It is evident that lack of (or poor) insurance coverage is a barrier to access healthcare. Evidence that insurance status is linked to access to healthcare seems overwhelming: those with insurance always use substantially more than those without.
Economists tend to be more skeptical, for the following two reasons: they question the causality behind the observed link between coverage and utilization; they question the inference from differences in utilization to differences in access to care. The causality issue is currently not important and it is summarized briefly in Section Health Insurance Increases Utilization. The distinction between utilization and access is currently a matter of scientific investigation among economists and social epidemiologists, and this review of the literature will mostly focus on this issue. Section Interpreting the Causal Effect of Insurance: Moral Hazard or Access? summarizes the theoretical debate on the inference question, which can be described as follows: Is the difference in utilization resulting from insurance coverage a matter of moral hazard – the insured use more than they need – or access – the uninsured do not use what they need? It is shown that the empirical answer depends on how healthcare need is defined and measured. Sections Effect of Insurance on the Subjective Assessment of Unmet Need by Survey Respondents, Insurance and Utilization of Medically Necessary Care, and Effect of Insurance on Health Outcomes: Adverse Events and General Health and Mortality then review the empirical evidence on the impact of insurance on the utilization of care that is needed, using three different definitions of need. In Section Effect of Insurance on the Subjective Assessment of Unmet Need by Survey Respondents, a subjective definition (what is perceived as unmet need) is used; in Section Insurance and Utilization of Medically Necessary Care, a more objective definition of need as what is clinically recommended to survive or maintain good health is used; last, in Section Effect of Insurance on Health Outcomes: Adverse Events and General Health and Mortality, an outcome-oriented definition of need and evidence on the effect of lack of coverage on mortality and health status is used. Section Policy Implications concludes and draws policy recommendations.
Health Insurance Increases Utilization
The causality issue is as follows: When we observe differences across insurance it is noticed that individuals are not assigned to a given health insurance status but they make their own decisions on whether to be insured or not. Of course, these decisions are constrained, by how much individuals can spend overall compared to the price of health insurance, but, nevertheless, individuals at the same level of income and faced with the same premiums make different decisions regarding coverage (Bundorf and Pauly, 2006). If that decision is somehow linked to their utilization of healthcare services in a way that is not observed (in the survey used by the analyst), the correlation between insurance status and utilization may be spurious and it would be wrong to infer causality from it. For example, if individuals were to buy health insurance only because they wanted to commit to visit a doctor once a year, and get their tension and cholesterol checked, the correlation between insurance status and utilization of these services would be perfect. However, that would not mean that covering the uninsured would change their behavior: if the reason why they do not buy insurance is as they do not value the services it covers, they then might not be interested even if the services were free of charge at the point of use.
One way to address the issue is to run a social experiment: the health insurance experiment (HIE), conducted by the RAND Corporation randomly assigned approximately 2000 households to a variety of plans with varying cost-sharing arrangements (Newhouse and the Insurance Experiment Group, 1993). Because individuals were assigned to the plans rather than choosing them, any difference in utilization can be safely interpreted as causal. The results from that social experiment indicate a clear causality from coverage to utilization: individuals assigned to plans with lower copayments used more outpatient services, prescription drugs, and even inpatient services. The latter finding has been recently disputed by Nyman (2007), who argued that it is an artifact because of attrition (those who are poorly covered through the experiment and need hospital care quit the experiment and revert to their former plan); Newhouse et al. (2008) responded that subjects have no incentive doing that because they are more than compensated for the loss if (and only if) they stay in the experiment. It is true that the attrition rate was much higher in the higher coinsurance plan than in the free plan but it remains undecided whether subjects left the experiment (although they had no interest doing it) when in need of hospital care and not well covered (Nyman’s suggestion) or whether they left for other reasons (the HIE Group’s response to Nyman).
Beside social experiments, which are costly and constrained by ethical issues (it is not feasible to assign subjects to no coverage at all and some stop loss must be put in place, which does not allow the researchers to test the effect of not being insured), economists use a variety of econometric strategies to test causal inference in observational studies and all find a causal link from insurance status to utilization pattern.
Interpreting The Causal Effect Of Insurance: Moral Hazard Or Access?
It is evident that coverage influences utilization and it can be said that not being insured causes lower levels of utilization of healthcare services. The remaining issue is one of interpretation: Do the uninsured use less because they cannot afford the services when they are ill? Or do the uninsured buy exactly the amount of healthcare they need, whereas the insured overconsume healthcare because they do not have to pay for it at the point of use? Or is it that both interpretations are partially true: Some among the insured ‘overconsume’ and some among the uninsured cannot access the care they need. To understand the issues underlying the difference in interpretations we need to go back to the economic theory of health insurance and introduce concepts such as moral hazard. As will be clear at the end of this section, a key concept for the understanding of the access versus moral hazard controversy is the concept of need: if we could tell what is needed and what is a matter of preferences in healthcare services utilization, we could tell which part of the variation in utilization across insurance status is a problem of access for the uninsured and which is moral hazard of the insured.
Andersen (1995), and most social epidemiologists, equated access to utilization: if one uses fewer services it is because they cannot use as much. He distinguishes between ‘potential access’ (enabling factors such as availability of services, coverage, regular source of care, travel costs, and waiting time) and ‘realized access’ (actual utilization). But the economists disagree on the proposed theory. As noted by Hurley (2000), access is a process-oriented concept and is unrelated to actual use: the difference between such a conception and Andersen’s is that, for a given level of accessibility, individuals with different preferences s make different choices. For most economists, access is similar to ‘opportunity,’ and individuals are always free to use opportunities as they see fit. Some of the difference between the insured and the uninsured is a matter of access (the medical need of the uninsured is not met), and some is a matter of want (the insured use nonneeded care).
The objective is of course to evaluate the respective roles of access and want in the difference in utilization across insurance status. To do so, one needs to understand the way health insurance works and interferes with decisions made by individuals regarding their utilization of healthcare services. The following is drawn from Nyman (2003).
Although standard (nonhealth) insurance pays a lump sum in case a detrimental event occurs (life insurance pays a given sum in case the insured dies), health insurance typically pays back through reduced prices of healthcare. Being covered by health insurance, therefore, means gaining access to discounted healthcare services. Some plans have a limit on reimbursement, but most public plans do not set such limits on reimbursements for acute care (hospitalizations, visits to a family doctor, and drugs prescribed by a doctor).
As a result, insured individuals live ‘in a different world’ than the uninsured, a world with lower prices of healthcare services. Proponents of the moral hazard hypothesis posit that because the uninsured are faced with the true price of healthcare, they buy units of healthcare services until they reach a level at which the marginal value of an extra unit is less than the price they have to pay. The insured do the same, but because they face a lower price of healthcare they buy more than what would satisfy them (to be exact, what would maximize their satisfaction) if they were not insured. The analysis of health insurance is similar to the analysis of subsidies for specific goods (e.g., food): when a price is artificially lowered, individuals do not get the right information about the relative values of goods and favor the subsidized one to the detriment of nonsubsidized goods.
The economic theory of health insurance is not only about this substitution effect but also involves what economists call income effects: If we compare two individuals with the same level of income, one benefiting from a discount on the price of one specific good but not the other one, it is clear that the former has more purchasing power than the latter. In that sense, they are richer and can make the decision to allocate that extra purchasing power as they see fit. If they decide to buy more healthcare services, because they are sick and made richer by their health insurance coverage, they are not substituting away from other potential uses of their money. They make a rational decision to allocate their extra purchasing power where it is needed.
The moral hazard story goes as follows: ‘‘Being insured means I will take advantage of lower prices of healthcare to use more of them, whether I am sick or not, need it or not. It is the fact that they are cheaper than if I was uninsured that motivates me the most.’’
The income transfer story is as follows: ‘‘Being insured means that when sick and in need of care, I will be richer than if I was uninsured. I will then spend more on healthcare because this is what I need to do (I am sick) and I can afford it. It shows clearly that the ‘income effect’ is the translation in economic theory of the access problem of social epidemiology.’’
It is of course impossible to separate these two mechanisms empirically on the basis of the difference in utilization across insurance status: they both predict the exact same difference in utilization.
The only notion being observed that would allow to separate the two mechanisms is ‘need’: Recall that the income effect occurs because the insured benefits from an income transfer when sick, whereas the substitution effect is independent of health states. One useful way to look at access versus moral hazard would, therefore, be to look at the differential effect of coverage on care that is ‘needed’ versus care that one could go without.
So far, we have only moved the question one step further and still need to define what ‘needed’ means in healthcare. As shown by Culyer (1998) and the literature on equity in healthcare utilization, need is an elusive concept, and it is impossible to provide a theoretical definition of need that would satisfy most. Rather, need is defined as how it is measured in empirical studies.
How do we measure need? Here, three ways of defining needed care are suggested:
- Subjective: Do they feel they could not access care they needed?
- Objective, process-oriented: Needed care is the type of care that is clinically necessary to maintain health.
- Objective, outcome-oriented: Access barrier can be inferred from lower utilization if and only if lack of coverage causes poorer health outcomes.
These questions were investigated in the RAND HIE: the objective was not only to measure the causal link between coverage and utilization but also to describe which services were underused by the less well covered (or overused by the better covered) and to measure the impact of being less well covered on health (a 2–4 years follow-up was included in the experiment). It is very often stated that the RAND shows a strong difference in utilization as a result of differences in coverage but no difference at all in health outcomes. Some use that often stated conclusion to infer that 100% of the difference in utilization is because of moral hazard and nothing to access problems. Interestingly, this is not the interpretation of the HIE group members themselves: first, they show that the insured utilize more of both clinically recommended and futile care than the uninsured, implying that the difference is due in part to both access problems and moral hazard. Second, they observe that in some groups (the poor and the sick) being less well covered has consequences on health. However, the effect is offset on average because the better covered also seem to suffer (surprisingly) from ‘too much healthcare.’ The combination of these two effects is the often cited ‘no effect on health’ but the RAND experiment itself does not conclude to the absence of a link between being less well covered and deteriorating health. In a sense, there must be an effect because one of the result of the RAND is that those in the plans with higher copayments used less inpatient care, and it is hard to imagine that the better covered would be admitted to a hospital to receive treatments with absolutely no effect on their health, simply for the sake of staying in a hospital.
Effect Of Insurance On The Subjective Assessment Of Unmet Need By Survey Respondents
A simple way to assess needed care is to directly ask respondents of a survey to state whether they had to forgo care they needed in the recent past (typically 12 months). The price to pay for such simplicity is the subjective component of the perception of need: if subjective perceptions of need correlate in a systematic way with decisions not to buy insurance, the value of such subjective assessment is low. Also, it must be noted that unmet needed care can be the result of many factors beyond lack of insurance (lack of time, procrastination, and fear).
An idea to test a causal link between coverage and perception of unmet need that should not be affected by systematic variations in how subjective need is defined is to take advantage of exogenous changes in health insurance coverage. One such shock is the 1996 Reform of Welfare in the US that led to reductions in the caseload of the temporary assistance for needy families (TANF). Women who lost TANF also lost public health insurance after 12 months and follow ups show substantial increases in self-reported unmet need for a variety of healthcare services.
Insurance And Utilization Of Medically Necessary Care
To overcome the subjectivity of self-reported unmet need, we can define needed care as what is necessary to maintain health. A stringent definition is that care is needed if and only if not receiving it would lead to death or severe disability, and the evidence on the causal effect of coverage on utilization of such care is reviewed (see Section Care That is Needed in Life-Threatening Situations or When Quality of Life Would Be Greatly Affected without Treatment). A more lenient definition is that care is needed as long as clinical consensus is that not receiving that type of care would affect intermediary health outcomes and the evidence based on that clinical definition of need is reviewed in Section Differences in Utilization of Clinically Recommended Care.
Care That Is Needed In Life-Threatening Situations Or When Quality Of Life Would Be Greatly Affected Without Treatment
A first approach is to describe what individuals facing a health shock (an illness or injury necessitating treatment if the patient wants to recover) do when they are not covered. Most of the literature on insurance and the economic consequences of health shocks is recent and from lowand middle-income countries; the literature on health shocks in rich countries is mostly about health and labor supply, and the case of the uninsured is less often considered because in most rich countries, to the possible exception of the US, public insurance covers potentially catastrophic health shocks.
In low-and middle-income countries (Ethiopia, Vietnam, and Laos), the uninsured pay for medical care in case of health shocks necessitating catastrophic spending through informal insurance mechanisms (microfinance schemes, informal lending, or transfers), drawing from their assets and savings, or cutting back on other consumption items. The only exceptions seem to be China, where the uninsured spend less out-of-pocket than the insured in case of health shocks, and Thailand, where the poor who need treatment for end-stage renal disease use therapeutic strategies or less frequent dialysis, which have side effects but keep them alive.
In the US, bankruptcy can be used to protect assets in case of large medical bills. Approximately 1 million households filed for bankruptcy caused by medical bills in excess of US$1000 in the US in 2001. Bankruptcy is not enough, though, and 61% of them also had to cut back healthcare.
In the US, as in China, the uninsured spend less out of pocket than the insured in case of a severe health shock, suggesting that lack of insurance makes medical services less affordable and, therefore, reduces access. Of course, another way to spend less out of pocket is to receive care free of charge, through charity. It is documented that public and not-for-profit hospitals in the US deliver care free of charge to patients unable to pay for care in cases of severe illnesses and accidents.
A less stringent definition of health shocks is ‘nonavoidable hospitalizations.’ These are hospitalizations that cannot be avoided by effective, timely, and continuous outpatient (ambulatory) medical care for certain chronic conditions – they are also called admissions for non-ambulatory care sensitive conditions (non-ACSCs). Among adults, their necessary character can be disputed: for instance, a cataract excision is a non-ACSC (no primary care can really prevent cataract), can be ‘necessary’ in some cases (to cure near blindness) but can also be discretionary in other cases (when vision quality is diminished); similarly, a hip replacement can be needed (the patient cannot walk without it) or discretionary (the patient can walk but feels some pain or discomfort). However, in the case of children (younger than 15 years of age), it can be argued that what is not preventable is more likely to be needed to prevent future health problems.
A study of non-ACSC pediatric admissions from 1983 to 1996 based on the US National Hospital Discharge Survey uses exogenous expansions of the Medicaid program between 1983 and 1996 (increase in children population covered by 16% points overall but at different times in different states) to estimate a causal link, rather than a simple correlation, between Medicaid coverage and use of hospital care for non-ACSC. If utilization of non-ACSC hospitalizations increases with enrollment in Medicaid, this is an indication of a causal link between lack of coverage and difficulties to access needed care. They find that Medicaid expansions led to an increase in non-ACSC admissions: any increase in enrollment by 1% increases the probability of admission for a non-ACSC by 0.81%. Therefore, there was an access problem to inpatient care for children without insurance before the expansion. When admitted, these newly covered children also receive more procedures than when they were not covered.
Similarly, the implementation of a universal National Health Insurance for the elderly in Taiwan had a stronger effect on low and middle-income elderly than on high-income elderly individuals, suggesting that there was an access problem linked to ability to pay for treatment without insurance.
Differences In Utilization Of Clinically Recommended Care
Although ambulatory care services are less expensive, some authors consider that they are ‘needed’ when proven to be effective, in the sense that not using them negatively affects health. As a result, if the uninsured can be shown to use less preventive services than the insured that could be interpreted as a problem of access to care. What is known on the causal effect of insurance on the utilization of clinically recommended services (such as mammography) or intermediary clinical outcomes (such as blood pressure) is now reviewed.
Changes in insurance status in longitudinal studies identify both a causal effect of copayments on mammography and a causal effect of loss of coverage on postemergency room visit to an ambulatory care doctor in the US.
Levy and Meltzer (2001, 2004, 2008) reviewed studies on insurance and intermediary health outcomes. Studies testing for a causal link are selected. Some of the studies reviewed in Levy and Meltzer will be reviewed in Section Effect of Insurance on Health Outcomes: Adverse Events and General Health and Mortality (those on final outcomes such as mortality, self-assessed health, or functional ability). They show a clear effect of loss of coverage on blood pressure, but some of these studies cannot conclude at any substantial effect of coverage on intermediary health outcomes.
Effect Of Insurance On Health Outcomes: Adverse Events And General Health And Mortality
Introduction of copayments in public schemes (medication insurance for the elderly and welfare recipients in Quebec in 1996 or the California Public Employees Retirement System (CalPERS) in 2001) reduces utilization and substantially increases the probability of adverse events (more than double in Quebec).
Moving to studies testing the effect of insurance on mortality and general health; most studies do not measure the effect of insurance on utilization and infer access problems directly from detrimental effect of lack of insurance on health outcomes. Historical data (European countries in 1870–1914) show that an increase of 10% points in the proportion of population covered by health insurance led to a reduction in mortality by 0.9–1.6 per 1000. The 1.6 effect is certainly implausibly high but it should be kept in mind that expansions of coverage were usually targeted at individuals toward the lower end of the income distribution, where mortality was very high and at a time when their income did not allow them any contact with a doctor. As a result, these estimates are of effects at the maximum rate of return of coverage on access and of access on health. On the contrary, the introduction of Medicare in 1965 had no discernible effect on the change in mortality around 1970: Regions in the US with lower rates of insurance after the age of 65 years did not see any more substantial decrease in their mortality than regions with higher rates (which were less affected by Medicare as a result). The fact that Canadian provinces did not implement universal coverage at the same time (between 1962 and 1972) can be used to identify a significant effect of universal coverage: a reduction of 4% in infant mortality and of 1.3% in low birth weight.
Another approach uses the exogenous discontinuity in insurance status for most Americans when they turn 65 years: There is indeed a decrease in the mortality rate (compared to the trend before the age of 65 years) of approximately 13%, but it is hard to attribute it entirely to Medicare (Americans tend to retire at the age of 65 years as well, which can be good for health). Moreover, the effect does not vary at all across race and location or self-employed status although insurance status pre-Medicare varies substantially across these variables. A randomized trial in Oregon studies the effect of getting coverage on health outcomes (30 000 low-income individuals were randomly selected to benefit from Medicaid coverage and 10 000 applied – these are compared to similar individuals on the waiting list who were not selected) and finds an effect on self-assessed health at 1 year follow-up. The data are still under analysis and more should be known soon about objective measures such as blood pressure.
Studies using instruments (variables that affect health through insurance but are not subject to the endogeneity issue of insurance status, such as spouse’s union status, immigration status, and number of years in the US, work loss in the previous 5 years, or state-level unionization rates or Medicaid eligibility and generosity of benefits) find large and significant effects of insurance on health (self-assessed health, general mortality, and human immunodeficiency virus-acquired immunodeficiency syndrome-related mortality), but the quality of the instruments can be discussed.
One particular relationship has been studied in more detail and remains disputed in the empirical literature: the effect of insurance on infant and children health. The effect of expansions of health insurance for pregnant women, infants, and children in the 1980s (1979–92) in the US on birth outcomes and children health is estimated as strong and negative on mortality (expansions yielded a decrease in mortality by almost 40%) by Currie and Gruber (1996). However, Dave et al. (2008) rightly pointed out that this is implausibly high. Their objection is that the quasiexperiment is not methodologically sound: if some unobserved variable explains that states where efforts on public maternal and natal health were made also were those states where Medicaid expansions took off first, using eligibility by year and state will overestimate the effect of insurance on mortality).
The study of expansions in insurance for infant and pregnant women finds a weak effect on birth weight (likely because of crowding out: overall, the expansions led to only a 10% points increase in the proportion insured) but a substantial effect on infant mortality (expansions decreased it by 8.5%). Last, the study of expansions in insurance for pregnant women on infant mortality found that the effect was strong for infants whose mother lived closest to a hitech hospital. It is also found that better educated women (not dropouts or teen mothers) actually used less hitech care (notably caesarian section) after the expansions, likely because of the fact that they switched from a private insurance to Medicaid, but without any notable effect on their infant’s health (a case of futile care because of private insurance and generous coverage).
Overall, the lack of insurance increases the probability of adverse events and is the cause of poorer self-assessed health and higher infant mortality. Its effect on adult mortality and low birth weight is less clearly documented.
Policy Implications
It can be safely concluded that access problems are part of the difference in utilization across insurance status: It is not only about moral hazard and the difference also stems from the fact that the uninsured do not benefit from an income transfer when sick and, as a result, cannot access needed medical care. They access charity care if the intervention is a matter of survival or to prevent disability, delay recommended care such as follow-up after emergency admission or ambulatory care after new symptoms of a chronic condition, and are much less likely to be screened for cancer, or have their blood pressure or cholesterol measured. As a result, not being insured has consequences on health, documented by downstream adverse events, self-assessed health, and, possibly, longevity.
From a normative perspective, this means that some of the difference in utilization between the insured and the uninsured is welcome: it does not mean that the insured spend ‘too much’ on care because they are over covered but that the sick who are insured can use the income transfer they receive from the healthy to access needed care. This review shows that inpatient services that are nonavoidable and expensive should enter a universal plan; it also shows that preventive services and ambulatory care services that meet clinical recommendations should also be covered for the less well-off, who cannot afford it if not covered. There is no literature that would allow us to determine what is not affordable if not covered. It has been suggested in some countries with public insurance that affordability should be the main criterion for coverage: for instance, the ‘bouclier sanitaire’ (health shield) discussed in France in 2007–08 was a project to replace the current universal public plan with various copayments for various services (and exemptions for the chronically ill) with a full coverage plan with a deductible set at 10% of income. In Ontario, Canada, the same idea forms the basis of a tax deduction for those who have to spend more than a share of their taxable income on prescription drugs out of the pocket. The issues with such attempts at solving moral hazard (through deductible) and access (universal coverage and no copayment beyond the deductible) are threefold: first, there is no clear definition of affordability and the 10% threshold is rather arbitrary (Glied 2008); letting physicians determine what is needed without imposing any cost sharing on patients seems to be a more promising avenue to solve moral hazard and access simultaneously (the difficulty being to provide doctors with the right incentives to deliver services that are needed only). Second, the chronically ill with low or middle income will reach the deductible every year and will be penalized for being chronically ill though they are not at fault. In the US, this would prove a progress compared to the situation before the latest reform (where preexisting condition were a cause for exclusion of coverage), but in most European countries and Canada that would be a regression. Third, the deductible set at 10% of income would not address the issue of access to preventive care: in the case of preventive care, the issue does not seem to be that individuals cannot pay for it (except the very poor), but rather that the benefits (positive effect on health) accrue in the future, whereas the cost is borne immediately.
References:
- Andersen, R. M. (1995). Revisiting the behavioral model and access to medical care: Does it matter? Journal of Health and Social Behavior 36(1), 1–10.
- Bundorf, M. K. and Pauly, M. V. (2006). Is health insurance affordable for the uninsured? Journal of Health Economics 25(4), 650–673.
- Culyer, A. J. (1998). Need – Is a consensus possible? Journal of Medical Ethics 24, 77–80.
- Currie, J. and Gruber, J. (1996). Health insurance eligibility, utilization of medical care, and child health. Quarterly Journal of Economics 111(2), 431–466.
- Dave, D. M., Decker, S., Kaestner, R. and Simon, K. I. (2008). Re-examining the effects of Medicaid expansions for pregnant women. NBER Working Paper 14591. Cambridge, MA: National Bureau of Economic Research.
- Glied, S. (2008). Universal public health insurance and private coverage: Externalities in health care consumption. Canadian Public Policy 34(3), 345–357.
- Hurley, J. (2000). An overview of the normative economics of the health sector. In Culyer, A. J. and Newhouse, J. P. (eds.) Handbook of health economics, vol. 1, Part 1, ch. 2, pp. 55–118. North Holland: Elsevier.
- Levy, H. and Meltzer, D. (2001). What do we really know about whether health insurance affects health? University of Chicago School of Public Health, Unpublished document, December.
- Levy, H. and Meltzer, D. (2004). What do we really know about whether health insurance affects health? In McLaughlin, C. (ed.) Health policy and the uninsured: setting the agenda, pp. 179–204. Washington, DC: Urban Institute Press.
- Levy, H. and Meltzer, D. (2008). The impact of health insurance on health. Annual Review of Public Health 29, 399–409.
- Newhouse, J. P., Brook, R. H., Duan, N., et al. (2008). Commentary: Attrition in the RAND health insurance experiment: A response to nyman. Journal of Health Politics, Policy and Law 33(2), 295–308.
- Newhouse, J. P. and the Insurance Experiment Group (1993). Free for all? Lessons from the RAND health insurance experiment. Cambridge: Harvard University Press.
- Nyman, J. A. (2003). The theory of demand for health insurance. Stanford, CA: Stanford University Press.
- Nyman, J. A. (2007). American health policy: Cracks in the foundation. Journal of Health Politics, Policy and Law 32(5), 759–783.
- Culyer, A. J. and Wagstaff, A. (1993). Equity and equality in health and health care. Journal of Health Economics 12, 431–457.