A recent study published in Pediatrics reveals risk factors for self-harm among children and adolescents, identifying four risk profiles based on psychiatric comorbidity, age, and gender. The research, conducted between April 2016 and March 2020, examined 1,098 children aged 5-18 hospitalized for neuropsychiatric events, with 37% diagnosed with self-harm. The high-risk profiles included males aged 10-13 with ADHD, autism, and other disorders, and females aged 14-17 with depression, anxiety, and substance- and trauma-related disorders. The low-risk profile consisted of younger children (5-9 years) without mental health diagnoses, while the moderate risk profile lacked depressive disorders. Lead author Mert Sekmen noted that most children who die by suicide have at least two psychiatric diagnoses. Further research is required to validate these risk profiles and to develop decision support applications for real-time risk assessment.
Senior author James Antoon, MD, PhD, MPH, assistant professor of Pediatrics and Hospital Medicine at Monroe Carell Jr. Children’s Hospital at Vanderbilt University Medical Center, set aside some time to discuss the study with SCINQ. (This interview has been edited for length and clarity.)
What prompted this study?
It’s no secret that we’re facing a national mental health crisis among children and adolescents in the United States, which has been ongoing for several years. This crisis has prompted various national organizations, including the American Academy of Pediatrics, to declare national emergencies regarding mental health issues in children.
As a Pediatric Hospitalist, I often see children admitted to medical facilities, such as Vanderbilt Children’s Hospital, while they await placement in psychiatric facilities. The reality is that there’s a national shortage of available beds in these facilities for both adults and children, particularly children. Consequently, children often stay in a hospital for days, even weeks, waiting for a bed to open up in a psychiatric facility.
Apart from the mental health crisis unfolding in outpatient settings, we’re witnessing an escalating burden within hospitals. Many children end up waiting so long that they’re discharged without even reaching a psychiatric facility.
One of the driving factors for this study was the need to better identify children at high risk for self-harm events. Who needs to be prioritized for hospital admission? Who is safe to go home with outpatient services? Can we alleviate some of this burden on the healthcare system? Predicting which children will experience a self-harm event and which won’t is an incredibly difficult task.
Before delving into the study’s uniqueness, it’s essential to define what we mean by a “self-harm event”. This refers to instances when a child intentionally or uncontrollably attempts to harm themselves or succeeds in doing so. This can range from serious attempts at suicide and cutting behaviors to a child with severe autism who engages in biting and head banging, which can be managed at home.
These events are categorized under self-harm, which is one of the most common and serious reasons a child is hospitalized for a mental health condition—particularly when it involves a deliberate attempt to self-harm.
Previous studies have focused on examining the independent effect of certain exposures or conditions and their relation to self-harm outcomes. For example, they might look at a child with bipolar disorder and evaluate the degree of association with self-harm, aiming to categorize the child’s risk level.
We know that 90% of children who experience serious self-harm, leading to hospitalization, have at least one psychiatric diagnosis. However, over 70% have more than two. These conditions aren’t independent of each other.
For instance, a child with bipolar disorder, severe type 1 diabetes, cancer, depression, and anxiety may present differently compared to a child with just anxiety and depression or just cancer and anxiety.
These conditions can interact, either exacerbating or mitigating each other, and we don’t yet fully understand how these complex factors influence predicting self-harm events.
Our study aimed to take a different approach. Instead of examining these factors independently, we considered the entire profile of each child: their psychiatric and non-psychiatric conditions, age, and sex. We sought to understand what a high-risk child looks like on average, compared to a low-risk one. Rather than assuming one condition increases risk, we aimed to see if a combination of conditions, when present together, could alter the risk profile. For example, we understand that a five or six-year-old with a trauma-related disorder is likely different from a 14-year-old female with depression and a trauma-related disorder.
You define what self harm is in broad terms. What were your parameters for this study? What was your definition of self-harm for the study?
We previously conducted a study aimed at validating the identification of neuropsychiatric events in hospitalized children. This study involved reviewing the cases of over 1,000 children.
In this process, multiple physicians conducted a detailed chart review to determine if the presence of certain diagnosis codes and other factors could reliably identify these neuropsychiatric events. Our aim was to see if this method was as effective as a physician going through the chart and confirming that an event had indeed occurred.
The result was the development of an algorithm that could accurately identify children who had experienced these events. We used the physician-confirmed cases from this initial study to conduct the present one.
In this current study, we examined over 1,100 children whose charts had been thoroughly reviewed by a physician. The objective was to determine if an event was present at the time of admission and directly related to the hospitalization. Each case was validated by a physician, ensuring a confirmed outcome.
Your research identified patterns in mental health issues among kids and teenagers. Can you elaborate on these patterns and how they can possibly influence the way doctors treat their patients?
Yes, it’s a bit complicated and challenging to summarize for the layperson due to the complexity of our methodology. Rather than isolating factors from each other, we included both the presence and absence of certain conditions in our analysis. For instance, we determined that a child aged 10-13 years old with anxiety and depression, but without an underlying medical condition or a specific other disorder, constituted a particular risk profile.
The advantage of this approach was that we considered not only the presence of factors but also their absence. We found that not having a certain condition could be significant.
This led to complex risk profiles, combining the presence and absence of multiple factors. These profiles fell into four risk categories, ranging from about 5% to 75% risk of experiencing a serious self-harm event.
Given the wide range, and considering that we started from a population of children with neuropsychiatric events, our findings probably overestimate the risk in the general population. The low-risk group, especially among otherwise healthy children with no conditions, is likely even lower.
We found some interesting trends related to age and sex that align with existing research. Trauma-related disorders were particularly important in our high-risk group. However, as you might have seen in the paper, multiple factors came into play. Physicians typically take all these factors into account when evaluating a patient, including the absence of certain conditions.
We envision implementing our findings through clinical decision support tools integrated into electronic medical health records. Such tools could be programmed to flag a patient if they have a particular set of conditions or a certain medical disorder alongside specific psychiatric diagnoses. This could prompt a warning to physicians that a child is at particularly high risk, even indicating a 75% chance of a serious self-harm event in the upcoming days. Conversely, it could suggest that a child is low risk and might not need to be hospitalized while awaiting a psychiatric evaluation.
In this way, we could operationalize this somewhat complex profile in electronic health records. Future studies might further refine and simplify it, making it more manageable at the individual level.
You employed a technique called Bayesian profile regression. Can you explain what this method is and why it was the right choice for this study?
This study represents one of the first clinical applications of Bayesian profile analysis, a relatively new statistical method that refines upon latent class analysis. To give you some context, many studies employ regression, which involves identifying an exposure and then accounting for multiple factors that might be associated with your outcome.
For example, if the exposure is anxiety, we know that depression is often associated with both anxiety and serious self-harm events. Many studies would use regression to remove the impact of depression, isolating the contribution of anxiety to the outcome.
However, Bayesian profile analysis, which is a branch of latent class analysis, takes a different approach. It looks at what groups or patterns of all exposures or underlying diagnoses coincide with the outcome. Rather than isolating factors, it identifies which combinations have a higher or lower incidence of the outcome.
One of the key advantages of Bayesian profile analysis over similar methods is that it doesn’t require preset categories or profiles. While other analyses may require you to specify a certain number of profiles (like five, seven, or three), Bayesian profile analysis identifies which groups have the strongest association with the outcome and creates the appropriate number of profiles based on that association. Thus, it avoids arbitrarily assigning the number of groups and instead identifies those most significantly related to the outcome.
Another unique aspect of Bayesian profile analysis is its ability to incorporate the absence of a condition into the analysis. Typically, studies focus on the presence of a condition and its significance. However, in our analysis, the absence of a condition was also considered important.
As a final note, I should mention that although I am not a statistician, we were fortunate to have an excellent statistician working with us on this study.
You discovered four distinct profiles. Can you describe these groups and what makes them more likely to harm themselves?
One profile we identified is a low-risk group comprised of younger children aged five to nine who have developmental delays or neurodevelopmental disorders, such as autism, or neurocognitive disorders, along with non-mental health diagnoses or chronic conditions.
This group was found to be at even lower risk than our baseline group, which itself contained common conditions like ADHD, anxiety, autism, and bipolar disorder. The baseline group was predominantly male, whereas in the lower risk group, gender didn’t appear to significantly influence the risk.
We also identified high-risk and very high-risk groups. These two profiles were similar, except for age differences. The very high-risk group was aged 10 to 13, while the high-risk group was 14 to 17. This age-based trend aligns with observations over the past decade, which have shown a significant shift in the high-risk category towards the 10 to 13 age group, compared to older adolescents. This group of younger adolescents, predominantly males, is experiencing increasing rates of hospitalization and suicide attempts.
The 14 to 17 high-risk group was predominantly female. It’s an interesting finding because, typically, females attempt self-harm more frequently than males, but males tend to be more successful in their attempts, leading to a higher mortality rate. What our study underscores is that not only is the younger male population at a very high risk, but there is also an older female population at substantial risk.
Older teenage girls have a mix of anxiety, depression, and substance abuse. Sometimes trauma disorders are the highest for school right. What can you extrapolate about the relationship between their issues and risk and their risk of self harm?
The high-risk group had a risk level of about 70%, while the highest risk group was closer to 80%. While this may not seem like a massive difference, it’s clinically meaningful and statistically significant. This highlights how age and gender interact with underlying conditions in meaningful ways.
We understand that both groups – children in the high-risk and highest risk categories – warrant concern for potential incidents. However, there are also several factors we couldn’t account for in this study that are undeniably important, many of which pertain to family and social support. These elements interact with the severity of the underlying conditions and can trigger or exacerbate these events.
While we have defined these risk profiles, the role of underlying family support, social support, and triggering events are likely different for each group. These profiles overlap to some extent, but there are also probably unique factors at play within each group that we don’t fully comprehend yet.
To answer your question more directly: why does a specific profile place an individual into a higher risk category? We do not entirely understand the interplay of these factors yet. And why are younger boys of a certain profile experiencing these events more often than older girls of a different profile, even when both are classified as high-risk? That’s something we don’t have a clear answer to at this point. Does this help clarify things for you?
How would you apply these findings health checkups and diagnosis?
I believe this analysis is most beneficial for emergency department physicians and potentially outpatient providers. It helps in identifying children at high risk of experiencing these events and enables us to intervene before an event occurs. This intervention could involve providing these at-risk children with resources and setting them up with psychiatric counseling and support. Alternatively, it could involve placing them in a safe environment where they can receive treatment, such as a children’s hospital or psychiatric hospital.
The primary goal here is early identification of children who are likely to experience these events, and then intervening before these events occur. As a hospitalist, by the time a child reaches me in the hospital, these events have usually already occurred. The child may have already attempted or threatened self-harm, or experienced something else that necessitates hospitalization.
But what if we could prevent these children from reaching that point? Could we reduce the frequency of exacerbations, triggering events, or prevent these triggers from escalating into more serious incidents before they need hospitalization? I believe the most effective implementation of our findings would involve informing providers about these patterns. Moreover, a clinical decision support tool could automatically identify these patterns whenever a child encounters a healthcare provider.
Do you see this as something that can be implemented in mental health diagnosis apps?
Yes, absolutely. This analysis could be invaluable even in school settings. If a school becomes aware that a child fits into one of these high-risk profiles, they might automatically arrange counseling sessions for that child. This would allow them to better understand the child and assess their risk. It’s important to remember that not every child in the high-risk group will experience an event, and not every child in the low-risk group is completely safe from experiencing an event – these profiles aren’t flawless.
So, having this information in an app, accessible to schools or even parents, could prompt a more focused care approach. It could signal a need for the parents to closely monitor their child in the immediate future, to provide additional care and attention, and to seek out supportive services.
This approach does not need to be limited to medical settings. With further refinement and validation, it could be implemented in various other settings such as schools, homes, and even in applications designed for smartphones. Interpreting this data can indeed be challenging for a single person, especially a non-expert. However, integrating this information into a user-friendly application, a school computer system, or an Electronic Health Record system is highly feasible.
IMAGE CREDIT: James Antoon.
If you enjoy the content we create and would like to support us, please consider becoming a patron on Patreon! By joining our community, you’ll gain access to exclusive perks such as early access to our latest content, behind-the-scenes updates, and the ability to submit questions and suggest topics for us to cover. Your support will enable us to continue creating high-quality content and reach a wider audience.
Join us on Patreon today and let’s work together to create more amazing content! https://www.patreon.com/ScientificInquirer