Interrupted time series analysis for the impact of integrated medical insurance on out-of-pocket hospitalization costs for catastrophic illness

Study region

The study was conducted in Jiujiang in the Chinese province of Jiangxi, located in southern China. The city has ten counties with a population of 4 million. Prior to 2016, patients covered by NCMS were reimbursed for all costs (including catastrophic medical insurance) prior to discharge18. From 2016, an integrated urban and rural medical insurance was set up19. In addition, since October 2017, patients in the city have been reimbursed for all benefits under a “one-stop instant reimbursement plan” before discharge, including basic health insurance benefits, catastrophic health insurance, supplemental health insurance for destitute populations, and medical assistance from the Bureau of Civil Affairs20. In addition, the local government levies no less than 5% of the fund pool for catastrophic medical insurance which aims to provide additional reimbursements of high medical expenses.21. The catastrophic medical insurance deductible is set at 50% of Jiujiang’s per capita disposable income, the reimbursement rate is 60%. The main differences and changes between NCMS and URRBMI in Jiujiang are listed in Table 1.22,23,24,25.

Table 1 Comparison of NCMS and URRBMI between 2015 and 2018.

In this study, we chose one tertiary hospital out of six tertiary hospitals for analysis. The hospital selected was one of the largest hospitals with over 1500 beds.

Data management

Data was extracted from the hospital’s information system, containing information on all patients diagnosed with catastrophic illness admitted from January 2014 to December 2018. Catastrophic illness included the following 20 illnesses and was covered by both URRBMI and catastrophic health insurance: acute myocardial infarction, cancer, cerebral infarction, cancer of the cervix, chronic myeloid leukemia, cleft lip and palate, colon cancer, colorectal cancer, end-stage renal failure, cancer of the esophagus, gastric cancer, hemophilia, opportunistic HIV infection, hyperthyroidism, hypospadias, lung cancer, thalassemia, multidrug-resistant tuberculosis, serious mental illness and type I diabetes19.

The “tidyverse” R package was used for data cleaning. Data over 5 years has been combined into a single file. 20 catastrophic illnesses were selected by matching the primary diagnosis with the ICD-10 code. Variables selected for study included socio-demographic characteristics, length of stay, ICD-10 diagnosis, insurance, medical costs. For medical expenses, outliers and missing data were explored. Complete cases were included in the final data. China’s Consumer Price Index (CPI) from 2014 to 2018 was obtained from the World Bank26. All medical costs have been standardized by the IPC taking the year 2014 as a reference.

Data analysis

Interrupted time series analysis (ITSA) was used to analyze the effects of integration. ITSA is considered a robust quasi-experimental approach in the absence of control groups. In interrupted time series studies, segmented regression analysis has been shown to be effective in estimating intervention effects27,28,29. In this study, the model was constructed as follows:

$${text{Y}} , = , beta_{0} + , beta_{{1}} times {text{ time }} + , beta_{{2}} times {text{ integration }} + , beta_{{3}} times {text{ time after integration }} + , varepsilon ,$$

where Y is the dependent variable representing the reduced percentage of unreimbursed hospital costs per capita, the total hospital expenditure per capita and the average reimbursement rate. Time, integration and time after integration were considered as the independent variables. Time referred to each month from January 2014 to December 2018. Integration was assigned to 0 before integration (January 2014 to December 2015) and to one after integration (January 2016 to December 2018). The time after integration was assigned to 0 before the integration and increased monthly after the integration (because there were 24 months before the integration and 36 after the integration, the value of this variable after the integration was fixed from 25 to 60). β0 is the level of the variables explained at the beginning. β1 estimates the baseline trend before integration. β2 estimates the effect of integration on the change in intercept. β3 reflects the changed trend after the integration (the change in slope). ε is the error term. All statistical analyzes were performed using R software version 3.5.2. R packages “Wats” and “nlme” were used for the ITSA, “ggplot2” was used to create the graphs.

Ethical statements

The study was conducted in accordance with the guidelines of the Declaration of Helsinki and approved by the Ethics Committee of Jiujiang University (REC: JJU20160116).

The patient’s consent was revoked by the Jiujiang University ethics committee because this study only involved anonymized secondary public access data.

Comments are closed.