Do high‐deductible health plans affect price paid for childbirth?

Abstract

Objective

To test whether out‐of‐pocket costs and negotiated hospital prices for childbirth change after enrollment in high‐deductible health plans (HDHPs) and whether price effects differ in markets with more hospitals.

Data Sources

Administrative medical claims data from 2010 to 2014 from three large commercial insurers with plans in all U.S. states provided by the Health Care Cost Institute (HCCI).

Study Design

I identify employer groups that switched from non‐HDHPs in 1 year to HDHPs in a subsequent year. I estimate enrollees' change in out‐of‐pocket costs and negotiated hospital prices for childbirth after HDHP switch, relative to a comparison group of employers that do not switch plans. I use a triple‐difference design to estimate price changes for enrollees in markets with more hospital choices. Finally, I re‐estimate models with hospital‐fixed effects.

Data Collection

From the HCCI sample, childbearing women enrolled in an employer‐sponsored plan with at least 10 people.

Principal Findings

Switching to an HDHP increases out‐of‐pocket cost $227 (p < 0.001; comparison group base $790) and has no meaningful effect on hospital‐negotiated prices (−$26, p = 0.756; comparison group base $5821). HDHP switch is associated with a marginally statistically significant price increase in markets with three or fewer hospitals ($343, p = 0.096; comparison group base $5806) and, relative to those markets, with a price decrease in markets with more than three hospitals (−$512; p = 0.028). Predicted prices decrease from $5702 to $5551 after HDHP switch in markets with more than three hospitals due primarily to lower prices conditional on using the same hospital.

Conclusions

Prices for childbirth in markets with more hospitals decrease after HDHP switch due to lower hospital prices for HDHPs relative to prices at those same hospitals for non‐HDHPs. These results reinforce previous findings that HDHPs do not promote price shopping but suggest negotiated prices may be lower for HDHP enrollees.

What is known on this Topic?

• People are not inclined to shop for health care services, even in high‐deductible health plans.

• Evidence about whether people pay lower prices for medical services after joining a high‐deductible health plan is mixed.

What this study adds?

• Women who live in places with more hospitals pay less for childbirth after they are enrolled in a high‐deductible health plan.

• Lower childbirth prices are largely the result of decreases in hospital‐negotiated prices for enrollees in high‐deductible plans, not enrollees shopping around for a cheaper provider.

1. INTRODUCTION

Enrollee cost sharing in health insurance benefit design has long been used to mitigate health service spending. ¹ , ² In recent years, its use has accelerated in large part through the growth of high‐deductible health plans (HDHPs), which use upfront cost sharing to increase enrollee price sensitivity with the idea it will mitigate moral hazard and encourage greater discernment about medical decisions ³ , ⁴ In 2006, 4% of U.S. workers with health insurance were covered by an HDHP; in 2019, 30% of workers were covered by such plans. ⁵ Evidence shows enrollees in HDHPs have lower total medical spending than enrollees in other types of health insurance plans, which persists over time and across service settings. ⁶ , ⁷ , ⁸ , ⁹

In aggregate, evidence suggests spending reductions are largely achieved as enrollees reduce the quantity of health services they use in HDHPs relative to lower‐deductible plans. ⁶ , ¹⁰ , ¹¹ Reductions in quantity can be problematic as consumers cut back on services important to maintaining good health such as high‐value preventive services. ⁶ , ⁸ , ⁹ , ¹² , ¹³ Health care spending may also decrease if enrollees in HDHPs realize lower prices, here specifically referring to the transaction price negotiated between the insurer and health care provider. HDHPs expose consumers to these transaction prices to a greater degree than low‐deductible plans through deductibles, co‐insurance (a fixed percentage of the negotiated price), and out‐of‐pocket maximums (the maximum amount an enrollee could pay in a plan year). HDHPs are more likely to have service‐based co‐insurance and higher out‐of‐pocket maximums than other common plan types. ¹⁴ Additionally, the plans are typically associated with an increased availability of financial management tools including price transparency tools, savings accounts, and educational materials. ¹⁵

Policy makers have been especially interested whether HDHPs lead to lower medical service prices, which may be less likely to adversely impact health than reductions in medical service utilization. (Theoretically, price reductions could harm health if price were positively correlated with clinical quality but evidence contradicts that. ¹⁶ , ¹⁷ ) Consumers could realize lower prices in two ways: either they can shop around for providers that have lower prices or an insurer can negotiate lower prices with a provider, meaning the consumer could see reductions without changing behavior. To date, evidence shows HDHP enrollees are unlikely to engage in price shopping, despite that price transparency tools are widespread and within‐market price variation is substantial. ¹⁰ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ Yet, there is little work specifically describing the behavior of HDHP enrollees in markets with more provider choices or shopping for hospital services, which represent a substantial portion of health spending. ²² , ²³ , ²⁴ Understanding the dynamics of price shopping across services and markets is important to understanding whether this mechanism may reduce spending and where it could be useful.

Consumer price sensitivity could also lead, in theory, to lower transaction prices via insurer–provider negotiations. A few studies have looked directly at transaction prices with mixed results. One cross‐sectional analysis found little evidence of lower transaction prices among HDHP enrollees in large employer plans. ²⁵ A more precisely identified study found evidence of lower prices for some services, including childbirth, in the 2 years after one large employer switched to an HDHP, though its generalizability may be limited due to the specificity of the setting. ¹⁰ Finally, recent work found modest decreases in office visit and imaging pricing for employees whose employers switched to an HDHP. ²⁶

By using a panel dataset that spans the United States, this study addresses current limitations in the literature and provides some of the first evidence of potential heterogeneity of response by market structure. Specifically, the study measures whether enrollment in an HDHP affects hospital prices for a moderately expensive hospital service, childbirth, and whether the response differs by the number of hospitals in a market. Reductions in medical service prices after HDHP enrollment, through either consumer shopping or changes in provider prices, may be an efficient mechanism through which HDHPs mitigate spending growth.

2. METHODS

The basic strategy employed by this analysis is to measure whether enrollees who give birth after an employer‐mandated switch to HDHPs have lower hospital‐negotiated prices for childbirth than prior to the switch, relative to a comparison group that maintains a low‐deductible plan. Specifics about the clinical setting, data, and analytic strategy are below.

2.1. Setting

About one‐third of all medical services are considered potentially shoppable, that is, they are not urgent and are discrete. ²⁷ However, not all shoppable services are equally likely to be shopped for. For example, services bundled with others, services so expensive that shopping is moot, or services done as specific referrals may be less amenable to shopping. ¹⁹ , ²¹ , ²²

While the specific amount a woman would pay for childbirth varies depending on her plan and the timing of the birth, enrollment in an HDHP may expose her to a substantial percent of the childbirth cost. The mean hospital‐negotiated price for vaginal delivery in these data is about $5000 in 2010; average family deductibles for employer‐sponsored HDHP enrollees in that year were about $2100. ¹⁴ After paying the deductible, women in HDHPs are typically subject to co‐insurance up to an out‐of‐pocket maximum, which, in 2010, averaged $6100. ¹⁴ Indeed, there is evidence that tools to comparison shop are used for this service, and that price differentials may change choice of the childbirth provider. Among people in the HCCI sample insured through Aetna, childbirth is the third most commonly searched‐for service in a price transparency tool. ²⁸ And, in a related setting, hospital networks tiered by price, women admitted for childbirth were found to be more price sensitive across hospitals than the general inpatient population. ²⁹

2.2. Data

This study uses administrative medical claims in the small or large group commercial insurance markets from 2010 to 2014 from three insurers: Aetna, Humana, or UnitedHealthcare. The data come from the Health Care Cost Institute (HCCI) and contain claims for about 50 million individuals or about 25% of the population in the employer‐sponsored market across all 50 U.S. states and the District of Columbia. The data contain diagnostic and procedural codes and information about the medical encounter. The dataset includes out‐of‐pocket spending amounts, negotiated prices, and encrypted provider identifiers. Importantly, HCCI data are one of the largest sources of negotiated transaction prices for medical services.

The HCCI enrollment data have age in bands of approximately 10 years, zip code, and enrollment at the month level. There is also limited information about insurance plan structure including the market in which it is sold (large group, small group, or individual), the type of network structure (e.g., point of service, preferred provider organization [PPO], health maintenance organization [HMO]), and a data flag that identifies a member as enrolled in an insurer‐defined HDHP. Finally, the data contain a group ID, which generally corresponds to an employer‐plan group though, depending on the data contributor, may signify an employer group, with employees in multiple plans. When a single group ID contains both HDHPs and non‐HDHPs, those observations are dropped (Table S1).

I merge these data via encrypted NPI numbers with the relevant years of the American Hospital Association Survey Dataset to identify individual facilities and, using zip code, with the U.S. Census Bureau Area Health Resource files to obtain information on population and median income at the core‐based statistical area (CBSA) level.

2.3. Outcome measures: Hospital prices for childbirth

I focus on one procedure, childbirth. To measure a homogenous set of procedures, I restrict my analysis to only vaginal deliveries without complications (DRG 775) and in which the primary procedural code is either manual assistance with delivery or repair of obstetrical tear (CPT 7359 or 7569). Previous studies have used the same codes to compare prices across providers. ¹⁶ I further limited the analysis to those in a hospital inpatient setting and for which length of stay was less than 5 days.

The main outcome measure is the negotiated transaction price for the medical service: the price paid by the combination of the payer (e.g., insurer) and patient to the hospital. I also measure the out‐of‐pocket price paid for childbirth (the amount paid by the patient) before and after the switch to HDHP. In this sample, about 80% of births are subject to some cost sharing.

2.4. Identifying HDHP rollovers and sample construction

The key identifying assumption is that the impetus for switching from a traditional (low‐deductible) health insurance plan to an HDHP is an employer, not individual, decision. Employer decisions are less likely made for reasons correlated to the outcome of lower prices, such as individual expected medical spending or a preference for price shopping. A similar strategy has been used in previous studies to identify the effect of HDHP structures. ³⁰ , ³¹ I cannot directly observe employer decisions in these data, so I infer it when I see an entire group of enrollees change insurance plan‐group numbers accompanied by a switch from a low‐deductible plan to an HDHP. Specifically, using the universe of data, I identify enrollees in plan‐groups with at least 10 individuals that do not have an HDHP at the beginning of 1 year and, in a subsequent year, have an HDHP and the group retains at least 95% of the same enrollees and is not larger than 105% of its original size. I follow each enrollee as long as he or she remains in that specific HDHP, up to 4 years, and does not move between hospital referral regions (HRRs). More details about the approach as well as the number of potentially eligible enrollees and final sample are in Table S1.

I form a comparison group by taking a random 25% sample of the entire universe of enrollees ages 18–64 who remain in the same low‐deductible plan throughout the study period and do not move between HRRs. The demographic makeup of the full eligible sample (without random selection) and the comparison group is almost identical (Table S2). I will refer to the randomly selected comparison group as simply the comparison group for the remainder of this paper. Comparison of individuals who are in an employer group that switches to an HDHP and those in stable benefit plans is shown in Table S3. As a final step, I flag for analysis women who gave birth among groups that switch to an HDHP and among those in the comparison group; these women make up the analytic sample and are used in all analyses in the remainder of this study.

2.5. Measurement of number of provider choices

The primary independent variable is the employer‐mandated switch to an HDHP, the construction of which is described above. Additionally, I analyze the differences in switchers' behavior in markets with few choices compared to markets with more hospital choices. I define a market using the Dartmouth Atlas Hospital Service Areas (HSAs), which are collections of zip codes that form local markets for common hospital services (dartmouthatlas.org). The ideal measure would be to look at the number of choices that each enrollee has. That depends on her insurance plan structure, the geography of her HSA, and perhaps other things, such as obstetrician admitting privileges. My data do not include variables that allow direct observation of these factors; instead, I use the total number of hospitals with obstetric services in an HSA as a proxy for the number of choices for each enrollee. I split HSAs into quartiles based on the number of hospitals and compare enrollees in the bottom quartile (with three or fewer hospitals) to those in the top three quartiles.

2.6. Analysis and estimation

Using a difference‐in‐differences framework, I leverage the panel nature of the dataset and breadth of national coverage to compare prices and out‐of‐pocket spending for births to women whose employers switch to an HDHP relative to those with stable benefit plans. I analyze the data at the birth‐year level with the following empirical specification:

$$Y_{\mathit{it}}=f\left(\alpha +{\beta }_{1}S{\mathit{\text{witchers}}}_{\mathit{it}}+{\beta }_2{\mathit{\text{SwitchersPost}}}_{\mathit{it}}+{\beta }_3{\mathit{X}}_{\mathit{it}}+{\gamma }_t+{\delta }_i+{\varepsilon }_{\mathit{kit}}\right).$$

The outcome is either the out‐of‐pocket cost or the negotiated transaction price for a birth to person, i, at time, t. The primary coefficient of interest for the main analysis is ${\beta }_2$, which picks up the difference in the spending outcome for those who switch to an HDHP after they have switched. In the equation, the $\mathit{\text{Switchers}}$ variable controls for the pre‐period‐level differences between the HDHP and comparison groups; ${\gamma }_t$ is year‐fixed effects, ${\delta }_i$ is regional‐fixed effects (Northeast, Midwest, south, west), and X is a vector of individual‐ and area‐level covariates: age (in bands), plan type (PPO, HMO, point‐of‐service [POS], other) mean area income in 2010 (by CBSA), and CBSA population in 2010. In robustness checks, I use CBSA‐fixed effects in place of the regional‐fixed effects and find it makes no difference in the significance or magnitude of the outcomes.

Histograms show severe skewedness of the main outcome variables (Figure S1). Thus, I use generalized linear models (GLM) for my primary specification, which allows more flexibility in the expectation of y. ³² To determine the correct distribution family for the GLM equation, I used a Modified Park Test. Following suggestions in the econometric literature, I used a log link function in all GLMs. ³² I use the margins command in Stata (version 14.2) to transform the difference‐in‐difference estimates from the models into real dollar amounts.

After running the main specification, I test whether there is a differential effect when an HDHP enrollee has more choices in her market, employing a triple‐difference strategy. The equation stratifies the effect of HDHP switch by market structure through a binary variable (1 = more provider choices) interacted with preswitch indicator for the HDHP group and a postperiod indicator for the HDHP group. I also interact year‐fixed effects with the market structure indicator to capture differential price trends by market structure. Once I obtain these results, I estimate a linear model by ordinary least‐squares with similar specifications and include provider‐fixed effects to test whether prices drop conditional on using the same provider.

To check the sensitivity of the main analysis, I run a series of robustness checks with varying functional forms and covariates. These checks include a nonparametric matching model, which matches each enrollee who later switches to an HDHP to its nearest neighbor in the comparison sample on plan type, CBSA residence, year, and whether an enrollee is of age 35 or older. I additionally perform matching on providers and the same set of demographic covariates to test whether prices change among individual providers after patients switch to an HDHP. Other robustness checks include analysis on only the subset of enrollees who have the most common plan type (POS), the subset of comparison and preperiod enrollees with HMOs plans that rarely use a co‐insurance structure, models with outliers trimmed, an individual‐fixed effects model on women with >1 birth, and specifications that control for differential preperiod trends.

2.7. Threats to internal validity

There are several potential threats to internal validity in this analysis: the assumption of stable unit treatment value, the possibility of nonparallel pre‐trends between the HDHP and comparison group, the assumption of common shocks in the post‐period, and, finally, the potential for selection into births after switch to HDHP. To satisfy the stable unit treatment value assumption, I restrict the HDHP sample to those in an insurer‐defined HDHP, or comparison group low‐deductible plan, again as defined by the insurer. I also test for differences in cost‐sharing between groups. While the common shock assumption is not directly testable, both groups are pulled from a large nationwide sample of employer‐sponsored plans, making it unlikely they are subject to systematically different macroeconomic shocks.

Generally, nonparallel trends in the pre‐period can signify a potential for unobserved differences between the treatment and comparison group that may bias the estimated effect of the treatment. ³³ , ³⁴ Here, pre‐trends appear parallel for the primary outcome variable, negotiated transaction prices. Figure 2(A) shows predicted yearly differences in negotiated transaction price from a GLM using observations from the comparison group and the group that switches to an HDHP prior to that switch. The graph shows roughly parallel lines; Table S4 gives the exact estimated difference in slopes—about 1% and not statistically different from 0. This analysis provides assurance that transaction price trends are approximately parallel in the pre‐period. Looking at pre‐trends for the sample split by the number of hospitals, as described above, predicted pre‐trends again appear similar when graphed (Figure S2b,c) and are not significantly divergent (Table S4; Model 3). As an additional check, I control for these pre‐trends in robustness checks.

FIGURE 2.

Average hospital transaction price for normal childbirth (in $). Estimates are based on predictions from GLM that uses a difference‐in‐differences framework to control for baseline differences between women who switch to an HDHP and those who remain in a stable benefit plan. Estimates above show the predicted prices based on whether women belong to an employer group that rolls over to an HDHP or the comparison group that remains in a low‐deductible plane. Markets with more choices are those in which there are more than three hospitals in a hospital service area providing birth services. GLM, generalized linear model; HDHP, high‐deductible health plan

By contrast, out‐of‐pocket spending does not show parallel pre‐period trends. As both the graph and table show, out‐of‐pocket spending is higher for the HDHP group in the pre‐period but growing more quickly for the comparison group (Figure S3, panel A; Table S3). This differential trend would bias the estimated effect of switching to an HDHP on out‐of‐pocket spending downward. Alternative specifications adjust for this bias using a sample matched on pre‐period out‐of‐pocket spending and other covariates to flatten differences in trends without affecting level differences (Figure S3, panel B), a technique that may yield less‐biased estimates. ³⁴

Another potential threat is that after a switch to an HDHP, there will be selection into medical treatment. Previous studies have found quantity reductions in medical service use after HDHP enrollment, which begs the question of whether, when services are used, the population using them is different. ⁸ , ¹⁰ , ³¹ Table S5 shows that the percent of each group giving birth remains at between 0.5% and 0.6% in both groups, though declines slightly after a group switches to an HDHP. Comparing the demographics between the pre‐ and post‐period among groups that switch to an HDHP, after the switch, childbearing women are slightly more likely to be younger and living in the south compared with the pre‐period (Table S6). To mitigate these differences, I adjust for demographic covariates in all analyses and test whether estimates are sensitive to alternative samples or specifications.

3. RESULTS

3.1. Descriptive statistics

The sample of women who gave birth includes 71,381 observations (births): 3065 births to women whose employers will or have switched to an HDHP and 68,316 in the comparison group. Because some women give birth more than once, the sample includes 67,099 individuals: 2805 individuals in the HDHP group and 64,294 individuals in the comparison group (Table 1). Relative to women in the comparison group, a lower percent of the HDHP group is in the 18–24 years' age group and a higher percent live in the Midwest or the west. POS network structures are more common among the HDHP group at baseline.

TABLE 1.

Descriptive statistics for analytic sample of women who give birth

	High‐deductible health plan group (n = 2085)	Comparison group (n = 64,294)	p‐value for test of statistical equivalence
Age band (%)
18–24	15.8	19.8	<0.001
25–34	68.4	65.5	0.002
35–44	14.7	12.4	<0.001
45–54	0.04	0.07	0.454
Insurance product (%)
EPO	1.8	5.7	<0.001
HMO	8.4	9.3	0.086
IND	0.07	0.05	0.559
POS	74.5	68.9	<0.001
PPO	15.2	16.1	0.235
Region of residence (%)
Northeast	20.2	17.3	<0.001
South	35.6	45.3	<0.001
Midwest	24.3	18.7	<0.001
West	20.4	19.0	0.077
Median CBSA income ($)	56,783	56,111	<0.001
Metro population a (in thousands)	4312.9	4737.9	<0.001
Number of births	3065	68,316
Number of births in markets with >3 (≤3) provider choices b	1404 (739)	36,509 (14233)

Note: All statistics are from the baseline year defined as the first year a person appears in the data (i.e., prior to switch to HDHP for the HDHP group). p‐values are taken from a t‐test of sample equivalence.

Abbreviations: EPO, exclusive provider organization; HMO, health maintenance organization; IND, indemnity plan; POS, point of service; PPO, preferred provider organization.

^a Metro population is defined as population within a core‐based statistical area (CBSA), a U.S. Census Bureau definition roughly corresponding to the cluster of population around an urban center.

^b The number of births stratified by market does not add up to the total number of births because not all births were able to be matched to a CBSA.

Unadjusted transaction prices for childbirth increase over the study period from about $5000 per birth for the HDHP and comparison groups to nearly $7000 in 2014 (Figure 1). These figures indicate some differences in nonadjusted price trends between the two groups; panels (B) and (C) of Figure 1 show that price trends differ after HDHP switch depending on the number of hospital choices in a market. Out‐of‐pocket spending shows a sharp jump in the post‐period for the HDHP group (Figure S4). While the HDHP group has slightly higher out‐of‐pocket spending relative to the comparison group prior to HDHP switch, after switching, this group's spending is distinctly higher such that by 2014, the mean is close to $1400 per year compared with about $900 in the low‐deductible group.

FIGURE 1.

Unadjusted trends in childbirth prices: These figures show the unadjusted mean negotiated transaction price by year for uncomplicated childbirth for three groups: the comparison group with a stable, low‐deductible health plan, the HDHP group prior to switch, and the HDHP group post switch. (A) Mean transaction price, full sample. (B) Mean transaction price, areas with ≤3 hospital choices. (C) Mean transaction price, areas with >3 hospital choices. HDHP, high‐deductible health plan

3.2. Changes in out‐of‐pocket price

Estimates from a linear regression model show out‐of‐pocket spending is an average of $227 more post‐switch than it would be had she not switched plan types (p < 0.001; Table 2). The parallel trends analysis suggested that this estimate may be biased downward; it is possible the actual amount is higher. Matching enrollees and employer groups on pre‐trends in out‐of‐pocket spending and other variables does indeed yield a slightly higher estimate: $238 (p < 0.001; Table S7). Other alternate specifications are consistent with the main results (Table S7), including when using an individual‐fixed effect model with the subset of women with more than one birth (Table S8).

TABLE 2.

Estimates of changes in out‐of‐pocket spending and hospital prices

	Model outcome
	Out‐of‐pocket spending ($)	Negotiated hospital transaction price
Baseline mean for comparison group ($)	790.10	5821.81
Estimated pre‐period difference in HDHP group ($)	146.69***	−81.87
Estimated change in post‐period for HDHP group net of change in comparison group (difference‐in‐differences estimate; $)	227.37***	−26.18
Relative change for HDHP group from pre‐period	24%	−0.46%
N (observations)	71,381	71,381

Note: Estimates in this table are from a GLM that uses a difference‐in‐differences design to control for baseline differences between women who switch to an HDHP and those who remain in a stable benefit plan. Covariates used: age, insurance type (e.g., PPO, EPO, HMO), region of residence, median income at core‐based statistical area (CBSA) level, and CBSA population. Year‐fixed effects included in all models. Observations are births (67,099 individuals). Standard errors are robust.

Abbreviations: EPO, exclusive provider organization; GLM, generalized linear model; HDHP, high‐deductible health plan; HMO, health maintenance organization; PPO, preferred provider organization.

*p ≤ 0.10; **p ≤ 0.05; ***p ≤ 0.001.

3.3. Changes in transaction prices

From a baseline price of $5821 for the comparison group in the pre‐period, regression models estimate the negotiated hospital price for childbirth decreases by a statistically nonsignificant $26 after a woman switches to an HDHP (p = 0.756; Table 2). However, there is heterogeneity by the number of hospitals in a market. In markets with more than three choices, women giving birth after switching to an HDHP pay an estimated $512 less than switchers in markets with fewer provider choices after the rollover (p = 0.028; Table 3). For women with fewer provider choices, prices increase by $343 after switch to HDHP, though that change is not statistically significant at the 95% level (p = 0.096). To get a better sense of the how these estimates might affect actual prices, Figure 2 shows the predicted prices for three groups: the comparison sample, HDHP sample pre‐switch, and HDHP sample post‐switch. As the graph shows, in markets with more hospital choices, women pay an average of $5702 prior to switching to an HDHP and $5551 after switching, a 3% relative drop.

TABLE 3.

Estimates of changes in hospital prices for markets with more or fewer providers

	Hospital prices	Hospital prices (including provider‐fixed effects)
Baseline mean price for childbirth for comparison group ($)	5806.11	4925.02
Estimated pre‐period difference in HDHP group ($)	−392.13**	−238.85**
Estimated pre‐period difference in markets with >3 hospitals in comparison group ($)	−424.12***	−69.77
Estimated pre‐period difference in markets with >3 hospitals in HDHP group ($)	404.64**	261.91
Estimated post‐period difference in HDHP group in markets <=3 hospitals (difference‐in‐differences estimator; $)	343.14*	253.65*
Estimated post‐period difference in HDHP group in markets with >3 hospitals relative to markets with fewer choices, net of comparison group change: (triple‐difference estimator; $)	−511.94**	−445.81**
Relative change in post‐period for markets with >3 hospital choices	−3.1%	−3.9%
N (observations)	52,885	52,885

Note: Estimates are from generalized linear or ordinary least‐squares regression models that use a difference‐in‐differences design to control for baseline differences between women who switch to an HDHP and those who remain in a stable benefit plan. Estimates above show the main effects of belonging to the HDHP group (row 2) and of market type (>3 childbirth providers; row 3), the difference in HDHP group markets with more choices relative to the comparison group in the pre‐period (row 4), the difference after HDHP switch in markets with three or fewer hospitals relative to the comparison group (row 5), and the difference in markets with more than three hospitals relative to those with fewer and to changes in the comparison group (row 6). The number of providers was determined by counting up the total number of hospitals that provided childbirth services in each hospital service area (HSA). Observations are births; the number is lower in these specifications than for main effects because not all individuals could be matched to an HSA.

Abbreviation: HDHP, high‐deductible health plan.

*p ≤ 0.10; **p ≤ 0.05; ***p ≤ 0.001.

In the model with hospital‐fixed effects, which tests whether prices change conditional on using the same hospital in the pre‐ and post‐period, I find a drop in prices of $446 after enrollees switch to an HDHP in markets with more choices relative to those with fewer choices (p = 0.0082; Table 3). A model with similar specifications without hospital‐fixed effects estimates a transaction price decrease of $482 post‐switch (Table S9, model 3). Comparing these two estimates shows that 93% of the decrease in prices for the HDHP group is explained by changes in prices conditional on using the same provider for childbirth services both before and after switching to an HDHP.

3.4. Robustness checks

I run a number of robustness checks to test the stability of the transaction price results. For estimates without market stratification, alternate specifications show results generally consistent with the main specification. In some cases, however, there is a negative effect on prices that is statistically significant (Table S7). For example, matching on demographic characteristics and CBSA, I find a decrease of $142 per birth or about 2% of the baseline price (Table S7, model 4). Additionally, matching on demographic characteristics and exactly on the provider without market stratification, I find a statistically significant decrease in price of $157 post‐rollover (Table S7, model 5). This result is consistent with the idea that changes in price are driven by provider price decreases rather than price shopping. Results are also consistent, limiting the sample to women who have a POS plan before and after HDHP switch, women who have an HMO plan in the pre‐period or comparison group, when controlling for pre‐period price trends, and when trimming outliers (Table S7, models 6–10). With an individual‐fixed effect model, using a subset of women who have >1 birth in the data, I find a $100 decrease in price, though it is not statistically significant (Table S8).

In the sample stratified by market choices, running models with other functional forms, using a subsample of women with a consistent network structure, controlling for pre‐period trends, and trimming outliers, all return generally consistent estimates (Table S9). When the sample is limited to those who had a POS plan through the entire study, the estimated decrease in transaction price is not statistically significant, and the magnitude is slightly smaller. However, the conclusion drawn from the full sample—that decreases are almost entirely accounted for by provider‐fixed effects—holds for this subset of the sample as well (Table S9, models 5–7).

4. DISCUSSION

Overall, this study shows that in markets with >3 hospital choices, enrollment in an HDHP leads to decreases in hospital prices for childbirth. Specifically, average price decreases from $5702 to $5551 after a woman switches to an HDHP, a 3% drop. Yet when hospital‐fixed effects are added to the equation, the results change very little, which suggests that the majority of the observed change in spending is due to a changes in hospital prices after the switch to HDHP and not to enrollee shopping behavior. Hospital prices could change for several reasons including because of a re‐negotiation of payer–provider contracts or because the employer, in addition to switching to an HDHP, shopped for an insurer with more favorable hospital contracts.These results are consistent with a growing body of literature that finds little enthusiasm for price shopping among consumers, even when they have shopping tools and substantial cost exposure. ²⁰ , ²¹ , ³⁵ , ³⁶ Additionally, as prices for childbirth are higher than average deductibles, an explanation of these findings could be that women in HDHPs and lower deductible plans are equally sensitive to prices once the deductible has been met. Regardless of enrollee behavior, the results from this analysis suggest negotiated hospital prices may be sensitive to enrollee out‐of‐pocket costs. Here, the results are aligned with theory developed by Craig, Ericson, and Starc, who predict that, with higher levels of out‐of‐pocket spending, demand for an insurer decreases in negotiated price level, implying firms choosing HDHPs may be more sensitive to negotiated price when evaluating contracts. ³⁷ More work needs to be done to better explore this mechanism and characterize the extent to which consumer cost exposure may put downward pressure on provider prices.

This project had limitations. The identification of an employer's choice to switch to an HDHP is not directly observable. In large administrative claims datasets, it is typically not possible to observe detailed decisions of employers, which may introduce unobservable selection bias. Nevertheless, the bias is mitigated compared to other strategies, such as measuring outcomes after an individual decisions, and, because of that, this strategy of implied rollover has been used previously to identify HDHP effects in similar datasets. ¹³ Second, this study only looked at prices in the first 4 years, at most, after rollover to HDHP; long‐run effects may be different. In particular, if these results estimate a one‐time savings, employer cost reductions may be limited. Third, there are no insurer identifiers in the data that allow for testing of whether a rollover includes a change in the insurer or whether the insurer is constant, but plan type is changing. Existing research shows there is substantial variation in hospital prices across insurers and within insurers by plan type. ¹⁶ , ³⁷ , ³⁸ Finally, the identification strategy does not allow for separation of contracting or network effects, including changes in network size or structure, from consumer shopping. The results with provider‐fixed effects and matching on the provider are suggestive that contracting plays a role in observed decreases in prices in some markets. Nevertheless, this study could not test that directly.

Evidence presented in this paper is not consistent with consumer shopping among providers on price. While the intention of HDHPs is to encourage pricing pressure through direct consumer shopping, prices may also decrease if consumer price sensitivity leads to increased insurer bargaining power or if employers shop for favorable contracts when switching plans. This study was not set up to directly test these possibilities, and future work should look for these effects. If this finding generalizes to other procedures and holds up when directly tested, it may be a mechanism through which HDHPs can control health spending.

Supporting information

Data S1. Supplementary information.

Cliff BQ. Do high‐deductible health plans affect price paid for childbirth? Health Serv Res. 2022;57(1):27‐36. 10.1111/1475-6773.13702