top of page

The Potential for Artificial Intelligence and Machine Learning in Developmental Healthcare

Updated: Mar 7

Artificial intelligence is the ability of a computer program to learn and make decisions for itself. That might sound like science fiction, but it's already being applied in a variety of industries and is contributing to life-changing innovations. In recent years, there have been rapid increases in the uses of artificial intelligence in healthcare, but it is still underutilized in the developmental healthcare sphere.

In this article, we provide an overview of artificial intelligence and its subset, machine learning; describe their uses and potential uses in healthcare; and predict how they could be used to fix some of the problems that contribute to a critical–and potentially detrimental–care gap in developmental healthcare.

In this article:

What is artificial intelligence and how does it work?

At its most basic level, an artificial intelligence (AI) system works by taking in huge amounts of data and using that data to improve its performance at a specific task.

For example, imagine that you're training an AI system to identify dogs in pictures. You would start by feeding it lots of pictures of dogs, along with pictures of other things (such as cats, houses, trees). The AI system would then use what it knows about the world (dogs have four legs, tails, etc.) to try to identify which pictures contain dogs.

Over time, as the AI system is exposed to more and more data, it will get better and better at identifying dogs, to the point where it can easily distinguish a dog from other objects.

By feeding an AI system enough data, we can teach it to do all sorts of things, from recognizing faces to driving cars.

What is machine learning?

There are two main types of AI systems:

  • rule-based systems

  • machine learning (ML) systems.

Rule-based systems are exactly what they sound like–they follow a set of rules that have been programmed into them by humans.

ML systems, on the other hand, don't rely on human-created rules, but teach themselves from examples. Simply put, ML is a method of teaching computers to learn from data, identify patterns, and make predictions. This is done through algorithms, which are a set of rules that can be followed to solve a problem.

So, going back to our dog example, an ML system would learn to identify dogs by looking at lots of pictures of dogs (and non-dogs), rather than being given a set of rules about what makes something a dog.

Importantly, ML algorithms generally improve as they’re given more data. This means that the more pictures of both dogs and non-dogs the algorithm is given to work with, the better it will become at identifying dogs.

ML is not a new field–it’s been around for decades. However, it has become more prominent in recent years due to the increasing amount of data that is available and the advances in computing power and storage.

Because it can handle–and even requires—such vast amounts of data, ML is often used for tasks that are too difficult or time-consuming for humans to do.

What is deep learning?

Deep learning is a subset of ML that enables computers to learn from data that is unstructured or unlabeled, also known as unsupervised machine learning.

This might sound confusing, so let's break it down further.

As we discussed above, ML is a method of teaching computers to recognize patterns. Deep learning takes this one step further by teaching computers to not only recognize patterns, but also to make predictions based on those patterns.

This is an important distinction because it means that deep learning can be used for much more than just classification tasks.

Deep learning can be used in applications like facial recognition, natural language processing, and even self-driving cars.

How are artificial intelligence and machine learning used in healthcare?

The benefits of AI technologies in healthcare generally

Analyzing complex data

In healthcare, AI technologies are being used to develop models that can predict everything from patient outcomes to disease severity.

Because deep learning algorithms are able to automatically learn and improve from experience without being explicitly programmed, ML is well-suited for complex problems where there is a lot of varied and unstructured data available.

For example, healthcare data can be both unstructured and text-heavy (such as medical professionals' clinical notes) and high dimensional, meaning it has a great deal of variables to consider.

AI-enabled software can quickly sift through large data sets, identify patterns, and generate insights that would be difficult for humans to spot.

This can help healthcare providers make more informed decisions about diagnosis and treatment, as well as improve the efficiency of their clinical workflows.

Increasing access

Additionally, ML increases access to quality healthcare. Patients or their families with limited education or English skills may not understand what constitutes certain milestones or which symptoms are significant and need to be reported to a medical professional.

They may also struggle to communicate their concerns or respond to questions from healthcare providers.

However, ML removes the challenges in reporting and automatically extracts relevant patient data.

The end result is that medical professionals gain the full data that they need to provide patient care.

Eliminating bias

One of the key advantages of using ML in healthcare is that it can eliminate bias in medical diagnosis and treatment.

For example, research has shown that Black children are more likely to be diagnosed with ADHD than white children, even when they have the same symptoms.

This is likely due to racial bias on the part of clinicians. But because ML algorithms are not influenced by a human’s personal biases, they can evaluate data objectively and provide more accurate diagnoses, assuming the data informing ML algorithms are unbiased and representative.

Current applications of machine learning in healthcare

ML is already used in a number of different ways in healthcare, including:

A word of caution

Despite its many benefits and applications, ML is not a foolproof solution. When building deep learning models for use in the healthcare industry, it is important to evaluate them for both interpretability and fairness.

Interpretability refers to how well humans can understand why the ML model made a certain prediction. If we don't understand the rationale behind the prediction, we won't be able to trust that it is accurate.

Fairness refers to whether or not the model treats all groups of people fairly, such as minority groups. This is important because if the model is biased against certain groups of people, it could lead to negative clinical outcomes for those groups.

The problem is that ML models learn from historical data, which often reflects the patterns of health care disparities caused by structural racism and classism. Without deliberate efforts to compensate for these inherent biases and test ML models for fairness, models trained on inaccurate historical data might perpetuate healthcare inequality.

Artificial intelligence and machine learning in pediatrics

Pediatrics has historically lagged behind other areas of medicine in adopting new technologies. However, this is beginning to change.

Recent innovative applications or proposed applications of ML in pediatrics include:

Mental health disorders

A pediatric behavioral health company, Cognoa, is using ML to diagnose mental health disorders in children. Their software screens for conditions like ADHD and autism by gathering data from a variety of sources, including information from parents, teachers, and clinicians, as well as data from wearable devices and electronic medical records. This data is then fed into an ML model that looks for patterns that can assist the software in predicting whether a child might have a specific condition.


In September 2022, the Duke Center for Autism and Brain Development received a $12 million research grant from the National Institute of Child Health and Human Development. Duke plans to use the grant to fund three separate projects that will study very different ways to use AI in autism screening.

  1. The first project will test the accuracy of a smartphone app that parents and caregivers can use to record children's interactions and behavior. The app uses AI to automatically code the videos and identify behavioral characteristics found in children who later receive an autism diagnosis.

  2. The second project utilizes AI to analyze over 200,000 health insurance claims from children aged 0-18 months, including some diagnosed with autism. Researchers will use this data to develop an algorithm that can be used to screen young children by identifying early medical conditions associated with a later autism diagnosis.

  3. The third project will utilize AI to monitor brain wave activity and synchronize it with videos of children diagnosed with autism. This will help researchers identify brain networks associated with behaviors characteristic of autism.

Other applications in pediatric healthcare

Researchers have also studied a wide variety of other potential uses for ML in pediatrics. Other applications for ML that have been shown to be successful in studies but are not yet used in a regular clinical setting include:

  1. Evaluating head trauma injuries without CT scans

  2. Detecting children with anxiety or depression

  3. Assessing the risk of serious bacterial infections in newborns with fevers

  4. Identifying children with a higher risk of asthma

Potential uses of artificial intelligence and machine learning in developmental healthcare

Current issues in developmental healthcare

In the United States, an estimated one in four children under age five are considered to be at moderate or high risk for developmental, behavioral, or social delay. And the prevalence of developmental disabilities is only increasing. However, currently, only 3% of all children received publicly-funded early intervention services (EIS) by three years of age. Multiple factors contribute to this disparity.

Failure to detect concerns early

The first five years after a child’s birth are particularly crucial to their health, well-being, and the overall trajectory of their life. In fact, 90% of a child's brain develops by age five. This is the time when the foundations for future learning, health, and behavior are established.

Early diagnosis and treatment is critical--because a child's brain grows so much during the first five years, this is the time when it has the highest neuroplasticity (the brain’s ability to essentially "rewire" itself to function in a different way).

This flexibility means that treatment has the greatest impact during the early years, especially birth to age three. The key is to identify concerns at the earliest possible age. However, in many cases the US healthcare system fails to do so.

The importance of tracking milestones

Developmental milestones are tasks or skills that most children can do by a certain age.

Generally, children achieve developmental milestones in a set pattern: for example, they crawl, then stand, then walk.

For some of the most significant developmental milestones, failure to achieve the milestone by a certain age range constitutes a "red flag," meaning it may indicate a developmental issue and warrants further evaluation by a medical professional. For this reason, among others, keeping track of developmental milestones is essential.

The problem with developmental milestones in their current form is that there are literally hundreds that a neurotypical child should achieve by the age of five. Many of these are poorly defined and depend on nuances that non-experts might not understand. This presents a significant challenge for parents looking to monitor their child's development.

The importance of screenings

This is why it is so crucial that healthcare professionals conduct formal developmental screenings. Developmental screening refers to a formal assessment of the milestones that a child has achieved at a very specific point in time, which provides important insights into how the child is developing, where they might need additional help, and what might come next.

The American Academy of Pediatrics (AAP) recommends every child be screened 5 times before the age of 3 for developmental health, including Autism.

Unfortunately, most primary care providers are not following through--nationally, developmental screening rates are as low as 17%, depending on the state. A major reason for this abysmally low rate is that physicians simply don't have time to conduct screenings during the typical 15-minute well-child visit.

Disparities in access to early diagnosis and intervention

And even if a parent or doctor detects a concern, there is a critical shortage of developmental healthcare providers in the US, especially in sub-specialities.

In fact, there are only 800 developmental and behavioral pediatricians in the entire country for a population of 23 million children under age five.

These professionals are particularly rare in rural and low-income areas–over 80% of US counties have no pediatricians at all. As a result, the wait time to be seen by a developmental and behavioral pediatrician in some parts of the country can be almost two years.

This means that precious treatment time is being wasted, worsening outcomes.

How artificial intelligence and machine learning can address these and other problems

Manual versus smart detection

Due to the short appointment times and limitations of a clinical setting, healthcare providers are unlikely to personally observe many milestones during a well-child visit.

As a result, the current model of milestone tracking depends almost entirely on parent observation, interpretation, and reporting.

Because parents have a natural incentive to believe that their children are developing normally, their reporting can be heavily biased.

All of this means that if the parent does not (1) understand what constitutes a certain milestone, (2) see their child complete the milestone or notice that they have not reached it, and (3) report the achievement or lack thereof to their healthcare provider in an accurate, unbiased, and timely manner, the provider will not have reliable data to assess the child's development.

When trained with the right medical data, ML could remove the uncertainty and bias inherent in manual detection of milestones. An ML model could be trained to detect precise nuances of a milestone and apply them the same way to every child.

Additionally, unlike a self-reported model, an ML model would automatically detect every instance of a milestone in the input provided.

Best of all, with each milestone it detects, the ML model would only get smarter.

Overwhelming volume of healthcare data

Thanks to new technology and the digitalization of medical data, the amount of available healthcare data is increasing exponentially.

Currently, around 30% of the world’s data is created only by the healthcare industry and this data is expanding faster than in any other industry. As a result, it is becoming less and less practical for healthcare providers to examine and analyze every piece of data.

ML models can sort through data, identify trends, and highlight anything notable far more efficiently than any human.

This has particular utility in developmental healthcare, which combines the need to assess hundreds of individual milestones with a need for prompt diagnosis and intervention.

Applying ML models to detecting developmental concerns could lead to thousands of children receiving earlier diagnoses and beginning treatment during the ages that they would most benefit.

Healthcare provider burnout

Between 30% and 50% of physicians, nurse practitioners, and physician assistants are thought to experience burnout. A major cause of burnout among healthcare professionals is the rising amount of time spent on administrative tasks coupled with the need to analyze large amounts of data and still maintain a doctor-patient connection. In extreme cases, burnout can contribute to worse patient outcomes and even serious medical errors.

This is of particular concern in the developmental healthcare field. As noted above, there are only 800 developmental and behavioral pediatricians in the US for 23 million children under age five. These professionals are stretched very thin.

But ML could play an important role here as well. An ML model could deal with all of the data collection and evaluation, leaving only the complex analysis and insight to healthcare providers.

This would take a huge amount of work off of physicians' plates, freeing up more time to spend with patients and reducing burnout.

The takeaway

The potential for AI and ML in developmental healthcare is immense. With the help of these technologies, we can continue to improve our understanding of how children develop, identify problems earlier, and provide targeted interventions that can make a real difference in children’s lives.

Pathfinder Health is at the forefront of this initiative. Our smart detection feature is designed to empower parents and help doctors provide the best possible care for children with developmental delays.

With ML, we can make faster and more accurate diagnoses, recommend interventions, and track progress over time–all while providing more access to developmental healthcare, reducing bias, and minimizing the burden on physicians.

We believe that every child deserves a chance to reach their fullest potential, and we hope that our tools will help make that a reality.


2. Achenie L, Scarpa A, Factor RS, et al. A machine learning strategy for autism screening in toddlers. J Dev Behav Pediatr. 2019;40(5):369–376.

3. Bridgeman PJ, Bridgeman MB, Barone J. Burnout syndrome among healthcare professionals. Am J Heal Pharm. 2018;75(3):147–152.

4. Dash S, Shakyawar SK, Sharma M, Kaushik S. Big data in healthcare: management, analysis and future prospects. J Big Data. 2019;6:54–54.

5. Matsushita FY, Jornada Krebs VL, Brunow de Carvalho W, Artificial intelligence and machine learning in pediatrics and neonatology healthcare. Rev Assoc Med Bras (1992). 2022 Jun; 68(6): 745–750.

6. Rajkomar A, Hardt M, Howell MD, et al. Ensuring Fairness in Machine Learning to Advance Health Equity. Ann Intern Med. 2018 Dec 18; 169(12): 866–872.

7. Zablotsky B, Black LI, Maenner MJ, et al. Prevalence and Trends of Developmental Disabilities among Children in the United States: 2009-2017. Pediatrics. 2019 Oct;144(4):e20190811.


Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page