Surviving Burn Injury: Drivers of Length of Hospital Stay

Chimdimma Noelyn Onah, Richard Allmendinger, Julia Handl, Ken Dunn

TL;DR

This study reveals that while burn severity is a key predictor of hospital stay length, a patient's socioeconomic status and mental health significantly influence recovery. By segmenting patients based on these factors, researchers found that the impact of burn severity diminishes in individuals with adverse mental or socioeconomic conditions, suggesting a need for personalized care plans that consider these non-physical aspects.

What Problem Does This Paper Address?

Globally, burn injuries remain a significant cause of trauma and long-term disability. As mortality rates decrease, the length of hospital stay (LOS) has become a crucial metric for assessing patient outcomes and resource management. While factors like burn severity (e.g., Total Burn Surface Area - TBSA) and age are known to influence LOS, there's a recognized gap in understanding how a patient's mental state and socioeconomic environment contribute to their recovery trajectory. Anecdotal evidence suggests these factors play a role, but a systematic, data-driven approach was needed to quantify their impact and inform better patient care and resource allocation. This paper aims to bridge this gap by investigating the interplay between socioeconomic status, mental health, and burn severity in predicting LOS.

Methods

The researchers employed a multi-stage analytical pipeline. First, they performed extensive data cleaning and preprocessing, including removing irrelevant cases and variables, and one-hot encoding categorical features. To address their hypotheses, they utilized two primary segmentation strategies:

  1. Socioeconomic Status (SES) Segmentation: Patients were grouped into five quintiles based on the Index of Multiple Deprivation (IMD), a measure of socioeconomic disadvantage.
  2. Mental State (MST) Segmentation: Patients were clustered into groups based on the presence and number of adverse mental health conditions using hierarchical agglomerative clustering (HAC) with a Gower distance metric.

For each segmented group (SES and MST clusters) and a combined "unitary" group (all adult patients), they built predictive models for LOS using LASSO (Least Absolute Shrinkage and Selection Operator) regression. LASSO is a regularization technique that performs variable selection, effectively shrinking the coefficients of less important predictors to zero, thus creating more parsimonious and interpretable models. To ensure robustness, they used k-fold cross-validation for model fitting and evaluation. Model performance was assessed using weighted root mean square error (wRMSE) and R-squared (R²).

Datasets

The study utilized data from the International Burn Injury Database (iBID), a comprehensive dataset containing anonymized patient-level information from all burn units in England. The dataset covers the period from 2003 to 2019 and includes information on over 50,000 admitted patients. After applying exclusion criteria (e.g., age < 16, non-survivors, no overnight stay, outliers), the final analysis included approximately 19,837 adult patients.

Key Insights and Ideas

  • Beyond Physical Severity: The study strongly suggests that a patient's recovery from a burn injury is not solely dictated by the physical extent of the injury but is significantly influenced by their social and psychological context.
  • Homogeneity through Segmentation: Segmenting patients based on mental health or socioeconomic status creates more homogenous groups in terms of LOS, leading to more interpretable models.
  • Adverse Conditions Blunt Severity's Impact: As a patient's adverse mental health or socioeconomic status increases, the predictive power of burn severity (like TBSA) on LOS appears to decrease. This implies that these non-physical factors can mediate or even overshadow the direct impact of physical injury.
  • Living Circumstances Matter: For patients in more deprived socioeconomic groups, factors like living circumstances and the source of the injury become more prominent predictors of LOS, highlighting the importance of the patient's environment.
  • Collaborative Model Development: The researchers emphasized a collaborative approach with burn care specialists throughout the process, ensuring the models were clinically relevant and interpretable.

Key Results

  • Unitary Model Predictors: In the overall (unitary) model, TBSA, age, and inhalation symptoms were identified as the top predictors of LOS, consistent with previous research.
  • Mental Health Impact: The number of adverse mental health conditions was a significant predictor of LOS, with patients experiencing more mental health issues having a longer median LOS. Patients with self-inflicted or suicide-attempt-related burns also showed longer LOS.
  • Socioeconomic Status Impact: Higher deprivation (lower IMD score) was associated with increased LOS. The most deprived group (SES_5) had the highest median LOS (13 days).
  • Model Performance: While there was no significant difference in predictive performance (wRMSE) between the segmented and unitary models, the segmented models offered greater interpretability and simpler explanations.
  • R-squared Improvement: Except for one cluster (MST_1), the cluster-specific regression models showed a higher R² score than the unitary model, indicating a better ability to explain LOS variability within those specific patient groups.
  • Reduced Severity Influence: In the MST-based clusters, the proportion of significant severity variables decreased as the number of mental disorders increased. Similarly, for SES-based clusters, the proportion of significant severity variables decreased with increasing deprivation.

Limitations and Future Work

  • IMD as a Proxy: A key limitation acknowledged is the use of the Index of Multiple Deprivation (IMD) as a proxy for socioeconomic status. IMD reflects area-level deprivation, not individual-level status, which could provide a more nuanced understanding. The study also used the 2004 IMD score for all patients, suggesting future work should use individual-level indicators and more up-to-date data.
  • Data Availability: The study's reliance on the iBID database means future research could explore other datasets or combine data sources.
  • Future Directions: The authors suggest that future research should focus on developing individual-level socioeconomic indicators and utilizing more current data. The findings also point towards the need for personalized care pathways that integrate mental health and socioeconomic considerations into patient triaging and management, potentially leading to improved outcomes and more efficient resource allocation.

Summary updated

Abstract

With a reduction in the mortality rate of burn patients, length of stay (LOS) has been increasingly adopted as an outcome measure. Some studies have attempted to identify factors that explain a burn patient's LOS. However, few have investigated the association between LOS and a patient's mental and socioeconomic status. There is anecdotal evidence for links between these factors; uncovering these will aid in better addressing the specific physical and emotional needs of burn patients and facilitate the planning of scarce hospital resources. Here, we employ machine learning (clustering) and statistical models (regression) to investigate whether segmentation by socioeconomic/mental status can improve the performance and interpretability of an upstream predictive model, relative to a unitary model. Although we found no significant difference in the unitary model's performance and the segment-specific models, the interpretation of the segment-specific models reveals a reduced impact of burn severity in LOS prediction with increasing adverse socioeconomic and mental status. Furthermore, the socioeconomic segments' models highlight an increased influence of living circumstances and source of injury on LOS. These findings suggest that in addition to ensuring that patients' physical needs are met, management of their mental status is crucial for delivering an effective care plan.
BibTeX
@article{onah2021surviving,
  author = {Chimdimma Noelyn Onah and Richard Allmendinger and Julia Handl and Ken Dunn},
  title = {{Surviving Burn Injury: Drivers of Length of Hospital Stay}},
  journal = {International Journal of Environmental Research and Public Health},
  year = {2021},
  doi = {10.3390/ijerph18020761},
}
Figure