Published: 08-20-2020

Shari Messinger Cayetano, Ph.D, and J. Sunil Rao, Ph.D., co-directors of the University of Miami Clinical & Translational Science Institute (CTSI) Biostatistics, Epidemiology and Research Design (BERD) program, have spent their careers developing health-related statistical models, applying statistical expertise to clinical and translational research, and drawing important conclusions from data.

Now, they have leading roles in COVID-19 research — driving greater understanding of virus prevalence, treatments and outcomes.

Getting a Handle on COVID-19 Data

Gathering the right data to make inferences and predictions has been challenging throughout the COVID-19 pandemic.

“This novel disease has exposed a lot of things that still need to be studied,” says. Dr. Rao, who is also professor and director of the Division of Biostatistics at the Miller School of Medicine.

For his part, Dr. Rao is focused on the fundamentals while also creating new tools in the fight against COVID-19 and future pandemics.

Estimating Prevalence

One critical piece of information that nearly everyone seeks is COVID-19 prevalence. Knowing the total number of cases at any given time would help states as they try to reopen and health systems as they plan for patient volumes and supplies.

The best way to estimate the prevalence of a virus is through widespread testing. But, says Dr. Rao, inherent in testing data is a certain amount of bias. This bias is created through many factors, including geography and exclusion criteria (rules that might prevent someone from being tested and entered into the sample).

Because bias contaminates inferences you might make from the data, Dr. Rao and colleague Daniel Diaz, Ph.D., research assistant professor of biostatistics, set out to create a way to “bias correct” COVID-19 testing data. Applying another bias correction solution — one used for publication bias — to testing data, Dr. Rao and Dr. Diaz have demonstrated dramatic differences in disease prevalence estimates. A paper outlining their work is now under review and their efforts have caught the attention of a large life sciences company. The company plans to use the solution for testing bias correction in an upcoming COVID-19 testing study.

Perfecting Predictions

Predictive models are also key to pandemic planning. But projections, says Dr. Rao, are notoriously difficult to do.

“If you look closely at these projection models — and other models attempting to make predictions outside the range of training data — you’ll see that the prediction intervals or standard error bars are extremely wide and take up most of y-axis of the graph,” says Dr. Rao. “I wanted to know if I could do something statistically to improve model projections.”

It turns out that, in some situations, significant improvements can be made. But in order to do so, Dr. Rao and his Ph.D. student Mengying Li had to develop a new statistical construct.

They developed what he calls a “predictive sponge” — something that can absorb the unknown parts of predictive models to reduce uncertainty. For instance, when one tries to project into the future, it’s highly likely the underlying model might change, and this is not accounted for using current techniques. Models based on an underlying mechanistic framework can do better, but even they fail badly when trying to model things like human behavior changes, which are notoriously difficult to “pin down.”

A paper describing their work together with a theoretical justification of the ideas has just been completed and submitted for review.

Dr. Rao’s work on this new statistical construct was driven not just by COVID-19, but also by a family member’s cancer diagnosis.

“I got involved in discussions about this person’s care and realized their particular situation put them outside the range of available data we were seeing for treatment outcome predictions,” says Dr. Rao. “I wanted to create something that could fill in the holes so we could make more informed decisions from the data.”

Understanding Disparity Drivers of COVID-19

Accurate real-time prevalence data and predictive models take us further in our understanding of COVID-19, but these data don’t necessarily explain why certain people — specifically communities of color — are impacted differently by the virus.

Hoping to better understand outcomes disparities for COVID-19, Dr. Rao went on a hunt for good data. Specifically, data with a level of granularity that would allow enough probing into these issues. A former colleague pointed him to West Virginia University (WVU), where a team has developed a robust COVID-19 patient registry.

“The degree of detail WVU includes is far greater at this point than other such registries I have seen,” says Dr. Rao.

He’s now finalizing a collaboration agreement with WVU that will allow him to apply a new statistical method he’s developed — called Disparity Driver Analysis — to the registry and draw conclusions. He will be working with BERD member Alejandro Mantero, Ph.D., and biostatistics doctoral student Hang Zhang.

“I want to understand the degree of the problem and then rank-order the drivers of such disparities,” says Dr. Rao. “Perhaps then, strategies could be developed to mitigate what is happening.”

Data Scientists at the ‘Bedside’

While Dr. Rao looks for novel biostatistical solutions to solve health problems, Dr. Messinger Cayetano uses biostatistics to find and test new therapies.

She is partnering with long-time University of Miami research collaborator Camillo Ricordi, M.D. to study the use of mesenchymal stem cells (MSC) as a treatment for COVID-19-associated lung inflammation.

An integral part of the research team, Dr. Messinger Cayetano drove early decisions around study design. She guided the group to a stratified block randomization approach, which ensures patients enter the study split evenly between treatment groups and according to severity of symptoms. Importantly, this design also allows the research team to account for ever-changing COVID-19 treatment standards.

Though small (the trial is enrolling just 24 patients through mid-August), this particular study, says Dr. Messinger Cayetano, has come with a sense of urgency and level of team science she’s not experienced before.

In twice-per-day meetings, research colleagues have turned to Dr. Messinger Cayetano to help maintain study integrity, troubleshoot study design challenges and keep the team focused on study endpoints.

“This is the first time in my career that I’ve felt decisions I’m making right now will affect patients tomorrow,” says Dr. Messinger Cayetano, who also is professor of biostatistics and director of the Biostatistics Collaboration and Consulting Core at the University of Miami Miller School of Medicine.

Also making this research experience unique for Dr. Messinger Cayetano is the direct connection she’s had to patients throughout the course of the study.

“I’ve been on Zoom calls with patients in the ICU during the study consent process,” she says. “This brings a level of involvement I’ve not experienced before. This is not a part of research that most statisticians will ever be a part of.”

The Interconnectedness of Research and Statistics

Dr. Messinger Cayetano believes that statistical expertise should be woven into the fabric of every research project.

“In many cases, a biostatistician isn’t brought in from the beginning,” she says. “When I’m brought in midway or at the end of a project, I’m really doing ‘patchwork.’ This isn’t always ideal in terms of scientific rigor.”
Dr. Ricordi agrees.

“All of my studies have been done with biostatistical support from Dr. Messinger Cayetano,” Dr. Ricordi says. “I would not take the risk to spend time, funding and resources to perform a clinical trial to then realize halfway through, or at the end, that the research strategy and protocol development were inadequate and invalidate the credibility or validity of the results.”

Other Biostatistical COVID-19 Research

There are many other faculty members within the Division of Biostatistics in the Department of Public Health Sciences also doing important COVID-19 related research. Get more information on these projects.

Can the BERD Team Help Your Research?

The BERD program of the Miami CTSI advises, educates, and trains researchers in the design and analysis of clinical translational research studies and develops new methods for data analysis.

The program offers expert consultation and research support through the Biostatistics Collaboration and Consulting Core and educational opportunities through workshops, roundtables and graduate programs to train the next generation of clinical researchers and biostatistical scientists.

If you are in need of biostatistical support for your research, visit the Biostatistics Collaboration and Consulting Core’s webpage.




If you would like to receive emails about the latest Miami CTSI events, funding opportunities, services and news, sign up to be on our mailing list.