Missed findings: One symptom of burnout in radiology
How artificial intelligence (AI) can potentially identify high-value findings and help ease the burden for radiologists
In the three minutes it will take you to read this blog post, a radiologist will review an estimated 45-60 images.1 And they must continue to review an image every 3-4 seconds during each eight-hour shift, five days a week, all year long.
This pace is not sustainable, even for the most experienced radiologists working under ideal conditions. As image volume continues to increase, there simply aren’t enough of us radiologists to make this formula work over the long term.
We’re struggling to balance productivity and quality, and it’s contributing to physician burnout. Radiologists are acutely aware of the potential toll it takes on their work. For example, according to one survey in the United Kingdom, one in three consultant clinical radiologists report experiencing work-related stress that negatively affects their work.2 And according to a recent Medscape survey, nearly one in two clinical radiologists in the United States experiences symptoms of burnout, and fewer than 25% of us are happy at work.3
Of the many potential costs of burnout, the one that most often keeps me up at night is the increased risk of medical error and missed findings. We, as a profession, look at more than 84 million studies and 90 billion images each year. With an estimated 3-5% miss rate among all radiologists, the numbers, and potential for harm, is staggering.4
As an example, let’s say a 50-year-old patient shows up in the emergency department with chest and abdomen pain. The attending physician suspects appendicitis and appropriately orders a chest/abdomen CT. The diagnostic radiologist confirms appendicitis, but he misses a centimeter-wide solid nodule in the right lung. One year later, the patient presents with chest pain and physicians discover the lung nodule, which has now doubled in size and there are now enlarged mediastinal lymph nodes.
Reducing the rate of missed findings
Radiologists are only human, and especially with an increasingly demanding workload, (not to mention the frequent interruptions and distractions), mistakes are inevitable. With the estimated daily rate of errors and discrepancies of 3% and 5%,4 if I review 100 cases today, the averages tell us that I could miss four potentially important findings. At the end of the workweek, this adds up to 20 findings — by the end of the year, it could be more than 900. There are at least 30,000 diagnostic radiologists in the United States today. Errors are understandable, but these rates are still uncomfortably high.
Technology can help. AI’s capacities in imaging have been evolving over several years. Today, AI-enabled solutions have been trained on vast data sets to detect anomalies through organ-specific analytics. This capability can help radiologists identify high-value findings that could otherwise have been missed.
People working with AI technology can outperform either one working alone. Humans excel at applying common sense, solving dilemmas, and giving compassionate care, while AI systems excel at identifying patterns, locating knowledge and providing endless capacity. I believe AI systems hold the key to identifying missed findings, to enabling providers to work more effectively and joyfully, and to ensuring that our patients and clinical colleagues are comfortable that they are receiving safe, accurate and high-quality care. Watson Health is committed to building AI to help identify potential missed findings.
- Based on and estimated blog reading time of 3 minutes and a study that estimates that radiologists interpret one image every 3-4 seconds: The effects of changes in utilization and technological advancements of cross-sectional imaging on radiologist workload. McDonald et al Acad Radiol. 2015 Sep; 22(9): 1191–1198. Published online 2015 Jul 22. doi: 10.1016/j.acra.2015.05.007 https://www.ncbi.nlm.nih.gov/pubmed/26210525
- Clinical Radiology UK Workforce Census 2018 Report
- Error and discrepancy in radiology: inevitable or avoidable? https://link.springer.com/article/10.1007/s13244-016-0534-1