In the past decade, the explosion of data collection, analysis and mining tools have created tremendous improvements in our understanding of various influences on health such as place, trauma and health-related behaviors. The promise of “big data” – extremely large data sets that may be analyzed computationally to reveal patterns, trends, and associations, especially relating to human behavior and interactions – to also inform how healthcare identifies, profiles and stratifies risk among populations is alluring.
The outcome of health for an individual reflects many factors including genetics, health behaviors, exposures and experiences over a lifetime. Understanding health outcomes for pediatric populations can present additional challenges, as child health outcomes are also integrally influenced by adult choices and behaviors. While data exist for all of those spheres of influence, datasets are typically disparate, incomplete and are not linked at the level of the individual, which makes data analysis time consuming, costly and cumbersome for health professionals.
Thus, the emergence of machine learning, which uses sophisticated algorithms to “train” machines to recognize patterns in extremely large data sets, has been touted as potentially transformative for healthcare. Sophisticated algorithms have been used to identify populations at risk for congestive heart failure disease and have detected relevant factors associated with rare conditions so that appropriate interventions can be made earlier.
Some examples of machine learning applied to pediatric populations follow:
Clinicians and researchers at Seattle Children’s Hospital and the University of Washington - Seattle utilized machine learning tools in quantitative image analysis to detect structural features that were subtly associated with hydrocephalus, a potentially life-threatening condition that could require surgical intervention.
Children’s Hospital of Atlanta researchers utilized a vast array of clinical data to create an application that enabled clinicians to understand the impact of surgical treatment for retinopathy of prematurity (ROP) on infants in real-time.
Drexel University and Children’s Hospital of Philadelphia announced a collaborative partnership that would leverage geographically-based population data with individual patient-level data to explore environmental, family and individual-level predictors of pediatric health. Researchers ultimately hope to use insights from this “Big Data” project to reduce disparities in pediatric health outcomes.
While machine learning can identify patterns and associations within and across large datasets, clinical and population-based expertise is still essential for the algorithms to be meaningful and for appropriate action to be taken based on identified patterns and associations. For example, if machine learning identified an association between housing age and an increased risk of asthma among pediatric populations, clinical expertise would be necessary to determine if there was a plausible physiologic explanation for the association, which in turn, should inform future modifications to the algorithm.
The real promise of using big data and machine learning is to predict the occurrence of preventable conditions so that healthcare systems can actually deliver and support health, and may yet be in our future. Until then, clinicians, researchers, payers and healthcare consumers can all benefit from continued exploration of the most effective way to utilize data to improve the health of populations.
Genesis Health Consulting has the expertise to help your organization create and analyze data sets to improve child health and business outcomes. For more information on how Genesis Health Consulting can help you achieve your analytic goals, click here.