IDENTIFYING CLINICAL VARIATION USING MACHINE INTELLIGENCE: A PILOT IN COLORECTAL SURGERY
CLOUD COMPUTING & BIG DATA
Author: Piyush Mathur
Coauthor(s): Kamal Maheshwari,MD,MPH
Status: Completed Work
Standardized clinical pathways are useful in reducing clinical variation and improving quality of care. Utilizing organization specific data, machine intelligence can help discover clinical variation and create valid clinical pathways for specific populations at a fraction of the cost and time of traditional methods. We report our experience with the use of machine intelligence software to identify practice patterns associated with better outcomes and lower costs in colorectal surgery patients.
In this quality improvement project to identify and manage clinical variation, we analyzed the inpatient care of 1,786 patients undergoing colorectal surgery from 2015-2016 across the Cleveland Clinic System. Data from multiple information subsystems was loaded in the Ayasdi Clinical Variation Management (CVM) application (Ayasdi, Inc., Menlo Park, CA). The Ayasdi CVM software application utilizes machine learning algorithms to enable unsupervised learning from multidimensional data sets.
Our general approach was to identify high-level trends in the “All patients” cohort of 1,786 patients and then pursue these findings in sub-cohorts to see if these associations remained. The software auto-generated 9 distinct groups of patients based on similarity analysis (Figure 1).
Overall, ketorolac use was associated with lower LOS and cost per case in the all-patients groups and in multiple sub-cohorts (Figure2). We found a strong correlation between lower intra-operative fluid use (<2000 cc) and lower length of stay and lower cost groups. We found significantly higher rates of readmission among inflammatory bowel disease patients compared to colon cancer. Discussion We learned that this technology has several advantages over traditional analytic approaches: 1) Analysis across disparate data sets – The machine intelligence software enabled joining of disparate data sets at the patient level for ease of analysis. 2) Unsupervised discovery – The software algorithms rapidly identified patients who had received similar care across numerous hospitals and identified the differentiating treatment pathways for each group of patients. 3) Speed and auto-generation of clinical pathways – We rapidly created a consensus clinical pathway from our best performing subgroups of patients. 4) Ease of use by team members – The software allows easy comparison of groups and sub-cohorts of patients across a wide range of categorical and continuous variables. 5) Adherence reporting – The system creates detailed reports at both the physician and hospital levels comparing performance against a selected consensus pathway Conclusion Using near real-time clinical data, novel methods like machine learning can help identify and monitor clinical variation. Additionally, the interventions associated with better outcomes can be identified improving quality of care. References 1. Schrijvers G, van Hoorn A, Huiskes N. The care pathway: concepts and theories: an introduction. Int J Integr Care. 2012;12(Spec Ed Integrated Care Pathways):e192. 2. Li L, Cheng WY, Glicksberg BS, et al. Identification of type 2 diabetes subgroups through topological analysis of patient similarity. Sci Transl Med. 2015;7(311):311ra174. 3. Gunnar Carlsson PhD, Francis X Campion, MD FACP. Machine Intelligence for Healthcare Vol 1: CreateSpace Independent Publishing Platform; 2017.