By Jonathan Pabalate, DNP, CRNA, APRN | Founder, JPCAIC — JPC Anesthesia Informatics Corp | Nurse Anesthesia Faculty, University of North Florida
It is after hours in an operating room in rural Iowa. No case, no patient, nobody on the table. Just me, a camera on a tripod, and almost 2 score of nurse anesthesia residents watching from a thousand miles away, because I cannot fly them out here and this is the closest I can get them to a real machine in a real room.
I reach over to the ventilator and start walking the flow trigger down. Three liters a minute. Then one. Then half a liter.
And the ventilator starts breathing.
Not once. Rhythmically, about seventy times a minute, the waveform changing color with every cycle to announce that the patient had just taken a spontaneous breath. My heart beats around seventy. It was beating next to my lungs, pushing just enough gas back and forth in the circuit to clear the threshold I had lowered a moment earlier.
There was no patient. I made the machine say there was.
Purple does not mean the patient is breathing
Every fall I tell a new class the same thing, and every year somebody in clinical un-tells them within a month. You will be told that if the screen turns purple on one machine, or shifts from yellow to blue on another, the patient is breathing.
That is not what that means.
What it means is narrower, and far more useful. The machine has decided that gas moved toward the patient faster than the flow trigger you set, and sustained it longer than the trigger window you set, inside the slice of the respiratory cycle it was watching. Three conditions. You chose two of them.
So the color is not an observation that your patient breathed. It is the result of a rule you configured, run against a signal from a flow sensor. Those are very different claims, and only one of them survives a bad day.
The threshold belongs to you
The flow trigger is a number in liters per minute, and on most adult anesthesia machines it lives around two, the smallest inspiratory flow a healthy adult produces without much trouble. The trigger window is a percentage, usually around twenty-five, describing how much of the expiratory pause the machine spends listening and how long an effort has to persist before it counts.
Set that number low and the machine will believe almost anything. It will believe a hand releasing a reservoir bag. It will believe a heartbeat. Set it high and you buy certainty, and you pay for it by missing the patient who is genuinely trying and cannot make the bar.
There is no correct setting. There is only which error you would rather make in this room, with this patient, right now. That is the other end of the argument, and it is the part that never makes it onto the slide. It is also the first sensitivity and specificity tradeoff most residents ever hold in their own hand and turn with a knob.
Somebody has already measured the trade
In 2018 a group in São Paulo did the bench experiment properly. Plens and colleagues, in Respiratory Care, drove five ICU ventilators with simulated cardiogenic oscillations at escalating amplitudes and asked what it takes to stop the machine from inventing breaths. For every additional centimeter of water of oscillation, the trigger had to be desensitized by about 2.51 liters per minute to abolish false triggering. And every liter per minute of desensitization cost about 4.79 milliseconds of response time to the patient’s real effort.
That is the tradeoff with a number on it. You do not eliminate the error. You move it, and you pay in latency.
Now the counterweight, because this is the part that usually gets left out. In Thille and colleagues’ 2006 study in Intensive Care Medicine, twenty-four percent of ventilated ICU patients had an asynchrony index above ten percent, but autotriggering accounted for under one percent of those events. Ineffective triggering was eighty-five percent. At ordinary settings the common failure is not the machine hallucinating a breath. It is the machine missing one, and that failure was associated with a less sensitive trigger.
Both errors live on the same dial. I demonstrated one end of it in an empty room. Most of your patients are quietly suffering from the other.
The stakes are not hypothetical
McGee and Mailloux reported nine patients in Neurocritical Care with catastrophic brain injury who appeared to be breathing spontaneously on patient-triggered modes, despite absent cranial nerve function. When support was withdrawn and formal apnea testing performed, there were no spontaneous respirations. Minimal changes in circuit flow, unrelated to any respiratory effort, had been cycling the ventilator. The authors name the consequences plainly: delayed recognition of brain death, with downstream effects on organ donation and on supporting the family. Dodd-Sullivan and colleaguesreported the same pattern in Progress in Transplantation.
The same physics I used as a classroom party trick has, in other rooms, told families something that was not true.
Sometimes you move the threshold on purpose
I recently anesthetized a man in his eighties with severe COPD and emphysema. His baseline saturation on room air sat in the low nineties, and that was his normal, not his emergency. The plan was spontaneous ventilation on a supraglottic airway under a heavy propofol infusion. He was going to breathe shallowly, because that is what propofol does, and I knew it before I drew it up.
At a two liter per minute trigger, he would have gotten almost nothing. The machine would not have believed him.
So I took the flow trigger down to about one and gave him generous pressure support. When the propofol went in he dropped into the mid eighties, then sat around eighty-eight for ten minutes. I was content with eighty-eight, because he lives at ninety-two.
I deliberately made the machine more credulous. I accepted a higher rate of false breaths in order to catch his real ones, because on that man a missed effort cost more than a phantom one. That is not a technical adjustment. It is a clinical decision about which failure you can afford, made with your understanding of how the equipment supports the pathophysiology of whoever landed on your table that morning. It is not an exact science. It takes a fair amount of tinkering. But you have to know what the tinkering does to the sensors, or your ventilation will be all over the place.
A flat capnograph is also a claim
Everything so far has been a machine reporting something that did not happen. The same logic runs the other way, and that direction is worse.
The checkout on most modern machines asks you to occlude or disconnect the gas sampling line so it can test the circuit for leaks. Near the end, it asks you to put it back. It does the same thing with the APL valve, which you spun up to 50 a few steps earlier to see whether the circuit holds pressure, and which the checklist then tells you to bring back down, because a patient breathing against 50 of PEEP is a different kind of morning. The checkout does not only verify the machine. It reconfigures it, and every reconfiguration has to be undone before anyone gets wheeled in.
I once sat back and let a resident run an induction alone. Go fly, I said, I am just going to sit here. They preoxygenated, they pushed drugs, and I watched the capnograph stay flat.
The sampling line was on the floor. It had never been reconnected after the check.
The ASA Standards for Basic Anesthetic Monitoring name carbon dioxide as the arbiter: correct airway positioning “must be verified by clinical assessment and by identification of carbon dioxide in the expired gas.” So when the patient stops breathing and you reach for the mask, your preceptor asks you one question. Are you ventilating? And the way you answer it is that you see CO2.
That resident was never going to see CO2. Not because the patient was not being ventilated, but because the line carrying the answer was lying on the tile.
The purple waveform was true to the sensor and false about the patient. A flat capnograph on a well-ventilated patient is the same error running backward, arriving at the one moment you have the least time to work out which of the two you are looking at. It is a very easy detail to be tricked into forgetting when you are rushing.
Every number on that monitor is somebody’s threshold
Once you see the flow trigger for what it is, you cannot unsee it anywhere else. A threshold you set, a valve you spun, a line you unplugged. Every reading is downstream of all of it. The averaging window on your pulse oximeter is a configuration decision. So is the stimulation current on your nerve monitor, the smoothing on your processed EEG, and every alarm limit in the room, including the ones your partner silenced before you walked in.
Drew and colleagues counted 2,558,760 alarms over thirty-one days in an ICU population and found 88.8 percent of the annotated arrhythmia alarms were false. That is not a broken monitor. That is a monitor configured, reasonably, to miss nothing, and the cost of missing nothing is being wrong almost nine times in ten.
Which is where this stops being a ventilator story. The predictive models arriving in perioperative care bring one new thing and one very old thing. The new thing is that the signal is more complicated than a propeller spinning in a limb. The old thing is that somebody still chooses the operating point, and it is almost never the person standing at the head of the bed.
If you understand why my heartbeat drove a ventilator at seventy breaths a minute, you already understand why the deterioration alert fires forty times a shift. Same dial. Bigger machine. Fewer people in the room who know it can be turned.
What the screen cannot hand you
Later in that same session I turned the ventilator off and put my hand on the reservoir bag, which is what most of us do the moment a patient starts to fight the vent. There is less work of breathing on the bag. There is also something else.
I do not need the screen to tell me what that is. I can feel it there.
So I told them what is going to happen to them this year. You will be standing at the head of the bed at the end of a case, staring at the monitor, waiting for it to tell you something. And a good preceptor is going to reach over, take your hand, and put it on the bag. Put it right here. Just rest it right here like this. And you will feel it.
I am the informatics person. I build the models, I teach the technology, I made them an AI-generated podcast about vaporizers so the material would follow them into the car. And the most trustworthy respiratory monitor in that operating room, the one with no threshold to set, nothing to configure, and no way on earth to be talked into a breath that never happened, was a palm resting on a rubber bag.
That one does not travel over a camera. It has to be handed to you, by somebody standing next to you, in a room where a real patient is waking up.
I told them it begins this semester. It does.
Works Cited
Dodd-Sullivan, Rebecca, et al. “Ventilator Autotriggering: A Caution in Brain Death Diagnosis.” Progress in Transplantation, vol. 21, no. 2, 2011, pp. 152–55. https://doi.org/10.
Drew, Barbara J., et al. “Insights into the Problem of Alarm Fatigue with Physiologic Monitor Devices: A Comprehensive Observational Study of Consecutive Intensive Care Unit Patients.” PLOS ONE, vol. 9, no. 10, 2014, e110274. https://doi.org/10.
McGee, William T., and Patrick Mailloux. “Ventilator Autocycling and Delayed Recognition of Brain Death.” Neurocritical Care, vol. 14, no. 2, 2011, pp. 267–71. https://doi.org/10.
Plens, Glauco M., et al. “Effect of Cardiogenic Oscillations on Trigger Delay during Pressure Support Ventilation.” Respiratory Care, vol. 63, no. 7, 2018, pp. 865–72. https://doi.org/10.
“Standards for Basic Anesthetic Monitoring.” American Society of Anesthesiologists, Committee on Practice Parameters, approved 21 Oct. 1986, last amended 15 Oct. 2025, www.asahq.org/standards-
Thille, Arnaud W., et al. “Patient-Ventilator Asynchrony during Assisted Mechanical Ventilation.” Intensive Care Medicine, vol. 32, no. 10, 2006, pp. 1515–22. https://doi.org/10.