By Melissa Jackson | Spring 2021


In the beginning, our cells knit us together with extraordinary precision and timing, following the script outlined in the origin story that is our DNA. No one quite knows what sometimes trips up this process, disrupting a healthy cell’s cycle of growth and division and causing it to silently spin off copies containing deadly inaccuracies.

doctor aiming at targetBut science has found ways to prevent cancerous mutations from becoming the automatic death sentence they were for previous generations. Clinical trials have demonstrated the potential of regular cervical, colorectal, breast and lung cancer screenings to identify many of these abnormalities before symptoms appear. Coupled with treatment advances, early detection can lead to better patient outcomes and lower medical costs.

And yet, convincing the public to take advantage of these life-saving opportunities remains a challenge. Participation rates across the board are below the U.S. Department of Health and Human Services’ Healthy People 2020 targets, with substantial disparities among underserved populations.

For instance, colorectal cancer is the third-leading cause of U.S. cancer deaths. Despite strong evidence that early detection through screenings saves lives, roughly a quarter of people ages 50 to 75 have never been screened. In fall 2020, the U.S. Preventive Services Task Force cast a wider net by lowering the recommended screening age to 45. The hope is that by increasing the eligibility pool, doctors can catch more cases at a highly treatable stage, especially among subgroups with historically low participation rates.

It’s a vexing problem for clinicians. Millions of dollars are spent each year on prevention and education campaigns, and yet patients continue to miss out on the benefits of preventative screenings.

University of Notre Dame marketing researcher Yixing Chen can’t cure cancer, but he believes the tools of his trade can help health care providers shift patient behavior and potentially save lives — while also maximizing the return on outreach campaigns. Chen, an assistant professor of marketing at Mendoza College of Business, studies the social impact of marketing interventions. He taps into the power of big data using applied econometrics and machine learning methods in order to assess and improve the effectiveness of marketing spending in health care, K-12 schools and other nonprofit settings.

Through Dr. Amit Singal, a professor of internal medicine and chief of hepatology at the University of Texas Southwestern Medical Center in Dallas, Chen became aware of the troubling cancer screening participation rates. “He said, ‘We’ve heard marketing can help,’” Chen recalled. He and a team of researchers began working with Singal on a National Cancer Institute-sponsored study to quantify the effectiveness and financial impact of cancer screening outreach programs. What they found led them to develop a machine learning-informed personalized patient outreach strategy.

Their study, “Improving Cancer Outreach Effectiveness through Targeting and Economic Assessments: Insights from a Randomized Field Experiment,” was published in the Journal of Marketing in 2020. 


The team first designed a randomized field experiment to understand the impact of marketing interventions on the screening completion status of patients identified as high risk for hepatocellular carcinoma (HCC), the most common type of liver cancer.

“A very small percentage of people are taking advantage of screening as a preventative measure,” Chen said, explaining the significance of early diagnosis with HCC. Symptoms are rarely noticed until the advanced stage when surgical options are less successful. Statistically, early-stage liver cancer patients who undergo surgery have a five-year survival rate between 60%–70%, while the five-year relative survival rate for liver cancer is 18%. Despite this, fewer than 1-in-5 high-risk patients — those with chronic liver diseases — undergo screening, and the rate is even lower among low-income and racial minority subgroups.

Chen’s team initiated a trio of outreach strategies among a diverse group of 1,800 high-risk HCC patients between December 2014 and March 2017:

  • Group 1 received no additional intervention beyond the usual medical care.
  • Group 2 received a “light touch” — a screening invitation in the mail and additional follow-up by phone.
  • Group 3 received the mailing, phone follow-up with a script and additional contact from a “patient navigator” who helped address barriers to screening.

When the results were tabulated, they found that both Group 2, which received a light-touch approach, and Group 3, which received both the light touch outreach and the more costly patient navigator intervention, increased in screening likelihood to 45% and 49%, respectively, compared with the control group’s 25% screening rate.

With a 4-percentage-point difference in screening rates between the two outreach groups, the lower-cost strategy seems like the way to go — assuming that all high-risk patients will respond similarly. But as every good marketer knows, all customers differ.


Using advanced machine-learning techniques, the researchers were able to drill down into patient compliance based on various characteristics, such as age, gender, neighborhood socioeconomic status, visit history, health status and so on, and map out what outreach strategy worked best. Using an algorithm, they personalized the outreach strategy and assigned each patient to the most suitable intervention.

headshotFor example, they found that Caucasian men with pre-existing conditions who live in less populated neighborhoods and have fewer primary care visits were less responsive to light-touch outreach. But that same strategy works well with Hispanic women who are covered by medical assistance and live near the test site and receive hepatology care.

“We can show that by sorting out how responsive each patient is to each outreach type, we can save the health care system a lot of money,” Chen said. “If you involve a patient navigator, it’s extremely expensive per hour. And so we want to know who are those people who we can just use a light touch to help.”

That’s not to say that the more intensive outreach has no place, he adds. Rather, advanced analytics offers a way to identify those who would benefit most from the higher-cost outreach strategy, enabling the medical center to better allocate resources.

“While modern health care has implemented personalized medicine using genetic information, most health care outreach and educational programs still rely on untailored communications,” they wrote. “Practitioners who manage these programs should recognize that the use of a large number of patient characteristics can substantially improve the outreach responsiveness through a tailored approach.”

The team found that they could improve screening rates by an additional 14% by matching patients with the type of outreach they are more likely to act upon.

The increase in the number of patients completing screenings has financial implications, as well. The researchers did an outreach cost-benefit analysis and calculated a net loss of $840 per patient in the group that received no outreach. The light touch outreach generated a gain of $1,192 per patient and the patient navigator outreach group showed a $1,635 gain per patient. When they extrapolated these gains to the 3,217 eligible patients at the Dallas medical center, they found a $2.1 million total net gain. And when the personalized patient outreach results were calculated, the team found it would further enhance net returns by an additional $1.6 to $2 million.

Chen and his research colleagues, Ju-Yeon Lee of Iowa State University, Hari Sridhar of Texas A&M University, Vikas Mittal of Rice University and Amit Singal and Katharine McCallister of the University of Texas Southwestern Medical Center, urged medical providers to invest in patient-centric marketing and encouraged outreach to underrepresented populations, which are more responsive to such messages.

“Hospitals and health care practitioners should realize that a ‘one-size-fits-all’ outreach program is neither effective nor economic,” they wrote. “The use of machine learning can power data-driven patient-centric outreach programs that are also dynamically adaptive.” Noting, however, the additional resources and training such campaigns would require, they also suggested that policy makers explore ways to incentivize hospitals to do so.

Photo by Barbara Johnston, Illustration by Errata Carmona

YIXING CHEN teaches marketing analytics to business students in the undergraduate, MSBA and MBA programs. He also serves as a faculty affiliate of the Mike and Josie Harper Cancer Research Institute (HCRI) and the Eck Institute for Global Health (EIGH), University of Notre Dame.

Improving Cancer Outreach Effectiveness through Targeting and Economic Assessments: Insights from a Randomized Field Experiment,” Journal of Marketing (May 2020), Yixing Chen, Ju-Yeon Lee, Hari Sridhar, Vikas Mittal, Amit Singal and Katharine McCallister