VAD Support Simulator Experimental

One modelled adult on a HeartMate 3-type LVAD or BiVAD. Adult only.

1. The model running

Configuration

Pick a scenario, or move any control while it runs. Each scenario's size is an illustration, not a measured patient.

Patient states sizes are illustrations

More controls RV, rhythm, aortic valve, reflex
FlowL/min; shading marks suction
  • —— LVAD
  • ···· Aortic valve
Pump flow …

Modelled flow through the LVAD, not the controller's estimate.

Flow pulsatility (model) …

Not the pump's PI.

Arterial pressure mmHg …

Systolic/diastolic (mean).

LV pressure–volume loopmmHg against mL
  • —— Last beat
  • - - - Starting state, 5,300 rpm (the fitted patient, not a target)
  • ···· Same modelled heart, no pump, default settings, fixed (not a stopped pump, not this patient before implant)
CVP (mean right atrial pressure), per beatmmHg, last 60 s
  • —— Mean CVP of each beat

2. What this shows, and what it cannot

It teaches how a continuous-flow pump and a failing heart share the work: what speed, blood volume, afterload, the right ventricle, rhythm and a leaking aortic valve do to pump flow, filling pressures and suction. Adult only.

Flow pulsatility (model) is not the pump's PI. The HeartMate 3 shows a pulsatility index computed from motor power over 15 s, on a scale of about 2 to 6. This page shows the swing in modelled pump flow divided by its mean. Its direction can teach; its value cannot be set beside a controller's. With the reflex on, it moves the opposite way to Belkin 2022's table in aortic insufficiency and in severe dehydration.

Model, equations and sources

What is being solved

A closed-loop lumped circulation: four chambers with time-varying elastance, valves as orifices, systemic and pulmonary arteries with inertance, arterioles, capillaries and veins (Korakianitis & Shi 2006). An LVAD takes blood from the LV to the proximal aorta; in BiVAD a second pump takes it from the right atrium to the pulmonary artery. Each pump is the HeartMate 3 pressure-flow curve of Boes 2019 (Table I), with its inertance, its periphery and its backflow region. Suction is Simaan 2009's inflow resistance, which rises steeply once inflow pressure falls below 1 mmHg. Twenty-six ordinary differential equations (sixteen for the circulation and pumps, nine for the reflex, one for the phase of the beat) are integrated in your browser with a fourth-order Runge–Kutta step of 0.1 ms. Halving the step changes no readout in the third decimal, and blood volume is conserved to within a millionth of a millilitre.

The adult patient: five parameters (LV and RV end-systolic elastance, stressed blood volume, systemic and pulmonary resistance) are fitted exactly to the five baseline means of Uriel 2017 Table 3 at 5,307 rpm. Heart rate 80/min is assumed: the paper prints none. Nothing in the ramp response is fitted.

The baroreflex

Ursino 1998, as printed: a lead-lag filter on carotid sinus pressure, a sigmoid for afferent firing, a falling exponential for sympathetic and a rising sigmoid for vagal efferent firing, and for each effector a pure delay, a logarithmic gain and a first-order lag (Eqs. 30–42, Table 3). Every parameter is Ursino's. How it is coupled to this patient is ours:

Ursino's reflex is a normal adult's. Whether an LVAD patient's matches it, and how a nearly flat arterial pulse changes it, was not checked: his own model shows pulsatile perfusion lowering the reflex gain.

Checks against published data

HeartMate 3 speed ramp, change per 100 rpm from 5,000 to 6,000 rpm. Patients: Uriel 2017, n 16, mean ± SD.
Per 100 rpmAbstractTable 4Model, reflex offModel, reflex on
Cardiac output, L/min+0.08 ± 0.08+0.14 ± 0.15+0.10+0.11
Wedge, mmHg−0.48 ± 0.27−0.48 ± 0.28−0.26−0.79
CVP, mmHgnot significant−0.06 ± 0.10+0.12+0.20
Mean arterial pressure, mmHgnot given+0.34 ± 0.83+2.02+0.96
Mean PA pressure, mmHgnot given−0.56 ± 0.470.00−0.50
Heart rate, /minnot givennot given0 (fixed)−1.5

With the reflex, mean arterial pressure rises half as fast and mean PA pressure now falls at the patients' rate. Wedge now falls too fast, and CVP still rises: nothing in the model lowers pulmonary resistance or helps the RV as the LV unloads, which is Uriel's own explanation for a flat CVP.

Direction of change against Belkin 2022 Table 2 (HeartMate 3), reflex on, from 5,300 rpm (pump 4.25 L/min, flow pulsatility 0.31). Sizes are ours.
State (model change)Flow, Belkin / modelPulsatility, Belkin / modelModel flow, L/minModel pulsatility
Dehydration, 400 mLdown / downeither / down3.940.20
Severe dehydration, 1,000 mL, with suctiondown / downup / down3.340.26
Fluid overload, 750 mLup / upeither / up4.800.42
Hypertension, arteriolar resistance ×1.6down / downup / up3.820.33
Severe RV failure, RV contractility 20%down / downvariable / down3.910.25
VF, no contractiondown / downvariable / zero3.600
Aortic insufficiencyup / updown / up5.480.35
Cardiac recovery, LV contractility ×2variable / upup / up4.410.61

Flow goes the published way in every row. Pulsatility goes the published way in 2 of the 4 rows where Belkin gives one direction. With the reflex off it is 4 of 4: the reflex's faster heart rate and stronger contraction raise the swing in aortic insufficiency, and in severe dehydration it mobilises enough venous blood that suction starts later, from a lower pulsatility. Below the suction point, pulsatility falls as volume is lost (0.31 to 0.05 at 750 mL removed); it jumps once suction starts.

BiVAD against Xiao 2026. Patients: after implant, mean ± SD, different patients in each group. Model: the same patient with RV contractility 20%, LVAD 5,200 rpm, reflex on, without and with an RVAD at 4,000 rpm (RA to PA), the speeds Xiao simulated.
After implantLVAD, n 49BiVAD, n 13Model, LVADModel, BiVAD
Right atrial pressure, mmHg7 ± 46 ± 4145
Wedge, mmHg11 ± 614 ± 61117
PA systolic/diastolic, mmHg28/1237/2225/1729/29
Mean PA, mmHg19 ± 828 ± 92129
LVAD flow, L/min4.7 ± 0.15.2 ± 0.23.84.3
RVAD flow, L/minnone4.0 ± 1.3none4.3

The directions Xiao reports, in patients and in their own simulation, come out here: PA pressure higher with BiVAD, wedge higher (in patients a trend, p = 0.07), native RV output lower, and, at 3,000 rpm where the model's RV still ejects, RVAD flow slightly lower while it does. Sizes do not all match: LVAD flow is about 1 L/min below the patients' in both configurations, right atrial pressure on the LVAD alone is twice theirs (this model's RV is set at 20%), and the PA pulse on the LVAD alone is 8 mmHg against their 16. The PA pulse in BiVAD does not come out at all: in the model the RV stops, or almost stops, ejecting through the pulmonary valve once the RVAD runs at about 4,000 rpm, so PA systolic and diastolic pressure meet. Read PA pulse pressure in BiVAD as a gap, not a result.

The pump and the left ventricle on their own

The HeartMate 3 curve reproduces Boes 2019's Table IV slopes to within 0.1 mmHg per L/min. Simaan 2009's LV and pump model reproduces its Fig. 5 pressures within 2 mmHg, and its suction-onset speeds within 220 rpm (20 to 220) across five resistances, unfitted.

Errors in the papers, and how they are read here

Why there is no child setting

Scaled to a child's size by Lundquist 2026's allometric rules and run on Tran 2024's paediatric mock-loop inputs, this model gets LVAD-supported flow within 5 to 16% and in the right order, but LV peak pressure about 50% too high and mean arterial pressure 15 to 30% too high. Unassisted, the scaled infant is hypertensive (systolic 174 against 82 mmHg). The arteries come out too stiff, and the source that scales arterial stiffness by age does not print the adult value it scales from. The reflex is a 70 kg adult's, and the only pump curve is an adult device's. A child setting would put adult or higher pressures, and a loop drawn from them, beside a 15 kg child. A label cannot fix that, so the setting is left out.

Why an RVAD on its own is not offered

Xiao 2026, the one paper with measured haemodynamics on durable right-sided support that we read, studied only BiVADs: every RVAD ran beside an LVAD. With nothing to check an isolated RVAD against, it is left out.

Sources

Verification

Experimental and not independently verified. There is no official calculator to compare a physiology model with, so the comparators are the published results above. The scenario sizes and the coupling of the reflex are this page's own choices, stated where they are used.

The model runs in this browser; nothing you set on this page is stored or sent.

Educational use only. A teaching model of one adult, not a patient-specific prediction, not a device setting and not a treatment recommendation.