Using Analytics to Diagnose Cancer – Kogod Now
single,single-post,postid-4119,single-format-standard,mkd-core-1.1,kogodbiz-child-ver-1.0.0,optimizewp-ver-1.3.1,mkdf-smooth-scroll,mkdf-smooth-page-transitions,mkdf-ajax,mkdf-blog-installed,mkdf-header-standard,mkdf-sticky-header-on-scroll-up,mkdf-default-mobile-header,mkdf-sticky-up-mobile-header,mkdf-dropdown-default,mkdf-light-header,mkdf-header-style-on-scroll,mkdf-search-dropdown,wpb-js-composer js-comp-ver-4.12,vc_responsive
Kogod Now / Faculty Research  / Using Analytics to Diagnose Cancer

Using Analytics to Diagnose Cancer

In 2014, approximately 233,000 new cases of prostate cancer will be diagnosed in the United States, and about 29,480 men will die of the disease, according to the American Cancer Society. Even with such overwhelming numbers, most men face their annual prostate exam with trepidation.

Unfortunately, not only is prostate cancer’s prevalence spreading, but it’s also hard to diagnose. Professor Edward Wasil and his colleagues are trying to alleviate the anxiety that comes with the prostate check up by applying health care analytics and mathematical models to diagnose the disease using a minimally-invasive method.

A doctor currently has three options to search for prostate cancer: a prostate specific antigen (PSA) blood test, a digital rectal exam, and a trans-rectal extended biopsy.

The latter two options are unpleasant for the patient, and all three have drawbacks. While the biopsy is more accurate, it is highly invasive and expensive. And the PSA test, which came under scrutiny from the US Preventative Services Task Force (USPSTF) recently, has a high false-positive rate and can be inaccurate. The digital rectal exam suffers from high false-positive and negative rates.

Wasil, a faculty member in the Information Technology department, partnered with researchers from the University of Maryland to test the accuracy of diagnosing prostate cancer using a fourth method— magnetic resonance imaging (MRI).

Their work is part of the new world of health care analytics—a buzzword that packs a mathematical punch. It involves using and analyzing data to drive down health care costs and improve quality of care.

“The focus for my research is to show from an academic’s point of view that we can take techniques that we know and apply them to important health care problems,” he said. “Lots of people in my profession are starting to focus on health care analytics.”


A non-invasive imaging technique, the MRI uses magnetic fields and radio waves to generate images of organs and structures in the body. While it is being used in some prostate evaluations, it has not yet become common practice.

Wasil thinks their research models may help clinicians. “There are researchers working on trying to make the determination of a prostate [being] cancerous using MRIs, and our technique would help them.”

The research team set out to test whether MRI screening is more accurate than a biopsy. They worked with a radiologist to assess 223 slices of biopsied, cancerous prostates from 28 patients using an MRI.

The team used the resulting data to build a logistical regression model. This first model performed with 65 percent accuracy—relatively the same as the existing PSA test.

Next, they used a classification technique called the “nearest neighbor” method, where they tried to find other cancerous slices that were closest in proximity and similar to a particular slice (resulting in 77 percent accuracy).

Their third and final model was a combination of the logistical regression and nearest neighbor models, which performed with the highest accu- racy of 79 percent in detecting cancer. All three models used the same data of cancerous prostates, provided by the University of Maryland Medical System.


Since Wasil and his colleagues’ models were using only biopsied slices of cancerous prostates, they now must determine how to build models of the whole prostate in vivo and diagnose whether or not it is cancerous.

“What we’ve shown is we can certainly use MRIs to identify prostate cancer,” he said. “The next step is to build models of the entire prostate.”

Using MRIs to diagnose prostate cancer “is a new area that deserves certainly much more attention,” Wasil said. David Anderson, another author on the study, agreed.

“Commonly used methods now, like the PSA test, only give an estimate of whether there is cancer or not,” Anderson explained. “By using MRIs to screen, we can locate the potential cancer.”

“This information could allow biopsies to be done more precisely, therefore potentially decreasing discomfort and side effects,” Anderson said.

Wasil said MRI screening would be especially effective since the USPSTF advised against screening for prostate cancer using PSA-based tests in a 2012 recommendation.

“There is a very small potential benefit [from the PSA-based screening] and significant potential harms,” the recommendation concluded. “A better test and better treatment options are needed.”

The potential good news for insurance companies is that MRI screenings cost between $500- $700 each, whereas biopsies can cost up to $2,100.

“If you could reduce the number of biopsies by using MRIs, that would lead to huge savings, not only in terms of costs, but in terms of pain. The needle-biopsies are quite painful and have potential side effects,” Wasil said.

While MRIs would be more expensive than a PSA test, which costs about $100, Wasil said the increased accuracy of MRIs would outweigh its costs.

To take their MRI study to the next level, Wasil said he and his colleagues want to gather additional data from an experimental study with patients and combine the results from a PSA test and an MRI screening to make a diagnosis.

Anderson, who is an assistant professor at the City University of New York Baruch College, said he predicts MRI screenings can be added to standard of care for prostate screenings.

“I can see the standard being screening with PSA tests, and then screening any potential positive PSA tests with an MRI,” he said. “Both [can be used] for additional screening and locating the cancerous region before confirming with a biopsy.”

“As of now, MRIs are probably too costly to be used as a broad screening method,” he added.


Wasil said in the past five years, the field of operations research has been working more on health care analytics.

So, in a separate endeavor, Wasil and other colleagues are working on a paper assessing how different applications, studies and operations research tools—like statistical methods and optimization methods—apply to prostate cancer research.

“We want to give researchers a sense of what kinds of problems have been looked at—where there have been success stories, where tools have been useful, and give a nice background as to how each of these tools have been used,” he explained. They’ve done a synopsis of 40 articles so far that were published since 2000 and plan to send the paper out for review in a journal.

“As you can imagine, there are lots of studies published in both the academic literature and medical literature,” he said.

He said they want to get published in a main- stream academic journal that focuses on quantitative modeling.


Wasil said while their three models for prostate cancer diagnosis through MRI may not transfer to other types of cancer research, the use of logistical regression and nearest neighbor models are commonly used throughout health care analytics.

“Hospitals are beginning to see that health care analytics is an important area that they have to work on,” he said. “In many cases they don’t have expertise in house, and they have to work with outside experts to help solve these important problems.”

So as health care analytics grows and more people are covered under the Affordable Care Act, operations researchers will be kept busy in the years ahead, predicted Wasil.

“My sense is that operations research-based models will continue to be used in ever-increasing numbers by health care professionals,” he said. KN mark


Prostate-specific antigen (PSA) is a protein produced only by the prostate. In healthy men, small amounts of PSA in the blood are normal, while elevated levels can indicate the presence of cancer. In the past, a blood PSA test was used as a pre-screening for prostate cancer, but in recent years the accuracy of these tests has come under scrutiny.

Doctors screen a blood sample to determine the level of PSA in the blood.

  • • Four nanograms of PSA per milliliter of blood are considered normal. Anything more than that amount could indicate cancer.
  • • About 15 percent of patients with normal or low PSA levels are false negatives—meaning they do have cancer.
  • • About 75 percent of patients with elevated PSA levels are false positives—meaning they do not have cancer.

In 2011 the US Preventive Services Task Force recommended against using a PSA screening to detect prostate cancer.

Other Factors Linked to HIGH PSA Levels

  • • Age: PSA levels naturally increase as men age
  • • Cycling: Some studies indicate regular bicycling can increase PSA levels
  • • Enlarged prostate due to older age
  • • Bacterial infection
  • • Male hormone replacement therapy

Other Factors Linked to LOW PSA Levels

  • • Obesity
  • • Use of herbal supplements
  • • Use of statins, cholesterol-lowering prescription drugs such as Lipitor or Crestor
  • • Use of aspirin regularly, as in the case of heart disease prevention
  • • Use of Thiazide diuretics, also known as water pills, for treatment of high blood pressure


*   All facts and data from American Cancer Society

“Predicting Prostate Cancer Risk Using Magnetic Resonance Imaging Data,” David Anderson,
Bruce Golden, Edward Wasil and Hao Zhang, was published online in the
Information Systems and e-Business Management Journal in February 2014.

No Comments

Post a Comment