Modelling unbalanced catastrophic health expenditure data by using machine‐learning methods

• Published date: 01 October 2020
• DOI: http://doi.org/10.1002/isaf.1483
• Date: 01 October 2020
• Author: Songul Cinaroglu

RESEARCH ARTICLE

Modelling unbalanced catastrophic health expenditure data by

using machine-learning methods

Songul Cinaroglu

Faculty of Economics & Administrative

Sciences, Health Care Management,

Hacettepe University, Ankara, Turkey

Correspondence

Songul Cinaroglu, Faculty of Economics &

Administrative Sciences, Health Care

Management, Hacettepe University , Ankara,

Turkey.

Email: songulcinaroglu@gmail.com

Summary

This study aims to compare the performances of logistic regression and random

forest classifiers in a balanced oversampling procedure for the prediction of

households that will face catastrophic out-of-pocket (OOP) health expenditure. Data

were derived from the nationally representative household budget survey collected

by the Turkish Statistical Institute for the year 2012. A total of 9,987 households

returned valid surveys. The data set was highly imbalanced, and the percentage of

households facing catastrophic OOP health expenditure was 0.14. Balanced

oversampling was performed, and 30 artificial data sets were generated with sizes of

5% and 98% of the original data size. The balanced oversampled data set provided

accurate predictions, and random forest exhibited superior performance in

identifying households facing catastrophic OOP health expenditure (area under the

receiver operating characteristic curve, AUC = 0.8765; classification accuracy,

CA = 0.7936; sensitivity = 0.7765; specificity = 0.8552; F

1

= 0.7797).

KEYWORDS

catastrophic health expenditure, logistic regression, random forest

1|INTRODUCTION

Out-of-pocket (OOP) health expenditure refers to the payments made

by households at the point they receive health services; the amount

of OOP health expenditure depends mostly on where households

sought care (Van Damme et al., 2004). OOP health expenditure

includes gratuities, any in-kind payments, therapeutic appliances, and

other goods and services purchased with the primary objective of

enhancing or restoring health status. Doctors’consultation fees, medi-

cation purchases, and hospital bills are some examples of OOP health

expenditures. OOP payments include expenditures on alternative

and/or traditional medicine but not expenditures on health-related

transportation and special nutrition (Minh et al., 2013). Fees or

co-payments for health care may be excessively high in relation to

personal income that such payments may result in ‘financial catastro-

phe’for the individual or the household. High health-care expendi-

tures can mean that people have to cut down on necessities, such as

food or clothing. Three factors account for the increase in cata-

strophic health expenditure: OOP payments required to access health

services; low household capacity to pay (CTP); and nonexistent

prepayment mechanisms for risk pooling (Xu, 2005). The increase in

medical expenditures and OOP health expenditures can be impo-

verishing (van Doorslaer et al., 2005) and may affect a household's

welfare (Mchenga et al., 2017). The rates of catastrophic health-care

expenditure are usually higher in countries with limited prepayment

systems or resources (Xu et al., 2003). Given that the problem of cata-

strophic health payments will not simply disappear with increasing

income, the complex process of developing social institutions to effec-

tively pool financial risk must be considered (Xu et al., 2003). The

health and finance departments of developing countries, as well as

other government bodies, have an important role in regulating and

financing the health sector (Sulku and Caner, 2011).

Turkey is a developing country and has provided accessible health

services under the effect of the Health Transformation Program since

2003 (Akdag, 2012; Okem and Cakar, 2015). According to statistics

from the World Bank in Turkey, thepercentage of the current health-

care expenditure in gross domestic product decreased from 4.40% in

2013 to 4.22% in 2017 (World Bank, 2020a). On the other hand, the

Received: 26 March 2020 Revised: 11 July 2020 Accepted: 28 September 2020

DOI: 10.1002/isaf.1483

168 © 2020 John Wiley & Sons, Ltd. Intell Sys Acc Fin Mgmt. 2020;27:168–181.wileyonlinelibrary.com/journal/isaf