Objective. Ambulatory Care-Sensitive Conditions (ACSCs), conditions for which ambulatory care may reduce, though not eliminate, the need for hospital admission, have been used as an index of adequate primary care. However, few studies of ACSC have focused on children. We estimated national hospitalization rates for ACSC among children and examined the behavior of the index between subgroups of children.
Methods. We used data from the 1990–1995 National Hospital Discharge Surveys (NHDS), the US census, and the National Health Interview Survey (NHIS) to calculate hospital discharge rates. Rates were estimated as the number of condition-specific hospital discharges from the NHDS divided by the population at risk, as estimated from the US census and NHIS.
Results. Predictably, ACSC hospitalization rates were significantly higher among children who were younger, black, had Medicaid insurance, and lived in poorer areas compared with their counterparts. However, the relationship between ACSCs and income and the distributions of conditions within the index varied significantly between children.
Conclusions. ACSCs may indicate disparities in access and utilization of health care, however, the differing behavior of the index between subgroups suggests that inferences from examining rates of ACSCs may not be comparable for all children.ambulatory care-sensitive conditions, hospitalization rates.
Access to health care for children has emerged as an important policy issue in recent years.1 Many reports and studies have documented insufficient access to care among poor children and among children without health insurance in the United States.2–13 For example, children living in poorer neighborhoods have fewer ambulatory care visits than children in wealthier neighborhoods4 and children with lower family incomes are less likely to have a regular source of medical care than children with higher family incomes.6 St Peter and colleagues7 found poor children were less likely to receive routine care in a physician's office than children with higher family incomes. Newacheck8 reported associations between children's use of physician services and poverty, maternal education, and lack of health insurance coverage. Stoddard and colleagues10 found that children without insurance were less likely to receive physician care for specific acute conditions than children with health insurance. Children with Medicaid have higher hospitalization rates than privately insured children.12The variety of outcomes used in these studies underscores the complexity of measuring access to care.13,14
One index that has been used to assess access to health care is hospitalization rates for potentially ambulatory care-sensitive conditions (ACSCs),14 calculated as the estimated number of hospital discharges in a year with an ACSC as a primary diagnosis, divided by the estimated size of the population at risk. From the link between access to health care and hospitalization rates, previous studies have compiled sets of medical conditions for which hospitalizations could be reduced if adequate primary care were received.15,16 Using these lists, researchers have demonstrated that ACSCs are associated with health insurance,16–18 perceived access to health care,19 and income.15,17,19 However, few of these studies have focused on children.
In one study of ACSCs in children, Casanova and Starfield20 found no relationship between ACSCs and socioeconomic status in Spain, a country with universal access to health care. They further reported lower hospitalization rates for ACSCs among children in Spain compared with children in the United States. Casanova and Starfield concluded that the provision of universal access to care was associated with lower hospitalization rates for conditions amenable to adequate primary care. Supporting their conclusion, a Maryland study found that poor children enrolled in the state's Medicaid Managed Care plan had fewer hospitalizations for ACSCs than their similarly disadvantaged, but nonenrolled, counterparts.21 In addition, a recent Institute of Medicine report, America's Children: Health Insurance and Access to Care, reports that children living in high-income areas are hospitalized less often for ACSCs than children living in low-income areas.3
For this study, our first objective was to calculate national estimates of hospitalization rates for ACSCs among children. Toward this objective, we estimated hospitalizations for ACSCs among a large nationally representative sample of children in the United States, overall and within subgroups.
Our second objective was to evaluate the general behavior of the ACSC index for children. If differences in the ACSC index are to be used to compare access to health care, then the index should act similarly and have similar components between groups. For this study, we do not attempt to justify the use of ACSCs as an index of primary care; to explicitly validate that relationship more specific data linking primary care characteristics to hospitalizations would be needed. Instead, we investigated the behavior and composition of the index among different demographic subgroups of US children.
For our evaluation, we compared the relationship between income category and hospitalization for ACSCs among different subgroups of children categorized by age, race, region of the country, and expected source of hospital payment. Because income is a strong predictor of access and utilization of health care for children, we expected that children living in lower income areas would have corresponding higher rates of hospitalization for ACSCs than children living in higher income areas, and that this relationship would be evident among all subgroups of children. We assumed that similar relationships among different subgroups of children would lend some support, though not prove definitively, the validity of the index; conversely, we assumed that dissimilar relationships would imply that either ACSCs or income (or both) are differentially related to access to or utilization of health care for different groups of children. Next, we examined the distribution of the diagnoses within the ACSC index between these subgroups of children. Again, similar distributions would support the idea that the ACSC index represents a consistent concept for different groups of children, although more targeted studies are required to firmly establish the use of ACSCs as a measure of access or utilization of primary care.
We used data from the National Hospital Discharge Surveys (NHDS) from 1990–1995.22,23 The NHDS consists of a sample of inpatient records from a national sample of nonfederal general and short-stay specialty hospitals in the United States. For each of the years of the survey used for this study, the survey collected about 15 000 medical records for children.
Our list of ACSCs was defined based on existing lists16,20and careful review (K.C.S.). The final list included 6 categories: 1) asthma (International Classification of Diseases, Ninth Revision [ICD-9] 493); 2) pneumonia (ICD-9 481, 483, 485, 486, 482.2, 482.3); 3) other upper airway conditions (ICD-9 381, 382, 460, 461, 462, 463, 465, 490, 472, 473, 474, 034.0, 079.9, 466.0), not including specific procedures (ICD-9 828, 823, 286, 200.1); 4) gastroenteritis and dehydration (ICD-9 558.9, 276.5, 008.8, 008.6); 5) cellulitis (ICD-9 680–684, 686, 289.3); and 6) seizures (ICD-9 345, 780.3). Conditions commonly on other lists of ACSCs but not deemed appropriate for an index for children were omitted. Congestive heart failure and diabetes, for example, are important contributors to the existing ACSC lists currently applied to adults. However, because of their relatively low prevalence and high severity among children, they were not included on our list. It is important to note that ACSCs refer to medical conditions for which appropriate ambulatory care is expected to reduce the risk of hospitalization. Consequently, a condition's presence on this list does not imply that all hospitalizations for that condition are avoidable or inappropriate.
Three broad conditions that may be expected to result in similar rates of hospitalization among groups, regardless of their level of access to ambulatory care were defined as nonambulatory care-sensitive conditions (non-ACSCs): appendicitis without rupture (ICD-9 540.9), congenital heart disease (ICD-9 745, 746, 747), and malignant neoplasms (ICD-9 140–208, 230–234, 284, 287, 288.0–288.2). This grouping was used primarily as a comparison for the main analysis.
Zip code-based median income was used to provide a measure of socioeconomic status.17,24 The NHDS does not collect income or other individual measures of socioeconomic status on its medical abstract form. However, the residential zip code of the patient is collected. For confidentiality reasons, zip code is not included on public use NHDS data files, however, on in-house files, under restrictive conditions, residential zip code can be used for analysis.17 Approximately 4% of the discharges could not be linked to a zip code and were excluded.
We used zip code-level median family income from census data25 for 1990 to link to the 1990–1992 NHDS data files and zip code level median family income estimates for 1994 from Claritas (Arlington, VA), a commercial marketing firm, to link to the 1993–1995 NHDS data files. Quartile values of income were derived from linking the zip code-level median income values from 1990 and 1994 to the National Health Interview Survey (NHIS) for the corresponding years. Although respondent zip code is not released on the public use files of the NHIS, as the NHDS, the variable is collected for administrative purposes and available, under specific circumstances, on in-house computer files. Because the NHIS is weighted to be nationally representative,26 the distribution of zip code-level income among children could be estimated. Based on distributions from the NHIS, 1990 income was broken into 4 quartiles: <$23 500, $23 500–$29 500, $29 500–38 500, and ≥$38 500. Similarly, 1994 income (in 1990 dollars) was divided with the following cutpoints: $27 500, $33 500, and $42 500. Because we had too few discharges for stable estimates in the highest income quartile, the 2 highest groups were combined.
Although the NHDS does not collect detailed health insurance information, the expected source of payment is collected on the medical abstract form. For our analysis, we divided the source of payment into 3 groups: Medicaid; uninsured, recorded as self-pay on the discharge abstract; and some other source of payment, which was primarily private health insurance, but included the categories other payer (5.3%) and unknown source of payment (3%).17,27,28
Hospitalization rates and health care utilization are known to vary by race.4 However, on the NHDS, race is missing for about 20% of the children. There is evidence that a relatively large proportion of the children with missing race data live in areas that are largely white.29 Based on this evidence and previous hospitalization utilization data, which show a gap in utilization between white and black children, we assigned all the missing race data to white. The effect of this assignment on our results is to narrow differences by race. Children with a reported race other than white or black were not included in the race-specific analysis because reliable discharge rates by income could not be obtained for these groups. Because practice patterns and hospitalization rates vary regionally,30,31 we examined discharges for ACSCs by geographic region: North, Midwest, South, and West. There were relatively few discharges in the West because of to the lower number of responding hospitals in that region.22,23
To calculate discharge rates, we obtained annual population counts, by age, race, insurance status, and region, from the US census.32 These population values were considered free of sampling error for variance calculations. To estimate the number of children within each income category for each of the demographic variables, the estimated proportion of children within each cell was obtained from the NHIS. Standard errors of these proportions from the NHIS were used to obtain standard errors of the populations totals, by income level.
We used SUDAAN33 to control for the complex sampling design of the NHDS. Variances of rate ratios were considered reliable when the relative standard errors of both the numerator and denominator were <10% or the relative standard error of the denominator was <5%.34 For the majority of estimates in this study, these criteria were met. For completeness, however, we present some estimates that fail these criteria and indicate them when appropriate.
To examine the relationship between income group and the discharge rate for ACSC linear categorical models using the CATMOD procedure in SAS (SAS Institute, Inc, Cary, NC).35 As a general test for trend, the estimated discharge rates were modeled as a function of income and resulting parameter estimate was assessed for statistical significance. These models allowed us to incorporate the standard errors of the discharge rates derived from SUDAAN. We were able to fit these models overall and within demographic subgroups. However, we had insufficient power to estimate rates or test trends adjusted for >1 variable simultaneously—the relative standard errors of the discharge rates exceeded NCHS standards for statistical reliability and could not be modeled. To determine if the relationships between income group and the proportion of discharges attributed to ACSCs were statistically significant, tests of association based on χ2statistics were calculated.
Over one-third of all hospital discharges occurred to children living in zip code areas with a median family income in the bottom one-fourth of the income distribution (Table 1). In contrast, <40% of discharges occurred in the areas that make up the top half of the income distribution. About 10% of discharges were for uninsured children and nearly one-third were for children with Medicaid coverage listed as their expected source of payment. Although the younger children comprised <30% of the population 1 to 14 years old, they made up nearly half of the discharges.
There were 29.4 hospital discharges per 1000 children in the United States during 1990–1995 (Table 1). We found significantly more hospitalizations among children living in the lowest income areas than those living in higher income areas. Discharge rates were lowest among the uninsured and highest among children with Medicaid. There were significantly more discharges among the younger (ages 1–4) than among the older (ages 5–14) children. Black children had a slightly higher discharge rate than white children, despite the addition of the children with unknown race into the white group; this difference, however, was not statistically significant. Discharge rates were highest in the Northeastern region of the United States and lowest in the Southern region.
There were 10.9 discharges for ACSCs per 1000 children, about one-third of all hospitalizations (Table 2). The ACSC rates varied, similar to the overall hospitalization rates; rates were higher among the younger compared with the older children, higher among lower income compared with higher income children, higher among black compared with white children, and higher among children with Medicaid insurance than among other children.
The relationship between income group and ACSC hospitalizations was significant overall and after individually adjusting for each of the demographic variables. However, there was no significant relationship between income group and ACSC hospitalization within some subgroups of children. Among uninsured children or among children with Medicaid, the discharges for ACSCs in the lowest quartile was higher, if not significantly higher, than for those in the higher income groups. Black children in the lowest income group had a slightly higher rate of discharge for ACSC than those in the higher income groups, but the trend was not statistically significant. The relationship between income and ACSC hospitalization was significant among children in 3 of the 4 regions (not the West), but the underlying ACSC hospitalization rates differed. However, estimates for income groups within regions had high relative standard errors and are unstable, particularly for the West.
As a percentage of overall hospital discharges, over one-third of discharges were for ACSCs (Table 3). However, this proportion differed among subgroups. Nearly half of all hospitalizations among younger children (48.0%) were for an ACSC compared with 28.1% among older children. Statistical tests and patterns of the relationships between income and the percent of discharges for ACSCs were similar to those observed between income and the ACSC rate; however, the differences in percents between lower and higher income children were smaller.
The composition of the ACSC index varied significantly, if not always substantially, between subgroups of children (Table 4). For example, although asthma discharges comprised about 27% of all ACSCs, they were a much larger component of the ACSC index among black children than among white children, and a larger component of ACSCs among children in the West compared with children in the South.
Discharges for conditions not considered ambulatory care-sensitive also varied (Table 5). There was no clear relationship between non-ACSCs and income; nonetheless, children in lower income areas had slightly higher discharge rates for these conditions. In contrast, black children had discharge rates considerably lower than other groups; children in the South and Midwest as well as those without health insurance also had somewhat lower rates than other children. The lack of statistical significance for some groups, however, could be attributable to the relatively low number of discharges and resulting low statistical power.
These results confirm the clear relationship between socioeconomic status and hospital discharges for potentially avoidable or ACSCs. Consistent with results from other studies,3,15–17children living in areas with lower median incomes had higher discharge rates for ACSC than children living in higher income areas. Within subgroups of children, those living in the lowest 25th percentile of income areas had higher discharge rates for ACSCs than other children, although these rates were not always significantly higher. We also found children living in lower income areas had slightly higher discharge rates than children in higher income areas for conditions thought to be not sensitive to preventative or ambulatory care. However, the rates were low and the differences were small and not statistically significant. These findings may reflect disparities in access to care and health insurance, differing practice patterns, and differences in underlying health status.
Younger children had much higher discharge rates for ACSCs than older children. Even as a proportion of all hospitalizations, younger children had proportionally more ACSC discharges than older children. This is primarily attributable to discharges for respiratory conditions. Although these conditions are more common among younger children, it is also possible that these conditions are more responsive to primary care in older compared with younger children. However, the income gradient suggests that at least some of these discharges could be prevented.
Like Pappas and colleagues,17 we found children with private/other health coverage had lower rates of discharges for ACSCs than children with Medicaid, within each income category. In fact, children with Medicaid coverage living in areas with the highest income level had discharge rates for ACSC over twice as high as the overall rate (22.2 vs 10.9/1000; Table 2). This finding is not surprising given the large variation in individual-level socioeconomic status in each of our income groupings. It also could be attributable to different eligibility criteria for Medicaid—children in wealthier areas may be more likely to be eligible based on a medical condition, whereas, children in poorer areas may be more likely to meet the criteria based on financial need. However, we found uninsured children had the lowest discharge rates, overall and for ACSC. This result differs from previous studies that reported higher ACSC hospitalizations for the uninsured under 65 population relative to those with private insurance.16,17 However, Weissman16 reported uninsured children in Maryland had lower rates of hospitalizations for ACSCs than privately insured children in his 2-state study, although this result was not apparent in Massachusetts. The proportion of discharges attributed to ACSCs was similar for children without any insurance coverage and for children with private/other coverage but was higher for children with Medicaid coverage. If we assume that uninsured children are disadvantaged within the health care system and may have a higher percent of discharges for ACSC, it is likely that some of our findings for uninsured children could be attributable to the misclassification of insurance status based on expected source of payment.
We found that the relationship between income category and ACSCs differed between subgroups. For example, the relationship was much stronger among children with private/other insurance than among children with Medicaid coverage, and stronger among white than among black children. We also found that the distribution of conditions comprising the ACSC index differed between children. Similar relationships or similar compositions among the different subgroups would have supported the hypothesis that similar access problems are being comparably measured by the ACSC index for each group. From these results, however, definitive conclusions cannot be made about the behavior of the ACSC index. Users of an ACSC index for comparative purposes need to consider differences in its implications for different groups to make reasonable conclusions about the quality or accessibility of primary care.
Although hospitalization for an ACSC is generally looked on as a measure of ambulatory care access and utilization, it is important to consider other factors that may contribute to the variation in ACSC hospitalization. For example, rates of ACSC hospitalization vary according to the underlying prevalence of the conditions included in the index. Billings and colleagues,15 however, examined underlying asthma rates and concluded that differences would have minimal impact on the inferences from ACSCs. Furthermore, the difference in the prevalence of asthma among children in the Billings study, between those in the higher and lower income areas, was lower than the differences for the other age groups. Additionally, from a policy perspective, the underlying prevalence of all conditions that contribute to the index would likely not be controllable in comparisons between, for example, health plans or demographic groups, as well.
Another specific factor that may influence a group's access is the quality of ambulatory care that is available to that population. Work by Homer and colleagues36 suggests, not surprisingly, that hospitalization for asthma and other conditions is influenced by the type of care received by a population, not by the simple receipt of care. Unfortunately, because of data limitations, measurement of access to health care is often limited to quantitative measures, such as the number of ambulatory visits, with no consideration of content of care.
Another limitation of this study is the degree of comparability between the numerators and the denominators for rate calculation, particularly for rate estimates by child's race. As mentioned above, child's race is not reported on about 20% of the discharge records. Given the lower discharge rates for the white children in this study relative to the black children, our assigning all of these children with unreported race into the white category likely made our estimated differences between the groups smaller than the actual difference.29
Comparability between the numerator and denominator also affected estimates by insurance status. Using the expected source of payment from the NHDS may not be consistent with actual insurance status.27 For example, children without health insurance may become eligible for Medicaid once admitted to the hospital, which may partially explain the lower discharge rates for children without insurance coverage and higher rates for children with Medicaid coverage. There is also evidence that continuity of Medicaid coverage is low,37 further reducing the compatibility between our numerator and denominator estimates. In addition, our inclusion of children with other and unknown source of payment in the private/other insurance category would also bias the rate estimates for uninsured children downward if some of these children are uninsured. However, given the low proportion of children in the unknown and other category, this bias is likely to be small. In this study, estimates of proportions of children in the population by health insurance status and income category were derived from the NHIS and official census estimates. The correspondence between the groupings used by the census and the expected source of payment is unknown.
The use of zip code-level income as an indicator of socioeconomic status, rather than family level income, is also a limitation. Although there is a growing research literature documenting the advantages and disadvantages of community level variables relative to individual level variables,24,38 we had no individual level data or additional community level data for comparison.
Despite these limitations, we found clear disparities in discharges for ACSCs between children, which suggest corresponding inequalities in adequate primary health care. Furthermore, our estimates are based on a large, nationally representative sample of hospital discharges in the United States making them more widely applicable than smaller studies. However, the divergent associations between ACSCs and income, as well as the varying distributions of conditions within the ACSC composite, imply that the index may be differentially related to access or utilization of health care between subgroups of children. It is not surprising that a single measure cannot convey the myriad elements that contribute to problems in access to health care among children. Several access to care measures, in addition to ACSCs, are required to more fully identify high-risk children and their health care needs.
- Received March 20, 2000.
- Accepted June 29, 2000.
Reprint requests to (J.D.P.) Infant and Child Health Studies Branch, National Center for Health Statistics, 6525 Belcrest Rd, Room 790, Hyattsville, MD 20782. E-mail:
- ACSCs =
- ambulatory care-sensitive conditions •
- NHDS =
- National Hospital Discharge Survey •
- ICD-9 =
- International Classification of Diseases, Ninth Revision •
- NHIS =
- National Health Interview Survey
- Starfield B. Family income, ill health, and medical care of US children. J Public Health Policy. 1982(Sept):244–259
- Institute of Medicine. America's Children: Health Insurance and Access to Care. Washington, DC: National Academy Press; 1998
- Pamuk E, Makuc D, Heck K, Lochner K. Socioeconomic Status and Health Chartbook. Health United States, 1998. Hyattsville, MD: National Center for Health Statistics; 1998
- Weigers ME, Weinick RM, Cohen JW. Children's Health, 1996. Rockville, MD: Agency for Health Care Policy and Research; 1998. MEPS Chartbook No. 1. AHCPR Publ. No. 98-0008
- Simpson G, Bloom B, Cohen RA, Parsons PE. Access to health care. Part 1: children. National Center for Health Statistics. Vital Health Stat. 1997;10(196)
- Newacheck PW,
- Hughes DC,
- Stoddard JJ
- Weinick RM,
- Weigers ME,
- Cohen JE
- Homer JC,
- Perrin JM,
- Kemper K,
- Freeman J
- Center for Health Economics Research, Robert Wood Johnson Foundation. Access to Health Care: Key Indicators for Policy. November Princeton, NJ: Center for Health Economics Research, Robert Wood Johnson Foundation; 1993
- Weissman JS, Epstein AM. Falling Through the Safety Net. Baltimore, MD: Johns Hopkins University Press; 1994
- Billings J, Zeitel L, Lukomnik J, et al. Impact of socioeconomic status on hospital use in New York City. Health Aff. 1993(Spring):162–173
- Billings J, Teicholz N. Uninsured patients in the District of Columbia hospitals. Health Aff. 1990(Winter):158–165
- Gadomski A, Jenkins P, Nichols M. Impact of a Medicaid primary care provider and preventive care on pediatric hospitalization. Pediatrics. 1998;101(3). URL: http://www.pediatrics.org/cgi/content/full/101/3/e1
- Graves EJ. National Hospital Discharge Survey. Annual summary, 1993. National Center for Health Statistics. Vital Health Stat. 1995;13(121)
- Gillum BS, Graves EJ, Wood E. National Hospital Discharge Survey: annual summary, 1995. National Center for Health Statistics. Vital Health Stat. 1998;13(133)
- Geronimus AT,
- Bound J
- Census of Population and Housing, 1990: Summary Tape File 3b (machine-readable data files)/prepared by the US Bureau of the Census. Washington, DC: US Bureau of the Census (producer and distributor), 1992. Available at: http://www.census.gov/td/stf3/contents.html
- Massey JT, Moore TF, Parsons VL, Tadros W. Design and estimation for the National Health Interview Survey, 1985–94. National Center for Health Statistics. Vital Health Stat. 1989;2(110)
- Graves EJ. Expected Principal Source of Payment for Hospital Discharges: United States 1990. Advance Data From Vital and Health Statistics, No. 220. Hyattsville, MD: National Center for Health Statistics; 1992
- Kozak LJ,
- Norton C,
- McManus M,
- McCarthy E
- Kozak LJ. Underreporting of Race in the National Hospital Discharge Survey. Advance Data From Vital and Health Statistics, No. 265. Hyattsville, MD: National Center for Health Statistics; 1995
- Goodman DC,
- Stukel TA,
- Chang C
- US Census Bureau, Population Division, Population Distribution Branch. National Population Estimates. Available at: http://www.census.gov/population/www/estimates/uspop.html
- Shah BV, Barnwell BG, Bieler GS. SUDAAN User's Manual: Software for Analysis of Correlated Data, Release 6.40. Research Triangle Park NC: Research Triangle Institute; 1995
- Cochran WG. Sampling Techniques. 3rd ed. New York, NY: John Wiley and Sons, Inc; 1977
- SAS Institute Inc. SAS/STAT Users Guide, I. Version 6. 4th ed. Cary, NC: SAS Institute Inc; 1989
- Homer CJ,
- Szilagyi P,
- Rodewald L,
- et al.
- Copyright © 2000 American Academy of Pediatrics