Using Data To Predict Mortality
A new predictive app that culls through medical data can offer a better answer to many cancer patients’ top question: “What’s my prognosis?”
Feb 11, 2019

Using Data To Predict Mortality

How unlocking the power of predictive tools can make a difference

Feb 11, 2019
As Featured In 
Production and Operations Management

Data is powerful. But how can you use it to offer a more-informed prognosis to cancer patients? New research from Maryland Smith’s Margrét Bjarnadóttir offers a roadmap.

In their work, Bjarnadóttir and three co-authors developed a new predictive app that culls medical data to offer a better answer to a cancer patient’s top question: “What’s my prognosis?” It aims to help patients choose the treatment or care that will work best for them, and helps doctors and hospitals make smarter recommendations and treatment plans.

The researchers explored a comprehensive dataset from all colorectal cancer patients in California, choosing that disease because of its prevalence – the third most common cancer in the U.S. – and its varying treatment options and outcomes. Bjarnadóttir and her co-authors used the data to create predictive models to estimate short-term and medium-term survival probabilities for patients based on their clinical and demographic information.

Currently, almost half of colorectal cancer patients receive no prognosis information from their doctors, and the predictions they do receive are often inaccurate or biased, according to the researchers. They contend that using predictive modeling and visualization tools makes prognosis information easier for doctors to present and patients to understand. They say having accurate information about their prognosis helps patients make better decisions about their course of care, and increases their satisfaction with their doctors.

The researchers also say using predictive tools can save health care dollars by reducing the tendency many doctors have to overprescribe and over treat conditions that won’t see any benefit to those actions. Typical colorectal treatments include surgery, chemotherapy and radiation, which have risks and often-severe side effects, but may not improve outcomes for a patient. Hospitals can also use the predictions to anticipate loads on their systems and adjust staffing levels accordingly and make better resource allocations for things like surgical facilities, beds and equipment.

The researchers say their app is a user-friendly tool to help doctors and patients have better conversations about treatment options and outcomes, and a solution that could be applied to many other types of cancer and medical diagnoses. The app takes a patient’s information and displays both predicted survival curves and the actual survival data for the 20 most similar patients in the database to help the patient understand the possible range of outcomes to make better treatment decisions.

The study provides “a roadmap for modeling disease outcomes based on existing data,” the researchers say. “These models may provide an unbiased view of the likely trajectory of a patient's disease, offering patients and providers a base for data‐driven decision‐making that personalizes treatment and improves quality of care.”

Read more: Predicting Colorectal Cancer Mortality: Models to facilitate patient-physician conversations and inform operational decision making is published in Production and Operations Management.

About the Author(s)

Margrét Bjarnadóttir

Dr. Margrét Vilborg Bjarnadóttir, is an Assistant Professor of Management Science and Statistics in the DO&IT group. Dr. Margrét Bjarnadóttir graduated from MIT's Operations Research Center in 2008, defending her thesis titled "Data Driven Approach to Health Care, Application Using Claims Data". Dr. Bjarnadóttir specializes in operations research methods using large scale data. Her work spans applications ranging from analyzing nation-wide cross-ownership patterns and systemic risk in finance to drug surveillance and practice patterns in health care. She has consulted with both health care start-ups on risk modeling using health care data as well as governmental agencies such as a central bank on data-driven fraud detection algorithms.

More in


The Pros and Cons of Open Source
New Maryland Smith research presents an economic model to study how software licensing attributes affect a software maker’s decisions, along with insights for policymakers.
Apr 09, 2019
Why There's An App For That
A new study examines what happens when a platform owner enters the market with a product that directly competes against rivals in the app store. Does it help or hurt innovation?
Mar 05, 2019
A Tool for Closing the Gender Pay Gap
Equal pay for equal work is a simple notion, but surprisingly hard to implement without buy-in from upper management and quantitative tools for decision support.
Feb 28, 2019