Observations from the Real World

A new statistical method provides a way to better approximate lab settings when analyzing data.
November 7, 2013 | Research Feature
Print this page

In an ideal world, researchers investigate cause and effect relationships in the controlled environment of a laboratory, carefully assigning subjects to specific interventions or treatments and then analyzing the results. In reality, however, most experiments fall short of this benchmark, because creating such an environment would be impractical, unethical, or simply too expensive. This is often the case in business, medicine, and social science. Topics such as the effect of ad campaigns on sales or the relationship between prisons and crime are usually approached using observational rather than experimental data.

Research by Professor José Zubizarreta, who has a background in statistics and an interest in health research, offers a new method to analyze cause-effect relationships. His method provides a better way to approximate the structure of a lab experiment than current standard methods, allowing researchers to conduct fine-grained adjustments of variables that could confound results, approximating a lab experiment.

Zubizarreta used this method to address a healthcare question: Are patients who are both elderly and obese at a greater risk of acute kidney injury following surgery? In 2008, more than 30 percent of Americans older than 60 years were obese; research has returned conflicting results about the association between the elderly obese and post-opereative kidney injury. Working with medical doctors Rachel R. Kelz, Caroline E. Reinke, and Jeffrey H. Silber of the University of Pennsylvania, he performed a case-controlled study of more than 500 patients who experienced renal failure after a hip or knee replacement or thoracic or colon surgery. Using the new method, the researchers matched these patients with control subjects similar in operation type, age, sex, race, and a variety of medical factors. The study, which also relied on data from Medicare claims and chart reviews, included patients from three states who were treated at 47 different hospitals. Zubizarreta’s sample balancing method ensured that for each of these hospitals, patients who experienced an acute kidney injury were matched extremely closely on all of the demographic characteristics and medical factors to patients who were part of the control group.

The study showed that after surgery, obese patients had a 65 percent greater chance of experiencing an acute kidney injury within 30 days of their initial hospital admission. (While obesity is known to be an independent risk factor for chronic kidney disease, the relationship between excess weight and postoperative acute kidney injury has been unclear.) Given the prevalence of obesity among elderly patients, the authors recommend that hospitals increase their efforts to monitor the kidney function of at-risk groups following surgery. Acute kidney injury is associated with prolonged hospital stays, hospital readmissions, and lower long-term survival rates.

For Zubizarreta, the study achieved one of the primary goals that links most of his research. “I have always wanted to use statistics, mathematics, and computation to address questions of social relevance,” he says. “And in healthcare, there are so many of these issues.” In a separate project, he is applying his new method to a study on the effects of C-sections on premature infants with very low birth weights. He has also started a project that seeks to determine whether schools should be allowed to operate as for-profit businesses; this project is studying the relationship between a school’s for-profit or non-profit status and the performance of its students on standardized tests, using data from his home country of Chile.

Zubizarreta’s method can be applied to almost any topic within the realm of business, economics, and public policy. “In operations research, there has been a surge of interest in using these kind of methods to evaluate interventions,” he says. “By making this statistical contribution, we hope to provide a way of looking at some of the most important questions facing countries around the world.”

José Zubizarreta is professor of decision, risk, and operations at Columbia Business School.

Jose Zubizarreta

Professor Zubizarreta teaches statistics to MBA and PhD students. From a methodological perspective his research focuses on new statistical methods for causal inference (impact evaluation) in randomized experiments and observational studies. From a substantive perspective, he is very interested in using these methods to address questions in health care and public policy. Besides the Division of Decision, Risk and Operations, Zubizarreta is also an...

View full profilePersonal Website

Read the Research

Jose Zubizarreta

"Using Mixed Integer Programming for Matching in an Observational Study of Kidney Failure after Surgery"


Download PDF View abstract/citation