Enterprise AI Analysis
Evaluation of postoperative bleeding risk after dental extractions in patients on antithrombotic medication: A comparison of machine learning and clinical experience
This study compared machine learning (ML) algorithms with an experienced oral surgeon's decisions for predicting postoperative bleeding after dental extractions in patients on antithrombotic medication. The ML algorithms outperformed the surgeon in balanced accuracy, with KNN achieving the highest. While surgeons tend to overestimate bleeding risk, ML algorithms provide an objective assessment, helping identify high-risk cases and guiding postoperative observation periods. The study highlights the potential of AI as a complementary tool in complex medical decision-making.
Executive Impact & Key Metrics
Our analysis reveals the following critical performance indicators, demonstrating the potential for significant advancements:
0
Incidence of Hemorrhagic Events
0
ML Outperformance (Balanced Accuracy)
0
False Positive Reduction (XGB vs. Surgeon)
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
A growing share of patients is taking antithrombotic medication, and the impact of the various medications on the risk of postoperative hemorrhage after oral surgery is sometimes difficult to anticipate, even for experienced surgeons. Identifying bleeding risks is highly complex due to a broad range of influencing factors like osteotomies, number of teeth removed, and specific medications. Clinical experience alone often leads to overcaution, as prior studies showed many preventively admitted patients did not bleed.
Algorithms based on machine learning (ML) and neural networks are increasingly used in dentistry, providing objective decisions based on trained data, potentially surpassing human experience and clinical heuristics. Various ML algorithms like Logistic Regression (LR), Random Forest (RF), eXtreme Gradient Boosting (XGB), and K-nearest neighbors (KNN) can be applied as prediction models for binary dependent variables like bleeding risk. A sufficient sample size is necessary for training these models, especially when the required effect (bleeding risk) is low.
This retrospective clinical study included 2000 dental extraction procedures from patients on antiplatelet or anticoagulant therapy, spanning January 2014 to August 2024. Data collected included patient demographics, medication type (monotherapy, dual, triple), number of teeth extracted, method of removal, surgical region, and incidence of postoperative bleeding. A standardized local hemostatic protocol was followed, which likely contributed to the low overall bleeding incidence (4.35%). The data was split into 80% for training and 20% for testing the ML algorithms and an experienced oral surgeon.
62%
KNN's Highest Balanced Accuracy, Outperforming All Other Algorithms and the Surgeon
Machine Learning Workflow for Bleeding Prediction
Collect Dental Extraction Data (n=2000)
→
Split Data into Training (80%) & Test (20%) Sets
→
Train LR, RF, XGB, KNN Algorithms on Training Set
→
Algorithms Evaluate Test Set
→
Senior Surgeon Assesses Test Set
→
Compare Algorithm & Surgeon Predictions
| Metric | Senior Surgeon | Best ML Algorithm (KNN) |
|---|---|---|
| Balanced Accuracy | 53% | 62% |
| Sensitivity (Identified Bleeding Cases) | 5 out of 17 (29.4%) | 9 out of 17 (52.9%) |
| False Positive Predictions | 92 | 114 |
| Overall Accuracy | 74% | 70% |
| While KNN had more false positives, its higher sensitivity means it identified more actual bleeding incidents, making it safer in a medical context where preventive overcaution is preferred over underestimation. | ||
AI in Action: Improved Risk Assessment in Oral Surgery
Context: A university hospital faces challenges in accurately predicting postoperative bleeding for patients on antithrombotic medication, often leading to overcautious inpatient admissions and extended observation periods.
Challenge: Existing methods, including experienced surgeons' judgments and traditional scores like HAS-BLED, are insufficient to precisely identify high-risk individuals, resulting in inefficient resource allocation and patient inconvenience.
Solution: Implementation of a machine learning model, specifically K-Nearest Neighbors (KNN), trained on a large dataset of dental extraction procedures, to objectively assess bleeding risk based on various patient and procedural parameters.
Result: The KNN model demonstrated a 62% balanced accuracy, outperforming human surgeons (53%) and identifying 23.5% more actual bleeding incidents. This allows for more targeted resource allocation, reducing unnecessary inpatient stays and ensuring better patient safety through data-driven risk stratification.
Impact: The AI model serves as a valuable complementary tool, supporting less experienced surgeons and providing objective guidance for postoperative care planning, optimizing both efficiency and patient outcomes.
Calculate Your Potential ROI
Estimate the financial and operational impact AI could have on your enterprise by adjusting key parameters.
Estimated Annual Savings
$0
Annual Hours Reclaimed
0
Your AI Implementation Roadmap
A strategic, phased approach ensures successful integration and maximum impact. Here’s a typical journey:
Phase 1: Data Integration & Model Training
Consolidate existing patient and procedural data into a centralized, anonymized dataset. Train initial ML models (LR, RF, XGB, KNN) on historical dental extraction outcomes.
Phase 2: Pilot Deployment & Validation
Integrate the highest-performing ML model (KNN) into a pilot clinical decision support system. Conduct a prospective validation study to compare its predictions with clinical outcomes and refine the model based on real-world feedback.
Phase 3: Full-Scale Integration & Monitoring
Fully deploy the validated ML model across all relevant oral surgery departments. Establish a continuous monitoring system to track model performance, identify potential drifts, and retrain the model with new data periodically.
Phase 4: Advanced Feature Development & Scalability
Explore incorporating additional variables (e.g., incision type, preoperative inflammation, surgeon experience) and developing multi-modal AI capabilities (e.g., integrating imaging data) to further enhance predictive accuracy and expand to other surgical specialties.
Ready to Transform Your Operations?
Discover how customized AI solutions can drive efficiency and innovation in your enterprise. Book a free consultation today.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat