Connecting the Dots: From Diagnosis to Delivery of Care

Whenever Columbia radiologist Stella Kang opens a scan, the first thing she looks at is the patient’s name and date of birth.

It’s something she learned from one of her mentors during her radiology residency. “The information is practical,” says Kang, professor of radiology at the Vagelos College of Physicians and Surgeons, “but it’s also to ingrain the idea that you’re meeting a patient and seeing their story. As radiologists, we are often the first physicians to encounter a patient’s specific diagnoses and understand what challenges they are facing.”

Kang joined Columbia at the end of 2024 to serve as vice chair of clinical research for the Department of Radiology. Her own interest in the stories that are contained in a patient’s data has expanded beyond imaging to include the masses of data contained in patients’ electronic health record, clinical trials, and epidemiological studies.

Kang is also director of Columbia’s new Center for Advanced Diagnostic Research (CADRe)—launching this June—which focuses on developing ways to take advantage of this information as an integrated part of advanced care pathways.

“We’ve recently seen an explosion of data in genetics, epidemiology, pathology, diagnostic imaging, and other areas,” Kang says. “But all this data usually isn’t integrated in a meaningful way to guide patient care.”

With her center, Kang aims to leverage patient data to better meet the needs of patients who are in different states of health and need to explore a broadened spectrum of therapeutic options.

We spoke to Kang recently about her interest in diagnostic imaging, translating new diagnostics for precision medicine to the bedside, and the goals of CADRe.

You’re in the midst of launching CADRe, Columbia’s Center for Advanced Diagnostic Research. What do you hope to accomplish with this center?

The overarching goal of CADRe is to advance diagnostics, both the innovation of new diagnostics and the translation of emerging diagnostics to the clinic.

The idea is to support translational work that breaks down barriers between traditionally siloed specialties so that we can better synthesize and make use of diagnostic information. There are tremendous opportunities for precision medicine if we can leverage the new forms of diagnostic data that we now have at our disposal to create better diagnostic capabilities. Equally important is to bridge innovative new imaging techniques to the personalized treatment decisions that should follow.

Another goal is to take existing tools, such as AI algorithms, and see if we can get even more out of them in particular settings and patient populations to characterize disease and treatment response. So the performance of a tool or test may vary overall but in a particular subgroup it performs in an outstanding way. We want to identify populations that would benefit from a particular test or testing regimen, and tailor it to that subgroup. In this era of molecular targeted therapies, we also want to identify the patients who are most suited to certain therapies.

We’re partnering with the Herbert Irving Comprehensive Cancer Center, neurology, pathology, and researchers across multiple basic science labs at Columbia. The center will be organized around opportunities we see to advance new diagnostics into the clinic. One avenue would clearly be through clinical trials, so that would mean finding opportunities to take a new biomarker, for example, or a new diagnostic imaging technique, and put it in the direct decision-making pathway for precision treatments.

There’s so much new knowledge about treatments and disease subtypes, and there are opportunities for us to look at ways that new MRI sequences, or new radiopharmaceuticals, might yield information that we haven’t been able to access before and that can guide treatment decisions.

What can the CUIMC community expect to see from CADRe over the next year?

One of the things we’re currently working on are tools for early detection and radiology/pathology correlation for many purposes such as quality improvement or health care delivery tools.

Another project that my own lab on mathematical modeling is working on is identifying new cancer screening and surveillance paradigms that take into account the newest cancer treatments. We use in silico trial design and simulated trials. These disease simulation models factor in the capabilities of the screening or surveillance technology, the timing and frequency of testing, and the effectiveness of treatments. This helps us identify strategies that have the potential to lead to better quality of life and life expectancy for patients, or more cost-effective approaches.

Your own research is focused on diagnostics, but also on the treatment that follows a diagnosis. Why is it important to connect the two?

Looking at processes of care delivery bridges radiology with each of the important steps that follow, such as selection of the right treatment. It’s best realized as a chain of events. By evaluating which diagnostics fit best and where in that chain, we can begin to iterate on improving processes to get the most out of timely and technically optimized testing.

For example, patients are increasingly aware of diagnostics, through public interest in artificial intelligence and through direct access to their own radiology images and reports. We need to be thinking about how treatment decisions can better involve patients’ understanding of their own diagnosis. And so we know that many people respond to having their information presented in pictures. A patient who is told they have an incidentally found kidney tumor will have a very different understanding of what that means if they are able to see how big the tumor is. A doctor will likely discuss how the tumor’s size and location might affect the recommended treatment and whether that means the patient would lose none of their kidney, part of their kidney, or their entire kidney. Using pictures in that discussion can really help a patient understand what they are going to be going through. They can help patients participate in a discussion with more lucidity.

Disease simulation modeling is another innovation to help patients make decisions after their diagnosis. Disease simulation is a process of mathematically representing a decision at hand and where the best decision lies probabilistically. It involves building a mathematical model of disease natural history. We create a structure that represents the progression of patients’ risks of developing a particular disease or set of diseases, and then we input the transitions to the different health states. This simulation can occur at multiple scales. Then we can layer on events that might alter the course of natural history and measure the projected differences according to testing and treatment decisions and the effectiveness of different population strategies.

To go back to the kidney tumor example, my group is involved in the development of a web-based tool for patients with kidney cancer who are trying to decide on a course of treatment. There’s an analytic engine behind the tool so that when patients enter their information—age, tumor characteristics, comorbidities—they get a quantitative comparison of risks and potential benefits. The tool isn’t making a specific recommendation, but it’s giving patients and their physicians information about their options based on the patient’s condition, patient preferences, and the latest research on treatments and prognosis.

The models offer additional possible uses in terms of analysis and perspectives—and these methods are used in evidence-based medicine, economic evaluation, policy development, and risk assessment, and quality of life analyses—to make informed choices about clinical care and health policy. We can also use them to help guide clinical trial designs, for example identifying markers of futility or projection of outcomes based on efficacy and adherence.

What interests you about radiology?

Radiology offers a direct view of structural and functional pathophysiology—a truly unique perspective on the spectrum of normal and diseased tissues. The images and derived metrics thus provide opportunity to study disease and guide therapeutic decisions. The field is constantly evolving. Now, leaps in computing power and innovation in functional and quantitative imaging at different scales provide a tremendous opportunity to characterize an individual’s health. When we combine these technical capabilities with advancements in therapy, the path to early detection and individualized care can be realized. It’s critical that we properly assess these new technologies so that health systems can actualize benefits.