Naihui Zhou

Graduate Research Assistant, Friedberg Lab
Naihui Zhou


Academic Background
B.S., Information and Computing Science, 2013, Beijing Jiaotong University, China
M.S., Statistics, expected 2015, Columbia University

Research Experience
Intern at the New York City Health and Hospitals Corporation, worked on discovering associations between beta blocker usage and hospital admissions, on a population of MetroPlus health plan patients.

Limited research analyzes the effects of beta blockers on respiratory functions of COPD patients. Since beta blockers and beta agonists oppose each other’s effects, patients who need beta agonists are usually excluded from clinical trials meant to prove the safety and efficacy of beta blockers. Our study examines a large group of patients (n=384,002), attempting to understand the effects of beta blockade on hospital admission and cost. We include possible confounding variables to help explain variation. Some literature has treated admission as count data and used log linear models such as Poisson regression. However, serious overdispersion and an excess of zeros were present.

A zero-inflated Poisson model, sometimes called a Bernoulli-Poisson mixture model, was proposed, where first, the probability of a value being zero is estimated, and for those that are not zero (with probability ! ), a Poisson model can be applied. In our research, a zero-inflated negative binomial was a better fit in terms of BIC and Vuong’s test. We included comorbidities, gender, age, socio-economic factors, as well as medication as explanatory variables. A number of comorbidities, such as hypertension, dysrhythmia and diabetes are found to be correlated with increased number of hospital admissions and cost. However, we have found no specific statistical association between increased hospital admission and beta blocker usage for COPD patients, although COPD is often considered a contraindication for beta blockade therapy. There are a lot more to look into in this research, such as better model selection methods, algorithms to estimate the parameters and so on, and I wish to continue this research to tackle these challenges.

The abstract of this paper has been submitted to the Health Policy Statistics section at the 2015 Joint Statistical Meeting. Not many real-life large datasets perfectly meet the assumptions required for certain statistical models, as is the case in this project. The jobs of scientists in statistics are not only applying models on data, but also developing new models that apply to new datasets. The zero-inflated model is a perfect example. Ideas like this are what make bioinformatics a flexible and vibrant discipline. As a statistics student, it would be great for me to join the bioinformatics community and make my own contributions to developing new models and ideas.

Other Interests: mathematical biology - interested in applying mathematical tools such as differential equations, optimization, combinatorics and game theory, to understanding complex biological systems.

Area of Expertise: 
biological data mining