JOSEPH A. BOSCARINO INTRODUCTION. Pergamon

PD: S1353^«50S(96)00056-7 i353-»505/96 sis.oo+o.oo

Patients' Perception of Quality Hospital Care and Hospital

Occupancy: are there Biases Associated with Assessing

Quality Care Based on Patients' Perceptions?

JOSEPH A. BOSCARINO

Vice President for Outcomes Research, Department of Outcomes Research, Sisters of Charity of Nazareth Health System, 2020 Newburg Road, Louisville, KY 40205, USA

There currently is interest in evaluating medi-cal outcomes based on patient perceptions. However, in the U S there may be biases associated with these perceptions because of past marketing activities and other factors, such as facility location. The research question examined is whether perceived overall quality could predict hospital occupancy. To assess this, the quality ratings of 155 local hospitals by over 20000 household beads surveyed in 20 US states were analyzed using an ecological research design. Facility image and hospital occupancy were assessed after controlling for community, facility and quality care differences between facilities. Results indicated that hospitals in more urba-nized areas (p = 0.003), with lower costs O» = 0.0001), that were non-teaching (p = 0.033) and those with more employees per bed (j>< 0.0001) had higher occupancies, but that perceived quality did not predict admissions after facility differences were controlled (p = 0.302). However, further analysis suggested both positive and negative biases may exist: controlling for community, facility, and quality care differences, facilities with "high" ratings appeared to have consistently higher occupancies, those with "low" ratings consistently lower occupancies, and facilities with "average" ratings appeared to be unaffected. Based on this finding, an interaction effect was tested and confirmed for

community rating x facility size (f= 0.015), suggesting that smaller hospitals with low ratings had lower than expected occupancies. Although this study has limitations, it was suggested that researchers should use quality indicators based on patients' perceptions with caution and be open to additional scientific research, until these mea-sures are better understood. Copyright © 1996 Elsevier Science Ltd.

Key words: Quality care, outcomes research, perceived quality, patient-centered care, hospital quality, health services research.

INTRODUCTION

Evaluating patient outcomes represents a major challenge for health services researchers [1-3]. While progress has been made, a major obstacle has been the lack of a framework for assessing medical outcomes upon which there is agreement [4-6]. Difficulties include research problems related to process vs outcome, imper-fect data, administrative vs clinical data, adjust-ments for risk, episodes of care and inclusion of "quality of life" factors [5]. Nevertheless, pres-sure for medical accountability is growing [7-9]. Recently, the Health Care Financing Adminis-tration (HCFA) announced the use of the mass media to inform Medicare and Medicaid bene-ficiaries about quality issues [10]. Previously, HCFA released hospital mortality data with the expectation that beneficiaries and others would use these data to make decisions about medical care [11]. The US government's policy in this area is unequivocal: many medical procedures are being evaluated with the intention of widely disseminating these results [12].

Received 22 January 1996; accepted 12 June 1996.

467

by guest on May 8, 2016

The aspect of quality care assessed in this article relates to patients' perception of quality. The research question is: do hospitals with higher perceived quality have higher patient occupancies, in part, because of biases asso-ciated with a facility's image? Many experts believe that perception of quality should be a component of evaluating medical outcomes [13-15]. In addition, there is a recognition in the US that perceived quality, satisfaction, and consu-mer choice should be key in structuring health care, because these should encourage organiza-tional innovation and efficiency [16-19]. In the past, some health services researchers and eva-luators have routinely used global patient rat-ings to assess outcomes [20,21]. If these perceptions proved biased, evaluations based on these could be misleading. In addition, it could mean that more specific quality indicators based on perceptions also are biased and may require closer surveillance.

There are at least five reasons why perceived quality and hospital occupancy should be related in the US health' care market—some of these are related to quality care and some are not. First, over the past decade hospitals have extensively marketed their services and pro-moted their images [16] and it is likely that some of these efforts have succeeded [7]. Second, some studies have suggested that hospitals with better quality staff and more advanced technol-ogies should have higher quality care, better community reputations and should attract more productive and qualified physicians [19,22,23]. Third, to attract patients many hospitals in the US have recruited physicians and have pro-moted their services [7,17-19,22]. Fourth, hospi-tals located in lower-income neighborhoods or in rural areas may have poorer quality images because of these locations [24]. As a conse-quence, these hospitals may be perceived to have lower quality care and may have fewer patient admissions. Fifth, hospitals have focused on improving patient satisfaction and patient accommodations to attract patients [25,26]. It is possible that these efforts have succeeded, even though the quality of medical care has not improved.

In summary, there are a host of reasons why hospital image may be associated with patient admissions in the US, not all of which are quality related. In addition, studies suggest that

patients have become more involved in hospital choice [27]. Furthermore, while many patients do not choose a hospital directly, findings suggest that they often do this indirectly through choice of a personal physician [27,28]. As suggested, in the US patient choice and facility image have become salient issues over the past decade because many hospitals have aggressively promoted their facility's image, marketed their services, and promoted their medical staff [7]. Within this context, it is possible that health care marketing activities and other trends in the US over the past decade may have biased patients' perceptions of qual-ity. Furthermore, although this research was conducted in the US, it also has public health significance for other countries considering adapting, in whole or in part, the US "market model" of health care delivery.

The public's perception of hospitals and its motivations for hospital choice in the US have been studied [27-31]. Research documents that most adults believe they can discern higher quality hospitals from lower quality ones [30,31]. Furthermore, this research indicates that "patient relations," the medical staff, the nursing staff, convenience, and technology are critical in the public's perception of quality. Typically, however, "patient relations" are very high on the public's quality list [30,31]. The latter finding is troubling, since many of the hospital-based outcomes recommended by the Joint Commission on the Accreditation of Healthcare Organizations (JCAHO) are heavily premised on clinical outcomes related to morbidity and mortality [6,32].

It has been reported previously that house-hold heads rate non-rural, larger, tertiary care, teaching, higher patient occupancy, better staffed and lower mortality facilities higher in quality [24]. Hospitals that had higher employee salaries and those that were more costly also receive higher quality scores [24]. Multivariate analyses indicated that these factors combined could account for over 50% of a facility's overall quality rating. In addition, high- and low-quality hospitals could be correctly classify 85% of the time using these variables [24]. The most impor-tant variables discriminating perceived quality here were level of care, average patient occu-pancy, employee salaries and teaching status [24].

by guest on May 8, 2016

METHODS

Data base

The data for this study are from 40 separate US community surveys conducted between 1984 and 1988. These surveys were administered in hospital services areas in 20 US states (these included: CO, IA, IL, IN, MA, MD, MI, MO, MT, NC, NH, NJ, NV, NY, OH, OK, PA, RI, SC and TX). Combined, the surveys included telephone interviews with more than 20000 adult household heads. The average community response rate for the surveys was estimated to be about 65%. After the surveys were combined, community ratings were available for 155 short-term, non-government, medical/surgical hospi-tals. Government, non-acute and non-medical/ surgical facilities were excluded. While not a random sample, the 155 hospitals represented a cross-section of institutions, including urban, suburban, and rural facilities (see below). More-over, the surveys were conducted with adult "household heads", not just recent hospital patients or any adult in the household. Recent patient households were included in the surveys, insofar as they were represented in the general household populations surveyed. This figure is not insignificant in most US communities, however. For example, national surveys in the US have reported that 35% to 40% of US households have had someone hospitalized in the past two years [33]. In addition, government studies estimate over 143 acute-care, non-Fed-eral hospital discharges occurred per 1000 population in 1986 [34]. Furthermore, these hospital experiences are not only concentrated in older households, since the most common reason for a hospital admission in the US is for a routine obstetrical delivery [34]. Among the household heads surveyed, about 65% reported an inpatient or outpatient hospital experience in the household in the past two years.

The hospital sample in this study had the following profile. US geographic region: north = 34%, south =19%, mid-west = 44%, west = 3%. Location: rural = 21%, suburban = 54%, urban = 25%. Bed size: 200 or less = 29%, 201-400 = 42%, 401 or more = 29%. Teaching status: non-teaching = 78%, teaching = 22%. Religious affiliation: secular = 72%, non-secu-lar =28%. In comparison to national hospital statistics [35], this sample under-represents

hos-pitals from the West and smaller hoshos-pitals. For the purposes of this study, however, we assume the data are representative of US hospitals, with the exception of smaller facilities. Additional information on the study design and data base are available elsewhere [24,36].

Study variables and analytical approach

The hypothesis evaluated is that hospitals with better community perceptions have higher than expected occupancies, possibly due to the "marketing effects" associated with hospital image. The "placebo" effect is well documented in clinical research [37] and this effect has also been noted in psychotherapy outcomes [38], as well as epidemiologic research [39]. The "image effect" in marketing [40], advertising [41] and public relations [42] has been documented. This outcome should also be observed in hospital utilization. To test this hypothesis, the associa-tion between hospital image and hospital occu-pancy was examined after key covariates were controlled. These covariates included commu-nity, institutional and quality variables. It was expected that a positive association would remain between image and occupancy, after controlling for these factors, because the hospi-tal's image tends to outweigh other factors. Study variables are defined below. Unless noted, all data were derived from the same year householders were surveyed in the community.

"Community factors" are associated with the hospital's geographic area. These include socio-economic status (SES INDEX), urbanization level (URBAN), percent African-American (BLACK), and percent Hispanic (HISPANIC). SES INDEX, BLACK and HISPANIC were from The Sourcebook of Zip Code

Demo-graphics [43] and represent the demographic

composition of a hospital's service area based on US census data. Geographic region (REGION) and the average hospital bed size in the state (STATE BED) were used as control variables to adjust for sample bias. REGION was based on US census categories. STATE BED was based on American Hospital Associa-tion data [35] and represents the average hospital bed size in a state.

"Institutional Factors" are related to the type and size of a facility and include the number of

by guest on May 8, 2016

beds (BEDS), religious affiliation status (AFFILIATION), cost per admission (COST/ ADM), average employee salary (SALARY) and hospital case-mix index (CASE-MIX). BEDS, AFFILIATION, COST/ADM and SALARY were from American Hospital Asso-ciation data [35]. AFFILIATION is an indicator variable and was based on whether the facility was associated with a religious order or not. COST/ADM represents the average cost per admission. SALARY represents the average annual salary per facility employee. CASE-MIX was based on data reported by HCFA in

1986 [44].

"Quality Care Factors" include reported mortality rate (MORTALITY), level of tertiary care provided (CARE LEVEL), teaching status (TEACHING) and the number of employees per bed (EMPL/BED). CARE LEVEL was based on the number of tertiary services offered by a facility and was described in detail elsewhere [36]. Higher scores on this variable represented the provision of more tertiary care services. MORTALITY was based on the 1986 HCFA report on overall patient mortality [45]. EMPL/ BED represents the number of employees per bed, based on American Hospital Association data [35]. CARE LEVEL and TEACHING were also derived from American Hospital Associa-tion data [35].

"Hospital Quality Rating" (RATING) was derived from each of the 40 surveys conducted. All of these surveys had the same hospital quality rating question presented to the survey respondents. This question was based on a standard five-point Likert scale format [46], where respondents were asked to rate each local hospital as follows: "Based on experience or what you may have heard, how would you rate the following hospital overall? Would you say that" " Hospital was much worse, some-what worse, the same, somesome-what better, or much better than other area hospitals overall?" Each respondent was asked to rate an average of about four local hospitals. Studies have shown that while the public's image of local hospitals is multifaceted [27,29], often there are common service and facility features used to evaluate hospitals [28,33,47,48]. For example, as noted above, studies indicate that the public's percep-tion of individual courtesy, the medical and nursing staff, convenience and higher technology

are important in defining its perception of quality hospital care [30,31].

The hospital quality rating used in this paper was an overall score that represented the average rating of a hospital by the adult household heads in the local community. Not all adults rated local hospitals, especially if they were unfamiliar with particular faculties. In these cases, responses were classified as "don't know" responses and were excluded from the data base. To minimize bias, no hospital ratings were based on fewer than 100 respondents. Using this method (with five being the highest score) the "average" hospital studied received a community rating of 3.54 (95% C.I. = 3.47-3.61). The median rating for hospitals was 3.56. Low-quality ratings (defined as the 25th percentile) were 3.26 or lower. High-quality ratings (defined as the 75th percentile) were 3.87 or higher. The distribution of these ratings approximated a normal curve (standard deviation = 0.421, standard error = 0.034, kurtosis = 0.112, skewness= -0.234). All surveys included in this study were conducted between 1984 and 1988.

"Hospital Patient Occupancy" (OCCU-PANCY) for each hospital was based on average yearly patient "census" figures reported by the American Hospital Association for the year the community survey was conducted [35]. This figure represents the average percent of hospital beds occupied over the past year and, hence, indicates both the number of admissions and the level of patient activity at a facility. Hospital patient census is synonymous with hospital occupancy rate.

Statistical methods

Descriptive and multivariate statistical ana-lyses for this paper were performed using SPSS® for Windows Version 6.01 [49] and STATA® Release 3.1 [50]. Bivariate tests for statistical significance for dichotomous predictor variables were based on conservative two-tail f-tests [51]. The main multivariate analyses were based on stepwise multivariate regression, in which aver-age patient occupancy was regressed on commu-nity ratings, commucommu-nity factors, institutional factors and quality factors in four steps [52]. Because this study was based on ecologic-level data, the main assumptions underlying "least squares" regression were closely examined This

by guest on May 8, 2016

examination indicated that none of these appeared violated [52]. For example, a normal-ized probability plot of the standardnormal-ized resi-duals by the observed resiresi-duals indicated that the full regression model (with all the variables) did not deviate from a multivariate normal distribution. Furthermore, a scatterplot of the predicted residuals by the standardized residuals indicated that the variance was constant. Finally,- a check for statistical leverage, using Cook's Distance, indicated that this was not a problem either [52]. Analysis of covariance (ANCOVA) also was used to analyse the results in terms of deviations from the grand mean [53]. To do this, hospital quality rating was divided into approximate thirds and then coded as an effects variable [53]. The stepwise model described above was then rerun. This permitted examination of the data in terms of deviations from the grand mean as blocks of variables (i.e. community, institutional and quality factors) were added to the model.

RESULTS

Table 1 shows that higher rated hospitals have higher occupancies (p<0.0001), as do hospitals in urban areas (p< 0.0001), with more beds (p< 0.0001), and those providing tertiary care (p< 0.0001). In addition, hospitals that are teaching facilities (/? = 0.002), have more employees per bed (/?<0.0001), have higher costs (j> = 0.009), have higher employee salaries (p = 0.001) and those that have higher case-mix indices (/? = 0.001), also have a higher patient occupancy. Based on previous reports, many of these associations are anticipated [24,27]. While useful, though, these associations are misleading because most of these variables are interrelated. For example, larger hospitals are often in urban areas, are more likely to be teaching hospitals and often have higher costs. In addition, Table 1 indicates that there are regional differences in this sample (/? = 0.008), as well as in terms of the average hospital size in a state (/> = 0.04). For the above reasons, a "stepwise" approach is used to adjust for intercorrelations between variables and for these two potential sample biases [53].

Table 2 shows the regression results. Model 1 indicates that community ratings predict patient

occupancy (j> < 0.0001), which is expected based on Table 1 results. Model 2 shows that commu-nity factors have a modest impact on reducing this association; only urbanicity (/? = 0.004) is statistically significant. However this is not the case in model 3, when institutional factors are also added to the model. In this case, hospital case-mix index is associated with patient occu-pancy (/> = 0.046), as is urbanization level

(p = 0.03). While reduced in magnitude,

com-munity rating is still significant in model 3 (see Table 2). In model 4, the situation is altered further. When quality-related factors are added to model 4, the association between quality rating and patient occupancy is not significant (/; = 0.302). Urbanization (p = 0.003), cost per admission (/? = 0.0001), teaching status (/> = 0.033) and employees per bed (j><0.0001) now are, however. The beta coefficients asso-ciated with this model indicate that occupancy is positively associated with urbanicity (beta = 0.273) and employees per bed (beta = 0.677). Occupancy is also negatively associated with costs per admission (beta= —0.441) and teach-ing status (beta= —0.166). Noteworthy here is that model 4 explains a substantial proportion of the variance in patient occupancy (R-square = 0.585).

Figure 1 shows the multivariate analyses depicted in terms of analysis of covariance (ANCOVA) results. For this, as noted, commu-nity hospital ratings were stratified into thirds, representing "low," "average," and "high" community ratings and then re-parameterized as an effects-coded variable [53]. Figure 1 shows the unadjusted occupancy results for this stra-tification (far left), followed by statistical adjustments for community factors, community institutional factors, then for community plus institutional plus quality-related factors (far right). Similar to the regression results, patient occupancy remains significant, until quality factors are added. While the final adjusted differences for the three ratings strata were not significant (p = 0.120), two trends are apparent here. First, the occupancy levels at "average-rated" hospitals remained constant at about the grand mean. Second, while the occupancy at low-rated hospitals increased and the occu-pancy at high-rated hospitals decreased as covariates were controlled, a gap of about 5% still appears to have remained.

by guest on May 8, 2016

TABLE 1. Ayerage hospital patient occnpancy by background factors

Background variable* (N) Group means SD t o r F p-valuef

Hospital rating} Low Moderate High SES index 99 or < 100 or > Urban level Rural Suburban Urban Black 19% or < 20% or > Hispanic 4% or < 5% or > Beds 200 or < 201-400 401 or > Care level Primary Secondary Tertiary Teaching No Yes Empl/bed 3.4 or < 3.5 or > Affiliation Secular Nonsecular Mortality 13% or < 14% or > Cost/Adm $3 800 or < $3 801 or > Average salary $20 300 or < $20 301 or > Region North South Midwest West Case-mix index § Low High

State bed average

230 or < 231 or > (52) (51) (52) (76) (79) (33) (83) (39) (113) (42) (124) (31) (45) (65) (45) (59) (52) (44) (121) (34) (102) (53) (111) (44) (119) (36) (76) (79) (70) (85) (52) (30) (68) (5) (83) (72) (59) (96) 64.64 69.42 74.62 67.91 71.14 59.04 71.99 73.28 68.90 71.34 70.21 66.97 60.28 70.55 77.41 64.30 70.52 75.47 67.97 75.20 66.10 76.21 68.99 70.98 70.29 67.14 66.97 72.05 65.98 72.51 72.99 64.69 69.72 60.92 66.52 73.07 67.02 71.12 13.01 11.45 8.47 12.94 11.16 12.64 10.48 10.09 12.00 12.46 12.00 12.54 13.45 9.45 7.19 13.85 10.60 7.79 12.58 8.32 12.31 8.56 12.89 9.95 12.10 12.09 13.76 9.78 13.47 10.07 10.19 14.68 11.24 15.56 13.13 9.84 13.65 •10.88 9.81 p<0.0001 -1.67 -1.11 1.33 -3.16 -5.34 -0.92 1.37 -2.66 -3.45 4.08 0.098 19.78 p<0.0001 = 0.267 0.186 31.96 p< 0.0001 12.56 /><0.0001 p = 0.002 p<0.0001 p-0.359 p -=0.173 /> = 0.009 p = 0.001 p = 0.008 -3.47 p = 0.001 -2.07 /> * See method section for additional information on variables.

t All p-vahies are based on two-tail tests.

X Low, moderate and high ratings were defined by dividing community ratings into approximate thirds.

§ Low case-mix index defined as 1.1470 or below [44].

by guest on May 8, 2016

TABLE 2. Multiple regression predicting average patient occupancy from hospital background factors*

Background variable

Model 1 Model 2 Model 3 ModeU

b beta /rvalue b beta /7-value b beta />-value b beta /7-Value

Rating RACE(SQRT) HISPANIC(LOG) SES INDEX URBAN BEDS AFFILIATION COST/ADM CASE-MTX AVG.SALARY CARE LEVEL TEACHING EMPL/BED MORTALITY Intercept R R2 Adj.R2 Stan.Error /7-Value 11.38 -25.478 = .517 - . 2 6 7 = .243 -10.561 <.0001 0.395 p<. 0001 9.66 .16 - 1 . 5 6 .05 3.87 20.210 .591 .349 .308 10.091 <.0001 0.336 /7<.0001 5.14 0.179/7 = 0.040 2.36 .033 /7 = .746 .06 .013/7-.893 - . 0 7 - . 1 3 8 /7 = .06O - 1 . 4 7 -.129/7-.O8O - . 1 3 .130 /7-.122 .07 .159/7 = .O77 .03 .285 /7 = .OO4 3.16 .232/7 = .O3O 3.72 .01 - . 7 7 - . 3 8 17.83 .13 .185 p".082 - . 0 2 9 / 7 - . 6 8 4 - . 0 4 4 p = .686 .179^=.046 .036 p = .730 .01 - . 8 6 - 3 . 7 9 10.24 .59 13.549 .633 .401 .341 9.852 <.0001 - . 2 1 - 4 . 8 5 9.57 — J22 14.892 .765 .585 .530 8.314 <.0001 0.082 p- 0.302 -.016/)=-.854 - . 0 1 1 / j = .859 .065/7 = .400 .273/7 = .003 .178/7-.134 - . 0 3 2 / 7 = .595 - . 4 4 1 / 7 = .0001 .103/7-.228 .163/7 = .O71 - . 0 5 2 / 7 = .690 - . 1 6 6 / 7 - . 0 3 3 .677/7<.OO01 - . 0 3 5 / 7 = .572

*AJ1 regression models include geographic region (REGION) and the average hospital bed size in the state (STATE BED). REGION was coded as a four-category dummy variable (with one of these used as the reference category). STATE BED was coded as a five-point ordinal scale. Since RACE and HISPANIC were not normally distributed, square root and logistic transformations were used, respectively, to normalize the variables [52].

DISCUSSION

Initial bivariate analysis suggests that hospi-tals with better community ratings, in non-rural areas, with more beds and employees, that are teaching institutions, with more tertiary care, with higher case-mix indices and hospitals that are more expensive, have higher patient occu-pancies. Multivariate analysis suggests that hospitals in more urbanized areas, with lower costs, that are nonteaching facilities and that have more employees per bed, have higher patient occupancies. However, while the multi-ple regression analysis suggested that patient occupancy was not associated with community image after differences between institutions were controlled, analysis of covariance evaluating "high," "average," and "low" hospital ratings suggested that both positive and negative biases could exist.

Based on the ANCOVA results, an interaction was suspected between hospital rating and hospital size. To test this hypothesis, a

RATING x BEDS interaction term was added to the final regression model (Table 2, model 4). This interaction term was found to be significant 0 = 0.0152). To interpret this effect, a condi-tional effects plot was constructed employing the technique recommended by Cohen and Cohen [53]. This method uses the intercept and the b coeflBcients derived from the regression model with the interaction term and uses these values to generate a plot of "patient occupancy" by "hospital size" as follows. For this plot, RATING is used to predict the "Y-hat" for patient occupancy based on "high" (JCI = 4.2), "average" {x2 = 3.6), and "low" (x3 = 2.8) values for this variable, using the b coefficients for the model with the interaction term, together with the mean values for the other x variable values, with the variable BEDS plotted at 100, 250, 400 and 550 beds, respectively (see Fig. 2). This plot shows that controlling for differences between facilities, smaller hospitals with low ratings are more likely to have lower patient occupancies, which is not the case for larger hospitals with

by guest on May 8, 2016

6.0 4 . 0 2 . 0 -0.0 -2.0- -4.0--6.0 5.1 4.9 (F=9.81 df=2/152 p<001) -0.1 -4.9 (F=9.91 df=2/139 p<.001) -0.2 -4.7 3.3 (F=3.35 df=2/134 p..038) -0.3 -3.0

Unadjusted Community Facility

2.5 (F=2.20 df=2/130 p=.12O) -0.3 -2.2 S Low Ratings • Average Ratings • High Ratings Quality

'Result! represent dertatloa from (rand mean and are cumolatlTe, left to rljht.

FIGURE 1. Deviations in hospital occupancy by perceived quality, controlling for community, facility and quality differences, based on analysis of covariance*.

these ratings (Fig. 2). Furthermore, smaller hospitals with high ratings have consistently higher occupancies, so there appears to be an admissions bias for smaller hospitals with low ratings, not all hospitals in general as hypothe-sized.

It is believed that the findings presented are

significant, because assessment of medical effec-tiveness by experts and the public's perception of quality are increasingly related issues. This is a consequence of the trend to use satisfaction and perceived quality data to define medical out-comes [2,13]. Health researchers have recently developed patient-reported outcomes surveys

80 CO o. u u

o

2

2

4

Number of Hospital Beds

FIGURE 2. Hospital occupancy by bed size and perceived quality interaction effect, based on multiple regression.

by guest on May 8, 2016

that appear to be reliable and appear to have acceptable validity [15,54]. However, as pre-viously noted, research also indicates that patients often perceive quality and select medical care based on not only the services evaluated, but also on the "subjective" aspects of such care [27-31,55]. In addition, research has shown that the public is often not aware of quality differ-ences between facilities, unless they have been covered widely in the local media [56]. For example, in the US when HCFA reported hospital mortality statistics in the 1980s [11], the public was found to be generally unin-formed, though this report received extensive national media coverage [57].

The meaningful assessment of patients' out-comes typically requires solving two problems [58]. These include the measurement of out-comes themselves and the adjustment for "risks" for various outcomes. In medicine, the "gold standard" for solving these has been randomized clinical trials or RCTs [37]. How-ever, because of practical impediments, few randomized clinical trials can be performed in health care settings [58]. Consequently, much medical outcomes research must rely on non-experimental data collected among subjects who have been exposed to events outside of the control of investigators [59]. Thus, the usefulness of these assessments rely not only on whether adjustments for "risks" allow proper compar-ison, but also on how well the clinical issues have been defined and whether the data capture the necessary clinical features [60]. An important recent development in this regard has been the Medical Outcomes Study (MOS) approach [54]. In contrast to earlier methods that relied on outcomes related to medical morbidity and mortality, the MOS method focuses on psycho-metric measures of functioning and well-being that give more weight to patient-relevant out-comes [54]. Another noteworthy development also includes the Patient-Centered Care (PCQ approach, which focuses on assessing patient satisfaction based on the patient's perception, rather than that of the health care professional [61]. Together, both the MOS and PCC approaches, in part, represent alternative efforts to capture the critical "clinical" elements of the treatment encounter or episode, given the restrictions of conducting RCTs in health care settings. Recently, however, as suggested here,

some investigators have warned that patient-reported quality indicators are often neither understood statistically nor methodologically and that caution is advised when evaluating specific clinical outcomes based on these mea-surements [62]. As shown here, it is possible that smaller facilities with poor images may be disadvantaged based on subjective patient rat-ings.

This paper has some limitations. The first relates to the so-called problem of "ecological fallacy" [39]. That is, it is possible that this study was biased because it was conducted at the institutional and community levels. The results may have been different if this study was conducted with individual householders or patients. For example, the results presented may have been confounded because key vari-ables were not included, such as the hospital affiliation of the householder's physician. Also, the patient's specific proximity to the hospital in the service area was not known. This may have been important for some of the analyses con-ducted. A second limitation is that the data were not based on a random sample of hospital service areas, but service areas in which market-ing surveys were conducted. These communities may be different from communities where these studies were not undertaken. A third limitation is that this paper was based on data from 1984 to 1988, a period characterized by deregulation and increasing competition in the US. It could be argued that the 1990s represent a different health care marketplace, although there is still some consensus that future reforms in the US should include consumer choice and patient-oriented data [17]. Fourth, the quality factors studied (mortality, care level, teaching status, and employees per bed) were limited in terms of depth and breadth and it probably would have been useful to have included other quality indicators, such as JCAHO accreditation level.

Nevertheless, given these limitations, it is clear that medical outcomes based on patient self reports, including information of a subjective nature and use of nonexperimental research designs, will be used to evaluate medical care in the US and other countries in the future. In addition, in whole or in part, market-based models involving patient choice, marketing promotion, and public service information pro-grams will also be employed in many places to

by guest on May 8, 2016

enhance public health and encourage organiza-tional efficiency in the foreseeable future. Based on the findings reported, however, it is suggested that while use of patient perceptions in outcomes is important, these indicators may have certain biases associated with them and will likely require future methodological investigations.

Acknowledgements: Support for this research was provided, in part, by the National Institute of Mental Health Grant # MH-19105. A version of this paper was presented at the Association for Health Services Research, 12th Annual Meeting, Chicago, IL, USA, June 1995.

REFERENCES

1. Cleary P D and McNeil B J, Patient Satisfaction as an Indicator of Quality. Inquiry 1988; 25: 25-36.

2. Lohr K N, Yordy K D and Thier S O, Current Issues in Quality Care. Health Affairs 1988; 7: 5 -18.

3. US Congress, Office of Technology Assessment. The Quality of Medical Care: Information for Consumers. US Government Printing Office (OTA-H-386), Washington, DC, 1988.

4. Patrick D L and Erickson P, Health Status and Health Policy: Quality of Life in Health Care Evaluation and Resource Allocation. Oxford Press, New York, NY, 1993.

5. Grady M L and Schwartz H A, editors. Medical Effectiveness Research Data Methods. US Department of Health and Human Services (AHCPR Pub. No. 92-0056), Rockville, 1992. 6. Wenzel R P, editor. Assessing Quality Health

Care: Perspectives for Clinicians. Williams and Wilkins, Baltimore, MD, 1992.

7. Boscarino J A and Steiber S R, The Future of "Marketing" Health Care Services. In: Health Care Marketing: A Foundation for Managed Quality, 3rd ed. Edited by Philip D. Cooper. Aspen Publications, Gaithersberg, MD, pp.73-83, 1994.

8. Meier B, Health Plans Promote Choice, But Decisions May Not be Easy: Hurdles Loom on Rating Doctors and Hospitals. New York Times, pp. 1 and 8, 31 March 1994.

9. Shortell S M and Reinhardt U E, Improving Health Policy and Management: Nine Critical Research Issues for 1990s. Health Administration Press, Ann Arbor, MI, 1992.

10. American Public Health Association. HCFA Moves to Get Word Out The Nation's Health, p. 4, (August), 4, 1994.

11. Brinkley J, US is Releasing Lists of Hospitals with Abnormal Mortality Rates. New York

Times, pp. 1 and 22, 12 March, 1986.

12. Salive M E, Mayfield J A and Weissman N W,

Patient Outcome Research Teams and the Agency for Health Care Policy and Research. Health Services Research 1990; 25: 697-708. 13. Davies A R and Ware J E, Involving Consumers

in Quality of Care Assessment. Health Affairs 1988; 7: 33-48.

14. Nelson E C, Hays R D, Larson C D and Batalden P B, The Patient Judgment System: Reliability and Validity. Quality Review Bulletin 1989; 15: 185-191.

15. Stewart A L, Sheldon G, Hays R D, Wells K, Rogers W H, Berry S D, McGlynn E A and Ware J E, Functional Status and well-being of patients with chronic conditions: results from the medical outcomes study. Journal of the American Medical Association 1989; 262: 907-913. 16. Cooper P D, editor, Health Care Marketing: A

Foundation for Managed Quality, 3rd cd. Gaithersburg, MD: Aspen Publications, 1994. 17. Shortell S M, Morrison E M and Friedman B,

Strategic Choices for America's Hospitals: Man-aging Change in Turbulent Times. Jossey-Bass Publishers, San Francisco, CA, 1990.

18. Shortell S M and Zajac E J, Health Care Organizations and the Development of Strategic-Management Perspective. In: Innova-tions in Health Care Delivery: Insights for Organizational Theory. Edited by Stephen S. Mick and Associates. Jossey-Bass Publishers, San Francisco, CA, pp. 144-180, 1990.

19. Griffith J R, Principles of the well-managed community hospital. Hospital and Health Ser-vices Administration 1989; 34: 457-470.

20. National Committee for Quality Assurance. Executive Summary for Report Card Pilot Pro-ject. National Committee for Quality Assurance,

Washington DC, 1995.

21. Group Health Association of America. GHAA's Consumer Satisfaction Survey and User's Manual, 2nd ed., Group Health Association of America, Washington DC, 1991.

22. Begun J W, Luke R D and Pointer D D, Structure and Strategy in Hospital-Physician Relationships. In: Innovations in Health Care Delivery: Insights for Organizational Theory. Edited by Stephen S. Mick and Associates. Jossey-Bass Publishers, San Francisco, CA, pp.116-143, 1990.

23. MacStravic R S, Outcome Marketing in Health Care. Health Care Management Review 1988; 13: 53-59.

24. Boscarino J A, The public's perception of quality hospitals II: implications for patient surveys. Hospital and Health Services Administration 1992; 37: 13-35.

25. Steiber S R and Krowinski W J, Measuring and Managing Patient Satisfaction. American Hos-pital Publishing, Chicago, IL, 1990.

26. Strasser S and Davis R M, Measuring Patient Satisfaction for Improved Patient Services. Health Administration Press, Ann Arbor, MI, 1991.

by guest on May 8, 2016

27. Boscarino J A and Steiber S R, Hospital Shopping and Consumer Choice. Journal of Health Care Marketing 1982; 2: 15-23.

28. (a). Lane P M and Lindquist J D, Hospital Choice: A Summary of the Key Empirical and Hypothetical Findings of the 1980s. Journal of Health Care Marketing 1988; 8: 5-20.

(b). Jensen J, Choosing a Hospital. American Demographics 1987; 9: 45-47.

29. Smith S M and Clark M, Hospital image and the positioning of service centers: an application in market analysis and strategy development. Jour-nal of Health Care Marketing 1990; 10: 13-22. 30. Arnwine D, Buyers are smarter, looking for

quality. Modern Healthcare 1987; 17: 32. 31. Hays M D, Consumers Base Quality Perceptions

on Patient Relations. Staff Qualifications. Modem Healthcare 1987; 17: 33.

32. Joint Commission. The Measurement Mandate. Joint Commission on Accreditation of Health-care Organizations, Oakbrook Terrace, IL, 1993. 33. Steiber S R, Making use of patient survey data. Health Care Strategic Management 1989; 7:12-14. 34. National Center for Health Statistics. Utilization of ShortStay Hospitals. United States. 1986 -Annual Summary. Public Health Service, DHHS Publication No. (PHS) 88-1757, Hyattsville, MD, 1988.

35. American Hospital Association. AHA Guide to the Health Care Field. 1988 Edition. American Hospital Publishing, Chicago, IL, 1989. 36. Boscarino J A, The public's rating of hospitals.

Hospital and Health Services Administration 1988; 33: 189-199.

37. Pocock S J, Clinical Trials: A Practical Ap-proach. Wiley, New York, 1983.

38. Frank J D and Frank J B, Persuasion and Healing: A Comparative Study of Psychotherapy, 3rd ed., Schocken Books, New York, NY, 1991. 39. Last J M, A Dictionary of Epidemiology. Oxford

Press, New York, NY, 1988.

40. Kotler P, Principles of Marketing, 3rd ed., Prentice-Hall, Englewood Cliffs, NJ, 1986. 41. Schultz D E, Martin D and Brown W P,

Strategic Advertising Campaigns, 2nd ed., Crain Books, lincolnwood, IL, 1984. 42. Cutlip SM, Center AH and Broom GM,

Effective Public Relations. 6th ed. Prentice-Hall, Engelwood, NJ, 1985.

43. CACI. The Sourcebook of Zip Code Demo-graphics, 7th ed., CACI Marketing Systems, Fairfax, VA, 1990.

44. US Government. Medicare program; changes to the inpatient hospital prospective payment sys-tem and fiscal year 1987 rates. Federal Register 51 (no. 170, September 3): 31454-31603, 1986. 45. Center for Study of Services. Consumers' Guide

to Hospitals. Center for Study of Services, Washington DC, 1988.

46. Green P E and Tull D S, Research for Marketing Decisions, 4th ed. Prentice-Hall, Engehvood Cliffs, NJ, 1978.

47. Steiber S R, How consumers rate health-care quality. Health Care Strategic Management 1988; 6:6-8.

48. Newton M and Steiber S R, Hospital image based on quality, fact and perception. Health Care Strategic Management 1990; 8: 16-19. 49. SPSS, Inc. SPSS for Windows (Release 6.01).

SPSS, Inc., Chicago, IL, 1994.

50. Stata Corporation. Stata (Release 3.1). Stata Corporation, College Station, TX, 1993. 51. Blalock H M Jr, Social Statistics, 2nd ed.,

McGraw-Hill, New York, NY, 1972.

52. Neter J, Wasserman W and Kutner M H, Applied Linear Statistical Models, 3rd ed., Irwin, Homewood, IL, 1990.

53. Cohen J and Cohen P, Applied Multiple Regres-sion/Correlation Analysis for the Behavioral Sciences, 2nd ed., Lawrence Erlbaum Associates, Hillsdale, NJ, 1983.

54. Stewart A L and Ware J E Jr, editors. Measuring Functioning and Weil-Being: The Medical Out-comes Study Approach. Duke University Press, Durham, NC, 1992.

55. Javalgi R G, Rao S R and Thomas E G, Choosing a hospital: analysis of tradeoffs. Journal of Health Care Marketing 1991; 11: 12-22.

56. Boscarino J A, Measuring the impact of negative publicity and repairing the damage. Health Care Strategic Management 1988; 6: 10-11.

57. Gross P A and Schaffer W A, Consumer awareness of hospital mortality data. Journal of Health Care Marketing 1989; 9: 52-55.

58. Iezzoni L I, editor. Risk Adjustment for Measur-ing Health Care Outcomes. Health Administra-tion Press, Ann Arbor, MI, 1994.

59. Cook T D and Shadish W R, Social Experi-ments: Some Developments Over the Past Fif-teen Years. In: Annual Review of Psychology. Edited by Lyman W. Porter and Mark R. Rosenzweig. Annual Reviews, Inc., Palo Alto, CA, pp.545-580, 1994.

60. Markson L E, Nash D B, Louis D Z and Gonnella J S, Clinical outcomes management and disease staging. Evaluation and the Health Professions 1991; 14: 201-227.

61. Cleary P D, Edgman-Levitan Walker J D, Gerteis M and Delbanco T L, Using patients reports to improve medical care: a preliminary report from 10 hospitals. Quality Management in Health Care 1993; 2: 31-38.

62. Testa M A and Nackley J F, Methods for Quality-of-Life Studies. In: Annual Review of Public Health, Edited by Gilbert S O, Fielding J E and Love L B. Annual Reviews Inc, Palo Alto, CA, pp. 535-539, 1994.

by guest on May 8, 2016