The Dangers of Hospital Closures Ch 4

Appendix 1 Multivariate analysis of predictors of high hospital costs per person across the states As described briefly in the text, we have performed a multivariate regression analysis to try to tease out the main independent predictors of high hospital costs across the 50 states. We compiled data on all 50 states on a number of hospital, physician, population, and other characteristics. We selected eleven specific variables for analysis. These were chosen mainly in light of their conceptual interest as predictors of variation in cost. The eleven variables chosen were: Hospital occupancy rate Acute hospital beds/1,000 residents In-hospital surgeries/1,000 residents ER visits/1,000 residents OPD visits/1,000 residents Average length-of-stay Operating margin Share of state population > age 65 COTH hospitals’ share of total admissions Labor share of total hospital expenses Practicing office-based physicians/10,000 people The main results of the regression are presented in Appendix Exhibit 1-A.50 Five of the eleven independent variables formed associations with hospital cost per person that would be found by chance no more often than five times in one hundred. That is, they had a statistical significance of p = 0.05 or lower. The independent variables are ranked by their Beta values. The Beta value indicates the association between a one percent change in the variable in question and the change in the dependent variable that’s being analyzed—hospital cost per person in a state. The value of each Beta is controlled for the values of the other independent variables in the regression equation. So, for example, a one percent change in COTH share of hospital admissions, across the 50 states, is associated with a 0.458635 percent rise in hospital cost per person, controlling for OPD visits/1,000 residents, and the other independent variables listed in Appendix Exhibit 1-A. Ranking the predictors’ importance by their Beta’s indicates that COTH share of admissions was most important, followed by OPD visits per 1,000 residents, beds per 1,000 residents, the ratio of practicing office-based physicians per 10,000 residents, and labor share of total expenses. The Beta is a useful way to rank the predictive power of the different independent variables. In other words, hospital beds per 1,000 residents was statistically significant but was only the third-most-useful predictor of differences in hospital cost per person across the states. The p-level of each independent variable is its statistical significance—the probability that the observed relationship between the independent variable in question and hospital cost per person would occur by chance (controlling for the values of the other independent variables). The relation between COTH share of admissions across the states and hospital cost per person across the states is very close—it would be found by chance only 4 times in 1 million—0.000004. Inspection of Appendix Exhibit 1-A reveals that, generally, the greater the Beta, the more statistically significant the variable (the lower the p-level). The regression equation as a whole was highly statistically significant. More important, it was practically powerful. The five independent variables explained fully 81.3 percent of the difference in hospital cost per person across the 50 states (R2 = 81.3 percent). Appendix Exhibit 1-A Results of Regression Analysis of Hospital Cost/Person across the 50 States Independent variable Intercept COTH share of admissions OPD visits/1000 residents Beds/1000 residents Practicing Office-based physicians/pop. Labor share of total expenses Beta B t (44) p-level 0.458635 0.416987 0.388850 0.322196 0.248460 -867.212 893.309 0.189 112.493 31.369 2313.306 -1.86821 5.30452 5.08607 5.27010 3.55126 2.96012 0.068400 0.000004 0.000007 0.000004 0.000928 0.004938 Note: The regression equation was significant at p = 0.000000. Its adjusted R2 = 81.3 percent. The data in Appendix Exhibit 1-B further assess the five independent variables’ usefulness in predicting differences among the states in hospital cost per person. The variable’s tolerance indicates its separate predictive power. The higher the tolerance, the less redundant is the variable with other independent variables—in a sense, the less it overlaps or correlates highly with other variables in the regression equation. All of these tolerances are above 0.50. The variable’s partial correlation measures its unique contribution to predicting the value of the dependent variable. The five variables’ partial correlations rank in the same order as did their Beta’s. A variable’s semi-partial correlation squared captures share of the total variance in that is uniquely accounted for by this variable. We see that the five variables, together, sum to almost 0.44—44 percent—of the total variance.51 COTH hospitals’ share of admissions and the rate of OPD visits/1,000 residents each accounted for about 12 percent of the total variance. Hospital beds/1,000 residents accounted for 11 percent. The ratio of practicing office-based physicians to population accounted for about 5 percent, and labor share of total hospital expenses accounted for about 4 percent. Appendix Exhibit 1-B The Independent Variables’ Tolerances, Partial Correlations, and Semi-partial Correlations Tolerance Partial correlation Semi-partial correlation Semi-partial correlation squared 0.568847 0.781109 0.632641 0.516609 0.603593 0.624545 0.622064 0.608475 0.471987 0.407519 0.345911 0.343667 0.331666 0.231580 0.193032 0.1197 0.1181 0.1100 0.0536 0.0373 Independent variable COTH share of admissions OPD visits/1000 residents Beds/1000 residents Practicing Office-based physicians/10K Labor share of total expenses Sum of semi-partial correlations squared 0.4387 This analysis strongly suggests that efforts to reduce hospital cost per person by trying to cut hospital beds/1,000 people would not be very effective because—after controlling for other characteristics—beds do not strongly influence costs. Further, the close correlation between COTH share of hospital admissions and cost per person indicates that if beds are closed at non-teaching hospitals, the teaching hospital share of hospital admissions will rise. This could be expected to result in an increase in hospital costs per person in a given state. This relation can be quantified with reasonable precision. As noted earlier, COTH share of hospital admissions had a Beta of 0.458635, controlling for the other independent variables in the regression equation. Thus, a one percent rise in the share of admissions to teaching hospitals will predict a rise of 0.458635 percent in hospital cost per person statewide. Hospital admissions totaled a bit over 2.5 million in New York State in 2004. And some 45.3 percent of admissions were to COTH hospitals (text Exhibit A). Statewide, we have calculated, admissions per hospital bed averaged almost exactly 43 admissions in 2004.52 Suppose, then, that 2,000 beds are closed at non-COTH hospitals and that the patients formerly admitted to these hospitals are shifted to COTH hospitals.53 We are thereby shifting some 86,000 admissions. Teaching hospitals admitted some 1,139,000 patients in New York State in 2004. This rises to 1,225,000, or 48.7 percent of admissions as a result of the closing of 2,000 beds in non-teaching hospitals. That’s a rise of 7.6 percent in teaching hospitals’ share of admissions (48.7/45.3 = 1.076). Consequently, hospital cost per resident of New York State rises by 7.6 * 0.458635 (the Beta), or almost 3.5 percent. A 3.5 percent rise in hospital spending per person would have been $76.70 in 2004. This would have translated into an actual increase of $1.5 billion in 2004’s statewide hospital spending—raising spending from $42.0 billion to $43.5 billion. Appendix 2 Regression analysis of hospital closings in 52 U.S. Cities, 1980 – 2003 We studied hospital closings from 1980 to 2003 as functions of hospitals’ 1980 or 1990 characteristics, focusing on the six characteristics that significantly predicted closings during the full 23 years from 1980 to 2003, and also hospital efficiency. Only two of these—number of beds and fund balance per patient— were significant during all periods (Appendix Exhibit 2-A). Area percent black significantly predicted chance of closings during all periods but one, 1990 to 1997. Greater efficiency never predicted survival. Appendix Exhibit 2-A Significance of predictor characteristics during five periods Significance level of each variable during each period b Initial year characteristic a 19802003 Medical school affiliation Beds Area percent black Hospitals within one mile Fund balance per patient Case mix index, Medicare patients Efficiency 19801990 19901997 0.031460 0.000000 0.000841 0.018967 0.000210 0.004696 0.150235 0.014292 0.000000 0.000096 0.002387 0.021249 0.804076 0.095338 0.161555 0.000062 0.412443 0.438814 0.000866 0.140895 0.852061 0.288121 0.001826 0.016508 0.335435 0.003237 0.624833 0.635725 0.090647 0.000002 0.008305 0.197952 0.000013 0.218536 0.796208 Adjusted R2 from regression 26.1% 16.7% 10.5% 6.9% 16.0% number of hospitals in 52 cities Hospitals closing during period Hospitals remaining open at end of period Percent of hospitals closing during period 692 252 440 36.4% 701 123 578 17.5% 623 72 551 11.6% 550 68 482 12.4% 621 139 482 22.4% Note for Exhibit 65: Hospitals with 25 or more beds in the initial year are included. a b 1980 for 1980-2003 and 1980-1990 analyses, and 1990 for the others. Characteristics significant at 0.05 or better are italicized. 19972003 19902003 Notes 1 Judy Wessler, Analysis of the Approved Federal-State Health Reform Partnership (FSHRP) Waiver, New York: Commission on the Public’s Health Care System, October 2006. 2 Richard Perez-Pena, “In Move to Cut Hospitals, U.S. Will Pay New York $1.5 Billion,” New York Times, 3 October 2006. 3 This sum is well below the adjusted R2 of 81.3 percent of the entire regression equation. The difference would seem to be accounted for by interactions between and among the variables. 4 This analysis assumes that the average occupancy rate and annual admissions/bed at closed hospitals and surviving teaching hospitals are equivalent. 5 Owing to the attrition of non-teaching hospitals over many decades in New York City, for example, surviving hospitals are increasingly likely to be teaching hospitals and COTH members. 6 Milton Roemer and Max Shain, Hospital Utilization under Insurance, Chicago: American Hospital Association, 1959. See also Milton Roemer, “Bed Supply and Hospital Utilization: A Natural Experiment,” Journal of the American Hospital Association, Vol. 35 (1 November 1961), pp. 36-42. 7 Paul B. Ginsburg and Daniel M. Koretz, “Bed Availability and Hospital Utilization: Estimates of the ‘Roemer Effect,’ “ Health Care Financing Review, Vol. 5, No. 1 (Fall 1983), pp. 87-92. 8 U.S. General Accounting Office, Available Research Shows that Capacity Is Only Weakly Related to Volume, Washington: GAO, 31 January 1991, GAO/PEMD-01-7. 9 Calculated from American Hospital Association, Hospital Statistics, 2002 Edition, Chicago: AHA, 2002, Table 1. 10 Walter McClure, Reducing Excess Hospital Capacity, Excelsior, Minnesota: InterStudy, 15 October 1976, prepared for the Bureau of Health Planning and Resource Development, Department of Health and Human Services, under Contract HRA-230-760086. The material cited is from the executive summary. 11 Walter McClure and Lenore Kligman, Conversion and Other Policy Options to Reduce Excess Hospital Capacity, Washington: Bureau of Health Planning, Health Resources Administration, September 1979, under Contract No. HRA 230-77-0033, DHEW Publication No. (HRA) 79-14044. The material cited is from Section V, especially pp. 99-100. 12 Witness, for example, the Massachusetts Hospital Association’s dozen years of opposition to proposed state legislation that would identify and stabilize all needed hospitals. The law’s stabilization fund would be financed by a one-quarter of one percent assessment on the revenues of all hospitals. 13 Richard Perez-Pena, “In Move to Cut Hospitals, U.S. Will Pay New York $1.5 Billion,” New York Times, 3 October 2006. 14 Barbara Caress and Ronda Kotelchuck, “Politics Makes Strange Beds: New York as a Laboratory for Cost Controls,” Health PAC Bulletin, No. 77 (July/August 1977), pp. 110, 19-21. 15 Some of these forces are discussed in Richard Perez-Pena, “Hospital Business in New York, Once Prized, Braces for a Crisis,” New York Times, 11 April 2005. 16 For a look at some of the problems at one hospital, see Katherine E. Finkelstein, “Layoffs Feared as Consultant Examines Mount Sinai’s Budget,” New York Times, 29 November 2001; and Janny Scott with Mary Williams Walsh, “Mt. Sinai on a Path away from the Past,” New York Times, 25 August 2002. 17 Richard Perez-Pena, “Hospital Business in New York, Once Prized, Braces for a Crisis,” New York Times, 11 April 2005. 18 Stephen Berger, Third report of the Working Group on Medicaid and health care, 17 November 2004, especially pp. 16-19. 19 Stephen Berger, Third report of the Working Group on Medicaid and health care, 17 November 2004, p. 17. 20 Consumer Price Index data are the annual figure for all items, U.S. city average, Series ID CUUR0000SA0, access confirmed 21 October 2006. 21 Alan Sager, “D.C. General Must Be Resuscitated,” op-ed., Washington Post, 3 December 2000; Alan Sager, Saving D.C. General Hospital: Why and How? Boston: Health Reform Program, Boston University School of Public Health, 5 October 2000; Alan Sager, D.C. General Hospital Should Be Renewed, Not Closed, Testimony before the Washington, D.C. City Council, Boston: Boston University School of Public Health, 18 September 2000. 22 See, for example, Alan Sager, The Proposed Closing of Rancho Los Amigos National Rehabilitation Center Endangers the Health of Disabled Medi-Cal Patients Who Reside in Los Angeles County, Statement of expert witness prepared in the case of Rodde, et al. v. Bonta et al, 3 April 2003. 23 Report to the Princeton Health Care Task Force on the University Medical Center at Princeton’s Capital Needs and Location Options, Boston: Boston University School of Public Health, 1 February 2005, 31 pages 24 This section relies heavily on Barbara Caress and Ronda Kotelchuck, “Politics Makes Strange Beds: New York as a Laboratory for Cost Controls,” Health PAC Bulletin, No. 77 (July/August 1977), pp. 1-10, 19-21. 25 Barbara Caress and Ronda Kotelchuck, “Politics Makes Strange Beds: New York as a Laboratory for Cost Controls,” Health PAC Bulletin, No. 77 (July/August 1977), pp. 110, 19-21. 26 Health and Hospitals Planning Council of Southern New York, “Unneeded Facilities for General Hospital Care in New York City,” New York: The Council, 5 February 1976, p. 2, cited in Barbara Caress and Ronda Kotelchuck, “Politics Makes Strange Beds: New York as a Laboratory for Cost Controls,” Health PAC Bulletin, No. 77 (July/August 1977), pp. 1-10, 19-21. 27 Barbara Caress and Ronda Kotelchuck, “Politics Makes Strange Beds: New York as a Laboratory for Cost Controls,” Health PAC Bulletin, No. 77 (July/August 1977), pp. 110, 19-21. 28 Cited in Barbara Caress and Ronda Kotelchuck, “Politics Makes Strange Beds: New York as a Laboratory for Cost Controls,” Health PAC Bulletin, No. 77 (July/August 1977), pp. 1-10, 19-21. 29 New York Times, 18 February 1977, cited in Barbara Caress and Ronda Kotelchuck, “Politics Makes Strange Beds: New York as a Laboratory for Cost Controls,” Health PAC Bulletin, No. 77 (July/August 1977), pp. 1-10, 19-21. 30 Walter McClure and Lenore Kligman, Conversion and Other Policy Options to Reduce Excess Hospital Capacity, Washington: Bureau of Health Planning, Health Resources Administration, September 1979, under Contract No. HRA 230-77-0033, DHEW Publication No. (HRA) 79-14044, pp. 108-109. 31 Howard French, “Crime and Fear Follow Crack in Hospitals,” New York Times, 10 May 1989. 32 Wolfgang Saxon, “Hospital Bed Shortage Tied to Type of AIDS Cases,” New York Times, 25 December 1987. 33 Jesse Green, Madeleine Singer, Neil Wintfeld, Kevin Schulman, and Leigh Passman, “Projecting the Impact of AIDS on Hospitals, Health Affairs, Fall 1987, pp. 19-31, Exhibit 3. 34 Niccie L. McKay and Fred H. Dorner, “The Effect of Rural Hospital Closures on the Financial Performance of Neighboring Rural Hospitals,” Inquiry, Vol. 33 (Fall 1996), pp. 271-282. 35 Donald S. Shepard, “”Estimating the Effect of Hospital Closure on Areawide Inpatient Hospital Costs: A Preliminary Model and Application,” Health Services Research, Vol. 18, No. 4 (Winter 1983), pp. 513-549, at p. 513. 36 See, for example, Robert Mechanic, Kevin Coleman, and Allen Dobson, “Teaching Hospital Costs: Implications for Academic Missions in a Competitive Market,” Journal of the American Medical Association, Vol. 280, No. 11 (16 September 1998), pp. 10151019. 37 As growing proportions of lower-income urban residents, who had formerly been admitted to now-closed non-teaching hospitals, are served in teaching hospitals, the cost of their care will rise if they receive more care or more esoteric care than formerly. Further, lower-income patients and their payors may be asked to assume greater shares of the cost of training and research conducted in teaching hospitals. Even if total cost of care in the community remains unchanged, health care for lower-income people will look more expensive to Medicaid, to other insurers, and to private employers worried about their bills. 38 Arnold M. Epstein, Robert S. Stern, and Joel S. Weissman, “Do the Poor Cost More? A Multihospital Study of Patients’ Socioeconomic Status and the Use of Hospital Resources,” New England Journal of Medicine, Vol. 322, No. 16 (19 April 1990), pp. 1122-1128. 39 Niccie L. McKay and Fred H. Dorner, “The Effect of Rural Hospital Closures on the Financial performance of Neighboring Rural Hospitals,” Inquiry, Vol. 33 (Fall 1996), pp. 271-282. 40 Alan M. Garber, Victor R. Fuchs, and James F. Silverman, “Case Mix, Costs, and Outcomes: Differences between Faculty and Community Services in a University Hospital,” New England Journal of Medicine, Vol. 310, No. 19 (10 May 1984), pp. 1231-1237. 41 Alexander P. Frick, Anne Grisez Martin, and Michael Shwartz, “Case-mix and Cost Differences Between Teaching and Nonteaching Hospitals,” Medical Care, Vol. 23, No. 4 (April 1985), pp. 283-295. 42 Donald S. Shepard, “Estimating the Effect of Hospital Closure on Areawide Inpatient Hospital Costs: A Preliminary Model and Application,” Health Services Research, Vol. 18, No. 4 (Winter 1983), pp. 513- 43 Margaret O’Malley, “We Deserve More,” Gloucester Daily Times, 29 April 1998. 44 Andrew Bindman, Dennis Keane, and Nicole Lurie, “A Public Hospital Closes: Impact on Patients’ Access to Care and Health Status,” Journal of the American Medical Association, Vol. 264, No. 22 (12 December 1990), pp. 2899-2904. 45 National Center for Health Statistics, Health United States, 2005, Table 88, http://www.cdc.gov/nchs/data/hus/hus05.pdf. 46 This conservatively considers only those who designated themselves as black race alone on the U.S. Census. 47 David Elesh and Paul T. Schollaert, “Race and Urban Medicine: Factors Affecting the Distribution of Physicians in Chicago,” Journal of Health and Social Behavior, Vol. 13 (September 1972), pp.236-250; Sheila Joroff and Vicente Navarro, “Medical Manpower: A Multivariate Analysis of the Distribution of Physicians in the United States,” Medical Care, Vol. 9, No. 5 (September-October 1971), pp. 428-437; and Donald Dewey, Where the Doctors Have Gone: The Changing Distribution of Private Practice Physicians in the Chicago Metropolitan Area, 1950-1970, Chicago: Illinois Regional Medical Program, 1973. 48 Even turmoil or fears of a bankruptcy or closing can induce physician departure. See Scott Hensley, “Flight from Philly: Allegheny Collapse Further Fuels Doctors’ Exodus,” Modern Healthcare, 31 August 1998, p. 12. 49 Only the rapid rise in the number of hospital-based physicians in poverty areas was sufficient to offset the substantial drip in office-based physicians, particularly in officebased primary care physicians. There was a substantial drop in the physician-topopulation ratios in areas changing from non-poverty to poverty status over time. David A. Kindig, Hormoz Movassaghi, Nancy Cross Dunham, Daniel I. Zwick, and Charles M. Taylor, “Trends in Physician Availability in 10 Urban Areas from 1963 to 1980,” Inquiry, Vol. 24 (Summer 1987), pp. 136-146. 50 The regression analysis, like the Pearson product-moment correlation analyses that generated the R2 values cited in the text, was performed in Statistica 6.1. 51 This sum is well below the adjusted R2 of 81.3 percent of the entire regression equation. The difference would seem to be accounted for by interactions between and among the variables. 52 This analysis assumes that the average occupancy rate and annual admissions/bed at closed hospitals and surviving teaching hospitals are equivalent. 53 Owing to the attrition of non-teaching hospitals over many decades in New York City, for example, surviving hospitals are increasingly likely to be teaching hospitals and COTH members.