An Adaptive Tutor for Improving Visual Diagnosis
Abstract
Rationale and Objective: The diagnostic process continues to be an area with considerable potential for improvement. The Institute of Medicine, in its 2015 report "Improving Diagnosis in Healthcare," called for efforts to "enhance healthcare professional education and training in the diagnostic process." An important component is ensuring that healthcare professionals have and maintain the competence needed for effective diagnostic performance. This problem with diagnostic errors is compounded in the military context, where deployment requires that clinicians who have grown comfortable with one spectrum of patients, apply what may be rusty skills to a new type of patient. In this proposal, we tackle the issue of visual diagnosis in which a clinician makes a diagnosis by interpreting an image such as an x-ray, photo or electrocardiogram (ECG). The clinician typically looks at the image and declares it either normal (free of disease) or abnormal, in which case s/he describes the possible diagnoses. Using ECGs as an example, we propose to look closely at how clinicians can best maintain their visual diagnosis skill. The ECG allows the diagnosis of life-threatening heart conditions such as myocardial infarction and dysrhythmias. ECG findings are key components of deciding who gets rushed to cardiac catheterization or requires cardioversion. Fast, accurate diagnosis can be life-saving. However, our methods for learning and maintaining this skill do not take advantage of the tremendous advances in learning technologies. We propose to create an online adaptive ECG tutor that is based on the electronic records of thousands of patients. We would collect authentic examples of ECGs in the exact proportions that they are seen in real life. Using natural language processing systems like those used in cell-phone voice interpretation, we would extract the diagnosis of each ECG along with the outcome of each patient. This training set of over 20,000 emergency department cases would then be calibrated on a number of factors such the case difficulty, case complexity, the exact nature of the diagnosis and whether errors had been documented. In essence, we would have 20,000 diagnostic dilemmas and 20,000 answers. We know we cannot simply train clinicians on all of the cases. The key product of this research is to develop a method of presenting cases in the exact quantity and order that ensures efficient and effective learning. We aim to answer the question of how many cases of what diagnosis and difficulty need to be completed before we are assured that the clinician s skill is sufficient and optimized. This requires using statistical methods to predict the difficulty of each case and how well it matches the current ability of the clinician. Our objective however is not simply to create a single tutor for one type of ECG. Instead, we will create a method of learning and assessing visual diagnosis to allow military health professionals to rapidly create new tutors also based on large collections of images in other clinical settings. What Types of Patients Will This Help and How Will It Help? This is a demonstration project that will directly benefit patients with cardiac illness that requires ECG diagnosis. In particular, patients with heart attacks due to myocardial infarction or rhythm problems could benefit by having fewer errors made in the timely diagnosis of their conditions. However, the "recipe" for developing these types of online learning tutors for visual diagnosis should work for many other types of clinical images that can be downloaded from the electronic health record. Our emphasis is not only on efficient training with these tools but also on the efficient building of the tools themselves. What Are the Potential Clinical Applications, Benefits and Risks? We envisage a day when mobilized clinicians perform simulated clinical diagnoses on cases that are directly relevant to their deployment. The
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 31, 2017
- Source ID
- W81XWH1610797
Entities
People
- Martin Pusic
Organizations
- Grossman School of Medicine
- United States Army