By clicking “Accept All Cookies”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Privacy Policy for more information.
Thought leadership

Webinar: How Seattle FD focuses on Quality Improvement with Corti

Henrik Cullen
Webinar: How Seattle FD focuses on Quality Improvement with Corti

How does quality improvement help you build a world-leading PSAP?

At this webinar, Henrik Cullen from Corti, and Dr Cathrine Counts from Seattle Medic One, took us under the hood of the QI process at Seattle Fire Medic One and the accompanying artificial intelligence software, that helps Seattle and its call-takers remain among the best in the world. Missed the live session? We've included a summary of the key points below, about:

  • The problem with manual QI processes
  • The most common misconceptions about quality improvement
  • Interventions SFD has taken to optimisations to actually change behaviour 
  • How Corti helps SFD make quality assurance analysis  
  • How Corti’s Artificial Intelligence (AI) can be integrated to create a continuous quality improvement flow 

You can watch the full webinar here.

For Seattle Fire Department, the old ways of reviewing calls involved a lot of back and forth and manual processes which included:

  • Excel spreadsheets, and a manual process of downloading the audio on a desktop player, and pausing he audio to check the timestamp, then paste it.
  • One 911 call could last 30-40 minutes to review.  
  • Essentially the time it took to analyse calls takes 4-5x longer than an average dispatch call due to the back and forth of playing, pausing, and replaying the audio to understand what was being said.
  • Complaint calls could take up to several hours of piecing together data across CAD systems, with multiple audio recordings.

Given the novel technologies used at SFD, all of these processes are faster because you can do all of the review tasks within one single platform using a drag and drop feature, so the processes are more streamlined in the end. 

Four favorite misconceptions about quality improvement 

Misconception 1: We’re too busy for quality improvement 

  • If you don’t invest in QI as a process, it’s going to cost you time on the back end anyway. The reality is, if you want to prioritise QI into your workflow, you can. It just takes some intentionality and recognition that you are going to shake up the usual process.
  • Theoretical framework of QI, the occurrence of a worst case scenario in a 911 call where a cardiac arrest isn’t identified during the call is not the result of a singular event or process, but it is typically a combination of failures
  • Sometimes things are out of our control such as a secondhand caller or language barriers - these are known as “latent conditions” - we can’t really prevent these from happening but there are solutions to alleviating their detrimental impact
  • On the other end of the spectrum, there are the more active failures, often phrased as ‘human error’. These can induce not appreciating agonal respirations, or thinking a fall is just a fall without confirming consciousness. These mistakes can often be forgiven on an individual level, but it doesn’t mean there isn’t something we can do at the organizational or structural level to safeguard against them from happening again. 

With this misconception, either we invest in front end evaluation, or we pay with back end corrections.

Misconception 2: Identifying bad apples is the most important use of time 

  • Reflects a rather kind of old school mentality of ‘this is how we’ve always done things around here” so it sparks the question how we can change behaviour within an organisation
  • The idea that the source of all of our problems can be blamed on a select group of individuals or a single individual is rarely true, but we waste an inordinate amount of time and resources believing otherwise
  • Nothing wrong with setting thresholds or holding people to certain standards but if you only focus on the low performers, then you don't move the needle on the system as a whole
  • You just make your low performers look more like your middle of the pack performers. And often with a substantial investment in time and energy. There's a reason your low performers are low performers. If instead we focus our energy on the entire system of performers, and thus help everyone improve, we move the fuel gauge rather than just the needle. 

Misconception 3: We have no way to link outcomes, there is nothing to improve

  • With out of hospital cardiac arrest, this misconception simply doesn't work. Because 9 times out of 10, the dispatch centre is going to get immediate feedback that CPR is in progress, and thus you know the diagnosis 
  • OHCA is so well studied. You can use benchmarks created via research to evaluate your systems performance.
  • In Seattle there is a tiered response system with a huge focus on sending the right level of resources to the right patient in the right amount of time 
  • Current dispatch data isn’t equivalent to a gold standard diagnosis - there should be some variation and indication of the frequency a dispatcher gets the diagnosis “correct”. This enables to turn to the high performers to evaluate their performance for better understanding on what good looks like and modify the system form there 

Misconception 4: Quality Assurance is too difficult 

  • Just because something is hard, doesn't mean it isn't the right thing to do
  • Let’s be real, the science behind QA is the easy part, it’s the culture change that allow you to change behavior is the hard part 
  • The Iceberg analogy applies here - what you see above the surface are the formal rules and regulations of an organisation such as the protocols etc. While these play a part in how an organisation and its members behave, the real power is below the surface where perceptions, traditions, and the unwritten rules dictate behaviour. 

Sometimes the right thing to do is also the hard thing to do. 

How Corti helps SFD make quality assurance analysis  

  1. Removes the manual labour and long hours of analysis where the outcome is a fraction of what can be analysed with the help of Corti  e.g flagging time to detection, listing the order of questions asked to patients
  2. Standardizes processes between two EMS systems to allow for comparability (e.g. Seattle and Perth) 
  3. Accelerates the time it takes to review, evaluate and label calls which otherwise would take too long and thus not dedicated resources are scarce 

How Corti’s Artificial Intelligence can be integrated to create a continuous quality improvement flow 

  • A typical improvement process or framework model is called PDSA
  1. Start with PLANNING  
  2. DEVELOP a hypothesis and define the problem 
  3. SET a baseline design experiment and carry out the intervention 
  4. Study that it's a measure if that intervention had any results 
  5. Understand if it’s a fit for you running the same analysis at the planning stage
  6. See if there was an improvement 
  7. See if the intervention had the desired effect
  8. ADJUST and iterate 
    - If it didn’t have the desired effects (Go back to the priming phase) 
    - If it did, should you do it more, or is there a next step? 

This is pretty useful for analysis and quality assurance – how?  

  1. You can map out all the questions you would typically ask in a call, particularly the ones you’d ask before dispatch if you’re trying to optimise to reduce dispatch 
  2. Corti’s software offers you an overview that visualises the distribution across all your calls for the time at which a call taker asked a particular question
  3. If you want to start tracking address time which would take months and months of analysing data manually, with Corti’s AI you can create this query at any time to see patient consciousness questions across the entire data 
  4. You can see the median time it took to learn if a patient was conscious 
  5. You can investigate further and/or move onto the next question for analysis 
  6. Instead of guessing, you can rely on the computer’s identification to know what you should invest your time and resources in to improve
  7. You can create an intervention to run an experiment on something that was detected as unusual 
  8. You can look at optimising question phrasings and how this affects speed and quality of diagnosis 
  9. You can enhance your protocols by flagging cases where there are deviations to understand where and how are people getting the best results 
  10. You can use Corti’s machine learning model that was developed together with SFD, where the AI can analyse a call and then predict what level of severity on a five point scale the call is  
  11. You still need to listen to the audio, as we’re not quite at the point where AI can do all the job for us. So we still need to be part of the process. 

For more info contact us or book a demo at: