Page 51 - ISMCON souvenir 2021

ISMSCON - 2021

OS10: Modelling disease susceptibility with a linear re-
parameterized model of Multiple group latent class analysis

Ankita Dey, Diganta Mukherjee, Sugata Sen Roy
Affiliation of presenting author:
Ms. Ankita Dey,
Statistician, Biostatistics section,
National Institute of Tuberculosis and Respiratory Diseases,
Sri Aurobindo Marg,
New Delhi – 110030
Email: ankitadey14@gmail.com
Keywords: Linear Reparameterization, Multiple group latent class analysis, Maximum likelihood, Risk
factors, Cardiac events.
The present study describes a unique approach of latent class modelling in data related to risk factors
of a disease by developing a new model of linear re-parameterized multiple group latent class analysis
and demonstrates the model in the context of a data set consisting of several risk factors related to
cardiac events. A new set of parameters, reflecting a patient’s item response pattern and latent classes,
are incorporated in the model along with the parameters of latent class size. The latent class sizes are
also assumed to vary between observed classes of the population. The present study uses a secondary
data set on patients referred to the UCLA Adult Cardiac Imaging and Hemodynamics Laboratories for
Dobutamine stress echocardiography (DSE) between March 1991 and March 1996. Variables on risk
factors of cardiac events viz. gender, history of hypertension, history of diabetes, history of smoking
are taken to be the observed or manifest variables for the analysis. Age of the patients are used to
stratify the population into two observed classes. The multiple group latent class model with a linear
re-parameterization achieves parsimony i.e. latent classes may be identified with a lower number of
parameters. The likelihood function is maximized with respect to all the parameters to obtain the maximum
likelihood estimates (MLE) of the parameters and results are compared for a traditional model (baseline
model) and the proposed extended model.

OS11: VALIDATION OF DIETARY ASSESSMENT METHOD

Ann Maria Moncy1, Tinku Thomas2, Dr. Sumathi Swaminathan3, Smitha Joseph4, Sivakami
Sundari S5
1,2,4,5Division of Epidemiology and Biostatistics, 3Division of Nutrition, St. John’s Research Institute, Bengaluru.
Email id: annmoncy01@gmail.com
Background: It is well recognized that a nutritionally-balanced diet is critical for ensuring normal growth
and development in children. Nutritional assessment is thus an integral part of optimal paediatric care. 24-
hour dietary recall methods and Household Consumption and Expenditure Surveys (HCES) are helpful
for estimating individualized dietary intakes. But the validation of the dietary intake estimates using these
different methods were not given much importance. To address these gaps in knowledge, we compared
estimates of individualized dietary intakes from household data to 24-hr recall dietary estimates among
the children in the rural areas of Bihar.
Objective: To validate the Household Consumption and Expenditure Survey (HCES) using the 24 hour
dietary recall data containing the nutrient intake of children below five years of age.
Methods: Correlation coefficients, scatter diagrams and simple linear regression were used to find
the relation between nutrient intakes from 24 hr recall data and household survey data. The nutrients
considered were protein, fat, carbohydrate, energy, thiamine, riboflavin, niacin, vitamin B6, total folate,
calcium, iron and phosphorous. Bland Altman plots were used to check the agreement between the two
methods. The agreement in quartiles of nutrient intake from HCES and 24 hour recall data examined

CONFERENCE SOUVENIR 49