User login
In a previous Letter from Maine I wrote about a study performed in China in which more than half a million patients were diagnosed by an artificial intelligence (A.I.) system that was able to extract and analyze information from their electronic medical records. The system was at least as accurate as physicians who had access to the same data (“A.I. Shows Promise Assisting Physicians,” by Cade Metz, The New York Times, Feb. 11, 2019). I ended my column with the hopeful assumption that despite incredible advances in A.I., the practice of medicine always would include a human element. However, I left unexplained exactly how physicians would fit into the post-A.I. revolution. In the weeks since I submitted that column, I have been searching for roles that might remain for physicians after A.I. has snatched their bread and butter of diagnosis and management.
I easily can envision a system in which the patient enters her chief complaint and current symptoms into her smartphone or tablet. Using its database of the patient’s past, family, and social history, the system generates a list of laboratory and imaging studies, some of which the patient may be able to submit directly from her handheld device. For example, the system may be able to use the patient’s phone to “examine” her. The A.I. system then generates a diagnosis.
If the diagnosed condition and management is simple and straightforward, such as a rash, the information could be communicated to the patient directly, with a short paragraph of explanation and list of persistent symptoms that would indicate that the condition was not improving as expected. A contact dermatitis comes to mind here.
However, suppose the A.I. system determines that the patient has a 90% chance of having stage IV pancreatic cancer, with a life expectancy of 6 months. Is this the kind of information you would like to learn about yourself by clicking “Your Diagnosis” box on your phone while you were having lunch with a friend? Obviously, a diagnosis of this severity should be communicated human to human, even though it was generated by a highly accurate computer system. And this communication would best be done in the form of a dialogue with someone who knows the patient and has some understanding of how she might understand and cope with the information. In the absence of a prior relationship, the dialogue should occur in real time and face to face at a minimum. I guess we have to acknowledge that FaceTime or Skype might be acceptable here.
Fortunately, stage IV cancers are rare, but there are a bazillion other conditions that, while not serious, require a nuanced explanation as part of a successful management plan that takes into account the patient’s level of anxiety and cognitive abilities. A boilerplate paragraph or two spit out by an A.I. system isn’t good health care. Although I know many physicians do rely on printed handouts for conditions they feel is a no-brainer.
The bottom line is that even when a machine may be better than we are at making some diagnoses, there always will be a role for a human to help other humans understand and cope with those diagnoses. At this point, physicians would appear be the obvious choice to fill that role. How we will get reimbursed for our communication skills is unclear.
Dr. Wilkoff practiced primary care pediatrics in Brunswick, Maine for nearly 40 years. He has authored several books on behavioral pediatrics, including “How to Say No to Your Toddler.” Email him at [email protected].
In a previous Letter from Maine I wrote about a study performed in China in which more than half a million patients were diagnosed by an artificial intelligence (A.I.) system that was able to extract and analyze information from their electronic medical records. The system was at least as accurate as physicians who had access to the same data (“A.I. Shows Promise Assisting Physicians,” by Cade Metz, The New York Times, Feb. 11, 2019). I ended my column with the hopeful assumption that despite incredible advances in A.I., the practice of medicine always would include a human element. However, I left unexplained exactly how physicians would fit into the post-A.I. revolution. In the weeks since I submitted that column, I have been searching for roles that might remain for physicians after A.I. has snatched their bread and butter of diagnosis and management.
I easily can envision a system in which the patient enters her chief complaint and current symptoms into her smartphone or tablet. Using its database of the patient’s past, family, and social history, the system generates a list of laboratory and imaging studies, some of which the patient may be able to submit directly from her handheld device. For example, the system may be able to use the patient’s phone to “examine” her. The A.I. system then generates a diagnosis.
If the diagnosed condition and management is simple and straightforward, such as a rash, the information could be communicated to the patient directly, with a short paragraph of explanation and list of persistent symptoms that would indicate that the condition was not improving as expected. A contact dermatitis comes to mind here.
However, suppose the A.I. system determines that the patient has a 90% chance of having stage IV pancreatic cancer, with a life expectancy of 6 months. Is this the kind of information you would like to learn about yourself by clicking “Your Diagnosis” box on your phone while you were having lunch with a friend? Obviously, a diagnosis of this severity should be communicated human to human, even though it was generated by a highly accurate computer system. And this communication would best be done in the form of a dialogue with someone who knows the patient and has some understanding of how she might understand and cope with the information. In the absence of a prior relationship, the dialogue should occur in real time and face to face at a minimum. I guess we have to acknowledge that FaceTime or Skype might be acceptable here.
Fortunately, stage IV cancers are rare, but there are a bazillion other conditions that, while not serious, require a nuanced explanation as part of a successful management plan that takes into account the patient’s level of anxiety and cognitive abilities. A boilerplate paragraph or two spit out by an A.I. system isn’t good health care. Although I know many physicians do rely on printed handouts for conditions they feel is a no-brainer.
The bottom line is that even when a machine may be better than we are at making some diagnoses, there always will be a role for a human to help other humans understand and cope with those diagnoses. At this point, physicians would appear be the obvious choice to fill that role. How we will get reimbursed for our communication skills is unclear.
Dr. Wilkoff practiced primary care pediatrics in Brunswick, Maine for nearly 40 years. He has authored several books on behavioral pediatrics, including “How to Say No to Your Toddler.” Email him at [email protected].
In a previous Letter from Maine I wrote about a study performed in China in which more than half a million patients were diagnosed by an artificial intelligence (A.I.) system that was able to extract and analyze information from their electronic medical records. The system was at least as accurate as physicians who had access to the same data (“A.I. Shows Promise Assisting Physicians,” by Cade Metz, The New York Times, Feb. 11, 2019). I ended my column with the hopeful assumption that despite incredible advances in A.I., the practice of medicine always would include a human element. However, I left unexplained exactly how physicians would fit into the post-A.I. revolution. In the weeks since I submitted that column, I have been searching for roles that might remain for physicians after A.I. has snatched their bread and butter of diagnosis and management.
I easily can envision a system in which the patient enters her chief complaint and current symptoms into her smartphone or tablet. Using its database of the patient’s past, family, and social history, the system generates a list of laboratory and imaging studies, some of which the patient may be able to submit directly from her handheld device. For example, the system may be able to use the patient’s phone to “examine” her. The A.I. system then generates a diagnosis.
If the diagnosed condition and management is simple and straightforward, such as a rash, the information could be communicated to the patient directly, with a short paragraph of explanation and list of persistent symptoms that would indicate that the condition was not improving as expected. A contact dermatitis comes to mind here.
However, suppose the A.I. system determines that the patient has a 90% chance of having stage IV pancreatic cancer, with a life expectancy of 6 months. Is this the kind of information you would like to learn about yourself by clicking “Your Diagnosis” box on your phone while you were having lunch with a friend? Obviously, a diagnosis of this severity should be communicated human to human, even though it was generated by a highly accurate computer system. And this communication would best be done in the form of a dialogue with someone who knows the patient and has some understanding of how she might understand and cope with the information. In the absence of a prior relationship, the dialogue should occur in real time and face to face at a minimum. I guess we have to acknowledge that FaceTime or Skype might be acceptable here.
Fortunately, stage IV cancers are rare, but there are a bazillion other conditions that, while not serious, require a nuanced explanation as part of a successful management plan that takes into account the patient’s level of anxiety and cognitive abilities. A boilerplate paragraph or two spit out by an A.I. system isn’t good health care. Although I know many physicians do rely on printed handouts for conditions they feel is a no-brainer.
The bottom line is that even when a machine may be better than we are at making some diagnoses, there always will be a role for a human to help other humans understand and cope with those diagnoses. At this point, physicians would appear be the obvious choice to fill that role. How we will get reimbursed for our communication skills is unclear.
Dr. Wilkoff practiced primary care pediatrics in Brunswick, Maine for nearly 40 years. He has authored several books on behavioral pediatrics, including “How to Say No to Your Toddler.” Email him at [email protected].