Page 30 - ISMCON souvenir 2021

ISMSCON - 2021

New and Improved Meta-Analysis

Approaches to Identify Risk Factors

of Diseases

Debashree Ray
PhD, MStat
Assistant Professor

Johns Hopkins University, USA

ABSTRACT
With the rise of large collaborative studies of human diseases and traits, analyses are commonly done
by pooling cohort-level results (summary statistics) using meta-analysis techniques. A pooled analysis of
individual-level raw data from all cohorts (mega-analysis) in modern observational epidemiologic studies
and genome-wide association studies (GWAS) is often impossible due to onerous computational needs
of big data, and other logistical, ethical and privacy concerns. The type of meta-analysis technique used
is primarily driven by the research question and the challenges posed by the data. In this talk, I will
describe two different novel meta-analysis approaches– one for any epidemiologic study and the other
for GWAS– which may be applied by any researcher using publicly available software (https://github.
com/RayDebashree/).

A practical challenge in meta-analyzing studies is that important confounders are likely not measured
across all cohorts since each cohort may have been independently funded with independent study
protocols. Some may report adjusted estimates while others report unadjusted estimates, and there is
no consensus on how to synthesize these estimates. There exist naïve solutions such as meta-analyzing
only unadjusted estimates, or only adjusted estimates, or do both and qualitatively assess conclusion
from each. We have proposed CIMBAL, a practical yet valid statistical method for meta-analyzing
independently sampled cohorts/studies with imbalance in measurement of confounders. It imputes
the adjusted estimates for studies with missing confounders and provides a meta-analyzed adjusted
estimate that appropriately accounts for the dependence between estimates arising due to borrowing of
information across studies. I will discuss assumptions behind CIMBAL and illustrate its performance in
both simulated and real data.

Meta-analysis approaches are routinely employed in genetic epidemiologic studies due to easier access
to summary statistics than raw data. An important research question in GWAS is to identify and study
shared genetic basis of human diseases. The phenomenon of one genetic region influencing risk of two or
more human diseases/traits, known as pleiotropy, is increasingly being observed. Pleiotropy provides new
opportunities, as well as challenges, for diagnosis, therapeutics, and intervention on diseases. We have
proposed PLACO (pleiotropic analysis under composite null hypothesis) that can be applied to summary
statistics available from GWAS of two traits and can account for potential correlation across traits, such
as that arising due to shared controls in case-control studies. I will discuss statistical considerations
behind PLACO and illustrate its advantages over other approaches. Finally, I will present exciting findings
from prostate cancer and type 2 diabetes.

28 CONFERENCE SOUVENIR