Insurance Type and Withdrawal of Life-Sustaining Therapy in Critically Injured Trauma Patients

This cohort study examines patients’ insurance type and the decision to withdraw life-sustaining treatment in trauma patients who are critically injured using data from level I and level II trauma centers in the US.


Introduction
Withdrawal of life-sustaining therapy (WLST) decisions are complicated and multifactorial for trauma patients who are critically injured.Ideally, WLST is an individualized shared decision made by health care clinicians and the patient's substitute decision-makers (SDMs) in accordance with the patient's previously outlined advanced directives and values, with integration of cultural competencies and illness prognosis. 1,2However, the nature of severe traumatic injury means that trauma patients are typically younger, less likely to have preexisting care directives, more likely to be estranged from their families and SDMs, and more likely to belong to marginalized social populations compared with the general critical care population. 3These factors complicate WLST decisions and may increase the likelihood of practitioner, caregiver, or institutional biases impacting decisions and timing.
In 1986, the Emergency Medical Treatment and Labor Act was enacted to ensure that all patients who are critically ill be provided optimal care regardless of financial means or insurance status. 4 However, the decision to proceed with escalating care or heroic measures after initial stabilization of the injured patient may have significant financial implications for patients, their families, and institutions when a patient is uninsured.Over 70% of patients who are uninsured admitted for trauma are at risk of catastrophic health expenditures. 5[8] When patients are unable to pay for health care expenditures, the financial responsibility for the care provided may land on the institutions themselves.Trauma centers across the United States face a disproportionately high burden of providing care for patients who are uninsured.A 2013 study estimated an annual care cost of over $2.8 billion from uninsured trauma care. 9Accordingly, there may be disincentive from an institutional perspective in deciding to treat or escalate care for the patients who are uninsured or have Medicaid in comparison to patients who are privately insured. 10,11th this study, our primary objective was to determine if patient insurance type (private insurance, Medicaid and uninsured) is associated with time to WLST in critically injured adults cared for at US trauma centers.We hypothesized that patients who are uninsured would undergo earlier WLST compared with patients who are privately insured.Secondarily, we aimed to analyze other baseline factors that were associated with time to WLST.

Study Design, Data Source, and Research Ethics
We conducted a registry-based retrospective cohort study using data from the American College of

Study Outcome and Exposure
The primary outcome of this study was time to WLST, as defined by days after admission.WLST is defined by TQIP as a documented "decision to remove or withhold further life supporting intervention." 40

Other Potential Factors Affecting WLST
We considered a broad range of patient, treatment, and clinician factors that have previously been shown to or could be thought to affect insurance status or WLST for inclusion in our statistical modeling a priori.4][15] Patient race and ethnicity are recorded in TQIP as self-reported by either the patient or a family member in accordance with the classifications in the US Census Bureau.Workplace injury, injury intent, hospital size, teaching status, and profit status were included because we felt these variables may also secondarily affect family and clinician treatment decisions.Comorbidities were included as a composite score using the trauma comorbidity index to account for the additive effect of multiple comorbidities in worsening prognosis. 16

Statistical Analysis
Descriptive statistics were calculated and presented as mean (SD) or median (IQR) for continuous variables where appropriate and relative frequencies with percentages for categorical variables.
Standardized mean differences (SMDs) were used to compare characteristics between patients of different insurance types, with absolute differences greater than 10% indicating meaningful differences. 17 calculated adjusted hazard ratios (HRs) between insurance type and time to withdrawal of care using a shared frailty model (also known as mixed-effects survival model) using lognormal distribution with random effects and clustering by facility ID to account for institutional differences. 18ansformations of continuous variables were explored to best fit, and age was represented as a restricted cubic spline with 3 knots at the 10th, 50th, and 90th percentile.Relevant interaction terms  were explored for significance (coma with insurance type, race with insurance type, and hospital profit status with insurance type), though none were left in the final model as they were either not statistically significant (race with insurance type, and hospital profit status with insurance type) or not felt to aid in explaining the data and results (coma with insurance type).Missing data were excluded from analyses.
We planned a priori sensitivity analyses.The first was excluding all patients who died within 48 hours of presentation to account for survivor treatment assignment bias, which has previously been documented for insurance status in trauma patients as hospitals are less likely to obtain insurance coverage for trauma patients uninsured who die early. 14,19The threshold of 48 hours was chosen because prior work investigating this phenomenon determined almost all patients who would ultimately be insured at discharged were insured by day 3. 14 The second was to formally account for the competing risk of death before WLST using cause specific and subdistribution (Fine and Gray) hazard models.
All analyses were performed using SAS version 9.4 (SAS Institute).Analyses were performed on December 12, 2023.Statistical significance was set at P < .05,and all tests were 2-sided.
Compared with patients with private insurance, patients who were uninsured were younger; more likely to be male and from minoritized racial and ethnic groups; had more substance abuse; had lower rates of known chronic illness; and were more likely to be victims of assault (Table 1).In most data categories, patients with Medicaid were most similar to patients who were uninsured.
In total, 12 962 patients (4.2%) underwent WLST during their admission, with a higher Our sensitivity analysis including the subset of patients who survived longer than 48 hours did not meaningfully change our results (eTable in Supplement 1).Our exploration of nonwithdrawal death as a competing risk did not yield results indicating the need to change our analysis model to a cause-specific hazard or subdistribution hazard model (eTable in Supplement 1).

JAMA Network Open | Surgery
Insurance Type and Withdrawal of Life-Sustaining Therapy in Critically Injured Trauma Patients

Discussion
In this retrospective study of trauma patients, we found that patients who are uninsured underwent earlier WLST compared with those with private or Medicaid insurance.This association was robust to sensitivity analyses accounting for delays to obtaining insurance and the competing risk of nonwithdrawal deaths.Our study suggests that a patient's ability to pay may be associated with a shift in decision-making for WLST, which could indicate that socioeconomic status informs end-oflife care for some patients with critical injuries.
To our knowledge, our study is the first large cohort study to examine the association between insurance status and WLST.1][22] These prior works have detailed the potential social factors that may contribute to discrepancies in outcomes between patients are uninsured and insured.One factor to consider as it related to WLST is the availability and level of involvement of SDMs in the clinical decisions.Patients belonging to socially disadvantaged groups may be less likely to have SDMs who are available, able, or willing to participate in treatment discussions.In this setting, clinicians may be more likely to proceed with earlier WLST in patients with a poor prognosis given the inability to have extensive discussion and family consultation before deciding, which can be time consuming.This represents 1 of the possible contributing social factors as to why patients who are uninsured were more likely to undergo earlier WLST.Our present study also suggests that a contributor to the increased rates of mortality in these prior studies could be from decisions to withdrawal or not proceed with treatment among patients who are uninsured.7][8] When presented with decisions about whether or not to proceed with tests, procedures, or care continuation, institutions and/or SDMs may both have concerns with the cost of care and be less likely to pursue extensive measures, resulting in earlier mortality.However, we similarly found higher rates of nonwithdrawal deaths in patients who are uninsured in our cause-specific hazards analysis, demonstrating patients who are uninsured have higher risks of both WLST and death from other causes.
A 2015 study by Osler et al 14 questioned the legitimacy of the association between insurance status and mortality in trauma patients, stating that previous studies reporting this association lacked confounder control (specifically gunshot mechanism and shock on admission) and did not account for survivor treatment assignment bias in the insurance variable.The survivor treatment assignment bias phenomenon exists due to the patients' insurance status being considered as a baseline characteristic in studies, when in fact it is recorded at their discharge or death, with patients who are uninsured who live longer being more likely to become insured during their stay than those who die early.After controlling for these issues in their models, the authors found no association between insurance status and mortality.In a sensitivity analysis that mirrored their approach, we found that the association between insurance status and earlier WLST still held true.[25] When comparing those insured by Medicaid with patients who are privately insured, there were minimal differences in rates of WLST.This finding was true despite Medicaid patients appearing most similar to the uninsured group at baseline.This finding would suggest that insurance status itself may be a more important estimator of WLST timing than patient demographics or characteristics.
However, it is also important to consider that patients enrolled in Medicaid may have greater support structures or face fewer systematic barriers to obtaining insurance than those who remain uninsured, and these social differences might contribute to discrepancies in WLST decisions.Furthermore, while discrepancies in hospital renumeration may exist between private and Medicaid insurance payouts

JAMA Network Open | Surgery
Insurance Type and Withdrawal of Life-Sustaining Therapy in Critically Injured Trauma Patients for equivalent care, this did not result in earlier WLST for Medicaid patients.The potential financial concerns for SDMs of patients who are uninsured considering care continuation are likely not as present for SDMs of patients whose medical costs will be covered by Medicaid.From an institutional and clinician perspective, there may be a larger perceived difference when comparing the financial costs of pursuing heroic measures for an insured patient with an uninsured patient, who is unlikely to pay, than there is when comparing the incremental differences between renumeration amounts in patients that are differently insured.
Secondarily, we analyzed the association between other factors included in our model and time to WLST.We found that Asian, Black, and Hispanic patients were less likely to undergo WLST than non-Hispanic White patients.In an exploratory manner, we also examined the interplay between race and insurance status and found that no significant interaction existed, indicating there may be similar WLST practices within racial groups across different insurance types.7][28][29] In our study, the decreased likelihood of WLST was countered by an increased likelihood of nonwithdrawal death for these populations, suggesting that mortality may remain similar across races after controlling for injury severity and comorbidities despite differences in end-of-life decision-making.1][32] In population or large database studies, race and ethnicity may be a surrogate for how cultural and religious beliefs inform the patient and family wishes during end-of-life care and WLST decision-making.
In 2014, the Patient Protection and Affordable Care Act introduced a federal mandate for personal health insurance with a financial penalty for those who were uninsured without exemption.
This individual mandate penalty was repealed in 2019 allowing individuals to remain uninsured without direct disincentive. 33A recent US Centers for Disease Control and Prevention publication 34 estimates that 12.2% of Americans aged 18 to 64 years are uninsured.The groups at highest risk of being uninsured included those of Hispanic ethnicity, those with family incomes less than twice the federal poverty line, and those who lived in non-Medicaid expansion states. 34In our study, and as previously documented, patients who are uninsured experience disproportionately high rates of traumatic injury, with the uninsured representing a greater proportion of the trauma population compared with the general population. 35Our findings suggest that there is a fundamental problem for equity of care, as the ability to pay may affect decisions to withdraw life sustaining therapy for patients who are uninsured.

Limitations
This study has limitations.First, as with all retrospective cohort studies, our modeling and risk adjustment are limited to variables that are recorded and accessible within the TQIP database.Thus, we were not able to include personal or family income or religious and spiritual beliefs in our analysis, which we believe may have aided in our ability to further explain our findings.Similarly, controlling for geographic location of the treating hospital may have helped account for some cultural influence on both patients and clinicians as well as state-based influences of Medicaid expansion eligibility, 34,36 though clustering by institution likely reduced some confounding in this regard.No data for marital status, language ability, or education level were available for inclusion.Second, we included comorbidities in our model to control for prognostic differences; however, the diagnosis of a comorbid condition often depends on the prior use of primary care or the health care system.
Patients who are marginalized or uninsured are less likely to seek regular primary care, and thus are less likely to have documented chronic disease diagnoses, such as hypertension. 37This phenomenon may partly explain why hypertension and smoking status are often found to be protective in trauma prediction models, 16,38,39 since they either require prior documentation or a length of stay long Trauma Quality Improvement Program (TQIP) over 4 years (2017 to 2020).TQIP is a registry of more than 700 participating trauma centers across the US.Patients with at least 1 severe injury (abbreviated injury score [AIS] of 3 or more in at least 1 body region) are included.Data quality is ensured through trained data abstractors collecting and validating data entry and interrater reliability external audits performed.This project was approved by the St Michael's Hospital research ethics board.The need for patient informed consent was waived due to the deidentified nature of the data.This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. 12Population Patients aged 18 to 64 years treated at level I and II trauma centers participating in TQIP between January 1, 2017, and December 31, 2020, were considered.January 2017 was selected as the starting point because this corresponds with the introduction of the WLST variables within the TQIP registry.Assembly of the study cohort is shown in the Figure.We included Medicaid, private insurance, and uninsured trauma patients who required admission to the intensive care unit.Patients with missing insurance status and those with Veterans Affairs insurance or Medicare (in patients younger than 65 JAMA Network Open | Surgery Insurance Type and Withdrawal of Life-Sustaining Therapy in Critically Injured Trauma Patients

Figure .
Figure.Assembly of the Patient Cohort

Table 1 .
Cohort Characteristics by Insurance Type

Table 2 .
Associations Between Variables of Interest and Time to Withdrawal of Life-Sustaining Therapy a Cubic spline.bOther race is inclusive of patients whose race was unknown, not reported or missing.