The study population was estimated by the annual incidence of patients aged 25 years or older in Portugal evaluated by their GP with new signs and symptoms suggestive of CHF. The incidence rate by age group was estimated by Gouveia et al. based on the DisMod method, incorporating EPICA prevalence data.2 As more recent estimates were not available, epidemiologic data was updated accounting for changes in the demographic composition of the population in mainland Portugal in 2019, per recent data published by the National Institute of Statistics (INE).9

The model decision tree was based on current clinical guidelines published by the ESC, along with the local clinical practice in Portugal. Three scenarios were compared in the analysis. First, in the current standard of care (SoC) scenario, clinicians were assumed to rely on clinical judgment (patient history, risk factors, physical exam) to assess disease probability and request ECHO to confirm or exclude HF diagnosis. The SoC was compared to two scenarios where NT-proBNP was used according to the ESC algorithm, prior to ECHO evaluation, including when performed at an external lab facility (Lab) or performed at the health center with a point-of-care device (PoC).

The primary outcome was the estimated budget impact in terms of change in total costs for the NHS, including diagnosis (medical visits and exams), treatment and hospitalization related to decompensated CHF. The model only considered direct medical costs. A visual depiction of the decision tree analysis is shown in Figure 1.

Figure 1.

Model Structure.

(0,27MB).

In each of the three scenarios, once diagnosis was confirmed by ECHO, either as CHF with reduced (HFrEF) or preserved ejection fraction (HFpEF), the patient began treatment over the remaining of the one-year period. Also, following the ESC guidelines,3 it is assumed that HFrEF patients were treated with angiotensin-converting-enzyme inhibitors (ACEi) and betablockers (βB), whereas patients with HFpEF were treated with diuretics. Treatment costs for comorbidities, e.g., hypertension or coronary artery disease, were not considered. The time to return with test results from NT-proBNP (external lab) and ECHO was assumed to be 15 days after the test was ordered (per expert's opinion).

The clinical parameters evaluated in this analysis included the diagnostic test accuracy data for NT-proBNP, along with hospital and mortality data over the one-year time horizon of the model. The default values for each of these parameters are shown in Table 1.

All unit costs for diagnostic testing, consultations and hospitalizations (Diagnosis-Related Groups (DRG) 194 – heart failure) were extracted from Portuguese official/public sources.16–18 Hospitalization cost was based on the average of grades 1-2 assuming an average length of stay of 8-9 days.

The same cost for the NT-proBNP test (code 22578 – Natriuretic peptide – Type B) was applied to both the PoC and Lab settings, and according to the source methodology, this cost included the procedure performed by the healthcare professional.

The treatment cost was applied for the duration of patient treatment. Drug treatment assignment/market share was estimated based on 1) the expert panel opinion on the therapy of choice in clinical practice; 2) patient distribution from official drug monitoring platform from NHS19; and 3) recommended dosages from the ESC Guidelines,3 assuming the average dose when a range of dosages was presented. Only monotherapy formulations were considered, and therapy cost was weighted by the pharmacologic class administered to each patient group (HFpEF and HFrEF) and the market share of the main drugs within each therapy class. All unitary costs were collected from official data source corresponding to the reimbursement percentage of out-patient prices supported by the Portuguese NHS (Table 2).20

A one-way deterministic sensitivity analysis was conducted by varying different model parameters individually and observing the effect on the results in order to assess the impact of each variable and the robustness of the findings.21 The selected parameters and range over which the default values were varied can be seen in Table 3. The outcome tested was the incremental difference in costs for SoC vs. Lab and PoC.

The estimated annual incidence of patients with newly suspected CHF in Portugal was 81 012, of which 42 190 were expected to be confirmed with either HFrEF or HFpEF. Regarding resource utilization, compared with the SoC, using NT-proBNP in a Lab setting was estimated to save 2.5% (5620) medical visits, 26.8% (21 137) ECHOs, and 14.4% (1890) hospitalizations. In the PoC setting, the reduction of ECHO and hospitalizations was similar, but more medical visits were avoided, 86 633 (38%), since NT-proBNP results were immediately available at the first visit.

Time to diagnosis was shortened in the NT-proBNP groups, with nearly all true CHF cases (99%) being diagnosed and starting treatment at one month for NT-proBNP groups, compared to 38.6%, 43.2%, 18.2% at one, two and after six months in the SoC, respectively. Furthermore, 100% of subjects without CHF had the disease correctly discarded at one month for NT-proBNP groups (60% NT-proBNP-; 40% NT-proBNP+ECHO-), compared to 32% in the SoC. As more patients have CHF ruled-in or ruled-out in earlier phases for NT-proBNP groups, fewer medical visits were overall needed compared to SoC.

Earlier treatment initiation in the NT-proBNP groups was associated with positive effects on hospitalizations and mortality, as more patients benefited from being more time on treatment. Importantly, the number of patients diagnosed with CHF at first hospitalization due to decompensation was estimated to fall from 5.2% in the SoC to less than 1% in NT-proBNP arms. The effect on mortality was also noticeable, with rates of 2.9% and 2.5% for SoC and NT-proBNP arms, respectively, representing 340 fewer deaths at one year.

The total annual expected cost for the current SoC scenario was EUR 37 660 910 (Table 4). The introduction of NT-proBNP in the Lab setting was expected to save an estimated EUR 935 657 in annual costs, which is a 2.5% reduction in the estimated budget spent in the SoC scenario. When NT-proBNP was performed in PoC setting, it was expected to save an estimated EUR 2 982 443 in annual costs, or 7.9% of the SoC budget.

The Lab and PoC settings led to similar reductions of hospitalization costs (EUR 2 702 825) and ECHO (EUR 860 265), but the reduction of office visits related costs was higher in the PoC setting (EUR 2 685 574 compared to EUR 174 213), resulting in overall higher savings compared to Lab.

The results of the model were found to be robust to all changes in parameter values (Figure 2). The use of NT-proBNP retained savings in nearly all scenarios – the only exception being in Lab vs. SoC analysis, where the sensitivity of NT-proBNP for CHF was reduced to 57% (Table 3), causing an increment of 26.8 thousand euros. All other scenarios generated savings ranging from 489.6 thousand to 1.3 million euros in the Lab setting, and 2.2 million euros to 3.4 million euros in the PoC settings.

Figure 2.

One-way sensitivity analysis results (tornado diagram - cost savings estimation).

The base case budget impact analysis yielded EUR 2 982 443 in savings in a point of care setting (a) and EUR 935 657 in the laboratory setting (b), with the bars on the tornado diagram representing the deviation from this base case when the default parameter value was changed. Positive values indicate costs that are avoided (cost savings).

CHF: congestive heart failure; HF: heart failure.

(0,42MB).

Overall, there was a much greater magnitude of change in the results (about three times greater) when varying parameter values in the Lab vs. PoC sensitivity analysis. This is because of the number of visits in correlation with epidemiological data, e.g., a 1% increase in the prevalence data corresponded to 810 more cases, which in turn corresponded to 810 additional visits to the Lab, compared to the PoC arm.

The model is a simulation of the CHF diagnostic journey in Portugal, which implies some simplification of the clinical pathway that is intended to apply to the majority of patients with suspected CHF. Furthermore, there are complexities and additional factors that may not have been considered in this model that apply to certain clinical scenarios and subpopulations.

Some conservative assumptions were made in this analysis that could potentially lead to greater cost savings. First, the analysis covers a short time period of only one year and as CHF is a chronic condition, the potential economic advantage of earlier diagnosis and treatment of CHF patients could go well beyond a one year by slowing down disease progression and avoiding costly downstream hospitalizations and procedures. Moreover, this analysis only includes the consequences of diagnosing and treating CHF patients earlier – thereby avoiding unnecessary ECHO testing and hospitalizations. However, greater cost savings and clinical benefits are expected when considering the pool of patients without CHF that would receive their proper diagnosis and treatment earlier. In addition, emergent treatment classes (angiotensin receptor neurolysin inhibitors and sodium-glucose transport protein 2) can lead to greater decreases in hospitalizations and mortality beyond that anticipated by our model. Aligned with the perspective of the NHS, this model only captures direct costs, and therefore does not account for indirect costs, such as lost productivity or impact on quality of life, for which there could also be a substantial benefit.

The main limitation of this analysis is that performance data for NT-proBNP are not available in the primary care setting in Portugal. We selected the 125 pg/mL cut-off value recommended by the ESC and used sensitivity/specificity data from a recent meta-analysis based on three studies in primary care with NT-proBNP.12 As different cut-offs may be used in other countries, this must be considered when comparing the results versus international literature.28 Also, the sensitivity/specificity data used refer to NT-proBNP from a specific manufacturer (Roche®), and therefore results are not directly transferable to other commercially available natriuretic peptides.

Several other variables were inputted from different sources that may introduce some bias to the model. However, the most critical variables – prevalence, distribution of visits, hospitalization rate, test sensitivity and specificity – were tested in different sensitivity analysis over a broad range of values that showed preserved savings in most scenarios, suggesting the results of the base case are robust.