Lee Rivers Mobley
Department of Economics
School of Business Administration
Oakland University

Spatial Modeling in Health Economics

What are my research interests in the areas of inequality and equity, and what spatial dimensions do I currently or potentially see in them?

My healthcare research interests in the areas of inequality and equity include: 1) examination of impacts from policy reforms on access to healthcare, via their impacts on hospital and medical services availability; 2) integrating spatial modeling into examination of medical treatment utilization and effectiveness, in order to develop heterogeneous, population-based protocols for treatments and services; 3) analysis of the cost-effectiveness of medical treatments, with explicit consideration of costs to the patient in obtaining care (time, transportation, travel, work release, day care for dependents, etc.).

Current: In my published work, I have examined how policy changes which affect hospital markets have impacted access to care. One paper focused on public hospital closures in California following MediCal reforms, looking at how this impacted distances traveled by the poor for hospital care. This paper used unconditional analysis comparing mean distances traveled before and after reforms (Mobley, 1998). Another series of papers models space explicitly, using a competing-destinations model, to determine whether the penetration of managed care insurance plans has affected distance traveled for hospital inpatients. There are two effects from managed care penetration: the direct effect on constituents of managed care plans, and an indirect effect on market extent via impact on hospital services offerings and adoption of technology. We find that both direct and indirect effects are significant and economically meaningful, but are surprised to find that managed care apparently shrinks markets. We conclude that managed care plans apparently place high priority on convenience, and contrary to mainstream theoretical expectations, have apparently revived the medical arms race rather than stimulated specialization into (fewer, more widely dispersed) centers of excellence (Mobley and Frech, 2000). A third body of research in progress tests this latter supposition directly. We examine how managed care has impacted the so-called 'medical arms race', i.e. the broad availability of services and technologies. Preliminary results suggest that managed care has increased the overall breadth of services offered by competing hospitals, while encouraging greater homogeneity (less specialization) in services offered among them. This is consistent with the notion that managed care favors 'one-stop-shopping' convenience for constituents over specialization, which is surprising, because better outcomes are known to occur in more specialized, higher-volume settings (see Chernew, Scanlon, and Hayward, 1996, and Grossman and Banks, 1998 for evidence from open heart surgery, and Tilford et. al., 2000, for pediatric intensive care services).

Potential: There is tremendous potential from integrating spatial modeling into examination of medical treatment utilization and effectiveness, and analysis of the cost-effectiveness of medical treatments. Many medical research studies include demographic characteristics of patients, either directly from patient records or interpolated from the U.S. Census based on patient zipcode, but few actually include geographic dimensions in their models. For example, in examination of medical treatment utilization (screening mammography use), researchers have included household income based on zipcode from the U.S. Census as a partial determinant of inadequate follow-up, in order to better understand the shortcomings of existing delivery systems (McCarthy et. al., 1996, 1997). Another study looks at strategies for reducing potentially avoidable hospital admissions among home-care clients, including demographic characteristics (race, income) as partial determinants of the odds of avoidable admissions (Weissert et. al., 1997). Burgess and DeFiore (1994) examine the impact of distance in patient choice among various VA outpatient facilities, but the patient geo-demographic characteristics included are scant.

Integrating the spatial dimension and spatially-referenced information into medical research can be extremely important, although to my knowledge, this is rarely done. An exception which illustrates the importance of this integration is a recent cost-effectiveness analysis of different management strategies for an ongoing, chronic-disease therapy (the three strategies include in-hospital, in-clinic, or in-home therapy). This study includes patient-related costs, such as time and travel costs, in estimating the relative benefits of the three treatment strategies. Not surprisingly, in-home treatment was most effective from the patient's perspective (when patient time and travel costs were included) while the clinical delivery site was most effective from an institutional perspective (ignoring patient time and travel costs) (Lafata et. al., 2000). This study illustrates the importance of including geographic factors when doing cost-effectiveness analysis of competing delivery mechanisms. Failure to model geography in a system where space clearly matters in explaining observed outcomes is shortsighted, and can yield misleading policy conclusions.

Currently, there is funding available from the National Institutes of Health for medical studies in cancer surveillance, cancer prevention, cancer screening, and cancer care (http://www.ahrq.gov/fund/99014.htm; http://www.ahrq.gov/fund/99015.htm).

In my opinion, these areas of medical research could be enhanced by integration of spatial modeling into the analyses. For example, neighborhood effects - such as lack of good private or public transportation, exposure to environmental contaminants, prevalence of low-income jobs, low educational attainment, many single-headed households with children, low rates of health insurance and/or prevalence of insurance with narrow coverage (high co-payments and deductibles), distance to available hospitals or clinics - can contribute to poor utilization and outcomes even when the best possible services are available. Because of patient confidentiality, many of these dimensions are not available from patient records, and are sometimes obtained via expensive surveys. With the Census 2000 available soon, some of these factors can be modeled directly using data from the Census matched to patient's zipcode or neighborhood. Similarly, GIS software can be used to measure distance (travel time) between patient address and hospital/clinic for direct inclusion in the model. An alternative modeling strategy (rather than direct inclusion of many geo-referenced demographic variables) is to estimate a spatial regression model on patient-level data, which exploits geographic location (of patients) via a distance matrix, to filter out omitted neighborhood effects which might otherwise bias policy-variable parameter estimates (Anselin and Bera, 1998). To my knowledge, this alternative strategy has not been exploited in medical research. If we ignore these spatial dimensions, we are ignoring information that helps determine patient outcomes, and information about potential data complexities. This can affect efficiency, sufficiency, bias, and consistency - the four properties of statistical estimators.

What kind of spatial data, models, techniques, software, etc. do you use or have considered using in your research. Which of these work well for you? Where do you see problems and/or shortcomings?

Data I have used: census data, hospital and patient data by zipcode, EPA data on air quality by city, Area Resource File data by county. Models and techniques: I have modeled space directly using spatial interaction models (competing destinations models), and I have modeled space indirectly using spatial regression with lattice data, and some ad-hoc combinations of these two methods. I have used ArcView with SpaceStat software, and Atlas GIS software. These work well enough in combination: ArcView has limited database capabilities that can be bypassed using Atlas GIS and Excel. SpaceStat works well with ArcView, but would be improved if the regression residuals could be mapped directly in ArcView.

Can you point out any "best practice examples" of spatially-oriented research in your field?

My paper with Frech regarding impact of managed care on distance traveled by hospital inpatients is the best example I know about which directly applies spatial modeling to healthcare markets (Mobley and Frech, 2000). No other research by health economists that I know of explicitly uses spatial interaction models, or spatial regression techniques. In general, economists are slow to embrace rigorous spatial modeling. Health policy analysts have used GIS to do simple things, like map univariate health characteristics and distributions of health professionals and disease clusters (Murray et. al., 1998; Rickets et. al., 1994). Others have computed simple access measures and compared them by disparate groups across regions such as cities, MSAs, counties, and states (i.e., see regularly disseminated policy reports from the UCLA Center for Health Policy Research, and the Public Policy Institute of California; Rickets et. al., 1994). Some health economists are using distance between hospitals and patients as explanatory variables. Some are modeling hospital (destination) choice, with simple hospital-patient distance as a partial determinant. However, most use conditional logit-type models (Burns and Wholey, 1992; others cited in Mobley and Frech, 2000) rather than spatial interaction models, despite the demonstrated shortcomings in these logit models due to their IIA properties (Fotheringham and O'Kelley, 1989, pp 78-80). To my knowledge, competing-destinations-type models, which do not suffer from IIA, are rarely used to predict hospital choice. Some exceptions are work by Burgess and DeFiore (1994) and Kessler and McClellan (2000).

Do you have any suggestions for Learning Resources that CSISS might provide? Workshops we might offer?

Please continue subsidizing tuition for the spatial modeling and spatial regression workshops with Luc Anselin. I have attended two, and both were excellent. I am not as familiar with your other workshops, but tuition subsidies are helpful to academics (like myself) whose fiscal gatekeepers may be skeptical about these "new" methods.

References

Anselin, A. and Bera, L., "Spatial Dependence in Linear Regression Models With an Introduction to Spatial Econometrics", in Giles, D. and A. Ullah (eds.), Handbook of Applied Economic Statistics (New York: Marcel Dekker, 1998).

Burgess, J. and DeFiore, D., "The Effect of Distance to VA Facilities on the Choice and Level of Utilization of VA Outpatient Services", Social Science and Medicine, v 39 (1)(1994), pp 95-104.

Burns, L., and D. Wholey. 1992. The Impact of Physician Characteristics in Conditional Choice Models for Hospital Care. Journal of Health Economics 11(1): 43-62.

Chernew, M., R. Hayward, and C. Scanlon. 1996. Managed Care and Open-Heart Surgery Facilities in California. Health Affairs 15 (1):191-201.

Fotheringham, A. S. and O'Kelly, M. E., Spatial Interaction Models: Formulations and Applications, Kluwer: Boston (1989).

Grossman, J., and D. Banks. 1998. Unrestricted Entry and Nonprice Competition: The Case of Technological Adoption in Hospitals. International Journal of the Economics of Business 5 (2):223-246.

Kessler, D. and McClellan, M., "Is Hospital Competition Socially Wasteful?", The Quarterly Journal of Economics, May 2000, pp 577-615.

Lafata, J. and Martin, S. and Kaatz, S. and Ward, R., "The Cost-Effectiveness of Different Management Strategies for Patients on Chronic Warfarin Therapy", Journal of General Internal Medicine, v 15 (January 2000), pp 31-37.

McCarthy, B., Yood, M., MacWilliam, C., and Lee, M., "Screening Mammography Use: The Importance of a Population Perspective", American Journal of Preventive Medicine, v 12 (2)(1996), pp 91-95.

McCarthy, B., Yood, M., Janz, N., Boohaker, E., Ward, R., and Johnson, C., "Evaluation of Factors Potentially Associated with Inadequate Follow-Up of Mammographic Abnormalities", Cancer, v 77 (10((May 15, 1996), pp 2070-2076.

McCarthy, B., Yood, M., Bolton, M., Boohaker, E., MacWilliam, C., and Young, M., "Redesigning Primary Care Processes to Improve the Offering of Mammography", Journal of General Internal Medicine, v 12 (June 1997) pp 357-363.

Mobley, Lee, "Effects of Selective Contracting on Hospital Efficiency, Costs, and Accessibility", Health Economics, v 7 (1998), pp 247-261.

Mobley, Lee and Frech, H.E. III, "Managed Care, Distance Traveled, and Hospital Market Definition", Inquiry, v 37 (1)(Spring 2000), pp 91-107.

Mobley, Lee and Frech, H.E. III, "Managed Care and Hospital Markets: HMOs vs. PPOs vs. Kaiser", Antitrust Report (July 2000), pp. 11-28.

Murray, C. and Michaud, C. and McKenna, M. and Marks, J., U.S. Patterns of Mortality by County and Race: 1965-1994, Harvard School of Public Health, Cambridge (1998).

Ricketts, T. and Savitz, L. and Gesler, W. and Osborne, D., Geographic Methods for Health Services Research, University Press of America, New York (1994).

Tilford, Simpson, Green, and others, "Volume-outcome Relationships in Pediatric Intensive Care Units", Pediatrics, v 106 (August 2000), pp 289-294.

Weissert, W. and Lafata, J. and Williams, B. and Weissert, C., "Toward a Strategy for Reducing Potentially Avoidable Admissions Among Home Care Clients", Medical Care Research and Review, v 54 (4)(December 1997), pp 439-455.