Environmental Statistics Seminar

Edgar Castro
PhD Candidate, HSPH

Naeem Khoshnevis – Software Engineer
Pycausalgps – Matching on generalized propensity scores with continuous exposures (Python Package – Phase I: Design Proposal

Riccardo Cadei – Visiting Master’s Student
Tutorial on Causal Rule Ensemble R Package: Interpretable Inference of Heterogenous Treatment Effects

Presenter: Sarika Aggarwal (PhD Student)
Title: Associations of Floods with Zipcode-Level Hospitalization Rates in the United States

Presenter #1: Ellen Considine – PhD Student
Title: Can Reinforcement Learning Improve Strategies for Issuing Heat Alerts?

Presenter #2: Kevin Lee Chen – PhD Student
Title: Estimating Heterogenous Causal Effects Under Bipartite Network Interference: An Application to Air Pollution Regulatory Policy

Presenter #1: Michael Cork – PhD Student
Title: Curving Emissions – Comparing Methods for Evaluating Response Curves

Presenter #2: Sofia Vega – PhD Student
Title: Expanding Bayesian Methods to Estimate Causal Effects of Environmental Exposures on Childhood Cancers

Kevin Josey

TITLE: Causal Inference in an Ecological Regression Setting
 
ABSTRACT:  Ecological regression is an important tool for analyzing relationships with aggregate data, especially in environmental health studies where only group-level data, rather than individual-level data, are available. By combining ecological regression and causal inference, researchers can overcome many limitations inherent with traditional ecological studies, and produce more reliable and valid results. However, there is limited understanding on how to perform a causal analysis with aggregate data. In this presentation, I will demonstrate how to use aggregate data to estimate causal effects, specifically the exposure response curve relating annual average fine particulate matter (PM2.5) with mortality, enabling inferences on the individual-level and further overcoming issues with ecological bias. We cover minor differences in the indentifying assumptions required to obtain individual-level parameter estimates as opposed to group-level parameter estimates supposing that the exposure assignment is determined at the group-level. We compare the group-level causal estimates in a simulation study and in a real world data-analysis examining the association PM2.5 with mortality

Jenny Lee

TITLE: Air Pollution Epidemiology using Medicaid Claims: Causal Exposure-Response Function Estimation in Clustered Data with Surrogate Measures 

ABSTRACT: The methods development of this project is motivated by a case study where we are interested in estimating the causal exposure-response function (ERF) in the context of long-term exposure to (PM$_{2.5}$) and respiratory hospitalizations in socioeconomically disadvantaged children enrolled in the Medicaid program. We need to develop new methods to overcome the following challenges: 1) socioeconomic information, which is known to be a confounder and an effect modifier, is not available for the Medicaid population under study; 2) data are clustered, that is zip codes (our unit of the analysis) are nested within states; and 3) we have available two useful surrogates: the median household income at the zip code of residence of each Medicare enrollee and the state-level Medicaid family income eligibility thresholds for children. In this paper, we introduce a customized approach that leverages these two surrogates for addressing our motivating question. More specifically, we introduce \textit{MedMatch} that builds on generalized propensity score (GPS) matching methods for causal inference in the context of clustered data. We conduct simulation studies to evaluate the performance of our proposed method. We apply our method to estimate the average causal ERF between long-term PM$_{2.5}$ and first respiratory hospitalization among low-income children in Medicaid from 2000 to 2012.

Moderator: Ellen Considine

Kate Hu
Research Associate, Department of Biostatistics, HSPH

Robbie Parks
Post-Doctoral Research Fellow