How to Steer Clear of Statistical Heartbreak

Study design can be make or break for research projects. Poor study design or inappropriate statistical analysis can be reasons that research isn’t funded or accepted for publication. When biases aren’t recognized in time, years of research can be wasted. In one case, a researcher finished collecting data only to realize that a major confounding variable had been overlooked. By then it was too late.

While this kind of statistical miscue is rare, many research projects can benefit from some expert statistical guidance. The Mailman’s Biostatistics Consultation Service does just that. Four or five times a week, they sit down with investigators to formulate abstracts, rework rejected articles, or develop grant proposals. In tandem with the Design and Biostatistics Resource at the Irving Institute for Clinical and Translational Research, the group offers free short-term consultations to faculty across Columbia University Medical Center.

While the Service is open to all researchers, junior investigators who lack the resources and connections of their more senior counterparts may benefit the most. “If there is one piece of advice I can give people, it’s to include a statistician early,” says Shing Lee, director of the Consulting Service and assistant professor of Biostatistics. “If it is a badly designed study, there might be a bias we can’t undo at the end.”

Lee, her co-director Codruta Chiuzan, and biostatisticians Jimmy Duong and Vivian Zhang, meet with clients to learn their research goals then find the right statistical tools, steering clear of pitfalls along the way. “There might be issues researchers don’t consider,” says Lee. For instance, accuracy in data collection is essential to maintaining the integrity of research. And many researchers aren’t aware of newer study designs that can be a better fit to address an investigator’s scientific question, she adds.

While the group can do a lot to fix problems with data collection and design, they can’t do it all. If a research idea or proposal isn’t feasible, they will let their clients know, but they will do all they can to help reformulate the question and re-design the study.

Recently, the Biostatistics group assisted researchers in the Department of Radiation Oncology with statistical analysis so they could investigate the accuracy of preoperative imaging for breast cancer patients as a tool for radiation therapy. In another case, the consultants worked with the Division of Pediatric Cardiology to select the right variables and methodology to evaluate a computer algorithm that uses MRIs to measure blood flow through a patient’s heart. In the latter, the project evolved into a longer-term relationship on a fee-for-service basis.

Their mission goes beyond single projects, however. Lee and Chiuzan’s larger vision is to promote statistical literacy and have more people understand how good study design and analysis can give an original hypothesis wings.

“The motivation is simple,” says Lee. “We want to contribute to improving the science on campus.”