top of page

Projects

As a student and graduate student researcher, I have led and been a team member on projects related to health services and policy research.  I am experienced in completing all aspects of a research project, including literature search, observational study design, grant/proposal writing, data extraction/manipulation (SAS), descriptive and predictive analytics, manuscript writing, and communicating results.

The Relationship Between Postoperative Prescriber Networks and High-Risk Opioid Prescribing

EN_AHPoster.jpg

Principal Investigator

Abstract Introduction Postoperative opioid prescribing is common in the United States and poor coordination of prescriptions may lead to higher opioid doses and increased risk for overdose. Our study aimed to use social network analysis to evaluate the impact of prescriber connections on postoperative high-risk opioid prescriptions, including subgroups with chronic opioid use (COU), serious mental illness (SMI), and substance use disorder (SUD). ​ Methods Using administrative claims, we sampled patients who underwent surgery in 2018. We created patient-prescriber networks utilizing prescription claims and prescriber-prescriber ties via medical claims with shared patients. Network measures included density (proportion of existing ties to possible ties, scale 0-1) and mean prescriber and opioid prescriber tie strength (number of shared patients, scale 0-3).  Measures were categorized into 4 levels for values of 0 and terciles (Levels 0-3). High-risk prescribing was defined as high-dose prescribing (cumulative opioid doses >90 MME daily) and concurrent prescribing (>30 days of concurrent opioid and benzodiazepine prescriptions).  We evaluated associations between prescription drug monitoring program (PDMP) legislation and high-risk prescribing across varying levels of network measures. We used mixed effects logistic regression models and generalized estimating equations models to evaluate associations between high-risk prescribing and network measures. Results Our final study cohort included 53,273 patients (COU n=4,170; SMI n=9,673; SUD n=1,948; SMI & SUD n=1,783). The majority of the sample was white (70.24%), female (67.90%) with a mean age of 54.36 years. Overall, network density and mean prescriber and opioid prescriber tie strength was low (density 0.5, prescriber tie weight 0.10, opioid prescriber tie weight 0.11).  Compared to patients in Level 0, those in Level 1 were more likely, and those in Level 3 were less likely, to have high-risk prescribing.  Exact relationships varied by network measure, patient subgroup and outcome.  In disconnected networks, having prescribers in a strong PDMP state (compared to weak) reduced the odds of high-risk prescribing by 10% (OR 0.90, 95% CI 0.87, 0.93). Discussion Having more and stronger connections between prescribers reduced high-risk prescribing in the year after surgery. Our findings suggest that healthcare networks and policies that enhance communication between prescribers may improve opioid prescribing outcomes.

EN_SBposter.jpg

Hospital Nursing and Patient Outcomes: The Effect of Nurse-Patient Assignments

Analyst - Graduate Student Researcher

The goal of this project was to evaluate associations between shift-level nurse-patient staffing ratios and nursing-sensitive quality outcomes.  As the analyst, I used SAS to:

  1. Clean EHR encounter-level data to establish a study cohort

  2. Format existing data to create input variables for AHRQ QI software

  3. Adapt software from the AHRQ QI bundle and prior research projects to fit existing data and study goals

  4. Provide output datasets and data documentation to the PI

CodingExample2.jpg
CodingExample3.png
Delivery of Behavioral Health Medication Management By Nurse Practitioners Compared to Physicians

Analyst - Graduate Student Researcher

The goal of this project was to evaluate trends in behavioral health (BH) medication management by nurse practitioners and compare their care patterns with other providers.  As the analyst, I used SAS to:

  1. Clean claims data and establish a study cohorts

  2. Assign a primary BH provider to beneficiaries based on claims

  3. Code outcome variables

  4. Present descriptive data on cohorts and outcomes

  5. Consult with statisticians to identify appropriate predictive models

EN_Poster_AH23.jpg

Behavioral Healthcare and Postoperative Opioid-Related Outcomes

Prinicpal Investigator

Abstract Research Objective: In the United States, more than 90% of patients are prescribed opioids following common surgical procedures. Prolonged postoperative opioid use (PPOU) was recently identified as the most common postoperative complication, with individuals with a substance use disorder (SUD) or a serious mental illness (SMI; ie Major Depressive Disorder, Bipolar Disorder, Schizophrenia and other psychotic disorders) at higher risk. The goal of the current research project was to evaluate the role of mental health services in mitigating prolonged postoperative opioid use. Study Design: This is a retrospective cohort study using de-identified insurance claims from Optum Market Clarity Data. The primary outcome of interest is PPOU (the fulfillment of an opioid prescription greater than 90 days after surgery). We identified behavioral health visits in the year prior to the index surgery using HCPCS codes for any counseling or behavioral health assessment visits, or medication management visits with a mental illness as the primary diagnosis. We calculated rates of PPOU using raw percentages, stratified by the presence of a SMI or SUD. We estimated risk of PPOU using binary logistic regression models, controlling for the presence of SMI/SUD and preoperative opioid use. Population Studied: We evaluated claims from commercial and Medicare Advantage beneficiaries undergoing 1 of 14 common minor and major surgical procedures (index surgery; e.g. laparoscopic appendectomy and cholecystectomy, carpal tunnel release, colectomy). We included patients with continuous insurance coverage in the 6 months before and 6 months after the index procedure, and who filled an initial opioid prescription either 30 days before or 14 days after the index procedure. We excluded patients receiving hospice services in the year before or after the index surgery. Principal Findings: Our study cohort included 82,829 beneficiaries, whom were mostly female (70.9%) with a mean age of 49.5 (SD 18.3). Overall, 19.9% of the cohort developed PPOU (37.1% for those with SMI and 51.4% for those with a SUD). For the overall cohort, receiving any behavioral health visit resulted in a slight increase in risk for developing PPOU (OR 1.14, 95% CI 1.06, 1.22; controlling for presence of a SMI, SUD, and preoperative opioid use). Of beneficiaries with a SUD, those that received any behavioral health visit were 35% less likely to develop PPOU (OR 0.65, 95% CI 0.46, 0.92). There was no significant difference in risk for those with a SMI (OR 1.02, 95% CI 0.90, 1.16). Conclusions: PPOU is extremely common following common surgical procedures, especially for individuals with a SUD and a SMI. Those with a SUD are less likely to develop PPOU if they were engaged in behavioral health services prior to surgery. Implications for Policy or Practice: Risk stratification postoperative opioid-related outcomes should not rely solely on SMI or SUD diagnoses. When possible, preoperative engagement with behavioral health services may reduce the risk of PPOU for those with a SUD.

EN_Poster_AH22.jpg

Comparing Methods to Evaluate Risk of Prolonged Postoperative Opioid Use in Patients with Serious Mental Illness

Prinicpal Investigator

Abstract Research Objective: Patients with serious mental illness (SMI) are vulnerable to respiratory depression related to prolonged postoperative opioid use (PPOU; i.e. the fulfillment of an opioid prescription greater than 90 days after surgery) due to increased concomitant use of opioids and anxiolytics. The goal of this study is to estimate the risk for PPOU in patients with SMI by comparing two statistical approaches: propensity score methods (PSM; identifying treatment effects by balancing groups on observed covariates) and endogenous treatment effect modeling (ETEM; identifying treatment effects by controlling for unmeasured confounders through correlating errors of selection and outcome models). Study Design: This is a retrospective cohort study using a 5% sample of 2016-2018 fee-for-service Medicare Claims. The primary outcome of interest is PPOU. We calculated rates of PPOU using raw percentages and risk for PPOU using PSM (inverse probability weighting with regression adjustment) and ETEM. Both approaches assumed a probit model for SMI and poisson distribution for PPOU. Population Studied: We evaluated claims from beneficiaries who underwent 1 of 13 common minor and major surgical procedures (index surgery; e.g. laparoscopic appendectomy and cholecystectomy, carpal tunnel release, colectomy). We included beneficiaries that had continuous coverage (365 days before to 90 days after index) and filled an opioid prescription (30 days before to 14 days after index). We excluded beneficiaries with extended hospital stays (>30 days for index stay or >30 days total in the 90 days after index surgery), additional surgeries (365 days before to 90 days after index surgery), diagnosis of opioid use disorder or chronic pain disorder, who received hospice services before or after index surgery or filled an opioid prescription 31-365 days prior to the index surgery. Principal Findings: Our final study cohort of 7,369 beneficiaries was mostly female (58.3%) and white (83.0%) with a mean age of 71.3 years (SD=10.9). There were 1,944 (26.4%) beneficiaries with a SMI, 5.6% of which developed PPOU (versus 3.9% of the final cohort). We used best-subset variable selection to identify important covariates for SMI and PPOU. Using PSM, patients with SMI were 1.013 times more likely to develop PPOU than those without SMI (p=0.047, 95% CI: 1.000-1.025). There was a high degree of endogeneity (rho=-0.922, 2(1)=128.950, p

AHposter_6.3.jpg
bottom of page