Seattle Heart Failure Model

Enter ambulatory chronic HF data. Diuretic doses are summed as daily furosemide equivalents (see Formula tab). Use one selection for ACE inhibitor or ARB when on renin–angiotensin blockade.

Demographics & clinical

Age (years) *

Sex

NYHA class

LVEF (%) *

Systolic BP (mmHg) *

Values >160 are truncated to 160 for the model.

Weight (kg) *

Ischemic etiology (vs non-ischemic cardiomyopathy)

Diuretics (oral, total daily dose, mg)

Leave 0 if none. Equivalents: 80 furosemide = 40 torsemide = 3 bumetanide = 2 metolazone = 25 HCTZ.

Furosemide

Torsemide

Bumetanide

Metolazone

Hydrochlorothiazide

Laboratory

Sodium (mEq/L) *

Values >138 are capped at 138 for the sodium term.

Total cholesterol (mg/dL) *

Hemoglobin (g/dL) *

Lymphocytes (%) *

Values >47% are truncated to 47%.

Uric acid (mg/dL) *

Values <3.4 are set to 3.4.

Disclaimer: Educational purposes only. The SHFM supports but does not replace clinical judgment, guideline-directed therapy, and shared decision-making. Validation and calibration vary by population.

Seattle Heart Failure Model — how it works

Multivariable prognostic model for survival in chronic heart failure (Levy et al., Circulation 2006). This page implements the Cox linear predictor and exponential baseline survival encoded in reference software (including the R package cliot).

Linear predictor (SHFM score)

Each factor contributes additively to a score that acts as the log-relative-hazard term in a Cox model. Medications and devices that reduce hazard enter with negative weights; markers of severity enter with positive weights. The score is the sum of these terms (see calculator for inputs).

Diuretic dose (furosemide equivalents)

Daily oral doses are converted to furosemide equivalents:

80 mg furosemide = 40 mg torsemide = 3 mg bumetanide = 2 mg metolazone = 25 mg hydrochlorothiazide

The model uses diuretic dose per kg body weight (mg/kg per day furosemide equivalent).

Truncations

Systolic BP is capped at 160 mmHg; lymphocyte percentage at 47%; uric acid is floored at 3.4 mg/dL; serum sodium is capped at 138 mEq/L for the hyponatremia term—consistent with common SHFM implementations and calculator help text.

Survival function

With baseline hazard constant λ = 0.0405 per year, baseline survival S₀(t) = e^−λt, and hazard ratio HR = exp(score):

S(t) = S₀(t)^HR = e^−λt·HR

Median survival (time to 50% predicted survival) = ln(2) / (λ × HR). The University of Washington SHFM site also reports mean life expectancy (area under the survival curve); that value can differ from median survival.

Reference

Levy WC, Mozaffarian D, Linker DT, et al. The Seattle Heart Failure Model: prediction of survival in heart failure. Circulation. 2006;113(11):1424–1433. doi:10.1161/CIRCULATIONAHA.105.584102

Background and purpose

Chronic heart failure carries a wide range of prognoses depending on left ventricular function, congestion, renal–electrolyte profile, comorbidity burden, and the intensity of guideline-directed medical therapy and device support. Clinicians often need a structured way to translate these heterogeneous factors into individualized estimates of survival for counseling, advanced therapy triage (e.g., transplant, durable left ventricular assist device), implantable defibrillator discussions in context of competing risks, and research comparisons using a common scale.

The Seattle Heart Failure Model (SHFM) addresses that need. It is a multivariable prognostic model that combines readily available clinical, laboratory, pharmacologic, and device information into a single risk score and translates that score into predicted survival over time. The model was developed using a Cox proportional hazards framework in a large heart failure population and was subsequently evaluated across multiple external cohorts, where it demonstrated useful discrimination for mortality in broad ambulatory and referral populations.

Derivation and conceptual framework

The SHFM originates from analyses of patients with chronic heart failure, including the PRAISE study population, in whom long-term follow-up and detailed baseline characterization were available. The investigators modeled time-to-death using a Cox proportional hazards model with a set of covariates that could be measured in routine practice. The result is a linear predictor (often called the SHFM score in research): a sum of weighted contributions from each factor, where positive weights correspond to features associated with higher hazard (worse prognosis) and negative weights correspond to therapies or findings associated with lower hazard (better prognosis).

Because the hazard of death is modeled in proportional hazards form, each factor acts as a multiplicative modifier of risk relative to a reference pattern. Medications and devices that reduce mortality in trials—such as renin–angiotensin blockade, beta-adrenergic blockade, mineralocorticoid antagonists, and selected device therapies—enter the model with negative contributions to the linear predictor when present, reflecting their association with improved survival in the derivation framework and/or supporting literature incorporated in the original modeling approach.

What the calculator inputs represent

Demographics and functional status

Age captures the steep gradient of cardiovascular and non-cardiovascular mortality risk across the lifespan. Sex is included because heart failure epidemiology and risk patterns differ between men and women. NYHA functional class encodes symptomatic limitation—higher classes reflect more severe limitation from exertion and correlate with hemodynamic burden, congestion, and exercise intolerance. Together, these variables situate the patient’s baseline risk before finer physiologic and laboratory measures refine the estimate.

Left ventricular systolic function and hemodynamics

Left ventricular ejection fraction (LVEF) is a central measure of systolic performance; lower values generally reflect more advanced remodeling and are associated with worse outcomes, though the model’s relationship is implemented through the specific functional form used in the published parameterization. Systolic blood pressure contributes to the risk profile; in many implementations, very high values are treated with an upper bound for modeling stability and to reflect how extreme values were handled in the derivation datasets. Weight is required because diuretic exposure is scaled per kilogram, reflecting that loop diuretic doses are interpreted relative to body size when assessing congestion intensity and hemodynamic fragility.

Ischemic versus non-ischemic etiology

Ischemic cardiomyopathy identifies a substrate commonly associated with scar, ischemic ventricular arrhythmia risk, and atherosclerotic comorbidity burden. The model assigns a separate contribution for ischemic etiology versus non-ischemic dilated cardiomyopathy, acknowledging that prognosis and pathways of care differ between these groups even when LVEF and symptoms appear similar.

Diuretic load and congestion

Loop diuretics are a cornerstone of decongestive therapy, but higher diuretic requirements often reflect more refractory fluid retention, higher filling pressures, and renal–neurohormonal derangement. The SHFM incorporates diuretic exposure using furosemide equivalents: daily doses of furosemide, torsemide, bumetanide, metolazone, and thiazide-type agents are converted to a common scale so that different prescribing patterns can be compared fairly. The resulting equivalent dose is then expressed per kilogram per day, emphasizing that a fixed milligram dose carries different implications in a small versus large patient.

Laboratory profile

Serum sodium integrates neurohormonal activation, free-water handling, and diuretic effect; derangements toward lower sodium are associated with adverse prognosis in heart failure cohorts.
Hemoglobin reflects anemia, renal perfusion, chronic disease, and nutritional status; the model uses a piecewise formulation around a reference range to reflect both low and high hemoglobin patterns as captured in the published structure.
Lymphocyte fraction (percent) serves as a marker of inflammation and nutritional/immune state; implementations often cap very high values to avoid outlier influence.
Uric acid is linked to oxidative stress, diuretic use, renal function, and metabolic comorbidity; very low values may be floored in modeling to avoid extrapolation beyond the training distribution.
Total cholesterol contributes as a continuous variable; in heart failure, cholesterol is interpreted cautiously because it may reflect illness severity, cachexia, and reverse epidemiology, yet it remains part of the published prognostic surface.

Neurohormonal and adjunctive therapy

The model includes indicators for renin–angiotensin blockade (ACE inhibitor or ARB), beta-blocker, mineralocorticoid receptor antagonist, statin, and allopurinol. In clinical practice, ACE inhibitors and ARBs are not combined; the calculator should represent one active renin–angiotensin pathway blockade at a time. These therapies generally shift predicted survival toward longer life expectancy when present, consistent with the magnitude and direction of hazard modification encoded in the model.

Devices

Implantable cardioverter-defibrillator (ICD) and cardiac resynchronization therapy (CRT) are included as discrete markers of device therapy associated with outcome differences in the modeled population. Device decisions in real life require guideline indications, rhythm substrate, electrocardiographic criteria, and patient goals; the SHFM quantifies association with survival at the population-model level rather than replacing device eligibility criteria.

How the score translates into survival estimates

The SHFM linear predictor is combined with a baseline hazard parameterization to produce predicted survival at fixed horizons (commonly one, two, three, and five years) and an estimate of median survival under the specified assumptions. In exponential-baseline formulations, survival is related to a baseline hazard rate and a hazard ratio derived from the exponentiated score, yielding survival curves that decline over time. The calculator’s output includes predicted survival percentages and a median survival time (time to 50% predicted survival) under the chosen parameterization.

It is important to distinguish median survival from mean life expectancy reported by some proprietary implementations: mean survival integrates the entire survival curve and can differ from the median when the distribution is asymmetric.

Truncations and boundary handling

Many SHFM implementations apply boundary rules so extreme laboratory or hemodynamic values do not dominate predictions beyond the range supported by the derivation data. Typical rules include:

Systolic blood pressure capped at an upper threshold (e.g., 160 mmHg) for the modeled term.
Lymphocyte fraction capped at an upper threshold (e.g., 47%).
Uric acid floored at a lower threshold (e.g., 3.4 mg/dL).
Sodium incorporated into the hyponatremia-related term using a capped value relative to a reference (e.g., 138 mEq/L), so hypernatremia does not invert the intended physiologic interpretation.

These conventions are not arbitrary “fixes”; they are practical safeguards for calibration and stability when applying a model trained on observed clinical ranges.

Clinical uses

Clinicians use the SHFM to frame discussions about expected prognosis, to compare modeled outcomes under different treatment scenarios (for example, adding a device class or intensifying neurohormonal therapy in a hypothetical “what-if” analysis), and to stratify risk in cohorts for research. Hospitalists and specialists may use it alongside acute illness indices (which emphasize short-term inpatient risk) because the SHFM is primarily oriented toward chronic heart failure trajectories rather than single-admission mortality.

Limitations and caveats

Model performance varies by population, era of care, and treatment intensity. Patients with acute decompensation, severe renal failure, valvular emergencies, or active ischemia may not be well represented by ambulatory chronic HF cohorts. The SHFM should not be used as a standalone decision rule for device implantation, transplant listing, or palliative transitions; it should complement clinical judgment, patient preferences, and contemporary guideline pathways. Online tools may apply additional updates (e.g., intravenous therapies, mechanical support, or revised hazard assumptions) that differ from transparent implementations used for reproducibility.

CalcMD.

Seattle Heart Failure Model (SHFM)

Understand the Seattle Heart Failure Model: Cox-based survival prediction in chronic HF using NYHA class, LVEF, labs, diuretic equivalents, medications, and devices. Learn how the score works and how to interpret SHFM outputs.