On April 15, 2019, Notre Dame cathedral, the iconic and beloved monument, burned and was nearly lost in its entirety despite an enormously complex and expensive fire alarm system built exclusively to prevent such a disaster.
As a post-mortem examination has revealed, there are lessons for all of us who design, implement, and depend on warning systems – whether they be for detecting fires or detecting clinical events like sepsis.
Construction of the cathedral began in 1163 and was a multi-generational effort. A masterful example of Gothic architecture, its flying buttresses and other remarkable features have lent it recognition, admiration, and love the world over.
But architectural features like buttresses and the rib-vaulted roof were much more than visually appealing design elements: they served a functional purpose, allowing the downward pressure of the roof and other structures to be displaced. Without them, the massive ornate windows and open nave would not have been possible. The cathedral’s uniqueness and beauty were dependent upon complex interrelated elements, each serving a purpose, and each depending upon the other to maintain the building’s integrity.
Cathedrals, Hospitals, and Incident Detection
In many ways, the architectural systems that allowed Notre Dame to stand, and the ancillary systems needed to protect the structure, are no different than the systems required to maintain the integrity and functioning of a modern hospital – or any other complex system for that matter.
One of these systems is incident detection. In this context, incident detection is a broad term which includes not just systems that monitor structures, but also systems that monitor patients and other clinical activities.
While incident detection systems come in many varieties, the overriding purpose is the same for all: shorten the time-to-act (T2A).
Something (an event) has taken place which requires detection. Information regarding the detection must be reported to those imbued with the knowledge, ability, and responsibility to act. And, finally, action is required even if it is merely the cognitive acknowledgement that an event has taken place.
Each piece of this puzzle (recognize, report, respond) is required if timely action is to take place. Failure of any piece can bring the entire edifice down.
Where Hospital Incident Detection Fails
In the hospital setting, rapid response systems have proliferated for the purpose of shortening the T2A to improve outcomes for patients who deteriorate, because preventing outcomes such as unanticipated ICU transfer or death is very time dependent.
Classically, the rapid-response literature describes four components of a rapid response system, two of which are clinical (afferent and efferent limbs), and two which are organizational (process improvement and administration).
Focusing on the two clinical limbs, the afferent (a.k.a. recognition) contains elements which recognize and detect patient instability while the efferent (a.k.a. response) includes elements required for appropriate intervention.
Typically, reporting is embedded in the afferent limb. Because the reporting aspect of a rapid response system is often underemphasized or altogether overlooked, insufficient efforts are made to ensure its integrity and it is a common cause of failure.
Here, we see a similarity to the Notre Dame catastrophe.
Where Notre Dame’s Incident Detection Failed
The fire at Notre Dame was anticipated – almost expected – given the materials used internally to support the roof. Underlying the lead-tiled roof was a complex system of old oak timbers referred to as “The Forest.” The cathedral’s administrators were well aware that the Forest was a fire risk.
To head off such an event, tremendous efforts were expended to craft a customized afferent smoke and fire detection system combined with well-trained, well-rehearsed fire fighters as the complementary efferent response team.
The smoke detection system worked as designed and the Parisian fire department performed in an exemplary manner. Yet despite this, the cathedral was all but lost in its entirety.
The failures were confined to the reporting process, a series of communication failures that would be all too familiar if they had been part of a hospital’s root cause analysis of a sentinel event. As determined by journalists from the New York Times, here’s what occurred:
- The cathedral’s smoke detection system went into an alarm state at about 6:18 a.m.
- The alarm system’s notification went to a single station manned by a single attendant within a side building – a single point of failure. As luck (bad luck) would have it, the employee on duty that day was new; it was his third day on the job.
- The alarm attendant’s shift had ended but his replacement didn’t show up, so he was working an unanticipated double shift.
- The smoke detection system messaging component was an almost indecipherable mix of common language and numeric codes. This information had to be decoded and subsequently communicated to a security person within the cathedral who would manually evaluate the situation. Essentially this appears to be a process used to minimize false alarms and secondarily confirm a true positive event.
- The security person responded to the call from the alarm attendant by going to the wrong structure – to the Sacristy (another building to the side of the church) rather than the church nave. Perhaps this took place because the fire alarm zone was reported to be the “Attic Nave Sacristy”?
- Manual confirmation that a fire was in progress was delayed, so subsequent notification of the fire department was also delayed by 30 minutes – 30 minutes the cathedral didn’t have to spare.
It seems, in retrospect, to be a rather typical though unfortunate example of the ‘Swiss cheese’ model of accident causation.
The Lesson for Hospitals
If we transpose these events into a healthcare setting, what might we conclude?
- Certainly, the idea of funneling all information regarding a clinical event to a single inexperienced person will create weakness in the process. Should that person be experienced, well rested, and prepared to respond? Of course.
- Are there tools hospitals can use to notify more than one person simultaneously that a clinical incident (like patient deterioration or sepsis) has been detected? Yes.
- Can information about clinical incidents be transmitted clearly and with context so that “decoding” by the recipient isn’t needed? Yes.
- Can clinical alarms be passed directly to the nurse or team directly responsible to act, bypassing intermediaries and thus mitigating errors of delay and translation? Yes.
- Can a rapid response team (as but one example) be automatically and directly notified once an appropriate clinical alarm has been triggered? Yes.
- Can hospital staff determine what the positive predictive value of alarms will be – and use that to determine whether it is prudent to send alarm information directly to responding personnel? Yes.
Despite the extraordinary and expensive efforts expended by the staff of Notre Dame de Paris to recognize and respond to a fire, the dependence upon a manual reporting process led to catastrophic failure.
As you evaluate your hospital’s numerous alerting systems, can you say you’re in any better shape that Notre Dame was on their fateful day?
This article was originally published on Vocera and is republished here with permission.