Economic Evaluation of Palliative Management versus Peritoneal Dialysis and Hemodialysis for End-Stage Renal Disease: Evidence for Coverage Decisions in Thailand

Authors


Yot Teerawattananon, International Health Policy Program, Ministry of Public Health, Tiwanon Road, Nonthaburi 11000, Thailand. E-mail: yot@ihpp.thaigov.net or yot.teera@hotmail.com

ABSTRACT

Objective:  To examine the value for money of including peritoneal dialysis (PD) or hemodialysis (HD) into the universal health insurance scheme of Thailand.

Methods:  A probabilistic Markov model applied to end-stage renal disease (ESRD) patients aged 20 to 70 years was developed to examine the incremental cost-effectiveness ratio (ICER) of palliative care versus 1) providing PD as an initial treatment followed by HD if complications/switching occur; and 2) providing HD followed by PD if complications/switching occur. Input parameters were extracted from a national cohort, the Thailand Renal Replacement Therapy Registry, and systematic reviews, where possible. The study explored the effects of uncertainty around input parameters, presented as cost-effectiveness acceptability frontier, as well as the value of obtaining further information on chosen parameters, i.e., partial expected value of perfect information.

Results:  Using a societal perspective, the average ICER of initial treatment with PD and the average ICER of initial treatment with HD were 672,000 and 806,000 Baht per quality-adjusted life-year (QALY) gained (52,000 and 63,000 purchasing power parity [PPP] US$/QALY) compared with palliative care. Providing treatments for younger ESRD patients resulted in a significant improvement of survival and gain of QALYs compared with the older aged group. The cost-effectiveness and cost-utility ratios of both options for the older age group were relatively similar.

Conclusions:  The results suggest that offering PD as initial treatment was a better choice than offering HD, but it would only be considered a cost-effective strategy if the social willingness-to-pay threshold was at or higher than 700,000 Baht per QALY (54,000 PPP US$/QALY) for the age 20 group and 750,000 Baht per QALY (58,000 PPP US$/QALY) for age 70 years.

Introduction

The treatment of end-stage-renal disease (ESRD) is recognized as a major economic and political challenge in health care [1]. Renal replacement therapy is essential to many patients suffering from ESRD. It is, however, one of the most expensive health technologies [2]. Unsurprisingly, policy analysis and economic evaluation of ESRD treatment is among the first interventions to have been assessed, and evaluations have been performed regularly in many settings worldwide [3–15].

There are three major treatment modalities for patients with ESRD: peritoneal dialysis (PD), hemodialysis (HD), and kidney transplantation. A number of previous studies confirmed that kidney transplantation was the most cost-effective strategy and considered the preferable choice [3–8]. In many settings, however, including Thailand, the number of kidney donors is insufficient to meet demand. There are around 200 donated kidneys available each year compared with the current incidence of 10,000 ESRD patients per year [16,17].

Thailand has been providing universal health-care coverage through a tax-based universal health insurance scheme (UC) since 2001 [18]. The scheme protects a population of 45 million who are not eligible for Civil Servant Medical Benefit Scheme (CSMBS) or Social Security Scheme (SSS). Although all treatment modalities for ESRD are currently covered by CSMBS and SSS, none of them is included in the UC benefit package [19]. To date, there is strong pressure from various stakeholders to provide universal access to PD and HD for UC beneficiaries [20].

This study is one of a series of studies supported by the National Health Security Office (NHSO) to provide scientific evidence for policymakers to make decisions on whether to provide dialysis treatments for ESRD patients under UC. Using both the NHSO’s and a societal perspective, the objective of this study was to assess the value for money of providing PD and HD for UC patients versus the current practice, i.e., palliative care.

Design and Methods

Overview Options

The standard practice (“palliative care” option) includes restricted fluid intake, high-dose diuretic, antihypertensive drugs, calcium, bicarbonate, ferrous sulfate, blood transfusion, and hospital admission, if required. Previous evidence has indicated 50% mortality within 1 to 3 months [19].

Peritoneal dialysis has two main treatment varieties, either with manual exchange of dialysis fluid (continuous ambulatory peritoneal dialysis—CAPD) or with automated exchange of dialysis fluid at night. Only CAPD is offered in Thailand so it has been included in the analysis [17].

Hemodialysis can be carried out in a hospital, dialysis center, a satellite unit, or in a patient’s home. Hospital and in-center dialysis, however, are the only options currently available in Thailand [17]. At the end of 2004, 301 hospitals and dialysis centers, at least one in every province, offered HD [21].

Analyses and Model

The traditional approach of economic evaluation, incremental cost-effectiveness analysis, which compares additional costs and health outcomes of moving from one intervention to alternatives, may be inappropriate in this case [2]. Because PD and HD are not complete substitutes, patients assigned for PD may require HD while complications such as peritonitis occur, and vice versa. Hence, this study evaluated the incremental cost and effectiveness of moving from current practice, palliative care, versus 1) providing PD as initial care followed by HD if complications/switching occur; and 2) providing HD as initial care followed by PD if complications/switching occur.

We developed a decision-analytic model and applied it to ESRD patients aged 20 to 70 years. The model was used to quantify the costs and effects of the two alternative long-term managements for ESRD, starting either with PD or with HD, for each patient age group. The Markov model structure shown in Figure 1 illustrates the mutually exclusive health states that a patient commencing treatment on either PD or HD may go through, respectively. The states of health are denoted in the solid-line ovals. We also developed substates (dotted-line ovals) to reflect the difference in rates of complications between the two treatment modalities. An arrow indicates that movement from one state to another is possible. The movement between each state is determined by probabilities that were obtained from a national cohort, the Thailand Renal Replacement Therapy (TRT) Registry, and systematic reviews, where possible. Because both PD and HD are lifelong treatments, the model used a 1-year-cycle length for full health state and 1 month for complication substates.

Figure 1.

Schematic diagram of the Markov model.

In the model, patients may start either with PD or with HD and remain on the same treatment for the next cycle. Some patients, however, may experience complications for a month at anytime during a year cycle. The move between treatments or entry into the final state (death) may or may not be related with the occurrence of complications. Patients may die of non-ESRD causes, such as cardiovascular disease, or patients may move to another treatment because of dialysis complications, such as catheter-related infections (for HD) or peritonitis (for PD), or patients may have nonmedical limitations, such as moving to a new place where there is no hemodialysis available. In each case, it was assumed that the event would only happen at the end of each cycle.

Monte Carlo simulation was used to model costs and events over a 99-year period to cover the total period over which the whole cohort would be expected to survive.

To comply with the guideline set beforehand for conducting this health economic evaluation [22], and to be comparable with other economic evaluations [23], all costs and outcomes were discounted at the rate of 3.5%. We, however, also explored results with the discount rates of 0% and 6%.

Outcome Measures

From the review, we found only one small randomized controlled trial (38 patients) that investigated the relative efficacy of PD and HD and found no significant difference in 5-year survival (95% confidence interval of hazard ratio between 0.8 and 15.4) after adjusting for age and comorbidities [24]. A systematic review conducted by MacLeod et al. also shows that existing data are not available to allow reliable conclusions to be drawn about the relative effectiveness of PD and HD [25]. There are still conflicting results in the literature of nonrandomized observational studies, but when data are adjusted for comorbid conditions they show no significant difference in patient survival on PD and HD [25,26]. Our assumption, based on these sources, is that PD and HD are equally effective in terms of patient survival.

Using this assumption, we estimated survival rates for a hypothetical cohort of patients from the TRT Registry undergoing dialysis. The national database consisted of records of 6272 patients who underwent dialysis from 1997 to 2003. In our analyses, we assumed all patients as having similar treatment. This does not result in biased estimators of survival but may underestimate variance. Given limitations of the database, we can only focus on patients’ age at the start of dialysis as a covariate that might potentially affect patient prognosis.

Using the statistical software package STATA (Stata Corp, College Station, TX), we initially applied the nonparametric Kaplan–Meier approach [27] to fit Kaplan–Meier curves and plot graphs of log{−log[S(t)]} against log(time), which were generally linear, indicating that a Weibull survival model would adequately fit the data [28]. We consequently used the “streg” module of the software to perform maximum likelihood estimations for parametric regression of the Weibull survival models (see Table 1).

Table 1.  Means and standard error (SE) of input parameters
ParametersMeanSEParameter distributionData source
  1. TRT, Thailand Renal Replacement Therapy Registry.

Weibull survival
 Constant value for baseline hazard−11.17710.2439LogNormalTRT
 Age coefficient for baseline hazard0.03470.0021LogNormalTRT
 ln(γ)0.07220.0247LogNormalTRT
Transitional probabilities
 Monthly probability of dying among patients with palliative care0.3310.1509Beta[19]
 Annual rate of having complications among PD patients0.32940.5739Gamma[32]
 Annual rate of having complications among HD patients0.26980.5194Gamma[19]
 Annual probability of switching from PD to HD0.02700.0054BetaTRT
 Annual probability of switching from HD to PD0.00640.0011BetaTRT
Direct health-care costs
 Lifetime costs of palliative care18,00018,000Gamma[19]
 Set-up costs for PD, e.g., peritoneal catheter implantation47,00015,000GammaSurvey
 Set-up costs for HD, e.g., arteriovenous graft21,0007,000GammaSurvey
 Annual maintenance cost for PD356,00045,000Gamma[33] and survey
 Annual maintenance cost for HD380,000132,000Gamma[38]
 Total (monthly) cost of treating PD complications32,00024,000Gamma[33] and survey
 Total (monthly) cost of treating HD complications15,00015,000Gamma[33] and survey
 Annual cost of treating comorbidities, e.g., diabetes40,00029,000Gamma[33,39]
Direct non–health-care costs, e.g., travel costs
 Lifetime cost paid by household with palliative care35,00035,000Gamma[39]
 Annual cost paid by household with PD5,0001,000Gamma[39]
 Annual cost paid by household with HD33,0008,000Gamma[39]
Indirect non–health-care costs, e.g., costs of sick leave
 Annual cost paid by household with PD3,000600Gamma 
 Annual cost paid by household with HD41,0008,000Gamma[39]
Utility parameters
 Utility for PD without complication0.720.08Beta[33–36]
 Utility for HD without complication0.680.10Beta[33–36]
 Utility for patients with palliative care/dialysis with complications0.600.15Beta[33–36]

For the Weibull distribution, the survival function, which describes the probability of survival as a function of age [29], is

S(t) = exp[−H(t)]

and

H(t) = λtγ

where H(t) is the cumulative hazard; λ (lambda) is the scale parameter; t is time in days; and γ (gamma) is the shape parameter that describes the instantaneous death rate, the hazard rate h(t), which increases with age if γ > 1. λ depends on the covariate, age, according to the formula

λ = exp[(age_coefficient × Age) + cons]

The transitional probability of dying during the cycle, tp(c), is therefore estimated from the following formula (where c is the number of cycle):

tp(c) = 1 − exp[H(t − c) − H(t)]

The survival graphs shown in Figure 2 reveal that the mortality rates were predicted to be high among older patients and increased with age. In this figure, however, it can be observed that for all age groups a great proportion of the cohort reaches 100 years or more. For example, about 20% of the age 20 group survives up to 100 years, which is not consistent with the national figures for life expectancy in the general population [30].

Figure 2.

Survival graphs by age group (up to 4 years and projected using the Thailand Renal Replacement Therapy Registry).

This is a common problem found in the modeling of survival when the length of follow-up of empirical data is insufficient to capture enough events (4 years for this case) [31]. An explanation is that the hazard rate used in the study is for all-cause mortality, so that for younger dialysis patients the total mortality rate is mostly mortality due to ESRD, whereas for older patients there will be a significant risk of mortality due to other causes, e.g., cardiovascular problems. Using the mortality hazard for age 20 years starting dialysis to estimate their survival for 99 years may underestimate their risk of mortality due to non-ESRD causes.

As an alternative, we estimated another survival model that allowed patients to move between estimated survival curves in Figure 2 as their age increases. The adjusted survival graph from the new model is presented in Figure 3, which shows more realistic estimations. We applied survival from both models in turn in our analyses.

Figure 3.

Adjusted survival (allowing patients to move between survival curves in Figure 2 as their age increases) using sensitivity analysis.

Analysis of the national database indicated that the annual probability of switching from PD to HD was 0.0270 (SE 0.0054) and the probability of switching from HD to PD was 0.0064 (SE 0.0011).

The rate of complications among those with PD and HD was not routinely reported in the TRT Registry. Therefore, we did a systematic search of national literature from MUCC-OPAC (Mahidol University), Chulalinet (Chulalongkorn University), CMUL OPAC (Chiangmai University), INNOPAC web (Khon Kaen University), and OPAC PSU (Prince of Songkla University) using the keywords “dialysis,”“renal dialysis,”“hemodialysis,” and “peritoneal dialysis.” Two independent researchers reviewed whether all abstracts and full manuscripts matched the inclusion criterion previously set by the team. If more than one study were identified, meta-analysis of data was proposed.

The annual rate of having peritonitis among patients with PD was 0.3294 (SE 0.5729) [32] and the annual rate of having vascular access-related complications among patients with HD was 0.2698 (SE 0.5194) [19].

Based on an assumption of a short half-life of patients with palliative care (1–3 months) [19], the mean and SE of the monthly probability of ESRD patients dying without dialysis were 0.3331 and 0.1509, respectively.

Quality-Adjusted Survival

We adjusted outcomes for quality of life (QoL) using published utility estimates from a systematic search of similar databases and keywords used in outcome measures. We identified 12 studies reporting QoL for ESRD patients but subsequently excluded eight studies in which results were not presented to compare QoL among treatment strategies and/or were only qualitative descriptions. Finally, we included the four remaining studies, all of which interviewed ESRD patients. All studies employed multiattribute utility measurements, e.g., Health Utility Index, EQ-5D [33–36]. Meta-analysis using a random effects approach found that the QoL of ESRD patients without dialysis was 0.60 (SE 0.15). Without complications, the QoL of patients with PD was 0.72 (SE 0.08) and the QoL of patients with HD was 0.68 (SE 0.10). No study reported the QoL of patients with dialysis and complications. We applied the QoL of patients without dialysis to those with dialysis and complications.

Costs

Costs were derived from a systematic search of Thai published and gray literature, selecting the most up-to-date reports. They included all items of resources used, i.e., capital, labor and material costs of health-care providers, and real and opportunity costs lost by patients and relatives, i.e., patient treatment time, time for informal care, and cost for sick leave.

In addition, although most economic evaluations include future costs only for related illnesses [37], we considered that the cost of unrelated medical care might be considerable in this particular case because the interventions increase length of life and the patient population is likely to have comorbidities, e.g., diabetes or cardiovascular disease. Hence, we counted the cost of treating comorbidities as a program cost for each extended life-year (LY).

Briefly, the provider’s costs for offering HD and PD were from a national costing survey carried out by Tisayathikom et al. [38] and from a master’s degree thesis by Sriwajana [33], while the patient’s costs were from a study conducted by Homvijitkul [39]. We also conducted a micro-costing survey in two public hospitals (Prince of Songkla hospital and ChiangKhum hospital) and one private hospital (Viphavadee hospital) for cost items where no published data were identified.

All costs shown in Table 1 are reported in 2004 Thai Baht for each state of health within the Markov model. For intercountry comparison, costs have been converted into international dollars using purchasing power parity (PPP) US$ exchange rates at 1 US$ (2004) = 12.868 Thai Baht [40].

Uncertainty Analysis

A probabilistic sensitivity analysis using a second-order Monte Carlo simulation was carried out in Microsoft Office Excel 2003 (Microsoft Corp., Redmond, WA) [29]. All input parameters were assigned a probability distribution to reflect the feasible range of values that each input parameter could attain [41]. The beta-distribution was the choice of distribution for probability and utility parameters, which were bounded zero—one, and the gamma-distribution, which ensures positive values, was modeled for all rate and unit cost parameters. The normality on the log-odds scale with covariance matrix and Cholesky decomposition [42] was applied for survival parameters.

The simulation then drew one value from each distribution simultaneously and calculated cost and effectiveness pairs. This process was repeated 1000 times to provide a range of possible values given the specified probability distributions. Because there is no linear relationship between inputs and outputs in the Markov model [43], all costs, outcomes, and cost-effectiveness ratios provided in the Results section are expressed as the average value from the probabilistic sensitivity analysis.

Furthermore, because net benefits are not always normally distributed, a cost-effectiveness acceptability curve may wrongly present some strategies as suboptimal, but which have a higher probability of being cost-effective [44]. Thus, the results are presented as a cost-effectiveness acceptability (CEAcc) frontier, which determines the intervention giving the maximum expected net benefit for each value of the ceiling ratio [29].

We also determined an uncertain parameter’s importance using the utility theory framework. The analysis of partial expected value of perfect information (EVPI) was performed to determine whether different values of a particular input parameter lead to different optimum decisions, and if so, how much the expected loss under alternative optimum decisions varies [45,46].

If θ is the set of parameters for the model, with defined prior probability distributions; t is the set of possible decisions or strategies; and NB(t, θ) is the function of net benefit for decision t and parameters θ, the expected net benefit given no further information can be written as

maxt[EθNB(t, θ)]

The expected net benefit given full information is

Eθ[maxtNB(t, θ)]

and overall EVPI equals to

Eθ[maxtNB(t, θ)] − maxt[EθNB(t, θ)]

This is expressed in Baht per patient. To give a final value for the Thai context, we then multiplied by the incidence of 10,000 new ESRD cases per year and assumed that the current technology would be in operation for 10 years [19].

To quantify the value of obtaining further information on chosen parameters, partial EVPI is the difference between the expected value of a decision made with perfect information about a particular vector of the parameters (θ) and the current optimal decision [47].

With perfect information θi is the known vector of the parameters of interest, θ; then the expected benefit of a decision made would now be found by averaging over the uncertainty in θ that remains once we know θi and then by selecting the optimal treatment that provides maximum expected net benefit:

image

At this stage, however, we do not have perfect information on θi, so the expected value of a decision made with perfect information about θi is found by averaging over the uncertain ranges of the parameters θi and can be presented as

image

The additional value of collecting perfect information on a subset θi of uncertain model parameters is therefore given by the following equation:

image

The analysis of partial EVPI requires an explicit statement of the value of the ceiling ratio, and because, unfortunately, there is no such accepted threshold for adopting health technologies in Thailand, we applied the threshold that is recommended by the commission on Macroeconomics and Health, which suggests the use of three times of gross domestic product (GDP) per capita as the threshold for consideration in developing countries [48]. This would indicate a ceiling value in Thailand of 270,000 Baht per quality-adjusted life-year (QALY) based on 2004 Thai GDP and population [49].

Results

Using the government perspective, the total program cost of palliative care was 72,000 Baht, accounting for 32% of the program cost from a societal perspective (224,000 Baht). The palliative care option yielded 0.34 life-year (LY), or 0.20 QALY. These costs and outcomes of palliative care were equivalent for all age groups.

Compared with providing PD as the first-line treatment, the introduction of HD offered more health benefits, LYs saved, and QALYs gained, and was more costly for all age groups (Tables 2 and 3). Total lifetime costs borne by the NHSO accounted for 95% for offering PD as an initial treatment, and 85% for offering HD as an initial treatment, for all age groups, resulting in comparable economic evaluation results both using either the NHSO’s or a societal viewpoint. Given the effects of discounting on long-term costs, it is noteworthy that the lifetime costs derived from the model with unadjusted survival and the model with adjusted survival were similar. This is shown in Tables 4 and 5.

Table 2.  Lifetime cost for providing peritoneal dialysis and hemodialysis as an initial treatment by age group and view point of analysis
Age
(year)
Unadjusted survivalAdjusted survival
Peritoneal dialysisHemodialysisPeritoneal dialysisHemodialysis
Societal perspectiveNHSO’s perspectiveSocietal perspectiveNHSO’s perspectiveSocietal perspectiveNHSO’s perspectiveSocietal perspectiveNHSO’s perspective
  1. Costs are given to nearest 1000 Baht, 2004 price levels.

  2. The figures are not incremental but absolute numbers. Conversion to incremental values compared with palliative care can be performed by subtracting 224,000 Baht for a societal perspective and 72,000 Baht for the NHSO’s perspective for all age groups.

  3. NHSO, National Health Security Office.

208,277,0007,804,0009,286,0007,956,0008,277,0007,804,0009,286,0007,956,000
307,262,0006,823,0008,206,0007,013,0007,262,0006,823,0008,206,0007,013,000
406,135,0005,737,0006,997,0005,807,0006,135,0005,737,0006,997,0005,807,000
505,062,0004,826,0005,771,0005,028,0005,062,0004,826,0005,771,0005,028,000
604,096,0003,901,0004,696,0004,011,0004,096,0003,901,0004,696,0004,011,000
703,286,0003,126,0003,775,0003,235,0004,233,0003,126,0004,353,0003,235,000
Table 3.  Health outcomes of providing peritoneal dialysis and hemodialysis as an initial treatment by age group
Age
(year)
Unadjusted survivalAdjusted survival
Life-years gainedQALYs gainedLife-years gainedQALYs gained
Peritoneal dialysisHemodialysisPeritoneal dialysisHemodialysisPeritoneal dialysisHemodialysisPeritoneal dialysisHemodialysis
  1. The figures are not incremental but absolute numbers. Conversion to incremental values compared with palliative care can be performed by subtracting 0.34 for life-year gained and 0.20 for QALY for all age groups.

  2. QALYs, quality-adjusted life-years.

2017.6517.6512.2711.8917.6517.6512.3211.89
3015.3215.3210.6710.2515.3415.3410.7110.26
4012.9612.96 9.03 8.6712.9312.93 9.01 8.65
5010.6010.60 7.44 7.1410.5810.58 7.42 7.11
60 8.45 8.45 5.92 5.70 8.41 8.41 5.92 5.65
70 6.53 6.53 4.60 4.38 9.21 9.21 6.43 6.13
Table 4.  Results of economic evaluation of providing peritoneal and hemodialysis as an initial treatment compared to palliative care, using the societal perspective
Age
(year)
Unadjusted survivalAdjusted survival
Baht per life-year savedBaht per QALY gainedBaht per life-year savedBaht per QALY gained
Peritoneal dialysisHemodialysisPeritoneal dialysisHemodialysisPeritoneal dialysisHemodialysisPeritoneal dialysisHemodialysis
  1. Costs are rounded up to nearest 1000 Baht, 2004 price levels.

  2. QALY, quality-adjusted life-year.

20466,000525,000667,000777,000466,000525,000667,000777,000
30470,000533,000671,000795,000470,000533,000671,000795,000
40470,000539,000672,000803,000470,000539,000672,000803,000
50473,000543,000672,000806,000473,000543,000672,000806,000
60480,000555,000677,000821,000480,000555,000677,000821,000
70497,000575,000700,000850,000497,000575,000700,000850,000
Table 5.  Results of economic evaluation of providing peritoneal and hemodialysis as an initial treatment compared to palliative care, using NHSO’s perspective
Age
(year)
Unadjusted survivalAdjusted survival
Baht per life-year savedBaht per QALY gainedBaht per life-year savedBaht per QALY gained
Peritoneal dialysisHemodialysisPeritoneal dialysisHemodialysisPeritoneal dialysisHemodialysisPeritoneal dialysisHemodialysis
  1. Costs are rounded up to nearest 1000 Baht, 2004 price levels

  2. NHSO, National Health Security Office; QALY, quality-adjusted life-year.

20447,000456,000641,000675,000447,000456,000641,000675,000
30451,000463,000645,000691,000451,000463,000645,000691,000
40449,000455,000643,000678,000449,000455,000643,000678,000
50463,000483,000657,000714,000463,000483,000657,000714,000
60472,000486,000670,000716,000472,000486,000670,000716,000
70495,000512,000696,000759,000495,000512,000696,000759,000

Both for models with adjusted and nonadjusted survival, the incremental costs of providing the “PD first” option ranged from 466,000 Baht per LY saved or 667,000 Baht per QALY gained for patient aged 20 years to 497,000 Baht per LY or 700,000 Baht per QALY gained for patient aged 70 years using a societal perspective. Slightly higher, i.e., less favorable, cost-effectiveness and cost-utility ratios were observed if “HD first” is introduced in the UC benefit package.

Uncertainty Analysis

The results of the probabilistic sensitivity analysis using a societal viewpoint are presented by CEAcc frontier in Figure 4. We compared the simulation results for patients aged 20 years (black line) with those for patients aged 70 years (gray line), which are quite similar, with one exception. The CEAcc frontier suggests that for ceiling ratios less than 650,000 and 700,000 Baht per QALY for age groups 20 and 70 years, respectively, providing care without dialysis for ESRD patients was the most appropriate. If policymakers are, however, willing to pay more than 700,000 Baht per QALY for age group 20 years and 750,000 Baht per QALY for age group 70 years, providing “PD first” was the optimal choice.

Figure 4.

Cost-effectiveness acceptability frontier.

With the existence of uncertainty around input parameters of alternative treatment modalities, Figure 5 illustrates the expected opportunity loss of making a wrong decision for patients aged 50 years (the average age among ESRD patients in Thailand). Overall EVPI of treating 10,000 new ESRD cases per year and for a 10-year time period was highest (260,000 million Baht) at a ceiling ratio of 650,000 Baht per QALY.

Figure 5.

Population expected value of perfect information for a model using the 50 years age group with adjusted survival.

Because it was certain that palliative care was the best option at the ceiling ratio of 270,000 Baht per QALY (Fig. 5—overall EVPI = 0 at the ceiling ratio 270,000 Baht), there is no partial EVPI at that particular threshold. We, however, decided to examine the importance of each input parameter in the model, at least, as a proxy for future research in other settings. Thus, we took the point of maximum overall EVPI (ceiling ratio of 650,000 Baht per QALY) to quantify the expected additional value of collecting perfect information on each input parameter, partial EVPI.

Figure 6 reveals that, among all uncertain model parameters, the maintenance cost of the PD and HD options were the parameters with the first and second highest partial EVPI, respectively. Utility values for patients treated by PD and HD, costs of treating comorbidity, and survival function were among the important parameters.

Figure 6.

Expected value of perfect information (EVPI) for input parameters. pDoNot, probability of dying among patients with palliative treatment; pComPD, rate of having complications among PD; pComHD, rate of having complications among HD; pHDtoPD, probability of switching from HD to PD; pPDtoHD, probability of switching from PD to HD; SurvFunction, survival estimation; cPalliative, health-care cost of palliative care; cChroHD, health-care cost of hemodialysis; cChroPD, health-care cost of peritoneal dialysis; cPeritonitis, health-care cost of treating PD complications; cCoMorbid, health-care cost of treating comorbid conditions; cIdMDonot, direct non–health-care cost of palliative treatment; cIdMHD, direct non–health-care cost of hemodialysis; cIdMPD, direct non–health-care cost of peritoneal dialysis; cPatientHD, indirect non–health-care cost of hemodialysis; cPatientPD, indirect non–health-care cost of peritoneal dialysis; uPDnoCom, utility for PD without complication; uHDnoCom, utility for HD without complication; uCom, utility for patients with complications.

Because it is debatable whether to include future unrelated medical care costs resulting from lifesaving interventions when performing economic evaluations [37,50], in our analysis we thus assessed the impact of excluding the costs of treating comorbidities for each LY gained. We found that, without the costs of treating comorbidities, all ICERs of both PD and HD in terms of Baht/LY gained and Baht/QALY gained were reduced around 9% to 10% across age groups (because the model used no age-specific average medical care costs).

In the analysis using the alternative discount rates of 0% and 6%, we found that the rates above and below the actual discount rate used for the reference case (3.5%) had no impact on the overall conclusions—offering PD as an initial treatment was always superior to HD for all age groups, and providing treatment to younger patients was superior to that for elders (see Fig. 7). Furthermore, although there was not much difference among ICERs using different discount rates, it can be seen that the lower the rate of discounting the lower the ICER.

Figure 7.

Incremental cost-effectiveness ratios (ICER) using different discount rates by age group. Note: to present the differences of the results, we modified scale in x-axis starting with 500,000 Baht per QALY.

Discussion

The results indicate that the government should not include dialysis services for ESRD patients unless social willingness to pay or the ceiling ratio is at or higher than 700,000 Baht per QALY for young patients (age 20 years) and 750,000 Baht per QALY for older people (age 70 years). Although, there is no agreed threshold for adopting health technologies in Thailand, the threshold of 700,000 is almost three times higher than that recommend by the commission on Macroeconomics and Health.

If decision-makers are willing to pay 700,000 Baht per QALY or more, then providing PD as an initial treatment is the optimal decision because it dominated the HD option at any value of the ceiling ratio given the assumption of equal survival between PD and HD.

In subgroup analysis, it is interesting to note that providing treatments for younger ESRD patients resulted in a significant improvement of survival and gain of QALYs compared with the older age group. We found, however, that cost-effectiveness and cost-utility ratios among all subgroups were relatively similar. This could be explained, as Manns et al. [50] pointed out, by the fact that the inclusion of future costs, i.e., costs of dialysis of extended survival, would largely affect the economic evaluation results, so that even relatively inexpensive interventions that extend the survival of dialysis patients may not be cost-effective, as long as the treatment must be continued.

At this stage, no final decision has been made by Thai government whether to include PD or HD in the UC benefit package. We believe that this evidence needs to be considered alongside other policy issues, e.g., equity grounds, ethical principles, or financial implications, all of which are being considered. The NHSO is also considering launching a process of national consensus, in 2006, through a series of public hearings, to solicit public opinions on the rationing of care for ESRD patients in Thailand.

The results of this study are not in agreement with the conclusion made by Winkelmayer et al. [2]. They reviewed 13 economic evaluation studies on renal replacement therapy published between 1968 and 1998 and concluded that the cost-effectiveness of center hemodialysis were within a narrow range of US$55,000 to US$80,000 per LY saved (all figures were adjusted using PPP US$ exchange rate). Our study found a lower figure for Thailand (US$43,000 per LY saved). This finding could possibly be explained by the fact that the study of Winkelmayer et al. was dominated by evidence from developed countries, because 12 out of 13 studies were conducted in United States, United Kingdom, Canada, Sweden, New Zealand, and The Netherlands.

Nevertheless, our results are still higher than the results of a study conducted in Brazil [5], which was included in the work of Winkelmayer et al., and a newer publication from Malaysia [51]. Using a provider (NHSO) perspective, our study finds that the cost-effectiveness of PD and of HD were US$37,000 and US$38,000 per LY saved. The Brazilian study indicated that the cost-effectiveness of PD and of HD were US$12,000 and US$11,000 per LY saved, respectively, and a recent report by Hooi [50] revealed that the cost-effectiveness of PD and of HD in Malaysia were US$8000 and US$9000 (2001) per LY saved. These differences may partly be explained by the fact that the Brazilian and Malaysian studies did not include future medical costs of treating comorbidities from extended LYs.

There are some limitations regarding the availability of data used in the model. We carried out subgroup analysis by modeling age-specific survival from the TRT Registry, but we have no information available to model age-specific expenditures. Thus, the model applied the same provider and patient costs for all subgroups. Using non–age-specific costs may have underestimated or overestimated some cost items, e.g., cost for sick leave or cost of treating comorbidities, which are expected to vary across age groups.

Nevertheless, it is arguable that the magnitude of cost-effectiveness and cost-utility ratios were largely dependent on the estimated future cost of dialysis, PD or HD, and these costs were likely to be comparable across all subgroups. Also, although costs applied in the analysis included all items of resources used, the costs of staff training, which would be large for this particular form of care, were excluded.

Furthermore, we assumed independence between the occurrence of complications and switching between treatment modalities in the model. Thus, it is likely to have had an impact on the probabilistic sensitivity analysis and the CEAcc frontier.

Because we wanted to use Thai data and perform sub-age group analysis, and because there is no effectiveness evidence from randomized controlled trials or meta-analysis available to perform that, we had to estimate effectiveness from the registry database, which gives an equal survival for PD and HD. This approach, however, does not allow analysts to explore the uncertainty surrounding the effectiveness of each treatment and to quantify its impact and value for obtaining more information. As a result, we recommend that a high quality synthesis of the evidence on PD versus HD survival should be conducted.

Finally, the utility value for dialysis with complications is assumed equal to the utility of having no dialysis, which may have underestimated the uncertainty in this parameter. Nevertheless, given the rare incidence and relatively short duration of dialysis with complications, we still believe that the utility of dialysis with complications would have little effect on the overall results, as indicated by the partial EVPI analysis.

We believe that the methods and results of this study provide an important contribution to the evaluation of the value for money of alternative modes of management of ESRD patients in Thailand and other health-care settings, especially in developing countries, where such evidence is rarely available. The methods used in this study can be applied to future evaluations of health-care technologies.

There are two features of the methods that deserve particular emphasis. First, rather than evaluating PD or HD alone, we evaluated the value for money of providing PD or HD as the first-line treatment, which permits patients subsequently to move between treatment modalities. For example, we found that among people whose initial treatment modality was PD, 77% of their time on dialysis (LY saved) was from PD, while among people initially treated by HD 96% of their LY saved was from HD. This approach allows the evaluation of policy options known as “adoption decision” rather than assume adherence to a particular technology. In our model, both technologies would need to be available but used at different levels.

Second, the study extensively evaluated the effects of uncertainty around input parameters, which is common in decision modeling. The model is fully probabilistic and the results are comprehensive to handle the real-life situation faced by policy decision-makers who need to make a judgment under suboptimal information. We hope that this study would also facilitate the wider use of such methods in future health economic evaluation studies so that policymakers can take the benefits from the approach.

Acknowledgments

The authors would especially like to acknowledge and thank the review panel members whose expertise was invaluable throughout this study. We also thank Miss Sanya Srirattana and Miss Suwanna Mukem for their help in the literature review. Lastly, we are indebted to Prof. Andrew Briggs, Prof. Virasakdi Chongsuvivatwong, and three anonymous reviewers of the journal whose comments significantly improved our article.

Sources of financial support: We appreciate grant support from the ACCEPT (Assessing Cost and Cost-Effectiveness for Priorities in Thailand) project by the NHSO and the World Health Organization (WHO) country office, Thailand. The first author is currently supported by the WHO under the Fellowship Program award to study at the University of East Anglia.

Ancillary