A secondary data analysis was used in this study. As a first step we retrieved the data for the period between January 1, 2006 and December 31, 2016 from the Health Data Center (HDC) in Roi Et Province. HDC is a health database system that stores monthly health data since 2007 from all health care services belonging to the Ministry of Public Health, Thailand. The health data comprised demographic data, household data, community (village, activity), health facilities in the community, disability details, provider, service, outpatient department (diagnostic date, procedure, drug and laboratory), inpatient department (admission date, diagnostic date, procedure, drug and laboratory) and the prevalence of communicable diseases. The data does not contain identifying information and cannot be linked to individuals. We selected participants with an estimated GFR (eGFR) of less than 15 ml/min/1.73m ² (CKD stage 5), aged ≥18 years and with universal health coverage were included in our study.

The Khon Kaen University Ethics Committee for Human Research approved this study (permit no.: HE642084).

The rate of access to renal replacement therapy (RRT) services was calculated by the number of participants accessing treatment for RRT (peritoneal dialysis, hemodialysis) among all dialysis modalities. This data does not include kidney transplant patients. We aggregated the data for each month during the study period. Therefore, the data were created for120 monthly study points between January 1, 2006 and December 31, 2016.

We also collected information including gender (male, female), age (18 to 44 years, 45 to 60 years, > 60 years), and the primary cause of renal failure (hypertension, diabetes, heart disease).

In the ITS analysis, segmented regression was used to test the effect of implementation before, and immediately after the intervention. The strength of the model distinguished the effect of the intervention from secular change. Time series data often have many issues that may affect to the robustness of the analysis such as seasonality, time-varying confounders, use of control, and for other more complex ITS designs, over-dispersion and autocorrelation. ¹² Seasonality refers to some events occurring more frequently than others, which may affect to the results. Time-varying confounders may occur because of the longitudinal data and age variation across time. Finally, autocorrelation is the correlation between the values of the same variables across different observations (period of the study). ¹⁶

The dataset for the segmented regression analysis comprised the rate for the event and the time point when the intervention was implemented. An interrupted time series analysis measures trends as slope changes before and after implementation of the intervention, and also the effect seen when the intervention is implemented (the instant effect). Our analysis constructed a series of quarterly rates of access to treatment for renal replacement therapy for CKD stage 5 patients from January 2006 to December 2016. The 24-month pre-intervention period was from January 2006 to December 2007, while the 96-month post-intervention period took place from January 2008 to December 2016. We aggregated the data with 3-month interval. Therefore, there have an 8-time points in the pre-intervention period and 32-time point in post-intervention.

The equation for the ITS analysis ^{17 , 18} was as follows: Y = β 0 + β 1 × time before intervention x 1 + β 2 × intervention x 2 + β 3 × time after intervention + e t where Y is the rate of access to renal replacement therapy treatment, β 0 is a constant term, and β 1 is the coefficient of the time before intervention. Regarding the trend of access to the clinic before implementing the policy, β 2 is the coefficient of the intervention immediately (level change), and β 3 is the coefficient of time after intervention that refers to the effects of the intervention over time (a different slope before and after intervention). For autocorrelation, we used robustness to fit the model. Stata software, version 12.0, was used to perform all the analyses.

The total yearly incidence of RRT increased by an average of 14.8% after the implementation of the universal coverage policy for renal replacement therapy (known first as CAPD) and the yearly incidence of all RRT modalities increased by an average of 34.8% in 2007 to 2009. ²⁰ A report by Thai Renal Replacement Therapy (TRT) stated that the number of end-stage renal failure patients receiving renal replacement services increased from 68.34 per million in 2006 to over 181 per million in 2009 (or more than 11,500 new patients per year). The number of people receiving renal replacement services increased significantly from 419 per million to over 639 per million (or more than 40,000 in 2009). ²¹ Our analysis results also showed that the linear trend of RRT after implementing the continuous policy slightly increased. Despite the expected increased volume of patients, the year-by-year growth rate of patients in all RRT modalities seemed to diminish over time. ²² due to the CAPD first policy. HD is performed only for those who are unable to undergo peritoneal dialysis or who have a medical indication prohibiting PD. Therefore, HD dialysis will be covered. ²³ Some patients who meet the criteria for HD decide not to enroll for CAPD due to lack of readiness for the treatment required, such as inappropriate accommodation, lack of caregivers, being unsure of self-cleaning processes at home and fear of infection after the procedure. ²⁴

A slight increase in access to RRT services was in line with the assessment of access to services and the provision of renal replacement services under the health insurance system in Thailand. The expected number of patients who accessed the services was more than 35,000 in 2011, but there are only 19,000 cumulative cases under the UHC scheme. At the end of the fiscal year 2012, the number of patients was estimated at more than 48,000 patients, but the cumulative number of cases was only about 23,000. ²⁵ In terms of service provision, some hospitals are unable to participate in the universal health insurance program or are unable to provide services. Due to the availability of human resources and location, a health care service might require patients to go to other hospitals that can provide services but are far from their homes, which made it inconvenient and costly for traveling. This might be a consideration when making the decision to select a treatment.

Our study had some limitations. Firstly, data retrieved from other hospitals might have different methods that could have impacted data quality. Secondly, only two hospitals currently incorporate this scheme and one is a private hospital, so some data were incomplete. Finally, our results might not be generalizable to Thailand as a whole because we analyzed data from only one province. However, the strength of our study is an analysis using the interrupted time series model that accurately interpreted the results.