CHDA logo with Harvard shield. Logo is a half gear and lines in blue and yellow shaped like a CThe Center for Healthcare Data Analytics (CHDA) was established in 2016 by the faculty and staff of the Department of Health Care Policy after a realization that our current servers could not accommodate the data analytics on large public or private data sets. We also needed a more secure environment to acquire and process over a decade worth of Centers for Medicare and Medicaid Services (CMS) data files that would form the foundation for additional research with CMS data.

The Center's core faculty members are nationally recognized for their work in data analytics, Bayesian methodological approaches, statistical methods for health services, and regulatory policy research. Our research includes broad topics on financing and delivery of health care, quality of care, studies on special and disadvantaged populations (including those with mental disorders), and access to care. In addition, we manage large-scale efforts related to benefit design and the care of the elderly. In many cases, these applied studies are augmented by fundamental contributions in statistics and biostatistics.

Expertise

The Center has a rich pool of statisticians and programmers from the Department of Health Care Policy (HCP) with many years of experience working with healthcare claims and survey data. We have four full-time PhD-trained statisticians/biostatisticians whose expertise covers a broad spectrum of statistical methods from regression analysis and simulations to machine learning and big data analysis. Their research portfolios cover the strategic methodological areas necessary for making scientifically valid conclusions on the basis of large, complex routinely collected databases. Among these, three dimensions are of particular relevance for big data: causal inference, predictive inference, and multi-leveled inference. These areas are critical for discovery: running fast algorithms for data visualization and information retrieval are important, but inferring cause and effect, identifying what treatments work for whom, and determining new areas of investigations require careful, particular expertise.  

Our statistical faculty is at the forefront of many of these larger methodological endeavors. Our research success has benefited over the years from a rich collection of public and private administrative and clinical data and integral relationships between HCP faculty members and government policy makers. Faculty expertise in medicine, economics, and statistics allows us to approach complicated problems related to cost, quality and access in a rigorous, unbiased, and timely fashion. Interactions with researchers in other schools at Harvard or in other universities expands our expertise and skill sets.

Support Staff and Environment

In addition, we support over 30 programmers, analysts, and data managers that work with our statisticians and research Principal Investigators in HCP. Many of our team have 20 or more years of experience analyzing healthcare data and code in multiple languages, including SAS, R, Stata, Python, SPSS, WinBUGS, and MPlus. Their expertise covers a broad spectrum of statistical methods from regression analysis and simulations to machine learning and big data analysis.

Our research staff benefit from a high-performance Harvard level 4 certified computational environment, which includes 3 compute servers, one with GPUs, and nearly 2 petabytes of storage. This environment is also certified by the Centers for Medicare and Medicaid Services (CMS) to utilize their data on-premises at the Harvard Medical School. 

Having the CMS certification allows the department to fund a keystone CMS data use agreement (DUA) that includes mostly 100% Medicare Fee-for-Service, Medcare Advantage, and Medicaid cohorts from 2008 to present, with 20% cohorts for Carrier and Part D event data files. The keystone DUA allows HCP researchers to request reuse of these files at a dramatically reduced cost, free for dissertations, for approved research protocols.