Text

theses

Background: When delivering evidence-based care at the bedside, learning with anchors is a proven method of efficiently learning statistically driven patient phenotypes. Such libraries currently learn with anchor terms that are universally suspect in their ability to support evidence-based practice for physical therapists (PT) within electronic medical records (EMRs).

Methods: A definition of anchor terms for venous thromboembolism (VTE) was developed using structured and unstructured retrospective data from PT documentation within two separate EMRs. The learned PT specific VTE phenotype anchor terms were compared against the published PT clinical practice guideline and clinician documentation for consistency. The learned PT specific VTE phenotype anchor terms were then evaluated against the published learned anchors derived from physician-based documentation.

Results: Two of the top 25 anchors showed a statistically significant correlation with the presence of VTE: ‘vessel’ (P <0.001) and ‘pe’ (P < 0.05). The top 20% most frequently appearing learned anchor terms in descending order of total observed frequency was ‘boot’ (12.2%), ‘movement’ (10.4%), ‘develop’ (10.1%), ‘cad’ (9.5%) and ‘pulmonary’ (9.2%).
Discussion: This research provides new insight into the relationship between anchor terms and the documentation of PT. The data indicate that the top 20% of discovered physical therapy derived phenotype terms for VTE anchors did not match the existing physician derived phenotype definition for VTE. Based on the existing physician derived anchor terms, clinical decision support tools for VTE would not have been triggered if used by the PT.

Conclusion: The delivery of patient-centered care requires an interdisciplinary team of clinicians to achieve optimal patient outcomes. Evidenced-based practice is enhanced through the presence of clinical decision support tools in the clinical workflow of the modern healthcare system. Before this research, there did not exist an established set of anchor terms with a likelihood of detecting the presence of VTE within the profession of physical therapy. An initial listing of such anchor variables has now been discovered. Further research is needed to expand the ability of machine learning classifiers to identify patients both at risk and with active disease.

ETD_10415

Ph.D.

Includes bibliographical references

doi:10.7282/t3-942s-x435

ETD doctoral

The author owns the copyright to this work.