Can you hear the sirens?

Presenting ambulance response patterns during a coronavirus surge using multimedia

Yaning Wu
7 min readMay 13, 2021

During London’s third coronavirus lockdown this winter, I became more aware than ever of the sounds that dominate our urban spaces when they aren’t filled with people.

Some of the most vivid were the whining ambulance sirens that increasingly punctuated my weekly grocery runs. Treating them as a mainstay of city life before the pandemic, I began in January to mentally plead with the passengers of each passing vehicle to stay alive. There seemed to be so many of them, some transporting critically ill children (indicated with special signage) and others racing back and forth from hospitals across the capital to reach all kinds of people experiencing the scariest moments of their lives.

Maybe you’ve heard these sirens and felt the same way during your country’s most serious COVID-19 outbreak. I didn’t know if my ears were playing tricks on me, so I decided to see what the data said about ambulance dispatch patterns during severe waves of COVID-19.

1. Gathering data

I wanted to use the most granular version of data available to analyse time-series trends in ambulance incidents, so I searched for records of every single ambulance call in London during the past year. Though the UK’s National Health Service has comprehensive information on ambulance quality indicators, it only releases weekly statistics, so I had to look for a different country’s data.

I ultimately chose another place important to me: the New York City borough where I grew up. Queens, made up of several districts on the city’s east side and the most linguistically diverse place on Earth, is shown in grey in the map below (my old aptly-named neighbourhood is coloured dark blue). After the Bronx, Queens was the NYC borough hit hardest by COVID-19 during the March — April 2020 outbreak.

NYC Open Data provides more than 20 million live-updated data points on ambulance incident dispatches (one point for each incident). I used these data to quantify the impact of COVID-19 on the number of incidents and ambulance response times in Queens between February 15th and April 30th in 2019 and 2020, filtering and keeping only incidents of severity level 1–3 (i.e. life-threatening events such as cardiac arrests or traumas) for respective final N’s of 109,780 and 128,279. I define response times as the length of time between the initial logging of the incident in the dispatch system and the arrival of first responders at the scene of the emergency.

2. Visualising the data

In the first plot below, two lines show the progression of total daily ambulance dispatches from mid-February to the end of April for both 2019 and 2020. Even without the legend, it may be obvious which year each line represents.

In 2020, the number of dispatches peaked on April 6th, with paramedics attending to 2,782 incidents on that day. In 2019, on the other hand, peaks are hardly noticeable, with fluctuations seen throughout the timeline.

The dashed red line roughly located in the middle of the plot represents the day when New York State’s governor declared a “stay-at-home” order. On March 22, non-essential retail and gatherings were halted, physical distancing was mandated, and individuals were urged to limit outdoor recreational activities and their use of public transport. We know that many incidents requiring an ambulance response happen outside the home: in public spaces, during travel, or in large gatherings. With all these potential risks reduced compared to 2019, Queens residents were still in greater need of genuine emergency medical help for more than three weeks in 2020 than 2019. It’s difficult to conceive of a more telling picture of COVID-19’s devastation in this community.

The next plot illustrates another difference in ambulance statistics between 2019 and 2020, this time using response times.

Median ambulance response times in 2019 during the months specified remained between six and seven minutes, while they reached 9.41 minutes at their peak in 2020 on April 6th (the same date when incident numbers peaked). Again, median response times began inching upwards right after the March 22 stay-at-home order (or perhaps even before). Ambulance services in high-income countries like the U.K. and U.S. have set an approximate standard of eight minutes for life-threatening calls, which Queens services may have met prior to the pandemic but struggled with during those critical weeks in March and April. Consider also that road blockages and traffic delays would have been minimised from the end of March, which would normally reduce transit times. Removing these obstacles failed to substantially ease the burden on medical personnel.

Having seen the number of dispatches Queens first responders faced during the toughest weeks of the pandemic, it would be a mistake to blame their competency for these longer response times. But what caused this delay?

Reporting from Bloomberg and the New York Post suggests that patient pile-ups at emergency rooms, personnel shortages, and worried residents were contributing factors during the first wave of the pandemic.

These delays have likely cost lives, no matter their statistical significance. They have also left first responders feeling overwhelmed at each helpless situation they encounter, with many arriving in homes where COVID-19 patients could not be saved.

Despite the tragedy that these plots can’t fully display, inklings of hope exist for ambulance response during COVID-19. After initial spikes in both incident numbers and response times immediately after March 22, figures began to decline rapidly, and data from the fiscal year ending last June 30 suggest that response times actually decreased compared to 2019 in the entire city. With the easing severity of the outbreak and less congestion on the road, this is good news that we shouldn’t be surprised by. Nevertheless, a robust health infrastructure, especially surrounding emergency services, is vital to ensuring better ambulance outcomes.

3. Can I display this data differently?

Finally, having presented and interpreted Queens’ ambulance data, let’s return to my original question of interest to discuss more unconventional ways of informing an audience about New York City’s crisis.

I was curious about whether the overwhelming number of sirens I heard this winter in London were indicative of a relationship between COVID-19 outbreak timelines and ambulance incident numbers (and by extension, response times). I chose a different city than I first intended, but nonetheless demonstrated a pattern for both response variables using descriptive visualisations.

But the sounds of these ambulances often signal their presence before their flashing lights, as they are designed to do. So why not translate this auditory experience into data visualisation? Not visualisation — sonification.

The video below shows what sonifying one specific dataset, ambulance response times in Queens between February 15 and April 30, 2020, could sound like. Please turn up your volume and use earphones if you have them — I should warn you that these sounds aren’t pleasant.

Does that sound familiar?

I believe that this uncommon form of data presentation brings two benefits to the table:

Accessibility

The very nature of data visualisation can exclude people with visual impairments all the way from those who are legally blind to those who are colourblind. If done well, data sonification can bring to life a line or area chart through only auditory means to allow users with low vision to explore time-series visualisations just like anyone else. For example, here’s a chilling sonification project by a Queen’s University Belfast professor describing the early stages of the COVID-19 pandemic:

The replication of human experiences

Again, only if done well, sonified data can add humanity to large datasets. This being my first time with this style of presentation, I used only basic methods to tell a story about ambulance incidents. Please let me know whether I’ve been successful!

  • First, I set the frequency limits (Hz) of the sonification to 300 (min) and 1550 (max) to approximate how emergency sirens typically sound.
  • Then, I placed background white noise behind the main sonification to both differentiate median response times below eight minutes (the aforementioned common standard for first responders) and emulate the sounds of walking along a busy street. With this dataset, I was lucky for the purposes of this work that the cut-off of white noise occurred exactly one day after NY’s stay-at-home order, meaning that someone going outside for groceries may have heard exactly what I’ve tried to depict: a quiet neighbourhood punctuated only by blaring sirens.
  • Finally, I chose to indicate greater median response times with higher frequencies because of the urgency that these frequencies convey.

Appendix: Notes on code and methodology

  • Here are the R packages I used to clean my data and produce the outputs shown in this article:
library(data.table) # to separate a date and time object into two columns for date and timelibrary(ggplot2) # to plot grouped line chartslibrary(gganimate) # to animate grouped line chartslibrary(ggthemes) # to make plots "look nice"library(sonify) # to sonify data
  • I used Canva to produce the YouTube video linked above, and Mac’s SketchBook app to produce the first ambulance sketch.
  • For ambulance response times, I chose the measure of median instead of mean because the former is less influenced by outliers (these were plentiful in the data — some ambulances took 20 minutes to arrive).
  • I excluded incidents with missing incident response time values, reducing my sample size slightly.

Thank you for reading! I would welcome any critiques of my methods and results :)

--

--

Yaning Wu

she/her. Population Health student @ UCL. Perpetual dataviz nerd. Published on Towards Data Science and UX Collective.