Abstract Healthcare "super-utilizers"—the small percentage of patients who account for a massive share of emergency department visits—leave a distinct digital footprint. By analyzing the Facebook posts of these patients, researchers found they post twice as often as matched controls and use language significantly more associated with confusion, loneliness, and psychosomatic distress. These findings offer a potential "digital window" for proactive health interventions, moving beyond the static snapshots provided by traditional electronic health records.
Introduction
In the United States healthcare system, a stark disparity exists: approximately 5% of the population accounts for nearly 50% of total healthcare expenditures. These individuals, often termed "super-utilizers," typically possess a complex interplay of physical, behavioral, and social needs that result in frequent Emergency Department (ED) visits. While traditional management strategies like intensive care coordination have been implemented, they often yield mixed results. A primary reason for this is their focus on "downstream" logistical issues, such as transportation, rather than the "upstream" daily lived experiences and emotional stressors that drive health-seeking behavior.
Current clinical tools, specifically Electronic Health Records (EHR), are limited by their episodic nature, capturing only a snapshot of a patient's life during a clinical encounter. This study explores the utility of social media language as a continuous, naturalistic data source to fill this gap, providing insights into a patient’s mental state and social environment outside the hospital walls.
Methods
The researchers conducted a retrospective case-control study utilizing the "Social Mediome Cohort," a dataset linking historical Facebook posts with EHR data from 1,830 patients at an urban academic ED. To isolate the specific effects of "super-utilization," the team employed propensity score matching to create a balanced control group.
As shown in the demographic data, 109 super-utilizers (defined as having >6 ED visits per year) were matched with 109 controls based on age, gender, and the Charlson comorbidity index. Linguistic analysis was performed using three distinct Natural Language Processing (NLP) approaches:
- Dictionary-based (LIWC): The Linguistic Inquiry and Word Count tool categorized words into 73 psycholinguistic categories.
- Open-vocabulary (LDA): Latent Topic Allocation identified 200 "topics" or clusters of co-occurring words.
- Predictive Models: Validated text-regression models were applied to score users on mental health markers such as anger, stress, and loneliness.
Results
A primary finding was the high level of digital engagement among super-utilizers, who posted roughly twice as often as the control group (1,537 posts/user vs. 963). This suggests that digital platforms are a viable channel for reaching this high-need population.
Linguistic analysis revealed significant differences across several domains. Super-utilizers frequently exhibited what researchers termed a "digital cry for help," using language categorized as "seeking help and attention" (e.g., "somebody save me," "Lord help me"). This indicates that social media may serve as a primary support system when traditional healthcare systems feel inaccessible.
The strongest linguistic differentiator was "Confusion and Negativity" (Cohen’s D = 0.65). Patients in the super-utilizer group frequently expressed daily frustration and cognitive friction. Furthermore, their posts often reflected somatic manifestations of distress, using colloquial terms for physical pain ("Tummy and Head Hurt"). This aligns with clinical observations that high-utilizers often experience psychosomatic symptoms driven by underlying stress or mental health conditions. As detailed in the linguistic style analysis, super-utilizers also used significantly more first-person pronouns and present-tense verbs, suggesting a high degree of self-focus and immediate emotional processing.
Discussion
This study demonstrates that social media language can provide a "digital window" into the lives of healthcare super-utilizers. However, several limitations must be considered. The sample was primarily composed of younger, African American females in an urban setting, which may limit the generalizability of the findings to other demographics. Additionally, the "over-sharer" effect suggests that patients willing to share their data may be inherently different from those who are more private.
Ethical considerations are paramount; the authors emphasize that "hovering" over patients' social media lives requires strict guardrails to prevent stigmatization or the violation of patient-provider trust. Future research should expand beyond text to include image analysis and "lurking" behavior to develop a more holistic understanding of patient well-being and to refine proactive intervention strategies.
Summary updated