Automated drug delivery system
In this study, by extending the systems that we reported previously [15–17], we developed an automated drug delivery system to control the infusion rates of landiolol and dextran, and the injection of furosemide to reduce MVO2 without inducing circulatory collapse in subjects with acute HF. Figure 1 shows the scheme of our system. The details of the system are described in Supplemental Material. In brief, the user sets the target values of mean AP (AP*) and mean PLA (PLA*). The system measures AP, CO, PLA and right atrial pressure (PRA), and then low-pass filtered these hemodynamic variables at a cut-off frequency of 0.1 Hz to calculate mean values to be used in the closed-loop feedback system. From the measured hemodynamic variables, the system calculates the slope of the Frank-Starling curve for left (SL) and right (SR) ventricles, systemic vascular resistance (R) and stressed blood volume (V). From AP* and PLA* and the measured hemodynamic data, the system determines the target values of SL (SL*) and V (V*). To minimise the difference between SL* and SL, a proportional-integral feedback controller adjusts the infusion rate of landiolol. To minimise the difference between V* and V, a nonlinear feedback controller adjusts the infusion rate of dextran or the injection of furosemide. Using these feedback controllers, the system administers landiolol and dextran or furosemide to bring mean AP and mean PLA to the preset target values.
Animal experiments
Rapid pacing-induced heart failure model
We used five adult mongrel dogs weighing 21.0-29.5 kg (male/female, 4/1). We induced anaesthesia with intravenous thiamylal sodium (25 mg·kg-1), performed endotracheal intubation, and maintained an appropriate anaesthesia level during the experiment by continuous inhalation of isoflurane (1‒2%). Body temperature was maintained between 37 and 38°C. We performed transthoracic echocardiography to measure left ventricular end-diastolic dimension (LVDD), left ventricular end-systolic dimension (LVDS) and ejection fraction under normal conditions. After inserting a bipolar pacing lead (Model BT-60P, Star Medical Inc., Tokyo, Japan) to the right ventricular apex through the right jugular vein, we connected a generator (Model SIP-501, Star Medical, Tokyo, Japan) to the pacing lead and implanted it in a subcutaneous pocket at the neck [16]. We closed the incisions and withdrew anaesthesia. One day after implantation, we started rapid ventricular pacing at a rate of 230 beats·min-1 and continued for three weeks to induce HF.
Experimental preparation and automated drug control
We performed experiments the day after discontinuing rapid pacing. Under anaesthesia induced as described above, we performed transthoracic echocardiography to assess LVDS, LVDD and ejection fraction, and recorded an electrocardiogram (ECG) to calculate HR. We placed 8-Fr sheath introducers in the right femoral artery to measure AP, and in the right and left femoral veins for infusing dextran and landiolol, respectively, and placed a 10-Fr sheath introducer in the right jugular vein to measure PRA. We inserted a catheter to the coronary sinus via the right jugular vein under fluoroscopy. After a left thoracotomy and pericardial incision, we introduced a catheter into the left atrial appendage to measure PLA. We placed ultrasonic flow probes at the ascending aorta (20PS; Transonic, Ithaca, NY) and the left circumflex artery (2.5PS; Transonic, Ithaca, NY) to measure CO and coronary flow, respectively.
We attached an infusion pump (CFV-3200, Nihon Kohden, Tokyo, Japan) for administering landiolol, and a roller pump (Minipulse 3, Gilson, Middleton, WI) for administering dextran. We controlled these pumps via a laboratory computer (LC-72N10, Logitec, Tokyo, Japan). We used the sheath introducer at the femoral vein for injecting furosemide according to a command signal from the computer. We digitised all hemodynamic data at 200 Hz with an analogue-to-digital converter (AD 12-16, Contec, Osaka, Japan) and stored the data in a dedicated laboratory computer system.
Experimental Protocol
After stabilisation for 30 min, we connected the closed-loop system to the animal. We set AP* as 10 to 15 mmHg lower than baseline AP, but not lower than 70 mmHg. We set PLA* as baseline PLA, but not higher than 18 mmHg. After activating the system by closing the feedback loops of drug administration (Fig. 1), we recorded the infusion rates of landiolol and dextran and the injection of furosemide on the computer. The performance of the system was monitored for 60 min, and arterial and coronary sinus blood samples were collected simultaneously at baseline (0 min), 30, and 60 min after system activation.
After completion of the protocol, the dogs were euthanized with an intravenous injection of pentobarbital and potassium chloride. We measured left ventricular weight after excision of the adjacent right ventricular muscle and valvular tissues.
Myocardial oxygen consumption and blood gas analysis
We measured oxygen contents of the arterial and coronary sinus blood samples using a co-oximeter (AVOXimeter 4000; Instrumentation Laboratory, Bedford, MA). According to Fick’s principle, the product of coronary flow and the difference between arterial and coronary sinus oxygen contents yields MVO2. We normalized MVO2 by 100 g left ventricular weight (LVW). We also performed blood gas analysis of arterial blood samples using a blood gas analyser (ABL800 FLEX; Radiometer, Tokyo, Japan) to assess pH, electrolytes, lactate, and partial pressure oxygen and carbon dioxide.
Data analysis
Efficacy of the automated drug delivery system.
To evaluate the precision and stability of the system, we calculated the performance error (PE), median PE (MDPE), median absolute PE (MDAPE), and wobble by the following equations [19].
where t represents a time unit. Divergence is the slope of the regression line between |PE(t)| and t (min). MDPE, MDAPE, wobble, and divergence indicate the bias, accuracy, stability, and trend of the absolute error, respectively. Since haemodynamics was stabilised after approximately 15 min, we calculated PE for AP and PLA from 15 to 60 min after the system was activated.
Statistics
Data are expressed as median (interquartile range). We used Friedman’s test followed by the post hoc Conover’s test for multiple comparisons among different time point data. We performed all statistical analyses using R version 3.4.3 (R Foundation for Statistical Computing, Vienna, Austria). We considered differences to be significant at p < 0.05.