Center for Operations Research in Medicine and Healthcare

Advancing knowledge, empowering decisions, transforming organizations

Research Overview

Research projects in the center focus on the design of novel mathematical modeling and advanced theory and computational algorithms with a primary emphasis on applications to bio/medical and healthcare problems. Specific applications include large-scale bio/medical informatics, modeling, and computing for:
  • Genomic analysis
  • Predictive health, medical prediction and diagnosis
  • Optimal treatment design and drug delivery
  • Early detection, target intervention, monitoring and controlling of disease
  • Public health and medical preparedness, emergency response and disaster medicine
  • Health informatics, quality and operations advances in healthcare delivery systems
The collaborative research program of Operations Research in Medicine and HealthCare at Georgia Tech is the first of its kind in the operations research community. Students and faculty in the center work closely with medical and healthcare researchers on each research project We would like to acknowledge research support by the National Science Foundation, the Whitaker Foundation, the Centers for Disease Control and Preventation, and the National Institutes of Health.



Novel pattern recognition and classification algorithms for health risk analysis, early disease diagnosis and prediction, target therapeutic intervention, and disease monitoring

In this area, we focus on design and development of novel predictive models and algorithms that are suitable for developing predictive rules for large heterogeneous biological/clinical/healthcare data sets. Our general-purpose system simultaneously incorporates:
  • The ability to classify any number of distinct groups
  • The ability to incorporate heterogeneous types of attributes as input
  • A high-dimensional data transformation that minimizes noise and errors in medical and biological data
  • The ability to incorporate constraints to limit the rate of misclassification between groups, and a reserved-judgment
  • region that provides a safeguard against over-training
  • Successive multi-stage classification capability to handle data points placed in the reserved judgment region
The study facilitates predictive health (health risk prediction); early disease diagnosis, intervention, and prognosis; design of novel treatments; and treatment outcome prediction. Models have been successful in predicting early stage atherosclerosis, cardiovascular risk, human cancer diagnosis; novel ultrasonic-assisted drug delivery; tumor shape identification in cancer treatment; predicting the immunity of vaccines; early diagnosis of diabetes, pre-mature aging, cognitive impairment (that will advance to Alzheimer's disease), macular degeneracy, tumor metastasis, and nutritional deficiency in humans.

Selected highlights:



Analysis of clinical treatment modalities and design of individualized (patient-centered) optimal treatment regimens

This research focuses on optimal treatment design and delivery for various types of diseases. Models and computational engines are designed to study and investigate: Selected highlights:



Public health and medical preparedness, emergency response and disaster medicine, critical infrastructure defense

This research effort include various projects involving publich health informatics and large-scale response effort for all hazard events, including biological, chemical, radiological events, as well as infectious disease outbreaks. Much of this effort is joint with the Centers for Disease Control and Prevention and hundreds of local/state public health departments. Topics include: Selected highlights:



Clinical outcome analysis, comparative effectiveness, and development of prediction rules for treatment effectiveness, and design of improved treatment regimens

This research focuses on outcome prediction and comparative effectiveness. The key lies on development of predictive rules that correlates staging of diseases versus demographics versus treatment modalities versus observed clinical outcome. This allows for subtype of individuals and the design/selection of best treatment modality for personalized medicine.
Selected highlights:
  • Long-term results of angioplasty versus stenting in cardiac transplant recipients with allograft vasculopathy
  • MDCT of Thoraco-Abdominal Trauma: An Evaluation of the Success and Limitations of Primary Interpretation
  • Using Multiplanar Reformatted Images versus Axial Images
  • Vaccine outcome prediction



Biological modeling and large-scale computing for multi-modality optimal cancer treatment design

MRS can identify regions within tumors (e.g., prostate tumors) that have denser populations of tumor cells.
  • NMR studies have indicated that choline is elevated in rapidly growing tissues such as tumors
  • Choline metabolism may be related to tumor proliferation.
  • 1H MRS can be used to image location of prostate cancer in the gland: differentiates cancer from benign tissue by
  • the ratio [Cho+Cr]/Cit (Cho=choline, Cr=creatine, Cit=citrate) of the respective peaks in the MR spectrum. The ratios are calculated on a spatial grid covering the prostate tissue.
  • Tumor control probability (TCP) is largely determined by the response of fast-proliferating, radioresistant cells in
  • the tumor.
  • Goal: Incorporate biologic tumor information into planning procedures (e.g., brachytherapy, IMRT) in the treatment
  • ofprostate carcinoma
  • Conjecture: improved TCP may be obtained by emphasizing not only the clinical prescribed dose but also appropriate
  • escalated dose within the prostate.
  • Challenges: morphing of various imaging information into treatment images, design of optimal treatment plans.
MRS data for a patient with Gleason score 7, PSA= 8 ng/ml. The spectral voxels correspond to the grid overlaid on the image.

Example of dose escalation around the tumor region.




Large-scale cancer modeling and biocomputing intensity-modulated radiation therapy

To produce IMRT treatment plans, sophisticated optimization methods must be used for designing the optimal beam geometry, fluence pattern and delivery parameters. This research explores novel optimization and biological models, and large-scale parallel algorithms for real-time clinically relevant plan design. Furthermore, issues related to organ motion and continuous tumor shrinkage are investigated.

For an overall discription of IMRT process, refer to The IMRT Process: From Planning to Delivery presented by Dr. Margie Hunt at ASTRO IMRT Practicum 2004




Genomic pattern recognition and prediction of epigenetic silencing phenomenon in human cancer

Aberrant methylation of normally unmethylated CpG islands occurs frequently in human cancers andleads to inappropriate gene silencing. Methylation-mediated silencing plays a role in progression of human caners by inactivating genes thought to suppress invasion/metastasis. TMS-1 has been shown to be silenced by aberrant methylation of CpG islands in breast carcinomas. This silencing results in the loss of TMS-1 promoting activity in apoptosis

This project focuses on developing computational models and algorithms, based on artificial intelligence and machine learning techniques, to understand genomic structures of CpG islands and their role in epigenetic cancer. Specific goals include:
  • Explore importance of CpG island methylation in cancers at the genomics level
  • Develop predictive rules for methylation status associated with TMS-1 in human cancer
  • Potential for reactivating genes being silenced by reversing DNA methylation, thus providing an exciting molecular
  • target for chemotherapeutic intervention
  • Develop novel treatment strategies aimed at blocking or reversing methylation status
  • Develop methylation markers for cancer prediction, treatment and prognosis





Novel disease prediction via microvascular networks

Recent development of fluorescence micro- angiography offers a cost-effective, imely technique for imaging the functional microvasculature in tissues and in multiple organs with superb resolution

Challenges: Quantitative analysis of large amounts of complex capillary patterns that may allow the identification of discriminatory patterns to provide fingerprints distinguishing between normal and perturbed microvascular perfusion.

Goals:
  • To develop automated fingerprinting algorithms through large-scale imaging analysis followed by predictive
  • modeling for microvascular networks that could be used to investigate the potential perturbing effects of conditions such as cardiovascular disease, aging, genetic deficiency, diabetes, and cancer on the microvascular structure in relevant tissues
  • "Angioprinting" will contribute to the understanding of angiogenic mechanisms, and could potentially be utilized
  • in the early diagnosis of microvascular deficiencies, as well as for monitoring and allowing for design of a better therapeutic regimen
The study offers novel approaches for early disease diagnosis and intervention, treatment monitoring and prognosis, and therapeutic models.




Evolutionary models, DNA sequencing and genomic analysis

Goals: To develop efficient mathematical models and robust algorithms for large-scale sequence analysis problems arising in computational genomics, protein structural analysis, and evolutionary biology. Specifically, key issues under investigation involve designing models and algorithms that are:


Decision support system for mass dispensing for pandemics and bioterror attacks

Goal: To develop a computerized decision support system to help U.S. state, city and county healthcare departments create and test more efficient plans for emergency response logistics and treatment of infectious illness, whether it's a natural or man-made outbreak. The software, 'RealOpt©' , allows for operational and strategic planning and offers the capability to:
  • Design customized and efficient clinic/dispensing facility floor plans for regional needs via an automatic graph-
  • drawing tool.
  • Determine optimal labor resources required and provides the most-efficient placement of staff at individual
  • stations within the facility. The results maximize the number of people who can be treated, minimize the average time patients spend in the clinic, and equalize utilization across clinic stations.
  • Determine the best location for emergency clinics based on population density and road accessibility, the most
  • efficient facility layout, the number of health care professionals needed in certain areas, the number of medical countermeasures and supplies (e.g. vaccinations, antibiotics) needed and the time it will take to treat patients.
  • Process data in real time as the emergency treatment occurs, and determine dynamic changes when necessary.
  • Perform disease propagation analysis and derive dynamic response strategies to mitigate casualties.
  • Provide a framework for emergency healthcare managers to assess current resources and determine minimum
  • needs to prepare for readiness in emergency situations for their regional population.
  • Allow large-scale virtual drills and performance analysis, and assist in the study, training and enhancement of
  • emergency response, planning and treatment from terrorism, infectious disease outbreaks, and natural disasters.
In collaboration with the Centers for Disease Control and Prevention, the development and testing of RealOpt© began in 2002. Since 2004, RealOpt© has been used for planning of Anthrax emergency exercises, and actual vaccinations for seasonal flu, and for Hepatitis A booster shots. The system currently includes clinic layouts for emergencies involving Anthrax, Flu pandemic, Hepatitis, and Smallpox. It is used by thousands of public health and emergency coordinators across the 50 states for infectious disease planning and biodefense exercise drills.

Request RealOpt© Usage





Comprehensive analysis of patient triage in urban pediatric emergency departments

The Emergency Severity Index (ESI) system is a patient triage model recently adopted at Children's Healthcare of Atlanta (CHOA) for implementation in two urban pediatric emergency departments (ED) starting in August 2006. The decision to use ESI is primarily due to the current four-tier model's poor reproducibility, poor correlation to resource utilization, and tendency to misclassify both critical and non-critical patients.

To be able to provide quality and timely care to pediatric ED patients, it is critical to investigate effective patient triage methods as well as improve resource utilization. This urgency is compounded by the planned CHOA ED expansion in each of its two campuses, Scottish Rite and Egleston. This project brings together researchers at Georgia Institute of Technology and CHOA to investigate the efficacy of Emergency Severity Index (ESI) triage in determining resource utilization, and short- and long-term staffing needs while maintaining a high quality of care.

Primary aims of this project include:
  1. Compare immediate performance outcomes of the conventional four-tier triage versus the ESI system
  2. Evaluate long-term ESI performance outcomes
  3. Analyze how proper triage categorization affects staffing projections and resource utilization/optimization
  4. Develop an operational planning tool to forecast optimal staffing, appropriate resource utilization, and future ED expansion needs
The broad impact of this proposed research includes:
  1. Improvement of correct triage categorization of patients upon admission into the emergency care service
  2. Decrease in the number of patients who leave without being seen by a physician due to incorrect "fast-track" categorization and inefficient resource utilization
  3. Improvement in timeliness of care, in directing proper care, and thus in overall quality-of-care for pediatric ED patients
  4. Improvement of hospital resource utilization and cost-effectiveness of staff and resources assigned to patient care
  5. Improvement of operations efficiency related to billing