Coverage Expansions and Utilization of Physician Care: Evidence From the 2014 Affordable Care Act and 1966 Medicare/Medicaid Expansions

Abstract

Objectives. To evaluate the effects of the 2 major coverage expansions in US history—Medicare/Medicaid in 1966 and the Affordable Care Act (ACA) in 2014—on the utilization of physician care.

Methods. Using the National Health Interview Survey (1963–1969; 2011–2016), we analyzed trends in utilization of physician services society-wide and by targeted subgroups.

Results. Following Medicare/Medicaid’s implementation, society-wide utilization remained unchanged. While visits by low-income persons increased 6.2% (P < .01) and surgical procedures among the elderly increased 14.7% (P < .01), decreases among nontargeted groups offset these increases. After the ACA, society-wide utilization again remained unchanged. Increased utilization among targeted low-income groups (e.g., a 3.5-percentage-point increase in the proportion of persons earning less than or equal to 138% of the federal poverty level with at least 1 office visit [P < .001]) was offset by small, nonsignificant reductions among the nontargeted population.

Conclusions. Past coverage expansions in the United States have redistributed physician care, but have not increased society-wide utilization in the short term, possibly because of the limited supply of physicians.

Public Health Implications. These findings suggest that future expansions may not cause unaffordable surges in utilization.

Randomized trials^1,2 and observational studies^3,4 indicate that individuals with comprehensive health insurance coverage use more physician care than do uninsured persons or those whose coverage requires higher copayments or deductibles. As a consequence, some economic analyses project that implementing universal coverage in the United States—where 29 million remain uninsured—would cause a costly surge in patients’ use of care.⁵ Yet coverage expansions might have smaller, or even negligible effects on society-wide utilization (and, hence, cost) if they produce offsetting “indirect effects” on persons whose coverage remained unchanged.^6,7

Studies of universal coverage’s implementation in England and Wales in 1948⁸ and Quebec in 1970⁹ found such offsets: physician visits increased among lower-income groups who gained coverage but fell among higher-income individuals. In Quebec, population-wide visit rates did not change.⁹ In the United States, many analyses indicate that large-scale coverage expansions had the “direct effect” of increasing health care utilization by targeted groups,^10,11 but relatively fewer have explored indirect effects on those whose coverage remained unchanged,^6,7,12,13 or on the society-wide utilization of physician care.

Some analyses have examined how Medicare’s 1966 implementation affected hospital utilization,^14,15 including a recent study that identified offsetting indirect effects,¹⁵ but few have delineated Medicare’s effect on physician utilization. Decades-old tabulations of doctor visits lack formal statistical analyses or controls for changing demographics.^16,17 Studies of more recent expansions have come to mixed conclusions regarding indirect effects. Two recent analyses found evidence of indirect effects on utilization by continuously insured individuals after pre–Affordable Care Act (ACA) Medicaid expansions,^12,13 but another found no indirect effect from the ACA’s Medicaid expansion, although it focused on only 1 segment of the already-insured population.¹⁸

We studied the utilization effects of the implementation of Medicare and Medicaid on July 1, 1966, and the ACA on January 1, 2014. We had 3 aims: (1) to assess trends in overall society-wide utilization of physician services in the short run, when growth in physician supply is limited; (2) to explore the effect of these expansions on the distribution of use by analyzing both “direct” effects on groups whose coverage was most affected by the expansions and “indirect” effects on groups whose coverage was unaffected; and (3) to determine whether indirect effects were associated with perceptions of reduced access after the 2014 expansion, especially among the nontargeted population.

METHODS

For both the 1966 Medicare/Medicaid and the 2014 ACA expansions, we analyzed the National Health Interview Survey (NHIS), which collects nationally representative data on utilization of physician services visits and surgical procedures by the noninstitutionalized US population.¹⁹

For the 1966 expansion, we analyzed 3 years of NHIS data before and after implementation. The study population included children and adults. The NHIS transitioned from a fiscal to a calendar year survey during the study period; preyears were fiscal years 1964 (i.e., July 1963–June 1964) through 1966, and postyears were fiscal years 1967 and 1968 and calendar year 1969. Only 1 year of preimplementation physician-visit microdata was available for analysis; the NHIS did not collect visit data for 1965, and the fiscal 1966 doctor visit file has been lost. For 1966, we plotted published tabulations for visits January through June of 1966 from the subsequently lost file,¹⁶ but could not analyze visits further. Surgical procedure data were available for all years. We excluded 345 individuals with weight equal to 0.

For the 2014 ACA expansion, we analyzed “sample adults” in the 2011–2016 NHIS. We also used the NHIS to examine patients’ perceptions of supply strain before and after the ACA’s implementation. Of 205 493 adults, 3323 had missing data on office visits and 2709 on surgical procedures.

Supplementary Analyses

For supplementary analyses, we used physician surveys to produce estimates of visit rates that, unlike the NHIS, rely on physicians’ contemporaneous recordings rather than patients’ recall, and that include visits provided to institutionalized patients. For the 1966 expansion, we used a report from the National Disease and Therapeutic Index (NDTI),²⁰ a national survey of randomly selected private physicians. For the 2014 expansion, we analyzed data from the 2011–2015 National Ambulatory Medical Care Survey (NAMCS), which began in 1973. Data from a companion survey to the NAMCS, of hospital outpatient departments, has not been released since 2011. We excluded community health center visits for 2011 because the NAMCS excluded them in subsequent years. We also used NAMCS to assess trends in physician-reported metrics of supply strain.

Primary Analysis Plan

The 1966 coverage expansion.

We computed annual doctor visit and surgical procedure rates (Appendix, note 1, available as a supplement to the online version of this article at http://www.ajph.org) before and after the implementation of Medicare/Medicaid, both society-wide and among 3 overlapping groups that the expansion targeted: (1) the elderly (age > 64 years), (2) low-income persons of any age, and (3) the low-income elderly. We also created a “combined” subgroup that included persons aged older than 64 years or those with low incomes of any age. We defined low income as being in the bottom third of the income distribution (Appendix, notes 2 and 3). We adjusted doctor-visit rates for small changes in NHIS methodology (Appendix, note 4).

We used univariate linear regressions to test the significance of pre–post expansion utilization changes. We then performed multivariable linear regressions to assess overall utilization changes adjusted for age, sex, family size, region, race, marital status, education, employment, income category, and a pre–post expansion dummy variable (Appendix, notes 5 and 6).

To assess whether the expansion had a differential effect on target versus nontarget populations, we performed multivariable difference-in-difference (DiD) analyses, which additionally adjusted for the interaction between the target population indicator and the pre–post dummy. Linear trends in surgical procedure rates were inspected and appeared parallel among subgroups before the implementation of Medicare/Medicaid.

The 2014 coverage expansion.

Using the 2011–2016 NHIS, we tabulated the number of physician or other health care practitioner visits (hereinafter, “doctor visits”) and surgical procedures for adults each year. The NHIS reports annual visit numbers in categories, which we converted into counts²¹ (Appendix, note 7). We calculated annual visit and procedure rates and the proportion of persons with 1 or more visits, both society-wide and for each of 3 overlapping targeted groups: (1) adults aged 18 to 64 years with incomes less than or equal to 138% of the federal poverty level (FPL; the new threshold for Medicaid eligibility set by the ACA), (2) nonelderly adults with incomes less than or equal to 250% of the FPL (the threshold for cost-sharing reductions), and (3) nonelderly adults with incomes less than or equal to 400% of the FPL (the threshold for premium subsides). The nontarget populations consisted of adults with family income greater than 400% of FPL, or those aged older than 64 years of any income. We also calculated uninsurance rates for each subgroup before and after reform (insurance data were available in only a single year of our Medicare-era data; hence, we provide only some published coverage estimates for that era).

As with the 1966 expansion, we tested the significance of pre–post expansion changes by using univariate linear regressions, and used multivariable linear regression models adjusted for age, sex, race, education, marital status, family size, region, health status, family income, employment status, and a pre–post dummy variable (see Appendix, notes 8 and 9, for details). We also examined changes within each region (Northeast, Midwest, South, West), because a disproportionate number of Southern states rejected the ACA’s Medicaid expansion, and repeated our adjusted analyses with a region by pre–post expansion dummy interaction term.

To assess whether utilization trends differed for target and nontarget populations, we performed adjusted multivariable DiD models that included an interaction term between 1 of 3 target population indicators and the pre–post dummy variable, with separate models for each target population specification (Appendix, note 9). Visual inspection suggested that linear trends for targeted and nontargeted populations were parallel before 2014.

We also used the NHIS to assess trends in the proportion of individuals with 4 patient-reported indicators of supply strain: difficulty making a timely appointment, trouble finding a general doctor, long waits in a doctor’s office, and being told that an office or clinic would not accept them as a new patient (Appendix, note 7). For each indicator, we produced adjusted estimates from logistic regression models for the overall population and for targeted and nontargeted subgroups (Appendix, note 9).

For ACA-era analyses, we used the NHIS multiple imputed data files for income (and employment). However, because imputation may introduce bias (particularly in analyses stratified by an imputed variable), we performed sensitivity analyses restricted to respondents with nonimputed income data.

Supplementary Analyses—Plan

For our supplementary analysis of the 1966 expansion, we used published NDTI visit estimates²⁰ and population estimates from the Census Bureau²² to calculate per-capita physician visit rates for July through December 1965 and July through December 1966, which we annualized.

For our supplementary analysis of the 2014 expansion, we tabulated office visits for each year from the NAMCS, and calculated per-capita rates using Census Bureau population estimates.²³ We also assessed pre–post expansion changes in 3 physician-reported indicators of supply strain included in NAMCS: wait time for an appointment greater than 1 month, mean visit length, and not accepting new patients (Appendix, note 7). We used logistic regressions to evaluate changes in the proportion of physicians with long wait times or not accepting new patients, adjusted only for specialty, and linear regressions to assess changes in visit length, controlling for patient age, sex, race, chronic conditions, physician specialty, and whether it was an initial visit (Appendix, note 9).

We performed all analyses with Stata/SE 15.1 (StataCorp LP, College Station, TX) procedures appropriate for complex survey designs (and multiple imputed data), together with person-level weights (for the Medicare-era NHIS) and sample-adult weights (for the ACA-era NHIS) to generate nationally representative estimates.

RESULTS

We present results for the 1966 coverage expansion followed by those for the 2014 expansion.

1966 Expansion

The 1964–1969 NHIS surveys included 812 673 individuals. Appendix Table A presents pre- and postexpansion characteristics of respondents. As compared with the pre-expansion period, respondents in the postexpansion period were slightly older, better educated, and higher income; they were also less likely to be female or living in large families. Published figures demonstrate increasing insurance coverage during the 1960s, with an especially sharp increase among the elderly. For instance, surgical insurance coverage rose from 62.0% in 1962–1963 to 79.4% in 1968, but from 37.1% to 94.6% among the elderly.²⁴

Figure 1 presents society-wide utilization trends for the years bracketing each coverage expansion; Table 1 presents detailed pre- and postexpansion estimates. Society-wide utilization changed little after the implementation of Medicare/Medicaid. The visit rate was 426.5 visits per 100 persons in years before implementation, and averaged 424.8 visits per 100 persons in the 3 years after implementation (P = .69). Surgical procedure rates fell slightly after 1966, from 7.0 to 6.7 per 100 persons (P < .01). Adjusted analyses showed similar small reductions in visits and procedures.

FIGURE 1—

Trends in the Society-Wide Utilization of Physician Services in the Periods Before and After the 1966 (Medicare/Medicaid) and 2014 (Affordable Care Act) Coverage Expansions by (a) Doctor Visits and (b) Surgical Procedures: United States

Note. ACA = Affordable Care Act; NAMCS = National Ambulatory Medical Care Survey; NDTI = National Disease and Therapeutic Index; NHIS = National Health Interview Survey. Dashed line indicates date of coverage expansion implementation—for the Medicare/Medicaid expansion, this was July 1, 1966; for the ACA, this was January 1, 2014. The NAMCS survey excludes visits to hospital outpatient departments. For 2011–2016, our NHIS estimates included visits to other health care professionals but excluded visits by children.

Source. Authors’ analyses of the NHIS, NAMCS, and ACA; NDTI rate was calculated by using a published report.²⁰

TABLE 1—

Overall Utilization of Physician Services Before and After the July 1, 1966, Implementation of Medicare/Medicaid and the January 1, 2014, Implementation of the Affordable Care Act: United States

		Unadjusted Results				Adjusted Results
	No.	Before	After	Difference	P	Effect Estimate (95% CI)	P
1966 expansion
No. office visits/100 persons	534 274	426.5	424.8	−1.72	.69	−15.7 (−25.0, −6.5)	.001
No. surgical procedures/100 persons	812 328	7.0	6.7	−0.28	.004	−0.21 (−0.40, −0.01)	.036
2014 expansion
No. office visits/100 persons	202 170	372.4	372.1	−0.2	.93	−1.0 (−6.1, 4.1)	.71
≥ 1 office visits last y, %	202 170	80.6	82.2	1.68	< .001	1.3 (0.8, 1.8)	< .001
No. surgical procedures/100 persons	202 784	16.1	15.9	−0.2	.5	−0.3 (−0.9, 0.4)	.39

Note. CI = confidence interval. See Methods as well as Appendix notes 5, 6, 8, and 9 for details on covariate treatment and model specification.

Source. Authors’ analysis of the National Health Interview Survey. 1966 expansion data are from various years: for office visits, the “before” year is 1963–1964 and “after” years are 1966–1967, 1967–1968, and 1969; for surgical procedures, “before” years are 1963–1964, 1964–1965, and 1965–1966, and “after” years are 1966–1967, 1967–1968, and 1969. 2014 expansion data are from the National Health Interview Survey 3 years before (2011–2013) and after (2014–2016) the Affordable Care Act.

Figure 1 also provides pre- and post-Medicare visit rates calculated from the NDTI report. As expected, these rates (which include institutional visits) were higher than our NHIS figures, but similarly changed little, from 663.5 visits per 100 persons before Medicare to 647.3 per 100 after.

Appendix Figures A and B display annual trends in utilization for subgroups targeted by the 1966 Medicare/Medicaid expansion; Table 2 presents detailed pre–post estimates. While overall utilization remained stable, groups targeted by the expansion experienced increased utilization, with offsetting reductions among others. Doctor visit rates for low-income persons (of all ages) rose by 22.3 per 100 persons after the expansion (P < .01), although they did not change significantly among the elderly or among the combined target population. Surgical procedure rates, meanwhile, increased by 0.95 per 100 among elderly persons (P < .01), and by 0.98 per 100 among low-income elderly persons (P = .01), although not among low-income persons of all ages. By contrast, visits fell by 23.5 per 100 persons in the top income tertile (P = .01), while the surgical procedure rate fell by 0.40 per 100 persons in those aged younger than 65 years (P < .001), by 0.41 per 100 persons in the middle-income group (P = .01), by 0.34 per 100 persons in the top-income group (P = .05), and by 0.46 per 100 persons in the combined nontarget population (P < .001).

TABLE 2—

Utilization of Physician Services Before and After the July 1, 1966, Implementation of Medicare/Medicaid by Age and Income Groups: United States

		Unadjusted Results				Adjusted Results
	No.	Before	After	Difference	P	DiD Estimate (95% CI)	P
Doctor visits/100 persons
By age, y
< 65	485 292	406.5	407.7	1.1	.79	0 (Ref)
≥ 65	48 982	624.9	590.8	−34.1	.09	−26.8 (−67.6, 14.0)	.2
By income
Bottom tertile	174 636	358.0	380.2	22.3	.002	52.1 (28.6, 75.5)	< .001
Middle tertile	171 822	433.0	428.0	−5.0	.55	15.0 (−10.1, 40.1)	.24
Top tertile	160 157	501.5	478.0	−23.5	.01	0 (Ref)
By income, among the elderly
Bottom tertile	24 500	561.9	585.3	23.4	.34	105.1 (−10.0, 220.1)	.07
Middle tertile	11 545	671.4	598.4	−73.1	.07	11.7 (−125.2, 148.6)	.87
Top tertile	9119	717.4	639.4	−78.0	.15	0 (Ref)
By combined target status: age ≥ 65 y or bottom income tertile
Not target	311 315	449.8	441.5	−8.3	.18	0 (Ref)
Target	199 118	399.3	407.5	8.2	.28	26.9 (6.4, 47.4)	.011
Surgical procedures/100 persons
By age, y
< 65	738 208	7.05	6.65	−0.40	< .001	0 (Ref)
≥ 65	74 120	6.50	7.45	0.95	.002	0.90 (0.31, 1.49)	.003
By income
Bottom tertile	270 464	6.73	6.72	−0.01	.95	0.36 (−0.09, 0.81)	.12
Middle tertile	255 453	7.50	7.09	−0.41	.008	−0.13 (−0.59, 0.33)	.58
Top tertile	246 028	6.92	6.58	−0.34	.046	0 (Ref)
By income, among the elderly
Bottom tertile	36 833	5.70	6.68	0.98	.005	−0.41 (−1.81, 0.99)	.57
Middle tertile	17 368	7.27	7.55	0.28	.64	−1.07 (−2.79, 0.65)	.22
Top tertile	14 327	7.68	9.11	1.43	.032	0 (Ref)
By combined target status: age ≥ 65 y or bottom income tertile
Not target	469 786	7.20	6.74	−0.46	< .001	0 (Ref)
Target	307 751	6.79	6.91	0.11	.45	0.51 (0.16, 0.85)	.005

Note. CI = confidence interval; DiD = difference-in-difference. Adjusted models included variables for age, sex, family size, region, race, marital status, education, employment, income category, and a pre–post 1966 expansion dummy (Appendix, note 5). Adjusted DiD models “by age” also included an age-65-years-and-older dummy variable and an interaction term between this variable and the pre–post expansion dummy variable; these models excluded the continuous age variable. Adjusted DiD models “by income” also included the income tertile variable and an interaction term between this variable and the pre–post 1966 expansion dummy variable; these models excluded the categorical income covariate (see Appendix, notes 5 and 6, for details on models and treatment of covariates). A total of 345 individuals with weight = 0 were excluded from all analyses.

Source. Authors’ analysis of the National Health Interview Survey for various years. For office visits, the “before” year was 1963–1964 and “after” years were 1966–1967, 1967–1968, and 1969. For surgical procedures, “before” years were 1963–1964, 1964–1965, and 1965–1966, and “after” years were 1966–1967, 1967–1968, and 1969.

The findings of the adjusted DiD models were consistent (Table 2). Low-income individuals increased their visits after the 1966 expansion compared with high-income individuals while surgical procedure rates rose for the elderly relative to younger individuals. The combined target population had an increase in both visits and procedures compared with the nontarget population. Other comparisons were nonsignificant.

2014 Expansion

Utilization.

The 2011–2016 NHIS included 205 493 adults; their pre- and postexpansion characteristics are presented in Appendix Table B. Compared with the pre-ACA period, post-ACA respondents had higher incomes and were slightly older, less likely to be White, better educated, and in better health. Coverage rates increased overall (Appendix Table C), most sharply among low-income groups. For instance, after the ACA’s implementation, the uninsurance rate fell from 40.1% to 26.7% among nonelderly adults earning less than or equal to 138% of the FPL, while the rate among those in the nontarget population (i.e., those aged ≥ 65 years or with a family income > 400% of the FPL) fell from 3.5% to 2.6%.

Society-wide utilization did not change in the wake of the ACA (Table 1 and Figure 1). Visit rates remained stable (372.4/100 persons before ACA vs 372.1/100 after ACA; P = .93) as did surgical procedure rates (16.1/100 before ACA vs 15.9/100 after ACA; P = .50). However, the proportion of persons with 1 or more visits in the past year rose by 1.68 percentage points (P < .001). Multivariate analyses adjusting for demographic changes yielded virtually identical results for the overall populations. Trends differed by region; adjusted results showed small increases in visits in the West relative to the South (Appendix Table D).

Figure 1 also presents our supplementary analysis of visit rates to physicians’ offices (i.e., excluding visits to hospital outpatient departments and community clinics) based on the NAMCS. Like the NHIS, this NAMCS data showed no increase in total office visits after the ACA’s implementation, with rates falling slightly from 299 to 293 per 100 persons (data not shown).

Table 3 and Appendix Figure C present utilization before and after ACA implementation by target-population status. As in the Medicare era, the target and nontarget groups experienced somewhat different trends. All targeted groups had statistically significant increases in the share with 1 or more annual visits. For instance, nonelderly adults with incomes less than or equal to 400% of FPL had a 2.25-percentage-point increase in the proportion with 1 or more visits in the past year (P < .001). Targeted groups also had nonsignificant increases in visit and procedure numbers. Increases were largest in the lowest income subgroup (≤ 138% FPL) compared with the other targeted subgroups. By contrast, the nontarget group had nonsignificant trends toward fewer office visits (P = .06) and surgical procedures (P = .08), but no change in the proportion with 1 or more visits.

TABLE 3—

Utilization of Physician Services in the 3 Years Before and After the January 1, 2014, Implementation of the Affordable Care Act (ACA) According to Target Population Status: United States

		Unadjusted Results				Adjusted Resultsb
	No.a	Before	After	Difference	P	DiD Estimate (95% CI)	P
No. office visits/100 persons
≤ 138% FPL	41 608	350.7	361.1	10.3	.1	12.5 (−0.6, 25.5)	.06
≤ 250% FPL	71 540	334.4	339.9	5.6	.22	11.6 (0.8, 22.4)	.036
≤ 400% FPL	102 089	330.6	333.0	2.4	.53	10.0 (0.0, 20.1)	.05
Not target	100 081	416.0	408.7	−7.3	.06	0 (Ref)
≥ 1 office visits, %
≤ 138% FPL	41 608	69.5	73.0	3.50	< .001	2.7 (1.4, 4.0)	< .001
≤ 250% FPL	71 540	70.2	73.6	3.36	< .001	2.8 (1.8,3.8)	< .001
≤ 400% FPL	102 089	73.1	75.3	2.25	< .001	1.9 (1.0, 2.8)	< .001
Nontarget	100 081	88.4	88.7	0.35	.23	0 (Ref)
No. surgical procedures/100 persons
≤ 138% FPL	41 766	13.2	14.2	1.0	.11	1.7 (0.1, 3.2)	.032
≤ 250% FPL	71 756	12.8	13.5	0.6	.19	1.4 (0.1, 2.7)	.035
≤ 400% FPL	102 343	13.1	13.2	0.1	.77	1.0 (−0.2, 2.2)	.09
Nontarget	100 441	19.4	18.5	−0.9	.08	0 (Ref)

Note. CI = confidence interval; DiD = difference-in-difference; FPL = federal poverty level, based on US Census thresholds from the year an individual was surveyed. Affordable Care Act target population = those aged 18–64 y with family income as a percentage of the FPL below threshold; not-target population = those aged ≥ 65 y of any income or aged 18–64 y with income > 400% FPL.

Source. Authors’ analysis of the 2011–2016 National Health Interview Survey.

^aAnalyses used multiply imputed income data, which resulted in slight variations in the number of observations around the no. shown in the table.

^bAdjusted for age, sex, family size, region, race, marital status, health status, education, pre–post ACA dummy variable, employment status, ACA target population status, and ACA target population status*pre–post ACA dummy variable interaction term. See Appendix notes 8 and 9 for details on treatment of covariates.

Adjusted DiD analyses (Table 3) indicated that the ACA affected target and nontarget groups differently. The proportion with 1 or more visits rose faster for target versus nontarget populations (e.g., by 1.9 percentage points; 95% confidence interval [CI] = 1.0, 2.8 more for those with incomes below 400% of FPL as compared with the nontarget group). Similarly, office visits rose by 11.6 per 100 persons (95% CI = 0.8, 22.4) and surgical procedure rates by 1.4 per 100 persons (95% CI = 0.1, 2.7) for those below 250% of FPL relative to the nontarget group. Sensitivity analysis limited to those with nonimputed income data (Appendix Table E) yielded similar results.

Measures of supply strain.

Appendix Table F presents patient-reported metrics of pre- and post-ACA supply strain. Changes were mostly small and nonsignificant. However, the targeted population defined at the 138% FPL threshold reported a significant reduction (i.e., improvement) in the odds of “not being accepted as a new patient” or “trouble finding a general doctor.” By contrast, the nontargeted population had a small increase in “trouble finding a general doctor,” from 1.6% to 2.1% (adjusted odds ratio = 1.29; 95% CI = 1.15, 1.46; P < .001). In sensitivity analyses excluding persons whose income was imputed, the changes in “trouble finding a general doctor” remained significant (Appendix Table G); surprisingly, those analyses also showed a population-wide reduction in “waiting too long for an appointment.”

Appendix Table F also shows physician-reported indicators of supply strain, which changed little. The proportion of physicians not accepting new patients was 4.9% before versus 4.8% after ACA, and the proportion with an average wait for an appointment of 1 month or more was 7.1% before versus 8.2% after ACA; these changes were nonsignificant in adjusted analyses. Mean appointment length remained 22 minutes and also showed no change in adjusted analyses (P = .33; data not shown).

DISCUSSION

After the 1966 and 2014 US coverage expansions, targeted populations increased their use of some physician services. These direct effects are in keeping with previous studies, including the RAND Health Insurance Experiment,^2,25 the Oregon Health Insurance Experiment,¹ projections from observational studies,^3,4,26,27 and previous analyses of the ACA,^10,11 which all found that more insurance coverage increases the use of physician care.

However, we also found evidence of indirect effects of these coverage expansions, similar to studies of other domestic^12,13 and international expansions.^8,9 Utilization declined slightly among nontargeted populations, offsetting the direct effects. As a consequence, the society-wide utilization of physician services shifted, but did not increase, at least in the short-run.

Supply constraints (i.e., the limited number of hours that physicians are willing to see patients and operate) may in part explain these results. A larger supply of facilities and physicians is known to boost utilization,²⁸ a phenomenon dubbed “Roemer’s Law”²⁹ in the case of hospital care and more generally known as “supplier-induced demand.” When supply is constrained, increased use by persons with new or upgraded coverage might crowd-out utilization by others. Older reports on physician and hospital utilization in the United States,^16,20,30 Quebec,⁹ and the United Kingdom⁸ are consistent with such a redistributive effect, as is 1 study of Massachusetts’s 2006 health reform,⁶ together with a recent analysis of the effects of the 1966 and 2014 coverage expansions on hospital utilization.¹⁵

The NDTI physician survey, meanwhile, indicated that, although the elderly received more doctor visits in hospitals and nursing homes (but not offices) after Medicare’s implementation, doctor visits declined for the nonelderly.²⁰ (Of note, this pattern of utilization increases among the elderly is consistent with our finding that the elderly received more inpatient surgical procedures, but not more ambulatory visits, after Medicare’s implementation.) Similarly, 2 recent studies of pre-ACA Medicaid expansions found that increased insurance coverage produced small reductions in utilization for the already insured,^12,13 although 1 study examining the ACA’s Medicaid expansion did not.¹⁸

In light of capacity constraints, coverage expansions could produce offsetting “indirect effects” on use for the already-insured in 2 ways.³¹ If the physician supply is already strained, increased utilization by the newly insured may impede appropriate access for the already-insured (e.g., by increasing waiting times). Alternatively, if supply is ample, and providers were previously keeping their schedules full by “inducing demand” among the well-insured, increased utilization by the newly insured might be offset by reductions in unnecessary care.

A study of pre-ACA Medicaid expansions suggests that the latter explanation may now predominate in the United States; in communities with larger expansions, Medicare beneficiaries’ visit rates declined without any deleterious health effects, and patients perceived that they were receiving less unnecessary care.¹³ Similarly, although we could not assess the medical necessity of visits and procedures, we found that nontargeted groups, whose visit rates fell slightly after the ACA, mostly perceived no worsening of their access to care, a finding that mirrors a previous study.³² Although no survey data are available on perceptions of supply strain with Medicare’s start up, newspaper reports from that era suggest that the waits for medical care that many had predicted never materialized.³³ We did, however, observe a slight increase in the proportion of persons in the nontargeted population with difficulty finding a general doctor after the ACA, a finding that underscores the importance of ensuring an appropriate supply and mix of providers when expanding coverage.

Limitations

Our analysis has limitations. For both expansions, the NHIS excludes institutionalized persons and relies on respondent recall. However, our supplementary analyses of physician-reported data from the NDTI and NAMCS, which include visits of institutionalized persons, showed similar trends.

For the Medicare era, as described in Appendix, note 4, our adjustment for small changes in NHIS methodology¹⁷ used estimates that apply to population-wide utilization, but whose accuracy for subgroups is uncertain. Although the Medicare-era income variable was categorical and not inflation-adjusted, it was sufficient to allow categorization of respondents by income tertiles. Finally, our “non-target” control groups (e.g., those aged < 65 years) included some individuals who were affected by the expansion (e.g., through Medicaid), although this would bias our DiD findings toward the null.

For the ACA era, the NAMCS data exclude visits to hospital outpatient departments. However, because outpatient departments accounted for just 11.3% of all physician visits in 2011 (the last year of outpatient department data available),³⁴ only an improbably large increase in outpatient department visits could substantially change our findings. Although more detailed utilization data are available from the Medical Expenditure Panel Survey, a 2014 change in that survey’s methodology created a discontinuity in visit utilization trends that would bias analyses such as ours.³⁵

Finally, we analyzed only 3 years of data after each expansion; longer-term trends may differ, particularly if expanded coverage leads to growth in provider supply.

Public Health Implications

Our results imply that studies of coverage expansions affecting a small portion of the community (e.g., the RAND Health Insurance Experiment and Oregon Health Insurance Experiment) may not accurately predict the society-wide effects of large-scale coverage expansions. Projections of changes in utilization (and costs) for future reforms should consider the possibility that, for supply constrained services, increased use among newly covered individuals may be offset by small reductions among those whose coverage remains unchanged. Together with the experiences of Canada⁹ and the United Kingdom,⁸ our findings suggest that future coverage expansions may be achievable without an unaffordable surge in the utilization of physician care.