Page 47 - ISMCON souvenir 2021

ISMSCON - 2021

OS4: Comparison of Conventional and Propensity Score Matched Methods for Reducing
Confounding Effect for Breast Cancer Risk Estimation

Aleyamma Mathew1, Sneha Mary George2, Preethi Sara George1, Beela Sarah Mathew3,
1Cancer Epidemiology & Biostatistics, 2Statistics Trainee, 3Radiation Oncology, Regional Cancer Centre,
Thiruvananthapuram

Abstract
Background: Association between risk factors and cancer may be confounded by several covariates. In a
case-control study, the conventional method is that the confounding factors are included in the regression
model (covariate adjustment method). Propensity score (PS) matching is a statistical tool for minimizing
the baseline difference between cases and controls and to make the groups more homogeneous. Aim of
the study is to determine the risk of lack of physical activity and obesity associated with the development
of breast cancer in a case-control study using conventional and PS matched methods and examined the
advantages and pitfalls of different methods.
Materials & Methods: A total of 467 breast cancer cases reported in Regional Cancer Centre,
Thiruvananthapuram and 563 controls from the community were included. Propensity Score 1:1
matching (PS matching), Inverse Probability of Weighting (IPW), PS as a single covariate in the model
and compared these results with those using classic covariate adjustment and validated these methods
using AIC and BIC criteria.
Results : Age of the patients ranged from 22 to 88 years with mean 52.4 (SD=10.8) years. 34.6%
of cases and 44.7% of controls were pre-menopausal women. The covariates (age, education, and
menopausal status) are significantly associated with risk of breast cancer. After the PS matching, 365
cases and controls were included. The risk factors (lack of physical activity and obesity) were significantly
associated with breast cancer. PS 1:1 matching method showed the least AIC and BIC values and hence
the corresponding ORs and 95% CI’s for the were chosen as the risk estimates.

Conclusion: PS matching appears to be a reliable method, in that it provides excellent covariate balance
in the case-control study. These insights will guide researchers to make wise choices in their use of
Propensity Score methods for selecting controls homogenous to the cases according to the covariates.

OS5: ORDERED THREE CLASS ROC ANALYSIS USING
PARAMETRIC METHOD

R. Amala1, G. Kumarapandiyan2, Rituparna Sen
1Department of Biostatistics, JIPMER, Puducherry, amalar.statistics@gmail.com
2Department of Statistics Madras Christian College, Chennai, India
3 Indian Statistical Institute, Bangalore, India

Abstract
In clinical screening and diagnosis one of the basic steps is to evaluate the biomarkers performance.
Receiver Operating Characteristic Curve (ROC) is a renowned tool which is used to evaluate the
biomarkers performance. Numerous models have been developed to assess the biomarker in case of
two class (normal and abnormal) problem. Now, the interest has been extended to study the ROC curve
for a three-class diagnostic problem (Normal, Suspicious, Abnormal). This paper deals with developing
the three class ROC model based on a parametric method using Rayleigh distribution. The ROC model
under Rayleigh distributional assumption, Volume under the ROC surface (VUS), asymptotic variance
and confidence intervals for an estimated VUS have been discussed and the same is also compared

CONFERENCE SOUVENIR 45