Infant Mortality Among Supplemental Security Income Applicants

Introduction

Infant mortality (defined as the death of a child before 1 year of age) and neonatal mortality (defined as the death of an infant at 0–27 days of age) are important markers of a nation's health and well-being. Although significant intergroup disparities remain—for example, by race/ethnicity or across geographic areas (Mathews, MacDorman, and Thoma 2015; MacDorman, Hoyert, and Mathews 2013; Alexander and others 2003)—overall neonatal and infant mortality rates in the United States have steadily declined since at least the late 1960s. At the national level from 1985 to 2013, neonatal mortality decreased from 7 deaths per 1,000 live births to 4 deaths per 1,000 live births and infant mortality decreased from slightly more than 10 deaths per 1,000 live births to about 6 deaths per 1,000 live births (National Center for Health Statistics 1987; Mathews, MacDorman, and Thoma 2015).¹ Despite the declines, U.S. rates are higher than those in most developed countries (Organisation for Economic Co-operation and Development 2018), and some subpopulations in the United States have much higher infant mortality rates than others, as indicated above.

One such group includes infants who apply to the Social Security Administration (SSA) for Supplemental Security Income (SSI), a means-tested program that provides monthly income support and a link to public health insurance (Medicaid). As we will show, the infant mortality rate among SSI applicants is five times that of the general population.

To qualify for SSI, a child's family must have low income and limited resources and the child must have a severe disability. Newborns who spend time in a neonatal intensive care unit or are otherwise confined to a medical institution are not subject to the income and resource tests for SSI and can receive a small, fixed monthly benefit of $30 (as well as Medicaid coverage, if applicable in their state).² Anecdotal evidence suggests that some hospitals assist families in applying for SSI and Medicaid. In addition, several online resources point new parents (especially those of children born prematurely) to information and guidance on SSI and Medicaid. Parents may also find out about these programs through various other means before or after the child's birth: personal contacts or experiences, community resources, social workers, and the like.

Low birth weight (LBW) is defined by SSA either as a weight of less than 1,200 grams or as “at least 1,200 grams but less than 2,000 grams and small for gestational age”³ (Table 1). It is among the conditions that SSA considers to be “functionally equivalent” to criteria contained in its medical Listing of Impairments, which the agency consults to establish disability eligibility for SSI payments.⁴ Research shows that birth weight and length of gestation are the two most important determinants of infant survival. Nearly 25 percent of very LBW infants (less than 1,500 grams) die during the first year of life (Singh and van Dyck 2010). Impairments that are more common among premature and LBW infants are retinopathy of prematurity, chronic lung disease of infancy, intraventricular hemorrhage, necrotizing enterocolitis, and periventricular leukomalacia. Other disorders affecting this group include poor nutrition and growth failure, hearing disorders, seizure disorders, cerebral palsy, and developmental disorders (SSA n.d.).

Chart 1 shows the neonatal and infant mortality rates among SSI applicants and all U.S. children from 1985 through 2015. Although the neonatal mortality rates among SSI applicants and all U.S. children have been roughly similar, the infant mortality rate for SSI applicants has been several orders of magnitude greater than the overall infant mortality rate. The gap was widest in 1985, declined substantially until the early 1990s, and has been closing gradually ever since. In 1991, the infant mortality rate for SSI applicants was about 50 deaths per 1,000 compared with the overall rate of about 9 deaths per 1,000. By 2000, the infant mortality rate among SSI applicants had declined to about 40 deaths per 1,000 while the overall rate had declined to 7 deaths per 1,000. Between 2007 and 2014, the infant mortality rate among SSI applicants declined to about 30 deaths per 1,000, which was still roughly 5 times the overall rate of about 6 deaths per 1,000.⁵

Several policy changes during the 1990s affected the eligibility of infants for SSI. In 1991, SSA revised its regulations for determining disability in children in response to the 1990 U.S. Supreme Court ruling in Sullivan v. Zebley. Along with establishing an individualized assessment of functioning for children with impairments that do not “meet or equal” the criteria in the Listing of Impairments, the revised regulations established LBW as functionally equivalent to meeting a medical listing. In 1993, SSA began allowing presumptive disability awards for LBW using the criteria described above. In addition, SSA began targeting LBW children for continuing disability reviews (CDRs) as part of a broader effort to reduce the SSI CDR backlog. In the mid-1990s, the SSI eligibility cessation rate following CDRs for LBW children was around 40 percent. Finally, welfare reform legislation enacted in 1996 (and modified in 1997) required redetermination of SSI eligibility at or around a recipient's first birthday if LBW was a contributing factor material to the disability determination. These changes establish the boundaries of four distinct policy regimes, which we identify by date range: pre-1991, 1991–1993, 1994–1996, and post-1996. Box 1 summarizes the policy regimes, and Charts 1 and 2 indicate their boundaries.

This article has three primary objectives. The first is to document the trends in infant and neonatal mortality among SSI applicants. The second is to explore whether ease of access to public health insurance (Medicaid) improves mortality outcomes. The third is to explore whether mortality rates differ among SSI applicants by type of disability. These questions are important given the substantial federal, state, and local outlays for this population. SSA awarded SSI payments to more than 30,000 infants in 2015.⁶ More than half of those children were considered LBW,⁷ and of those, virtually all were eligible for public health insurance through Medicaid in addition to their monthly SSI payment—in most states, Medicaid eligibility is automatic for SSI recipients. Although we do not estimate the causal role of SSI on mortality or public health outlays, understanding the use of these programs is important from both fiscal and public health policy perspectives.

Data and Methods

We obtain our data from SSA's Supplemental Security Record (SSR), the master file for SSI program data. We organize the data by effective date of application (the first day of the month following the date on which the application is filed, or on which the individual first becomes eligible, whichever is later), which we shorten hereafter to “application date.” For each annual cohort, we use the SSR to identify the applicant's sex, type of disability, living arrangements, state, type of Medicaid access, and other characteristics, along with whether payments were awarded or denied. We link these data to SSA's master file of Social Security number holders, referred to as the Numident file. We use the Numident, which includes the Death Master File, to obtain mortality outcomes including date of death.

We categorize SSI applicants and recipients by policy regime in effect as of the application date (pre-1991, 1991–1993, 1994–1996, and post-1996), state of residence at the time of application, sex, type of disability (LBW, all others), living arrangements (as defined by SSI rules),⁸ and type of Medicaid access (defined by state policy at the time of application).⁹

Our analysis is largely descriptive. We present mortality trends among infant SSI applicants and compare them with trends for all U.S. infants for 1985 through 2015. We focus some components of the analysis on subsamples such as SSI recipients (awardees), applicants awarded before age 1,¹⁰ or the 2015 application cohort. We describe the characteristics of infant SSI applicants and examine neonatal and infant mortality rates by applicant characteristics including award or denial. We then present mortality-rate estimates with regression adjustments for SSI award, state of residence, living arrangement, and type of Medicaid access. We also consider the role of age at SSI application on mortality because applications filed very soon after birth may differ from those for older infants in terms of medical conditions, treatment needs, levels of support, and other factors. Finally, we run Cox proportional hazard models to explore the relationships between applicant characteristics and the hazard of mortality by age 1.

Our data are limited in that we use only the primary impairment code from the SSR to identify impairments, including LBW. We do not use the secondary impairment code from the SSR, nor do we have access to information on other comorbid conditions or the cause of death. In terms of geography, we are limited to state of residence at the time of SSI application. To the extent that SSI applicants may have moved in the time between application and award, or after award, our measure of the type of Medicaid access will include some classification errors. In the cases of SSI denials, we know applicant characteristics only as of application date.

In addition to data limitations, our analysis of SSI applicants and awards likely suffers from selection effects. Many or most children with the highest neonatal mortality will never apply for SSI simply because there is not enough time to apply and establish eligibility. Children who apply for SSI generally have lived through the period of highest likelihood of death. Further, children who are awarded SSI payments generally have more severe disabilities (and thus higher mortality) than do those whose applications are denied, but they also are more likely to have survived the application processing period (and are thus now at comparatively lower risk of mortality). The potential bias from these selection effects is difficult to determine deductively, and we do not attempt to control for these selection effects in our analysis.

Results

Chart 2 disaggregates the population of SSI child applicants into awardees and denials (for cases decided before the applicant reaches age 1) and presents neonatal and infant mortality rates for 1985 through 2015.¹¹ Infant mortality rates are much higher than neonatal mortality rates regardless of case outcome. Before 1997, the infant mortality rates for awardees (SSI recipients) and denials varied widely, sometimes swinging in opposite directions. Thereafter, the infant mortality rate for awardees loosely tracked the infant mortality rate for denials, with the rate for awardees exceeding the rate for denials by roughly 13 per 1,000 after 1997. After averaging roughly 48 deaths per 1,000 between 1997 and 2008, the infant mortality rate among awardees began to decline, dropping to less than 36 deaths per 1,000 in 2014 and 2015. Conversely, the neonatal mortality rate was higher for denials than for awardees in most years. Although the absolute difference in neonatal mortality rates between awardees and denials is smaller than that for infant mortality rates by case outcome, the neonatal mortality rates in many years differed by a factor of 3 or more.¹²

Table 2 presents estimated infant mortality rates over time and across policy regimes. We use Cox proportional hazard models in which the risk (hazard) of infant mortality is expressed as a function of the four policy regimes described earlier and summarized in Box 1. There is no right-censoring in the model because we observe all SSI applicants through at least age 1, which captures all instances of infant mortality. Each model adds control variables sequentially: type of Medicaid access, application outcome, living arrangement, age at application, and state of residence at application. Table 2 presents the adjusted infant mortality rates for each policy regime, conditional on the specific set of control variables included in the model. Three conclusions emerge. First, the adjusted infant mortality rates are largely invariant to the combination of control variables included in the model. Second, the adjusted infant mortality rates show a clear decreasing trend across the policy regimes. Finally, controlling for age at application (models 5 and 9) results in a marked increase in mortality rates in the early policy regimes.

2015 Cohort

The remainder of the analysis focuses on the 2015 cohort of SSI applicants and examines their characteristics and the relationships between those characteristics, application outcome, and infant mortality. Table 3 shows the characteristics of children who applied for SSI before age 1 in 2015.¹³ Nearly 54 percent of those applicants were awarded SSI payments before reaching age 1 and the denial rate for this cohort of SSI applicants was 42 percent. Slightly more than half of applicants and of awardees were male. About 35 percent of applicants and nearly 62 percent of pre–age 1 awardees were diagnosed with LBW. Nearly 82 percent of applicants in our study population lived in states where SSA determines Medicaid eligibility, as authorized by section 1634 of the Social Security Act; another 4 percent lived in states that make their own Medicaid-eligibility determination following the SSI criteria. The remaining 14 percent of applicants in our study population lived in states that determine Medicaid eligibility using criteria that are more restrictive than the SSI criteria, as authorized by section 209(b) of the Social Security Act. Most applicants and awardees lived with their parents, although 10–11 percent lived in Medicaid institutions and around 5–7 percent lived with others (such as a nonparent relative or in foster care).

We find that more infant applications were filed within 1 week of birth than in any subsequent week; after week 1, the frequency of application steadily declined with additional weeks of age. Although we observe this pattern among infants regardless of disability, it is more pronounced for LBW cases than for others. Among infant SSI applicants diagnosed with LBW, about 23 percent applied within 1 week of birth versus only about 6 percent of cases with a primary impairment other than LBW (not shown).

Table 3 presents two different breakouts for age at application: (1) each of the first 4 weeks versus the rest of the first year and (2) less than 180 days (about 6 months) versus 180 to 365 days. Overall, about 10 percent of infant applications occurred in the first week after birth. Awardees before age 1 were more likely than denials to have applied in each of the first 4 weeks after birth. More than 80 percent of infant applications were filed before 6 months of age. Again, the proportion of applicants who filed before 6 months of age was greater for awardees than for denials. Note that older infants are likely not to be presumptively disabled because of LBW and so would likely have a longer disability determination process by default.

Chart 3 presents infant mortality rates by state for all infants and for SSI applicants in 2015. The states are ordered by the infant mortality rate among SSI applicants. Vermont had no observed deaths among infant SSI applicants in 2015. Of the remaining states, the rates ranged from a low of 11.4 deaths per 1,000 applicants in Rhode Island to a high of 69.0 deaths per 1,000 applicants in the District of Columbia. In all states except Vermont, the infant mortality rate for SSI applicants far exceeded the rate for infants overall. Singh and van Dyck (2010, Figure 8) find modest regional patterns in overall infant mortality rates in 1970 and 2007, with higher rates generally concentrated in the southeastern states. For SSI applicants in 2015, several southeastern states had above-average infant mortality rates, but there is no dominant geographic pattern, suggesting that factors other than geography are at play.

Table 4 presents mortality rates for infant SSI applicants in 2015 disaggregated by disability diagnosis, type of Medicaid access, living arrangement, and age at application. Among all SSI infant applicants, the mortality rate is significantly higher for those with LBW than for those with all other disability diagnoses, as indicated by the p-value from the likelihood ratio tests of the homogeneity of the mortality rates. However, this result masks differences in infant mortality rates for LBW infants by application outcome. The vast majority of SSI denials are not diagnosed with LBW and experience a very low mortality rate (19.8 per 1,000, or about 2 percent). For the 3.7 percent of SSI denials with LBW (Table 3), the infant mortality rate is extremely high (195.4 per 1,000, or about 20 percent; Table 4). Among the much larger group of SSI awardees diagnosed with LBW, the infant mortality rate is 28.3 per 1,000. In fact, the infant mortality rate for SSI awardees with LBW is significantly lower than the 47.4 per 1,000 infant mortality rate for SSI awardees with other disability diagnoses.

It is possible that denials diagnosed with LBW lack access to the health care that is available to others who acquire Medicaid eligibility by virtue of an SSI award.¹⁴ However, we observe no statistically significant differences in infant mortality rates among SSI awardees or denials by type of Medicaid access. We do, however, see statistically significant differences in infant mortality rates by living arrangement for all applicants, awardees, and denials. In all groups, those living in Medicaid institutions have the highest infant mortality rates: For awardees, the infant mortality rate is 185.0 per 1,000; for denials, it is 150.1 per 1,000.

Infant mortality rates by age at application decline steadily with each successive week after birth within the first month. Among all applicants, rates decline from about 52 deaths per 1,000 first-week applicants to 43 deaths per 1,000 fourth-week applicants. We see a similar pattern among denials, although the mortality rates are much higher in the first 2 weeks for those applicants. From one week to the next, the infant mortality rate is about the same for all eventual infant awardees who applied in the first month after birth. At some point, the mortality rates of denied and awarded applicants cross. This is likely due to the changing nature of predominant disabilities among older infant applicants.

Table 5 presents infant mortality hazard ratios by selected characteristics of 2015 SSI applicants as estimated from Cox proportional hazard models. The models use a continuous measure of time under which a child is at risk of dying each day through day 365. The hazard ratios are interpreted as the cumulative risk of death by age 1 relative to a reference group. Thus, hazard ratios greater than 1.0 represent greater relative risk of infant mortality; hazard ratios less than 1.0 represent lower relative risk of infant mortality. In the first model, we estimate each variable's effect on infant mortality, independent of the effects of any other variables for the full sample of SSI applicants. However, because the variables may affect infant mortality in different ways for awardees and denials, we also estimate hazard ratios using three other models. The second model fully interacts the application outcome and applicant characteristics for the full sample of SSI applicants. The third and fourth models estimate two independent equations—one for awardees and one for denials. Although this pair of equations is qualitatively similar to the fully interacted model, it is often more easily interpreted; however, the reduction in sample size in each equation makes the second method less statistically efficient.

Among all applicants, awardees in the uninteracted model have a significantly greater estimated risk of death before age 1 than the reference group of denials, by a factor of about 1.9. Infants who live in Medicaid institutions also have significantly greater estimated risk of death before age 1 than do children who meet the SSI definition of living alone.¹⁵ Infants with a LBW diagnosis are less likely to die than are those with other impairments. Finally, children who applied more than 27 days after birth are estimated to be less likely to die before age 1 than are those who applied in the first week after birth.

When we consider interactions between SSI award and the other independent variables, we find that the estimates for awardees and infants with LBW are intertwined. In the interacted model, although children with LBW are more likely to die overall (hazard ratio = 6.33), those with LBW who are awarded benefits are less likely to die (hazard ratio = 0.04). This is seen in the separate models, too, where the hazards reflect different directions of risk.

The estimates by living arrangement are also intertwined with award, as awardees living with nonparents or in Medicaid institutions have significantly higher risk of infant mortality than do those living alone. Living in a Medicaid institution increases the risk of mortality in both the awardee- and denial-only models, all else equal.

Discussion

In this article, we show that the neonatal mortality rate among SSI applicants closely resembles the neonatal mortality rate among all U.S. children, but that the infant mortality rate among SSI applicants is roughly five times that of infants overall. Infant mortality among SSI applicants decreased dramatically from 1985 to 1993, a period in which SSI program changes increased the likelihood of awards to LBW infants soon after birth. Among SSI applicants who filed from 2007 to 2015, infant mortality decreased by about one-quarter.

For SSI infant awardees, we find differences in mortality by how the state of residence administers Medicaid access for SSI recipients. We also find that infant awardees diagnosed with LBW are less likely to die by age 1 than are children with other impairments. Denied applicants with LBW, on the other hand, are significantly more likely to die by age 1 than are denied applicants with other impairments.

Living arrangement appears to be a primary factor associated with mortality, as infants living in Medicaid institutions have the highest mortality rates and those living with their parents have among the lowest mortality rates. Of course, it is unlikely that living arrangement is itself the cause of this differential. Individuals living in a Medicaid institution are, by definition, in poor health and thus at greater risk of death.

This study has limitations that future studies should work to overcome. First, our analysis includes only SSI applicants. Many families may not have been informed of potential SSI eligibility in the neonatal period or they may have erroneously expected their infant's health status to improve quickly. We do not know their outcomes relative to those of our study population. Second, we have limited information on types of disability, especially for denied applicants. SSA records for that group may not indicate an impairment type, which introduces potential error into our measure of disability diagnosis. Similarly, we cannot measure Medicaid and other health insurance coverage among denied applicants.

We stress that our findings do not present a causal effect of SSI on infant mortality. Nor does this article determine the role of SSI in other outcomes. In addition to mortality, LBW and lack of health care access have been linked to poor school outcomes and potentially poor adult employment outcomes (Behrman and Rosenzweig 2004; Black, Devereux, and Salvanes 2007; Oreopoulos and others 2008). One purpose of SSI is to mitigate the effects of poverty and disability on disabled children and their families so that the child recipients can eventually lead self-sufficient lives. Guldi and others (2017) provide some evidence that SSI mitigates some of the disadvantages borne by LBW children of mothers with low educational attainment. Additional studies should assess the effectiveness of SSI on the long-term outcomes for this vulnerable population.