Page 55 - ISMCON souvenir 2021

ISMSCON - 2021

taken based on these cut-offs which are based on any outcome such as morbidity or mortality. Developing
a new cut-off based on risk of death as an outcome can give better classification of malnutrition.
Aims and objective:

Create a graded classification of malnutrition based on the risk of death as an outcome
Methods:
Pooled data from 5 interventional and non-interventional studies across India were used for the analysis.
A hierarchical Bayesian Spline regression models have been used to find the association of height and
weight measurements with a latent risk of death derived from the morbidities experienced by the child with
the assumption that the association changes for different risk groups. The knots in the spline regression
are used to identify the risk groups.
Results:

Risk of death is negatively associated with anthropometric measures. There is decreased risk of death
after a particular point of anthropometry measure.

Conclusions :
Current cutoff for malnutrition using -2, -3 Z scores leads to heteroginty in outcomes as the cut offs
are not based on outcomes. A finer categorization of malnutrition based on risk of death can be used to
develop and then deliver tailored optimized therapeutic options for what is essentially a far more electic
group than what is captured by a current classifications of undernutrition.

OS16: COMPETING RISKS IN MULTISTATE MODEL

Bhrigu Kumar Rajbongshia Abhips Tripathy Atanu Bhattacharjee Gajendra K. Vishwakarmaa
b c
a
a
aDepartment of Mathematics & Computing, Indian Institute of Technology Dhanbad, Dhanbad-826004, India, Mail
id: kumarbhrigu536@gmail.com
bSection of Biostatistics, Centre for Cancer Epidemiology, Tata Memorial Centre, Navi Mumbai-410210, India,
cHomi Bhabha National Institute, Mumbai, India.

Competing risk is an event whose occurrence precludes the occurrence of the primary event of interest.
For example, in cancer studies local-regional-control, first progression, distant progression are the
different intermediate stages that an individual passes through till the absorbing state. An individual can
progress to the final stage through a number of transitional stages or can directly move to the absorbing
stage, which is known to be the competing event. Robust estimation techniques are required to handle
data containing censored information to decrease the bias compared to crude analysis of the dataset.
This work aims to use the propensity score method (PSM) to update the censored observation using the
covariates and thereby to apply various techniques for competing risk model to evaluate the hazard rates
and survival function.
In this work, we have used the PSM method and prepared scores for each individual. A threshold
probability is assumed to compare with the distribution of the propensity scores and depending on the
score function the censored information is updated. After updating the dataset stratified Cox PH model is
applied to compare the hazard between both the genders. The proposed method have been applied on
a simulated dataset as well as a chemotherapeutic dataset.

Mean estimated regression coefficients for covariate x2 obtained before and after updation of dataset
using PSM are 0.0144 and 0.0486 respectively for cause 1. Mean square error of the regression coefficient
have reduced while using PSM. Efficacy of PSM method can be clearly seen from cumulative-incidence-
curve for male and female separately.

CONFERENCE SOUVENIR 53