The critical step to reduce diagnostic errors in medicine: addressing the limitations of human information processing

Research in the field of diagnostic reasoning and diagnostic error has a long history that started in the 1960s when researchers applied knowledge obtained from cognitive psychology to the diagnostic reasoning process [1, 2]. Interestingly, the diagnostic reasoning process seems to be a rather consistent process across the world and across time. Despite the developments and breakthroughs that have taken place in medicine (e.g., development of better diagnostic tests, accumulation of knowledge about diagnosing certain diseases and improved imaging techniques), the estimates of the incidence rates of diagnostic error (a diagnosis that was unintentionally delayed, wrong or missed) have remained fairly stable over time [3–5]. Furthermore, the same diseases remain the most likely to be associated with diagnostic errors over the years. These are not the extremely rare diseases that are difficult to diagnose, but common diseases such as pulmonary embolism, pneumonia and myocardial infarction [3, 4, 6, 7]. Considering that the environment in which physicians work nowadays involves extensive interaction with the hospital organization and with health information technology (Health IT), one might expect the causes of diagnostic error to be more system-related. However, the causes of diagnostic error are still mainly cognitive [8, 9].

To summarize, diagnosis is faster and more precise thanks to advances in diagnostic testing but diagnostic errors especially in common diseases, still occur as frequently as they did 50 years ago. Why is that?

One of the most crucial tasks of the physician in the diagnostic process is to select the right information and to interpret it correctly. This stage is critical to the differential diagnosis. The differential diagnosis is a strong indicator for the diagnosis that will eventually be established and transformed into a treatment plan. The last Diagnostic Error in Medicine meeting included a session with clinical case presentations. As a non-clinician, it was instructive to see how a master clinician, Dr. Gupreet Dhaliwhal solved one of these cases. What struck me the most was how he consciously assigned value to the information that he received. Symptoms were considered to have less diagnostic value if they were associated with too many possible diseases Other symptoms were given a higher diagnostic value based on the context in which they appeared and because they were more specific. I realized that one quality of being a great diagnostician was being able to discern the most valuable information from all the information available.

In contemporary medicine, selecting and interpreting information correctly may be even more difficult than it was 50 years ago, because physicians have access to so much information: information provided by the patient, information from previous health care providers, information from the patient record, and information available in the medical literature. Further information is obtained from laboratory tests and medical images. In general, having access to all this information can be useful, because it can provide a more complete picture of the case. However, having such an abundance of information also makes the physician’s diagnostic task more complex. First, because the relevant information needed to diagnose the patient has to be extracted and differentiated from the irrelevant information. Second, because the value of information needs to be interpreted (e.g., how sensitive or specific this piece of information is into establishing or ruling out a diagnosis). The more information is available, the more difficult it is to determine which pieces of information are relevant and which are not.

This process of selecting and interpreting information (information processing), is extensively studied in cognitive psychology. It is well known that human information processing is highly limited [10, 11]. This means that we cannot process large amounts of information at the same time. Important elements of information processing such as attention, perception and memory help us to function adequately in the world, but also have their limitations. The main limitation of human information processing is that it is selective. We do not process all information that enters through our senses, but we are able to direct our attention to the information that is relevant to our goals [12]. If you are, for example, searching for a friend with a yellow jacket in the crowd, you are able to actively ignore the people without a yellow jacket and quickly find the person you are looking for. To protect us from missing crucial events, our attention can also be captured by highly salient information such as the lights and sirens of an ambulance passing by [13, 14]. In short, we attend to the information that we think is important as well as to highly salient information that is presented to us. This information is processed on a deeper level than the information that we do not attend to.

These limitations of information processing can lead to errors in everyday life, but also to diagnostic errors. Specifically, in the diagnostic reasoning process this means that not all information that is important to the diagnostic reasoning process will be identified as such, either because:

1. The information is not selected based on our pre-set goals, for example because the information is not considered sufficiently relevant to the initial differential diagnoses or

2. The information is not sufficiently salient to capture our attention.

The selectivity in human information processing contributes importantly to the occurrence of diagnostic error [15]. The way we process information is deeply rooted and not easily changed. Every physician seeing a patient is selective during the diagnostic process. The selectivity is necessary in order to diagnose the patient on a timely basis and without putting him or her through a large number of unnecessary tests, but the same selectivity is an important cause of diagnostic error [15, 16].

Knowledge and experience will help us to better select the relevant information from the irrelevant information. Furthermore, literature has shown that some reasoning strategies have improved the diagnostic reasoning process (e.g., reflective practice) [17]. I agree that it can be beneficial to be critical throughout the diagnostic reasoning process by trying to falsify the differential diagnoses and incorporating moments of reflection in the diagnostic process. However, this reduces the number of diagnostic error only to small extent, because the intrinsic limitations of information processing will remain. Developing additional ways to overcome the limitations of human information processing is an absolute priority if we want to reduce the diagnostic error rate substantially.

The question is: How do we make sure the important information does get processed? How do we determine what information is relevant to diagnose a patient? How can we objectively determine the value of information?

System-related interventions have great potential to improve information processing because they can create an environment which takes the human limitations into account.

In my opinion, health IT will provide the best solution. Computers and their information processing system have strengths and limitations that are opposite of those of human information processing. Computers can process very large amounts of information quickly and can determine cut-off values, calculate base-rates, and integrate information into differential diagnoses based on objective probabilities. I do not think that computers can take over the diagnostic process. Computers can calculate information for a general population of ‘patients like this’ but do not look at the specific characteristics of a patient. This is something that humans can do very well. Therefore, computers and physicians can complement each other in the diagnostic reasoning process.

Health IT is a developing area, especially in the field of diagnostic reasoning [18]. Systems such as Isabel (Isabel Healthcare) and DXplain can provide physicians with information that is valuable to the diagnostic process. However, we have larger steps to take. What is needed are adaptive learning systems that can recognize frequently made errors or often overlooked laboratory results, and can provide physicians with valuable pre-processed information at the right time. I realize that we are still far from having such a system fully incorporated in health care. We need to work on creating an environment that takes the intrinsic limitations of human information processing into account. This is the most promising way to substantially reduce the number of diagnostic errors in medicine.