Andrea DeVries, PhD

 To examine how health care utilization and spending vary for low-income employees compared with high-income employees enrolled in an employer-sponsored high-deductible health plan (HDHP).

 We use commercial medical claims data and administrative human resource data from a large employer between 2014 and 2018. We link the administrative data, which include details on salary and other benefit choices, to each employee in each year with medical claims. Our variables of interest include medical spending and utilization outcomes grouped into different care settings.

 Using multivariate regressions, we estimate the association between salary buckets and health care utilization and spending, controlling for demographic characteristics, comorbidities of employees, human resource health plan benefits, and geography.

 Employees earning less than $75,000 show lower rates of utilization and spending on preventive measures, such as outpatient visits and prescription drugs, while having higher rates of utilization of preventable and avoidable emergency department visits and inpatient stays, resulting in lower overall health care spending among lower-salary employees.

 Low-salary employees enrolled in HDHPs have higher rates of acute care utilization and spending but lower rates of primary care spending compared with high-salary employees. Results suggest that HDHPs discourage routine physician-patient care among low-salary employees.

Social determinants of health are a growing topic of interest in the health insurance market because they have been shown to be associated with health care spending and health outcomes.

High-deductible health plans shift the cost of care to plan holders. This can lead to reduced utilization of care. High-deductible health plans have potential negative consequences for individuals earning a low salary.

We compared employees along the salary distribution to compare health care spending and utilization and show that low-salary individuals spend less on outpatient care but have more visits and higher spending at emergency departments.

Additional work is needed to align health care needs with access to care and to the appropriate provider setting.

The Medicare Prescription Drug Improvement and Modernization Act of 2003 established tax-sheltered health savings accounts (HSA) for high-deductible health plans (HDHPs) to incentivize individuals to save for health care needs and to consciously shop for health care services. The legislation led to a subsequent shift from traditional plans with low deductibles to HDHPs by employers and health insurance plans, with 15% of all private health plans classified as an HDHP in 2007 and 44% by 2017.

Proponents of HDHPs argue that they tend to have lower total premiums and lower out-of-pocket (OOP) premiums for employees and they retain the traditional value of insurance, protecting employees from large health care expenses with limits on total OOP spending. At the same time, they reduce health care spending by exposing consumers to the full price of care for low levels of medical spending. The RAND Health Insurance Experiment showed that increased cost sharing decreased utilization and spending without resulting in adverse health outcomes.

 A recent study among high-income earners has also shown that switching to an HDHP has resulted in limited differential changes in medical spending across income.

 HDHPs also provide incentives to shop for less costly service providers, given the widely documented variation of health care cost even within the same geographic area.

Opponents of HDHPs argue that exposing employees to the full cost of care discourages potentially necessary and preventable health care services, especially when the deductible for families can reach more than $4000.

 Studies that evaluate increases in cost-sharing provisions have found negative effects on medication adherence and even increases in mortality.

 These effects may be amplified among lower-salary employees, who may avoid or delay health care spending simply because they may not have enough savings to cover relatively minor procedures or because they anticipate high OOP costs, especially in a system that operates with unpredictable pricing and out-of-network billing tactics.

 Additional complications emerge through the well-established link between health and social determinants of health.

 Thus, lower-salary individuals generally have a higher prevalence of chronic conditions, and this factor, combined with OOP spending concerns, suggests that lower-salary individuals should be instead encouraged to utilize more health care services to adequately treat their comorbidities compared with high-salary individuals.

Therefore, the aim of this study was to evaluate health care utilization and spending along the salary distribution for employees enrolled in an HDHP. We use medical and pharmacy claims data linked to employer administrative data that include precise employee salary and a variety of human resource information to adjust for differences in sociodemographic characteristics, health plan benefit design (such as employee medical plan premiums), geography, and baseline health status of employees.

There is limited research describing how individuals continuously covered by an HDHP differentially utilize health care. The best evidence to date uses self-reported salary or US Census estimates.

 For example, one study found that HDHPs affected spending similarly for low-salary employees and high-salary employees.

 Other research has shown that lower-salary families enrolled in HDHPs were more likely than high-salary families to delay or forgo care due to costs.

 Research also found an association between lower socioeconomic status and inappropriate reductions in high-acuity emergency department (ED) visits and longer delays in breast cancer care.

 However, earlier work also found that women of low socioeconomic status enrolled in HDHPs experienced no statistically significant detectable changes in rates of cervical and breast cancer screenings.

We used commercial medical and pharmacy claims and administrative data from a large national employer between 2014 and 2018. The employer operates in the health care industry and has more than 60,000 employees residing in all states, most of whom can be considered as holding professional jobs. For each employee covered by the employer-sponsored health insurance plan, we followed their compensation history with the employer, as well as their health plan enrollment and their medical and pharmacy claims history. Medical and pharmacy claims data for the employer were obtained from the HealthCore Integrated Research Environment, a repository of fully adjudicated claims data for approximately 70 million health plan members across the United States. We limit the data to employees enrolled in a medical plan for the full plan year (January 1 to December 31) between 2014 and 2018.

The employer’s administrative data included demographic information on employee gender, age, annual salary (salary is the actual salary earned by the employee during the calendar year, exclusive of bonuses), employment status (part time vs full time), length of tenure with the company, and race/ethnicity (White, Black, Hispanic, Asian, other). Health plan benefit design elements included the chosen consumer-directed health plan (CDHP) (HDHP with an HSA), the tier (employee vs family plan), associated deductible, coinsurance rate, and OOP maximum. We also observed the health insurance premiums paid by the employee and employer and the HSA and flexible spending account (FSA) contributions by employee and employer.

Employees at the company had the choice of 3 CDHPs with a similar provider network in all years of the study. Plans differed only in cost sharing and employee premiums. Plan A was the most generous plan with the lowest deductible, lowest maximum OOP, and lowest OOP coinsurance rate, but highest employee premium. Plan B had about a 30% higher deductible, 17% higher maximum OOP, 10% higher coinsurance rate, and, on average, 25% lower employee premiums. Plan C was the least generous plan with the highest deductible, the highest maximum OOP, and the least generous coinsurance coverage, although it had the lowest employee premium. The plan tier selection (ie, employee vs family plan) affected the premium, deductible, maximum OOP, and employer contribution to the HSA/health reimbursement account (HRA) (although the employer contribution across all plans was similar and very generous). Premiums also increased with salary, although low-salary employees paid very similar premiums for all 3 plans, which should lead to higher enrollment in the most generous health plan compared with higher-salaried employees’ enrollment in the most generous health plan.

Our independent variables of interest are binary variables indicating the salary level of the employee for the year. We use the employee’s base salary for the year, excluding bonuses. The median salary was $61,013 with an IQR of $42,000 to $85,100, and the highest salaries were truncated at $200,000. We categorize employees into 4 salary bins: those earning less than $50,000, $50,000 to $74,999, $75,000 to $100,000, and more than $100,000. The inclusion of binary variables, rather than a continuous salary variable, has the advantage of allowing for nonlinear effects of salary on health care utilization. Results using alternative bins are presented in 

Our primary outcome variables measure health care utilization and spending trends for each employee in each quarter. Specifically, we create binary and continuous utilization measures for inpatient stays, outpatient visits, ED visits, and prescription fills. Health care spending is defined as the total amount (plan paid, patient paid, and any payments by secondary payers) of medical expenses associated with inpatient stays, outpatient visits, ED visits, and prescription fills. We log the spending variables to account for the skewness of high health care utilizers and to facilitate the ease of interpretation of the results.

To classify ED visits by medical urgency, we used the New York University ED algorithm.

 The algorithm assigns probabilities to ED visit as urgent (ED needed), emergent but preventable or avoidable (preventable ED care where the condition could have been prevented or avoided with timely and effective ambulatory care), emergent but primary care treatable (care could be provided in a primary care setting), and nonemergent (immediate care not required). We allocated each visit to 1 of the 4 categories related to emergency status when the assigned probability of the algorithm was 100% for said category. We also classified inpatient visits based on the Agency for Healthcare Research and Quality’s Prevention Quality Indicators (PQIs), which identify whether an inpatient stay could have been avoided had high-quality ambulatory care been provided.

 Ten different inpatient conditions can be marked as a preventable inpatient stay, and we evaluate the cumulative number of inpatient stays for all PQI conditions.

We create control variables from the medical plan information, including plan choice, tier, coinsurance rate, deductible level, OOP maximum, annual employee premium payment, annual employee and employer HSA contribution, and total FSA contributions. These variables account for differential behavior over time stemming from varying incentives to consume health care.

 We also include age, gender, and the Elixhauser Comorbidity Index score, which indicates diagnosis of a high-expenditure chronic disease or condition (eg, cancer, diabetes, mental health diagnoses).

 The administrative data provide additional control variables that include employees’ full-time employment status (reference group: part time), tenure with the employer (measured in years), and race/ethnicity (measured with binary indicators for White, Black, Hispanic, Asian, and other). We also include binary variables for the state of residence, year, and quarter to account for geographic differences in access and health care spending, seasonality of care seeking and sickness, and year-specific trends.

We used multivariate regression analyses to estimate the association of salary with health care spending and utilization at the employee-quarter level. The first set of regressions estimate linear probability models, where the dependent variable is a binary indicator for whether the employee receives any type of care within a quarter for the aforementioned health care utilization variables. The second set of regressions estimate ordinary least squares (OLS) models, with the number of encounters for each outcome as the dependent variable. For spending outcomes, we use generalized linear models, with a log-link and gamma distribution. We transformed the coefficients of interests with ecoefficient minus 1 to correctly interpret the percentage change. 

 present results from sensitivity analysis for our binary outcomes using logit models for binary outcomes, negative binomial models for continuous utilization outcomes, and log-transformed spending outcome OLS models. Our independent variables of interest are binary variables indicating whether the employee earned less than $50,000, $50,000 to $74,999, or more than $100,000 per year in salary, with $75,000 to $100,000 as the reference category. In all multivariate regressions, we included the full set of control variables to adjust for confounding that cover demographic, region, and health plan benefit elements. State, quarter, and year fixed effects were included to account for differences by state and time in spending and care utilization. We conducted all analyses using Stata 15 (StataCorp LP).

This study was conducted in full compliance with relevant provisions of the Health Insurance Portability and Accountability Act. Because researchers used only analytic files derived from a limited data set to perform the analyses as defined by the Privacy Rule 45 CFR 164.514(e), no waiver of informed consent or exemption was needed from an institutional review board.

In total, our study sample includes 448,012 employee-quarter observations, representing utilization patterns for 33,470 employees between 2014 and 2018. Individuals in the higher-salary groups (ie, salary of $75,000-$100,000 and > $100,000) were more likely to be men (27.5% and 50.2%, respectively) and more likely to identify as White (73.5% and 77.6%) or Asian (7.0% and 10.5%) than those in the lower-salary groups (

). Higher-salary employees ($75,000-$100,000 and > $100,000) were healthier relative to lower-salary employees, with a smaller proportion having 3 or more Elixhauser comorbidities (15.3% and 11.7%). Higher-salary employees were more likely to have family plans that include either their spouse, their children, or both (59% and 69%). In terms of health plan choice, approximately the same percentage of employees across the salary categories were enrolled in health plan B; however, the employees in the highest-salary group were much more likely to be enrolled in the least generous plan C (18.1%). Lastly, those earning more than $100,000 contributed more money to both their HSA/HRA and FSA ($2045 and $401, respectively) compared with lower-salary employees earning less than $50,000 ($1037 and $163).

In terms of health care utilization, 1.6% of employees had an inpatient stay during the quarter, 4.0% had an ED visit, 63% had at least 1 outpatient visit, and 65% had at least 1 prescription filled during the quarter. On average, an employee had 3.4 outpatient visits and 3.6 prescription fills per quarter. Across salary groups, outpatient care utilization was similar, although low-salary employees were more likely to use inpatient and ED care. Mean quarterly medical spending equaled $1535, large seasonality across years in spending between the last quarter of the year and the first quarter of the following year was prevalent (with high spending in the last quarter and low spending in the first quarter), and the majority (64%) of spending occurred in the outpatient setting ($989). Low-salary employees had higher quarterly average spending on inpatient ($500) and ED ($156) visits compared with those in higher-salary groups.

The linear probability model results show that employees earning less than $50,000 had a 0.14-percentage-point lower probability (

 = .01) of having an inpatient stay relative to employees earning $75,000 to $100,000 in a given quarter (

). The lowest-salary group also had a 1.8-percentage-point higher probability of visiting the ED during the quarter (

 < .01), a 4.3-percentage-point lower probability of having an outpatient visit (

 < .001), and a 1.8-percentage-point lower probability of filling a prescription during the quarter (

 < .01). On the other hand, the highest-salary group had a 1.3-percentage-point higher probability of an outpatient visit (

 < .01) and a 1.4-percentage-point higher probability of filling a prescription (

 < .01) relative to the $75,000-to-$100,000 salary group during the quarter.

The lowest-salary group had 0.022 more ED visits (

 < .01) per quarter relative to the reference group (Table 2). The lower 2 salary groups also had significantly fewer outpatient visits per quarter, with 0.43 and 0.16 fewer visits per quarter (

 < .01 for both), whereas the highest-salary group had 0.2 more outpatient visits (

 < .01) during the quarter. Lastly, the lower-salary groups had 0.22 and 0.13 fewer prescription fills (

 < .01 for both), whereas the highest-salary group had 0.12 more prescription fills (

For total medical spending, plan and member paid, only enrollees in the $50,000-to-$74,999 salary group had 6% lower spending (

 < .01) relative to employees earning $75,000 to $100,000. The highest-salary employees spent 8% more on their health care (

 < .01) relative to those earning $75,000 to $100,000 (

). Within service categories, the lowest-salary group spent 40% more on ED care (

 < .01) compared with the reference group, and the lower 2 salary groups spent significantly less on outpatient care (14% and 8%, respectively; 

 < .01 for both) and prescription drugs (10% and 5%; 

 < .02, respectively). The highest-salary group spent 7% more on outpatient visits (

 < .01) and 17% more for prescription drugs (

 < .01) compared with the reference group.

The probability and number of nonemergent ED visits were higher among the lowest-salary employees, with little differential visit trends in all other salary bins (

). Similar results were found for primary care–treatable and preventable ED visits, where the lowest-salary group had higher visit rates compared with the reference group. Findings also reveal that the lowest-salary group had more ED visits that require ED care, suggesting that they may have a higher prevalence of conditions such as chest pain. Comparing preventable inpatient stays across salary groups showed that the lowest-salary group had a higher probability and number of preventable inpatient admissions.

In this study we found that low-salary employees, especially those earning less than $50,000 per year, utilize significantly fewer outpatient services and prescriptions compared with employees earning $75,000 to $100,000 in salary. Low-salary employees, however, have significantly higher inpatient and ED service use. The lowest-salary employees had 40% higher total medical spending on ED care but less spending on outpatient care and prescription drugs. In contrast, those earning more than $100,000 a year were generally more likely to seek outpatient care and prescriptions compared with employees earning $75,000 to $100,000 per year. The utilization and cost results indicate that lower-salary employees utilize care differently than higher-salary employees in a way that suggests suboptimal care-seeking behavior.

Higher ED care use, especially for preventable conditions, and lower outpatient care use for low-salary employees may point to substitution of what would typically be considered a more appropriate setting for nonemergent conditions—physician offices. It is also possible that lower-salaried individuals do not receive the right amount of condition management. These findings are reiterated by the fact that we also found higher preventable inpatient stays. At the same time, a lower probability of pharmacy use and spending among low earners also suggests that the cost of prescriptions may be a barrier to filling prescriptions; however, we observe higher pharmacy fills, potentially implying that those who fill prescriptions are more likely to use cheaper generics. The differentially signed total pharmacy fills and spending results may reflect that those who consider filling a single prescription may be less likely to do so (extensive margin), whereas those who commonly fill prescriptions (intensive margin) are much more likely to fill prescriptions relative to employees earning $75,000 to $100,000. Additional qualitative and quantitative work is needed to understand the mechanism that could explain these differential care patterns between low- and high-salary employees.

To date, few studies have focused on vulnerable populations enrolled in HDHPs, which is important because they are especially exposed to high OOP medical spending relative to their income; this may be because of a lack of individual socioeconomic information matchable to claims data.

 Our findings are in accord with work that found that HDHPs reduced utilization and spending for individuals of low socioeconomic status and evidence that lower-salary families enrolled in HDHPs were more likely to delay or forgo care due to costs than high-salary families.

 Other work analyzed the impact of switching to an HDHP.

 Some studies found negative consequences after the switch, such as reductions in high-acuity ED visits, that were especially pronounced for low socioeconomic status groups, whereas other research did not find statistical evidence for differences in breast cancer and cervical cancer preventive care.

Compared with previous studies that have focused on low-income individuals using geographic socioeconomic data, we perform comprehensive analyses that assess health care utilization and spending patterns by salary for employees in HDHPs across time using precise individual level earnings data.

 We also used a much richer set of control variables compared with previous work to account for factors that influence health care plan choice and health care spending. The closest work to ours was performed by Sherman et al, who utilized wage data in 2014 from self-insured employers, many of them located in the South.

 They found significant disparities in spending, ED use, and hospitalizations, which are partially confirmed in our study. Medical spending was higher for the highest-salary group but lower for the lowest-salary group. We also observe significantly elevated levels of avoidable ED use for low-salary employees. However, the number of hospital admissions for the lowest- and highest-salary groups were both elevated relative to the middle-salary reference group.

The continued shift toward HDHPs can result in significant challenges for low-salary employees. Although CDHPs come with lower premiums for consumers, plans may realize higher medical expenses in the long run based on the complicated nature of their setups, misunderstanding of consumers, and consumers’ liquidity constraints to pay for OOP expenses.

 Additionally, evidence suggests that many employees prefer consumption smoothing (eg, consistent plan premium payments rather than large 1-time OOP medical payments). HDHPs reduce consumption smoothing by shifting high spending onto consumers at the time of care needs, whereas low-deductible plans shelter consumers from high OOP spending but require consistently high monthly premium payments. Consumers might even prefer low-deductible plans when total medical and premium spending would have been lower in an HDHP compared with the total medical and premium spending in a low-deductible plan.

Work by Wharam et al suggests that rising health care costs could exacerbate the negative consequences of HDHPs on preventive care use for individuals of low socioeconomic status.

 Payers and providers can counteract rising medical costs by improving price transparency, especially for preventive services that might reduce costs in the long run. Payers may also reformulate HDHPs by providing generous HSA account contributions that cover most of the deductible for low-income employees. Health insurance literacy interventions, however, might need to accompany such changes to ensure that low-salary employees are aware of the benefits of HSAs.

This study has several limitations. First, the salary data that we use include information for only the employees of the firm, whereas the employee’s household income may be higher due to the employment of other household members (such as the spouse). However, many studies on the value of HDHPs rely on Census estimates based on the geographic location of the employee’s residence. Second, it is important to understand the household decision-making process of employees because health insurance coverage decisions are normally made at the household level, but we do not know the marital status or whether the employee had dependents unless their health plan covers family members. Third, the employees in this study are employed by a single firm in a single sector of the economy. It is possible that employees in different types of jobs and different sectors of the economy use health care differently, although the employer covers a variety of jobs across the United States. Fourth, the study is limited to claims data billed through employer-sponsored coverage, so care and prescriptions purchased without insurance are not captured in our data. Fifth, differences in health care use between low- and high-salary groups may not necessarily suggest that low-salary individuals underutilize care. It is possible that higher-salary individuals overutilize care. However, previous work has shown little difference in health care use among low-salary and higher-salary groups, and in some cases there is evidence of overutilization for low-salary individuals.

This study fills an important gap in the literature by assessing the association of salary and care utilization when a HDHP is present. Understanding how HDHPs influence employees’ health care decision-making, particularly for low-salary employees who may not have the liquidity to pay for high deductibles, is important because associational evidence has shown that HDHPs reduce medical care utilization. In this study, we found that lower-salary employees enrolled in an HDHP with an HSA through their employer spent significantly less on outpatient services and prescription fills compared with higher-salary employees. At the same time, low-salary employees spent significantly more on ED care. These results suggest that low-salary employees may be forgoing outpatient services and instead spending more in the ED compared with higher-salary employees when HDHPs are present. This pattern of health care utilization may lead to delayed diagnosis of health conditions and potentially miss the window and benefits of early diagnosis or prevention.


 Texas A&M University (BU), College Station, TX; Anthem Inc (BU, MO, AD, GS), Indianapolis, IN; HealthCore (SE-P), Wilmington, DE.

 Dr DeVries and Ms Sylwestrzak are employees of HealthCore, a wholly owned research subsidiary of Anthem Inc, and own stock in Anthem Inc. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (BU, MO, AD, GS); acquisition of data (BU, SE-P, AD, GS); analysis and interpretation of data (BU, SE-P, MO, AD, GS); drafting of the manuscript (BU, SE-P, MO, GS); critical revision of the manuscript for important intellectual content (BU, SE-P, GS); statistical analysis (BU); provision of patients or study materials (GS); obtaining funding (GS); administrative, technical, or logistic support (BU, GS); and supervision (BU, GS).

 Benjamin Ukert, PhD, Texas A&M University, 212 Adriance Lab Rd, College Station, TX 77843. Email: bukert@tamu.edu.

1. Medicare Prescription Drug, Improvement, and Modernization Act of 2003, HR 1, 108th Cong (2003). Accessed June 9, 2020. https://www.congress.gov/bill/108th-congress/house-bill/1

2. Cohen RA, Zammitti EP. High-deductible health plan enrollment among adults aged 18-64 with employment-based insurance coverage. 

3. Manning WG, Newhouse JP, Duan N, Keeler EB, Leibowitz A, Marquis MS. Health insurance and the demand for medical care: evidence from a randomized experiment. 

4. Brot-Goldberg ZC, Chandra A, Handel BR, Kolstad JT. What does a deductible do? the impact of cost-sharing on health care prices, quantities, and spending dynamics. 

. 2017;132(3):1261-1318. doi:10.1093/qje/qjx013

5. Cooper Z, Craig SV, Gaynor M, Van Reenen J. The price ain’t right? hospital prices and health spending on the privately insured. 

. 2019;134(1):51-107. doi:10.1093/qje/qjy020

6. 2018 Employer Health Benefits Survey: section 8: high-deductible health plans with savings option. Kaiser Family Foundation. October 3, 2018. Accessed June 9, 2020. https://www.kff.org/report-section/2018-employer-health-benefits-survey-section-8-high-deductible-health-plans-with-savings-option/

7. Chandra A, Flack E, Obermeyer Z. The health costs of cost-sharing. National Bureau of Economic Research working paper No. 28439. February 2021. Accessed December 3, 2021. https://www.nber.org/papers/w28439

8. Chernew M, Gibson TB, Yu-Isenberg K, Sokol MC, Rosen AB, Fendrick AM. Effects of increased patient cost sharing on socioeconomic disparities in health care. 

. 2008;23(8):1131-1136. doi:10.1007/s11606-008-0614-0

9. Allen EM, Call KT, Beebe TJ, McAlpine DD, Johnson PJ. Barriers to care and health care utilization among the publicly insured. 

. 2017;55(3):207-214. doi:10.1097/MLR.0000000000000644

10. Cooper Z, Nguyen H, Shekita N, Morton FS. Out-of-network billing and negotiated payments for hospital-based physicians. 

. 2020;39(1):24-32. doi:10.1377/hlthaff.2019.00507

11. Lazar M, Davenport L. Barriers to health care access for low income families: a review of literature. 

. 2018;35(1):28-37. doi:10.1080/07370016.2018.1404832

12. Gaskin DJ, Thorpe RJ Jr, McGinty EE, et al. Disparities in diabetes: the nexus of race, poverty, and place. 

. 2014;104(11):2147-2155. doi:10.2105/AJPH.2013.301420

13. Palacio A, Mansi R, Seo D, et al. Social determinants of health score: does it help identify those at higher cardiovascular risk? 

. 2020;26(10):e312-e318. doi:10.37765/ajmc.2020.88504

14. Oates GR, Jackson BE, Partridge EE, Singh KP, Fouad MN, Bae S. Sociodemographic patterns of chronic disease: how the Mid-South region compares to the rest of the country. 

. 2017;52(1):S31-S39. doi:10.1016/j.amepre.2016.09.004

15. Safran MA, Mays RA Jr, Huang LN, et al. Mental health disparities. 

. 2009;99(11):1962-1966. doi:10.2105/AJPH.2009.167346

16. Haviland AM, Sood N, McDevitt R, Marquis MS. How do consumer-directed health plans affect vulnerable populations? 

17. Kullgren JT, Galbraith AA, Hinrichsen VL, et al. Health care use and decision making among lower-income families in high deductible health plans. 

. 2010;170(21):1918-1925. doi:10.1001/archinternmed.2010.428

18. Wharam JF, Graves AJ, Landon BE, Zhang F, Soumerai SB, Ross-Degnan D. Two-year trends in colorectal cancer screening after switch to a high-deductible health plan. 

. 2011;49(9):865-871. doi:10.1097/MLR.0b013e31821b35d8

19. Wharam JF, Zhang F, Landon BE, Soumerai SB, Ross-Degnan D. Low-socioeconomic-status enrollees in high-deductible plans reduced high-severity emergency care. 

. 2013;32(8):1398-1406. doi:10.1377/hlthaff.2012.1426

20. Wharam JF, Zhang F, Wallace J, et al. Vulnerable and less vulnerable women in high-deductible health plans experienced delayed breast cancer care. 

. 2019;38(3):408-415. doi:10.1377/hlthaff.2018.05026

21. Reed ME, Warton EM, Kim E, Solomon MD, Karter AJ. Value-based insurance design benefit offsets reductions in medication adherence associated with switch to deductible plan. 

. 2017;36(3):516-523. doi:10.1377/hlthaff.2016.1316

22. Wharam JF, Graves AJ, Zhang F, Soumerai SB, Ross-Degnan D, Landon BE. Two-year trends in cancer screening among low socioeconomic status women in an HMO-based high-deductible health plan. 

. 2012;27(9):1112-1119. doi:10.1007/s11606-012-2057-x

23. Billings J, Parikh N, Mijanovich T. Emergency department use in New York City: a substitute for primary care? 

24. Johnston KJ, Allen L, Melanson TA, Pitts SR. A “patch” to the NYU emergency department visit algorithm. 

. 2017;52(4):1264-1276. doi:10.1111/1475-6773.12638

25. Prevention Quality Indicators overview. Agency for Healthcare Research and Quality. Accessed June 9, 2020. https://www.qualityindicators.ahrq.gov/Modules/pqi_resources.aspx#techspecs

26. Sherman BW, Gibson TB, Lynch WD, Addy C. Health care use and spending patterns vary by wage level in employer-sponsored plans. 

. 2017;36(2):250-257. doi:10.1377/hlthaff.2016.1147

27. Zheng S, Ren ZJ, Heineke J, Geissler KH. Reductions in diagnostic imaging with high deductible health plans. 

. 2016;54(2):110-117. doi:10.1097/MLR.0000000000000472

28. van Walraven C, Austin PC, Jennings A, Quan H, Forster AJ. A modification of the Elixhauser comorbidity measures into a point system for hospital death using administrative data. 

. 2009;47(6):626-633. doi:10.1097/MLR.0b013e31819432e5

29. Agarwal R, Mazurenko O, Menachemi N. High-deductible health plans reduce health care cost and utilization, including use of needed preventive services. 

. 2017;36(10):1762-1768. doi:10.1377/hlthaff.2017.0610

30. Wharam JF, Zhang F, Eggleston EM, Lu CY, Soumerai S, Ross-Degnan D. Diabetes outpatient care and acute complications before and after high-deductible insurance enrollment: a Natural Experiment for Translation in Diabetes (NEXT-D) study. 

. 2017;177(3):358-368. doi:10.1001/jamainternmed.2016.8411

31. Lieu TA, Solomon JL, Sabin JE, Kullgren JT, Hinrichsen VL, Galbraith AA. Consumer awareness and strategies among families with high-deductible health plans. 

. 2010;25(3):249-254. doi:10.1007/s11606-009-1184-5

32. Kullgren JT, Galbraith AA, Hinrichsen VL, et al. Health care use and decision making among lower-income families in high-deductible health plans. 

33. Bhargava S, Loewenstein G, Sydnor J. Choose to lose: health plan choices from a menu with dominated option. 

. 2017;132(3):1319-1372. doi:10.1093/qje/qjx011

34. Charlesworth CJ, Meath THA, Schwartz AL, McConnell KJ. Comparison of low-value care in Medicaid vs commercially insured populations. 

. 2016;176(7):998-1004. doi:10.1001/jamainternmed.2016.2086

35. Schpero WL, Morden NE, Sequist TD, Rosenthal MB, Gottlieb DJ, Colla CH. For selected services, blacks and Hispanics more likely to receive low-value care than whites. 

. 2017;36(6):1065-1069. doi:10.1377/hlthaff.2016.1416

Articles

Lower-salary employees in high-deductible health plans underutilize outpatient care and overutilize emergency departments.

Disparities in Health Care Use Among Low-Salary and High-Salary Employees

 To evaluate the relationship between cost sharing for blood glucose testing strips and glycemic control rates.

 A retrospective observational study using medical and pharmacy claims data integrated with laboratory glycated hemoglobin (A1C) values for patients using insulin and testing strips. A new user study design was utilized to identify individuals from 14 commercial US health plans who filled testing strips with assumed intention to monitor blood glucose.

Patients were divided into low (<20% of annual testing strip cost; n = 3575) and high (≥20%; n = 3580) cost-sharing categories. We compared the likelihood of patients in low and high cost-sharing categories achieving glycemic control (A1C <8.0%) through modified Poisson regression models.

Patients with low cost sharing for testing strips had higher rates of control than those with high cost sharing (58.1% vs 50.3%; 

<.001). Low cost sharing was associated with greater probability of glycemic control (adjusted risk ratio [aRR], 1.14; 95% CI, 1.09-1.20; 

<.0001). Glycemic control was more likely for patients in areas with median household income greater than $60,000 versus less than $40,000 (aRR, 1.16; 95% CI, 1.07-1.25; 

<.01) and greater than $80,000 versus less than $40,000 (aRR, 1.18; 95% CI, 1.06-1.32; 

 We found a statistically significant correlation between cost sharing for testing strips and better A1C control for patients using insulin medication. Lower cost sharing for testing strips can remove a barrier to diabetes self-management and may lead to improved glycemic control at the population level. Future efforts should study the potential benefits of reducing diabetic complications and associated cost savings.

Lower cost sharing for blood glucose testing strips can remove a barrier to self-management of diabetes and potentially improve glycemic control.

Patients with low cost sharing were likely to achieve glycemic control 58.1% of the time versus 50.3% for those with high cost sharing.

Despite filling more testing strips than patients with high cost sharing, those with low cost sharing had lower out-of-pocket costs.

Given these results and the benefits of glycemic control in patients with diabetes, payers may want to consider placing glucose testing strips in lower cost-sharing tiers and including monitoring of testing strip fills as part of the care management process.

Diabetes management aims to maintain glycated hemoglobin (A1C) levels as close to the normal range as possible, and self-monitoring of blood glucose (SMBG) levels is an important tool to help patients achieve this goal.

 Based on existing evidence describing the effectiveness of SMBG for managing glucose levels in patients with insulin-dependent diabetes,

 the American Diabetes Association and others recommend SMBG at least 3 times daily.

 However, up to 23% of patients with insulin-dependent diabetes do not regularly monitor their glucose levels.

A recent claims analysis found that glucose test utilization varied according to patients’ type of health insurance, suggesting that differences in coverage and reimbursement affected SMBG.

 Out-of-pocket (OOP) costs for testing supplies, as an additional financial burden for patients with diabetes who may be taking multiple medications that are each subject to cost sharing, have been shown to be an important barrier to regular SMBG practices.

 However, a large cross-sectional study in Canada reported that implementing quantity limits for reimbursement of testing strips was not related to worsened short-term clinical outcomes.

 That study did not separately analyze the impact on insulin users.

Other than this study, few studies have quantified the relationship between cost sharing for testing strips and A1C values, most likely due to a lack of available data sources that include cost sharing information and A1C levels. Published cost analyses have focused on utilization of testing supplies and associated costs.

 The few studies that have focused on A1C levels were based on small patient samples.

 Our study evaluated the relationship between cost sharing for diabetic testing supplies and A1C levels among patients using insulin, because patients on insulin are more likely to benefit from SMBG than patients using oral antidiabetic drugs (OADs).

 We hypothesized that patients with low cost sharing for testing strips would be more likely to attain glycemic control.

Administrative medical and pharmacy claims data integrated with laboratory test results were queried from the HealthCore Integrated Research Environment, a repository of fully adjudicated claims data for approximately 25 million Blue Cross and Blue Shield health plan members across the United States. The A1C laboratory values were obtained from national reference laboratories. All study data were kept de-identified to safeguard patient confidentiality; researchers only accessed a limited dataset, devoid of individual patient identifiers, which exempted this study from investigational review board review per 45 CFR 164.514(e), Other Requirements Relating to Uses and Disclosures of Protected Health Information.

This retrospective cohort analysis focused on patients with diabetes using insulin. The study population consisted of commercially insured patients with 2 years of continuous medical and pharmacy eligibility during the study period (January 1, 2009, through December 31, 2013). Patients with diabetes were identified if they had at least 1 medical claim for 

International Classification of Diseases, Ninth Revision, Clinical Modification 

Patients were identified as new users of diabetic testing strips if they had at least 1 testing strip prescription filled in the intake period from January 1, 2010, through December 31, 2012, with no prior fill within the previous year. As our study aim was to examine the impact on glycemic control of low versus high cost sharing for testing strips vis-à-vis initiation of testing strip use, we employed the new user study design for 2 reasons.

 First, prevalent users of testing strips are “survivors” of early SMBG and their inclusion could introduce bias, as risk for outcome (glycemic control) varies with time. Second, covariates for testing strip users at study entry could plausibly be affected by the use of the testing strips themselves. Because observational studies cannot exhaustively identify and adjust for all possible covariates, inclusion of prevalent users of testing strips could introduce confounding. Patients were required to have at least 1 year of continuous medical and pharmacy eligibility both before (baseline) and after (follow-up) the initial testing strip fill. Additional age- (18-75 years) and condition-specific inclusion and exclusion criteria were adapted from the Comprehensive Diabetes Care guidelines from the National Committee for Quality Assurance (NCQA) (n = 164,456).

Among these patients, those with at least 1 pharmacy claim with an insulin medication fill during the study period and an insulin fill within 30 days from the start date of testing strip fill (index date) were identified (n = 30,445). Patients were required to have at least 1 A1C test result within the 1-year follow-up (postindex) period as the primary measure of glycemic control. A total of 7155 (23.5% of 30,445) patients met these criteria, forming the final study sample.

Patient cost sharing for blood glucose testing strips was calculated as the OOP cost percentage of total testing strip costs, created by dividing OOP costs by total testing strip costs and multiplying by 100%. The OOP cost was the sum of co-pays, coinsurances, and deductibles paid by the patient over the 1-year period from the first testing strip fill. The total testing strip cost was the sum of OOP and health plan—paid amounts over the same period. We chose to use cost-sharing percentages rather than dollar amounts because the latter would reflect more on adherence differences instead of benefit design differences, as patients with more testing fills (better adherence) are likely to pay more out of pocket than those with fewer fills (lower adherence).

Using the median cost-sharing percentage (20%) in our final sample, we generated 2 study groups: the low cost-sharing group (OOP cost percentage below the median) and the high cost-sharing group (OOP cost percentage at or above the median).

Existing comorbid conditions and resource utilization were assessed during the baseline period; A1C outcomes, utilization, and cost measures were assessed during the follow-up period. The primary clinical outcome was the most recent A1C laboratory value during the follow-up period. The number and percentage of patients achieving glycemic control per NCQA guidelines (ie, A1C <8.0%) were analyzed. For the subset of patients whose baseline lab results were available, we also reported their baseline A1C lab values.

Descriptive statistics, including means (± SD) and frequencies, were reported for continuous and categorical data, respectively. Differences in descriptive characteristics between the low versus high cost-sharing groups were assessed with Pearson’s χ

 tests or nonparametric tests for numeric data. All statistical analyses were conducted with SAS 9.4 software (SAS Institute; Cary, North Carolina). Alpha was set at 0.05.

The association between cost sharing and glycemic control was first examined in the aforementioned descriptive (bivariate) analysis, followed by multivariable analysis in all patients and then a sensitivity analysis in a subset of patients. In the multivariable analysis in all patients, the likelihood of patients achieving glycemic control in the low cost-sharing group was compared with that in the high cost-sharing group through relative risk ratios, estimated from the modified Poisson model with the associated standard errors obtained from the sandwich method.

 Patient characteristics, such as age, gender, and area median income, were adjusted for in the multivariable model. In the sensitivity analysis, among the subset of patients with available baseline A1C values, we modeled the change in the proportion of patients achieving glycemic control before and after initiation of testing strip use with a difference-in-differences (DID) analysis through generalized estimating equations.

As an exploration of the impact of testing strip use on glycemic results, we examined the proportions of patients achieving glycemic control in groups with different numbers of testing strip fills. This was conducted in the subset of patients with available baseline A1C values.

A total of 7155 patients were included in the analysis: 3575 in the low cost-sharing group and 3580 in the high cost-sharing group (

). The mean age of patients in the 2 groups was similar (49 years and 50 years, respectively; 

<.01), although a higher proportion of those in the high cost-sharing group were aged 45 to 64 years (67.3% of high cost-sharing vs 58.7% of low cost-sharing). The low cost-sharing group resided in areas with a higher median household income (based on the zip codes for the areas in which the patients resided) compared with the high cost-sharing group (mean = $56,473 vs $53,025, respectively; 

Baseline use of medications for metabolic disorders was high in both groups (74.9% high and 72.1% low cost-sharing; 

<.05). A greater proportion of patients with high cost sharing had baseline OAD use compared with those with low cost sharing (57.4% high vs 48.6% low cost sharing; 

Testing Strip Utilization and Other Postindex Characteristics

Testing strip utilization, as measured by the mean number of glucose testing strip fills during follow-up, was higher in the low cost-sharing group than in the high cost-sharing group (5.7 fills vs 5.0 fills, respectively; 

). More patients with low cost sharing had 4 or more testing strip fills compared with those with high cost sharing (65.2% vs 55.1%, respectively; 

<.001). Not surprisingly, patients in the low cost-sharing group also had more days’ supply of testing strips (mean = 200 days in low vs 152 in high cost-sharing group; 

<.001) and more testing strips dispensed (mean = 1054 in low vs 604 in high; 

<.001). Although the low cost-sharing group used more testing strips than the high cost-sharing group, their mean OOP costs were significantly lower ($89 vs $225, respectively; 

<.001). During follow-up, the mean number of insulin fills was similar between the groups (9.1 vs 8.9 fills, respectively; 

From descriptive (bivariate) analysis, in the year after initial testing strip use, patients in the low cost-sharing group achieved a higher rate of glycemic control compared with those in the high cost-sharing group (58.1% vs 50.3%, respectively; 

In the multivariable analysis, after adjusting for patient characteristics, such as age, comorbidities, and area median income, low cost sharing was associated with a greater probability of achieving glycemic control (adjusted risk ratio [aRR], 1.14; 95% CI, 1.09-1.20; 

Patients with diabetes receiving insulin treatment with lower cost sharing for blood glucose testing strips were more likely to achieve glycemic control than those with higher cost sharing.

Andrea DeVries, PhD

Articles

Disparities in Health Care Use Among Low-Salary and High-Salary Employees

Leveraging Benefit Design for Better Diabetes Self-Management and A1C Control

Impact of Medical Homes on Quality, Healthcare Utilization, and Costs