ISSUES AND RECOMMENDATIONS
Perhaps more important than anything else, the findings of this study reconfirm the premise that motivated it; namely, that neighborhoods do indeed make a difference for health outcomes and health policy. From our cross-site analysis, we can say that, although neighborhood correlation coefficients did decline in the 1990s in a few cases, patterns of association remain very strong overall. And even though the gaps seem to have diminished over the past decade, the most striking finding is that the health problems of high-poverty neighborhoods remain substantially more serious that those of non-poor neighborhoods in all cities for which we have data.
The implications for health programs are critical. Since conditions differ markedly by neighborhood, 'standardized solutions' applied uniformly in many different types of neighborhoods are unlikely to work everywhere and are likely to be wasteful. Targeting the right services to the places that really need them and adjusting delivery strategies in response to neighborhood differences should both enhance payoffs and may save money. At a time when resources are scarce, it would seem that more emphasis on taking neighborhood variations into account is warranted in public health programming.
We believe that this study has also demonstrated some promising and cost-effective ways for public health agencies and their partner organizations to design and implement more customized approaches. Three things in part account for that: the development of large computer-based information systems, technology that has dramatically reduced the cost of manipulating data, and new institutions and groups of professionals that have learned how to do so in an efficient and practical manner.
We also conclude that the completion of this study is a promising sign for cross-site analysis of changing health conditions at the neighborhood level in the future. A decade ago, obtaining comparable year-by-year data for so many indicators in five cities would not have been feasible (and even at this point we could not obtain data going back to the early 1990s for all five). But more and more cities are developing NNIP-type capacities and building year-by-year data series from vital records and other sources as they go along (see Annex A). Even a few years from now, it seems likely that cross-site analysis like this covering a much larger number of cities will be possible.
And future monitoring in this regard is important. Data already exist on health trends for cities as a whole, but where there are great internal disparities across neighborhoods (as is normally the case) citywide averages can be very misleading. National policy makers need richer information across many cities to realistically understand how health conditions are changing in America and, thereby, how to target interventions more efficiently.
In the remainder of this section, we first look at the prospects for expanding the development of neighborhood-level information systems for use in the health field. We then consider the implications of this work for policy and program development. While many details of what we present may be new, we believe the general approach is very consistent with the National Committee on Vital and Health Statistics' "Strategy for Building the National Health Information Infrastructure" (U.S. Department of Health and Human Services 2001).
NEIGHBORHOOD INFORMATION SYSTEMS AND PUBLIC HEALTH
The local partners in NNIP have all been able to build sizable neighborhood information systems for their cities, but until recently their holdings related to health have been comparatively weak. We believe the experience of this project shows that local partners will be able to expand both their data and analytic work in this field, and that this sort of capacity can be developed elsewhere as similar institutions are established in other cities.
In almost all of the case studies summarized in part 1, the partners faced difficulties in obtaining the data, editing them, and working with them analytically. But that is no different than what the partners faced in building the other components of their systems to date. In each case, progress was made and permanent health-related data expansions either have happened or seem very likely to happen over the coming year. In the process, positive relationships were established with data providers that should facilitate further expansions and joint analytic work in the future. In the paragraphs below, we assess prospects and offer recommendations for further systems development in four areas:
- Contextual variables
- Health conditions
- Health facilities and services
- Integrated health data systems
In this review, we often rely on an important resource that we recommend to all who want to learn more about potentials for local data: Claudia Coulton's 1997 catalog of data sources that are typically available at the local level and contain information that can be presented at the neighborhood level.
We have already referred to the wide variety of contextual information that has been assembled for small areas by the NNIP partners. The range is summarized in Kingsley (1999), and Coulton's review (1997) covers a yet broader set of opportunities.
In this project, the local partners in the five sites primarily relied on contextual variables from the census or from files that were an established part of their ongoing systems. There were two notable exceptions. The Providence Plan successfully developed small area indicators on children's mobility (section 6). This measure meets a data gap in most NNIP systems: better measures for the comparative stability of neighborhoods. We certainly recommend the development of similar measures elsewhere, but remember that they are only possible if local school and/or health systems maintain records that track address changes for individual children. Although inquiries should be made, we doubt this will be possible in many other cities at this point. In particular, it would be difficult for cities that have several separate school districts.
The second exception is Denver's Piton Foundation, which obtained new data on a series of environmental conditions and added indicators on violence related school suspensions and expulsions, as well as child abuse and neglect. Acquisition costs for the environmental data were not high but Piton was unable to use the data to create a reliable measure of "hazards." Nonetheless, the spatial patterns for some of the conditions were judged to be worth knowing about and were added to Piton's collection of maps of neighborhood "risks." We think it would probably be sensible for intermediaries in other cities to review the Piton experience (section 3) and purchase those data sets that seem appropriate for their needs. As to the other measures, we recommend efforts to acquire child abuse and neglect data (several other NNIP partners have done so) and the school suspension/expulsion measure (although, its viability would depend on the status of the local school districts' information systems).
Health conditions and health care utilization
The availability of neighborhood-level data on health conditions is low to nonexistent in most cities today. The NNIP partners all regularly assemble and report on indicators from vital statistics files. The benefit of these data is they are legally mandated--all births and deaths, along with the residential addresses of the mother or the deceased, are required to be reported--and standard definitions are normally employed virtually everywhere. These events can be geo-coded and sorted to develop indicators by neighborhood--indicators like those we examined for the five selected sites in sections 8 and 9 on rates of teen births, low-birth weight births, infant mortality, and age-adjusted deaths. Yet in most cities, even these measures are not recurrently calculated and published at the neighborhood level, as illustrated in parts of this study.
And for work in health, professionals, advocates, and neighborhood groups would like to know much more; particularly changes in the neighborhood-by-neighborhood incidences of various health problems such as asthma, tuberculosis, and AIDS. At this level, even the NNIP partners typically have little to offer so far. These conditions are not as consistently reported to public agencies as births and deaths. (43) Nonetheless, there are some sources for additional health data and NNIP partners expect to be taking more advantage of them in the future. Privacy concerns involved in utilizing some of these sources are addressed in a discussion of the Health Insurance Portability and Accountability Act (HIPAA) immediately following this section. As identified by Coulton (1997), these sources include the following:
1. Communicable Diseases. Diseases caused by the direct or indirect spread of pathogens from one person to another are called communicable diseases. Tuberculosis, syphilis, and AIDS are examples. By law, physicians are responsible for reporting incidences of many communicable diseases to local and state health officials. Although not always computerized, these data are maintained by departments of health and usually contain addresses that might allow small area analysis. However, confidentiality and generalizability issues require special justifications for release of such data. Furthermore, rare events concerns must often be addressed. Indicators that could be developed include the incidence in a given area, and the classification within the specific group of disease. Some information about emerging and other communicable diseases can be found at the WHO web site, http://www.who.ch/programmes/emc/emc_home.htm. The Center for Disease Control publishes information on communicable diseases in selected metropolitan areas at http://www.cdc.gov/publications.htm.
2. Emergency Medical Services. Emergency Medical Services (EMS) are services delivered with a sense of urgency to those in need of immediate attention. Accident victims and patients having heart attacks, for example, need immediate medical attention. When transport is via the public system, the reports will appear in 911 data, which is being accessed by NNIP partners in some cities. But many medical emergencies treated in emergency rooms do not appear as 911 calls. So emergency room records could be a more complete measure. While there is no common database for emergency room visits in most cities, interest in injury surveillance is considerable (Centers for Disease Control 1988). The availability of E codes in the International Classification of Disease System makes it possible to establish data systems, and a growing number of communities are exploring such systems. When many emergency rooms in a city collaborate on an injury registry system, it is possible to calculate injury rates for small areas. Among the important indicators are rates of intentional and accidental injury by age group (Rivara, Calonge, and Thompson 1989). Among youth in particular, injuries are a good indicator of health risk as well as social control in a community (Prothrow-Stith 1991). Finally, Dr. John Billings, Director of New York University's Center for Health and Public Service Research, has developed an algorithm for analysis of emergent versus non-emergent use of emergency rooms now being used in several communities (Billings 2002).
3. Immunization. The population's status on immunizations is considered an important measure of the adequacy of preventive health care. It is not only the protection afforded by the vaccine that is of interest but also the accompanying chance for a medical professional to examine the young child who is receiving immunizations according to the prescribed schedule. There is no law requiring states to collect data on immunization nor can the information included in the National Immunization Survey be used to access this data. Some states conduct a survey and estimate the number of children immunized and some communities are experimenting with computerized immunization tracking systems. Schools and day care centers are required to ask for proof of immunization before they admit students, so school district data systems may be a source of data on this topic. Some neighborhood indicators can be developed from available data. The immunization rate of children can be calculated to determine what percentage of children are not covered at time of entry into school or did not receive their vaccinations at the appropriate age. More information about immunization can be accessed through the Centers for Disease Control and Prevention's National Immunization Program web site, http://www.cdc.gov/nip/default.htm.
4. Hospital Discharge Files. Hospital discharge files contain information on hospitalizations generated when a patient is discharged. State hospital associations or government agencies may maintain such files. Although there are some common conventions for such data, they vary across locales. Many state hospital associations maintain and publish their own data on patients aggregated at the ZIP code level. Data about age, payer, clinical service, sex, length of stay, Diagnostic Related Group (DRG), hospitals, beds, admissions, and the like may be available for small geographic areas such as a ZIP code. Patient-level data exist but are guided by confidentiality issues for release of information. However, the Agency for Healthcare Research and Quality (AHRQ) is building a standardized health data system that houses longitudinal and administrative state-specific discharge information and a national sample of discharges from community hospitals. AHRQ's Healthcare Cost and Utilization Project (HCUP) is a federal-state-industry partnership that maintains hospital discharge abstracts for 29 states. Five databases, including an interactive web-based site, allow users to analyze clinical and nonclinical information that has been translated for cross-sectional comparisons between states. HCUP also distributes software tools for use on its data and other administrative databases. HCUP resources can be found at http://www.ahrq.gov/data/hcup/. Also, the American Hospital Association publishes Hospital Statistics, which provides some hospital information along with the address of the facility. Some small geographic area indicators that can be developed are average charges by severity, number of inpatients and outpatients, incidence of most prevalent preventable conditions per 1,000 people, average length of stay, number of beds in a small geographic area, and number of hospitals in a small geographic area. Researchers have compiled utilization rates across geographic areas using discharge data (Wennberg, Freeman, and Culp 1987). The National Association of Health Data Organizations (NAHDO) has addressed some of the data-related issues on its web site, http://www.nahdo.org/index1.asp.
5. Medicaid Claims. Medicaid provides medical assistance for certain individuals and families with low incomes and resources. As we learned from the work in Cleveland (section 6), claims filed by medical providers for services delivered under Medicaid may be a valuable source of data on medical conditions and services for populations in small geographic areas. Although Medicaid program administration varies from state to state, Medicaid claims data can likely be obtained from the state agency that administers the program. However, utilization data may be more difficult to obtain if a state's Medicaid program relies on managed care where payments are to plans on a per member per month basis. Data may include provider description, classification of illness, procedure codes, service dates, and service charges. The recipient's address, needed for small area analysis, may need to be merged into the claim from an eligibility file. Most states only enter a limited number of variables into the computerized system. Owing to confidentiality issues special requests justifying need may be necessary for release of data.
Assuming the availability of claims data, small area indicators could be developed from such files including annualized rates of types of medical care utilization (emergency, inpatient, ambulatory, etc.), annualized rates of Medicaid utilization by health status and age, and Medicaid utilization by type of service. Statistical information about Medicaid at the national level can be viewed at the 'Centers for Medicare and Medicaid web site http://cms.hhs.gov/medicaid/. Additional national-level Medicaid-related statistical information such as type of utilization can be found at http://www.census.gov/prod/1/gen/95statab/health.pdf.
6. Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) Files. Local WIC program data files on individual clients contain a great deal of information about health and other circumstances of vulnerable families and children. NNIP's partner in Milwaukee has already performed basic analyses of WIC files locally. They plan further analysis looking at the WIC data in relation to childhood obesity and socioeconomic conditions, nutritional status, and barriers to exercise participation.
7. Youth Risk Behavior Surveillance System (YRBSS). The YRBSS was developed in 1990 to monitor priority health risk behaviors that contribute markedly to the leading causes of death, disability, and social problems among youth and adults in the United States. These behaviors, often established during childhood and early adolescence, include: tobacco use; unhealthy dietary behaviors; inadequate physical activity; alcohol and other drug use; sexual behaviors that contribute to unintended pregnancy and sexually transmitted diseases; and behaviors that contribute to unintentional injuries and violence. The YRBSS includes bi-annual national, state, and local school-based surveys of representative samples of 9th through 12th grade students. CDC provides funding and technical support to states, territories, and major cities to conduct a Youth Risk Behavior Survey (YRBS) every two years. CDC's technical assistance includes: training for state and local coordinators; specialized software to guide states in selecting schools and classes; help with applying survey results to improve school health programs and policies. Sites can add or delete questions in the core questionnaire to better meet the interests and needs of their area. School-based surveys were last conducted in 2001 among students in grades 9-12 in 38 states, 19 large cities, and 7 territories. YRBSS, is fully described at the CDC web site, http://www.cdc.gov/nccdphp/aag/aag_yrbss.htm.
Health facilities and services
Considerable information is also now available in automated form on the locations of health care service providers at various levels. First there are data on physicians by address and specialty (from the American Medical Association) and on dentists by address and specialty (from the American Dental Association). Second, national data on hospital locations (and some characteristics) are available from the American Hospital Association. Finally, there are two national databases on community-based health providers. The Bureau of Primary Health Care keeps aggregate data at the state level for federally funded community health centers and their users in the Uniform Data System (UDS) located at http://www.bphc.hrsa.gov/uds/. The Urban Institute also maintains information on community health centers and other nonprofit providers from a national database on all nonprofit institutions that are required to report financial information to the IRS (non-religious organizations with $25,000 or more in gross receipts). Institutions are classified by the work they do and several health- related categories are included on the list.
Via geo-coding of addresses, most of these files permit the user to count the number of providers of various types in small areas such as census tracts and to calculate rates (e.g., number per 1,000 people). In a few cases, some characteristics of the provider are included as well. For example, in the Urban Institute's nonprofit database, information on the size of the organization in financial terms is available from IRS returns. However, little such information is available in automated form. Users need to recognize that the number of providers in a neighborhood does not by itself provide a reliable measure of the quality of care available to its residents. Still, knowing this much is an important step toward finding out more.
Combined local data systems
Perhaps the most promising developments for work of this type noted in the Site-Specific Analyses were the Indianapolis area's Regenstrief Medical Records System (RMRS, discussed in section 4) and the Rhode Island Department of Health's KidsNet Database (section 6). Both systems are based on agreements among large numbers of local care providers and agency officials to share a wide range of information about individual patients. The agreements, of course, include detailed rules and procedures designed to protect client confidentiality. Records for individuals may include data on birth outcomes, and then information about a string of subsequent interactions with providers (e.g., inoculations, lead screenings, and treatments for various illnesses and diseases).
The Indianapolis and Providence cases, as well as similar efforts being implemented in other communities, illustrate the substantial potential value of these systems. Analysts can cut across case records to examine the spatial patterns of particular health problems (such as obesity) and they can see how those patterns are changing over time. Knowing both the pattern and the trends should not only enable them to target responses more efficiently in the short term, but also to learn more about why the problem develops as it does. They can also analyze various patterns of service performance. Adding links to Medicaid files where possible, like those analyzed in Cleveland, would add to the power of these systems to support more effective health program implementation and planning.
Based on these findings, we suggest that federal and state agencies collaborate with local players in the health field to continue to work toward the development of similar systems in other metropolitan areas.
Addressing confidentiality issues
Advanced information systems have substantial power to improve the performance of health initiatives but they contain sensitive information about individuals that could be very troubling for them if made public. It is not surprising that agency officials responsible for large data files like these are nervous about releasing them to outside organizations. The NNIP partners have established good working relationships with the agencies that provide data to them based on the partners' solid track record in honoring their pledges to protect the confidentiality of any records on individuals they use. Nonetheless, agencies are sometimes wary about releasing such information even to these organizations.
The Health Insurance Portability and Accountability Act of 1996 (HIPAA) represents a major development in the federal regulation of privacy of individually identifiable health information, and will govern the ability of public health authorities that are entities covered by the Act to share individual level data with data-intermediary organizations like the NNIP partners. The Department of Health and Human Services released the final regulations regarding privacy in December 2000. Entities covered under the act (certain health care providers, health plans, and health care clearinghouses that use electronic transactions for which the Secretary has adopted a standard under HIPAA) must be compliant by April 2003.
Under HIPAA, a covered entity must protect the privacy of a person's identifiable health information (PHI) (44). Specifically, a covered entity must:
- Tell the person how their health information will be used or disclosed
- Obtain express permission whenever a use or disclosure is not permitted or required by the rule
- Share only the amount of information necessary for disclosure
- Enable the person to revoke permission at any time.
The regulations permit entities to release limited data sets (those not completely de-identified) with a "Data Use Agreement." The limited set can include dates of admission, discharge or other services; dates of birth or death; age of participant; and some larger geographic information (such as zip code, city, county). The agreement specifies the permitted use of data and prohibits re-identifying or contacting the individuals. In addition to the limited data set, the rule permits release of de-identified information - information can be determined de-identified statistically or by the removal of 18 specified identifiers.
While the regulations govern the disclosure from covered entities of the individual data to an outside entity, they are not intended to interfere with public health surveillance and interventions or the collection of data by public health authorities such as vital statistics or disease registries for the purpose of disease prevention. However, NNIP partners report that some local officials are more guarded about the release of this data in light of HIPAA. As with any new regulatory regime, there are questions and implementation concerns that will need to be addressed. HIPAA will certainly affect the process for sharing public health data, but at this point it is unclear what the long-term implications of HIPAA will be for local data intermediaries. More information about HIPAA is available at http://www.hhs.gov/ocr/hipaa.
HIPAA is one tool for addressing legitimate concerns about confidentiality of medical information. With the increased technological capacity to link data about individuals from many different sources there are good reasons for officials to be cautious. However, we believe many are not sufficiently taking into account the fact that the new technology also provides the basis for stronger confidentiality protection. A simple example is the new approach to mapping used in the Oakland study. With the old system, providing data on a small number of individuals in a small area could well constitute a confidentiality breach, but with the isopleth data, it would be impossible to trace information back to individual cases. Analysts have found other statistical methods that allow researchers to release more detailed information in ways that cloud links to individuals (Lane et al. 2002).
Implications for Program Development
As mentioned at the beginning of this section, our most important finding may be that neighborhoods still make a significant difference. There were indeed important improvements in health outcomes in America's cities in the 1990s, but the health problems of high-poverty neighborhoods remain substantially more serious than those of non-poor neighborhoods in all cities for which we have data.
Addressing health problems in these areas is a well-recognized national priority, particularly since doing so is likely to have positive impacts on other national objectives such as improving educational outcomes and moving a higher share of welfare recipients into employment. This study was not expected to contribute to the debate about national health policy alternatives, but it does have implications for local programs and policies. In the remainder of this section, we discuss those implications first and then offer some ideas about how state and federal policies could better promote the themes we suggest for local action.
Implications at the local level
The first theme at the local level is the potential of the spatial differentiation of interventions. As suggested earlier, since conditions differ markedly by neighborhood, "standardized solutions" applied uniformly in many different types of neighborhoods are unlikely to work everywhere and are likely to be wasteful. Targeting the right services to the places that really need them and adjusting delivery strategies in response to neighborhood differences should both enhance payoffs and perhaps save money. Our second theme is to emphasize the notion of using data and analysis not just to evaluate, but also to motivate and manage change--for example, the potential use of Cleveland's and Providence's data on past service provision to improve future scheduling and targeting.
County health agencies and other entities involved in public health at the local level (public and private) should help build capacity needed to develop and expand information systems like those established by the partners in NNIP. To be helpful to health professionals, the system needs to cover many topics beyond those directly related to health. Therefore, health professionals may need to partner with other local groups in the process of system building.
Such systems will provide information to inform strategies that take neighborhood conditions into account. This information should provide a factual basis for designing alternative approaches in at least four interrelated types of interventions:
1. Prevention. Prevention most often means efforts to increase awareness of health problems among local residents and to offer education about hazards, harmful behaviors, and service/resource options. In cases like these, it is critical to take neighborhood differences into account because local culture is usually the key to getting attention and establishing confidence in case managers and service providers. This often implies working through community groups, and offering new data and analysis can be an effective way for public agencies to establish positive relationships with such groups (see examples in National Neighborhood Indicators Partnership 1997). Fresh analysis can both motivate interest and give grassroots groups and local nonprofits clues about how they should proceed.
2. Strengthening Health Care Services. The availability of facilities and services can vary markedly across neighborhoods. Residents do not need to be able to receive all care they need within their own neighborhoods, but they need access to good quality services within reasonable distance from their homes. Given their lower automobile ownership rates, physical proximity is more important for the poor than for the rest of society. Nationally, the Area Resource File (ARF) updated by HRSA provides data on comparative service availability at the county level. Such information is hardly ever available at the neighborhood level, even though inter-neighborhood variations in service resources may well exceed inter-county variations.
The capabilities of the NNIP data intermediaries, as evidenced in sections 2-6 of this study, suggest that it may now be possible to build credible local service inventories at the neighborhood level. At the minimum this would identify the locations of various health service providers, but available descriptive should be included for each: e.g., budget and staffing levels, size and geography of service areas, measures of service quality. Without better quantitative measures of differences in service levels across different parts of the city, it is very difficult for anyone (public health officials or interest groups) to know how to allocate resources in a cost-effective manner. With such data, there should be less disagreement about what and how much should be allocated where.
3. Reducing Hazards and Stressors. High rates of violence and other crimes, environmental hazards, deteriorated housing, and public infrastructure have all been shown to have deleterious effects on health. And once again, the incidence of each varies dramatically across a city's neighborhoods. Some of these problems are hard to measure reliably (witness Denver's effort trying to collect data on environmental hazards), but NNIP partners are generally expanding their data in these areas, and the development of a credible ongoing set of small-area indicators along these lines would seem a realizable goal for most cities over the coming decade.
Effective interventions to reduce such hazards and stressors, however, are not easy to develop and they certainly do not fall solely under the control of health professionals. To help motivate progress in these areas, public health officials need to affiliate with broader partnerships for neighborhood improvement--partnerships that should include representatives from resident associations as well as other civic groups and public agencies, including law enforcement.
In this area too, data can be one of the drivers. Knowing reliably how one neighborhood compares with others on these indicators, knowing which type of hazards and stressors warrant the highest priority in different neighborhoods, and knowing specifically where within the neighborhood these problems occur can both motivate interest in responsive actions and design actions that are likely to work in the specific environment at hand.
4. Program Monitoring and Management. Cleveland used Medicaid claims data to show how the timeliness of children's preventive health care visits differed in the city's neighborhoods. Providence used a local integrated health database to track similar measures at the neighborhood level. If senior officials had easy-to-read reports that compared a variety of program performance measures across neighborhoods in their cities, they may have a better basis for understanding the effectiveness of their resource targeting and assessing the performance of their staffs within that system.
Implications for state and federal governments
We have suggested that a state-of-the-art neighborhood data system would be a powerful tool for health policymaking and program design and implementation in individual cities and metropolitan areas. Furthermore, the development of such systems is proving more and more feasible as technology has dramatically reduced the cost of acquiring and manipulating the data, and new local institutions and groups of professionals are learning how do these things efficiently and practically.
Accordingly, the most important implication for state and federal officials at this stage is that they should take steps to encourage this type of activity at the local level. This direction is fully consistent with the National Committee on Vital and Health Statistics' recommended "Strategy for Building the National Health Information Infrastructure" (U.S. Department of Health and Human Services 2001), which states
State and local health agencies should invest in the collection and analysis of population health data to permit real-time small area analysis of acute public health problems and to understand new or rapidly growing populations and health disparities, and they should combine health data sources for population analysis. (p. 42)
What could state and local agencies do to provide such support? First, in localities where it is not already being done, they could encourage local officials in the health field to move in directions outlined in the paragraphs above. This can occur in a variety of ways. They could, for example, develop and offer relevant technical assistance and training and support the activities of coalitions of localities toward these ends. Efforts in this regard may have to be preceded by a review of laws and regulations to identify changes that would either promote or eliminate barriers to activities like these at the local level.
Second, they can work out sound methods of sharing data from larger information systems with locals who have the capacity and want to use it for these purposes. This, of course, is primarily relevant for state officials. Medicaid claims files can be particularly useful, but other types of state maintained data (as discussed earlier in this section) should also be considered.