Safety has been an important problem in healthcare for a long time. The release of “To Err Is Human” by the National Academy of Medicine in 1999 raised awareness in the public eye.1 That report estimated that 44,000 to 98,000 deaths occur annually in the U.S. alone as a result of safety issues in healthcare. There has been considerable controversy around this figure, but other estimates support it and suggest that the problem may be even bigger.2-4
Most healthcare harms in hospitals are accounted for in the following eight domains: healthcare-associated infections, adverse drug events, venous thromboembolism, surgical complications, pressure ulcers, falls, decompensation, and diagnostic errors.5 We elected to focus on these harms in our review. Overall, it is thought that roughly half of harms occur inside hospitals with the other half occurring outside, although this distribution is not certain. However, more serious harms are more likely to happen today in hospitals.
One of the reasons that patient safety has not received more attention is that organizations do not have good approaches for routinely measuring many of these major types of harm. While they do have a good sense of what the numbers are for some of the more readily identifiable issues like healthcare-associated infections and falls, the figures are much less precise for other issues like adverse drug events and diagnostic errors. As a result, this problem has been relatively easy to ignore.
Over the last 20 years, safety organizations have made many advances, especially in hospital settings. For healthcare-associated infections, safety bundles, such as those focused on central venous catheters, have been shown to dramatically reduce the risk of infections. Similar benefits have been achieved for some other problems like ventilator-associated pneumonia.
Artificial intelligence has great potential to improve healthcare.6 Many of the benefits so far have been realized in radiology. However, it is likely that artificial intelligence will be leveraged to address many problems, and safety may be a particularly attractive target. One great advantage of artificial intelligence is that it can be used to bring together many types of disparate data to make predictions, for instance, to estimate the likelihood that something of importance like an adverse event will occur. In the paper we published in npj Digital Medicine, we reviewed the main types of harm, and assessed what has been done so far in using artificial intelligence in any of these harm domains.7 We also considered the incidence and preventability to estimate the potential impact of artificial intelligence for reducing these main types of harm.
We concluded that artificial intelligence is likely to be a powerful tool for improving the safety of healthcare going forward (Figure 1). We suspect that the most mileage early on will be made by focusing on some of the individual types of harm, especially those that have been relatively resistant to improvement using the types of data routinely captured today. New data sources such as continuous sensing and genomic sequencing could significantly improve our ability to predict which patients are at risk of specific types of harm. If these estimates can be made accurately and conveyed to clinicians in ways that they find trustworthy and actionable, there is potential to dramatically improve the safety of care which is being delivered. Today much of the riskiest care is delivered in hospitals, but we expect that in the future, care will increasingly be done in patients’ homes, and technology like the Internet of things will enable interventions which do not seem possible today. Care delivery outside of healthcare institutions represents an exciting new frontier for safety.
Figure 1. Summary of the eight harm domains and key points of the article.
- Kohn, L. T., Corrigan, J. M. & Donaldson, M. S. To Err is Human: Building a Safer Health System (National Academies Press, 2000).
- Shojania, K. G. & Dixon-Woods, M. Estimating deaths due to medical error: the ongoing controversy and why it matters. BMJ Qual. Saf. 26, 423–428 (2017).
- Makary, M. A. & Daniel, M. Medical error—the third leading cause of death in the US. BMJ 353, i2139 (2016).
- James, J. T. A new, evidence-based estimate of patient harms associated with hospital care. J Patient Saf. 9, 122–128 (2013).
- Jha, A. K. et al. The global burden of unsafe medical care: analytic modelling of observational studies. BMJ Qual. Saf. 22, 809–815 (2012).
- Topol, E. J. High-performance medicine: the convergence of human and artificial intelligence. Nat. Med. 25, 44–56 (2019).
- Bates, D. W. et al. The potential of artificial intelligence to improve patient safety: a scoping review. npj Digit. Med. 4, 54 (2021).