Mixed-Methods Researcher |
Veterans Health Administration |
Ph.D., RAND Graduate School
Summative Evaluation
of Costs and Utilization
Background and Motivation
Why Focus on Seriously Ill Patients?
Seriously ill patients have multiple medical, behavioral, and socially complex needs and are at high risk for hospitalization or death.
​
The healthcare system is interested in improving outcomes while also reducing costs: the 5% highest cost patients account for ~1/2 of the healthcare system's total spending.
Piloted an intensive care coordination program for seriously ill patients in primary care setting.
Constraints and Challenges
Research In Real World Conditions
Project originally conceptualized as a randomized controlled trial.
But low participation and no significant decrease in acute care utilization lead to altering enrollment criteria under real-world evidence (RWE).
Challenges:
1. Evaluate project as implemented in real-world conditions without randomization.
2. Evaluation accommodate 100+ outcomes on costs & utilization.
Project Aim
Quasi-Experimental Design Mimics Randomized Control Trial Conditions
Evaluate effects of healthcare costs and utilization by comparing high-risk patients referred intensive care management services to propensity-score matched high-risk patients who were at the same facility.
My Role
My Role: Lead Quantitative Researcher
My role: “full-stack” data scientist: from data retrieval, wrangling, modeling & evaluation:
1. Lead evaluator – develop propensity-score matched models
2. Evaluate the effects on costs and utilization, using a differences-in-differences
3. Led team feedback sessions (build consensus)
Serve as the
4. 1st-author on deliverables to stakeholders
5. 2nd-author on the published manuscript
Methods
Predictors for Propensity Score
Propensity score: the probability of being referred to intensive care coordination.
Choose ~500 propensity-matched patients among a pool of ~15,000
Statistically equivalent on:
Demographics: age, gender, race/ethnicity. marital status, disability, insurance, urban residence, distance to nearest PC clinic
Social vulnerabilities: next of kin, history of falls
Baseline utilization: primary care, emergency room visits, and hospitalization
Comorbidities: 27 conditions, ranging from anxiety, substance use, frailty, hypertension, and stoke.
Propensity-matched patients:
​
Not statistically different (i.e. p-value > 0.05) on the 40 characteristics - demographics, social vulnerabilities, baseline utilization & comorbidities
The only difference between the referred patients and their propensity-matched patients is their referral status.
Essentially, we are mimicking randomization in a real-world conditions.
Methods
Advantage of Propensity-Score Matching: Evaluate Many Outcomes
Insights and Deliverables
Insights
-Nothing is ever for free-must weigh trade-offs.
-Cost-shifting in utilization and costs.
-Higher emergency department and inpatient healthcare utilization for patients referred to coordination program.
Deliverables
-Publication in peer-reviewed medical journal.
-Slide deck presentation of insights to leadership and key stakeholders.
-Comprehensive written reports to funders.
-Poster presentation at national conferences.
Impacts and Lessons
Study findings
-Inform program scale-up aimed at reducing costs while maintaining positive patient experiences.
​​
-Inform the development of approaches to care for seriously ill patients.
​