AI is changing pediatric surgery, but with strong ethical concerns

Before a child goes into the operating room, a large screen displays the risk score. This score predicts potential complications, provides an estimated time to recovery, and recommends a course of action. While the numbers seem accurate, the process that goes into creating them is hard to see.

Artificial intelligence (AI) is rapidly advancing pediatric surgery with greater accuracy in diagnosis, greater surgical planning capabilities, and less paperwork for the clinician. However, the same technology that has enabled doctors to simplify care has also introduced a new source of uncertainty for parents and caregivers when making decisions about children who cannot speak for themselves.

The benefits of AI in the operating room (OR) are obvious. For example, machine learning programs have been developed that allow the identification of surgical risks and complications. Other machine learning systems can interpret imaging studies and predict the likelihood of complications after a complex surgical procedure. There are also applications that are programmed to listen during patient visits and create clinical notes for the surgeon, enabling surgeons to focus on the family instead of the computer screen.

In practice, AI technologies are already having an impact on the way surgical care is provided. The American College of Surgeons’ National Surgical Quality Improvement Program (NSQIP), introduced in 2016, provided surgeons with risk estimates based on specific patient characteristics using traditional statistical models. Over the past few years, the model has evolved into a machine learning program. This now provides the surgeon with a better ability to capture the interactive nature of variables related to risk and outcome.

As a result of the introduction of machine learning technology, difficult conversations with families about risk and uncertainty have become easier. Families evaluating treatment options for conditions such as biliary atresia or neuroblastoma may benefit from accurate predictions. This allows them to weigh their choices fairly. For surgeons, automated documentation and summary tools reduce time spent reviewing charts. This activity has been shown to increase burn rates.

Pediatrics offers a different patient demographic than adult medicine. Children constitute a small, highly diverse group of patients, and their data are often not represented in large databases. Consequently, gaps in this information can affect the predictability of technology.

The ethical issue of autonomy begins with the idea that families should make informed decisions about their child’s medical care. In pediatric surgery, parents or guardians are responsible for making these decisions, often in very stressful situations. To help alleviate some of the stress or anxiety during consent discussions, AI tools may provide useful insights. These tools can translate complex medical language into layman’s terms.

Most ideas are about raising awareness. Some include creating a system that recognizes signs of distress or levels of family involvement. These systems can then automatically notify doctors when families need additional help. However, the authors caution against replacing traditional communication methods with the use of technology.

Families should understand the potential impact of AI technology on diagnosis and treatment for their children. They should also say how their data will be used to train future AI technologies. Not agreeing to future AI technology training should not create barriers to family care.

The increasingly popular use of surgical robots creates another layer of complications. Many current surgical technology systems are developing capabilities that allow them to operate in a partially autonomous capacity. Li and colleagues (2018) describe a classification system for how robotic systems can operate with different degrees of autonomy, from 1 to 6, depending on how much human support is required.

Until now, most robotic systems used in surgery are supervised by humans. This means that surgeons are responsible for all aspects of surgery. If fully autonomous devices are approved for use, it will raise questions about who is responsible if something goes wrong.

AI provides benefits in promoting positive outcomes and preventing negative outcomes in healthcare. Tools already exist that allow AI to quickly identify diagnoses. For example, deep learning algorithms developed for use during surgery could help surgeons diagnose Hirschsprung’s disease faster than current practices. Faster diagnosis allows for shorter operating room time, which benefits the patient.

However, the potential for misdiagnosis due to over-reliance on AI-generated results may put patients at greater risk. These risks include undergoing unnecessary surgery or receiving incomplete treatment. In the case of Hirschsprung’s disease, the wrong amount of bowel may be surgically removed. This has potential long-term consequences.

With the advent of AI, the concept of responsibility or accountability becomes less simple. If an AI-assisted robotic device plays a role in a patient complication, the potential for liability may extend beyond the surgeon to include the hospital and manufacturer. The authors suggest that ways to address the harm caused by these technologies must be defined before widespread adoption.

Another consideration is the principle of justice, which calls for fairness. For pediatric patients, fairness depends on having access to adequate information. Currently, information related to the pediatric population is underrepresented in most datasets used to develop medical imaging products. This makes it less likely that the algorithm will perform well for pediatric patients. In the case of appendicitis, this can lead to a delay in diagnosis and an increase in complications.

Another area of ​​concern about AI and equity is geographic change. Most machine learning algorithms are trained using data from only a limited number of states. This reduces their relevance in many geographic locations. Collecting representative samples from underrepresented groups or modifying the model output to better reflect geographic distribution have been explored as possible solutions. However, each solution has its drawbacks.

Expanding datasets by including previously untapped sources creates new challenges regarding data privacy. There has been an increase in cybercrime and data breaches associated with larger and more centralized datasets. This may put pediatric patients at greater risk due to long-term follow-up of their medical records.

One way to create accountability for AI systems is to improve their ability to provide transparency into their internal decision-making processes. AI systems that were previously seen as a “black box” present a challenge to healthcare professionals. Clinicians can see both the input and output of the system, but not the reasoning behind these recommendations.

Lack of transparency makes it difficult for healthcare workers to explain AI-driven decisions to patients and their caregivers. It also makes it difficult to identify and fix potential errors. Interpretable AI may address this issue by improving transparency in decision-making.

However, the authors worry that over-reliance on automated recommendations could lead to a decline in clinicians’ critical thinking skills. Accessibility to AI-assisted healthcare is also a challenge. This is due to the increasing use of digital healthcare tools and limited access to reliable internet and digital literacy.

The continued development of AI in healthcare may exacerbate existing disparities. This is especially true for people who do not have access to Internet-based devices or sufficient digital literacy.

Ultimately, the authors conclude that AI-assisted tools in pediatric surgical practice should support human decision-making rather than replace it. Surgeons should be actively engaged in the development and evaluation of these technologies.

Pediatric surgery is at a critical juncture as AI capabilities rapidly evolve. At the same time, systems for regulation and supervision are still developing.