2014. I am an epidemiology and statistics researcher at the INSERM at Paule de Viguier Hospital. Xavier Alacoque, an anesthesiologist specializing in cardiac surgery, enters my office: - "Christophe, I have collected a fabulous dataset, we will write great articles, I need your help with the statistics." I sigh and point to a stack of files on a corner of my desk. - "Dear Xavier, here is a ton of "fabulous" data collection that I already have to analyze." To be noted, I help my colleagues with their statistics, but my real job is being a researcher in statistics, specializing in the clustering of longitudinal data. Hence the idea of creating statistical analysis software that is simple to use is born.
The scenario I have just described is a classic: in almost every hospital, there is a severe lack of statisticians. Doctors have ideas, they design protocols, collect data thinking that the essentials are done, and they then find themselves stuck when it comes to analyzing the data. Bio-statisticians' are overwhelmed.
First option, befriend a statistician that can help you. That might not always be easy. Also, statistical analysis involves a lot of trial and error: doctors ask me if there is a correlation between the level of education and the father's salary. My answer: no. "And with the mother's salary?" Neither. From there, we will try several features: the average between the two salaries, the level of education, the size of the house ... In the end, the mother's level of education is the best predictor. To find this, I spent half a day, with the doctor next to me saying "what if we tried this feature instead?"
Alternative solution (I want to say "the last resort"): get by with a little Excel, a little R and a little SPSS. In short, it's a hassle. Yet the analyses needed are simple: data cleansing, a few statistical tests, occasionally a ROC curve, a regression, or a little Survival curve. In theory, this shouldn't be complicated. What doctors need is more autonomy. Adapted software. Easy to learn. Easy to use. Easy to re-learn after 6 months without use.
The solution comes from research called HCI, Human-Computer Interaction. The principle is as follows:
During a first brainstorming, doctors present their complaints. They have to answer the question: "What is really difficult with your current software? What takes you a full day when initially you thought it would take an hour? What is boring, delicate, time-consuming, risky, error-prone? ». We note everything.
Then, the group decides to work on a specific issue. In our case, the detection of outliers. Small teams are created, and they deliberate independently. The goal is to come up with an interface that would tackle the issue. Users do not care about any technical details and only focus on what would be really simple and user-friendly for them.
For example, a person suggests "If my mouse is hovering a column, it brings up a graphical representation of this column". Another person says: "I would like a button that displays the statistical summary of each column." In the end, we combine the two ideas, and we get "a button that brings up a graph for each of the columns".
When the interfaces are ready "in the mind", they are transformed into small films. Using papers, post-its, markers, scissors, we draw the interface "of our dreams" and film it. Then, we present it to the rest of the group, to get their feedback: does the solution seem interesting to them? Friendly? Does it solve the original problem? Better yet, these mini-films often raise new ideas that will allow you to go even further, towards a more efficient interface.
In the end, in 3 hours, we managed to:
1- Identify the real problems of users,
2- Find and draft solutions,
3- Get feedback from the group.
To conceive R++, we conducted about twenty video prototyping sessions with a hundred doctors. We identified all their constraints, and we gradually devised solutions. The result is R++: an ergonomic statistical analysis software, designed by and for doctors.
Hard to believe? Want to see for yourself? Or are you just curious to see how our interface can revolutionize statistics?
Our team is committed to contact you within 24 hours.
