Safety Communication: You Get One Call (15 mins)

Important notices are common in healthcare, but opportunities for feedback can be hard to find. Yet, communication about problems is essential to a robust patient safety program. This episode looks at changes in technology that could improve safety communication across the healthcare industry.
Key Points:
-Safety communication in space
-Healthcare feedback loops
-Overwhelming public comment

References:
Sull, D; Sull, C (Oct. 18, 2023). How Do Nurses Rate Their Employers? MIT Sloan Management Review. https://www.sloanreview.mit.edu/article/nursing-satisfaction-index/.

Transcript

After creating a flawless, foolproof plan, NASA launched Apollo 13 on another mission to the moon in 1970. But they didn’t just sign off and watch the trip. They had prepared a feedback loop. A communication system is incomplete without an opportunity for the person receiving the message to send a response. It was in this feedback system that the most famous words of the entire mission were contained: “Houston, we have a problem” was a mission-critical message. It became more important than all the picture perfect scripts that envisioned the crew landing on the moon safely, and it became the impetus to a daring plan to circle the moon and return without landing after an oxygen tank exploded. Not only did the communication loop provide an opportunity to create a rescue plan, but it informed future missions. In my community, we watch SpaceX Starships launch from the beach with more weight, more power, and more miles planned for missions to Mars, but all of that knowledge is built on the missions of the past and the lessons learned from invaluable feedback both on the ground and in flight.

Yet, in healthcare, we have a terrible habit of building systems with no feedback loop. There is no way to raise red flags or share downtimes or update other departments on failures in another one. It’s not that the industry is lazy, it’s just that there is a lot going on. In complicated systems, there can be too much data and it can overwhelmed the receiver. But if NASA could pull it off in space, surely we can find a way to do it too? This podcast is all about solutions discovered in various industries, including healthcare, that could significantly benefit the communication systems of healthcare facilities.

For example, last year, researchers gathered free text data from 150,000 Glassdoor reviews written by U.S. nurses and classified the text into two dozen themes which were then used to predict overall ratings of an employer.[1] The researchers used Machine Learning AI to pull the four factors that were the strongest predictors for nurse satisfaction in the workplace. The resulting study basically had the power of 150,000 qualitative interviews and nurses across the nation nodded their heads in agreement. Most nurses tell HR that they are leaving because of better pay, but a significant trend in their online comments was the presence of a toxic work culture – so much that it was more than twice as predictive as compensation or workload in revealing their overall satisfaction.

Another fantastic use of machine learning is the Amazon summary of reviews. These days, people are loathe to buy a product from a seller without reviews – no reviews is even worse than poor reviews, because at least the poor ones mean you’ve been in business for awhile. The trouble is that there can be thousands of reviews on a popular product and it can be difficult to wade through it all to find out if the business is serving customers effectively. Further, if you asked a customer what they look for in a good shoe, they might say color and fit, for example, but that still doesn’t really tell you much.

So Amazon’s AI system now provides a helpful summary of trends noted in the existing reviews. The summary just highlights repeat comments, so if you’re looking for a narrow shoe and there are hundreds of complaints that the shoes are a bit narrow, then you’ve found a contender for your closet. It’s easy to look through and brilliantly summarizes the things that customers cared most about when they actually bought the shoes.

It's clear that the business world is harnessing machine learning to help customers. And it makes sense. Helping customers find products that they love is generally the goal of business, so any effort to make that process painless, and even enjoyable, is an advantage to everyone involved. But does it work the same way in healthcare? After all, we want our patients to find a treatment plan that they love, something that makes them feel better and sends them bragging to their friends about how wonderful their care is.

Let’s take a look.

One of the mainstays of patient education is something called “return demonstration” where the patient teaches and demonstrates what they have just learned. One-way teaching without any evaluation of learning is strongly discouraged in healthcare because a small misunderstanding can have highly hazardous results. For example, when a nurse was teaching a patient to inject insulin, they practiced on an orange because the peel of an orange feels similar to the pressure of injecting into skin. When the patient returned a week later, she was frustrated. Her blood sugar was still elevated even though she had injected the orange every day, just like she was told to do. That poor patient! Clearly, the teach-back had not included the step of actually injecting the medication into her own body, and the story lives on as a warning to new nurses who want to skip over education quickly as though it is a low priority. It’s not. And that feedback loop where the patient shows what they know is incredibly revealing. It highlights every topic that was rushed or skipped over and reminds us just how confusing and complex our medical language and treatments can be for the people who now must use them on a daily basis. With a little bit of experience under our belts, nurses learn to say, “Oh, I’m not doing anything for this round. I’ll talk you through it, but it’s all you this time.” That way, the patient doesn’t go home only to realize that they have unanswered questions about how to care for themselves, and they feel less worried and more independent about the discharge process.

We know that works best, but at the same time, there are so many demands on our time. If we could, we would spend all day teaching everything we know to a single patient, but that simply isn’t possible in the real world. There are items that are timed and a small delay can affect a patient’s well-being, delay other departments, and cause significant problems across the board. So we have to feed the deadlines first, and then we squeeze education and that all-important feedback loop into whatever time is left because that is reality. Life is all about priorities, and education is often the item to be deprioritized in a crunch.

I think incident reporting can be similar because it offers a critical feedback loop for staff to share case studies, experiences, or report near misses that could have been much worse. Yet staff often struggle to actually report those incidents because they take time to write, and the folks in the patient safety department struggle to read them because they pile up in proportion to the number of employees providing feedback. I remember my friend in patient safety saying that it was going to be a late night because she had to go through incident reports. I looked at the stack of paper and it had to be thousands of pages. How can someone possibly investigate all of that material?

On the other hand, I recalled frustration as a nurse because some of the incidents I experienced did not have an incident type in the report dropdown. There was no category for “other” to free-type what happened, and anything related to the Electronic Health Record, for example, just didn’t have any real place in the reporting system. How could I provide feedback about the dangers I saw in the EHR workflow when there was no actual way to provide that feedback in our existing system?

And then imagine that we could add error types. Would that stack of thousands of incident reports grow into millions of pages? It seemed untenable as it was, so how could we possibly expand our feedback opportunities?

But here is the catch. Good communication must have a loop to return feedback. Every communication book on the planet will explain that communication is a loop between sending and receiving with various influences on that process. When one person speaks, the other could respond verbally or nonverbally. If I send an email, I expect a response. If someone asks for my name, I expect them to introduce themselves. This rule is so steadfast that even the absence of a response is considered a response in the world of communication. This means that communication that only flows in one direction is just an announcement or an advertisement or junk mail, but it’s not considered communication. To qualify as legitimate communication, there must be a response. Even marketing materials have a call to action where the response is to purchase a product or service.

This is the crux of the issue: I often see an expected response in healthcare, but I don’t always see a pathway to make that response.

For example, the CDC published material about what was known and not known regarding COVID-19. Clinicians around the world mobilized and worked around the clock to find solutions for their patients, but there was no legitimate place to share findings. I was reading hundreds of stroke charts at the time as a team lead for an abstraction company. I saw trends in those patient charts, and I was trained to look for those trends when I worked in the ICU, but when I looked for opportunities to report them, I found nothing. I saw patterns with electrolytes and fluid balance, I saw trends in the use or lack of antiplatelet therapy, and I saw opportunities in care around the documentation of frightened frontline workers. But it impossible to share observations, and the expected behavior was unquestioning compliance with a long list of unknowns and no collaboration.

Wouldn’t it be great if there was a forum to return the communication loop? What if a physician could share a case study or make comments about common trends in a case history the next time a novel virus invades our communities? What if they could relay the frontline observations they made?

Of course, that would be a whole lot of paperwork. In fact, it’s never really been feasible to allow such open feedback from the public because the paperwork would be overwhelming, right? Yet, regulations are required to be shared for public comment 100 days before implementation. Public comment, overwhelming as it may seem, is truly part of our culture and political heritage. So what would make it possible to improve the feedback loop between the CDC and the public? How could we improve collaboration, expand opportunities for solution discovery, and even build trust again with a public that has become immunized against our advice? It was impossible until now.

This is where I get excited, because today, machine learning allows us to cull through hundreds of thousands of comments and identify the most common trends for further study. If we opened a forum for public comment on a disease, clinicians could login and share findings, from vitamins and off-label medication use to a common history among infected individuals. Frontline providers could raise the questions that everyone is asking and machine learning could pull them into a neat, tidy summary for CDC officials to get a window into the clinic, the hospital, and the home without censorship or the invasion of our privacy.

And remember that giant stack of printed incident reports? Well, those incident reports were typed, not handwritten, and that means it was machine-readable. Can you imagine how helpful it would be for a hospital, or even a hospital system, to deep-dive with machine learning to discover trends and gaps in patient care? It really can be overwhelming to send an individual through all that data. It swallows patient safety professionals like a hungry ocean at high tide. But if it could be curated and summarized neatly by machine learning, it would be revolutionary to the field.

Patient advocacy is the job of everyone. It’s a basic facet of virtue and it’s not complicated to understand that we all benefit in improved patient care – because one day we’ll all going to be in the bed rather than beside it. But advocacy is not about asking for what we want for the patient. It requires listening. We have to get feedback from the patient, even if it takes time, even if it’s nonverbal, and even if it takes time and study to figure it out, because otherwise we are advocating for ourselves so we will feel better about the patient rather than advocating for the patient and their actual needs. So if machine learning could help us become better listeners, I would be all for it.

One thing I know is that the amount of data in healthcare today is untenable. It is simply not feasible to have a solid grip on patient safety today. The best safety professionals use prioritization and pick their battles carefully because it’s simply not possible to investigate everything that happens in a busy facility. But that prioritization process doesn’t have to be intuitive. We just need to adapt our technology and habits to pull what is most useful and then cut what’s not useful anymore (like fall bands, cough cough).

We cannot begin to hear our patients, or elevate diverse voices, or advocate for anyone until we tune our devices to the same channel. Identifying trends in healthcare’s big data could allow us to adjust the volume and tune into our patient’s needs more accurately, helping us rise above the static and interference to achieve real communication.

Personally, I’d like to see those results. The only remaining question is who will be the first to make history with a famous line like “Houston, we have a problem”?

[1] Sull, D; Sull, C (Oct. 18, 2023). How Do Nurses Rate Their Employers? MIT Sloan Management Review. https://www.sloanreview.mit.edu/article/nursing-satisfaction-index/.