Intervention, Population, and Setting
The RHP intervention is based on case-management of CKD patients through the frequent and regular observance of the estimated Glomerular Filtration Rate (eGFR) and microalbuminuria levels to prevent the progression of the renal disease, control of comorbidities status (diabetes and hypertension), and promotion of healthy lifestyles. The RHP was implemented in the Hospital E. Rebagliati Network, which included the hospital and primary care centers. Target patients were adults with a diagnosis of CKD. Patients were classified in 5 states of disease progression according to the KDIGO guidelines(10) which determines specific aspects of the intervention delivery. In each visit, they received attention from the physician, nurse, and nutritionist, and performed an eGFR test for microalbuminuria detection. Additionally, some patients received social assistant and psychology care. Patients were required to visit a primary care facility in a variant frequency depending on their CKD stage: once a year for stages 1 and 2, twice for stages 3a and 3b, and three times for patients in stage 4. Patients in the early stages of the CKD (1-3a) must receive attention in primary care, while late stage patients receive attention in specialized facilities.
Modelling approach
We performed a deterministic CEA comparing the cost and health consequences of RHP to the usual care. While the intervened patients follow a treatment protocol and are closely followed-up, usual care patients receive treatment based on their demand. Usual care patients do not receive regular laboratory tests nor CKD-specific case-management consultation in other clinical services beyond Nephrology. To be considered intervened, patients had to meet a threshold of attendance during the first year, defined by expert consultation and depending on CKD stage.
Our CEA used a Markov model defined by three health states: CKD (all stages), Dialysis and Death.(Figure 1) This study used the time-to-event variables from the previous RHP epidemiologic study to estimate both survival curves (to dialysis and overall mortality) in the control group, and the estimated hazard ratios to use them as treatment effects(7). The time scale in the previous study was days, in this study we used years. Due to the change in units, survival rates might suffer small changes that are negligible.
We used parametric methods that allow us to project the survival proportions beyond the five years, from 2013 to 2018, of observed data, corresponding to CKD diagnosed patients, beginning of RRT, and overall mortality. From the many potential distributions, we would select the one with the best goodness of fit using information criteria indicators the Akaike and Bayesian Information Criterion. Using the results of the selected distribution we estimate, for each outcome, lambda and gamma as: 1/(exp (intercept)^(1/scale)) and 1/scale, respectively. These two parameters would allow us to estimate the survival probabilities from CKD to dialysis and death for each cycle in the usual care following this formula: 1-exp (lambda * (cycle - 1)gamma-cyclegamma). This is the probability to transit from the first health state to the other ones, expressed as a function of the cycle that accounts for the increasing risk to event alongside the time. For the RHP alternative, we used the same transition probabilities in each cycle, but a treatment effect was applied to account for the effect of the intervention. The probability to transit from dialysis to death was estimated through literature review.
Since our objective was to model, the lifetime consequences of both competing alternatives and considering that the mean age of the population of interest is around 60 years old, our model ran for 30 cycles, where each cycle represents one year. Cost and outcomes were observed at the end of each cycle. We used a simulated cohort of 1,000 people for each strategy, replicating the same CKD stage distribution and diabetes prevalence as in the original dataset for representative purposes and internal validity.
Cost assessment
Costs were estimated from the payer perspective, EsSalud, considering direct medical costs for both alternatives. We considered all costs faced by the payer to provide treatment in one year. For the intervention, we considered the costs of nephroprotection treatment, outpatient visits, and laboratory tests. These costs are not homogeneous across stages of CKD but vary due to frequency of provision (hospital visits and lab tests), and type of facility (primary care facilities have cheaper provision costs than more specialized ones), and patients with diabetes receive additionally glycosylated hemoglobin tests. We also included the implementation cost of the intervention, including a first investment that includes the time utilized by the Renal Health Unit to develop the intervention, protocol, guidelines, personnel training, and a yearly operational cost that includes the time spent by the RHP team identifying, testing, referring, and keeping accurate records of the patients.
Patients on the usual care receive on-demand outpatient visits, and subject to medical indication for laboratory tests. There was not a fixed frequency for neither of those services, contrary to the intervention. Thus, healthcare utilization would depend on the patients’ behavior and preferences. Given the inherent randomness of this healthcare utilization variables, we decided to base our estimates in expert consultation. The nephrologists from the Renal Health Unit provided us with their best-educated guess of the number of outpatient visits that a regular CKD patient receives in one year. We also added the cost of one laboratory test per year without differentiation for diabetes condition or facility in which it would take place. This allow us to keep the usual care costs low, to obtain conservative estimations for the Incremental Cost-Effectiveness Ratio (ICER).
Since the costs vary across CKD stages in both alternatives, and across diabetes diagnose in the RHP, but the Markov mode used one compartment in the Markov model for all CKD stages, we use a weighted average to estimate the annual treatment cost per patient in each alternative. The weight was defined by the proportion of stages and diabetic patients in the observed data. Finally, we included the annual cost of hemodialysis by considering the cost of the session and the drugs provided in each one, and the number of sessions in a year.
We used two sources of cost estimation: the EsSalud General Management Office cost report (2018), and the report of resources use specifically for the intervention from the Renal Health Unit (2014). The first one provided the unit cost per activity, while the latter gives us the number of units per activity consumed to follow and treat a regular patient in each CKD stage. We used institutional costs to reduce uncertainty around the final estimations. Data collection was conducted in local currency, Peruvian Soles (PEN), while results are presented in United States Dollars (USD, $).
Health consequences
We sought to compare the differences in health outcomes between the alternatives using years lived free of dialysis (YL) and Quality Adjusted Life Years (QALY), to obtain a measure of the number of person-years avoided in dialysis and the number of person-years of perfect health gained, associated with the adherence to the intervention. The utility scores were obtained through literature review.
Analysis
We projected the costs and health consequences of each alternative separately during 30 cycles, each one equivalent to 1 year. Cost and health outcomes would be discounted by an annual rate of 3% in order to reflect the time-preferences of the economic agents. After these calculations, we aggregated the total costs, YL and QALY from each alternative and express them in per-person units.
To address the fundamental question of which alternative poses the highest economic value we use the Incremental Cost-Effectiveness Ratio (ICER) calculated as the difference in cost between RHP and usual care over the differences in health outcomes. Then, we can interpret the ICER as the additional cost for the payer associated to avoid one person-year in dialysis and to gain one QALY. Final estimations are made in local currency but converted to USD using a fixed exchange rate of 3.3 PEN per each USD. We use the CE threshold of 1 to 3 times the gross domestic product (GDP) per capita, estimated in $6,571 for Peru, according to the World Bank.
To assess the robustness of the ICER estimation, we performed a Probabilistic Sensitivity Analysis (PSA) based on a Monte Carlo simulation of 1,000 repetitions. In each repetition, the model randomly picked a value for each pre-defined changing parameter within its range value and considering its distribution, to create a unique context in which the RHP is evaluated against the usual care. The changing parameters were treatment effects (over both dialysis and mortality), cost of treatment, costs of dialysis, utility score of CKD, utility score of dialysis, and discount rate. The range of values for each parameter was as follows: the treatment effects would take the lower and upper values of the estimated confidence interval, the costs would vary 15%, the utilities would change 10%, and the discount rate would take 0% to reflect no discounting and 5% to reflect a scenario with higher opportunity cost. We summarized the results by descriptive statistics of the ICER per QALY distribution and a figure showing the incremental costs and QALYs for each simulation.