Derivation of a novel perioperative venous thromboembolism risk assessment model using National Surgical Quality Improvement Project data Öffentlichkeit
Mlaver, Eli (Spring 2024)
Published
Abstract
Venous thromboembolic events (VTE) remain a leading preventable cause of morbidity and mortality in postoperative patients despite nation-wide quality improvement efforts. Failure of consistent, standardized risk assessment contributes to these costly and devastating outcomes. Currently available risk assessment models (RAMs) are burdensome and lack procedural specificity or actionable thresholds for intervention; shortcomings which limit their utility within clinical workflows. The development of a parsimonious RAM designed for clinical workflows has the potential to increase adherence to risk assessment and prophylaxis administration, thereby improving the health outcomes of post-operative patients.
We applied multivariable logistic regression modelling with a clinically-guided forward selection process to the 2019 National Surgical Quality Improvement Project Public User File dataset in order to identify variables for inclusion in the new RAM. Procedural specificity was introduced by grouping Current Procedural Terminology (CPT) codes and creating a dichotomous variable capturing minimally invasive techniques. Integer point values were assigned to included variables to derive a VTE RAM. Model performance was compared to three currently available RAMs: the Caprini score, the COBRA model, and the American College of Surgeons (ACS) Risk Calculator.
Eleven variables were chosen for inclusion: age, BMI, functional status, American Society of Anesthesiologists Physical Status classification; history of steroid use, ascites, or cancer; pre-operative sepsis or blood transfusion; CPT group and minimally invasive surgery. The new FAST AS A CLOT model at a cutoff of 5 points has a lower c statistic (0.753) than the previously published c statistic for the ACS Risk Calculator (0.819), but has an 89% sensitivity for VTE outcomes in the NSQIP dataset as compared to 77% for COBRA and 62% for Caprini.
As it was derived with an emphasis on biological plausibility and face validity to clinicians, the FAST AS A CLOT model addresses many of the limitations of currently available RAMs. Validation within clinical workflows is still needed. If adopted, implementation within the electronic health may improve care quality and patient outcomes.
Table of Contents
1. Introduction. 1
2. Methods. 6
2.1 Data Source and Study Population. 6
2.2 Primary Outcome. 8
2.3 Data Cleaning and Establishing Variable Definitions. 8
2.4 Covariates for VTE Risk Modeling. 11
2.5 Univariable and Bivariable Analyses. 11
2.6 Logistic Regression Model Derivation. 12
2.7 Model Performance. 13
2.8 Scoring Model Building. 13
2.9 Secondary Aim – Transfusion Requirement 14
3. Results. 14
3.1 Study Population. 14
3.2 Candidate Selection. 15
3.3 Regression Model Derivation and Performance. 16
3.4 Risk Score Derivation and Performance. 17
3.5 Secondary Aim – Transfusion Requirements. 18
4. Discussion. 19
4.1 Comparison to Available RAMs. 20
4.2 Discussion of Specific Included Variables. 21
4.3 Statistical Idiosyncrasies. 24
4.4 Biological Plausibility and Face Validity. 24
4.5 Study limitations. 26
4.6 Future Research Opportunities. 27
4.7 Implementation. 28
5. Conclusion. 29
6. Tables. 30
7. Appendix: SAS Variable Definitions. 45
8. Disclosure. 48
9. References. 48
About this Master's Thesis
| School | |
|---|---|
| Department | |
| Degree | |
| Submission | |
| Language |
|
| Research Field | |
| Stichwort | |
| Committee Chair / Thesis Advisor | |
| Committee Members |
Primary PDF
| Thumbnail | Title | Date Uploaded | Actions |
|---|---|---|---|
|
|
Derivation of a novel perioperative venous thromboembolism risk assessment model using National Surgical Quality Improvement Project data () | 2024-03-28 11:48:28 -0400 |
|
Supplemental Files
| Thumbnail | Title | Date Uploaded | Actions |
|---|