Cost-Effectiveness of Rheumatic Heart Disease Echocardiographic Screening in Brazil: Data from the PROVAR+ Study

Introduction: In recent years, new technologies – noticeably ultra-portable echocardiographic machines – have emerged, allowing for Rheumatic Heart Disease (RHD) early diagnosis. We aimed to perform a cost-utility analysis to assess the cost-effectiveness of RHD screening with handheld devices in the Brazilian context. Methods: A Markov model was created to assess the cost-effectiveness of one-time screening for RHD in a hypothetical cohort of 11-year-old socioeconomically disadvantaged children, comparing the intervention to standard care using a public perspective and a 30-year time horizon. The model consisted of 13 states: No RHD, Undiagnosed Asymptomatic Borderline RHD, Diagnosed Asymptomatic Borderline RHD, Untreated Asymptomatic Definite RHD, Treated Asymptomatic Definite RHD, Untreated Mild Clinical RHD, Treated Mild Clinical RHD, Untreated Severe Clinical RHD, Treated Severe Clinical RHD, Surgery, Post-Surgery and Death. The initial distribution of the population over the different states was derived from primary echo screening data. Costs of the different states were derived from the Brazilian public health system database. Transition probabilities and utilities were derived from published studies. A discount rate of 3%/year was used. A cost-effectiveness threshold of $25,949.85 per Disability Adjusted Life Year (DALY) averted is used in concordance with the 3x GDP per capita threshold in 2015. Results: RHD echo screening is cost-effective with an Incremental Cost-Effectiveness Ratio of $10,148.38 per DALY averted. Probabilistic modelling shows that the intervention could be considered cost-effective in 70% of the iterations. Conclusion: Screening for RHD with hand held echocardiographic machines in 11-year-old children in the target population is cost-effective in the Brazilian context. Highlights: A cost-effectiveness analysis showed that Rheumatic Heart Disease (RHD) echocardiographic screening utilizing handheld devices, performed by non-physicians with remote interpretation by telemedicine is cost-effective in a 30-year time horizon in Brazil. The model included primary data from the first large-scale RHD screening program in Brazilian underserved populations and costs from the Unified Health System (SUS), and suggests that the Incremental Cost-Effectiveness Ratio of the intervention is considerably below the acceptable threshold for Brazil, even after a detailed sensitivity analysis. Considering the high prevalence of subclinical RHD in Brazil, and the significant economic burden posed by advanced disease, these data are important for the formulation of public policies and surveillance approaches. Cost-saving strategies first implemented in Brazil by the PROVAR study, such as task-shifting to non-physicians, computer-based training, routine use of affordable devices and telemedicine for remote diagnosis may help planning RHD control programs in endemic areas worldwide.

A couple of assumptions need to be made for the use of this formula. The first assumption is that the disease in the target population is in a steady state; prevalence remains stable over a longer period of time and the inflow and outflow of diseased are equal. While the study by Nascimento et al. shows this is not the case [3], the peak prevalence of the study (49 per 1.000 children for borderline cases and 9.3 per 1,000 for definite cases) was assumed to equal stable prevalence. The second assumption is that the average duration of disease for Asymptomatic Borderline RHD is estimated to be approximately 3.63 years, and the average duration of disease for asymptomatic definite RHD 10.72 years. This assumption is estimated by using the yearly probability of staying in state B: Undiagnosed Asymptomatic Borderline RHD and respectively state C: Asymptomatic Definite RHD and calculate at what time more than 99.5% of a hypothetical cohort stopped being affected by their respective affliction of RHD (disease could have either regressed to no RHD or progressed to worse health states, such as definite RHD). Then the average duration of disease time was calculated for 99.5% of this hypothetical cohort.
For adults above 20 years old it was assumed that the probability of being healthy to getting Asymptomatic Definite RHD is halved. Furthermore, since it is by definition not possible for adults to have Asymptomatic Borderline RHD, the pathway to this state was cut for this group. Lastly, the proportion of the population with the age of 19 that was still in the Asymptomatic Borderline RHD was transferred to state A: No RHD in the Markov model.

Almost all the transition probabilities for state B: Undiagnosed Asymptomatic
Borderline RHD to various states were derived from primary survival data from a study conducted in Uganda by Beaton et al. [26] using the msSurv package for the statistical program R [37]. The disease course of asymptomatic borderline cases from this study are equal to the disease course of Asymptomatic Borderline cases in this model. The transition probabilities were calculated by first calculating the probability of transitioning to different states between the start of the study and the third year, and then transforming that probability to yearly probabilities. The transition probability from state B: Undiagnosed Asymptomatic Borderline RHD to state Z: Death was assumed to be equal to State A: No RHD to state Z: Death. The reason for this is that the data showed a lower mortality rate than the standard mortality rate for a healthy individual in this model, which was deemed unrealistic. To control for the variance of state B: Undiagnosed Borderline RHD a dirichlet distribution was used. This distribution was centered on the yearly probabilities calculated above. To calculate parameters for the probabilistic sensitivity analysis a population size of 162 was used, which is the sample size of borderline RHD patients in the study by Beaton et [38] for their states. It was assumed that the Asymptomatic Definite RHD ceases identified in the study by Beaton et al. [26] were similar to the mild RHD cases identified in the paper by Cannon et al. [25]; in both cases one can speak of an asymptomatic, clinically trivial to mild valve disease, only with findings in auscultation. Under this assumption, it was possible to 'connect' both the databases as there is an overlap of Asymptomatic Definite RHD in both the Beaton et al. study [26] and the Cannon et al. study [25]. Yearly transition probabilities were calculated by using the formula mentioned at the start of the Appendix to estimate yearly rates based on 10-year transition probabilities, and then recalculate those rates to yearly transition probabilities. The transition probabilities for State D: Untreated Mild Clinical RHD are based on the moderate state in the study by Cannon et al. [25]. Transition probabilities for state E: Untreated Severe RHD were based on the severe state of the study [25]. However, given that the patients in the study by Cannon et al. [25] were under treatment, the transition probability from state E: Untreated Severe RHD to state X: Surgery was adjusted to mimic the natural discourse. This number was calculated based a study by Albuquerque et al. [39]. As an estimate, it was assumed that because of medication non-adherence, for every 70 patients an additional 30 patients need surgery, given that 30% of the heart failure decompensation could be attributed to medication non-adherence by the patients.  [40] shows that 65% of the children adhere to secondary prophylaxis in asymptomatic RHD, while a study by Mussi et al. [41] shows that in standard care 60% of the symptomatic participants adhered to their medication. In other words, 65% of the total transition from state C: Untreated Asymptomatic Definite RHD to Clinical RHD goes to state H: Treated Mild Clinical RHD and 60% of the total transition probability from state D: Untreated Mild Clinical RHD to severe RHD goes to state I: Treated Severe Clinical RHD. It was assumed that once patients adhered to a certain treatment that they would continue to adhere to further treatment if their diseased progressed. As an example, this means that all the patients for which the disease progresses in state H: Treated Mild Clinical RHD go to state I: Treated Severe Clinical RHD. This assumption was made because once the decision between adherence and non-adherence was made, people did not seem to change their minds as can be seen in the follow-up study by Mussi et al. [41].
State X: Surgery represents the proportion of the population in which RHD has developed to such an extent that surgery is needed. The yearly mortality risk of surgery (the risk of dying before, during and after the operation during one year) is derived from the study by Xavier et al. (Medium-term outcomes of 78,808 patients after heart valve surgery in a middle-income country: A nationwide population-based study). State X: Surgery is modelled in such a way that when a patient gets surgery, he can only stay for one year in this state. The reason for modelling a one-year transition state as opposed to an immediate surgery is the waiting times for surgery in the public health system in Brazil [42].
Transition probabilities from state K: Post Surgery to state Z: Death and state X: Surgery were derived from a study by Ribeiro et al. [43]. It was assumed that after surgery you stayed in a state that yielded high cost and disutility, because of medical consults, medicine and a not fully functioning heart. Patients in this state are also at risk of one or multiple resurgeries, which is reflected in the transition probability from state K: Post Surgery to state X: Surgery.

External validation
The model predicted a prevalence of Asymptomatic Definite RHD of 2.26 per 1,000 for the age group 11 to 13.9 and a prevalence of Asymptomatic Definite RHD of 6.36 per 1,000 for the age group of 14 to 17.9 in the target population. Observed prevalence [3) in these age groups were respectively 3.1 per 1,000 in the age group 11 to 13.9 and 9.3 per 1,000 in the age group 14 and older. The predicted prevalence of Asymptomatic Borderline RHD in the age groups 11 to 13.9 and 14 to 17.9 were respectively 33.62 per 1,000 children and 45.63 per 1,000 children. Observed prevalence [3] for these age groups were respectively 35 per 1,000 for the age group 11 to 13.9 and 49 per 1,000 for the age group 14 and older.

Costs
Costs for state R: Resolved RHD depend on from which state the specific patient For the intervention, it is assumed that 75 images are made per day, which amounts to 9.375 images per hour assuming an eight-hour workday. In these five years, it was assumed that the screening could be conducted for an 8.5-month period per year, when one takes into account the school holidays in Brazil and other national free days. It was assumed that in a week, the team could effectively screen for six hours per day for five days. By multiplying the total amount of work hours per year with the number of images made per hour (1,635 work hours per year, 9.375 images per hour) a total of 15,236 scans were assumed to be made per year.
In order to acquire this number of images, a team of two research nurses, an imagining technician, a biomedical technician and a physician (echo expert) are required. All of the cases with a positive handheld screening result will be referred to a cardiologist for further screening. There, they will get an echocardiogram and a cardiology consult. Cases diagnosed with definite RHD will receive penicillin. Due to the specificity of handheld screening (65%) [44], a proportion of the population will get a false-positive diagnosis. These cases will be evaluated as well, which results in extra costs. These costs, which are similar to cost of follow up for diagnosed borderline RHD are presented in Appendix Table 3 for false-positive and borderline RHD cases and Appendix Table 4 for definite RHD cases.
The variance of the cost parameters is based on the variance of the cost of surgery.
Given that surgery is relative complex compared to the other treatments, it can be expected that this variance is higher than the variance of the other treatments. This might result in an over-estimation of the uncertainty the costs of treatments other than surgery.

Sensitivity analysis
The sensitivity of the base-case treatment adherence (65%) for the treatment of asymptomatic definite RHD was tested by decreasing the adherence to 33% and increasing it to 100%. The sensitivity of treatment adherence of severe clinical RHD (base case 60%) was tested by decreasing the parameter to 30% and increasing it to 100%. A table of the base-case assumption and the sensitivity analysis around the base case can be found in Appendix Table   8. It was assumed that there was no loss to follow-up after surgery.
The full results of the one-way sensitivity analysis of the parameters can be found in Appendix Table 9. The results are ordered based on the magnitude of change; the parameters that showed the biggest change are at the top. An explanation about how the sensitivity analysis was conducted can be found in the methods section.  The source of the costs is primary data from the PROVAR project. The source of the costs is primary data from the PROVAR project. The source of the costs is primary data from the PROVAR project.