Healthcare Resource Uses and Out-of-Pocket Expenses Associated with Pulmonary TB Treatment in Thailand

This was a retrospective study based on patient medical records, TB registration records, and the billing database at three public referral hospitals in Songkhla province, southern Thailand: Songklanagarind hospital, Hatyai hospital, and Songkhla hospital, where pharmaceutical care, home visit, and modified DOT, respectively, have been adopted. Songklanagarind hospital is an 850-bed teaching hospital affiliated to the Prince of Songkla University (PSU). Hatyai hospital (700-bed, tertiary care) and Songkhla hospital (480-bed, secondary care) are the main provincial referral hospitals under the Ministry of Public Health. This study was approved by the Research Ethics Board at each of the participating hospitals.

Pharmaceutical care provides pharmacist-led patient education at every outpatient visit and telephone consultation at any time via mobile phone regarding disease, medications, possible ADRs, and lifestyle modifications. The home visit approach provides regular home visit until treatment completion. The modified DOT provides direct observation, but only for the first 2 weeks of TB treatment, followed by regular home visit. The description of each of these TB treatment adherence enhancement strategies is shown in Table 1.

The eligibility criteria for this cost analysis included age ≥18 years, pulmonary TB confirmed by a physician and classified bacteriologically as smear-positive or smear-negative, TB treatment started between October 2010 and September 2013, and a treatment period of ≥2 months in the study hospitals.

First-line anti-TB medications include isoniazid (H), rifampicin (R), pyrazinamide (Z), ethambutol (E), and streptomycin (S). The standard treatment regimen for drug-susceptible pulmonary TB is a daily combination of four first-line antibiotics (2HRZE/4HR) for at least 6 months divided into two treatment phases: an intensive phase (first 2 months) and a continuous phase (the remaining 4 months). Re-treatment patients are treated with an 8-month re-treatment regimen (2HRZES/1HRZE/5HRE), while patients with multi-drug-resistant tuberculosis (MDR-TB) require second-line agents (at least 20 months) as recommended by WHO [1].

The cost analysis was conducted from a healthcare sector perspective inclusive of out-of-pocket expenditures, which included direct healthcare costs to the public payer and to the patients. Costs were estimated separately for the pre-TB treatment period (from illness onset to TB diagnosis) and the TB treatment period (from the start to the completion of TB treatment). These costs were considered as costs associated with TB diagnosis and treatment if patients were diagnosed with at least one of the two ICD-10 codes for respiratory TB (A15 or A16). There are a few other relevant healthcare resource uses (e.g., treatment of ADRs, hospitalization, nutrition, and others). Therefore, costs incurred with some ICD-10 codes needed to be investigated concurrently with other information (e.g., type of outpatient service, laboratory investigation, patient’s medical record, and TB registration record) before they were determined to be associated with TB diagnosis and treatment (see Appendix Table 1 in the electronic supplementary material).

The Thai population has been covered under one of the three public insurance schemes: the civil servant medical benefits scheme (CSMBS), the social security scheme (SSS), and the universal coverage scheme (UC), all funded through general tax revenues [8]. The CSMBS covers government employees, pensioners, and their dependents in which insured persons can access, free of charge, medical services at all government hospitals. The SSS is for private employees who can freely access medical services at a contracted hospital. Every Thai citizen not covered under the CSMBS or the SSS is covered by the UC, but needs to pay all medical costs out of pocket if they seek healthcare services outside their primary residential area. As of 2015, 75% of the Thai population is covered under the UC [9].

Following the Thai Comptroller General’s Department’s healthcare service classification, we categorized direct costs into seven categories: medications, laboratory investigation and pathology, diagnostic radiology, medical services, supervision activities, hospitalization, and miscellaneous costs. Except medications costs, which were calculated using a mark-up method, unit costs of services in all remaining categories were set by factoring the costs related to labor, material, capital, overhead, and future development costs. Direct costs for each healthcare service in the hospitals were calculated by multiplying the natural units for the services with the corresponding unit costs. Costs associated with implementing supervision activities (i.e., the home visit at Hatyai hospital and the modified DOT at Songkhla hospital) were estimated separately as non-medical staff must be hired to provide these services. In addition to the salary paid from hospital, the global fund paid these staff 600 baht for each smear-positive drug-susceptible patient who successfully completed treatment, and 300 baht monthly for supervising each MDR-TB patient.

Methods for handling missing information varied depended on the type of data. Missing resource use items for laboratory investigation and diagnostic radiology were replaced by using each hospital’s standard procedures for pulmonary TB care, while missing quantities of medication used were estimated from patients’ drug regimen.

Out-of-pocket expenditures not covered by patients’ health insurance were also retrieved from hospital billing databases. These costs included copayment, non-essential drugs (NEDs), food nutrition, and non-covered medical costs (e.g., costs incurred to patients when accessing services provided outside their district).

All costs were calculated in Thai currency (Thai baht), adjusted with the cumulative inflation rate from the year of data collection to 2015 [10], and then converted to international dollars using the 2015 purchasing power parity (PPP) conversion factor of private consumption (LCU per international $) [11].

Descriptive analyses of demographic data were conducted to describe the patient population included in the cost analyses. Categorical variables were analyzed using a Chi squared test, unless expected cell counts were low, in which case Fisher’s exact test was used. Mean and standard deviation (SD) was calculated for continuous variables and differences were evaluated using analysis of variance (ANOVA).

For the cost data, measures other than the arithmetic mean (e.g., median costs, log transformed costs) do not provide information about the costs of treating all patients, which is more informative for healthcare policy decisions [12]. Therefore, we reported arithmetic mean costs with SD. Since the Shapiro–Wilk test indicated that all costs data were positively skewed, a nonparametric bootstrap technique was used to compare arithmetic mean costs and to derive the 95% confidence intervals (CI) [13].

Differences in the total costs between treatment strategies adjusted for baseline demographics were investigated by generalized linear models (GLMs) [14, 15]. The performance of GLMs with the inverse Gaussian and gamma distribution were investigated. Two different link functions (e.g., the identity and log link) were compared. Covariates act additively in the identity-link model, and multiplicatively in the log-link model [14]. Thus, GLMs using an identity link estimate the difference in costs, while those using a log link estimate a ratio. The model included six baseline covariates of treatment strategy, sex, age, habitat, health insurance, and HIV status. Model performance was assessed by using normal probability plots (P–P plots), normal quantile plots (Q–Q plots) of deviance residuals [14, 16], and Akaike information criterion (AIC) [14, 17]. Smaller AIC values indicate better-fitting models; differences of ten or more in AIC indicate that the better-fitting model should be strongly preferred [14]. Normal plots were assessed visually. When points fall close to the reference line this indicates that model assumptions are suitable [14, 16, 18]. Data analyses were performed using SPSS software [19].