Use of Payer as a Proxy for Health Insurance Status on Admission Results in Misclassification of Insurance Status among Pediatric Trauma Patients

PATIENTS WHO LACK health insurance have been shown to have worse outcomes after traumatic injury than those with insurance. Uninsured trauma patients arelesslikelytobeadmittedtothehospital,receivefewer referrals to specialized care and rehabilitation, and are more likely to die from their injuries.1-9 Despite the repeated findings of increased mortality among uninsured trauma patients, the reason for this relationship remains unclear.1, 4, 6, 8, 10-13 Explanations for this observation proposed by previous investigators can be categorized into either patient-level or hospital-level factors.

At the patient level, insurance status may be a surrogate for other factors that affect mortality such as health education, management of comorbidities, or risk-taking behaviors.8 A higher prevalence of untreated preexisting conditions among uninsured trauma patients may also contribute, as may differences in healthseeking behaviors among the uninsured.1, 4, 14

Explanations specific to children include lower health literacy and more language barriers among parents of uninsured children,4 and a lower degree of parental supervision leading to a delay in seeking care after injury.12 Factors at the hospital level include hospital efforts to conserve costs among patients who potentially cannot pay and different medical care received by uninsured patients that can lead to missed injuries or delays in treatment.4, 11 Identification of factors accounting for insurance disparities in the trauma domain, however, has perplexed researchers because providers often do not know the patient's health insurance status during the initial management and treatment of injuries.

Payer (the source of payment for patient's medical care) is commonly used as a proxy for true health insurance status (insurance on hospital admission) because this information is easily obtained from administrative data sources. The accuracy of using this approach, however, has not been studied. Erroneously equating payer with health insurance coverage at the time of injury may bias analyses investigating the impact of health insurance status on outcome. The objective of this study was to quantify the extent of misclassification of health insurance in a trauma registry of pediatric patients treated at a Level I trauma center and to determine factors associated with this misclassification.

Materials and Methods

Study Setting

Children's National Health System is a Level I pediatric trauma center verified by the American College of Surgeons Committee on Trauma, serving injured patients from the greater Washington, District of Columbia, region. All trauma patients less than 18 years old, who were treated at Children's National between 2008 and 2010 and met criteria for reporting to the National Trauma Data Bank (NTDB) were included. This study was approved by the Children's National Institutional Review Board.

Participation in the NTDB

Since 1988, Children's National has maintained a data registry for all trauma patients. Starting in 2008, Children's National began submitting data from pediatric trauma patients to the NTDB, a national registry complied by the American College of Surgeons. Inclusion criteria for the NTDB are ICD-9-CM discharge diagnosis 800.00 through 959.9, excluding 905 to 909 (late effects of injury), 910 to 924 (blisters, contusions, abrasion, and insect bites), and 930 to 939 (foreign bodies). These data are aggregated and used to produce hospital benchmark reports, data quality reports, and research datasets.

Data Collection

Demographic, medical, and financial information were abstracted from the hospital financial database, the trauma registry and the patient medical record. Variables collected included patient age, sex, race, mechanism of injury, injury severity, mortality (emergency department or inpatient death), and payer (source of payment).

Most patients at our institution are registered before receiving treatment, meaning demographic and insurance information is collected from the patient or the patient's family and entered into the medical record and hospital financial database. When critically ill or injured patients are brought from the scene and require immediate treatment, the usual registration process is deferred and patients are instead registered under a generic account with default values for demographic and payment information. For example, the record lists the date of birth as "Jan 1, 1900," the hospital address is included as a placeholder for home address, and the default method of payment is "self-pay." As demographic and health insurance information is obtained from family members, the patient's record is updated and every change is time-stamped in the patient's financial record. If the record is not updated, the default values remain unchanged. To document how payer changed during the hospitalization stay, all changes to the payer field in the financial database were abstracted.

Definition of Insurance Variables

Two insurance variables were created for each patient: true (insurance on hospital admission) and payer (source of payment for medical care). The value for payer in the trauma data registry that was submitted to NTDB was labeled as "payer." Twelve methods of payment were submitted to the NTDB, from which we created a three-level payer variable: 1) insured ("Blue Cross/Blue Shield," "Government," "HMO," "private health insurance," "Medicaid," "Medicare," "other," or any combination of these); 2) uninsured ("none," "selfpay," "bad debt," "Medicaid pending"); and 3) unknown ("unknown"). NTDB payer values of "no-fault automobile," "workers compensation," and "not billed" were not found in our dataset.

We then created a variable called "true" insurance, which represents the preinjury health insurance status of each patient. If the payer remained unchanged in the financial record during the course of a patient's hospital stay, true and payer were identical. Among patients who had differences between payer and the final paying entity and among those whose bills were not paid, the medical record and financial database were reviewed and all notes (by registration clerks, social workers, case managers, etc.) were recorded to determine the preinjury health insurance status for each patient. True insurance status was classified using the same three levels as payer: 1) insured, 2) uninsured, and 3) unknown.

The two health insurance variables (true and payer) were cross-tabulated to determine the degree of disagreement between a patient's true health insurance status and the insurance payer reported to NTDB. The main outcome, health insurance "misclassification," was a binary variable assigned a value of 0 when the true and payer values were concordant and assigned a value of 1 when they were discordant.

Statistical Analysis

Health insurance misclassification was cross-tabulated with patient and injury characteristics including age, gender, race, injury mechanism (blunt versus penetrating), injury acuity [measured by Injury Severity Score (ISS) and the Glasgow Coma Scale motor response score (GCSM)], and mortality (all deaths and only emergency department deaths). Multivariable logistic regression was used to determine which factors were independently associated with health insurance status misclassification after adjusting for patient and injury characteristics. To determine if correcting the misclassified health insurance records altered the relationship between health insurance and death, cross tabulations between death and health insurance (using both payer and true insurance variables, excluding "unknown") were calculated. Fisher's exact tests were used to determine if there was a statistically significant association between insurance and mortality. All analyses were conducted usingSASversion9.2(SASInstitute,Cary,NC).

Results

Factors Related to Misclassification of Health Insurance

Among the 3630 injured patients treated between 2008 and 2010, 123 (3.4%) had misclassified health insurance on admission as reported to the NTDB. Among the 123 misclassifications, 67 (55%) were falsely classified as being uninsured, 52 (42%) were falsely classified as being insured, and 4 (3%) were falsely classified as unknown (Fig. 1). During this 3-year study period, 36 patients died. Thirteen deaths (36%) occurred in the emergency department (Table 1). Health insurance misclassification among patients who died (16.7% among all deaths; 30.8% among emergency department deaths) was higher than misclassification among patients who did not die (3.3%; P < 0.001, both). Misclassification of health insurance also differed by race and mechanism of injury. The highest misclassification was among Hispanic (8.0%), followed by other/ unknown race (6.1%), blacks (3.5%), and whites (1.7%). Patients with penetrating injuries were almost twice as likely as those with blunt injuries to have a misclassified health insurance status (6.2% vs 3.3%; P 4 0.04). No differences in health insurance misclassification were detected by ISS, GCSM, gender, or age.

After adjusting for age, gender, race, injury mechanism, and injury severity, the odds of having misclassified health insurance were higher among patients who died. The odds were 6.7 [95% confidence interval (CI): 1.7, 26.6] times greater among patients who died compared to those who did not (Table 2), and 16.1 (95% CI: 3.2, 80.8) times greater among patients who died in the emergency department (Table 3). Race also remained statistically significant, with Hispanics and those of other/unknown race having odds of health insurance status misclassification about three times greater than whites.

Change in Payer during Hospital Stay

The payer listed in the financial database changed during and after many patients' hospital stays, with some records being modified as many as 10 times. Most patients (90%) had the final payer recorded within the first two days of admission, with a mean time to determination of six [standard deviation (SD) 4 30] days. For patients without health insurance before admission based on their "true" status, the mean time to final determination of payer was 46 (SD 4 69) days. Patient data were entered into the trauma registry within five to six weeks (mean 4 38; SD 4 35 days) after patient admission, with 79 per cent of data entered into the trauma registry within 60 days.

Several factors led to delay of final determination of payer such as misspelling of the patient's name, outdated or missing insurance information, multiple insurance providers, and pending applications for public insurance programs. The Medicaid application and approval process may take weeks or months, resulting in a patient's account information being updated multiple times as new information is received and verified.

Association of Insurance Status with Mortality

When using payer as a determinant of health insurance status, the risk of death was 2.54 per cent among uninsured patients and 0.83 per cent among insured patients [relative risk of 3.1, P 4 0.02 (Table 4)]. When using true health insurance status, the relative risk of death in the uninsured compared with the insured was lowered to 2.7 and was no longer statistically significant (P 4 0.05). No relationship between health insurance and inpatient mortality (excluding deaths in the emergency department) was observed for either payer or true values of health insurance.

Discussion

In our study, using payer as a proxy for health insurance status resulted in misclassification of insurance status among pediatric trauma patients, with misclassification being associated with mortality. Patients who died in the emergency department were 16 times more likely than those who survived to have incorrectly classified health insurance status. One possible cause of this misclassification may be differences in the registration process for patients who died compared to those who survived. All patients requiring immediate treatment are registered under a generic account with default values, but in cases where the patient is admitted to the hospital or discharged home from the emergency department, there is ample time for hospital staff to interact with the patient's family members and obtain accurate demographic and insurance information. When a patient registered under a generic account dies in the emergency department, the patient's family may be present in the hospital only briefly or not at all, and staff members may feel uncomfortable asking mourning parents for health insurance information. In some cases, hospital staff did not know the true identity of patients who died at the time of their treatment.

If similar findings of insurance misclassification are observed at other trauma centers, misclassification of health insurance status may partially explain the reported relationship between lack of health insurance and mortality among both adult and pediatric trauma patients. Uninsured pediatric trauma patients treated for moderate-to-severe injuries in hospitals throughout California had three times the odds of emergency department death compared to children with private insurance.1 Additional studies among children that use NTDB data have found similar results. One analysis of records of injured children from the NTDB data found that lack of health insurance was an independent predictor of death among children younger than 14 years old.15 Additional studies found that uninsured and publicly insured children and adolescents had higher mortality after injury compared to those with private insurance4 and self-pay patients had nearly three times the odds of death than the uninsured.12 All of these studies relied on payer as a proxy for health insurance status on admission.

Several other studies using NTDB data to investigate this relationship among adults have found significantly higher mortality rates among uninsured patients.6, 8 One study among adults and the elderly found no difference in mortality between uninsured patients and those covered by Medicare but found Medicaid predicted reduced mortality compared with no insurance.16 The authors hypothesize that their results may be affected by survival bias. Uninsured trauma patients who survive may be enrolled in Medicaid before discharge, leading to an underestimated mortality rate among patients with Medicaid and an overestimated mortality rate among the uninsured.

Studies using retrospective data from single institutions have mixed association between insurance status and mortality. One study from a single trauma center found that uninsured adult trauma patients with penetrating injuries did not have a greater risk of inhospital complications or mortality compared to those with insurance.17 Patients who were dead on arrival or died in the emergency department (16% of the sample) were excluded. The proportion of uninsured among the patients who were excluded was 88 per cent, compared with 48 per cent among patients who survived initial treatment and were admitted to the hospital. The authors suggested that this finding could have resulted from the impact of insurance status on the treatment of patients in the field and during transport. A retrospective review of adult patients admitted to a Level I trauma center found that, among black and Hispanic patients, the uninsured were more likely than those with insurance to die.18 A possibility raised by our study findings is that incorrect classification of health insurance status at this institution, particularly among those who died in the emergency department, may have contributed to these findings. Another study examining factors related to survival among motor vehicle accident victims treated in state-designated trauma centers, found that being uninsured was related to death within the first 24 hours of treatment but was not related to mortality risk after 24 hours.19 The authors attributed this disparity to high risk-taking behaviors among the uninsured, but this conclusion is not addressed in their study.

Our study has several limitations. Our conclusions are based on a single trauma center's trauma registry. The advantage of our study is that we were able to access the detailed financial records of all trauma patients treated at our institution and abstract multiple values of payer as this variable changed over time. This labor-intensive task cannot be conducted on databases with a single value of payer. The NTDB contains only one value of insurance, entered into trauma registries at one point in time, which may not reflect the final payer or the patient's health insurance status on arrival. Another limitation of our study is that the "true" insurance variable was derived based on available data in the patient chart, rather than from self-report by the patient or family member. Review of the patient chart, however, allowed us to determine the true health insurance status of 99 per cent of the sample. An additional limitation is that the extent of health insurance status misclassification at other pediatric and adult institutions or more globally in the NTDB is unknown. Although efforts have been made to standardize variable definitions and coding through the introduction of the National Trauma Data Standard, the NTDB is a compilation of data from many institutions, with each using their own methods of abstracting, coding, and verifying data. Finally, because of the small number of deaths in our sample, we cannot examine the relationship of payer and true health insurance status with mortality in multivariable models. A larger dataset with more deaths will need to be examined to determine the extent to which misclassification affects the relationship between health insurance and mortality.

Our results show that payer may not be an accurate measure of prehospital health insurance in trauma registries. Similar findings may be observed in datasets that record data from patients with diagnoses that have a high mortality, particularly when occurring in the emergency department or early after admission. To study the association between health insurance and mortality in trauma patients, studies that differentiate payer from health insurance status are needed. Payer is often used as a proxy measure of health insurance status because it is readily available from many administrative data sources, but it may not reflect the true health insurance status of the patient on presentation to the hospital. The ideal study for evaluating the relationship between health insurance and outcome would determine insurance status on patient arrival rather than to ascertain it retrospectively based on the entity that paid the patient's bill. In addition, choosing a definition of health insurance that is meaningful for the research question being asked and explaining the methods used to assign patients into various insurance categories should be a minimum requirement for publications related to health insurance and trauma outcomes. A plan for managing patients who die (especially emergency department deaths) in these types of studies is also needed, as these patients may be at a higher risk of having a misclassified health insurance status.