALIF-first group, although the differences did not reach statistical significance. Rates of other perioperative complications were no different.

CONCLUSIONS

In patients undergoing staged circumferential fusion with ALIF and PSF, there was no statistically significant difference in the rate of perioperative complications when comparing ALIF-first to PSF-first surgeries.

Evaluation of Predictive Models for Complications following Spinal Surgery

BACKGROUND

 Complications rates vary across spinal surgery procedures and are difficult to predict due to heterogeneity in patient characteristics, surgical methods, and hospital volume. Incorporation of predictive models for complications may guide surgeon decision making and improve outcomes.

METHODS

 We evaluate current independently validated predictive models for complications in spinal surgery with respect to study design and model generation, accuracy, reliability, and utility. We conducted our search using Preferred Reporting Items for Systematic Review and Meta-analysis guidelines and the Participants, Intervention, Comparison, Outcomes, Study Design model through the PubMed and Ovid Medline databases.

RESULTS

 A total of 18 articles met inclusion criteria including 30 validated predictive models of complications after adult spinal surgery. National registry databases were used in 12 studies. Validation cohorts were used in seven studies for verification; three studies used other methods including random sample bootstrapping techniques or cross-validation. Reported area under the curve (AUC) values ranged from 0.37 to 1.0. Studies described treatment for deformity, degenerative conditions, inclusive spinal surgery (neoplasm, trauma, infection, deformity, degenerative), and miscellaneous (disk herniation, spinal epidural abscess). The most commonly cited risk factors for complications included in predictive models included age, body mass index, diabetes, sex, and smoking. Those models in the deformity subset that included radiographic and anatomical grading features reported higher AUC values than those that included patient demographics or medical comorbidities alone.

CONCLUSIONS

 We identified a cohort of 30 validated predictive models of complications following spinal surgery for degenerative conditions, deformity, infection, and trauma. Accurate evidence-based predictive models may enhance shared decision making, improve rehabilitation, reduce adverse events, and inform best practices.

Artificial Intelligence Models Predict Operative Versus Nonoperative Management of Patients with Adult Spinal Deformity with 86% Accuracy

OBJECTIVE

Patients with ASD show complex and highly variable disease. The decision to manage patients operatively is largely subjective and varies based on surgeon training and experience. We sought to develop models capable of accurately discriminating between patients receiving operative versus nonoperative treatment based only on baseline radiographic and clinical data at enrollment.

METHODS

This study was a retrospective analysis of a multicenter consecutive cohort of patients with ASD. A total of 1503 patients were included, divided in a 70:30 split for training and testing. Patients receiving operative treatment were defined as those undergoing surgery up to 1 year after their baseline visit. Potential predictors included available demographics, past medical history, patient-reported outcome measures, and premeasured radiographic parameters from anteroposterior and lateral films. In total, 321 potential predictors were included. Random forest, elastic net regression, logistic regression, and support vector machines (SVMs) with radial and linear kernels were trained.

RESULTS

Of patients in the training and testing sets, 69.0% (n = 727) and 69.1% (n = 311), respectively, received operative management. On evaluation with the testing dataset, performance for SVM linear (area under the curve =0.910), elastic net (0.913), and SVM radial (0.914) models was excellent, and the logistic regression (0.896) and random forest (0.830) models performed very well for predicting operative management of patients with ASD. The SVM linear model showed 86% accuracy.

CONCLUSIONS

This study developed models showing excellent discrimination (area under the curve >0.9) between patients receiving operative versus nonoperative management, based solely on baseline study enrollment values. Future investigations may evaluate the implementation of such models for decision support in the clinical setting.

Machine Learning Algorithm Identifies Patients at High Risk for Early Complications After Intracranial Tumor Surgery: Registry-Based Cohort Study

INTRODUCTION

Reliable preoperative identification of patients at high risk for early postoperative complications occurring within 24 h (EPC) of intracranial tumor surgery can improve patient safety and postoperative management. Statistical analysis using machine learning algorithms may generate models that predict EPC better than conventional statistical methods.

OBJECTIVE

To train such a model and to assess its predictive ability.

METHODS

This cohort study included patients from an ongoing prospective patient registry at a single tertiary care center with an intracranial tumor that underwent elective neurosurgery between June 2015 and May 2017. EPC were categorized based on the Clavien-Dindo classification score. Conventional statistical methods and different machine learning algorithms were used to predict EPC using preoperatively available patient, clinical, and surgery-related variables. The performance of each model was derived from examining classification performance metrics on an out-of-sample test dataset.

RESULTS

EPC occurred in 174 (26%) of 668 patients included in the analysis. Gradient boosting machine learning algorithms provided the model best predicting the probability of an EPC. The model scored an accuracy of 0.70 (confidence interval [CI] 0.59-0.79) with an area under the curve (AUC) of 0.73 and a sensitivity and specificity of 0.80 (CI 0.58-0.91) and 0.67 (CI 0.53-0.77) on the test set. The conventional statistical model showed inferior predictive power (test set: accuracy: 0.59 (CI 0.47-0.71); AUC: 0.64; sensitivity: 0.76 (CI 0.64-0.85); specificity: 0.53 (CI 0.41-0.64)).

CONCLUSION

Using gradient boosting machine learning algorithms, it was possible to create a prediction model superior to conventional statistical methods. While conventional statistical methods favor patients' characteristics, we found the pathology and surgery-related (histology, anatomical localization, surgical access) variables to be better predictors of EPC.

Attitudes of Patients and Their Relatives Toward Artificial Intelligence in Neurosurgery

BACKGROUND

Artificial intelligence (AI) may favorably support surgeons but can result in concern among patients and their relatives. The aim of this study was to evaluate attitudes of patients and their relatives regarding use of AI in neurosurgery.

METHODS

In a 2-stage cross-sectional survey, a qualitative survey was administered to a focus group of former patients to investigate their perception of AI and its role in neurosurgery. Five themes were identified and used to generate a case-based quantitative survey administered to inpatients and their relatives over a 2-week period. Presented AI platforms were rated appropriate and acceptable using 5-point Likert scales. Demographic data were collected. χ² test was used to determine whether demographics influenced participants' attitudes.

RESULTS

In the first stage, 20 participants responded. Five themes were identified: interpretation of imaging (4/20; 20%), operative planning (5/20; 25%), real-time alert of potential complications (10/20; 50%), partially autonomous surgery (6/20; 30%), and fully autonomous surgery (3/20; 15%). In the second stage, 107 participants responded. Most thought it appropriate and acceptable to use AI for imaging interpretation (76.7%; 66.3%), operative planning (76.7%; 75.8%), real-time alert of potential complications (82.2%; 72.9%), and partially autonomous surgery (58%; 47.7%). Conversely, most did not think that fully autonomous surgery was appropriate (27.1%) or acceptable (17.7%). Demographics did not have a significant influence on perception.

CONCLUSIONS

Most patients and their relatives believed that AI has a role in neurosurgery and found it acceptable. Notable exceptions were fully autonomous systems, with most wanting the neurosurgeon ultimately to remain in control.

Utilization of Predictive Modeling to Determine Episode of Care Costs and to Accurately Identify Catastrophic Cost Nonwarranty Outlier Patients in Adult Spinal Deformity Surgery: A Step Toward Bundled Payments and Risk Sharing

STUDY DESIGN

Retrospective review of prospectively-collected, multicenter adult spinal deformity (ASD) database.

OBJECTIVE

The aim of this study was to evaluate the rate of patients who accrue catastrophic cost (CC) with ASD surgery utilizing direct, actual costs, and determine the feasibility of predicting these outliers.

SUMMARY OF BACKGROUND DATA

Cost outliers or surgeries resulting in CC are a major concern for ASD surgery as some question the sustainability of these surgical treatments.

METHODS

Generalized linear regression models were used to explain the determinants of direct costs. Regression tree and random forest models were used to predict which patients would have CC (>$100,000).

RESULTS

A total of 210 ASD patients were included (mean age of 59.3 years, 83% women). The mean index episode of care direct cost was $70,766 (SD = $24,422). By 90 days and 2 years following surgery, mean direct costs increased to $74,073 and $77,765, respectively. Within 90 days of the index surgery, 11 (5.2%) patients underwent 13 revisions procedures, and by 2 years, 26 (12.4%) patients had undergone 36 revision procedures. The CC threshold at the index surgery and 90-day and 2-year follow-up time points was exceeded by 11.9%, 14.8%, and 19.1% of patients, respectively. Top predictors of cost included number of levels fused, surgeon, surgical approach, interbody fusion (IBF), and length of hospital stay (LOS). At 90 days and 2 years, a total of 80.6% and 64.0% of variance in direct cost, respectively, was explained in the generalized linear regression models. Predictors of CC were number of fused levels, surgical approach, surgeon, IBF, and LOS.

CONCLUSION

The present study demonstrates that direct cost in ASD surgery can be accurately predicted. Collectively, these findings may not only prove useful for bundled care initiatives, but also may provide insight into means to reduce and better predict cost of ASD surgery outside of bundled payment plans.

LEVEL OF EVIDENCE

3.

Decision Tree-based Modelling for Identification of Predictors of Blood Loss and Transfusion Requirement After Adult Spinal Deformity Surgery

Background

Multilevel fusions and complex osteotomies to restore global alignment in adult spinal deformity (ASD) surgery can lead to increased operative time and blood loss. In this regard, we assessed factors predictive of perioperative blood product transfusion in patients undergoing long posterior spinal fusion for ASD.

Methods

A single-institution retrospective review was conducted on 909 patients with ASD, age > 18 years, who underwent surgery for ASD with greater than 4 levels fused. Using conditional inference tree analysis, a machine learning methodology, we sought to predict the combination of variables that best predicted increased risk for intraoperative percent blood volume lost and perioperative blood product transfusion.

Results

Among the 909 patients included in the study, 377 (41.5%) received red blood cell (RBC) transfusion. The conditional inference tree analysis identified greater than 13 levels fused, American Society of Anesthesiologists (ASA) score > 1, a history of hypertension, 3-column osteotomy, pelvic fixation, and operative time > 8 hours, as significant risk factors for perioperative RBC transfusion. The best predictors for the subgroup with the highest risk for intraoperative percent blood volume lost (subgroup mean: 53.1% ± 42.9%) were greater than 13 levels fused, ASA score > 1, preoperative hemoglobin < 13.6 g/dL, 3-column osteotomy, posterior column osteotomy, and pelvic fixation. Patients who underwent major blood transfusion intraoperatively had significantly longer length of stay (8.5 days) versus those who did not (6.1 days) (P < .0001). The overall 90-day complication rate in patients who underwent major blood transfusion intraoperatively was 49%, compared with 19% in those who did not (P < .0001). By multivariate regression analysis, patients with a preoperative hemoglobin > 13.0 were less likely to require major blood transfusion (odds ratio: 0.52, P = .046).

Conclusions

Using a supervised learning technique, this study demonstrates that greater than 13 levels fused, ASA score > 1, 3-column osteotomy, and pelvic fixation are consistent risk factors for increased intraoperative percent blood volume lost and perioperative RBC transfusion. The addition of having a preoperative hemoglobin < 13.6 g/dL or undergoing a posterior column osteotomy conferred the highest risk for intraoperative blood loss. This information can assist spinal deformity surgeons in identifying at-risk individuals and allocating healthcare resources.

Level of Evidence

3.

The Impact of Artificial Intelligence on Quality and Safety

As exponential expansion of computing capacity converges with unsustainable health care spending, a hopeful opportunity has emerged: the use of artificial intelligence to enhance health care quality and safety. These computer-based algorithms can perform the intricate and extremely complex mathematical operations of classification or regression on immense amounts of data to detect intricate and potentially previously unknown patterns in that data, with the end result of creating predictive models that can be utilized in clinical practice. Such models are designed to distinguish relevant from irrelevant data regarding a particular patient; choose appropriate perioperative care, intervention or surgery; predict cost of care and reimbursement; and predict future outcomes on a variety of anchored measures. If and when one is brought to fruition, an artificial intelligence platform could serve as the first legitimate clinical decision-making tool in spine care, delivering on the value equation while serving as a source for improving physician performance and promoting appropriate, efficient care in this era of financial uncertainty in health care.

Artificial Intelligence for Adult Spinal Deformity

Adult spinal deformity (ASD) is a complex disease that significantly affects the lives of many patients. Surgical correction has proven to be effective in achieving improvement of spinopelvic parameters as well as improving quality of life (QoL) for these patients. However, given the relatively high complication risk associated with ASD correction, it is of paramount importance to develop robust prognostic tools for predicting risk profile and outcomes. Historically, statistical models such as linear and logistic regression models were used to identify preoperative factors associated with postoperative outcomes. While these tools were useful for looking at simple associations, they represent generalizations across large populations, with little applicability to individual patients. More recently, predictive analytics utilizing artificial intelligence (AI) through machine learning for comprehensive processing of large amounts of data have become available for surgeons to implement. The use of these computational techniques has given surgeons the ability to leverage far more accurate and individualized predictive tools to better inform individual patients regarding predicted outcomes after ASD correction surgery. Applications range from predicting QoL measures to predicting the risk of major complications, hospital readmission, and reoperation rates. In addition, AI has been used to create a novel classification system for ASD patients, which will help surgeons identify distinct patient subpopulations with unique risk-benefit profiles. Overall, these tools will help surgeons tailor their clinical practice to address patients’ individual needs and create an opportunity for personalized medicine within spine surgery.

Applications of Machine Learning Using Electronic Medical Records in Spine Surgery

Developments in machine learning in recent years have precipitated a surge in research on the applications of artificial intelligence within medicine. Machine learning algorithms are beginning to impact medicine broadly, and the field of spine surgery is no exception. Electronic medical records are a key source of medical data that can be leveraged for the creation of clinically valuable machine learning algorithms. This review examines the current state of machine learning using electronic medical records as it applies to spine surgery. Studies across the electronic medical record data domains of imaging, text, and structured data are reviewed. Discussed applications include clinical prognostication, preoperative planning, diagnostics, and dynamic clinical assistance, among others. The limitations and future challenges for machine learning research using electronic medical records are also discussed.

Development and validation of risk stratification models for adult spinal deformity surgery

OBJECTIVE

Adult spinal deformity (ASD) surgery has a high rate of major complications (MCs). Public information about adverse outcomes is currently limited to registry average estimates. The object of this study was to assess the incidence of adverse events after ASD surgery, and to develop and validate a prognostic tool for the time-to-event risk of MC, hospital readmission (RA), and unplanned reoperation (RO).

METHODS

Two models per outcome, created with a random survival forest algorithm, were trained in an 80% random split and tested in the remaining 20%. Two independent prospective multicenter ASD databases, originating from the European continent and the United States, were queried, merged, and analyzed. ASD patients surgically treated by 57 surgeons at 23 sites in 5 countries in the period from 2008 to 2016 were included in the analysis.

RESULTS

The final sample consisted of 1612 ASD patients: mean (standard deviation) age 56.7 (17.4) years, 76.6% women, 10.4 (4.3) fused vertebral levels, 55.1% of patients with pelvic fixation, 2047.9 observation-years. Kaplan-Meier estimates showed that 12.1% of patients had at least one MC at 10 days after surgery; 21.5%, at 90 days; and 36%, at 2 years. Discrimination, measured as the concordance statistic, was up to 71.7% (95% CI 68%-75%) in the development sample for the postoperative complications model. Surgical invasiveness, age, magnitude of deformity, and frailty were the strongest predictors of MCs. Individual cumulative risk estimates at 2 years ranged from 3.9% to 74.1% for MCs, from 3.17% to 44.2% for RAs, and from 2.67% to 51.9% for ROs.

CONCLUSIONS

The creation of accurate prognostic models for the occurrence and timing of MCs, RAs, and ROs following ASD surgery is possible. The presented variability in patient risk profiles alongside the discrimination and calibration of the models highlights the potential benefits of obtaining time-to-event risk estimates for patients and clinicians.

The use of tranexamic acid in adult spinal deformity: is there an optimal dosing strategy?

BACKGROUND CONTEXT

ASD (Adult spinal deformity) surgery often entails complex osteotomies and realignment procedures, particularly in the setting of rigid deformities. Although previous studies have established the efficacy of tranexamic acid (TXA), data evaluating the widely variable dosing regimens remains sparse.

PURPOSE

To improve understanding of blood loss and transfusion requirements for low-dose and high-dose TXA regimens for adult spinal deformity (ASD) surgery.

STUDY DESIGN/SETTING

This is a retrospective cohort study of 318 ASD patients who received TXA. Outcome measures include estimated blood loss (EBL), perioperative transfusion requirement, and complications.

METHODS

A retrospective review was conducted on 318 ASD patients: 258 patients received a low-dose regimen of TXA (10 or 20 mg/kg loading dose with a 1 or 2 mg/kg/h maintenance dose) and 60 patients received a high-dose regimen of TXA (40 mg/kg loading dose with a 1 mg/kg/h maintenance dose, 30 mg/kg loading dose with a 10 mg/kg/h maintenance dose, or 50 mg/kg loading dose with a 5 mg/kg/h maintenance dose).

RESULTS

Compared with the low-dose TXA group, the high-dose TXA group had significantly decreased EBL (1402 vs. 1793 mL, p=.009), blood volume lost (30.3 vs. 39.4%, p=.01), intraoperative packed red blood cell (pRBC) transfusion (0.9 vs. 1.6 U, p<.0001), and intraoperative platelet transfusion (0 vs. 0.1 U, p<.0001). High-dose TXA was predictive of 515 cc less EBL (p=.002), 11.4% less blood volume lost (p=.004), and 1 U pRBC less transfused intraoperatively (p<.0001) than the low-dose TXA group. The high-dose TXA group had a higher incidence of postop atrial fibrillation (5 vs. 0%, p<.0001) and myocardial infarction (1.7 vs. 0%, p=.04).

CONCLUSIONS

Varying dosing regimens of TXA are utilized for ASD surgery, with a prevailing theme of dosing ambiguity. These data demonstrate that high-dose TXA is more effective than low-dose TXA in reducing blood loss and blood product transfusion requirement in ASD surgery. Importantly, rates of MI and postop AF were higher in the high-dose TXA group.

[A new classification of surgical complications in adult spinal deformity]

BACKGROUND

In the light of the increasingly aging population and the widespread understanding of the sagittal profile of symptomatic patients with adult spinal deformity (ASD), pervasive utilization of osteotomies on the vertebral column should be expected. These surgeries are accompanied with relatively high complication rates. However, there is no uniform definition or classification in terms of grading the severity or chronological incidence of complications after ASD surgery.

OBJECTIVES

The aim of this work is to give an overview of the different classifications described in the literature hitherto and to propose a standardized, clinically utile classification of complications after ASD surgery. Finally, the aim is to illustrate this classification using two case examples.

MATERIALS AND METHODS

We conducted a systematic PubMed search with the keywords: "adult spinal deformity", "surgery", "complications" and "classification". Results were screened by title, abstract and full-text article.

RESULTS

22 articles were included in this review. Regarding the systematic classification of the severity of a complication, the CTCAE classification (Common Terminology Criteria for Adverse Events v4.0) is a validated and well-established severity stratification tool used in oncologic treatment. Regarding chronological occurrence, complications can be categorized into three phases: intra-operative, peri-operative and post-operative.

DISCUSSION

The time of occurrence of a certain complication and its severity should constitute the cornerstones of a standardized and practical classification of complications after ASD surgery. To enable uniform reporting and coherent documentation of complications, spine surgeons should find consensus on a standardized classification. Future work needs to be directed towards defining and conducting an individual pre-operative risk stratification of adult spine deformity surgical candidates leading to a possible mitigation of surgery-related complications.

The use of tranexamic acid in spine surgery

Patients undergoing surgical procedures of the spine with associated large volume blood loss often require perioperative blood conservation strategies. Synthetic antifibrinolytic medications such as tranexamic acid (TXA) may reduce blood transfusion requirements and postoperative complications following spinal procedures. Studies investigating the role of TXA in spine surgery have presented promising results and have proven its safety and efficacy. However, further investigation is needed to determine the optimal dosing regimen of TXA. In this article, we provide an overview of the basic science and pharmacology of TXA. A comprehensive summary of the findings from clinical trials and a review of the literature that demonstrate the risks and benefits of TXA in spine surgery are also presented.

Machine learning applications to clinical decision support in neurosurgery: an artificial intelligence augmented systematic review

Machine learning (ML) involves algorithms learning patterns in large, complex datasets to predict and classify. Algorithms include neural networks (NN), logistic regression (LR), and support vector machines (SVM). ML may generate substantial improvements in neurosurgery. This systematic review assessed the current state of neurosurgical ML applications and the performance of algorithms applied. Our systematic search strategy yielded 6866 results, 70 of which met inclusion criteria. Performance statistics analyzed included area under the receiver operating characteristics curve (AUC), accuracy, sensitivity, and specificity. Natural language processing (NLP) was used to model topics across the corpus and to identify keywords within surgical subspecialties. ML applications were heterogeneous. The densest cluster of studies focused on preoperative evaluation, planning, and outcome prediction in spine surgery. The main algorithms applied were NN, LR, and SVM. Input and output features varied widely and were listed to facilitate future research. The accuracy (F_(2,19) = 6.56, p < 0.01) and specificity (F_(2,16) = 5.57, p < 0.01) of NN, LR, and SVM differed significantly. NN algorithms demonstrated significantly higher accuracy than LR. SVM demonstrated significantly higher specificity than LR. We found no significant difference between NN, LR, and SVM AUC and sensitivity. NLP topic modeling reached maximum coherence at seven topics, which were defined by modeling approach, surgery type, and pathology themes. Keywords captured research foci within surgical domains. ML technology accurately predicts outcomes and facilitates clinical decision-making in neurosurgery. NNs frequently outperformed other algorithms on supervised learning tasks. This study identified gaps in the literature and opportunities for future neurosurgical ML research.

Rigorous Correction of Sagittal Vertical Axis Is Correlated With Better ODI Outcomes After Extensive Corrective Fusion in Elderly or Extremely Elderly Patients With Spinal Deformity

STUDY DESIGN

Retrospective analysis of a prospectively collected consecutive case series.

OBJECTIVES

To determine the effect of spinopelvic correction on clinical outcomes and discuss the approach to target alignment in elderly or extremely elderly spinal deformity patients.

SUMMARY OF BACKGROUND DATA

Age-dependent target alignment during corrective fusion surgery in elderly patients remains controversial. Age-related target spinal alignment should be examined based on the outcomes data of patients with fused, nonphysiological spines.

METHODS

Consecutive adult spinal deformity (ASD) patients aged 45 years or older who underwent thoracolumbar corrective fusion of at least five levels were included. Spinopelvic radiographic parameters, health-related quality of life (Oswestry Disability Index [ODI]), and the scores on a numeric rating scale of low back pain were investigated before and after the operation. The patients were stratified into three groups according to age as follows: Middle-Age, 45-64 years; Elderly, 65-74 years; and Extremely Elderly, ≥75 years. We also stratified the patients into three groups according to lumber lordosis (LL) as follows: ideal (within ±5° of ideal LL), moderate (between -5° and -20° of ideal LL), and under (under ideal LL by -20°).

RESULTS

A total of 149 patients (Middle-Age, 38; Elderly, 68; and Extremely Elderly, 43) were included in this study. No significant difference was observed in any of the radiographic parameters in each age group. The ODI and numeric rating scale scores of the ideal-correction group at two years after surgery were significantly better than those of the undercorrection group across all ages. A significant correlation with ODI was observed between sagittal spinopelvic parameters in the Elderly and Extremely Elderly groups. A stronger correlation was observed in the Extremely Elderly group compared with the Elderly group.

CONCLUSIONS

Rigorous realignment of sagittal vertical axis is correlated with ODI outcomes, especially in very elderly patients.

LEVEL OF EVIDENCE

Level 4.

Artificial Intelligence Based Hierarchical Clustering of Patient Types and Intervention Categories in Adult Spinal Deformity Surgery: Towards a New Classification Scheme that Predicts Quality and Value

STUDY DESIGN

Retrospective review of prospectively-collected, multicenter adult spinal deformity (ASD) databases.

OBJECTIVE

To apply artificial intelligence (AI)-based hierarchical clustering as a step toward a classification scheme that optimizes overall quality, value, and safety for ASD surgery.

SUMMARY OF BACKGROUND DATA

Prior ASD classifications have focused on radiographic parameters associated with patient reported outcomes. Recent work suggests there are many other impactful preoperative data points. However, the ability to segregate patient patterns manually based on hundreds of data points is beyond practical application for surgeons. Unsupervised machine-based clustering of patient types alongside surgical options may simplify analysis of ASD patient types, procedures, and outcomes.

METHODS

Two prospective cohorts were queried for surgical ASD patients with baseline, 1-year, and 2-year SRS-22/Oswestry Disability Index/SF-36v2 data. Two dendrograms were fitted, one with surgical features and one with patient characteristics. Both were built with Ward distances and optimized with the gap method. For each possible n patient cluster by m surgery, normalized 2-year improvement and major complication rates were computed.

RESULTS

Five hundred-seventy patients were included. Three optimal patient types were identified: young with coronal plane deformity (YC, n = 195), older with prior spine surgeries (ORev, n = 157), and older without prior spine surgeries (OPrim, n = 218). Osteotomy type, instrumentation and interbody fusion were combined to define four surgical clusters. The intersection of patient-based and surgery-based clusters yielded 12 subgroups, with major complication rates ranging from 0% to 51.8% and 2-year normalized improvement ranging from -0.1% for SF36v2 MCS in cluster [1,3] to 100.2% for SRS self-image score in cluster [2,1].

CONCLUSION

Unsupervised hierarchical clustering can identify data patterns that may augment preoperative decision-making through construction of a 2-year risk-benefit grid. In addition to creating a novel AI-based ASD classification, pattern identification may facilitate treatment optimization by educating surgeons on which treatment patterns yield optimal improvement with lowest risk.

LEVEL OF EVIDENCE

4.

Surgical risk stratification based on preoperative risk factors in adult spinal deformity

BACKGROUND CONTEXT

Corrective surgery for adult spinal deformity (ASD) improves health-related quality of life but has high complication rates. Predicting a patient's risk of perioperative and late postoperative complications is difficult, although several potential risk factors have been reported.

PURPOSE

To establish an accurate, ASD-specific model for predicting the risk of postoperative complications, based on baseline demographic, radiographic, and surgical invasiveness data in a retrospective case series.

STUDY DESIGN/SETTING

Multicentered retrospective review and the surgical risk stratification.

PATIENT SAMPLE

One hundred fifty-one surgically treated ASD at our hospital for risk analysis and model building and 89 surgically treated ASD at 2 other our hospitals for model validation.

OUTCOME MEASURES

HRQoL measures and surgical complications.

METHODS

We analyzed demographic and medical data, including complications, for 151 adults with ASD who underwent surgery at our hospital and were followed for at least 2years. Each surgical risk factor identified by univariate analyses was assigned a value based on its odds ratio, and the values of all risk factors were summed to obtain a surgical risk score (range 0-20). We stratified risk scores into grades (A-D) and analyzed their correlations with complications. We validated the model using data from 89 patients who underwent ASD surgery at two other hospitals.

RESULTS

Complications developed in 48% of the patients in the model-building cohort. Univariate analyses identified 10 demographic, physical, and surgical risk indicators, with odds ratios from 5.4 to 1.4, for complications. Our risk-grading system showed good calibration and discrimination in the validation cohort. The complication rate increased with and correlated well with the risk grade using receiver operating characteristic curves.

CONCLUSIONS

This simple, ASD-specific model uses readily accessible indicators to predict a patient's risk of perioperative and postoperative complications and can help surgeons adjust treatment strategies for best outcomes in high-risk patients.

Treatment of adult thoracolumbar spinal deformity: past, present, and future

Care of the patient with adult spinal deformity (ASD) has evolved from being primarily supportive to now having the ability to directly treat and correct the spinal pathology. The focus of this narrative literature review is to briefly summarize the history of ASD treatment, discuss the current state of the art of ASD care with focus on surgical treatment and current challenges, and conclude with a discussion of potential developments related to ASD surgery.In the past, care for ASD was primarily based on supportive measures, including braces and assistive devices, with few options for surgical treatments that were often deemed high risk and reserved for rare situations. Advances in anesthetic and critical care, surgical techniques, and instrumentation now enable almost routine surgery for many patients with ASD. Despite the advances, there are many remaining challenges currently impacting the care of ASD patients, including increasing numbers of elderly patients with greater comorbidities, high complication and reoperation rates, and high procedure cost without clearly demonstrated cost-effectiveness based on standard criteria. In addition, there remains considerable variability across multiple aspects of ASD surgery. For example, there is currently very limited ability to provide preoperative individualized counseling regarding optimal treatment approaches (e.g., operative vs nonoperative), complication risks with surgery, durability of surgery, and likelihood of achieving individualized patient goals and satisfaction. Despite the challenges associated with the current state-of-the-art ASD treatment, surgery continues to be a primary option, as multiple reports have demonstrated the potential for surgery to significantly improve pain and disability. The future of ASD care will likely include techniques and technologies to markedly reduce complication rates, including greater use of navigation and robotics, and a shift toward individualized medicine that enables improved counseling, preoperative planning, procedure safety, and patient satisfaction.Advances in the care of ASD patients have been remarkable over the past few decades. The current state of the art enables almost routine surgical treatment for many types of ASD that have the potential to significantly improve pain and disability. However, significant challenges remain, including high complication rates, lack of demonstrated cost-effectiveness, and limited ability to meaningfully counsel patients preoperatively on an individual basis. The future of ASD surgery will require continued improvement of predictability, safety, and sustainability.

Long Term Outcomes and Effects of Surgery on Degenerative Spinal Deformity: A 14-Year National Cohort Study

Degenerative spinal deformity (DSD) has become a prevalent cause of disability and pain among the aging population worldwide. Though surgery has emerged as a promising option for DSD, the natural course, outcomes, and effects of surgery on DSD have remained elusive. This cohort study used a national database to comprehensively follow up patients of DSD for all-cause mortality, respiratory problems, and hip fracture-related hospitalizations. All patients were grouped into an operation or a non-operation group for comparison. An adjustment of demographics, comorbidities, and propensity-score matching was conducted to ameliorate confounders. A Cox regression hazard ratio (HR) model and Kaplan-Meier analysis were also applied. The study comprised 21,810 DSD patients, including 12,544 of the operation group and 9266 of the non-operation group. During the 14 years (total 109,591.2 person-years) of follow-up, the operation group had lower mortality (crude hazard ratio = 0.40), lower respiratory problems (cHR = 0.45), and lower hip fractures (cHR = 0.63) than the non-operation group (all p < 0.001). After adjustment, the risks for mortality and respiratory problems remained lower (adjusted HR = 0.60 and 0.65, both p < 0.001) in the operation than the non-operation group, while hip fractures were indifferent (aHR = 1.08, p > 0.05). Therefore, surgery for DSD is invaluable since it could reduce the risks of mortality and of hospitalization for respiratory problems.

Predicting Clinical Outcomes Following Surgical Correction of Adult Spinal Deformity

BACKGROUND

Deformity reconstruction surgery has been shown to improve quality of life (QOL) in cases of adult spinal deformity (ASD) but is associated with significant morbidity.

OBJECTIVE

To create a preoperative predictive nomogram to help risk-stratify patients and determine which would likely benefit from corrective surgery for ASD as measured by patient-reported health-related quality of life (HRQoL).

METHODS

All patients aged 25-yr and older with radiographic evidence of ASD and QOL data that underwent thoracolumbar fusion between 2008 and 2014 were identified. Demographic and clinical parameters were obtained. The EuroQol 5 dimensions questionnaire (EQ-5D) was used to measure HRQoL preoperatively and at 12-mo postoperative follow-up. Logistic regression of preoperative variables was used to create the prognostic nomogram.

RESULTS

Our sample included data from 191 patients. Fifty-one percent of patients experienced clinically relevant postoperative improvement in HRQoL. Seven variables were included in the final model: preoperative EQ-5D score, sex, preoperative diagnosis (degenerative, idiopathic, or iatrogenic), previous spinal surgical history, obesity, and a sex-by-obesity interaction term. Preoperative EQ-5D score independently predicted the outcome. Sex interacted with obesity: obese men were at disproportionately higher odds of improving than nonobese men, but obesity did not affect odds of the outcome among women. Model discrimination was good, with an optimism-adjusted c-statistic of 0.739.

CONCLUSION

The predictive nomogram that we developed using these data can improve preoperative risk counseling and patient selection for deformity correction surgery.

Artificial intelligence and machine learning in spine research

Artificial intelligence (AI) and machine learning (ML) techniques are revolutionizing several industrial and research fields like computer vision, autonomous driving, natural language processing, and speech recognition. These novel tools are already having a major impact in radiology, diagnostics, and many other fields in which the availability of automated solution may benefit the accuracy and repeatability of the execution of critical tasks. In this narrative review, we first present a brief description of the various techniques that are being developed nowadays, with special focus on those used in spine research. Then, we describe the applications of AI and ML to problems related to the spine which have been published so far, including the localization of vertebrae and discs in radiological images, image segmentation, computer-aided diagnosis, prediction of clinical outcomes and complications, decision support systems, content-based image retrieval, biomechanics, and motion analysis. Finally, we briefly discuss major ethical issues related to the use of AI in healthcare, namely, accountability, risk of biased decisions as well as data privacy and security, which are nowadays being debated in the scientific community and by regulatory agencies.

Relevance of Modified Debridement-Irrigation, Antibiotic Therapy and Implant Retention Protocol for the Management of Surgical Site Infections: A Series of 1694 Instrumented Spinal Surgery

Introduction: Management of surgical site infections (SSI) after instrumented spinal surgery remains controversial. The debridement-irrigation, antibiotic therapy and implant retention protocol (DAIR protocol) is safe and effective to treat deep SSI occurring within the 3 months after instrumented spinal surgery. Methods: This retrospective study describes the outcomes of patients treated over a period of 42 months for deep SSI after instrumented spinal surgery according to a modified DAIR protocol. Results: Among 1694 instrumented surgical procedures, deep SSI occurred in 46 patients (2.7%): 41 patients (89%) experienced early SSI (< 1 month), 3 (7%) delayed SSI (from 1 to 3 months), and 2 (4%) late SSI (> 3months). A total of 37 patients had a minimum 1 year of follow-up; among these the modified DAIR protocol was effective in 28 patients (76%) and failed (need for new surgery for persistent signs of SSI beyond 7 days) in 9 patients (24%). Early second-look surgery (≤ 7days) for iterative debridement was performed in 3 patients, who were included in the cured group. Among the 9 patients in whom the modified DAIR protocol failed, none had early second-look surgery; 3 (33%) recovered and were cured at 1 year follow-up, and 6 (66%) relapsed. Overall, among patients with SSI and a minimum 1 year follow-up, the modified DAIR protocol led to healing in 31/37 (84%) patients. Conclusions: The present study supports the effectiveness of a modified DAIR protocol in deep SSI occurring within the 3 months after instrumented spinal surgery. An early second-look surgery for iterative debridement could increase the success rate of this treatment.

Development of a validated computer-based preoperative predictive model for pseudarthrosis with 91% accuracy in 336 adult spinal deformity patients

OBJECTIVE

Pseudarthrosis can occur following adult spinal deformity (ASD) surgery and can lead to instrumentation failure, recurrent pain, and ultimately revision surgery. In addition, it is one of the most expensive complications of ASD surgery. Risk factors contributing to pseudarthrosis in ASD have been described; however, a preoperative model predicting the development of pseudarthrosis does not exist. The goal of this study was to create a preoperative predictive model for pseudarthrosis based on demographic, radiographic, and surgical factors.

METHODS

A retrospective review of a prospectively maintained, multicenter ASD database was conducted. Study inclusion criteria consisted of adult patients (age ≥ 18 years) with spinal deformity and surgery for the ASD. From among 82 variables assessed, 21 were used for model building after applying collinearity testing, redundancy, and univariable predictor importance ≥ 0.90. Variables included demographic data along with comorbidities, modifiable surgical variables, baseline coronal and sagittal radiographic parameters, and baseline scores for health-related quality of life measures. Patients groups were determined according to their Lenke radiographic fusion type at the 2-year follow-up: bilateral or unilateral fusion (union) or pseudarthrosis (nonunion). A decision tree was constructed, and internal validation was accomplished via bootstrapped training and testing data sets. Accuracy and the area under the receiver operating characteristic curve (AUC) were calculated to evaluate the model.

RESULTS

A total of 336 patients were included in the study (nonunion: 105, union: 231). The model was 91.3% accurate with an AUC of 0.94. From 82 initial variables, the top 21 covered a wide range of areas including preoperative alignment, comorbidities, patient demographics, and surgical use of graft material.

CONCLUSIONS

A model for predicting the development of pseudarthrosis at the 2-year follow-up was successfully created. This model is the first of its kind for complex predictive analytics in the development of pseudarthrosis for patients with ASD undergoing surgical correction and can aid in clinical decision-making for potential preventative strategies.

Generative models: an upcoming innovation in musculoskeletal radiology? A preliminary test in spine imaging

Background

Deep learning is a ground-breaking technology that is revolutionising many research and industrial fields. Generative models are recently gaining interest. Here, we investigate their potential, namely conditional generative adversarial networks, in the field of magnetic resonance imaging (MRI) of the spine, by performing clinically relevant benchmark cases.

Methods

First, the enhancement of the resolution of T2-weighted (T2W) images (super-resolution) was tested. Then, automated image-to-image translation was tested in the following tasks: (1) from T1-weighted to T2W images of the lumbar spine and (2) vice versa; (3) from T2W to short time inversion-recovery (STIR) images; (4) from T2W to turbo inversion recovery magnitude (TIRM) images; (5) from sagittal standing x-ray projections to T2W images. Clinical and quantitative assessments of the outputs by means of image quality metrics were performed. The training of the models was performed on MRI and x-ray images from 989 patients.

Results

The performance of the models was generally positive and promising, but with several limitations. The number of disc protrusions or herniations showed good concordance (κ = 0.691) between native and super-resolution images. Moderate-to-excellent concordance was found when translating T2W to STIR and TIRM images (κ ≥ 0.842 regarding disc degeneration), while the agreement was poor when translating x-ray to T2W images.

Conclusions

Conditional generative adversarial networks are able to generate perceptually convincing synthetic images of the spine in super-resolution and image-to-image translation tasks. Taking into account the limitations of the study, deep learning-based generative methods showed the potential to be an upcoming innovation in musculoskeletal radiology.

Potential of predictive computer models for preoperative patient selection to enhance overall quality-adjusted life years gained at 2-year follow-up: a simulation in 234 patients with adult spinal deformity

OBJECTIVE Patients with adult spinal deformity (ASD) experience significant quality of life improvements after surgery. Treatment, however, is expensive and complication rates are high. Predictive analytics has the potential to use many variables to make accurate predictions in large data sets. A validated minimum clinically important difference (MCID) model has the potential to assist in patient selection, thereby improving outcomes and, potentially, cost-effectiveness. METHODS The present study was a retrospective analysis of a multiinstitutional database of patients with ASD. Inclusion criteria were as follows: age ≥ 18 years, radiographic evidence of ASD, 2-year follow-up, and preoperative Oswestry Disability Index (ODI) > 15. Forty-six variables were used for model training: demographic data, radiographic parameters, surgical variables, and results on the health-related quality of life questionnaire. Patients were grouped as reaching a 2-year ODI MCID (+MCID) or not (-MCID). An ensemble of 5 different bootstrapped decision trees was constructed using the C5.0 algorithm. Internal validation was performed via 70:30 data split for training/testing. Model accuracy and area under the curve (AUC) were calculated. The mean quality-adjusted life years (QALYs) and QALYs gained at 2 years were calculated and discounted at 3.5% per year. The QALYs were compared between patients in the +MCID and -MCID groups. RESULTS A total of 234 patients met inclusion criteria (+MCID 129, -MCID 105). Sixty-nine patients (29.5%) were included for model testing. Predicted versus actual results were 50 versus 40 for +MCID and 19 versus 29 for -MCID (i.e., 10 patients were misclassified). Model accuracy was 85.5%, with 0.96 AUC. Predicted results showed that patients in the +MCID group had significantly greater 2-year mean QALYs (p = 0.0057) and QALYs gained (p = 0.0002). CONCLUSIONS A successful model with 85.5% accuracy and 0.96 AUC was constructed to predict which patients would reach ODI MCID. The patients in the +MCID group had significantly higher mean 2-year QALYs and QALYs gained. This study provides proof of concept for using predictive modeling techniques to optimize patient selection in complex spine surgery.

Predicting critical care unit-level complications after long-segment fusion procedures for adult spinal deformity

Background

To identify predictive factors for critical care unit-level complications (CCU complication) after long-segment fusion procedures for adult spinal deformity (ASD).

Methods

The American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database [2010-2014] was reviewed. Only adult patients who underwent fusion of 7 or more spinal levels for ASD were included. CCU complications included intraoperative arrest/infarction, ventilation >48 hours, pulmonary embolism, renal failure requiring dialysis, cardiac arrest, myocardial infarction, unplanned intubation, septic shock, stroke, coma, or new neurological deficit. A stepwise multivariate regression was used to identify independent predictors of CCU complications.

Results

Among 826 patients, the rate of CCU complications was 6.4%. On multivariate regression analysis, dependent functional status (P=0.004), combined approach (P=0.023), age (P=0.044), diabetes (P=0.048), and surgery for over 8 hours (P=0.080) were significantly associated with complication development. A simple scoring system was developed to predict complications with 0 points for patients aged <50, 1 point for patients between 50-70, 2 points for patients 70 or over, 1 point for diabetes, 2 points dependent functional status, 1 point for combined approach, and 1 point for surgery over 8 hours. The rate of CCU complications was 0.7%, 3.2%, 9.0%, and 12.6% for patients with 0, 1, 2, and 3+ points, respectively (P<0.001).

Conclusions

The findings in this study suggest that older patients, patients with diabetes, patients who depend on others for activities of daily living, and patients who undergo combined approaches or surgery for over 8 hours may be at a significantly increased risk of developing a CCU-level complication after ASD surgery.