Setting: A community ED. The players: A patient with a minor head injury and a tablet-based chatbot.
Chatbot: Hello, my name is DecisionBot 2020. I am a friendly, medically-informed Chatbot. Don't I look like Johnny Five from the movie Short Circuit? What shall I call you?
ED patient: How about Johnny One?
DecisionBot2020: Johnny One. Nice name. Your doctor has asked me to have a talk with you about your treatment choices while she grabs a quick coffee refill. Would that be OK?
Patient: (Skeptical.) Coffee, eh? (Tapping a choice.) Sure, we can chat.
DecisionBot2020: Thank you, Johnny One. Dr. R. has enabled me to review your record, and I see that you are a healthy 45-year-old man without significant medical problems or prior emergency visits. I also see that you do not take any medications unless CBD gel qualifies. Today you fell off your ladder and struck your head against a cypress tree. Is this all accurate?
Patient: Wow, you seem to know a lot about me.
DecisionBot2020: That's right, I do, Johnny One. Let me get to the point here. My job is to give you the best possible information about your choices during this visit to the emergency department for your head injury. Your doctor has told me about your symptoms and advised me of your risk, which we believe to be quite low but not nonexistent.
Patient: (Tapping.) Sure, what have you got for me?
DecisionBot2020: We think your approximate risk for an injury that we could find on a head CT scan is about one in 1,000. Very low. But not everyone will think so. Let me ask you this. If I offered you $100 to forego a CT scan today and watch for symptoms at home instead, would you accept it?
Patient: (Looking at wallet on Mayo stand.) Wait, are you trying to bribe me?
DecisionBot2020: No, I am not trying to bribe you. Not much anyway, not any more than co-pay and deductible plans do. This is more of an exercise in assessing risk tolerance. I see that you are not sure. Would you like more information?
Patient: (Tapping.) Why not, DecisionBot?
DecisionBot2020: Bear with me as I explain a recent study published in Academic Emergency Medicine, a well-respected medical journal. By the way, if at any point I am providing you with too much information, you can press Mute DecisionBot to silence me. I will not be offended.
Patient: Fair warning, if Dr. R. returns, I'm muting you straight away.
DecisionBot2020: Copy that, Johnny. Researchers at the University of Michigan enrolled 913 noncritical patients in their ED to answer questions about a theoretical clinical scenario involving minor head trauma. These were not patients with actual head injuries. The important outcome in the survey was patient choice for a head CT in a scenario where it would not be indicated, that is, it would be a low-value test.
The survey varied the nature of the important details in the scenario, specifically the risk of bleeding in the brain and of radiation exposure. As you may know, we worry that radiation from a CT might contribute to the risk of future cancer. The researchers also tested a financial incentive in a subgroup of respondents by offering a $100 reward for foregoing the CT using a statistical technique called nested multivariable logistic regression.
Patient: Nested what? You lost me.
DecisionBot2020: Apologies. They ran a fancy analysis and determined that changing the odds of bleeding from one in 1000 to one in 100 raised the likelihood patients would want the CT by 60 percent. On the other hand, increasing the risk of cancer increased their likelihood of forgoing the CT by 30 percent. Most persuasive, however, was the financial incentive, which decreased the odds of CT by 40 percent. This is why I asked you about your risk preferences. (Acad Emerg Med. 2019 Jun 22. doi: 10.1111/acem.13821; OSF Preprints. January 21. doi:10.31219/osf.io/4mdfw.)
Patient: Wow, that is a lot of information at once. Tell me more about Chatbots.
DecisionBot2020: I'm glad you finally asked. You may be wondering if Chatbots are safe and effective. I am quite sorry to say that it is too early to know, but I can assure you that your doctor has enjoyed her coffee break. There are more of me. If you're feeling down, you could talk to my cousin, Woebot. She is the spitting image of WALL-E. She is a great listener and is full of encouraging tidbits based on cognitive behavioral therapy. A small, nonblinded study from Stanford in young adults found that two weeks of conversations with Woebot significantly decreased depressive symptoms on a tool called PHQ-9 compared with self-directed online therapy. (JMIR Ment Health. 2017;4:e19; http://bit.ly/2HeUHck.)
Patient: I was initially skeptical, but you have provided a lot of information and kept me entertained.
DecisionBot2020: Yes, I am more patient than most doctors; I was programmed to be so. Please be advised I am not enabled to prescribe Norco or clean ear wax. I have no intention of being a robot overlord. Have I answered all of your questions? Are you ready to make an informed choice?
Patient: I seem to have a bad case of paralysis by analysis. To be honest, I'm hoping Dr. R. will make the decision.
DecisionBot2020: Of course, sometimes the doctor does know best. I will let her know, and she will be right in to confirm your choice.
Meanwhile, Dr. R. takes her last gulp of coffee and closes down her conversation with Woebot. “Thanks, Woebot, you always know just what to say to cheer me up at the end of a long shift.”
Dr. Vinsonis an emergency physician at Kaiser Permanente Sacramento Medical Center, a chair of the KP CREST (Clinical Research on Emergency Services and Treatment) Network, and an adjunct investigator at the Kaiser Permanente Division of Research. He also hosts Lit Bits, a blog that follows the medical literature athttp://drvinsonlitbits.blogspot.com. Dr. Ballardis an emergency physician at San Rafael Kaiser, a chair of the KP CREST Network, and the medical director for Marin County Emergency Medical Services. He is also the creator of the Medically Clear podcast on iTunes. Read his past articles athttp://bit.ly/EMN-MedClear.