Hospital ICU ventilator
An "autopilot" for the bellavista ventilator
Starting situation
The bellavista ventilator is used in intensive care units its area of application spanning from premature babies to adult patients. Because a patient’s condition can change rapidly, it must be possible to adjust the ventilation parameters (respiratory volume, respiratory rate, and oxygen supply) quickly. Uninterrupted patient monitoring by medical staff is not possible for economic reasons and due to limited human resources. The challenge for nursing staff is also to select the numerous parameters in such a way that the goal of ventilation is achieved as well as possible. This requires technical know-how as well as medical expertise.
Project objectives
The aim is to automate the ventilation settings in order to increase the quality of ventilation and simplify ventilator use by nursing staff. The number of possible settings is to be reduced and instead be determined automatically by the device. The remaining parameters should be easy to understand and should be able to directly influence the goal of ventilation (exchange of oxygen and carbon dioxide).
Technical implementation
After extensive research of the corresponding literature and analysis of interviews with physicians, the algorithm was broadly sketched. Because the minute volume (the product of respiratory rate and respiratory volume) is associated with the exchange of carbon dioxide, the user should specify the minute volume instead of rate and volume. The challenge is that there are an infinite number of solutions to achieve a minute volume based on frequency and volume. Consequently, from this multitude of solutions, the algorithm must select the solution that represents the best and optimal combination for the patient.
What is the “best” combination for the patient?
In principle, the algorithm needs a model that describes what is good and bad for the patient. This is known as a cost function in optimization theory. In other words, a function or model is needed that shows how much a ventilation setting “costs” (in terms of respiratory rate and respiratory volume) to minimize these costs accordingly. This is because current research has found that the released output (product of flow and pressure) from the ventilator to the patient may be associated with lung damage. Thus, the potentially hazardous output should be kept as low as possible. A model is needed that describes the output released to the patient. The output to be expected during a ventilation cycle can be derived using physical modeling as follows (equation 1):
\[ \dot{W}_{insp} = \frac {1}{2\cdot C} \cdot f \cdot \left( \frac {MV_A}{f} + V_D\right)^2 \cdot \left(1 + coth \left( \frac {T_I}{2 \cdot R \cdot C}\right)\right)\]
The search for the optimal combination of rate and volume—and thus the solution to the optimization problem—is conducted numerically in this case using an analytically derived fixed-point iteration. A fixed-point iteration means that the best combination cannot be directly calculated, but instead must be approximated in an iterative numerical process. The process can be stopped once the change between the iterations is no longer significant.
From model to adaptive ventilation
The equation above demonstrates that the cost function depends on the “resistance” (R), “compliance” (C), and dead space volume (VD) of the patient. Compliance and resistance are key performance indicators for lung elasticity and airway constriction, respectively. The dead space indicates how many regions of the respiratory system are not involved in gas exchange. It is thus clear that the cost function is dependent on current lung properties. In other words, the model must be repeatedly determined and analyzed during ongoing ventilation. This ensures that the rate and volume settings are continuously optimized and adapted to the patient’s condition.
The model parameters are determined using a statistical method based on the measured airway pressure, exhaled CO2, and output flow. The device learns to determine the status of the patient’s condition based on the measured data and responds accordingly by maintaining the released output to a minimum.
From idea to product
The initial step exclusively involved simulations. Here, the patient and ventilator were simulated using MATLAB and Simulink to execute the developed algorithm. This enabled the results to be reviewed and evaluated early on by clinical experts (Figure 1). Once the algorithm had been refined and perfected based on clinical feedback, the implementation on the device was started. Using lung simulators, a wide array of scenarios was then tested again to evaluate safety and reliability and increase them, wherever possible.
Clinical studies under controlled conditions were only able to be conducted after the entire development cycle had been completed (V-model) and all system tests had been successfully passed. Such studies were conducted in a hospital in order to validate the new operating principle. The ethical guidelines were of course adhered to in the process.
Results
Six out of nine setting options are now automatically defined. The adaptive ventilation mode (AVM) continuously adapts to the patient’s condition. The health status of the lung is estimated, and the respiratory effort and respiratory pattern are continuously analyzed and classified using a statistical method. Intelligent algorithms then ensure that the ventilation parameters are automatically adjusted. This guarantees the most optimal ventilation possible.
In order to protect this innovative solution and generate lasting value, a corresponding patent application (EP3656429B1) has been successfully filed for our customer.
