QUEUEING APPLICATIONS IN HEALTHCARE
|
New Drug Approval
The U.S. Food and Drug Administration (FDA) decides whether to approve or reject new drug applications based on statistical evidence of safety and efficacy determined through clinical trials. Despite undergoing rigorous evaluation, some FDA-approved drugs were subsequently found to be ineffective or harmful. In 2004, Vioxx was withdrawn from markets due to safety concerns, after more than 160,000 patients suffered heart attacks or strokes and 38,000 died. The tension between providing sick patients with potentially beneficial remedies, while protecting consumers from adverse events, plays an important role in the FDA's decision-making, yet the agency typically applies a 2.5% significance threshold across all diseases.
In collaboration with Fernanda Bravo and Taylor Corcoran, we develop a queuing model of the drug approval process, starting from development through evaluation, approval or rejection, and market exit. Each stage is modeled as a queue in a network, with arrival, abandonment, and service rates, and number of servers varying by disease. The objective is to maximize total expected benefits less the costs of committing type I or II errors. We find that diseases with low research and development (R&D) intensity—resulting from low initial investment in novel drug formulation, lengthy clinical trials, or high rates of attrition—should have less stringent approval standards to increase the number of drugs available to patients. Given high degrees of obsolescence (newer drugs replace older compounds) or substitution among available treatments, approval standards should be stricter as the market cannot support a large number of similar therapies. Using publicly available clinical trial data, we estimate model parameters and calculate the optimal significance levels for drugs targeting HIV, hypertension, or breast cancer.
In collaboration with Fernanda Bravo and Taylor Corcoran, we develop a queuing model of the drug approval process, starting from development through evaluation, approval or rejection, and market exit. Each stage is modeled as a queue in a network, with arrival, abandonment, and service rates, and number of servers varying by disease. The objective is to maximize total expected benefits less the costs of committing type I or II errors. We find that diseases with low research and development (R&D) intensity—resulting from low initial investment in novel drug formulation, lengthy clinical trials, or high rates of attrition—should have less stringent approval standards to increase the number of drugs available to patients. Given high degrees of obsolescence (newer drugs replace older compounds) or substitution among available treatments, approval standards should be stricter as the market cannot support a large number of similar therapies. Using publicly available clinical trial data, we estimate model parameters and calculate the optimal significance levels for drugs targeting HIV, hypertension, or breast cancer.
Flexible Drug Approval Policies
Fernanda Bravo, Taylor C. Corcoran, Elisa F. Long
Working paper, 2019
Fernanda Bravo, Taylor C. Corcoran, Elisa F. Long
Working paper, 2019
- First place, 2018 INFORMS Public Sector OR Paper Competition
- Third place, 2019 POMS Healthcare Paper Competition
- Finalist, 2018 Pierskalla Best Paper Award
- Finalist, 2018 M&SOM Student Paper Competition
Hospital Operations
I am interested in understanding the relationship between operational metrics (e.g., wait time for admission, discharge delays) and patient outcomes (e.g., mortality, hospital disposition, readmission) in the intensive care unit (ICU), which has the highest mortality rate in the hospital and costs the U.S. health care system more than $100 billion annually. Many hospitals experience chronic ICU bed shortages that are exacerbated by poor coordination with the emergency department and wards.
Using 2 years of patient-level data on ICU length-of-stay (LOS) and hourly bed occupancy at 2 major academic medical centers, we empirically examine the relationship between bed occupancy levels and LOS, 30-day readmissions, and mortality rates. Unlike prior studies, we split LOS into a medically-necessary "service time" and a discretionary "boarding time", during which a patient is deemed clinically ready for transfer to a lower level of care but physically remains in the ICU, potentially due to a lack of available ward beds. One main finding is that boarding time is accelerated during periods of high ICU occupancy and, conversely, prolonged when the receiving (ward) unit faces bed shortages, after controlling for patient covariates including severity-of-illness, primary diagnoses, age, insurance etc.
Using 2 years of patient-level data on ICU length-of-stay (LOS) and hourly bed occupancy at 2 major academic medical centers, we empirically examine the relationship between bed occupancy levels and LOS, 30-day readmissions, and mortality rates. Unlike prior studies, we split LOS into a medically-necessary "service time" and a discretionary "boarding time", during which a patient is deemed clinically ready for transfer to a lower level of care but physically remains in the ICU, potentially due to a lack of available ward beds. One main finding is that boarding time is accelerated during periods of high ICU occupancy and, conversely, prolonged when the receiving (ward) unit faces bed shortages, after controlling for patient covariates including severity-of-illness, primary diagnoses, age, insurance etc.
The Boarding Patient: Effects of ICU and Hospital Occupancy Surges on Patient Flow
Elisa F. Long, Kusum S. Mathews
Production & Operations Management, 2018, 27(12):2122-2143
Elisa F. Long, Kusum S. Mathews
Production & Operations Management, 2018, 27(12):2122-2143
Hospital managers decide how to assign ICU beds among waiting patients and choose among different triage policies (e.g., priority-based, FIFO). We use our empirical results to formulate a priority queuing model with exogenous service times but endogenous (state-dependent) boarding times, and we simulate a queue of multi-class patients arriving to an ICU with multiple server types. We compare policies, such as achieving 1-6 hour boarding times or building 4 additional ICU beds. Our simulation results closely mimic actual wait times collated by the hospital. This modeling framework could be used by other hospitals who wish to compare expansion or other policies.
A Conceptual Framework for Improving Critical Care Patient Flow and Bed Use
Kusum S. Mathews, Elisa F. Long
Annals of the American Thoracic Society, 2015, 12(6):886-894
Kusum S. Mathews, Elisa F. Long
Annals of the American Thoracic Society, 2015, 12(6):886-894