• Communications for Statistical Applications and Methods
• /
• v.27 no.6
• /
• pp.675-688
• /
• 2020

In survival analysis of observational data, the inverse probability weighting method and the Cox proportional hazards model are widely used when estimating the causal effects of multiple-valued treatment. In this paper, the two kinds of weights have been examined in the inverse probability weighting method. We explain the reason why the stabilized weight is more appropriate when an inverse probability weighting method using the generalized propensity score is applied. We also emphasize that a marginal hazard ratio and the conditional hazard ratio should be distinguished when defining the hazard ratio as a treatment effect under the Cox proportional hazards model. A simulation study based on real data is conducted to provide concrete numerical evidence.

• The Korean Journal of Applied Statistics
• /
• v.36 no.5
• /
• pp.381-397
• /
• 2023

The most representative design used in clinical trials is randomization, which is used to accurately estimate the treatment effect. However, comparison between the treatment group and the control group in an observational study without randomization is biased due to various unadjusted differences, such as characteristics between patients. Propensity score weighting is a widely used method to address these problems and to minimize bias by adjusting those confounding and assess treatment effects. Inverse probability weighting, the most popular method, assigns weights that are proportional to the inverse of the conditional probability of receiving a specific treatment assignment, given observed covariates. However, this method is often suffered by extreme propensity scores, resulting in biased estimates and excessive variance. Several alternative methods including trimming, overlap weights, and matching weights have been proposed to mitigate these issues. In this paper, we conduct a simulation study to compare performance of various propensity score weighting methods under diverse situation, such as limited overlap, misspecified propensity score, and treatment contrary to prediction. From the simulation results overlap weights and matching weights consistently outperform inverse probability weighting and trimming in terms of bias, root mean squared error and coverage probability.

• The Korean Journal of Applied Statistics
• /
• v.36 no.4
• /
• pp.323-335
• /
• 2023

In causal analysis of high dimensional data, it is important to reduce the dimension of covariates and transform them appropriately to control confounders that affect treatment and potential outcomes. The augmented inverse probability weighting (AIPW) method is mainly used for estimation of average treatment effect (ATE). AIPW estimator can be obtained by using estimated propensity score and outcome model. ATE estimator can be inconsistent or have large asymptotic variance when using estimated propensity score and outcome model obtained by parametric methods that includes all covariates, especially for high dimensional data. For this reason, an ATE estimation using an appropriate dimension reduction method and semiparametric model for high dimensional data is attracting attention. Semiparametric method or sparse sufficient dimensionality reduction method can be uesd for dimension reduction for the estimation of propensity score and outcome model. Recently, another method has been proposed that does not use propensity score and outcome regression. After reducing dimension of covariates, ATE estimation can be performed using matching. Among the studies on ATE estimation methods for high dimensional data, four recently proposed studies will be introduced, and how to interpret the estimated ATE will be discussed.

• Journal of the Korean Data and Information Science Society
• /
• v.26 no.1
• /
• pp.217-227
• /
• 2015

In observational study, handling confounders is a primary issue in measuring treatment effect of interest. Historically, a regression with covariate adjustment (covariate-adjusted regression) has been the typical approach to estimate treatment effect incorporating potential confounders into model. However, ever since the introduction of the propensity score, covariate-adjusted regression has been gradually replaced in medical literatures with various balancing methods based on propensity score. On the other hand, there is only a paucity of researches assessing propensity score methods compared with the covariate-adjusted regression. This paper examined the performance of propensity score methods in estimating risk difference and compare their performance with the covariate-adjusted regression by a Monte Carlo study. The study demonstrated in general the covariate-adjusted regression with variable selection procedure outperformed propensity-score-based methods in terms both of bias and MSE, suggesting that the classical regression method needs to be considered, rather than the propensity score methods, if a performance is a primary concern.

• Communications for Statistical Applications and Methods
• /
• v.29 no.2
• /
• pp.177-191
• /
• 2022

This paper addresses the use of machine learning methods for causal estimation of treatment effects from observational data. Even though conducting randomized experimental trials is a gold standard to reveal potential causal relationships, observational study is another rich source for investigation of exposure effects, for example, in the research of comparative effectiveness and safety of treatments, where the causal effect can be identified if covariates contain all confounding variables. In this context, statistical regression models for the expected outcome and the probability of treatment are often imposed, which can be combined in a clever way to yield more efficient and robust causal estimators. Recently, targeted maximum likelihood estimation and causal random forest is proposed and extensively studied for the use of data-adaptive regression in estimation of causal inference parameters. Machine learning methods are a natural choice in these settings to improve the quality of the final estimate of the treatment effect. We explore how we can adapt the design and training of several machine learning algorithms for causal inference and study their finite-sample performance through simulation experiments under various scenarios. Application to the percutaneous coronary intervention (PCI) data shows that these adaptations can improve simple linear regression-based methods.

• Korean Journal of Psychosomatic Medicine
• /
• v.31 no.2
• /
• pp.108-117
• /
• 2023

Objectives : Somatic symptom disorder (SSD) is characterized by the manifestation of a variety of physical symptoms, but little is known about differences in autonomic nervous system activity according to symptom severity, especially within patient groups. In this study, we examined differences in heart rate variability (HRV) across symptom severity in a group of SSD patients to analyze a representative marker of autonomic nervous system changes by symptoms severity. Methods : Medical records were retrospectively reviewed for patients who were diagnosed with SSD based on DSM-5 from September 18, 2020 to October 29, 2021. We applied inverse probability of treatment weighting (IPTW) methods to generate more homogeneous comparisons in HRV parameters by correcting for selection biases due to sociodemographic and clinical characteristic differences between groups. Results : There were statistically significant correlations between the somatic symptom severity and LF (nu), HF (nu), LF/HF, as well as SD1/SD2 and Alpha1/Alpha2. After IPTW estimation, the mild to moderate group was corrected to 27 (53.0%) and the severe group to 24 (47.0%), and homogeneity was achieved as the differences in demographic and clinical characteristics were not significant. The analysis of inverse probability weighted regression adjustment model showed that the severe group was associated with significantly lower RMSSD (β=-0.70, p=0.003) and pNN20 (β=-1.04, p=0.019) in the time domain and higher LF (nu) (β=0.29, p<0.001), lower HF (nu) (β=-0.29, p<0.001), higher LF/HF (β=1.41, p=0.001), and in the nonlinear domain, significant differences were tested for SampEn15 (β=-0.35, p=0.014), SD1/SD2 (β=-0.68, p<0.001), and Alpha1/Alpha2 (ß=0.43, p=0.001). Conclusions : These results suggest that differences in HRV parameters by SSD severity were showed in the time, frequency and nonlinear domains, specific parameters demonstrating significantly higher sympathetic nerve activity and reduced ability of the parasympathetic nervous system in SSD patients with severe symptoms.

• Health Policy and Management
• /
• v.34 no.1
• /
• pp.68-77
• /
• 2024

Background: Depressive disorders can be categorized into daily depression and clinical depression. The experience of depressive disorder can increase health care utilization due to decreased treatment compliance and somatization. On the other hand, the clinical depression group may also experience social prejudice associated with the illness, which can limit their access to health care utilization. In terms of the significance of health care utilization as a factor in individual and social issues, this study aims to compare the health care utilization of the clinical depression group with that of the non-depressed group and the daily depression group. Methods: The analysis utilized the inverse probability of treatment weighting based on the generalized propensity score. Results: As a result of the analysis, clinical depression and daily depression were higher among women, low-income groups, individuals with low education levels, and so forth. The clinical depression group was also higher among individuals who were not economically active, did not have private health insurance, or had multiple chronic diseases. The number of outpatient department visits in the depression group was significantly higher than in the non-depressed group. In addition, the number of outpatient department visits for the clinical depression group was significantly higher than that for the daily depression group. Outpatient medical expenses were higher in the depression group than in the non-depressed group, and there was no significant difference between the clinical depression group and the daily depression group. Conclusion: Health care utilization was higher in the depression group than the non-depressed group, it was also higher in the clinical depression group than the daily depression group.

• Therapeutic Science for Rehabilitation
• /
• v.12 no.3
• /
• pp.61-77
• /
• 2023

Objective : This study analyzed the correlation between cognitive function and oral health-related quality of life (OHQoL). Methods : Demographic and clinical characteristics were extracted and utilized for subjects aged 45 years or older who participated in the 8th Korean Longitudinal Study on Aging in 2020. The dependent variable was the Geriatric Oral Health Assessment Index, and the independent variable was the level of cognitive function classified by the Mini-Mental State Examination scores. The analysis method used inverse probability of treatment weighting (IPTW). Then, the association between cognitive function and OHQoL was analyzed by multiple regression analysis. Results : Among the participants, 4,367 (71.40%) had normal cognition, 1,155 (18.89%) had moderate cognitive impairment, and 594 (9.71%) had severe cognitive impairment. As a result of analysis by applying IPTW, there was a negative correlation between the cognitive function group and OHQoL (normal vs. moderate: β = -2.534, p < .0001; normal vs. severe: β = -2.452, p < .0001). Conclusion : After propensity score matching, mild cognitive impairment showed a more negative association than severe cognitive impairment. Therefore, patients with cognitive impairment require oral health management education to improve OHQoL regardless of the level of cognitive impairment.

• Korean Journal of Clinical Pharmacy
• /
• v.29 no.2
• /
• pp.109-114
• /
• 2019

Background: Rheumatoid arthritis (RA) is a systemic inflammatory disease that manifests as joint damage or athletic disability via sustained inflammation of the synovial membrane. The risk of cardiovascular disease (CVD) is higher in RA patients. This study aimed at evaluating the association between CVD comorbidities and RA by comparing a pharmacotherapy group with a non-pharmacotherapy group. Methods: Patient sample data from the Health Insurance Review and Assessment Service (HIRA-NPS-2016) were used. Inverse probability of treatment weighting (IPTW) using the propensity score was used to minimize the differences in patient characteristics. Logistic regression analysis was used to evaluate the risk of CVD comorbidities. Results: The analyses included 1,207,213 patients, of which 33,122 (2.8%) had RA. The odds ratios (OR) of CVD comorbidities were increased in RA patients; ischemic heart disease (IHD: OR 1.75; 95% CI 1.73, 1.77), cerebral infarction (CERI: OR 1.28; 95% CI 1.26, 1.30), hypertension (HTN: OR 1.44; 95% CI 1.43, 1.45), diabetes mellitus (DM: OR 2.04; 95% CI 2.03, 2.06), and dyslipidemia (DL: OR 3.49; 95% CI 3.47, 3.51). The ORs of IHD, CERI, HTN, and DM in the traditional DMARD and biologic treatment groups were decreased, compared with those in the non-pharmacotherapy group. Conclusions: Thus, CVD risk was higher in RA patients, considering age, sex, and socioeconomic status. Appropriate pharmacotherapy could decrease the risk of CVD comorbidities in RA patients.

• The Korean Journal of Applied Statistics
• /
• v.27 no.7
• /
• pp.1077-1095
• /
• 2014

Unlike randomized trial, statistical strategies for inferring the unbiased causal relationship are required in the observational studies. Recently, new methods for the causal inference in the observational studies have been proposed such as the matching with the propensity score or the inverse probability treatment weighting. They have focused on how to control the confounders and how to evaluate the effect of the treatment on the result variable. However, these conventional methods are valid only when the treatment variable is categorical and both of the treatment and the result variables are directly observable. Research on the causal inference can be challenging in part because it may not be possible to directly observe the treatment and/or the result variable. To address this difficulty, we propose a method for estimating the average causal effect when both of the treatment and the result variables are latent. The latent class analysis has been applied to calculate the propensity score for the latent treatment variable in order to estimate the causal effect on the latent result variable. In this work, we investigate the causal effect of adolescents delinquency on their substance use using data from the 'National Longitudinal Study of Adolescent Health'.