[ Main page of Report | Contents of Report ]
PURPOSE AND APPROACH
This section summarizes the site-specific analysis prepared by the Center on Urban Poverty and Social Change at Case Western Reserve University in Cleveland.(5) It focuses on the relationship between neighborhood conditions and residents' access to primary health care.
Access to health care is obviously recognized as an important determinant of health. It is also suspected that access is generally low for residents of poor neighborhoods, because sufficient facilities and services do not exist in or near those neighborhoods and/or because other barriers (e.g., the lack of health insurance, cultural barriers) prevent them from accessing those that do exist. One noted indication is the tendency of the poor to visit hospital emergency departments to deal with problems that more affluent Americans typically take to their regular family doctor. However, the lack of neighborhood-level indicators of health access has prevented significant research on the issue to this point.
Data Source: Medicaid Files
The Center's staff reasoned that Medicaid claim and encounter records would be a good source of data to shed more light on the issue--data from the state of Ohio's Department of Jobs and Family Services. A large share of all low-income families are enrolled in Medicaid, and the records for each family contain information on their address and on much of the health care they receive.
The Center has an established reputation with the state, having done a considerable amount of research for state agencies in the past, and it was able to obtain access to the files in time to conduct this analysis within our project's time constraints. The files for the study contained the records for all children from Cuyahoga County who were between birth and 6 years of age between July 1998 and June 1999.
Indicators and Methods
The staff selected two types of indicators that could be created from the files. The first was comprehensive preventative visits (CPVs) to doctors during a child's first year of life. They developed four specific indicators in this category: (1) percentage of newborns with a CPV before 3 months; (2) percentage of infants with no CPV from birth to age 1; (3) percentage of infants with five or more CPVs from birth to age 1; and (4) average number of CPVs for infants from birth to age 1. The second category contained two indicators dealing with visits to hospital emergency departments (EDVs): (1) monthly average percentage of children under 6 with an EDV and (2) annualized number of EDVs per child under 6.
The analysis consisted of relating each of the above indicators at the neighborhood level to a series of other neighborhood measures from the census and other administrative records as maintained in the Center's information system (indicators of age and racial/ethnic composition and family type along with various rates, such as inadequate prenatal treatment, low-birth weight births, child maltreatment, violent crime, poverty, and employment). The staff employed mapping analysis as well as correlation analysis to examine these relationships.
FINDINGS AND IMPLICATIONS
Hypotheses and main findings
The correlation coefficients resulting from this ecological analysis are presented in table 2.1, and maps showing spatial patterns for two of the indicators are presented in figure 2.1. Findings are noted under the four hypotheses posed by the staff.
| Early initiation of CPV | No CPVs | All CPVs | Average CPVs | Percent with ED visit per month | Annualized ED visits | |
|---|---|---|---|---|---|---|
| ECONOMIC INDICATORS | ||||||
| Poverty rate | -0.07 | 0.11 | -0.16* | -0.18** | 0.13** | 0.13** |
| Poverty rate for children under age 6 | -0.07 | 0.09 | -0.12 | -0.14* | 0.19** | 0.18** |
| Employment rate for males age 16 and over | 0.08 | -0.06 | 0.1 | 0.09 | -0.08 | -0.06 |
| Employment rate for females age 16 and over | 0.04 | -0.05 | 0.09 | 0.11 | -0.06 | -0.06 |
| RACE/FAMILY STRUCTURE INDICATORS | ||||||
| Percent white | 0.04 | -0.05 | 0.15* | 0.09 | -0.12** | -0.11* |
| Percent black | -0.04 | 0.03 | -0.13* | -0.07 | 0.10* | 0.10* |
| Percent hispanic | 0.03 | 0.13* | -0.12 | -0.14* | 0.17** | 0.16** |
| Percent of households with children under age 18 that are female-Headed | -0.04 | 0.09 | -0.20** | -0.18** | 0.23** | 0.22** |
| Note: CPV is an abbreviation for
comprehensive preventative visits. ED is an abbreviation for emergency
department. ** Correlation is significant at the .001 level. * Correlation is significant at the .05 level. |
||||||
Implications
All of the findings were certainly not as expected. The ideal of primary care has been for all children to have a "medical home"--a place where they can get regular preventive or well-child care and can also receive medical treatment for acute or chronic illnesses that are appropriately treated by pediatricians in their clinics and offices. The researchers had initially assumed that CPVs were indicative of access to preventive care, one aspect of primary care. However, they did not have a direct way of measuring access to primary care providers for illnesses.
Instead, the Center chose to measure a negative indicator: the frequency of EDVs. The assumption was that high EDV rates would be a proxy for the lack of access to primary care for illnesses. The Center recognized that some EDVs, especially for trauma or critical conditions, were appropriate, but it has not yet perfected a method for removing the "true emergency" EDVs from the counts.
In an additional analysis, the researchers showed that CPVs and EDVs were not correlated with each other at the census tract level. This suggests that these two types of indicators are measuring different things. It is possible that they are measuring two unrelated aspects of primary care access or that one of them is not an indicator of access but responds to other factors.
The researchers' assumption, though, is that families have more difficulty getting in to see a primary doctor when their children are ill than for well-child visits. Sick visits might be difficult for several reasons. Families that do not own an automobile may find it difficult to take a sick child to the doctor or clinic during normal business hours. They may need to wait until someone gets off work to drive them and, by that time, only the emergency services are open. Or these families may have a sense that illnesses are urgent and may not be comfortable waiting for an appointment with a primary care provider. Another possibility is that clinics and doctors' offices that serve these families for their well-child care are overcrowded and cannot readily schedule same-day sick visits, which are often needed for young children who are ill.
Another unexpected finding was that several of the measures of CPV rates were not correlated with poverty or other economic indicators at the neighborhood level. This may reflect the fact that local agencies and the managed care organizations have made a concerted effort to reach out to poor families to ensure that they receive their needed well-child visits. Thus, these indicators do not show any ecological correlations because the usual barriers have been removed. Deeper study (interviews and analysis of other administrative data) would be required to find out whether this is the correct interpretation. Whatever the result, it would seem advisable to monitor these relationships in the future.
Data limitations and potentials
Center staff encountered a number of difficulties in preparing these data for analysis. Extra work was required to avoid double counting where more than one claim was made for an event. Also, a simplifying assumption was needed to avoid excess complexity arising from the fact that children can move several times during their year of birth (it was decided to use the census tract at birth for all of a child's CPVs). Another difficulty was that data could not be provided for tracts with fewer than five children on Medicaid, to protect confidentiality. In addition, data for other tracts had to be removed from the analysis because of the rare events issue discussed earlier in this report--a problem that could be addressed in the future by averaging over several years, assuming the data will continue to be provided.
The Medicaid data also have some obvious coverage limitations. First, these files contain no data on children who are privately insured or are otherwise not enrolled in Medicaid. Nor do they have data on children who were not enrolled for periods of time for various reasons (to address the latter, staff included only children who were continuously enrolled during the year). Also, some events are missed because of services provided by public health clinics and other providers that do not seek reimbursement from Medicaid.
Nonetheless, these data do have a great deal to say about a large population of interest. Given the difficulties, the fact that Center staff were able to edit and analyze these data successfully is important evidence. It certainly suggests that more extensive use could be made of them in Cleveland and that local data intermediaries in other cities might be able to take advantage of them as well. Relevant offices in other states may be able to provide similar data for similarly controlled studies. For future work in Cleveland, the research would be even more valuable if the additional information sought by the Center but not received could be obtained: data on addresses of all Medicaid service providers and on non-EDV visits related to illnesses.
The data clearly fill a void in local knowledge--at present, local officials and advocates have virtually no recurrently provided information on how types and levels of primary health care vary across neighborhoods for low-income populations. Such data would appear extremely useful for program management, policymaking, and public accountability, especially if monitored recurrently.
Community Process
The Early Childhood Initiative (ECI) in Cuyahoga County has chosen as one of its goals a "medical home" for every child under age 6. Its initial focus has been on newborns and their parents. The ECI is very interested in using the indicators in this report as a way of measuring its own progress. It has already succeeded in expanding Medicaid enrollment to virtually all of the county's uninsured children. The new indicators of access to primary care will allow the ECI to determine whether there are particular neighborhoods that need to be targeted for assistance with access to primary care. Moreover, these indicators can be disaggregated in other ways, such as by age or program status, which will allow ECI to refine its approach. Since the indicators have just been reported at the time of this writing, it will be several months before all of the potential uses become apparent.
[Go To Contents]
PURPOSE AND APPROACH
This section summarizes the site-specific analysis prepared by the Piton Foundation in Denver.(7) It focuses on the possible effects of two types of conditions that are increasingly discussed as risks to health in urban neighborhoods: environmental hazards and exposure to violence.
Purpose
The Piton Foundation has a tradition of working closely with residents, activists, and community-based organizations located in or serving Denver's poorer communities. One focus of this work has been the search for more and better data to describe neighborhood realities and inform community action.
Denver Benchmarks is a new initiative embodying this theme in which Piton is partnering with residents, city officials, and other stakeholders to develop measures of community health and quality of life for all Denver neighborhoods. The two neighborhoods that are piloting the initiative (Cole in northeast Denver and Overland in south central Denver) have important socioeconomic differences, but both identified environmental hazards and community violence as their highest priority concerns. Accordingly, the Foundation sought to help these communities learn more about the influence of these problems so they could better address them.
Data sources and approach
The Piton team assembled a considerable amount of information that might serve as the basis for indicators in these areas. They planned to use correlation and mapping analysis to relate data on these problems to health outcomes and other neighborhood conditions.
As to environmental hazards, they purchased data on items 1 through 6 below from a commercial vendor (which had obtained and cleaned original data from various national sources). Data on items 7 and 8 were obtained from the Denver Department of Public Health and Environment. Information on all topics was provided for multiple years, and locations were geo-coded so they could easily be analyzed and related to neighborhood boundaries.
As to community violence, Piton already maintained data on Part I violent crimes by neighborhood (as noted in section 2). For this study, they obtained two new types of data. The first was information on suspensions and expulsions of students in the Denver public school system, coded by reason. They selected several of the reason codes as identifying violence-related offenses and geo-coded them for one academic year (2001/2002) by residence address of the student.(8) The second type of data was on confirmed cases of child abuse and neglect (by address), provided by the state Department of Human Services. For confidentiality reasons, Piton has agreed not to report such data for any area with five or fewer cases. To reduce the impact of this restriction for the analysis (and to address the broader rare events issue), Piton grouped the data for overlapping three-year periods (starting with 1995-1997 and running through 1998-2000--this allowed reporting for 57 of Denver's 77 defined neighborhoods).(9)
As to health outcomes,Piton already had the measures based on vital statistics from its own data system (as analyzed in section 4). However, the team obtained data on two new indicators from vital records: (1) Apgar scores, which represent the results of tests of physical functioning taken just after birth, and (2) the premature birth rate (live births with a clinical gestation period of 20 to 36 weeks as a percentage of all live births).(10)
FINDINGS AND IMPLICATIONS
This study had more difficulty in finding strong associations between variables at the neighborhood level than those in the other sites, but it offers useful lessons that could lead to development and use of more effective indicators in the future.
Hypotheses and findings: Environmental hazards
The team had hoped to test the following two hypotheses in this area, but they were unable to construct reliable measures of hazards needed to test either.
The team had assembled a considerable amount of environmental data, as noted earlier. The problem was that after working with the community to think through the meaning of the data, they mutually discovered that none of this environmental data, as given, yielded useable indicators of hazardous conditions at the neighborhood level (i.e., conditions that would let them reliably rate the extent of hazards in one neighborhood in comparison to another). There were several reasons:
The team did run correlations relating the location of these conditions to the neighborhood distributions of children, poverty, and minorities, but they found no significant associations. Piton already produces "asset and risk factor" maps for Denver, and this analysis has allowed them to add a new map to the risk factor section (showing Superfund contamination plumes, CERCLIS sites, and confirmed environmental complaints--figure 3.1). Beyond that, however, interpreting the nature and extent of environmental hazards in neighborhoods will require deeper information on conditions at each location.
Hypotheses and findings: Community violence
There are many reasons to expect a higher incidence of violence in neighborhoods distressed on other measures. The Piton team sought to test the following hypothesis with the new indicators they had obtained.
This hypothesis was supported by most of the available indicators. As expected, there was a high correlation between the rate of violent crime and the other measures: 0.624 with respect to poverty rates, and poverty is closely correlated with the minority percentage of total population at the neighborhood level (0.658). Correlations with poverty were even higher for two of the new measures: 0.719 for the rate of confirmed child abuse and 0.703 for violence-related school suspensions and expulsions. The latter relationship is also directly observable by comparing the map for the latter variable in Denver (figure 3.2) with that for poverty (figure 8.3).
Piton advises caution in the use of these measures. Even though Piton was conservative in selecting suspension/expulsion reason codes as violence related, one couldn't be sure that there was not bias in assigning the codes initially. Staff errors or misjudgments are possible in child abuse/neglect designations as well. Furthermore, it is suspected that abuse/neglect cases are recorded more often for low-income families because, as beneficiaries of various Department of Human Services subsidy programs, they have more contact with city social service professionals. Furthermore, Piton and the communities see reported violent crime as only a very limited starting point to understanding. To gain enough knowledge to plan a sensible response, more information about the perpetrators, the victims, and the circumstances of these crimes is indicated.
Community process
Using the relationships forged through Denver Benchmarks,Piton is beginning discussions with the Denver Department of Safety, the Colorado Judiciary, and the Colorado Department of Corrections to seek new and deeper data that could result in more powerful indicators of neighborhood safety and risk for violence. Piton will also be pursuing more information about the circumstances of environmental hazards and how to measure them more meaningfully.
Work with the communities continues and does seem to be having an influence. One example is a shift in orientation of the work in the Overland neighborhood. That community had come together initially with the sole objective of securing the rehabilitation of a Superfund site. Once that objective was achieved, the group might well have disbanded. However, with their participation in the Benchmarksprocess (and following the approach being taken in the Cole neighborhood), they have now embarked on their own strategy to create a comprehensive plan for community change.
Finally, Piton has also been working with other agencies on the broader policy front. The City Council has recently approved a resolution in support of Denver Benchmarks. In the resolution, the Council urges city agencies to share data and otherwise cooperate with Denver Benchmarks, and use their data in policymaking and resource allocation, and the Council pledges to use its own data in decisionmaking.
[Go To Contents]
PURPOSE AND APPROACH
This section summarizes the site-specific analysis prepared by the Polis Center at Indiana University-Purdue University at Indianapolis (NNIP's local partner in Indianapolis) in conjunction with the Children's Health Services Research Program in the Department of Pediatrics at Indiana University.(11) It focuses on the relationship between neighborhood conditions and risk of childhood obesity.
Purpose
In the past two decades, the prevalence of obesity has risen dramatically (Anderson 2000). Concern about this rise centers on the link between obesity and increased health risks that translate into increased medical care and disability costs. In the United States, total costs attributable to obesity exceeded $100 billion in 2000, or approximately 8 percent of the national health care budget (Wolf and Colditz 1998). Although the immediate health implications of obesity in childhood have not been examined extensively, obese children are likely to become obese adults, particularly if obesity is present during adolescence (Braddon et al. 1986; Serdula et al. 1993). However, adverse social and psychological effects of childhood obesity have been demonstrated (Stunkard and Burt 1967; Stunkard and Mendelson 1967). Furthermore, being overweight during adolescence has been shown to have deleterious effects on high school performance, educational attainment, psychosocial functioning, and socioeconomic attainment (Gortmaker et al. 1993).
The purposes of this research were to learn about and measure relationships between the prevalence of obesity and socioeconomic status, the presence of exercise opportunities, and exposure to social barriers at the neighborhood level during the late 1990s in Indianapolis (Marion County).
Data sources and approach
The researchers had access to a database for this work that is nationally known for its comprehensiveness: the Regenstrief Medical Records System (RMRS), which contains data on patient circumstances and care reported by a large number of care providers and other health entities in Indiana (now with data on 1.5 million patients since 1974).
The researchers recognized that the pediatric data in the RMRS reflected a population in which African Americans, Hispanics, and patients receiving Medicaid are overrepresented. While the selection bias affects the generalizability of the findings, it also works to the study's advantage given that several U.S. minority populations are disproportionately affected by obesity, particularly African Americans, Hispanics, and Native American women (Strauss 1999).
From the RMRS, the researchers obtained data on a random sample of children, ages 4 through 18, who had been seen by primary care clinics in the Indiana University Medical Group in Marion County from 1996 through 2000 and for whom simultaneous height and weight measurements were available. They classified all children in the sample according to body mass index (BMI) categories. Obese children were those with BMI above the 95th percentile according to a scale developed by the Centers for Disease Control and Prevention. They also had data on the children's addresses and other characteristics such as race. After excluding 17 percent of the initial group (because of unreasonable data or geo-coding difficulties), they were left with a sample of 17,871 children. The sample reflected the general age and gender distribution of the patient population.
Data were analyzed at the block group level. Block group characteristics included income and other socioeconomic variables from the 2000 census and information on physical activity opportunities (e.g., YMCAs, parks, after-school physical education programs) and crime rates from the Social Assets and Vulnerabilities Indicators database (SAVI), the neighborhood data system maintained by the Polis Center).
Patient characteristics and block group environmental data were first subjected to bivariate analysis. The researchers then conducted multivariate logistic regression analysis using only variables that evidenced significant associations in the bivariate analysis. They also prepared and examined maps of the key variables involved.
FINDINGS AND IMPLICATIONS
Hypotheses and main findings
Figure 4.1 shows that there are important differences by race and gender. The prevalence of obesity is generally higher for females than males. Among females, 25 percent of Hispanics and 23 percent of African Americans were obese, compared with 20 to 21 percent for those of other races. Among males, 26 percent of Hispanics, 21 percent of whites, 20 percent of blacks, and only 14 percent of those of other races were obese. (It should be remembered that whites and African Americans are by far the dominant population groups in Indianapolis.) The researchers formulated the following hypotheses:
| Parameter | Estimate | Standard error | P-value |
|---|---|---|---|
| Intercept (Obese) | -1.4933 | 0.1034 | < 0.0001 |
| Female | -0.0713 | 0.0636 | 0.2627 |
| African American | -0.0464 | 0.0583 | 0.4261 |
| Hispanic | 0.3622 | 0.1297 | 0.0052 |
| Other races | -0.4255 | 0.2105 | 0.0433 |
| Female & African American | 0.1752 | 0.0793 | 0.0271 |
| Female & Hispanic | 0.0269 | 0.1879 | 0.8863 |
| Female & Other race | 0.5892 | 0.2833 | 0.0375 |
| Age | 0.0746 | 0.0049 | < 0.0001 |
| Age & gender | -0.0109 | 0.0012 | < 0.0001 |
| Extremely low income | 0.3654 | 0.1359 | 0.0072 |
| Very low income | 0.4380 | 0.1035 | < 0.0001 |
| Low income | 0.3857 | 0.0988 | < 0.0001 |
| Moderate income | 0.2960 | 0.1127 | 0.0086 |
| Middle & upper income | 0.2890 | 0.1123 | 0.0100 |
3. That children living in areas with high exposure to social barriers (as measured by high crime rates, single-parent families, and those who are linguistically isolated) are more likely to be obese. Working with the larger sample, with all measurements at the block group level, bivariate analysis showed that none of these factors was significantly predictive of obesity. The block group averages for crime densities (Part I crimes per square mile, 1996-2001) were 603 for normal children, 605 for overweight children, and 603 for obese children. Block group averages for single-parent families as a percentage of all households were the same (29 percent) for all three weight groups. Block group averages for linguistic isolation (percentage of families with English language difficulties according to the 2000 census) were also the same (2.2 percent) for all three groups.
Implications
That obesity is strongly associated with SES (as measured by income) is noteworthy. To our knowledge, this is the first large population-based study to examine environmental (rather than individual) SES as a predictor of obesity in children. Characteristics of neighborhoods almost certainly have significant effects on the behavior of residents. Neighborhoods are where people make connections to others and become part of a social network. Therefore, neighborhoods can be seen as generating social and cultural capital that represents concrete targets for interventions aimed at improving self-management. If we can design strategies to combat the deleterious effects of low environmental SES, then we stand to empower vulnerable populations to make healthy choices.
The evidence did not support the other hypotheses, but that does not prove them incorrect. Other recent studies have found evidence to suggest that features of the physical environment (e.g., presence of sidewalks, enjoyable scenery) can be positive determinants of physical activity, and that other environmental factors (e.g., high rates of crime, the lack of a safe place to exercise) can be negative determinants (Brownson et al. 2001; King et al. 2000). It is likely that the measure used (linear distance) was too crude a measure/proxy for access to play space. Research using other, more sophisticated measures related to the other hypotheses seems warranted.
Data potentials
This study demonstrates that a large ongoing repository of data on patients and their care, supplied by a very large share of all local care providers (e.g., the Regenstrief System) can be used productively for policy analysis and, by logical extension, for helping to design and monitor program initiatives that may result from it. Few urban areas have mobilized and maintained the broad-scale agreement of providers to report in this way. The development of similar systems in other metropolitan areas is worth exploring.
Community process
This research has been timely for Indianapolis. The Alliance for Health Promotion there is forming a collaborative Strategic Thinking Coalition to address the issue of obesity. Stakeholders involved include the mayor's office, the United Way, health organizations, neighborhood organizations, educators, fitness and nutrition experts, members of the media, and local foundations. While the mission and goals of this new group are in the process of being defined, it is clear that it will become a strong forum for motivating action on the issue over the long term. The Coalition has decided to focus on both children and adults, recognizing the importance of affecting change through the family unit.
The Coalition is now reviewing the local report for this study. It expects to use it as a base for targeting resources to neighborhoods as well as understanding the issue and its determinants more broadly. The Coalition will also be involved in decisions on follow-up research to support the design of effective interventions.
The Polis Center has also met with the United Way of Central Indiana (its partner in SAVI) to discuss the implications of this analysis for local health initiatives. United Way plans to use the results to guide policy development in two of its impact councils: on Community Health and Well-Being and on Children and Youth. Plans are also being made to disseminate the report and its findings broadly within the local medical community.
[Go To Contents]
PURPOSE AND APPROACH
This section summarizes the site-specific analysis prepared by the Urban Strategies Council in Oakland.(12) It focuses on the relationship between neighborhood conditions and tuberculosis infection.
Purpose
The Council conduced the work in close coordination with the Alameda County Public Health Department (PHD), which is designing community-based interventions to reduce the risk of certain diseases, including tuberculosis. New drugs discovered in the 1940s brought a tentative cure for this disease, and through 1984 its incidence declined nationally. There was striking reversal of the trend thereafter, however, associated with the AIDS epidemic, increasing immigration from abroad, and other factors (Cowie and Sharpe1998; Lillebaek et al. 2001; Talbot et al. 2001;). Tuberculosis today disproportionately affects minorities (Centers for Disease Control and Prevention1990), and there has been research and speculation about the spatial concentration of TB in distressed urban neighborhoods and poverty as a potential risk factor (Barr et al. 2001).
Council staff wanted to find out about the spatial pattern of tuberculosis in Oakland to help guide PHD in resource targeting and to better understand its association with variations in neighborhood conditions in the city (specifically, with minority and foreign-born populations, poverty, and overcrowded housing). The Council had initially hoped to conduct similar analyses with respect to sexually transmitted diseases (STDs) and AIDS in Oakland as well, but PHD data administrators so far have been unwilling to release the relevant data owing to confidentiality concerns (an issue discussed further later in this section).
Data sources and approach
The Tuberculosis Control Unit of PHD provided the Council with data on reported cases of tuberculosis in Alameda County from 1997 through 2001. Only the 456 cases with residences in the city of Oakland were used in the analysis. Data on the independent variables were from the 2000 census.
As with most of our other site-specific analyses, Council staff ran correlation analysis (bivariate and multivariate) to test their hypotheses, and mapped spatial patterns as well. However, they used an innovative approach to do so. Normally in this type of work, analyses are run on data sets with values for all variables provided for comparatively small administrative or data collection areas (e.g., census tracts or block groups). In this case, the staff applied data originally provided (using addresses to pinpoint locations for the 456 tuberculosis cases and the centroids of Oakland's 337 block groups for the census variables) to then estimate values of all variables for all of the cells in a much more finely grained grid (69,181 square cells with 150-foot sides).
The methodology for any variable involves interpolating between the point locations for the values given in the original data to estimate values for cells in between. The work was done using the "single kernel density routine" in the CrimeStat II software package (originally developed for identifying "hot spots" of criminal activity).(13) The approach leads to a more accurate understanding of spatial patterns. The traditional approach yields chloropleth maps like that shown in figure 5.1, with colors or tones of indicating uniform intensities in each census tract. The new approach yields isopleth maps with contour intervals, as in figure 5.2, that can show more accurately how patterns of intensity change within tracts without the constraints of arbitrary political or planning boundaries.
FINDINGS AND IMPLICATIONS
Hypotheses and main findings
The work focused on the Urban Strategies Council's analysis of three main hypotheses concerning relationship between neighborhood conditions and tuberculosis:
Accordingly, the analyses measured the relationship between the rate of incidence of tuberculosis (cases per 100,000 population including both males and females, or TB100K), the dependent variable, and six independent variables:
The bivariate analysis showed that the relationships between all of the predictors and tuberculosis were positive except for age. The strongest of these was the immigrant percentage, +0.64. Coefficients from the multivariate analysis are given in table 5.1. Summary statistics are as follows:
R Square and Adjusted R Square = 0.481
Std. Error of the Estimate 719.081
Durbin-Watson = 1.625
Change Statistics:
| Unstandarized coefficients | Standardized coefficients | 95% confidence interval -B | |||
|---|---|---|---|---|---|
| B | Standard error | Beta | Lower bound | Upper bound | |
| (Constant) | 3127.74* | 60.17 | 3,009.80 | 3,245.68 | |
| Per_male | 2966.19* | 137.67 | (0.083) | (3,236.03) | (2,696.35) |
| Logit of Per_pov | 189.35* | 4.57 | 0.223 | 180.39 | 198.31 |
| Logit of Per_nonw | 37.80* | 3.27 | 0.062 | 31.38 | 44.21 |
| Logit of Per_imm | 635.07* | 5.91 | 0.505 | 623.48 | 646.65 |
| Square of c_crowd | 277.15* | 23.86 | 0.081 | 230.38 | 323.92 |
| Square of per_65 | 7725.61* | 323.76 | 0.193 | 17,091.04 | 18,360.18 |
| * Significant at the .0001 level | |||||
In sum, 48.1 percent of the variance in tuberculosis in Oakland can be explained by the six variables, indicating that this is a strong model. Immigration is the single most important variable that predicts tuberculosis, followed by poverty.
Mapping analysis
As noted in section 8, Oakland is not like the prototypical big city with high-poverty neighborhoods all clustered around the city center--its spatial pattern is much more complicated. The staff began the analysis by preparing a series of chloropleth maps of the incidence of tuberculosis (like the map at the tract level in figure 5.1) for various types of administrative areas: City Council Districts, PHD Health Team Areas, and ZIP Codes, in addition to census tracts. Because of the way the areas are configured, each map suggested that different areas of the city were experiencing the highest incidences of tuberculosis. Also, it was apparent that the areas were generally too large to provide useful information to guide intervention and prevention efforts.
The map in figure 5.2 is an isopleth map of the estimated density of the risk of tuberculosis infection. It shows spatial patterns in a more precise way, indicating variations within the larger geographies noted above. Therefore, it should support operational planning and management more effectively. While earlier maps highlighted fewer and larger geographies, the risk map shows that downtown, Chinatown, and neighboring areas reaching into West Oakland, as well as the Lower San Antonio and Fruitvale areas southeast of Lake Merritt, are where prevention efforts should be targeted. This analysis shows that combining regression analysis with kernel density estimation can be a powerful way to analyze disease infection geographically.
Implications and community process
As of this writing, the uses of the Urban Strategies Council's report are not fully planned, but some steps have been taken and others agreed to. The central staff of PHD have been briefed on the results of the analysis and its significance. A considerable amount of PHD work in the community occurs through Neighborhood Health Teams, and there is agreement that meetings will be held to discuss the findings with the Teams. These discussions will highlight the opportunities for action with the Teams that work in the neighborhoods identified in the analysis as having a high risk for tuberculosis. Also, a community-driven Health Working Group has recently been established in the Lower San Antonio area. The Urban Strategies Council has been working closely with residents and leaders in that area for years, as a part of the Annie E. Casey Foundation's Making Connections initiative and in other ways. Accordingly, special efforts will be made to brief the new Health Working Group on the report and its implications.
[Go To Contents]
PURPOSE AND APPROACH
This section summarizes the site-specific analysis prepared by the Providence Plan, working closely with the state of Rhode Island Department of Health (HEALTH).(14) It focuses on the relationship between residential mobility and childhood health and the way neighborhood conditions influence that relationship.
Background and purpose
A sizeable number of studies have established that highly mobile children (those who move from one house to another much more frequently than average) face serious educational disadvantages. These include poor verbal abilities, poor attendance, lower test scores, and repeated grades (see, for example, Buerkle 1999; Koehn 1998; U.S. General Accounting Office (GAO) 1994). It is believed that residential mobility disrupts learning because of the emotional and behavioral difficulties that accompany it. Children who move often are placed under more stress because of the loss of friendships and other social support systems.
While there has been less research on the topic (exceptions are GAO 1994; Morrow 1995), it seems reasonable to hypothesize that high mobility also has deleterious effects on children's health. The authors of this study wanted to find out the extent to which this was true in Providence, and how mobility and health effects varied with characteristics of local neighborhoods.
Data sources and methods
To conduct this study, the Providence Plan had access to unusually complete data on mobility and other aspects of the lives of children in Providence. First, for some time the staff have been conducting demographic and other analyses for the Providence School Department using its database, which now contains historic data on all students enrolled from 1987 through 2001. Information about each student includes all address changes as well as data on test scores, absenteeism, repeating of grades, and other indicators of educational performance.
Second, they were more recently given access to the state Department of Health's KidsNet Databases, which at the time of this project contained information on all children born in Rhode Island from 1997 through 2001--a group overlapping, but generally younger than, those in the School Department dataset. Each of the children has been tracked since birth, so this file also contains information on address changes as well as on birth-related measures (e.g., birth weight, prenatal care) and selected services by various local health care providers (e.g., well-child and illness-related visits to physicians, immunizations, and lead screenings).
In addition to these sources, the staff used various tract-level indicators from the U.S. census and local administrative records (e.g., on reported crime). Relationships were examined using tables, charts, and census tract maps.
FINDINGS AND IMPLICATIONS
Hypotheses and main findings
Introductory analysis showed that mobility was indeed high among children in Providence and Rhode Island. Information from the School Department database showed that on average, one-quarter of all students in Providence changed addresses at least once in any given year. Information from the HEALTH database showed that nearly one third-of the 65,800 children for whom there were records state-wide had moved at least once by their first birthday; 43 percent of all children born in 1997 had moved at least once by the end of 2001. Children born in the core cities of the state had the highest mobility rates. For example, 22 percent of Providence children had moved two times or more from 1997 to 2001, compared with 7 percent for those outside of the core cities.
Providence Plan staff explored a large number of hypotheses related to these measures, including some to reconfirm the problematic effects of high mobility on educational outcomes. Here we regroup the findings and report on three main hypotheses that emphasize health effects.
1. That young children who move often will have more disruptions to health care access than other children. Staff defined and examined three ways in which access might be disrupted: (1) when children have to shift from one care provider to another; (2) when children have fewer office visits for immunization; and (3) when children do not receive recommended blood level screenings in a timely manner. Results were mixed:
Figure 6.1:
Frequency of Visits to Service Provider and Number of Residential Moves
between 1997 ad 2001 for all Children born in 1997, Providence, RI
2. That children born to disadvantaged women, women of color, and single women are likely to move more often than other children. In this analysis, the staff defined "mobile" children as any born between 1997 and 2000 who had moved two or more times by December 2001 and any born in 2001 who had moved at all in that year. All results supported the hypothesis:
3. That neighborhoods with high levels of distress will also have higher rates of child mobility. For this analysis, Providence Plan staff identified 13 measures of distress and counted the number of times each census tract's score exceeded the mean for each measure by one standard deviation or more(16). The tracts that met this standard for at least two of the measures were classified as "distressed." These tracts, shown on the map in figure 6.2, are clustered to the west and south of downtown Providence and heavily overlap the high-poverty tracts mapped in section 3.
Mobility data support the hypothesis: (1) 92 percent of the distressed tracts have student residential mobility rates above the city average between 1996 and 2001; (2) the same share (92 percent) had residential mobility rates for young children above the city average (children age 5 and under as identified in the health database).
Looking at this relationship from the other direction, the staff also showed that neighborhoods with high rates of child mobility (see figure 6.3 related to younger children) have higher levels of neighborhood distress with respect to health. The seven tracts with the highest rates of child mobility (tracts 4, 5, 8, 9, 10, 11, and 19) are above the city's average on a number of measures of distress, including births to women with delayed or insufficient prenatal care; births to single women; risk-positive births; elevated blood lead levels; and violent crime.
Implications and community process
High rates of mobility are indeed prevalent among children from distressed and minority families and in Providence's most troubled neighborhoods. While the effects are not dramatic, they also appear to have a negative impact on health care. Coupled with the Providence Plan's previous research showing negative effects of high mobility on education outcomes, there is now substantial evidence to draw attention to the issue.
Staff members recognize that mobility cannot be lessened in isolation from the myriad of other problems facing distressed families. There is need to better understand how the circumstances of poverty and neighborhood-level stressors interact to cause high mobility and other problematic impacts on families and children, and then to approach the mix of problems holistically. Next steps have now been planned in conjunction with HEALTH.
First, now that high-mobility children have been specifically identified, it will be possible for trained "parent consultants" (e.g., peers) employed by HEALTH to have targeted discussions with the families both to find out more about why they move and raise awareness of deleterious effects. Providence has a system of grass-roots community health centers (operated by nonprofits), and this work will be coordinated through them. Consideration is also being given to involving an ethnographic researcher from Brown University to help understand the dynamics of the issue.
After better information is available, HEALTH will reassess how well all of its services meet the needs of mobile families. For example, families of risk-positive children are offered home visiting and early intervention services. The report showed that two-thirds of highly mobile children were identified as risk-positive at birth. Assessment of how well the home-visit approach is working and how it might be modified to better address mobility issues would be a priority. Starting from data in this report, further work on locational patterns would also appear warranted to allow HEALTH to develop a better system of indicators so it can more efficiently target resources to those most in need.
In addition, plans are being made to present the findings of this report to the staffs of housing agency and advocacy groups. It is recognized that growing housing affordability problems play a central role in expanded mobility for low-income families. Housing professionals need to be working more closely with community health centers and other grass-roots groups if the issue is to be addressed.
Data potentials
The detailed schools and health data systems available to the Providence Plan for this work are not readily available for policy analysis in most American cities, if they exist at all. However, they offer models that should be attractive to develop elsewhere. In particular, the KidsNet Databases (birth records coupled with records on subsequent health system interactions for each child) would certainly seem worth emulating. The potential of such databases to support effective program planning, resource targeting, and internal performance monitoring, as well as providing a basis for external accountability, should be substantial.
This analysis built on an existing, working relationship between the Providence Plan and HEALTH has enabled the organizations to jointly explore issues of concern to the health community for several years. Appropriate data sharing and confidentiality agreements have promoted the relationship. The Providence Plan had no problems in accessing the data and found the records to be orderly and reliable in general.
5. The study is fully documented in an as yet unpublished report of the Center on Urban Poverty and Social Change Mandel School of Applied Social Sciences, Case Western University: Neighborhood Health Indicators in Greater Cleveland: The Use of Medicaid Claims to Measure Children's Access to Primary Care, by Claudia Coulton, Kristen Mikelbank, Katherine Offutt, Engel Polousky and Siran Koroukian. 2002.
6. For a full discussion of the criteria and locations of Health Professional Shortage Areas, see the Health Resources and Services Administration web site at http://bhpr.hrsa.gov/shortage/.
7. The study is fully documented in an as yet unpublished report of the Piton Foundation: Indicators of Community Environmental Hazards and Community Violence in Relation to Neighborhood Conditions, 2002.
8. Three years of data had been provided by the school district, but analysis by Piton evidenced problems in the data for the first two years that made them unusable.
9. This rolling three-year average approach is the same method we used with vital records data in section 4. Piton staff also explored the idea of using the Confirmed Child Sexual Abuse Rate and the Rate of Deaths Due to Violent Causes, but the numbers of cases were too small to support neighborhood analysis.
10. Piton also asked the state's Department of Health and Environment for a file linking births data to Medicaid records for mothers and children eligible for that program. This would have supported a range of other types of analysis, including examination of service frequencies such as that done in Cleveland (section 6). The Department is attempting to create such a file, but it could not be completed in time for research in this project.
11. The study is fully documented in an as yet unpublished report of the Polis Center at Indiana University-Purdue University at Indianapolis: Indianapolis Site-Specific Neighborhood Health Analysis: Environmental Factors and Risk of Childhood Obesity, 2002, by Sharon Kandris and Gilbert Liu, 2003.
12. The study is fully documented in an as yet unpublished report of the Urban Strategies Council, Tuberculosis Infection in Oakland, CA 1997-2001: At-Risk Populations and Predicting "Hot Spots" An Analysis Method Combining Kernel Density Estimation Mapping with Regression, 2002.
13. The method of interpolation was the normal distribution. The choice of bandwidth was that of the fixed interval and their fixed interval was 0.253 square miles, the mean nearest neighbor distance between block group centroids in Oakland.
14. The study is fully documented in an as yet unpublished report of the Providence Plan: Development and Use of Neighborhood Health Analysis: Residential Mobility in Context, 2002. The work received support from the Rhode Island Department of Health as well as from this project.
15. The Rhode Island Department of Health establishes risk for developmental delay for all newborns based on a variety of maternal health and socioeconomic criteria and vital statistics gathered at birth.
16. The thirteen measures of distress were: student residential mobility, early childhood residential mobility, births to women without a high school degree, births to teens age 15 to 19, births to women receiving delayed prenatal care, births to women receiving insufficient prenatal care, low birth weight infants, lead exposure among children under 6 years of age, Part I violent and property crime rate, vacant dwelling units, renter-occupied housing units, and median family income.
Main Page of Report | Contents of Report
Home Pages:
Human Services Policy (HSP)
Assistant Secretary for Planning and Evaluation
(ASPE)
U.S. Department of Health and Human Services
(HHS)