Michael Reynolds died from a sore throat.
The 29-year-old father visited his GP with throat pain and difficulty swallowing. The doctor diagnosed tonsillitis, prescribed antibiotics and a nasal spray, and sent him home. Within 24 hours, Reynolds was dead from epiglottitis as his throat closed and oxygen couldn’t reach his lungs.
The December 2023 inquest concluded there was a “missed opportunity for urgent emergency treatment.” But I see something deeper in Reynolds’ death. It reveals a fundamental flaw in how we train physicians to recognize when common symptoms signal something deadly.
The Diagnostic Training Problem
Reynolds’ case exposes a critical blind spot in primary care training. We teach physicians to be efficient diagnosticians of common conditions, but we fail them when it comes to recognizing the subtle markers that separate routine cases from medical emergencies.
The problem isn’t that Reynolds’ GP didn’t know what epiglottitis was. Our current diagnostic framework doesn’t adequately prepare physicians to pause and consider life-threatening alternatives when they see familiar symptoms.
We’ve created a system that rewards quick pattern recognition and throughput. But epiglottitis and tonsillitis can look remarkably similar in the early stages. The difference often comes down to nuanced clinical judgment about severity and progression that requires a different kind of thinking entirely.
What’s particularly troubling is that difficulty swallowing, which Reynolds experienced, should have been a red flag demanding immediate escalation. But our training protocols don’t sufficiently emphasize these critical decision points where common presentations diverge into dangerous territory.
We’re teaching physicians to diagnose the most likely condition rather than systematically ruling out the most dangerous one first. Reynolds died because his GP was confident in the wrong diagnosis, not because he was uncertain.
The Scale of Diagnostic Failure
Reynolds isn’t an isolated tragedy. An estimated 795,000 Americans die or are permanently disabled by diagnostic error each year.
The numbers for epiglottitis specifically are even more damning. Between 35-50% of adult epiglottitis cases initially go misdiagnosed, making Reynolds’ case tragically representative of a broader pattern.
What’s happening is a cognitive bias issue. When a physician sees a sore throat, their brain immediately goes down the “common cold, strep throat, tonsillitis” pathway. Once you’re in that diagnostic tunnel, red flag symptoms get interpreted through that lens rather than as warnings to completely change direction.
Difficulty swallowing becomes “severe tonsillitis” instead of “potential airway emergency.”
Smart Triggers and Structured Doubt
The solution lies in what I call “smart triggers” and building “structured doubt” into clinical practice. These are specific symptom combinations that automatically activate a structured checklist, but only when certain thresholds are met.
For throat complaints, you could have a simple protocol: any patient with sore throat PLUS difficulty swallowing PLUS fever automatically triggers a 60-second rule-out checklist for epiglottitis, peritonsillar abscess, and other airway emergencies.
The physician doesn’t have to remember to be suspicious. The system makes them suspicious at the right moments.
This actually saves time for routine cases because physicians can move confidently through straightforward diagnoses, knowing the system will catch them when something’s potentially dangerous. It’s like having guardrails on a highway that don’t slow down normal driving but prevent catastrophic accidents.
We could also implement “progression protocols.” If a patient returns within 48 hours with the same complaint but worsening symptoms, that automatically elevates the case to emergency consideration. Reynolds might be alive today if there had been a hard stop requiring A&E referral for any throat complaint patient who returns with progression.
Technology Resistance and Legal Reality
The technology exists to build these decision support tools right into electronic health records. The challenge isn’t technical. It’s getting physicians to trust that these protocols enhance rather than replace their clinical judgment.
The resistance comes from two deeply ingrained aspects of medical culture. First, there’s “diagnostic pride.” Physicians are trained to see clinical reasoning as their core competency, so anything that feels like algorithmic oversight can be perceived as questioning their expertise.
Second, there’s legitimate concern about “alert fatigue.” Physicians have been burned by poorly designed systems that constantly interrupt them with irrelevant warnings.
But the liability question is actually what’s going to drive adoption faster than anything else. Right now, when a physician misses a diagnosis like Reynolds’ case, the legal standard is whether they met the “reasonable standard of care” for their profession. But once these decision support systems become available and proven effective, that standard starts to shift.
Research shows that more than half of all malpractice payments made by large academic health systems were connected to incidents that might have been preventable with clinical decision support.
We’re probably five years away from the tipping point where NOT using these systems could become the negligent act. Physicians who resist will find themselves explaining to a jury why they chose not to use technology that could have saved a patient’s life.
The Practical Path Forward
Healthcare leaders don’t need to wait for regulatory mandates. The most practical first step is diagnostic auditing: systematically reviewing cases where patients returned within 72 hours with the same complaint but worse symptoms.
Start by identifying every case in the past year where a patient came back with a progression of their original symptoms and ended up in emergency care. Look for patterns. Throat complaints that became airway emergencies. Chest pain that became heart attacks. Headaches that became strokes.
This gives you your organization’s specific “miss patterns” and shows you exactly where decision support would have the highest impact. Every healthcare organization has different miss patterns based on their patient population and physician training.
Reynolds-type cases are happening in your system right now. You just haven’t identified them yet.
Once you have that data, you can build simple protocols around your highest-risk scenarios. This approach builds internal buy-in because physicians can see their own near-misses rather than being told about theoretical problems.
The Stakes Are Clear
The goal isn’t to preserve some idealized version of clinical practice. It’s to create a system where 29-year-old fathers don’t die from misdiagnosed sore throats.
Reynolds died in the gap between what’s technically possible and what’s legally established as standard care. The technology exists today to prevent similar tragedies. We’re waiting for enough evidence to make these systems legally defensible while patients continue to die from preventable diagnostic errors.
The question isn’t whether this technological transformation will happen. It’s how many more Michael Reynolds cases we’ll tolerate before we make it happen faster.
Healthcare leaders who act now, while these systems are still seen as going above and beyond rather than legally required, will find themselves protecting both patients and their organizations from the diagnostic failures that continue to devastate families like the Reynolds’.
The technology to save lives like Michael Reynolds’ exists today. The only question is whether we’ll use it.
