Survivorship Bias in Stroke Care
Survivorship Bias in Stroke Care
PolicyViz Newsletter #47
Happy New Year, everyone! I know it’s mid-January, but this is my first newsletter of the new year. I hope you had a great holiday season and your 2024 is off to a great start. Anyways, here’s the newsletter……
Late last year, I presented my work on the Urban Institute’s Do No Harm project to analysts at a state department of health. (I’ve changed specific numbers in this post and am not naming the state or hospitals to keep the information confidential.) Following my presentation, one of the analysts who works on stroke care across the state told me about some of the data that had caused a bit of a stir in their data community.
At an early 2023 meeting, the department’s data team presented information about a worrying trend in stroke care at a specific rural hospital in the state. Hospitalization prevalence had increased from 8 percent of inpatient discharges in January 2021 to 10 percent in December 2021. The data showed that this hospital had more than 700 additional stroke patients discharged in 2021. But this analyst wasn’t worried because she recognized the increase as an example of survivorship bias—a logical error that can lead to false conclusions by concentrating on people (or things or communities) that pass a particular selection process.
Because this analyst deals with facilities specializing in stroke, she knew some key information about hospitals providing stroke care across the state. First, this hospital had added interventionists to do more thrombectomies between 2021 and 2023. Second, the hospital was planning to pursue a higher level of certification as a thrombectomy-capable center, which is a designation given to hospitals specializing in stroke care. The application for that designation was submitted in late 2022 and because it usually takes two to five years to pursue the designation, the hospital was clearly ramping up care in the years before.
There is also a geographic issue at play here. The analyst told me that if people in this community were experiencing more strokes, the hospital would not necessarily see an increase in the number of stroke discharges as there were only a handful of hospitals offering higher levels of stroke care in the state. Stroke patients, therefore, travel throughout the state (or to other states) to go to a thrombectomy-performing stroke center.
Thus, it wasn’t the case that the number of strokes in this community were increasing, but that the local hospital had increased its ability to provide more and better care for people with strokes. Patients would no longer have to travel to hospitals in other parts of the state or in other states as their local hospital was now able to provide a higher level of care. Instead of the increasing numbers representing a worrying trend, they actually demonstrated a community that had better access to health care.
More generally, the analyst told me that if she saw a hospital significantly increasing the number of patient discharges, she would bet it would be due to increased services, not a sudden change in the community.
One of the most famous examples of survivorship bias comes from the study of damage to WWII bombers by researchers at the Center of Naval Analyses. The original study concluded that additional protective armor should be placed where the returning planes had bullet holes. But that missed the point. Those bullet holes were found on the planes that returned from their runs—it was areas where there were no bullet holes that were downing other planes.
This example of hospital stroke care is similar: the data don’t represent an actual increase in the number of people needing stroke care; they represent better access to stroke care. As the analyst told me, “Data alone are just that—alone. Data need context.” It’s a good lesson for all of us to keep in mind.
PolicyViz at Home
We had a lovely holiday season in December—lots of relaxing at home, watching sports and movies, and a bunch of outdoor time. I was able to meet Washington Capitals’ play-by-play guy Joe Beninati and Caps’ Center Dylan Strome. I also got a behind-the-scenes tour of Washington Nationals’ ballpark and saw my buddies’ band, Nayan, play a show in downtown DC. I also read four(!) books and drank lots of tea and a lovely berry tea-inspired vodka cocktail. I hope you had an equally enjoyable holiday season!
Podcast: Exploring the Unbound Realms of Information Design with Sheila Pontis & Michael Babwahsingh
In the first episode of The PolicyViz Podcast of 2024, I speak with Sheila Pontis and Michael Babwahsingh, the authors of the new book, Information Design Unbound. The book, which contains over 400 images and input from more than 65 contributors, is designed to serve both students and professionals through exercises and activities. It specifically addresses how to teach design and information design to students without a traditional design background and explores different approaches to the subject based on students’ fields of study, as well as various team structures within the profession.
Things I’m Reading & Watching
Books
Indigenous Continent by Pekka Hämäläinen
Art of Insight by Alberto Cairo
Autographic Design by Dietmar Offenhuber
Sex and Gender by Alice Sullivan and Selina Todd
Gender Queer by Maia Kobabe
All Data are Local by Yanni Loukissas
Articles
How to Do Better with Gender on Surveys: A Guide for HCI Researchers by Spiel et al.
Limitations on Social Security Benefits for Black Retirees by Richard Johnson and Karen Smith
Because its power remains naturalized: introducing the settler colonial determinants of health by Wispelwey et al.
The Mortality Experience of Disabled Persons in the 1 United States During the COVID-19 Pandemic by David Weaver
Data Visualizations
2023 in Nine Charts: A Growing Inequality from the World Bank
Dot density map of the population in Europe from Eurostat
The 20 farming families who use more water from the Colorado River than some states from the Desert Sun
TV, Movies, Music, and Miscellaneous
Lessons in Chemistry, Apple TV
LA Confidential, Netflix
Crazy, Stupid, Love, TNT
Beef, Netflix
Note: As an Amazon Associate I earn from qualifying purchases.
Sponsor: Whisper Transcribe
Whisper Transcribe is a transcription game-changer! It turns a three-hour video into accurate, well-punctuated text in just 8 minutes, saving hours of work. Not only does it transcribe audio and video, but the underlying AI model will generate sample summaries, social media content, bullet points, and more. Check it out!
