Trends in charges and association with defaults on medical payments in uninsured Americans: a disproportionate burden in ethnic minorities - a retrospective observational study.

OBJECTIVE: To evaluate whether medical event charges are associated with uninsured patients' probability of medical payment default and whether there exist racial/ethnic disparity gaps in medical payment defaults.
DESIGN: We use logistic regression models to analyse medical payment defaults. Our adjusted estimates further control for a rich set of patient and medical visit characteristics, region and time fixed effects.
SETTING: Uninsured US adult (non-elderly) population from 2002 to 2017. PARTICIPANTS: We use four nationally representative samples of uninsured patients from the Medical Expenditure Panel Survey across office-based (n=39 967), emergency (n=3269), outpatient (n=1739) and inpatient (n=340) events. PRIMARY AND SECONDARY OUTCOME MEASURES: Payment default, medical event charges and medical event payments.
RESULTS: Relative to uninsured non-Hispanic white (NHW) patients, uninsured non-Hispanic black (NHB) patients are 142% (p<0.01) more likely to default on medical payments for office-based visits, 27% (p<0.05) more likely to default on emergency department visit payments and 82% (p<0.1) more likely to default on an outpatient visit bill. Hispanic patients are 46% (p<0.01) more likely to default on an office-based visit, but 25% less likely to default on emergency department visit payments than NHW patients. Within our fully adjusted model, we find that racial/ethnic disparities persist for office-based visits. Our results further suggest that the probabilities of payment defaults for office-based, emergency and outpatient visits are all significantly (p<0.01) and positively associated with the medical event charges billed.
CONCLUSIONS: Medical event charges are found to be broadly associated with payment defaults, and we further note disproportionate payment default disparities among NHB patients. © Author(s) (or their employer(s)) 2022. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

The study uses four nationally representative samples of uninsured patients from the Medical Expenditure Panel Survey (years 2002–2017) across office-based, emergency, outpatient and inpatient medical events.

The study uses logistic regression models to analyse medical payment defaults and racial/ethnic disparity gaps within these defaults.

Estimates are adjusted for a rich set of patient and medical visit characteristics, as well as region (and year) fixed effects, which are included to adjust for region-specific, time-invariant unobserved confounders.

The study examines charges and does not fully disentangle associations due to the event-specific volume of care received and chargemaster price levels.

Given the observational study design, estimates indicate associations rather than causal effects.

Introduction

High medical bills have been identified as a primary cause of personal bankruptcies in the USA.1–4 While the effect of medical bills on personal bankruptcies is experienced by both insured and uninsured patients,5 the risk of medical bills resulting in challenging payment decisions by uninsured patients is potentially higher as these patients do not have a third party with bargaining power to negotiate favourable rates on their behalf, and as such are billed based on providers’ chargemaster rates.6–8 Chargemaster rates are list prices that providers and health systems assign to each of their medical services, and these list prices are often several factors of magnitude higher than those extended to individuals with either private or public insurance.9–11 Furthermore, these prices have been increasing in relation to underlying provider costs for the past three decades.9–12

While provider policy often allows uninsured patients to negotiate down their chargemaster bills, many patients are unaware of this possibility as information pertaining to financial adjustment practices is hard to access and commonly not volunteered by providers to patients.7 13 14 The difficult task of navigating these charge negotiations as an uninsured patient means that many patients may resort to defaulting on their medical bills, which in turn may subject them to debt collection, having their credit tarnished and place them at increased risk of personal bankruptcy. Recent work has highlighted the causal link between default in medical bills from hospitalisations and personal bankruptcies,15 16 and while the literature on catastrophic health expenditures has noted vulnerabilities among uninsured patients with hospitalisations17–20 it should also be noted that the overall frequency and severity of the financial consequences of healthcare remain at large less settled.21 22

In this study we document trends in medical charge growth across office-based, emergency department, outpatient and inpatient medical events. We further test the hypothesis that medical event charges are associated with patients’ decision about defaulting on their medical payment, and we examine the prevalence of payment default disparity gaps across non-Hispanic black (NHB), Hispanic and non-Hispanic white (NHW) patients. We choose to focus on racial/ethnic disparity gaps as it is well established that racial/ethnic minorities are over-represented among the uninsured and because of the historical US context of structural racist policies that restricted African Americans from building intergenerational wealth, something that may cause these individuals to be at elevated risk of medical payment defaults.

Methods

Data sources and study sample

We use two sets of data and a total of five samples for our study. For the four main samples we pool together the Medical Expenditure Panel Survey (MEPS) consolidated patient condition data files, office-based physician visit event files, emergency department visit event files, outpatient visit event files and inpatient stay event files for the 2002–2017 period.23 24

Our patient inclusion criteria consist of adults under the age of 65 who report being uninsured for the full year during which they are surveyed and who report a race/ethnicity of NHB, NHW or Hispanic in order to ensure sufficient sample sizes. We further ensure our sample is that of fully self-paying (uninsured) patients by excluding any medical event that is recorded with having a payor source other than self-pay.

We restrict our focus to non-flat-fee payments as these may report payments of zero dollars in settings where the services have been prepaid. Lastly, we check the payment data for logical consistencies and exclude events with payments exceeding the actual charged amount (see online supplemental table S1 for further inclusion restriction details across each of our samples).

Our fifth and final analysis sample is used for the purpose of descriptive trend analyses and this sample consists of the MEPS consolidated patient condition data files alone for the 2002–2017 period. Our inclusion criterion for these data is that of adults (18 years of age and older) with positive medical charges and no missing data on our main covariates (listed in the Covariates section).

Patient and public involvement

Given the retrospective research design of our study, patients and the public were not involved in any way.

Study variables

Outcome variables

Total payment reflects the actual amount paid by the individual in conjunction with the medical visit and is a pure self-pay amount for the uninsured individual.

Beyond our examination of the total payment amount, we also examine the patient’s decision of whether or not to default on payment. Here, we code an encounter as a payment default if the patient’s payment for a visit is equal to zero and the charge is a strictly positive amount. A number of corrections were performed in order to ensure this zero payment is indeed due to the patient failing to pay the provider. First, we check that there is no other payment source for the service (public insurance, private insurance, etc). Second, visits covered under flat-fee agreements are omitted from the analysis since the payment for these are paid in full on the patient’s first visit and recorded as zero payments on all events after that. Third, since some follow-up events are sometimes provided free of charge, all these zero charge events are omitted from the analysis. The remaining zero payment events are those used within the analysis. While it is possible that some of the zero events observed may be the result of erroneous billing where a visit was included on another bill, the occurrence of this would likely be random and as such would only work to induce noise into the analysis.

Primary independent variables

Our charge variable is the total charge related to the patient’s medical event (ie, the charge pertaining to their office-based visit, emergency department visit, outpatient visit or inpatient stay). This charge amount reflects the amount that is billed to the patient for this event in full.

Additionally, we include three sets of race/ethnicity categories: (1) NHB, (2) NHW and (3) Hispanic. Other race categories are excluded from the analysis due to small sample size. NHW is used as the reference category within our main analysis.

Covariates

To account for factors that influence patients’ desire to seek medical care, we employ a rich set of controls that build on the healthcare utilisation framework of Andersen and Newman.25

The Andersen and Newman framework of healthcare utilisation categorises factors that influence utilisation as predisposing, enabling and need based.25 26 First, predisposing factors are characteristics that are present before the onset of illness and are associated with different patterns of service utilisation across patients. These factors include race/ethnicity, age, sex, marital status and family size within our analysis. Second, enabling factors capture the degree to which individuals have access to resources that allow them to obtain medical care. In our analysis, we capture the associations of enabling factors by controlling for patients’ poverty level/status (across the categories of being poor, near poor, low income, middle income or high income), and we also control for employment status, whether the patient has earned a bachelor’s college degree (or higher), and whether the patient reports having had to delay or avoid receipt of care due to anticipated costs. Third, need factors designate the perceived or evaluated presence of an illness that would provide the patient with a reason for seeking medical care. These factors are captured using information on the number of comorbidities that a patient has (measured as a count of whether the patient has a diabetes diagnosis, a diagnosis of high blood pressure, coronary heart disease, stroke, emphysema and/or arthritis). Additionally, we also include indicators of whether the patient reported needing help with activities of daily living and/or instrumental activities of daily living, along with indicators of whether the patient identifies themselves as having either fair or poor health across their physical and mental health.

Lastly, we also include controls for visit-specific contextual factors that we observe for office-based and inpatient stay visits. For office-based visits we control for whether or not the patient reports having seen a doctor, and for the inpatient stays we control for whether the patient had surgery and whether they were in the emergency department prior to being admitted for their inpatient stay.

Statistical analysis

We use multivariate regression methods to estimate the association between charges and our total payment outcome. We use logit regressions in order to measure the association between charges and race/ethnicity on patients’ medical payment default decision. In order to mitigate concerns of bias with our observational study design, we take the following two steps. First, we include controls for a broad set of patient and medical event characteristics that we believe are important for explaining payment amounts and final payment decisions by patients using the Anderson model (see Covariates section). Second, we control for census region and year indicators to account for potential confounding from unobserved geographical and time effects.

In terms of our payment regression model specification, this is given by the following:

In Equation (1), represents our patient-level outcome measure of interest (total payment), captures the event-specific total charge, is an indicator variable for whether the patient is NHB, is similarly an indicator variable for whether the patient is Hispanic, and the omitted category here is NHW. Additionally, is a vector of control variables (defined within Covariates section). Lastly, controls for census region fixed effects, while captures the year fixed effects.

Our second set of analyses examines whether charges are associated with patients’ medical default decision and whether there are racial/ethnic disparity gaps in the risk of medical payment default. This is evaluated using a logistic model specification:

In Equation (2) is an indicator variable for whether patient i defaulted on payment for the visit. The other variables are defined as in Equation (1).

Both sets of analyses are performed using Stata V.16, and in particular the built-in survey commands for mean estimates and logistic regression analysis that accounts for survey weights and provides nationally representative estimates. Statistical significance is noted at the level of p<0.01, p<0.05 and p<0.1 throughout the analyses.

Results

Time trends and sample descriptives

Figure 1 provides time trends in inflation-adjusted (in 2017 US dollars) aggregate annual (per capita) charges and expenditures (figure 1A) and information on the growth trends by the setting within which medical care is received (office-based, emergency department, outpatient and inpatient) (figure 1B).27 This figure is based on the fully consolidated MEPS data files and is included in order to provide context for the main analysis that will follow. These trends show us that total charges (on a per capita basis) have been steadily growing since 2002, and further that this growth is seen across all types of medical settings (office-based, emergency department, outpatient and inpatient); however, we also see that charges have been growing (relatively) the most within office-based and emergency department settings (with that said, we also note that there are significant level differences in terms of charges across these four event types; these level differences can be seen in online supplemental figure S1).

Figure 1

Trends in inflation-adjusted (in 2017 US dollars) charges and expenditures, 2002–2017. (A) Growth in inflation-adjusted charges, along with trend in expenditures, on a per capita basis. The bars around the mean estimates indicate 95% CI. (B) Growth in total charges per capita by type of care. ER, emergency room visit; IP, inpatient stays; OB, office-based visits; OP, outpatient department visits.

Given these trends, we next examine the summary statistics of our main study samples. Table 1 provides survey-weighted mean estimates for our four sets of samples that span office-based, emergency department, outpatient and inpatient medical events. Each of these samples is further stratified by race/ethnicity in order to facilitate mean comparisons. Looking across the office-based, emergency department and outpatient visit samples, we see a consistent disparity gap between NHB and NHW patients. These mean differences yield significantly higher risk of payment default for NHB patients for office-based visits (p<0.0001 relative to NHW, p<0.001 relative to Hispanic), emergency department visits (p=0.03 relative to NHW, p<0.0001 relative to Hispanic) and outpatient visits (p=0.04 relative to NHW, p=0.07 relative to Hispanic). Here we further note that our inpatient results yield the opposite trend; however, given the very limited sample size within our inpatient sample, we recommend against drawing any strong conclusions from this set of results. (Additional payment default time trends across race/ethnicity are provided in online supplemental figure S2).

Table 1

Sample demographics by medical event and race/ethnicity among uninsured US adults, 2002–2017

	OB visits			ER visits			OP visits			IP stays
	NHW	NHB	Hispanic	NHW	NHB	Hispanic	NHW	NHB	Hispanic	NHW	NHB	Hispanic
Payment action
Default (%)	20.3	38.4	25.4	68.6	74.2	60.9	40.2	57.6	44.1	80.7	76.3	60.9
Charge ($100)	1.5	2.2	2.1	17.0	17.5	22.9	8.8	8.4	7.3	156.1	162.2	187.4
Predisposing factors
Age	42.1	40.7	39.5	35.3	36.5	35.6	45.9	47.1	40.4	39.8	39.7	38.3
Female (%)	57.6	58.9	60.3	52.2	52.6	56.2	52.1	58.0	68.5	58.0	56.3	60.5
Married (%)	42.3	25.3	52.7	31.5	17.5	40.9	40.1	23.2	44.1	38.1	21.9	62.3
Family size	2.5	2.5	3.3	2.5	2.6	3.5	2.3	2.1	3.6	2.6	2.5	3.6
Enabling factors
Bachelor’s degree (%)	21.6	11.6	7.3	5.8	4.7	3.5	14.9	7.3	4.0	7.1	10.9	3.8
Employed (%)	62.8	55.5	62.4	58.7	54.9	61.1	64.5	43.3	56.4	41.6	44.7	52.0
Poverty level (%)
Poor	17.8	31.0	23.8	39.6	45.3	34.5	18.7	37.3	29.7	36.5	40.2	41.5
Near poor	5.6	9.9	8.4	8.2	8.8	11.4	3.7	5.9	8.7	7.2	3.2	11.0
Low income	18.8	20.9	26.2	19.3	20.3	23.7	17.2	22.1	26.7	25.8	18.1	24.0
Middle income	32.5	24.0	29.7	22.1	19.4	24.0	26.9	28.8	25.6	20.9	30.4	23.5
High income	25.3	14.1	11.9	10.8	6.2	6.5	33.5	6.0	9.4	9.7	8.2	0.0
Delayed care	19.5	12.9	8.1	23.7	17.1	12.3	17.7	5.8	20.5	33.8	7.6	8.6
Unable to get care	18.7	13.6	9.8	28.4	19.2	16.0	15.5	7.6	10.7	34.4	8.4	9.1
Need factors
Number of comorbidities	0.7	0.9	0.6	0.6	0.7	0.5	1.0	1.2	0.6	1.2	1.0	0.7
ADL help needed (%)	0.9	2.4	1.3	1.1	0.4	0.2	1.7	1.2	0.4	4.1	1.0	3.4
IADL help needed (%)	3.3	3.2	1.7	2.9	1.9	1.4	7.9	6.9	3.0	7.6	5.5	3.4
Self-health (poor/fair) (%)	22.3	26.6	28.0	28.0	33.1	32.4	28.2	36.0	32.3	43.7	43.4	38.8
Self-mental (poor/fair) (%)	12.9	11.8	11.0	18.1	17.1	10.0	14.1	10.9	13.5	29.0	16.3	9.6
Visit factors
Saw doctor at visit (%)	47.3	64.2	71.1	–	–	–	–	–	–	–	–	–
Had operation (%)	–	–	–	–	–	–	–	–	–	28.4	28.3	32.4
In ER prior to stay (%)	–	–	–	–	–	–	–	–	–	78.1	88.0	63.9
n	20 008	5170	14 524	1353	879	1018	716	372	608	133	75	126
Population size	15.6 million	1.9 million	4.5 million	0.9 million	0.3 million	0.3 million	0.6 million	0.1 million	0.2 million	85 000	32 000	32 000

ADL, activities of daily living; ER, emergency room; IADL, instrumental activities of daily living; IP, inpatient; NHB, non-Hispanic black; NHW, non-Hispanic white; OB, office-based; OP, outpatient department.

In order to explore these racial/ethnic disparity gaps further, table 2 provides disparity gap OR estimates that have been adjusted for census region and year fixed effects. Here we note qualitatively similar trends to those in table 1. That is, relative to uninsured NHW patients, uninsured NHB patients are associated with a 142% (p<0.01) higher likelihood of experiencing a medical payment default for office-based visits, a 27% (p<0.05) higher likelihood of experiencing one for emergency department visits and a 82% (p<0.01) higher likelihood of default on an outpatient visit bill.

Table 2

Racial/ethnic disparity gaps in payment default rates among uninsured US adults, 2002–2017

Sample	OB visits	ER visits	OP visits	IP stays
	(1)	(2)	(3)	(4)
	Pr (default)	Pr (default)	Pr (default)	Pr (default)
NHB	2.42***	1.27**	1.82*	0.77
	(0.36)	(0.16)	(0.57)	(0.24)
Hispanic	1.46***	0.75**	1.38	0.43***
	(0.18)	(0.09)	(0.35)	(0.08)
Observations	39 711	3257	1702	336
Year FEs	Yes	Yes	Yes	Yes
Region FEs	Yes	Yes	Yes	Yes

ORs are reported relative to NHW (the omitted category). SEs are reported within parentheses.

Reported estimates are based on using the MEPS sample weights, but the observation counts are based on actual (unweighted) observation counts.

*P<0.1, **P<0.05, ***P<0.01.

ER, emergency room; FEs, Fixed Effects; IP, inpatient; MEPS, Medical Expenditure Panel Survey; NHB, non-Hispanic black; NHW, non-Hispanic white; OB, office-based; OP, outpatient department; Pr, Probability.

Association of medical event charges and final payment amounts

Table 3 examines the association between charges (by the provider) and the resulting payment by patients. Column 1 indicates that a marginal increase in the chargemaster rate (by $100) for office-based visits is associated with an increase within the average total payment of about $13.6 (p<0.001). Column 2 presents the results for emergency department visits. Here we see that a $100 increase in charges is associated with an increase within the average payment amount by $17.1 (p<0.01). Similarly, in column 3, we note that a $100 increase in charges for outpatient services results in a $7.7 (p<0.01) increase in average payment. Lastly, in column 4, we see that a $100 increase in inpatient stay charges is associated with an increase within the average payment amount by $2.3 (p<0.01) on average (note: additional robustness check results that use an alternative generalised linear model, with a gamma distribution and log link, can be found in online supplemental table S2; a robustness check that pools office-based and outpatient visits can be seen in online supplemental table S3; and additional partial payment decision descriptives are provided in online supplemental table S4).

Table 3

Linear regression estimates for total payment among uninsured US adults, 2002–2017

Sample	OB visits	ER visits	OP visits	IP stays
	(1)	(2)	(3)	(4)
	Total payment	Total payment	Total payment	Total payment
Charge ($100s)	13.61***	17.05**	7.67***	2.35***
	(3.22)	(8.06)	(2.12)	(0.44)
Observations	39 711	3244	1702	336
R-squared	0.16	0.17	0.15	0.23
Controls included	Yes	Yes	Yes	Yes
Year FEs	Yes	Yes	Yes	Yes
Region FEs	Yes	Yes	Yes	Yes

SEs are reported within parentheses.

The control variables included are race/ethnicity, predisposing factors, enabling factors, need factors and the visit contextual factors from table 1.

Reported estimates are based on using the MEPS sample weights, but the observation counts are based on actual (unweighted) observation counts.

*P<0.1, **P<0.05, ***P<0.01.

ER, emergency room; FEs, Fixed Effects; IP, inpatient; MEPS, Medical Expenditure Panel Survey; OB, office-based; OP, outpatient department.

Default rates, charges and the racial/ethnic disparity gap

Table 4 provides our logistic regression OR estimation results across all four of our medical event samples. These fully adjusted models indicate two important sets of results. The first pertains to the association between medical event-specific charges and the probability of payment default. For office-based visits, we observe that a $100 increase in charges is associated with a 4.1% (p<0.01) increase in the odds of the patient having to default on payment. For emergency department and outpatient visits, a $100 increase in charges is instead associated with a 1.2% (p<0.01 and p<0.01) increase in the odds of a payment default. The same is not noted within our inpatient stays sample (within column 4); however, in online supplemental table S5, we show that restricting the charge distribution to lower inpatient charge events yields statistically significant (p<0.01) associations on par with those seen for emergency department and outpatient visit events. As such, we note that (overall) charge levels are associated with patients’ payment default decision and that this association may importantly depend on payment feasibility of the initial charge.

Table 4

Logit (OR) regression estimates for payment defaults by race/ethnicity among uninsured US adults, 2002–2017

Sample	OB visits	ER visits	OP visits	IP stays
	(1)	(2)	(3)	(4)
	Pr (default)	Pr (default)	Pr (default)	Pr (default)
Charged amount
Charge ($100s)	1.041***	1.012***	1.012***	1.000
	(0.007)	(0.005)	(0.004)	(0.001)
Race/ethnicity
NHB	2.232***	1.147	1.321	0.568
	(0.337)	(0.137)	(0.392)	(0.221)
Hispanic	1.488***	0.686***	0.967	0.482***
	(0.212)	(0.086)	(0.213)	(0.119)
Predisposing factors
Age	0.995	1.001	0.993	0.946***
	(0.005)	(0.005)	(0.010)	(0.012)
Female	0.902	0.984	0.962	1.187
	(0.090)	(0.099)	(0.168)	(0.288)
Married	0.634***	0.990	0.615**	0.952
	(0.076)	(0.131)	(0.144)	(0.229)
Family size	1.051	0.967	0.959	0.838***
	(0.041)	(0.033)	(0.062)	(0.042)
Midwest	0.846	0.810	1.268	4.549***
	(0.172)	(0.154)	(0.421)	(1.179)
South	0.665**	0.835	0.907	2.034**
	(0.121)	(0.150)	(0.288)	(0.605)
West	0.614**	0.604**	0.662	0.652
	(0.125)	(0.119)	(0.204)	(0.203)
Enabling factors
Bachelor’s degree	0.851	0.462***	0.842	2.635*
	(0.142)	(0.117)	(0.304)	(1.462)
Employed	0.848*	0.948	1.044	0.815
	(0.081)	(0.119)	(0.211)	(0.250)
Near poor	1.150	0.902	0.514*	1.466
	(0.189)	(0.156)	(0.203)	(0.824)
Low income	1.005	0.750**	0.820	0.689
	(0.140)	(0.101)	(0.214)	(0.258)
Middle income	0.652***	0.461***	0.524***	0.333***
	(0.089)	(0.066)	(0.120)	(0.083)
High income	0.656**	0.433***	0.151***	0.698
	(0.132)	(0.084)	(0.049)	(0.242)
Delayed care	0.967	0.905	2.627***	0.761
	(0.174)	(0.195)	(0.739)	(0.394)
Unable to get care	0.833	1.116	0.725	0.950
	(0.177)	(0.215)	(0.227)	(0.528)
Need factors
Comorbidity count	1.176***	1.061	1.077	1.587***
	(0.074)	(0.069)	(0.122)	(0.254)
ADL	0.765	0.348**	0.428	0.739
	(0.308)	(0.169)	(0.256)	(0.388)
IADL	1.584	1.387	1.090	0.431**
	(0.480)	(0.495)	(0.598)	(0.143)
Self-health (poor or fair)	1.333**	1.231	0.978	1.001
	(0.169)	(0.166)	(0.204)	(0.346)
Self-mental (poor or fair)	1.291*	0.966	1.922**	0.625
	(0.175)	(0.158)	(0.554)	(0.180)
Visit contextual factors
Saw doctor	0.704***
	(0.076)
Had surgery				0.263***
				(0.102)
At ER before IP stay				2.232***
				(0.522)
Observations	39 711	3244	1702	336
Year FEs	Yes	Yes	Yes	Yes
Visit type FEs	Yes	Yes	Yes	No
Reason for visit FEs	No	No	No	Yes

ORs are reported. SEs are reported within parentheses.

NHW is the omitted reference category for race/ethnicity, the northeast census region is the reference category for our geographical categories, and poor is the omitted reference category for the poverty category variable.

Reported estimates are based on using the MEPS sample weights, but the observation counts are based on actual (unweighted) observation counts.

*P<0.1, **P<0.05, ***P<0.01.

ADL, activities of daily living; ER, emergency room; FEs, Fixed Effects; IADL, instrumental activities of daily living; IP, inpatient; MEPS, Medical Expenditure Panel Survey; NHB, non-Hispanic black; NHW, non-Hispanic white; OB, office-based; OP, outpatient department; Pr, Probability.

The second and perhaps more interesting result in table 4 is seen within the office-based visit sample of column 1. Here, we see that relative to NHW, the odds of defaulting are 2.23-fold (p<0.001) higher for NHB and 1.49-fold (p<0.05) higher for Hispanic patients. It is important to note that the racial/ethnic disparity gap associations within this model persist even though we control for a rich set of predisposing, enabling and need factors, along with visit-specific characteristics, year, region and visit type fixed effects. Additionally, within columns 2 and 4, we see that Hispanic patients are associated with significantly lower likelihood of default when compared with NHW patients.

Limitations

First, it should be noted that our study design is retrospective, and as such our findings should be interpreted as associations rather than as causal effects. Second, while great effort has been made to ensure zero payments only reflect payment defaults (see the ethods section and online supplemental appendix B), it remains possible that some of these zero payments may be confounded with charity care events. We believe the risk of such confounding is ameliorated on account of (1) us controlling for a broad set of income and socioeconomic controls that we believe are able to (at large) capture any systematic variation in payment default that may emanate from charity care events (which tend to be given as a direct function of patient income); and (2) additional robustness checks indicating that our default measure is significantly associated with patients reporting difficulty paying their medical bills (see online supplemental table S6). Third, it ought to be recognised that in this study we focus on four sets of patient medical events as our primary unit of observation (across: office-based, emergency department, outpatient and inpatient events). The focus on an event, rather than a specific type of procedure, implies the possibility that we might have some across-event (within a given event type) variation in the type of care provided. While it is possible that such variations could act to influence patients’ default decisions and as such should be noted, we also want to highlight that we use a comprehensive set of control variables (based on the Andersen and Newman framework of healthcare utilisation) to adjust for potential differences in care needs. We also include controls for visit-specific contextual factors to ensure further homogeneity of the visits (note: the results based on the successive addition of these controls can be seen in online supplemental table S7). Additionally, in online supplemental table S8, we show that for a subsample of our office-based visit events (years 2002–2012 for which we have additional service location details), ensuring greater medical visit homogeneity (1) by only looking at visit events that took place at a doctor’s office/group practice, (2) where the care received was a diagnosis/treatment, and (3) where the patient report having seen/talked to a doctor still yields qualitatively similar results. Lastly, we note that while we include a broad set of socioeconomic controls, socioeconomic status still remains coarsely measured within this study.

Discussion

As noted in the Results section, this study has two primary findings. The first is that medical event charges are associated with the final payment outcome of uninsured patients. This is an important association to note given the trends of the past three decades where charges are (year-on-year) outpacing the growth of underlying medical cost.10 11 The second and perhaps most significant finding is that NHB patients are associated with significantly increased odds of default on medical payments than are NHW patients (this is seen for three out of the four event types studied). Our results show that this default gap association persists even after we account for patient characteristics, such as employment status and total annual income (along with a host of other patient, health, visit event, region and time controls) in the case of office-based visits. Additionally, we find that these gaps are a persistent phenomenon across the 16 years in our sample, which spans both pre and post Affordable Care Act periods (see online supplemental figure S1). Pertaining to Hispanic patients, however, the findings are importantly more mixed. For office-based visits, we find similar patterns as for NHB; however, for emergency department and inpatient stays, this pattern is reversed, with Hispanic patients being associated with lower odds of default than NHW patients.

We believe that these default gap results highlight an important area of structural inequity within our healthcare system, primarily so for African American patients. This structural inequality is important to highlight as it risks perpetuating the harms done by historical structural racist policies such as the practice of redlining, which greatly limited credit access for African Americans, and in turn their ability to become homeowners and therethrough build intergenerational wealth.28–32 This perpetuation can be seen via a number of factors. First, growing charges (relative to underlying costs) may disproportionately hurt racial/ethnic minorities as they are over-represented among the uninsured patients that are billed these charges in full and therethrough at increased risk of catastrophic health expenditure events.9–11 33 Second, the consequences of historical inequity can further be seen in the racial/ethnic default gaps documented within this study. Third, these default gaps put NHB and Hispanic patients at increased risk of having their credit scores affected, experiencing financial distress and/or having to file for personal bankruptcy. The potential outcome of this is historically familiar—loss of credit access, which restricts the ability to build intergenerational wealth and thus perpetuates the cycle.

Our hope is that this work can help spur further work within this area, and there appears to be a number of avenues for such efforts. First, we believe that it is important to examine the risk of default within the population with inpatient medical events using a longitudinal sample with more variables than are available within this study. Second, it is important to look at this problem within populations that prior work has identified as being at an increased risk of having multiple medical events (eg, patients with chronic conditions) and high medical expenditures.34–36 Third, there is a need to target interventions for individuals at risk of medical payment defaults and personal bankruptcy. Such targeted interventions could benefit from research examining the potentially heterogeneous payment action responses across more detailed (1) patient characteristics, (2) medical event types, (3) sites and geographical locations, as well as (4) at different levels of the overall charge distribution. Lastly, we need to investigate the effect of policies that can help reverse some of these trends. Such investigations may include analysis of how current policies aimed at expanding insurance coverage (via state-level Medicaid expansions and/or Affordable Care Act market-place design tied to individual mandates and premium subsidies) can be used to reduce the prevalence of medical payment defaults. Additionally, analysis pertaining to how state-level (as well as recent federal) chargemaster transparency initiatives may be leveraged to help curb the rapid growth of charges within the US healthcare system also appears warranted.37