Using machine learning in the prediction of symptomatic venous thromboembolism following ankle fracture

BACKGROUND

Venous thromboembolism (VTE) is a major cause of morbidity and mortality in the trauma setting, and both prediction and prevention of VTE have long been a concern for healthcare providers in orthopedic surgery. The purpose of this study was to evaluate the use of novel statistical analysis and machine-learning in predicting the risk of VTE and the usefulness of prophylaxis following ankle fractures.

METHODS

The medical profiles of 16,421 patients with ankle fractures were screened retrospectively for symptomatic VTE. In total, 238 patients sustaining either surgical or nonsurgical treatment for ankle fracture with subsequently confirmed VTE within 180 days following the injury were placed in the case group. Alternatively, 937 patients who sustained ankle fractures managed similarly but had no documented evidence of VTE were randomly chosen as the control group. Individuals from both the case and control populations were also divided into those who had received VTE prophylaxis and those who had not. Over 110 variables were included. Conventional statistics and machine learning methods were used for data analysis.

RESULTS

Patients who had a motor vehicle accident, surgical treatment, increased hospital stay, and were on warfarin were shown to have a higher incidence of VTE, whereas patients who were on statins had a lower incidence of VTE. The highest Area Under the Receiver Operating Characteristic Curves (AUROC) showing the performance of our machine learning approach was 0.88 with 0.94 sensitivity and 0.36 specificity. The most balanced performance was seen in a model that was trained using selected variables with 0.86 AUROC, 0.75 sensitivity, and 0.85 specificity.

CONCLUSION

By using machine learning, this study successfully pinpointed several predictive factors linked to the occurrence or absence of VTE in patients who experienced an ankle fracture. Training these algorithms using larger, more granular, and multicentric data will further increase their validity and reliability and should be considered the standard for the development of such algorithms.

LEVEL OF EVIDENCE

Case-Control study - 3.

Comparing machine learning algorithms for non-invasive detection and classification of failure in piezoresistive bone cement via electrical impedance tomography

At an estimated cost of $8 billion annually in the United States, revision surgeries to total joint replacements represent a substantial financial burden to the health care system and a tremendous mental and physical burden on patients and their caretakers. Fixation failures, such as implant loosening, wear, and mechanical instability of the poly(methyl methacrylate) (PMMA) cement, which bonds the implant to the bone, are the main causes of long-term implant failure. Early and accurate diagnosis of cement failure is critical for developing novel therapeutic strategies and reducing the high risk of a misjudged revision. Unfortunately, prevailing imaging modalities, notably plain radiographs, struggle to detect the precursors of implant failure and are often interpreted incorrectly. Our prior work has shown that the modification of PMMA bone cement with low concentrations of conductive fillers makes it piezoresistive and therefore self-sensing. When combined with a conductivity imaging modality such as electrical impedance tomography (EIT), it is possible to monitor load transfer across the PMMA using cost-effective, physiologically benign, non-contact, and real-time electrical measurements. Despite the ability of EIT for monitoring load transfer across self-sensing PMMA bone cement, it is unable to accurately characterize failure mechanisms. Overcoming this challenge is critical to the success of this technology in practice. Therefore, we herein expand upon our previous results by integrating machine learning techniques with EIT for cement condition characterization with the goal of establishing the feasibility of even off-the-shelf machine learning algorithms to address this important problem. We survey a wide variety of different machine learning algorithms for application to this problem, including neural networks on voltage readings of an EIT phantom for tracking the spatial position of a sample, specifying defect orientation within a sample, and classifying defect types, including cracks and delaminations. In addition, we explore the utilization of principal component analysis (PCA) for pre-treating impedance signals in each of these problems. Within the tested algorithms, our results show clear advantages of neural networks, support vector machines, and K-nearest neighbor algorithms for interpreting EIT signals. We also show that PCA is an effective addition to machine learning. These preliminary results demonstrate that the combination of smart materials, EIT, and machine learning may be a powerful instrumentation tool for diagnosing the origin and evolution of mechanical failure in joint replacements.

Determining the Key Predictive Factors for Non-Union in Fifth Metatarsal Fractures: A Machine Learning-Based Study

Category:

Midfoot/Forefoot; Trauma; Other

Introduction/Purpose:

Metatarsal fractures account for over 35% of all foot fractures, and of these 68% specifically involve the fifth metatarsal [1],[2]. Subgroups of fractures affecting the fifth metatarsal base may be at higher risk of nonunion and therefore benefit from early surgical fixation, but traditional predictive models focus on the location of the fracture and little else. In this study, we aimed to determine predictive factors associated with non-union of fifth metatarsal fractures to assist surgeons and patients, alike, in identifying those at higher risk of nonunion.

Methods:

A retrospective machine learning-based analysis of 1,000 patients, >=18 y/o, diagnosed with a fifth metatarsal fracture at three tertiary medical centers was conducted. The fifth metatarsal base fracture was confirmed radiographically. We gathered imaging and narrative data including demographics (age, height, weight, BMI, gender, race, smoking habits, activity level), medications, chronic conditions, and fracture status (fracture zone, displacement, treatment method, healing status, and healing time). Non-union was described as failing to heal within 180 days of initial injury [3]. A machine learning analysis together with Pearson's correlation test and T-test were utilized where applicable. Five imputation methods were used to fill in missing datapoints. P<0.05 was considered statistically significant.

Results:

Overall, this cohort of patients demonstrated a non-union rate of 22.4%. When divided by fracture zone, Zone 2 fractures results in a statistically significant increased rate of delayed union (17.2%) and non-union (8.6%), when compared to Zone 1 (10.8% and 5.8%, respectively) and Zone 3 (9.7%, 2.3%). Analysis of correlation between demographics data and union rates found no correlation with age, gender, race, or BMI. Our machine learning algorithm outcomes showed a significant correlation between nonunion and seven chronic diseases: diabetes, thyroid disease, hypertension, gastroesophageal reflux disease (GERD), irritable bowel syndrome (IBS) obstructive sleep apnea (OSA), and glaucoma. In terms of medications, significant correlation with nonunion was demonstrated with the use of levothyroxine, lisinopril, aspirin, steroids, and acetaminophen.

Conclusion:

Zone 2 fifth metatarsal base fractures have a significantly higher rate of nonunion as compared to other zones. Comorbid conditions including diabetes, thyroid disease, OSA, and glaucoma as well as medications, may also play a role, though their mechanism or correlative precipitants are yet to be determined. Although our results demonstrated correlation, causality can only be assessed using studies with limited confounding factors, such as clinical trials. Our outcomes suggest that physicians should pay attention not only to the location of the fracture, but also past medical history and medication use when making treatment decisions and discussing prognosis with patients.

Correlation of Statins Use with the Incidence of Venous Thromboembolism in Patients with Ankle Fracture

Category:

Ankle; Other

Introduction/Purpose:

Finding factors that can exacerbate or ameliorate the incidence of Venous thromboembolism (VTE) can affect the process of making decision on whether to start prophylaxis or not, especially when on the verge of whether to-give or not-to-give prophylaxis. Among each patient's profile, medications are of the most important factors influencing surgeon's decision on the prophylactic methods in VTE-vulnerable patients. Among medications, Statins were shown to reduce the incidence of VTE in patients who were receiving them for hyperlipidemia and cardiovascular conditions. However, none of the current VTE prediction methods, particularly in orthopaedic practice, have considered statins protective. Herein we aimed to determine any correlations between statin consumption and the incidence of VTE in ankle fractures and whether to include statins in prediction models of VTE.

Methods:

In this case-control machine learning-based study, approved by the Institutional Review Board (IRB), the ICD and CPT codes were used to identify the patients who were diagnosed with ankle fracture in the Mass General Brigham database from 2004 to end of May 2021. After screening approximately 16,421 patients with ankle fractures, a total of 1,176 patients who were suspect VTE according to their signs and symptoms were recruited, 239 had confirmed VTE (case group) and 937 did not have VTE (controls). Forty-nine cases and 396 controls were statin users. Using a semi-automated machine learning-based algorithm, patients' demographics, past medical and surgical history, fracture characteristics and weber classification, and statin consumption status were obtained, and values were organized in a numerical analyzable manner in the dataset. We used chi-squared and Pearson correlation tests where applicable, and outcomes were displayed and interpreted using p-value (p<0.05 considered significant) and odds ratio (OR).

Results:

The mean age and BMI in our case group were 55.1+-17.0 y/o and 30.0+-6.0, respectively; age and BMI in the controls were 69.4+-13.2 (p=0.09 vs. cases) and 29.2+-6.6 (p=0.12 vs. cases), respectively. Gender distribution is depicted in table 1. In addition, we found that in our population, a total of 239 patients had VTE, from which 49 (21%) were taking Statins and 190 (79%) were not. Out of the 937 patients who did not develop VTE, 396 (42%) were taking Statins whereas 541 (58%) were not. We found that patients taking statins had lower incidence of VTE after their ankle fracture, compared with patients not taking statins (OR=0.36, p <0.001). The distribution of statin users/non-users among cases and controls is shown in table 2. Moreover, using our machine learning algorithm, conditions that would necessitate the use of statins including cardiovascular diseases and hyperlipidemia showed negative significant correlation with VTE (p<=0.02).

Conclusion:

Several studies have suggested that hyperlipidemic blood is prone to a greater generation of thrombin, endothelial dysfunction, and higher platelet activity. By disturbing these mechanisms, statins play a protective role against VTE. Herein, using machine learning algorithms together with statistical analysis, we found that Statins were significantly associated with a lower rate of VTE in patients with an ankle fracture. These findings can be considered in future prediction models that are built based on patient-specific factors. Knowing the protective effect of statins can also help clinicians with deciding on prophylaxis administration in at VTE-risk patients.fig

Prediction of Venous Thromboembolism after Ankle Fractures using Machine Learning: To Give Prophylaxis or Not to Give, That is the Question

Category:

Ankle; Other

Introduction/Purpose:

The risk of venous thromboembolism (VTE) after foot and ankle surgery is significantly lower than rates after hip/knee arthroplasty, but it isn't zero. Specific subgroups of patients may be at higher risk, forcing patients and clinicians to navigate the risks and benefits of chemoprophylaxis with insufficient data. Efforts have been made to add clarity to such decision making using conventional data-analysis and risk-scoring methods, but none of these methods were patient-specific or built on robust models of a given patient's individual characteristics. In this study we used machine-learning to determine the potential risk factors for VTE after ankle fracture. We aimed to develop a patient-specific predictive model to assist clinicians and patients in deciding upon the use of postoperative chemoprophylaxis after foot and ankle surgery.

Methods:

In this preliminary machine-learning-based case-control study, 16,421 patients with ankle fractures were recruited retrospectively. We used an automated-string search method to find patients who were clinically suspected to have developed VTE. Among 1176 such patients, 239 had confirmed VTE within 180 days of sustaining an ankle fracture (cases) and 937 did not (controls). Groups were further sub-divided in patients who had been receiving chemoprophylaxis and those who hadn't. Over 110 factors and variables including patient demographics, past-medical and surgical history, fracture characteristics, treatment, medications, and laboratory values were included in our machine-learning dataset. Three analytical algorithms were used in our machine-learning methods including backward-logistic-regression, decision-tree-classifier (depth=5), and neural networks (two dense layers (n=16 and 4), two drop-out layers, and a sigmoid classifier). Conventional statistics were also used to compare the case and control groups (chi-squared, t-test, p<0.05 considered significant), and the odds-ratio (OR) was calculated for significant parameters.

Results:

Based on overall performance scores including specificity, sensitivity, area under the ROC curve, accuracy, PPV, NPV, F- 1 score, among the 3 machine-learning methods, the Backward-Logistic-Regression model was superior in predicting VTE post ankle fracture and in determining whether administering prophylaxis can be beneficial for the patient or not (Tables 1 and 2). Other than the previously suggested risk factors, our algorithms showed a positive correlation between the incidence of VTE and smoking (OR=2.09, p<0.001), age <55 y/o (p=0.001), open fracture (OR=2.49, p<0.001), male sex (OR=1.98, p<0.001), surgical versus nonoperative treatment (OR=1.88, p=0.001), and multiple fractures at the time of trauma (OR=1.9, p=0.001). Factors that showed negative correlation with VTE include statins use (OR=0.36, p<0.001), hypertension (OR=0.53, p=0.001), vitamin D (OR=0.43, 0.002), calcium supplementation (OR= 0.43, p= 0.01), hyperlipidemia (OR=0.55, p=0.006), cataract (OR=0.19, p=0.01), osteoporosis (OR=0.36, p=0.02), cardiovascular diseases (OR=0.54, p=0.02), hypokalemia (OR=0.26, p=0.03), and proton pump inhibitor use (OR=0.5, p=0.03).

Conclusion:

Our machine learning algorithms showed that factors such as tobacco use, younger age, open fracture, multi-trauma, operative treatment, as well as male sex heightened the risk of VTE. In contrast, certain factors such as vitamin D supplementation had negative correlation with VTE. Machine learning algorithms acted in a more complex manner and incorporated more factors in decision-making compared to conventional methods. External validation using larger and more granular datasets as well as using the algorithms in trial modes (shadow modes) are needed to build trust in this algorithm to assist clinicians in predicting/preventing VTE after foot and ankle surgeries.

Natural language processing for prediction of readmission in posterior lumbar fusion patients: which free-text notes have the most utility?

BACKGROUND CONTEXT

The increasing volume of free-text notes available in electronic health records has created an opportunity for natural language processing (NLP) algorithms to mine this unstructured data in order to detect and predict adverse outcomes. Given the volume and diversity of documentation available in spine surgery, it remains unclear which types of documentation offer the greatest value for prediction of adverse outcomes.

STUDY DESIGN/SETTING

Retrospective review of medical records at two academic and three community hospitals.

PURPOSE

The purpose of this study was to conduct an exploratory analysis in order to examine the utility of free-text notes generated during the index hospitalization for lumbar spine fusion for prediction of 90-day unplanned readmission.

PATIENT SAMPLE

Adult patients 18 years or older undergoing lumbar spine fusion for lumbar spondylolisthesis or lumbar spinal stenosis between January 1, 2016 and December 31, 2020.

OUTCOME MEASURES

The primary outcome was inpatient admission within 90-days of discharge from the index hospitalization.

METHODS

The predictive performance of NLP algorithms developed by using discharge summary notes, operative notes, nursing notes, physical therapy notes, case management notes, medical doctor (MD) (resident or attending), and allied practice professional (APP) (nurse practitioner or physician assistant) notes were assessed by discrimination, calibration, overall performance.

RESULTS

Overall, 708 patients were included in the study and 83 (11.7%) had 90-day inpatient readmission. In the independent testing set of patients (n=141) not used for model development, the area under the receiver operating curve of NLP algorithms for prediction of 90-day readmission using discharge summary notes, operative notes, nursing notes, physical therapy notes, case management notes, MD/APP notes was 0.70, 0.57, 0.57, 0.60, 0.60, and 0.49 respectively.

CONCLUSION

In this exploratory analysis, discharge summary, physical therapy, and case management notes had the most utility and daily MD/APP progress notes had the least utility for prediction of 90-day inpatient readmission in lumbar fusion patients among the free-text documentation generated during the index hospitalization.