Identifikasi Faktor yang Memengaruhi Layanan Rujuk Lanjutan Menggunakan Regresi Logistik Group LASSO

Abstract

Health is a very important need in life and is a human right to be able to live a decent, productive, and prosper life. For its importance in everyday life, it is necessary to have good health providers as stated in the 1945 Constitution of the Republic of Indonesia Article 28 H paragraphs 1 and 3. The government has made a program run by BPJS Kesehatan, namely the National Health Insurance which aims to improve access and quality of health services in order to achieve optimal welfare degrees effectively and efficiently. With the existence of the National Health Insurance, it is hoped that health services will no longer be centered in hospitals or FKRTL but will be carried out in stages according to medical needs. The high referral ratio is caused by the implementation of referrals that are not in accordance with the level flow, which will cause an accumulation of patients in one hospital, so that it has an impact on the decline in service quality. The high ratio of referral rates is a problem in implementing a tiered referral system, so it is necessary to find out what factors affecting the referral status of FKTP patients to FKRTL. The response variable used in this study was the patient's referral status of First Level Health Facilities (FKTP) whether or not including further referral to Advanced Referral Health Facilities (FKRTL). If the response variable is no longer a quantitative variable, but a categorical variable consisting of only a few values, then linear regression cannot be used. In high-dimensional data, the response variable is related to a small number of explanatory variables among a large number of possible variables. Thus, the selection of variables is important to do in identifying the relevant variables in high-dimensional data. One alternative method to deal with this case, the penalized logistic regression method was developed. Coefficient estimators in penalized logistic regression are carried out by maximizing the log of the probability function by adding a penalty to the function which causes the coefficient estimator to be biased, but reducing the variance of the coefficient estimator will increase the accuracy of the model prediction (Hastie et al. 2008). Therefore, penalized logistic regression is often referred to as the shrinking or regularization method. In this study, the LASSO Group regularization method was used because the explanatory variables in the research data were in the form of groups. Group LASSO is an extension of the LASSO method by adding group penalties to the LASSO method. To find out what factors affecting the patient's referral status to FKRTL, logistic regression is used, in which the response variable is -1 for outpatient treatment, recovery, internal referral, forced discharge, others; death and 1 for further referral to FKRTL, following the Bernoulli distribution. The covariates included in the process of variables are age, family relationship, gender, marital status, class of treatment, segment of participants, ownership of health facilities, type of health facility, type of health facility, and level of service. Dummy/dummy variables are variables used to quantify qualitative variables such as in this study, include age category, family relationship, gender, marital status, class of care, participant segment, ownership of health facilities, type of health facility, type of health facility, level of service, and poly. The dummy variable only has two values, namely the value 1 and the value 0 and is coded with the symbol D. A value of 0 usually indicates a group that does not receive a treatment and 1 indicates a group that receives treatment.