Development of a health economic model to evaluate the potential benefits of optimal serum potassium management in patients with heart failure

Abstract

Aims: Patients with heart failure are at increased risk of hyperkalemia, particularly when treated with renin-angiotensin-aldosterone system inhibitor (RAASi) agents. This study developed a model to quantify the potential health and economic value associated with sustained potassium management and optimal RAASi therapy in heart failure patients.

Materials and methods: A patient-level, fixed-time increment stochastic simulation model was designed to characterize the progression of heart failure through New York Heart Association functional classes, and predict associations between serum potassium levels, RAASi use, and consequent long-term outcomes. Following internal and external validation exercises, model analyses sought to quantify the health and economic benefits of optimizing both serum potassium levels and RAASi therapy in heart failure patients. Analyses were conducted using a UK payer perspective, independent of costs and utilities related to pharmacological potassium management.

Results: Validation against multiple datasets demonstrated the predictive capability of the model. Compared to those who discontinued RAASi to manage serum potassium, patients with normokalemia and ongoing RAASi therapy benefited from longer life expectancy (+1.38 years), per-patient quality-adjusted life year gains (+0.53 QALYs), cost savings (£110), and associated net monetary benefit (£10,679 at £20,000 per QALY gained) over a lifetime horizon. The predicted value of sustained potassium management and ongoing RAASi treatment was largely driven by reduced mortality and hospitalization risks associated with optimal RAASi therapy.

Limitations: Several modeling assumptions were made to account for a current paucity of published literature; however, ongoing refinement and validation of the model will ensure its continued accuracy as the clinical landscape of hyperkalemia evolves.

Conclusions: Predictions generated by this novel modeling approach highlight the value of sustained potassium management to avoid hyperkalemia, enable RAASi therapy, and improve long-term health economic outcomes in patients with heart failure.

Keywords:

• Hyperkalemia
• potassium
• heart failure
• renin-angiotensin-aldosterone system inhibitor
• economic modeling

Introduction

Potassium (K⁺), the most abundant cation in the human body, is vital to ongoing electrophysiological function. Serum potassium concentrations are tightly controlled through several renal and extra-renal mechanisms, in order to remain within the narrow homeostatic range of 3.5–5.0 mmol/L (3.5–5.0 mEq/L). Levels below or above this physiological range are defined as hypokalemia and hyperkalemia, respectively^1,2. Both are considered acute electrolyte imbalances, each associated with cardiac dysrhythmias, neuromuscular dysfunction, and increased mortality^3,4.

Hyperkalemia may result from excessive potassium intake, pathologies that increase intracellular potassium release, or, most notably, ineffective potassium clearance. Subsequently, renal insufficiency represents a significant predictive risk factor for the development of hyperkalemia, either alone or as a consequence of heart failure^5–7. Neurohormonal activation of the renin-angiotensin-aldosterone system (RAAS) is implicated in the pathogenesis of heart failure; chronic activation of this compensatory pathway consequently leads to organ fibrosis, renal dysfunction, and increased susceptibility to hyperkalemia⁸. Recent evidence suggests that heart failure patients who develop hyperkalemia are generally older and more comorbid than those who do not develop hyperkalemia⁹. Nevertheless, the incidence of hyperkalemia in heart failure patients has been associated with increased risks of hospitalization and mortality, and, consequently, imposes a significant burden on this population^9–13.

Given its role in heart failure and related cardiorenal conditions, the RAAS blockade is a widely advocated strategy to reduce heart failure symptoms, manage causative factors, and delay disease progression. On the basis of large, well-conducted clinical trials, RAAS inhibitors (RAASi), including angiotensin converting enzyme (ACE) inhibitors and angiotensin receptor blockers (ARBs), are established treatments for the first-line management of chronic heart failure^14–16, while mineralocorticoid receptor antagonists (MRAs) added to RAASi therapy in the second-line setting have demonstrated additional benefits^17,18. Treatment with these agents at guideline-recommended doses significantly improves cardiovascular morbidity and mortality; however, there is a disparity between clinical guidelines and the real-world implementation and dosing of RAASi and MRA therapies^19–21.

Hyperkalemia represents the major barrier for prescribing RAASi and/or MRAs at target doses, as RAAS blockade attenuates potassium excretion and exacerbates the risk of hyperkalemia in an already-vulnerable heart failure population^12,21,22. In patients who experience hyperkalemia whilst treated with RAASi and/or MRAs, clinicians often choose to down-titrate or discontinue guideline-recommended therapy^17,23. However, literature suggests that discontinuation or submaximal dosing of RAASi is associated with worsening clinical outcomes in patients with heart failure or other comorbidities^{14,20,24–27}. Effective management of serum potassium levels may, therefore, address an important clinical need in this population.

As the clinical importance of managing hyperkalemia risk is increasingly recognized, there is a need to quantify the long-term health economic burden of hyperkalemia in heart failure, the consequences of sub-optimal RAASi therapy, and the associated value of maintaining optimal serum potassium levels. Computer simulation modeling represents a valuable method to simulate disease progression over a lifetime horizon, predict long-term outcomes, and inform clinical and reimbursement decision-making. Despite this, models evaluating the inter-relationships between heart failure progression, serum potassium levels and RAASi and/or MRA use have not previously been published in peer-reviewed literature.

To our knowledge, we have developed the first natural history model for heart failure where serum potassium levels, in addition to RAASi and/or MRA use, were associated with long-term health economic outcomes. Model development was guided by available literature, in conjunction with an advisory panel with expertise in nephrology, cardiology, diabetes, general medicine, and health economic modeling. Following validation of its predictive capabilities, an application of the model sought to explore the health economic consequences of discontinuing RAASi therapy to avoid hyperkalemia, and quantify the value of optimizing both serum potassium management and RAASi therapy in heart failure patients.

Methods

Model development

A computer simulation model was designed to characterize the progression of heart failure across New York Heart Association (NYHA) functional classifications, and predict long-term health and economic outcomes according to serum potassium levels and/or RAASi use. Co-authors with clinical expertise, in collaboration with health economists, guided the structure of the model, data inputs, sensitivity analyses, transition probabilities, and assumptions; and established the clinical face validity of model outputs.

Model structure

A patient-level, fixed-time increment stochastic simulation was developed in Microsoft Excel to model the natural history of heart failure progression over a lifetime horizon. As shown in Figure 1, disease progression was modeled according to transitions between NYHA functional classifications (I–IV), using monthly probabilities sourced from Yao et al.²⁸. No relationships were modeled between heart failure progression and either RAASi use or serum potassium levels, since no suitable data were identified. Simulated patients progressed through the model until death from disease-specific or general causes.

Figure 1. Flow diagram summarizing heart failure health states and incident events implemented within the model. Abbreviations. NYHA, New York Heart Association; RAASi, renin-angiotensin-aldosterone system inhibitor.

Time-dependent risk factors

In addition to heart failure progression, the model simulates both natural decline in renal function and fluctuations in serum potassium concentration. In the absence of heart failure-specific data to inform the decline of estimated glomerular filtration rate (eGFR), a linear decline consistent with the general population was applied²⁹. To account for inter-patient variability in serum potassium levels, time-dependent potassium trajectories were modeled at the patient-level using mixed-effects regression models. Fixed effects parameters captured overall potassium trends at the cohort-level, while random effects terms allowed individual patients to exhibit a unique, fluctuating serum potassium profile that may deviate from the cohort average. Further details of the methods used to estimate fluctuations in serum potassium levels are provided in the Supplementary Material.

Modeled events

The incidence of hospitalization, arrhythmia, and mortality, and the association between serum potassium levels and/or RAASi use and such risks, were modeled using inputs sourced from published literature (Table 1). Monthly probabilities for hospitalization according to NYHA stage were informed by Ford et al.³⁰, while the likelihood of arrhythmia among heart failure patients was based on a monthly probability reported by Colquitt et al.³¹. The association between RAASi use and hospitalization events was modeled according to odds ratios consistent with Flather et al.³². Hyperkalemia and hypokalemia are each associated with arrhythmia and related morbidity and mortality; however, due to a paucity of heart failure-specific data describing the association between serum potassium and the risks of hospitalization and arrhythmia, incident rate ratios for hospitalization and major adverse cardiovascular events in patients with chronic kidney disease were applied, respectively³³.

Table 1. Model inputs relating to hospitalization and mortality events, according to heart failure stage, serum potassium sub-group, and RAASi use.

Event		Mean (SE)				Source
Hospitalization	Monthly probability of hospitalization by heart failure stage^a
	NHYA I	0.015 (0.0015)				Ford et al.^30
	NHYA II	0.024 (0.0024)
	NHYA III	0.024 (0.0024)
	NHYA IV	0.154 (0.0154)
	Odds ratio for hospitalization by RAASi use
	Any vs none	0.670 (0.033)				Flather et al.^32
	IRR for hospitalization by K^+ sub-group (mEq/L)^b
	eGFR (mL/min/1.73m^2):	<30	30–40	40–50	50–60	Luo et al.^33^d
	K^+ < 3.5	1.93 (0.45)	1.77 (0.34)	2.24 (0.38)	2.06 (0.27)
	K^+ 3.5–3.9	1.65 (0.22)	1.35 (0.16)	1.23 (0.13)	1.13 (0.08)
	K^+ 4.0–4.4	0.93 (0.09)	0.99 (0.08)	1.08 (0.07)	1.01 (0.05)
	K^+ 4.5–4.9	1 (0)	1 (0)	1 (0)	1 (0)
	K^+ 5.0–5.4	1 (0.10)	0.96 (0.09)	1.07 (0.09)	1 (0.07)
	K^+ 5.5–5.9	1.34 (0.18)	1.07 (0.14)	1.23 (0.18)	0.81 (0.11)
	K^+ ≥ 6.0	3.65 (0.58)	1.82 (0.44)	1.91 (0.56)	1.07 (0.30)
Arrhythmia	Monthly probability of arrhythmia
	Any NYHA class	0.008 (0.004)				Colquitt et al.^31
	IRR for arrhythmia by K^+ sub-group (mEq/L)^b
	K^+ < 3.5	1.89 (0.09)				Luo et al.^33^d
	K^+ 3.5–3.9	1.27 (0.04)
	K^+ 4.0–4.4	1.04 (0.01)
	K^+ 4.5–4.9	1 (0)
	K^+ 5.0–5.4	1.01 (0.02)
	K^+ 5.5–5.9	1.12 (0.04)
	K^+ ≥ 6.0	1.88 (0.12)
Mortality	Probability of mortality
	Any NYHA class	As calculated by SHFM				SHFM^34
	Hazard ratio for mortality by K^+ sub-group (mEq/L)^b
	K^+ < 3.5	2.19 (0.59)				Krogager et al.^11^e
	K^+ 3.5–3.8	1.91 (0.40)
	K^+ 3.9–4.2	1 (0)
	K^+ 4.3–4.5	1.10 (0.23)^c
	K^+ 4.6–5.0	1.47 (0.27)
	K^+ 5.1–5.5	2.28 (0.55)
	K^+ > 5.5	6.60 (1.70)

Abbreviations. eGFR, estimated glomerular filtration rate; IRR, incident rate ratio; K⁺, potassium; NYHA, New York Heart Association; RAASi, renin-angiotensin-aldosterone system inhibitor; SHFM, Seattle Heart Failure Model; SE, standard error.

^aSE assumed to be 10% of mean.

^bSE estimated from 95% confidence intervals.

^cNot statistically significant; value of 1 was applied in the model.

^dLuo et al.³³ reported IRRs for hospitalization and major adverse cardiovascular events in chronic kidney disease; these IRRs were applied to the risk of hospitalization and arrhythmia in heart failure, respectively.

^eKrogager et al.¹¹ reported all-cause mortality in heart failure patients following myocardial infarction, stratified by K⁺ intervals (90-day follow-up).

Mortality among heart failure patients was modeled through implementing the previously published Seattle Heart Failure Model, an externally-validated equation to predict all-cause mortality in this population³⁴. As serum potassium is not an explanatory variable in Seattle Heart Failure Model risk predictions, published hazard ratios were utilized to model the association between serum potassium category and mortality¹¹. Background all-cause mortality was applied if it exceeded Seattle Heart Failure Model predictions (adjusted for serum potassium level), and was consistent with gender-specific life tables published by the UK Office for National Statistics³⁵.

Whilst not influential to the present analysis, the model simulates acute hyperkalemia events when predicted serum potassium levels exceed a user-defined threshold (e.g. > 6.5 mEq/L); and uses published proportions to inform RAASi discontinuation and/or down-titration according to serum potassium levels²⁶, and the return to maximum RAASi dose over time once normokalemia is established³³.

In the Supplementary Material, a summary of the methods and data sources used to model disease progression and event incidence has been provided (Table S1), in addition to an illustration of predicted cumulative event incidence from the model for different patient characteristics (Figure S1).

Resource costs and health-related utilities

As the present study aimed to estimate the value of maintaining normokalemia irrespective of the strategy used to achieve this target, costs and utilities related to pharmacological serum potassium management were not considered. Nevertheless, a UK healthcare payer perspective was adopted with regards to all other costs and benefits applied in illustrative analyses. Health resource costs were obtained from published sources and inflated to 2014–2015 GBP using the Personal Social Services Research Unit Hospital and Community Health Services pay and price inflation index³⁶. Annual costs of RAASi use were based on maximum daily doses recommended in European Society of Cardiology Guidelines¹⁷, and costs sourced from the Monthly Index of Medical Specialities³⁷. Additional costs applied to modeled health states and events are presented in Table 2; it was assumed that there were no costs associated with death.

Table 2. Utilities and costs applied to modeled health states and events.

	Mean	SE	Source
Health state utility
NYHA I	0.855	0.005	Göhler et al.^38
NYHA II	0.771	0.005
NHYA III	0.673	0.006
NYHA IV	0.532	0.027
Event disutility^a
Arrhythmia	–0.025	0.005	Sullivan et al.^39
Hospitalization	–0.024	0.007	Göhler et al.^38
Treatment and event costs (£)^b
Annual cost of RAASi therapy	57.85	5.79^c	ESC^17; MIMS^37^d
Arrhythmia	1,453.32	145.33^c	Colquitt et al.^31
Hospitalization	2,444.80	244.48^c

Abbreviations. ESC, European Society of Cardiology; MIMS, Monthly Index of Medical Specialities; NYHA, New York Heart Association; RAASi, renin-angiotensin-aldosterone system inhibitor; SE, standard error.

^aEvent disutility values (subtracted from 1) were applied multiplicatively to baseline values, scaled to the current cycle length.

^bCosts were inflated to 2014–2015 values using the Personal Social Services Research Unit Hospital and Community Health Services pay and price inflation index³⁶.

^cSE values assumed to be 10% of mean.

^dAnnual cost of RAASi therapy reflects a weighted average, based on maximum daily doses recommended in ESC guidelines¹⁷ and costs sourced from MIMS³⁷. It was assumed that 90% of heart failure patients in UK clinical practice receive ACE inhibitors (assumed ramipril), 10% receive ARBs (assumed candesartan), and 70% receive MRAs (assumed spironolactone).

Utility values to inform health-related quality-of-life were sourced from published literature^38,39, and were applied to modeled health states and at the incidence of each event (Table 2). Health state utilities and event disutility values were applied multiplicatively to baseline values, sourced according to age from the EuroQoL Group⁴⁰.

Model validation

To assess the internal validity of its predictions, the model was used to estimate 90-day mortality rates as a function of serum potassium level, to be compared against rates published by Krogager et al.¹¹. External validation exercises compared model predictions of survival against rates reported in Pocock et al.⁴¹, and in studies previously used to validate the Seattle Heart Failure Model^34,42–47. In each validation scenario, modeled cohorts were initialized with baseline characteristics reported in each of the validation studies; where model inputs were unavailable, default model settings, consistent with those defined in the base case, were utilized. For this analysis, serum potassium levels were set to the mid-point of the normokalemia range at baseline (4.3 mEq/L), and were allowed to vary using a standard deviation of 0.5, based on an approximation of the variation observed by Krogager et al.¹¹.

Model application

To explore the long-term benefits of maintaining normokalemia and enabling optimal RAASi therapy in heart failure patients, the model was used to compare health economic outcomes between two hypothetical scenarios. Patients in both modeled cohorts were aged 73 years at baseline, with eGFR of 67 mL/min/1.73 m² and serum potassium of 4.5 mEq/L; 43% were female and patients were distributed across NYHA class I (40%), II (30%), III (20%), and IV (10%). Further details of baseline inputs used to inform model predictions are provided in the Supplementary Material (Table S2). Simulated patients who received optimal serum potassium management and ongoing RAASi therapy (treatment arm) were compared against those who discontinued RAASi treatment to avoid hyperkalemia (control arm). Key outcomes of the analysis included total discounted per-patient costs, quality-adjusted life years (QALYs), and life years accumulated over a lifetime horizon, in addition to net monetary benefit (NMB) based on willingness-to-pay (WTP) thresholds typically applied in the UK (£20,000–30,000 per QALY gained). As model analyses were conducted independent of pharmacological potassium management costs, incremental NMB represented the amount that could be spent to maintain normokalemia whilst remaining cost-effective at conventional WTP thresholds. In line with National Institute for Health and Care Excellence guidance⁴⁸, outputs were discounted at 3.5% annually, to allow future costs and benefits to be presented at their current value.

Sensitivity and scenario analyses

The sensitivity of model predictions to the following variable changes was assessed: age (73 ± 10 years); gender (male and female); NYHA class (I or IV); systolic blood pressure (100, 120, 140, or 160 mmHg); total cholesterol (190, 220, or 250 mg/dL); β-blocker use (all vs none); implantable cardioverter-defibrillator use (all vs none); cardiac resynchronization therapy defibrillator (biventricular implantable cardioverter-defibrillator) use (all vs none); RAASi efficacy for heart failure progression and events (base case ±20%); event costs (base case ±20%); event disutility (base case ±20%); NYHA class utility (base case ±20%); and time horizon (5 or 10 years).

Recent evidence suggests that sub-optimal RAASi therapy (defined as <50% of the guideline recommended dose) is associated with increased risks of hospitalization and/or death among heart failure patients^24,25. To explore the health economic impact of RAASi dosing in scenario analyses, model inputs describing the cost and efficacy of RAASi treatment (related to heart failure progression and events) were scaled from 0–100% of base case values. In the absence of evidence fully characterizing the relationship between RAASi dose and efficacy, costs and efficacy were scaled proportionally.

Results

Model validation

Internal validation exercises found that 90-day mortality rates predicted by the model were consistent with those described by Krogager et al.¹¹ (Figure 2(a)). Additional exercises that aimed to replicate survival estimates observed in external studies demonstrated that the model had a reasonable predictive capability (R² of 0.81); however, it was noted that the accuracy of survival predictions improved when the baseline serum potassium levels of cohorts with particularly high NYHA scores^42,46 were increased, producing an R² of 0.92 (Figure 2(b)).

Figure 2. Comparison of observed and predicted mortality rates (a) and survival rates (b) in validation exercises.

Model application

In simulated heart failure patients receiving optimal serum potassium management to enable ongoing RAASi therapy, predicted mean life expectancy was extended by 1.38 years compared to patients who discontinued RAASi in order to maintain normokalemia (Table 3). Ongoing RAASi treatment was associated with fewer hospitalization events over a lifetime (223 vs 249 hospitalizations per 100 patients). This result corresponded to a number needed to treat of four, which represents the average number of patients who require ongoing RAASi treatment in order to prevent one additional hospitalization event.

Table 3. Predicted value of optimal serum potassium management to enable ongoing RAASi therapy in heart failure patients.

Discounting	Outcome	Treatment arm	Control arm	Incremental benefit	Incremental NMB
					WTP at £20,000 per QALY	WTP at £30,000 per QALY
Undiscounted	Cost	£7,016	£6,977	£38	£14,753	£22,149
	QALY	4.53	3.79	0.74
	LY	8.31	6.93	1.38
Discounted	Cost	£5,734	£5,843	–£110	£10,679	£15,964
	QALY	3.72	3.19	0.53
	LY	6.79	5.81	0.99

Abbreviations. LY, life year; NMB, net monetary benefit; QALY, quality-adjusted life year; RAASi, renin-angiotensin-aldosterone system inhibitor; WTP, willingness-to-pay threshold.

Simulated patients were aged 60 years at baseline, with eGFR of 60 mL/min/1.73 m², serum potassium of 4.5 mEq/L, and distributed across NYHA class I (40%), II (30%), III (20%), and IV (10%). Analyses were conducted independent of costs and utilities related to pharmacological potassium management.

Over a lifetime horizon, optimal serum potassium management to enable ongoing RAASi therapy was associated with incremental per-patient costs of –£110 (+£38 undiscounted), and QALY gains of 0.53 (0.74 undiscounted) (Table 3). Additional per-patient costs associated with administering RAASi therapy in the treatment arm (+£393) were offset by a reduction in costs related to fewer hospitalization events (–£633), while additional per-patient costs related to arrhythmia (+£130) were attributed to improved survival and extended exposure to such risk (Figure 3). Overall, cost savings and QALY gains predicted in treatment-independent analyses corresponded to incremental NMB values of £10,679 and £15,964, which represent amounts that could be spent on cost-effective strategies to maintain normokalemia at WTP thresholds of £20,000 and £30,000 per QALY, respectively (Table 3).

Figure 3. Discounted cost breakdown for modeled heart failure patients receiving optimal serum potassium management to enable ongoing RAASi therapy (treatment arm), compared against patients not receiving RAASi to maintain normokalemia (control arm). Abbreviations. RAASi, renin-angiotensin-aldosterone system inhibitor.

Sensitivity and scenario analysis

The results of deterministic sensitivity analyses are summarized in Figure 4. Due to lower QALY gains in older patients, the predicted value of maintaining normokalemia and RAASi therapy decreased with increasing age; at a WTP threshold of £20,000 per QALY gained, incremental NMB ranged from £5,710 in heart failure patients aged 83 at baseline, to £13,232 in those aged 63. Although more favorable cost savings were estimated over shorter time horizons of 5 or 10 years, lower QALY gains were captured, which subsequently translated to lower incremental NMB than was estimated in the base case (£2,752 and £6,299 over 5- and 10-year horizons, respectively). Modeled efficacy of RAASi therapy (±20%) was also influential to predicted incremental costs (+£9 to –£245), QALYs (0.41–0.65) and NMB (£8,101–£13,300 at £20,000 per QALY gained).

Figure 4. Impact of model inputs on incremental discounted costs (a), QALYs (b), and NMB at £20,000 per QALY gained (c). Abbreviations. CRT-D, cardiac resynchronization therapy defibrillator; ICD, implantable cardioverter-defibrillator; NMB, net monetary benefit; NYHA, New York Heart Association; QALY, quality-adjusted life year; RAASi, renin-angiotensin-aldosterone system inhibitor; SBP, systolic blood pressure; TC, total cholesterol.

Figure 5 summarizes the results of scenario analysis in which the cost and efficacy of RAASi therapy were scaled from 0–100% of base case values. The lifetime incidence and cost of arrhythmia were predicted to rise with increasing RAASi efficacy, due to improved patient survival and longer exposure to such risk; however, this was offset by a reduction in the incidence and associated costs of hospitalization events. Overall, increasing RAASi efficacy was associated with improvements in discounted per-patient QALYs, life years, total costs, and NMB at conventional WTP thresholds.

Figure 5. Impact of RAASi cost and efficacy (0–100% of base case values) on discounted per-patient costs (a), QALYs and life years (b), and NMB at conventional WTP thresholds (c). Dotted line indicates 50% cost and efficacy; sub-optimal RAASi therapy is defined as <50% of guideline recommended dose; however, this may not equate to 50% efficacy. Abbreviations. NMB, net monetary benefit; QALY, quality-adjusted life year; RAASi, renin-angiotensin-aldosterone system inhibitor; WTP, willingness-to-pay.

Discussion

In the face of an aging population, heart failure is considered an emerging epidemic, associated with significant morbidity, mortality and healthcare expenditure. Despite this, prognosis following a heart failure diagnosis has improved to some extent during the last decade, due to efficacious treatment strategies including RAASi agents, β-blockers, and implantable cardiac devices. The risk of hyperkalemia is compounded in heart failure patients treated with cardiorenal protective RAASi agents; however, down-titrating and/or discontinuing RAASi therapy to avoid hyperkalemia has been associated with worsening clinical outcomes. This study developed a novel computer simulation model to explore the health economic consequences of discontinuing RAASi to avoid hyperkalemia in heart failure, and quantify the benefits to be gained by maintaining serum potassium levels within a narrow physiological range. Relative to maintaining normokalemia in the absence of RAASi, sustained potassium management and ongoing RAASi therapy were predicted to extend life expectancy by 1.38 years and reduce the incidence of hospitalization over a lifetime, from 249 to 223 events per 100 patients.

By predicting associations between serum potassium levels, RAASi use and long-term outcomes in heart failure patients, the model developed in this study represents a novel tool to quantify the health and economic value associated with sustained serum potassium management. This model was conceptualized by co-authors with health economic modeling and clinical expertise, constructed according to best practice guidelines⁴⁹, and validated against several data sources. Nevertheless, the predictive power of computer simulation modeling remains dependent on the source data used to inform model development. As new data describing the relationship between serum potassium and health economic outcomes in heart failure become available, further refinement and validation of the model will ensure the ongoing accuracy and generalizability of its outputs.

In the presented application of the model, improved life expectancy, discounted cost savings and QALY gains were predicted in heart failure patients who maintained normokalemia and received ongoing RAASi therapy, compared to those who discontinued RAASi to achieve equivalent serum potassium levels. The predicted value of optimal serum potassium management was driven by the established benefits of continued RAASi therapy in heart failure, including reduced risks of mortality and hospitalization events^32,34. These findings, in addition to the results of scenario and sensitivity analyses, highlight both the negative health economic consequences of RAASi discontinuation, and the potential value to be gained by alternative strategies that avoid hyperkalemia and allow optimal heart failure therapy.

Target doses of ACE inhibitors, ARBs, β-blockers, and MRAs, as recommended in recent European Society of Cardiology Guidelines, have been evidenced to reduce the risk of heart failure hospitalization and death¹⁷. However, the vast majority of heart failure patients do not receive guideline-recommended treatment in clinical practice, with hyperkalemia representing a leading reason for sub-optimal RAASi and/or MRA therapy¹⁹. Although this observation is consistent with clinical guidelines suggesting that acute hyperkalemia events (>5.5–6.0 mmol/L) necessitate the down-titration and/or discontinuation of such potassium-retaining agents^17,23, RAASi cessation and the prescription of submaximal RAASi doses have been associated with elevated risks of mortality and/or hospitalization, when compared to moderate RAASi therapy of at least 50% of the recommended target dose^24,25. Our results coincide with such findings, and emphasize the need to avoid hyperkalemia, and subsequent RAASi down-titration, in patients with heart failure.

Limitations of this study largely arise from a paucity of published literature. For this reason, the model does not currently simulate the impact of RAASi use on heart failure progression or arrhythmia events; and base case analyses reflect an ideal scenario in which patients were administered with guideline-recommended RAASi therapy¹⁷. The true health economic value of optimal serum potassium management may be under-estimated by model predictions, as some downstream implications (such as the potential avoidance of implantable cardioverter-defibrillator and cardiac resynchronization therapy defibrillator devices) were not considered, but may represent interesting avenues for future research. Furthermore, while modeled fluctuations in serum potassium levels may reflect several underlying mechanisms, including circadian rhythm and renal function⁵⁰, such factors were not implemented within mixed-effects regression equations. Finally, our study was not immune to the general limitations associated with modeling in healthcare. In the absence of suitable data, necessary modeling assumptions were made based on clinical expert opinion, and ongoing refinement of the model will ensure that its inputs are updated as appropriate data becomes available in the literature. Although external validation exercises proved the accuracy of model survival predictions was adequate with an R² of 0.81, they were not perfect, since modeling can only provide an approximation of clinical data rather than a perfect prediction of patient outcomes. Future studies validating the model against clinical data from a wider range of patient cohorts, treated in different countries and settings, and possessing different baseline characteristics, could provide a more detailed picture of the accuracy of model predictions and, if applicable, drive its refinement. In addition, extending the model to incorporate treatment effects will allow the value of novel serum potassium management strategies to be estimated in future health economic analyses.

Findings from the present study highlight the need for careful serum potassium monitoring in the heart failure population; and emphasize the importance of achieving normokalemia to maintain guideline-recommended RAASi therapy, avoid adverse events, and improve long-term outcomes. Emerging potassium-binding agents have been found to effectively lower serum potassium levels, control normokalemia, and enable continued RAASi therapy in patients at risk of hyperkalemia, and may provide significant health economic value over time^51–55. Data arising from this study do not support the adoption of a particular strategy to maintain normokalemia in heart failure, but may prove useful to inform cost-effectiveness analyses for future treatment-specific research.

Conclusions

In conclusion, this study developed a novel computer simulation model to quantify the potential health economic value of interventions that treat hyperkalemia, manage serum potassium levels, and optimize RAASi therapy in heart failure patients. In model analyses, controlled normokalemia and ongoing RAASi therapy were associated with the avoidance of hospitalization events, life year and QALY gains, overall cost savings, and NMB gains at conventional WTP thresholds, thus highlighting the consequences of under-utilizing RAASi to manage hyperkalemia risk in heart failure. Our analyses were conducted independent of pharmacological potassium management costs, and do not support the adoption of any one therapeutic strategy, but rather suggest that achieving and maintaining normokalemia is a valuable approach to enable optimal RAASi therapy and improve long-term outcomes in the heart failure population.

Transparency

Declaration of funding

Model development and medical writing support for this study were funded by AstraZeneca. The funding agreement ensured the authors’ independence in designing the study, interpreting the data, and preparing the manuscript for publication.

Declaration of financial/other interests

AB has received modest advisory honoraria from AstraZeneca in relation to this study. EP and LQ are full-time employees of AstraZeneca. CL has received significant research grant funding from AstraZeneca (awarded to institution), and modest speaker honoraria from Biotronik, Medtronic, Abbot, Novartis, and Vifor. HB and PM have received significant research grant funding from AstraZeneca (awarded to institution). HF has received modest research grant funding from AstraZeneca (as principal investigator), and modest expert witness funding from Amgen (as consultant). ME declares no conflict of interest. JME peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Previous presentations

Data for this study were presented in part at the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) 20th Annual European Congress, November 4–8, 2017, in Glasgow, Scotland.

Acknowledgments

Editorial assistance in the preparation of this manuscript was provided by Dr Karina Hamilton of Health Economics and Outcomes Research Ltd. The authors would like to also thank Dr Susan Grandy and Dr Klas Bergenheim of AstraZeneca Pharmaceuticals, for their continuous support, inspiration, and contribution to this work.