Development and Validation of a Markov Microsimulation Model for the Economic Evaluation of Methamphetamine‐Associated Heart Failure in the United States

Abstract

Background

Methamphetamine‐associated heart failure (MethHF) is an aggressive HF subtype, and its economic impact on the US health care system is not well defined.

Methods

A Markov microsimulation model from a limited societal perspective was developed to estimate the cost of illness and mortality burden of MethHF over a 20‐year time horizon. A hypothetical cohort of 10 000 patients with severe MethHF transitioned through 3 states: mild, severe, and death, with HF hospitalization as a transitional event. Utilities, measured as quality‐adjusted life years, and costs, expressed in 2024 US dollars, were based on published literature, and transition probabilities were derived from real‐world data.

Results

The age of the simulation cohort was 50±6 (mean±SD) years old. Through a median follow‐up duration of 5.25 (Quartile 1–Quartile 3: 2.92–9.83) life‐years (total 72 562.75 person‐years), 2.99 (1.67–5.27) quality‐adjusted life years were accrued. The 20‐year cost of illness per patient was $627 346±240 282, and the direct cost (from both HF hospitalization and outpatient care) per patient was $421 020±$313 258. Annual cost of illness per patient was $101 652 ($65 039–$171 216), out of which 61.7±29.7% was attributed to direct costs and 38.3±29.7% to premature death. HF hospitalization costs accounted for 87.8±13.3% of direct costs. A total of 8831 (88.3%) patients died from all causes, with age at death at 56±7 years old. The all‐cause mortality rate was 1217 deaths per 10 000 person‐years.

Conclusions

This Markov microsimulation study underscores the substantial cost and mortality burdens of MethHF. The field of HF care cannot afford to overlook the economic and societal impacts of MethHF, which require urgent interventions.

Nonstandard Abbreviations and Acronyms

COI

cost of illness

EV

expected value

HFH

heart failure hospitalization

MethHF

methamphetamine‐associated heart failure

PSA

probabilistic sensitivity analysis

QALY

quality‐adjusted life years

TP

transition probability

USD

US dollars

Clinical Perspective.

What Is New?

• This is the first study to use a real‐world data‐driven Markov microsimulation model to project the societal cost for methamphetamine‐associated heart failure.

What Are the Clinical Implications?

• The projected 20‐year cost of illness (including direct medical expenditures and productivity losses from premature mortality) for a patient with methamphetamine‐associated heart failure was estimated to be $627 346±$240 282, >4 times the cost incurred by a typical patient without methamphetamine‐associated heart failure.

• The findings extend prior epidemiological observations and emphasize the urgent need for multidisciplinary interventions to improve survival and contain costs in this aggressive yet preventable subtype of heart failure.

Methamphetamine, also known as “Meth,” is the most commonly used stimulant drug worldwide and carries one of the highest burdens of disease associated with drug use in the United States. ¹ Methamphetamine use has been associated with a wide range of cardiovascular diseases, with dilated cardiomyopathy and resultant systolic heart failure (HF) being the most commonly encountered and studied clinical entity. HF is traditionally a disease of older people, with 80% of HF‐related hospitalizations and 90% of HF‐related deaths occurring among patients aged 65 years or older. ² Over recent years, however, the US health care system has been caught in a surprising surge of methamphetamine‐associated heart failure (MethHF), which involves predominantly men in their 40s and 50ss, and with an aggressive HF subtype leading to high mortality, frequent HF hospitalizations (HFH), and rising societal costs. ¹ , ³ , ⁴ Patients with MethHF are also more likely to have other substance abuse problems, homelessness, as well as comorbid psychiatric disorders that make this group of patients challenging to manage and outreach. ⁵ , ⁶

From 2002 to 2020, national representative MethHF hospitalizations of all HFH nationwide increased exponentially from 0.05% in 2002, to 2.05% in 2008, to 7.36% in 2014, to 17.83% in 2020. ¹ , ⁶ , ⁷ The impact was particularly evident in the Western part of the United States, where MethHF in California increased by 585% between 2008 and 2018, with inflation‐adjusted hospitalization charges for MethHF increasing by 840%, a rate >10 times greater than that for all HFH. ³

Literature reported findings, which mostly consist of small single‐center case series and aggregate data from administrative databases, likely underestimate the true scope and severity of MethHF in the United States. Therefore, a health economic analytic framework was developed by using a Markov microsimulation model to project the cost of illness (COI) and mortality burden of MethHF in the United States over a 20‐year time horizon.

METHODS

Data Availability Statement

The authors declare that all supporting data are available within the article and its online supplemental files. The Markov model file is available from the corresponding author upon reasonable request.

Overview

This COI study, which measured the economic burden of MethHF in monetary units, assumed a limited societal perspective to include both direct medical expenditures and cost components beyond those captured by the health care sector perspective. ⁸ , ⁹ , ¹⁰ Specifically, the cost analysis incorporated nonmedical costs from loss productivity related to MethHF‐associated premature death but excluded spillover impacts affecting sectors other than health care such as education and criminal justice due to lack of data. ¹¹

The overall decision‐analytic approach was based on a Markov microsimulation model incorporating demographic and clinical data of real‐world patients with severe MethHF. The primary outcomes were COI, direct medical costs including both HFH‐related care costs and outpatient care costs, and productivity loss from premature death. Secondary outcomes included all‐cause mortality, life years, and quality‐adjusted life years (QALYs).

Model Description

A Markov microsimulation model was constructed to project the relationship between severe MethHF, its disease course, and associated costs over a 20‐year time horizon, or a total of 240 monthly cycles. TreeAge Pro Suite 2025.R2 (TreeAge Software Inc., Williamstown, MA, USA) software was used to develop the model incorporating patient and clinical characteristics into the modeling structure. The advantage of the microsimulation model over the traditional Markov cohort model is that microsimulation enables incorporation of clinical characteristics by randomly sampling from distributions to generate a wide range of possible outcomes. Additionally, individual disease history (or memory event that will in turn affect model parameters) can be integrated into microsimulation by using tracker variables. ¹² , ¹³

Patients with MethHF are distinctly different from patients with traditional HF, in that they are significantly younger and predominantly male, yet have a higher 5‐year hospital readmission rate for HF and age‐adjusted mortality. ¹ , ³ , ⁵ , ⁶ Therefore, we used real‐world data to estimate transition probabilities (TPs) between the health states, ¹⁴ , ¹⁵ and costs and utilities were derived from the literature. This Markov microsimulation model included 3 health states (mild MethHF, severe MethHF, and dead) and 1 transitional event (HFH). The mild MethHF state was defined as having a left ventricular ejection fraction (LVEF) between 35% and 55% and New York Heart Association Class II, and the severe MethHF state was defined as having an LVEF ≤35% and New York Heart Association Class III/IV. Patients with mild MethHF may undergo an HFH event and either progress to the severe MethHF state, remain in the mild MethHF state, or proceed to die from the disease. Similarly, patients in the severe MethHF state may undergo an HFH event, and either transition back to the mild MethHF state, continue to remain in the severe MethHF state, or die from the disease. The overall model structure is as shown in Figure 1A, and the state transition diagram shown in Figure 1B visually represents the transitions between the 3 health states as directed by the respective TPs.

Figure 1. Markov microsimulation model diagrams.

A, Markov microsimulation model structure. B, State transition diagram of this Markov microsimulation visually illustrates how a hypothetical patient proceeds through the various branches of the model and transitions between the 3 health states (mild MethHF, severe MethHF, and dead) as determined by their respective transition probabilities. MethHF indicates methamphetamine‐associated heart failure.

The model was set up to run with 10 000 iterations, representing 10 000 hypothetical patients, with an age distribution reflecting real‐world patients with MethHF. The mean age of the Program Evaluation and Review Technique distribution used in our microsimulation model was 50 years, with minimum age of 35 years and maximum age of 65 years. ⁵ , ¹⁶ Because most patients with MethHF entered the health care system at the advanced stage (severe MethHF), every patient in the model started in the severe MethHF state before cycling through the rest of the health states within the model, as directed by the TPs of each state. ¹ , ³ , ⁵ , ⁶ , ¹⁴ , ¹⁵

Model Input Parameters

Integration of Real‐World Data

TPs for HFH, death, and reversal from severe to mild MethHF were based on real‐world data. Santa Clara Valley HealthCare Institutional Review Board approved the reopening of 2 previously published studies on MethHF from 2017 and 2018, which provided the real‐world data to derive our model parameters. ¹⁴ , ¹⁵ Informed consent was waived for this retrospective study. Hospitalization rate and mortality data were updated as in 2024 (chart review by S. X. Z.) for TP calculations. The economic simulation model itself was exempt from institutional review board approval.

Between the cohorts with severe and mild MethHF, a total of 381 patients were identified, with a total follow‐up duration of 63.67 months (Q1–Q3: 33.0–101.5). During this time, 198 people died, resulting in an unadjusted all‐cause mortality of 52%.

Specifically, 123 patients identified as mild MethHF had complete data and at least 1 year of follow‐up. The number of patients lost to follow‐up (defined as last clinical contact ≥3 years and not known to be deceased) was 22. The median follow‐up for this group was 84.5 months (Q1–Q3: 43.5–111.8). The median age at diagnosis was 49 years (Q1–Q3: 43.0–56.0), 81.3% (100/123) were males, LVEF was 47% (Q1–Q3: 40–50), and 21.9% (27/123) were homeless. All‐cause mortality was 21.1% (26/123) (Tables S1 and S2).

A total of 258 patients were identified as severe MethHF with complete data and at least 1 year of follow‐up. The number of patients lost to follow‐up and not known to be deceased was 27. The median follow‐up duration was 57.7 months (Q1–Q3: 29.2–92.0), which was shorter than for mild MethHF due to higher all‐cause mortality. The median age at diagnosis was 48 years (Q1–Q3: 42–55); 85.3% (220/258) were male; LVEF was 25% (Q1–Q3: 20–30); and 30.2% (78/258) were homeless. All‐cause mortality was 66.7% (172/258) (Tables S3 and S4).

Transition Probabilities

Derivation of TPs for mild HFH, severe HFH, reversal from severe to mild, death from mild, and death from severe are outlined in Data S1. ¹⁷ , ¹⁸ , ¹⁹ , ²⁰

Monthly TP for Progression From Mild to Severe MethHF

Our source data did not include progression from mild to severe MethHF. Based on published literature, the rate of progression from Stage B (mild MethHF) to stage C/D (severe MethHF) HF was 1.4 cases per 100 person‐years of follow‐up (95% CI, 0.9–2.1). ²¹ The monthly progression rate was thus calculated as: 1.4/(100×12)=0.00116667. Using the rate to probability conversion equation P=1−exp^(−rt), monthly TP for progression from mild to severe MethHF was calculated as 0.001165986 (0.12%) with 95% CI of [0.0007497, 0.001748] or [0.07%, 0.17%].

NonCardiovascular (Background) Mortality Probabilities

Background mortality probabilities were based on the National Vital Statistics 2021 life tables. ²²

Costs

In this COI study for MethHF from a limited societal perspective, both direct medical costs and nonmarket productivity costs to society were estimated in our model. ¹⁰ All costs were inflated to 2024 US dollars (USD) using annual US Bureau of Economic Analysis Personal Consumption Expenditure Health Index deflators with an inflation factor according to the year of data. ²³ Details related to inflation adjustment of cost components are specified in Data S2. ²⁴

Direct Cost

Direct costs included outpatient costs, HFHs costs, and post HFH discharge costs including home care or nursing home/skilled nursing facility care.

Monthly outpatient costs, including costs for emergency department services, primary and specialty care physician office visits, and pharmacy costs, were derived from the literature. ²⁵ Based on Tables S1 and S3, the median age for this cohort was 48 years old. ¹⁴ Therefore, monthly costs of $498 and $561 (2024 USD) were used for the patients with mild and severe MethHF, respectively.

The cost of HFH was taken from a systematic review of medical costs associated with heart failure in the United States from 2014 to 2020. ²⁶ This study reported a mean national cost estimate for acute HF admission of $13 418 (interquartile limit [IQL], 11 125–15 667) in 2019 USD, derived from 17 studies. HF class‐specific estimates were made based on a regression model conditional upon survival for costs with prevalent HF, which estimated a 12.73% increase in costs for New York Heart Association class III/IV versus II. ²⁷ , ²⁸ Therefore, the cost for severe heart failure was estimated at $15 126 (IQL, 12 541–17 661) in 2019 USD. Inflation adjusted to 2024 USD yielded $16 128 and $18 182 for mild MethHFH and severe MethHFH, respectively.

The mean 30‐day postdischarge cost following each HFH was $6283 in 2017 USD, inflation‐adjusted to $7815 in 2024 USD. ²⁶ It was calculated as the difference between all costs incurred in the 30 days following discharge and the average 1‐month cost of a patient with HF. This 30‐day post‐HFH cost was added to each HFH (mild or severe HFH) for the purpose of cost computation in the model.

The last year of life in patients with HF often leads to significant increases in medical resource costs that are not fully included in HFH costs. ²⁹ Using literature estimates, the cost of all‐cause death was incorporated into the model as a 1‐time cost upon time of death. ³⁰ To determine this cost, a weighted average was calculated based on frequency of occurrence of each category of death. The costs of sudden cardiac death ($23 328 in 2008 USD, 72 out of 231 patients), HF death ($70 224 in 2008 USD, 80 out of 231 patients), other cardiovascular death ($119 374 in 2008 USD, 34 out of 231 patients), and noncardiovascular ($76 313 in 2008 USD, 45 out of 231 patients) were averaged to be $64 027 in 2008 USD. Furthermore, hospitalizations accounted for 84.3% of all‐cause medical costs in the last year of life of patients with HF. ³¹ To avoid double counting of costs related to HFH, the 1‐time cost of death was calculated as $64 027×(1−0.843)=$10 052 (in 2008 USD), inflation adjusted to $14 021 in 2024 USD.

Costs related to comorbid conditions, such as trauma, depression, or somatic conditions, and nonheart failure hospitalizations, were not included in the analysis due to lack of data. Costs related to LV assist device and heart transplant were also not included in the simulation given the rarity of such an event occurring in this population and the lack of data. We assumed that these costs would not have a substantial impact on the findings.

Productivity Cost

In this limited societal perspective analysis, some but not all costs outside of the medical care sector were included in the model. ⁸ , ¹⁰ Productivity costs were estimated primarily as the economic value of production associated with loss of paid employment (forgone gross earnings) in the form of absenteeism (work days lost) or presenteeism (productivity loss while at work) for those younger than 65 years. ²⁴ , ³² , ³³ Despite the young age of patients with MethHF, there is a high prevalence of polysubstance abuse, comorbid psychiatric illnesses, as well as a 30% homelessness rate (Tables S1 and S3). ³⁴ Therefore, the employment rate was assumed to be low. Thus, productivity loss due to morbidity was not included in our cost estimate.

Productivity loss in the form of forgone earnings due to premature death was included based on the recommendation from the Second US Panel on Cost‐effectiveness in Health and Medicine. ⁸ Productivity was further divided into market productivity (employment earnings) and nonmarket productivity (informal labor production such as volunteering and household services). Mean annual productivity was estimated to be $57 324 for US adults in 2016, including $36 935 in market and $20 389 in nonmarket productivity. ²⁴ Due to the assumed low employment rate of the population with MethHF, only the nonmarket productivity of $20 389 was included in the cost estimate. Nonmarket productivity cost was inflated to 2024 USD at $25 803 per year of life lost before age 65 years.

A summary of the cost structure is outlined in Table 1. ⁸ , ⁹ , ¹¹ , ²⁴

Table 1.

Cost Components Included in This Limited Societal Perspective Health Economic Analysis

Cost components	Description	Included?
Direct cost paid by patients or health care payers	Outpatient cost	Yes
	HFH cost	Yes
	30‐d post HFH discharge cost	Yes
	Last year of life health care cost	Yes
Direct nonmedical costs incurred by patients or caregivers	Examples include travel, food, and accommodation, etc to obtain care	No
Productivity cost for morbidity	Underemployment, absenteeism or presentism	No
Productivity cost for mortality	Market productivity costs from labor production (employment)	No
	Nonmarket productivity costs (from informal labor production such as volunteering and household production)	Yes

HFH indicates heart failure hospitalization.

Health‐Related Quality of Life (Utilities)

We measured effectiveness in QALY based on the product of health utility and time. Health utility reflects the patient's quality of life with values ranging from 0 (death) to 1 (perfect health). Health utility estimates were obtained from the published literature. We estimated a baseline utility of 0.83 for mild MethHF and 0.78 for severe MethHF. ³⁵ , ³⁶ We did not assign an annual disutility ²⁵ , ²⁸ to account for aging and disease progression in our model given the young age of the population with MethHF. We applied a disutility of −0.07 to patients who experienced an HFH event for both MethHF (mild and severe) states. ³⁷

All model input parameters (cost, utility, and TPs) are summarized in Table 2.

Table 2.

Model Inputs

Variable name	Description	Root definition	PSA distribution	References
totalCycles	Number of mo for model, 20 y or 240 mo	240
Age	Starting age— Program Evaluation and Review Technique distribution with a mean age of 50 y, range (35–65)	dist_StartAge+(_stage/12)		1, 3, 5, 6
cMild	Outpatient mild MethHF cost ($498 per mo)*	498	Gamma	25
cSevere	Outpatient severe MethHF cost ($561 per mo)*	561	Gamma	25
c_each_HFH_mild	Cost of each mild HFH ($16 128) and 30‐d postdischarge cost ($7815)*	16 128+7815	Gamma	26, 27
c_each_HFH_severe	Cost of each severe HFH ($18 182) and 30‐d postdischarge cost ($7815)*	18 182+7815	Gamma	26, 27, 28
c_death	Cost for premature deaths before 65 ($25 803 per y), and a 1‐time death cost ($14 021) to account for additional cost of death in last y of life*	if(Age ≥65; 14 021; (25 803×(65−Age)+14 021))		24, 30
uMild	QALY for mild state (base)	0.83	Beta	35, 36
uSevere	QALY for severe state (base)	0.78	Beta	35, 36
u_decrem_hosp_mild	Disutility associated with each HFH	0.07		37
u_decrem_hosp_severe	Disutility associated with each HFH	0.07		37
p_bkgd_mortality_rate	Annual background mortality rate converted to monthly	ProbToProb(p_die_backgnd[Age]; 1/12)		22
pHFH_Mild	TP for mild HFH	0.0872	Beta	Table S1
pHFH_Severe	TP for severe HFH	0.2038	Beta	Table S3
pProgress_base	TP for progression from mild MethHF to severe MethHF	0.0012	Beta	21
pProgress_calc	TP for progression increases with each successive mild HFH	if(t_HFH_mild=0; pProgress_base;choose(Min(t_HFH_mild;12); 0.0013; 0.0014; 0.0015; 0.0016; 0.0016; 0.0016; 0.0016; 0.0016; 0.0016; 0.0016; 0.0016; 0.0016))		21
p_Reversal_base	TP for reversal from severe MethHF to mild MethHF	0.0046	Beta	Table S3
p_Reverse_calc	TP for reversal decreases with each successive severe HFH	if(t_HFH_severe=0; p_Reversal_base;choose(Min(t_HFH_severe;12); 0.0045; 0.0044; 0.0043; 0.0042; 0.0041; 0.0040; 0.0039; 0.0038; 0.0037; 0.0036; 0.0035; 0.0034))		Table S3
pMild_Death_Haz	TP of death from mild MethHF	ratetoprob(rMild_Death_Haz; 1)		Tables S1 and S2
rMild_Death_Haz	Intermediary hazard rate from life table (Table S2) used to generate pMild_Death_Haz	Life_Table_Mild_Hazard[_stage+0.5]		Tables S1 and S2
pSevere_Death_Haz	TP of death from severe MethHF	ratetoprob(rSevere_Death_Haz; 1)		Tables S3 and S4
rSevere_Death_Haz	Intermediary hazard rate from life table (Table S4) used to generate pSevere_Death_Haz	Life_Table_Severe_Hazard[_stage +0.5]		Tables S3 and S4

HF indicates heart failure; HFH, heart failure hospitalization; MethHF, methamphetamine‐associated heart failure; PSA, probabilistic sensitivity analysis; QALY, quality‐adjusted life years; and TP, transition probability.

^*All costs were inflated to 2024 USD.

Discounting

Future costs and utilities were discounted to present value at 3% annually as recommended by the Second Panel on Cost‐Effectiveness in Health and Medicine. ⁸ Adjustment was made in the model to account for the monthly cycle length. ²⁰

Statistical Analysis

The Monte Carlo microsimulation was run in 10 000 interactions with each iteration drawn from an age distribution (Program Evaluation and Review Technique) with progression and reversal between severe and mild MethHF states being tracked with tracker functions built into the model. Numbers of HFHs were also tracked separately for mild HFH and severe HFH states, respectively. All continuous variables were presented as mean±SD, or median and interquartile range (Q1–Q3) as appropriate.

Sensitivity Analysis

A 1‐way deterministic sensitivity analysis was performed to assess the impact of individual variables on the base‐case result, and depicted by a tornado diagram, which combines multiple 1‐way deterministic sensitivity analyses and presents the results to show how the parameters' uncertainty ranges affect the expected values (EVs; mean COI or mean QALY) of the model. The input variables were varied 10% above and 10% below the base values.

Probabilistic sensitivity analysis (PSA) was performed to determine the impact of aggregate uncertainty on the TPs, costs, and utilities using a Monte Carlo microsimulation with 10 000 samples. In the PSA trials, all parameters were sampled from their respective distributions in each model calculation to generate a large set of estimates reflecting overall parameter uncertainty. TPs and utilities were assigned beta distribution, and costs were assigned gamma distribution. The distribution parameters and detailed methods are described in Data S3. The TPs for death from mild MethHF and severe MethHF were time‐dependent survival/hazard rate‐based and hence cannot be varied in the PSA. Spearman's rank correlation test, a nonparametric test, was used to assess the correlation between cost and QALY as their relationship is likely nonlinear and heteroskedastic.

Model Validation

The Markov microsimulation model structure was evaluated for its face validity by experts in the fields of HF and health economics. Additionally, the model underwent both internal and external validation testing. ³⁸

Internal validation of the overall model structure was carried out wherein all utility values were set to 1, decrement set to 0, and global discounting turned off; under these conditions, all patients still alive at the end of the simulation achieved the maximum QALY score of 20 QALYs (achievement of perfect health), as expected (data not shown).

Internal validation was also conducted by comparing the model's outputs with the expected mortality from the source data. ¹⁴ , ¹⁵ It was performed by running the microsimulation with the time horizon set as the median follow‐up duration of the source data cohort, at 64 months. This yielded 5082 deaths, thus with an all‐cause mortality of 50.8%, compared with all‐cause mortality in our source data of 52% (Table S5).

External validation of the model prediction was performed by comparing outputs with an independent data source. The all‐cause mortality of a series of 896 patients with MethHF from University of California, San Diego Health System was reported as 25% (±3%). ⁵ However, out of this cohort, only 33% had severe MethHF (LVEF <30%), whereas our simulation cohort consisted of 100% severe MethHF. Therefore, their observed 10‐year total mortality of 25% (±3%) was adjusted by a factor of 3, to an estimated 75% for the severe MethHF subgroup. The microsimulation model was then run with a 10‐year time horizon (120 months) and yielded 7542 deaths and thus an all‐cause mortality of 75%, which agrees with the data source and thus fulfilling the external validation (Table S6).

RESULTS

The starting age of the 10 000 hypothetical patients with severe MethHF was 50±6 (mean±SD) years old. Over the 20‐year time horizon, the median follow‐up duration was 5.25 (2.92–9.83) life‐years (total of 72 562.75 person‐years), with 2.99 (1.67–5.27) QALYs accrued. The number of HFH (combined severe and mild) per patient was 0.20±0.07 per month. The total number of reversals (from severe MethHF to mild MethHF) was 2999 from a sample size of 10 000, ranging from 0 to maximum of 5 reversals per patient with mean±SD of 0.30±0.59 per patient. This reversal rate is identical to the source data reversal rate of 0.3, ¹⁵ thus internally validating the model structure from this perspective. The mean number of progressions from mild to severe MethHF was 0.28±0.57 per patient (2794 progressions/10 000), ranging from 0 to maximum of 5 progressions per patient. The epidemiological parameters of the microsimulation outputs are presented in Table 3, and the summary Monte Carlo microsimulation output generated by TreeAge Pro is reported in Table S7.

Table 3.

Epidemiological Parameters of Microsimulation Outputs

	Mean	SD	Median	Quartile 1	Quartile 3
Age at entry, y	50.02	6.14	50.00	45.41	54.78
Age at death, y, n=8831	55.68	7.31	55.49	50.38	60.79
Duration of follow‐up, mo	87.08	71.08	63.00	35.00	118.00
Total no. HFH per patient	17.30	14.57	13.00	6.00	24.00
No. HFHs per patient per mo	0.20	0.07	0.20	0.16	0.23
No. progressions per patient	0.28	0.57	0.00	0.00	0.00
No. reversals per patient	0.30	0.59	0.00	0.00	0.00

HFH indicates heart failure hospitalization.

Over a 20‐year time horizon, 8831 (88.3%) patients died from all causes, with a mean age at death of 56±7 years old. The substantial number of deaths with a short follow‐up duration contributed to a median follow‐up time for all 10 000 participants of 5.25 (2.92–9.83) life‐years, despite a 20‐year time horizon. The all‐cause mortality rate can be derived from the 8831 deaths over 72 562.75 total person‐years to be 1217 deaths per 10 000 person‐years.

Probability distribution histograms showed a normal distribution of COI and a nonnormal distribution of QALYs (Figures S1 and S2). The mean value of the 20‐year COI per patient with MethHF was $627 346±$240 282. Annual median COI per patient was $101 652 ($65 039–$171 216), out of which 61.7±29.7% was attributed to direct costs, which included both outpatient and HFH costs, whereas the remaining 38.3±29.7% was attributed to productive cost due to premature death.

The 20‐year direct cost per patient was $421 020±$313 258, and annual median direct cost per patient was $59 437 ($49 956–$72 041). The annual median HFH cost per patient was $53 522 ($44 237–$65 985). HFH costs accounted for 87.8±13.3% of direct costs. All cost‐related statistics, including both mean±SD) and median (Quartile 1–Quartile 3), are presented in Table 4.

Table 4.

Results of Markov Microsimulation of MethHF

	Mean	SD	Median	Quartile 1	Quartile 3
20‐y COI per patient	$627 346	$240 282	$605 387	$451 903	$776 635
20‐y HFH cost per patient	$380 151	$284 880	$303 189	$153 890	$540 315
20‐y outpatient cost per patient	$40 869	$29 615	$32 496	$18 554	$56 815
20‐y direct cost* per patient	$421 020	$313 258	$340 727	$177 639	$596 629
20‐y productivity cost of premature death (n=8831)	$233 638	$159 075	$222 152	$101 616	$348 954
HFH cost/COI (%)	55.53	26.94	56.33	34.19	82.01
HFH cost/direct cost (%)	87.82	13.29	90.23	88.42	91.60
Direct costs/COI (%)	61.70	29.67	62.36	38.22	91.03
Premature death/COI (%)	38.30	29.67	37.64	8.97	61.78
Life‐y†	7.26	5.92	5.25	2.92	9.83
Annual COI per patient	$192 080	$430 100	$101 652	$65 039	$171 216
Annual HFH cost per patient	$56 223	$22 830	$53 522	$44 237	$65 985
Annual outpatient cost per patient	$5954	$476	$6124	$5716	$6321
Annual direct cost (HFH+outpatient) per patient	$62 176	$22 935	$59 437	$49 956	$72 041
QALY	3.75	2.74	2.99	1.67	5.27
COI per patient per QALY	$792 106	$12 260 755	$182 020	$125 380	$307 124
HFH cost per patient per QALY	$153 206	$1 348 620	$100 561	$79 269	$125 736
Outpatient cost per patient per QALY	$11 669	$20 158	$10 915	$10 385	$11 526
Direct cost per patient per QALY	$164 875	$1 368 279	$111 521	$89 635	$137 237

COI indicates cost of illness; HFH, heart failure hospitalization; MethHF, methamphetamine‐associated heart failure; and QALY, quality‐adjusted life year.

^*Direct cost is the combination of HFH and outpatient costs.

^† Life y=duration of follow up (mo)/12.

Comparisons between MethHF and literature derived non‐MethHF costs and all‐cause mortality are presented in Table 5. The 20‐year COI of MethHF was 4.3 times that of the average costs for non‐MethHF. The 20‐year direct cost was 2.8 times that of non‐MethHF. The COI per life‐year and COI per QALY of MethHF were 4.3 times and 5.7 times that of non‐MethHF, respectively.

Table 5.

Comparison of Cost and Mortality Between MethHF and Non‐MethHF

	MethHF	Non‐MethHF	Non‐MethHF (inflation adjusted to November 2024 USD)	References
Mean age, y	50.0	76.8	N/A	39
COI	$627 346 (20 y)	$106 000 (2012 USD) (20‐y)	$139 487	40
		$ 127 121 (2020 USD) (20 y)	$151 124	41
Direct cost	$421 020 (20 y)	$109 541 (2008 USD) lifetime	$152 818	39
COI per QALY	$182 020	$183 583/5.7=$32 207 (2020 USD)	$38 288	25
		$204 139/7.25=$28 157 (2021 USD)	$32 143	42
		$174 497/7.30=$23 904 (2022 USD)	$25 610	43
COI per life‐y	$101 652	$24 383 (median) (2020 USD)	$28 987	26, 44
		$183 583/7.6=$24 156 (2020 USD)	$28 717	25
		$204 139/11.26=$18 130 (2021 USD)	$20 697	42
		$174 497/10.99=$15 878 (2022 USD)	$17 011	43
HFH costs per life‐Y	$53 522	$15 879 (2020 USD)	$18 877	26, 44
% direct cost/COI	61.70%	71.3% (2020 USD projection)	N/A	41
% HFH/direct cost	87.82%	75%	N/A	40
		77%		39
		80%		41
All‐cause mortality	1217 per 10 000 person‐years	459 per 10 000 person‐years	N/A	45

COI indicates cost of illness; HFH, heart failure hospitalization; MethHF, methamphetamine‐associated heart failure; N/A, not applicable; and QALY, quality‐adjusted life year.

Sensitivity Analyses

Tornado diagrams, generated from multiple 1‐way deterministic sensitivity analyses, consist of a set of horizontal bars with each representing the impact of a specific input variable range on the mean EVs, are shown in Figure 2A and 2B (only the top 5 variables are presented).

Figure 2. Tornado diagrams representing multiple deterministic sensitivity analyses of the impact of input variable ranges on the mean expected values of cost of illness and quality‐adjusted life years.

A, The tornado diagram was generated for COI, based on the mean EV of $627 346. The variable with the largest impact on COI is presented on top, with descending order for subsequent variables. Numbers in parenthesis are 10% above and below the base value of each variable. B, The tornado diagram was generated for QALYs. The mean QALY of 3.75 was the EV. The variable with the largest impact on QALY is presented on top, with descending order for subsequent variables. Numbers in parenthesis are 10% above and below the base value of each variable. EV indicates expected value; HF, heart failure; HFH, heart failure hospitalization; MethHF, methamphetamine‐associated heart failure; and QALY, quality‐adjusted life year.

The variables are ranked so the widest bar (largest impact) is at the top, and the narrowest bar (smallest impact) is at the bottom. Each bar is split by the EV line, with the blue portion representing the part of the input range from the lower bound to the base case value, and the red portion represents the part of the input range from the base case value to the upper bound. This allows easy visualization of which inputs increase or decrease the EV.

The top 2 parameters with the strongest impact on COI were the cost of each severe HFH (variable name: c_each_HFH_severe) and the probability of severe HFH (variable name: pHFH_Severe). Similarly, the top 2 most impactful parameters on QALYs were the utility of severe HFH (variable name: uSevere) and the probability of severe HFH (variable name: pHFH_Severe).

The 10 variables and their respective distributions used in the PSA are as outlined in Table 2 and Data S3. A PSA scatterplot (Figure 3) was generated to reflect the overall COI and QALY spread, around the mean values of $414 310 (95% CI, 410 598–418 021) and 4.40 QALYs (95% CI, 4.38–4.41), respectively. Each data point represents a simulation scenario that incorporates uncertainty in both cost and effectiveness parameters within the model. Although there is moderate but well‐constrained uncertainty in effectiveness, with QALYs ranging between 1.8 and 6.2 QALYs and the highest density around ~3.8 to 5.0 QALYs, cost uncertainty is substantially larger, with a small number of simulations generating very high costs. The most extreme costs tend to occur at lower‐to‐mid QALYs, whereas very high QALYs are rarely associated with significantly high costs. In this PSA, costs and QALYs exhibited a weak inverse association with Spearman's rank correlation coefficient of −0.15, indicating that simulations with higher costs tended to yield slightly lower QALYs.

Figure 3. Probabilistic sensitivity analysis scatter plot displaying uncertainty of cost of illness and effectiveness, measured in quality‐adjusted life years, of MethHF.

Eff indicates effectiveness; HF, heart failure; MethHF, methamphetamine‐associated heart failure; and QALY, quality‐adjusted life year.

Undiscounted Outcomes

Analyses without applying the 3% global discounting to both costs and utilities were conducted and reported separately in Table S8.

DISCUSSION

MethHF has emerged as a particularly aggressive subtype of HF, notable for the predominantly younger target population, severe cardiac dysfunction, accelerated mortality, and high health care costs. ¹ , ³ , ⁴ , ⁵ , ⁶ , ¹⁴ , ¹⁵ In this Markov microsimulation model of 10 000 hypothetical patients with MethHF with LVEF ≤35% and New York Heart Association Class III/IV classification, global cost and mortality burdens throughout the 20‐year time horizon were projected, with the 20‐year COI of MethHF at 4.3 times that of non‐MethHF. Despite patients with MethHF being >20 years younger than the typical patients without MethHF, the all‐cause mortality derived from the microsimulation at 1217 deaths per 10 000 person‐years was 2.7 times that of non‐MethHF at 459 deaths per 10 000 person‐years.

Projecting these cost estimates to the broader US health system, the prevalence of MethHF nationwide can be extrapolated from the 2008 to 2019 National Inpatient Sample database, the largest US all‐payer database that captures 20% of stratified samples of US nonfederal hospitals, from which 160 000 adult patients (age >18 years) hospitalized with diagnoses of MethHF were identified. ⁴⁶ Therefore, there are currently an estimated 800 000 (160 000/0.2) patients with MethHF nationwide. With a mean 20‐year cost of $627 346 per patient, the projected total societal COI in the United States is $502 billion dollars over the next 20 years. Assuming the prevalence of severe MethHF in the United States remains constant, this translates to a projected annual cost of $25.1 billion for MethHF, which constitutes 35.8% of the projected $70 billion for total HF cost in 2030. ⁴¹ If the 20‐year direct cost of $421 020 per patient with MethHF is used instead of total cost, the projected annual direct cost of $16.8 billion still constitutes an astounding 24.1% of the total HF cost in 2030. ⁴¹

Direct medical expenditure accounted for 61.7% of the MethHF COI, which is lower than the projected 71.3% among the general HF population in the United States, due to the remaining 38.3% being attributed to productivity cost from premature death in this relatively young population with HF. ⁴¹ Annual HFH costs for patients with MethHF is nearly 3 times those of patients without MethHF. With HFH costs making up 87.8% of all direct costs in patients with MethHF, compared with 75% to 80% in non‐MethHF, the increased reliance on inpatient care by patients with MethHF is evidenced, which can be at least partially attributed to lack of outpatient engagement and poor social determinants of health, as reflected by the 30% homelessness rate, 29% of comorbid psychiatric conditions, and 44% of concurrent opioid abuse in patients with MethHF. ¹ , ⁵ , ⁶ , ³⁴ Despite these challenges, mitigating HFH in patients with MethHF is likely the most cost‐effective intervention. As demonstrated by the tornado diagrams in Figure 2A and 2B, the probability of severe MethHF related hospitalizations is one of the most impactful factors for both COI and QALY, suggesting that changing this variable's range holds the greatest potential “swing” or variability in the expected outcome.

The PSA scatter plot further illustrates the high level of cost variability present in patients with more advanced disease as reflected by the lower QALY values. The wider spread of COI at the lower end of the QALY spectrum reflects significantly more economic uncertainty in the management of MethHF. Understanding the drivers of this variation will be essential in guiding cost‐effective policy decisions. Our model highlights the complexity of MethHF management and the need for targeted approaches to improve both clinical and economic outcomes.

Not only is MethHF an extremely costly disease, this simulation also reveals high mortality among patients with MethHF. With 8831 out of 10 000 dead, the all‐cause mortality rate of 1217 deaths per 10 000 person‐years is significantly higher than that of non‐MethHF of 459 per 10 000 person‐years, which was derived from pooling 4 community‐based cohorts (CHS [Cardiovascular Health Study], FHS [Framingham Heart Study, PREVEND [Prevention of Renal and Vascular Endstage Disease], and MESA [Multi‐Ethnic Study of Atherosclerosis) with systolic HF and a mean age of 70. ⁴⁴ , ⁴⁵ Along with a high mortality rate, the population with MethHF also died young. The mean age at death was 56±7 years, which was much lower than the median age of death for patients with heart failure with reduced ejection fraction at 75 years from the Swedish Heart Failure Registry or 79 years from the FHS, consistent with the early mortality and aggressive progression known to happen to patients with MethHF. ⁴⁷ , ⁴⁸

The early and high mortality in MethHF reflects a combination of factors including advanced disease at diagnosis, high rates of HFH, and poor engagement with care due to both medical and social impediments. There is currently no established “methamphetamine replacement therapy,” although several agents have shown promise in preclinical and small clinical studies in attenuating the reinforcing effects of methamphetamine. ⁴⁹ , ⁵⁰ Additionally, multidisciplinary counseling coupled with contingency management programs have demonstrated some success. ⁵¹ Future health economic modeling should focus on cost‐effectiveness analyses comparing various pharmacological and nonpharmacological interventions for treating methamphetamine use disorder and MethHF. These analyses will help determine which treatment strategies provide the greatest clinical benefits relative to their costs, guiding policymakers and health care systems in allocating resources efficiently for both substance use and cardiovascular outcomes.

Limitations

The limitations of this paper include several assumptions made when integrating cost and utility parameters. Our source data were derived from a single health care system in a highly endemic area in Northern California, whereas states east of Mississippi generally have lower rates of methamphetamine use. ¹ , ⁴ , ⁶ However, the cost estimates are still likely underestimated due to the following factors: due to lack of data, emergency room visits, postacute rehabilitation care costs, and caregiver costs were only partially included. A recently published case series of home inotropic therapy may introduce an emerging area of cost for end‐stage patients with MethHF. ⁵² Non‐HFH for comorbid conditions, such as trauma, depression, or somatic conditions were not included due to lack of data. Advanced heart failure therapies such as mechanical circulatory devices/LV assist device implantation and heart transplant are exceedingly rare in patients with MethHF. Only anecdotal cases are communicated internally (no published data). The rate, however, is expected to increase as more advanced MethHF cases emerge over the next several decades. With an enlarging denominator, more patients with MethHF at end‐stage HF will meet the standard of advanced therapy such as sustained abstinence and adequate psychosocial support. The initial surgery and hospitalization costs for LV assist device and heart transplantation are estimated to be $247 553 and $200 130, respectively. ⁵³ Subsequent maintenance costs and high costs related to LV assist device and transplant death will only further push up the financial burden. Additionally, we counted only nonproductivity loss of wages due to premature death. Market productivity loss due to premature death, presenteeism, and absenteeism due to HF morbidity were not included in our cost estimates due to the 30% homelessness rate (a surrogate for low employment rate), 29% of comorbid psychiatric conditions, and 44% of concurrent opioid abuse. ¹ , ⁵ , ⁶ , ³⁴ In terms of mortality estimations, a combined 49 patients out of 381 in our source cohorts were lost to follow‐up and coded as alive for the purpose of survival analysis and transition probability calculations. The true mortality rate for severe MethHF may be even higher than projected by our model. Finally, these findings are generalizable only to the US population. However, these limitations are balanced by the strengths of the model design, where the key TPs were derived from real‐world cohorts. Overall, our model inputs were conservative. The simulation results, though already impressively high, are likely still underestimating the true severity of the disease and its economic impact.

CONCLUSIONS

This Markov microsimulation study demonstrates the heavy economic and mortality burdens of MethHF, with a 20‐year per patient COI 4.3 times and a mortality rate 2.7 times that of populations without MethHF. The high degree of cost variability and early mortality emphasizes the necessity for tailored, multidisciplinary interventions that address the economic and public health impact of MethHF on both patients and the health care system.

Sources of Funding

This study was not funded.

Disclosures

None.

Supporting information

Data S1–S3

Tables S1–S8

Figures S1 and S2

CHEERS Checklist