Understanding “Human Error”

Humans make mistakes. Any system that depends on perfect performance by humans is doomed to failure. In fact, the risk of an accident is more a function of the complexity of the system than it is the people involved. Humans are not the weak link in a process. We are a source resilience. We have the ability to respond to unpredictable inputs and variability in the system. The contents of this post are based on the work of Sydney Dekker in his book “The Field Guide to Understanding Human Error.”

Professor Dekker is a pilot and human factors engineer. Most of his work comes from analyzing industrials accidents and plane crashes. One such crash was the accident in May 1977 where one jet rammed into another killing 583 people. 

Now we can blame the pilot for the crash. Had the pilot performed better, this accident could have been avoided. If we remove such bad apples, the system works fine. 

However on deeper inspection there were multiple causes (non-standardized language, bad weather, overly crowded runway, equipment issues, etc). It was not simply “human error” that caused this crash, but a series of problems. Understanding all these causes reveals that pretty much any pilot could have made this mistake. The system needs to change to promote pilot success. 

Casting blame makes us feel like we’ve offered an appropriate response to a terrible event. However blaming does not improve the system so the next person doesn’t make the same mistake. In order to learn from our mistakes, we need to understand why they happened. 

Local Rationality and Just Culture

No one comes to work wanting to do a bad job. 

Sydney Dekker

The local rationality principle asks us to understand why an individual’s action made sense at the time. “The point is not to see where people went wrong, but why what they did made sense [to them].” We need to understand the entire situation exactly as they did at the time, not through the benefit of retrospection. 

We balance the need to keep people accountable while acknowledging that most adverse events are not due to “human error.” We emphasize learning from mistakes over blaming individuals. We need zero tolerance for blameworthy events like recklessness or sabotage while not unfairly blaming individuals for system problems. 

Just Culture Algorithm

The Just Culture algorithm asks a series of questions to determine the cause of an adverse event and offers an appropriate response. If an act was a deliberate act of sabotage, then severe sanctions are necessary. If reckless behavior led to the adverse outcome, the individual should be held accountable. However if the any individual’s actions in the same context could have led to the same result, then it is hardly fair to blame that person. 

  1. Did the individual intend to cause harm? Did they come to work in someway impaired? This is sabotage.
  2. Did the individual do something they knew was unsafe? This is reckless behavior.
  3. Does the individual have a history of similar events with similar root cause? This person is not learning from prior mistakes.
  4. Would three peers have made the same mistake in similar circumstances? This passes the substitution test. It is a no blame error. 

Analyzing Adverse Events

The single greatest impediment to error prevention in the medical industry is that we punish people for making mistakes.

Dr. Lucian Leape

The old school format of Morbidity and Mortality conferences pit the person who made the error against a room full of experts with the benefit of hindsight. This adversarial arrangement encouraged people to hide their mistakes. We needed a new approach if we wanted to encourage bringing errors into the light for analysis to learn from these mistakes. Dekker describes six steps. 

Step One: Assemble A Diverse Team

The team should include as many stakeholder perspectives as are pertinent. In medicine, we would include physicians, nurses, technicians, patients and others. This team needs to have expertise in patient care (subject matter expertise) and in quality review. The one group not included are those who were directly involved in the adverse event. Their perspective will be incorporated through interviews, but they do not participate in the analysis. 

Step Two: Build a Thin Timeline

In airplane crashes, investigators recover the flight recorder (black box) to create a timeline of events during the flight and conversations between parties. In medicine, we look at the chart to understand what happened and when. This is a starting point, but excludes the context needed to understand local rationality. 

Step Three: Collect Human Factors Data

Interview the people directly involved in the adverse event to understand what happened from their point of view. This is best done as early as possible as memory tends to degrade with time. Understand what was happening in the room, why did they make the choices they did, and what was their understanding of the situation and why. 

George Duoros presents a series of questions on the EMCrit Podcast to guide the collection of this human factors data. 

Collecting Human Factors Data (George Duoros)

Step Four: Build a Thick Timeline

With the human factors data in hand, overlay this on the thin timeline to build a thick timeline. This presents the events as they occurred within the context under which the providers were working. You may need to go back to interview providers until you can understand what happened as they understood it at the time. Then we achieve local rationality. 

Step Five: Construct Causes

We don’t find causes. We construct causes from the evidence we collect. The causes of the error are complex and are not readily available to be discovered. We need to work to understand and propose possible causes. One method of organizing the causes is in a Ishikawa diagram (or fishbone diagram). 

Ishikawa (fishbone) diagram to analyze potential causes of adverse events. The adverse event is placed at the fish’s head on the right. Off the spine are potential areas where errors may arise. From each rib, place the potential error and supporting details. 

Step Six: Make Recommendations

Brainstorm for potential solutions that would prevent others from having the same outcomes. Ideally recommendations are worded such that they are specific, measurable, achievable, relevant and time-bound. 

Final Thoughts

Remember that information is protected. It includes patient data and as such is protected under HIPAA. Do not put it in publicly available platforms such as Google Slides or Zoom. 

Additionally, the entire quality improvement process should be a safe space to encourage providers to examine their errors. As such, it is protected under the Patient Safety and Quality Improvement Act of 2005 (Public Law 109-41), signed into law on July 29, 2005. Use an approved slide template which includes the appropriate language, for example: 

This document is privileged and confidential under the Illinois Medical Studies Act and should not be shared or distributed other then through the Quality Assurance Committee structure.

Dr. Douras recommends the following agenda for a 30 minute M&M case: 

  • Introduction: Remind the group that this is about learning and identifying systemic problems, not about blame & shame.
  • Present the thin and thick timelines: this should take about 10 minutes, excluding extraneous information. It can be presented by a junior resident, but they would need the support of a senior facilitator to keep the discussion on track. 
  • Discuss the case: identify potential causes possibly using a fishbone diagram with the group. This should also last only about 10 minutes
  • Look for systemic problems and solutions: the goal of the exercise is to identify potential solutions that would prevent a similar mistake from happening again. The bulk of the time should be spent in this section: 10 to 15 minutes


  1. Sydney Dekker’s “Field Guide to Understanding Human Error”
  2. Angels of the Sky: Dorothy Kelly and the Tenerife Disaster
  3. EMCrit 249 – You Can Either Learn or You Can Blame – Fixing the Morbidity and Mortality Conference with George Douros
  4. The Patient Safety and Quality Improvement Act of 2005

Fuzzy Matches in Google Sheets

When trying to build our curriculum inventory I needed to match thousands of strings in Google Sheets. Doing this manually would have been tedious and prone to error. The EQ function would fail if there was one character different, forget about objectives that had words shifted around. Here’s my script that outputs the percentage of string A that is present in string B averaged with the percentage of string B that is in string A. I found that anything over 40% seems to be semantic match.

function howMuchMatches(string1, string2) {
  // clean both strings
  var array1 = string1.trim().toLowerCase().replace(/[^\w\s\d]/g,"").split(" ");
  var array2 = string2.trim().toLowerCase().replace(/[^\w\s\d]/g,"").split(" ");
  // how much of string1 is in string2
  var numberOfMatches12 = 0;
  for (i=0; i < array1.length; i++) {
    if (array2.indexOf(array1[i])>=0)
      { numberOfMatches12++; }

  // how much of string2 is in string1
  var numberOfMatches21 = 0;

  for (i=0; i < array2.length; i++) {
    if (array1.indexOf(array2[i])>=0)
      { numberOfMatches21++; }

  // convert those to percentages
  var percentMatch12 = numberOfMatches12/array1.length;
  var percentMatch21 = numberOfMatches21/array2.length;

  // return the average the two
  return (percentMatch12 + percentMatch21)/2; 

Infective Endocarditis

IE is an infection of native or prosthetic heart valves, endocardial surface or an indwelling cardiac device (such as a pacer or defibrillator). In recent years, we’re seeing it happen in older and more chronically ill patients. And more MRSA.

The diagnosis is tricky because the presentation can be fairly subtle. You need a high index of suspicion, fever and:

  • unclear infectious source
  • new regurgitant murmur
  • embolic events of unknown source

Get 3 sets of BCx in the ED before antibiotics and possibly a TTE. Patients should receive IV Abx and 1/2 of them require valve replacement or debridement surgery. Indications for surgery are:

  • development of heart failure
  • perivalvular abscess formation
  • uncontrolled infxn
  • Large, Mobile vegetations (that may flick off)


How to Take Smart Notes

In his book1, Sönke Ahrens describes two things to improve writing productivity:

  1. Routines for writing
  2. A system for organizing notes and ideas

He breaks down the daunting task of writing by pointing out:

  • Writing a book or manuscript would be easy if someone gave you the first draft.
  • Writing the first draft would be easy if someone gave you an organized set of ideas.
  • Generating the ideas would be easy if someone gave you a set of properly referenced notes from the literature.

Obviously, the “someone” in each of these scenarios is the same person. The routine comes from reversing the process, starting with taking a properly referenced set of notes – hence the title of the book.


True Grit: The Surprising, and Inspiring, Science of Success

The importance of perseverance

Angela Duckworth’s work suggests that perseverance is a predictor of success. During her graduate student days she created a “grit scale” which she subsequently tested throughout her career. She characterized “grit” as working hard and finishing what one begins and gives the example of Will Smith explaining in an interview that if he was in a competention on a treadmill, there would be only two outcomes: he would be the last one running or he’d die on that treadmill because he “will not be out worked.”


Non-Traumatic Lower Back Pain

Most of us will have lower back pain in our lives (80–90% lifetime prevalence) and it accounts for 2–3% of ED visits (so quite a bit).

As with everything in EM, there are benign and serious causes and we need to differentiate between the two. The serious causes include diagnoses in the back and those in the abdomen or retroperitoneum.

  • Benign: muscular and ligament strain, sciatica (posterolateral disk herniation) and spinal stenosis
  • Serious back: cancer, spinal epidural abscess, vertebral osteomyelitis, infectious diskitis, spinal epidural hematoma and giant (central) disk hernation (cauda equina syndrome).
  • Serious non-back: AAA, renal stones, renal infarct, tumor, pancreatitis, pancreatic cancer, PUD, cholecystitis, retroperitoneal hemorrhage, psoas abscess


Graham-Cassidy Bill Doesn’t Add Up

The latest Republican health care bill meant to replace the Affordable Care act comes from Senators Bill Cassidy (LA) and Lindsey Graham (SC). This one presents a more substantial change to the way things are run. Most of this comes from an article in the 9/23/2017 issue of the Economist.

  • The Federal government would play a much smaller role, giving money to the states proportional to the number of inhabitants between 50% and 138% of the Federal Poverty Level.
  • The decisions on how to structure care is passed down to the states. They can petition Health and Human Services to drop ObamaCare provisions such as the Essential Health Benefits that were meant to provide a basal level of care in each plan.

The thought is that this may encourage experimentation in each of the states, however three problems exist.

  1. States are now responsible for structuring care. There is no guarantee they can do a better job. States are also required to have a balanced budget, so the Medicare money may go instead to paying other debts.
  2. There is no increase in money if conditions change. If a state is hit with a disaster, say an opiate epidemic, and requires more money to address this… there is no more money.
  3. Doesn’t fix existing problems with the health insurance marketplace. Premiums have already risen under the ACA and insurers are leaving markets. This new plan has the potential of further raising premium prices (with healthy people leaving the market without the pressure of a mandate) and discouraging insurers from participating (with less healthy people, they are taking more risk).

My Email Rules

Email can consume hours of your day stealing opportunities to do real work. Several people have devised plans to reclaim this time, so I stole the ones that work for me.

These rules have one main goal: to respect my time and that of the receiver.

1. Keep as short as possible.

“I’m sorry to write you a long letter, as I did not have time to write a short one.”

Mark Twain