Do patient,hospital,and community characteristics predict variations in overall inpatient experience scores? A multilevel analysis of hospitals in California

Abstract

The purpose of this cross-sectional study was to examine how patient, hospital, and community characteristics explain variations in overall inpatient experience with care. We used data from the Patients’ Evaluations of Performance in California survey, the American Hospital Association annual hospital survey, and the Area Resource File. The sample consisted of 24,887 adult patients who received either medical or surgical inpatient care in 173 hospitals located in 46 California counties. A null hierarchical linear model for overall inpatient experience showed that 96.17%, 3.24%, and 0.59% of the variations were within hospitals, between hospitals, and between communities, respectively. Conditional models showed that patient characteristics (sex, age, education, health status, and service line) explained 10.95% of the within-hospital variations; hospital characteristics (teaching status, registered-nurse staffing intensity, and resources directed to patient care) explained 34.12% of the between-hospital variations; and community characteristics (hospital competition, teaching hospitals, per-capita income, and percentage of minority population) explained 99.33% of the between-community variations. These findings suggest that multilevel variations need to be considered when patient experiences are compared across hospitals. Larger future studies are needed to understand how patient experience with care may vary based on patient health-care provider communication across patient subgroups.

Keywords

community characteristics hospital characteristics hospital performance multilevel patient characteristics patient experience

Introduction

Patient experience is a key component of health care quality.^1,2 Yet the literature on the predictors of hospital inpatient experience with care is underdeveloped. Most extant studies have focused primarily on patient characteristics associated with patient experience,³ whereas only a few studies have considered the organizational and community context of these measures.^4,5 It has been suggested that the practice of ignoring the multilevel nature of patient experience data makes it difficult to distinguish the relative contribution of each data level to the overall variability in patient experience scores.⁶

A key reason for the limited multilevel analysis of patient experience with care is the historical lack of large-scale administration of standardized surveys across hospitals. The Picker Institute's—now the National Research Corporation (NRC)—inpatient survey⁷ is one of the first standardized surveys of patient experience with hospital care. The NRC/Picker inpatient survey has been used both in the U.S. and internationally.⁸

Using data from a multihospital state-wide sample of hospitalized medical and surgical patients from California, USA, this study extends the literature by examining the role of patient, hospital, and community characteristics on an overall inpatient experience with care composite consistent with the NRC/Picker methodology.

Methods

Data sources

We used three secondary data sources in this cross-sectional study. We derived patient survey data—overall inpatient experience with care measures and patient-level characteristics—from the 2002 Patients’ Evaluation of Performance in California (PEP-C) adult survey, conducted by the NRC/Picker with funding from the California Institute for Health Systems Performance and the California Health Care Foundation. We derived hospital-level characteristics from the 2002 American Hospital Association’s (AHA) survey of hospitals and derived community-level characteristics from the 2002 Area Resource File (ARF) data. No patient identifiers were included in the data file that we obtained, and the Office of Institutional Review Board at The Pennsylvania State University has determined the study as a “not human-subject” research.

PEP-C data

PEP-C was a mailed survey that targeted all general acute care hospitals in California. A total of 181 hospitals participated in the survey, accounting for approximately 47% of all eligible hospitals, 51% of hospital discharges, and 54% of the licensed beds in California.⁹ We assessed potential non-participation bias by comparing California hospitals that participated in the survey with those that did not. This comparison found no significant differences in ownership status, teaching status, occupancy rate, nurse-staffing levels, and maintenance of a separate nursing home unit between PEP-C participants and nonparticipants. However, PEP-C participant hospitals had some statistically significant differences compared to non-participant hospitals (P < 0.05): 6% more Medicare days, 8% lower Medicaid days, and 11% higher mean resources directed to patient care.

Participating hospitals provided the NRC/Picker with complete lists of their adult medical, surgical, and maternity patients who stayed for at least one night and were discharged between 1 July 2002 and 31 October 2002. Patients were excluded if they met any of the following exclusion criteria: admission for psychiatric or substance abuse treatment, admission for observational purposes, death during hospitalization, and discharge to a setting other than home. For each hospital, NRC invited a random sample of 600 patients—200 patients discharged from each of the three service lines—to participate in the survey. If a hospital had less than 600 discharges, all patients were invited to participate in the survey. The average response rate among patients across all participating hospitals was 45%.⁹

Sample

Our study focused on surveys from medical and surgical patients who received their care in acute short-term hospitals. We excluded all surveys from maternity patients because they differ from medical and surgical patients in many ways, and excluded uninsured patients because they accounted for a small proportion of the respondents. Our final analytic sample consisted of 24,887 patient respondents representing 173 hospitals in 46 California counties.

Variables

Dependent variable: The PEP-C survey had seven rating measures for inpatient experience with care: (1) How would you rate the courtesy of the staff that admitted you? (2) How would you rate the courtesy of your doctors? (3) How would you rate the courtesy of your nurses? (4) How would you rate the availability of your doctors? (5) How would you rate the availability of nurses? (6) How would you rate how well the doctors and nurses worked together? and (7) Overall, how would you rate the care you received at the hospital? These items used a 5-point Likert scale: poor = “1”; fair = “2”; good = “3”; very good = “4”; and excellent = “5”. Confirmatory factor analyses (CFA) with full information maximum likelihood (FIML) were fitted to test how these seven rating items load on an overall inpatient experience with care latent variable. The FIML estimator was used because it accounted for the presence of missing data. We used values of 0.90 or above to determine acceptable incremental model fit indices—Normed Fit Index (NFI), DELTA1, Relative Fit Index (RFI), Incremental Fit Index (IFI), Tucker-Lewis Index (TLI), and Comparative Fit Index (CFI)¹⁰—and used values of 0.06 or less to determine model fit using the Root Mean Square Error of Approximation (RMSEA).¹¹ We summed the items from the final CFA model to create an overall rating of inpatient care composite and used Cronbach’s alpha to calculate the internal-consistency reliability of this composite.¹²

CFA showed that the seven-item latent variable did not have acceptable model-fit parameters. Instead, the best-fitting model was the one that omits the “How would you rate the availability of your doctors?” item. The resulting six-item model had an overall model fit of χ² (9) = 8883, P < 0.01, but had excellent incremental model fit indices (NFI = 0.98; RFI = 0.95; IFI = 0.98; TLI = 0.95; CFA = 0.98; RMSEA = 0.09) and a high Cronbach’s alpha of 0.90. The standardized estimates and squared multiple correlations for this final model are presented in Table 1.

Table 1.

Confirmatory factor analysis of the final overall inpatient experience with care scale (n = 24,887).

Item^a	Standardized factor loading^b	Squared multiple correlations^b
How would you rate the courtesy of the staff that admitted you?	0.68	0.47
How would you rate the courtesy of your doctors?	0.64	0.41
How would you rate the courtesy of your nurses?	0.84	0.71
How would you rate the availability of nurses?	0.82	0.67
How would you rate how well the doctors and nurses worked together?	0.87	0.75
Overall, how would you rate the care you received at the hospital?	0.91	0.82

Omitted item: “How would you rate the availability of your doctors?”

Estimated are statistically significant (P < 0.05).

Patient-level characteristics: We considered the following patient-level independent variables: age, gender, education, race, self-reported health status, health insurance type, and line of service. We considered these characteristics because previous studies have suggested they were associated with patient experience with care.^3,13–15 Age was a continuous variable and the remaining variables were indicator variables (Yes = 1, No = 0). We measured gender as female versus male; education as less than high school versus high school or higher education; race as non-white versus white; self-reported health status as fair or poor versus excellent, very good, or good health status; health insurance as Medicare or Medicaid versus private insurance; and measured line of service as surgical versus medical service.

Hospital-level characteristics: We considered the following hospital-level independent variables: ownership status, teaching status, occupancy rate, payer mix, nurse staffing intensity, and resources directed to patient care. We measured ownership status as an indicator variable: whether a hospital was government-owned versus not government-owned. We measured teaching status as a summary score of three indicators: whether a hospital was a member of the Council of Teaching Hospitals or not; whether a hospital had a residency training program that is approved by the Accreditation Council for Graduate Medical Education (ACGME) or not; and whether the hospital had a medical school that reported to the American Medical Association (AMA) or not. We calculated occupancy rate as total inpatient days divided by the product of staffed beds and 365. We used two measures of payer mix: Medicare inpatient days and Medicaid inpatient days, calculated as percentages of the total inpatient days for each hospital, respectively. We measured nurse staffing intensity as registered nurse (RN) full-time equivalents (FTEs) and licensed practical nurse (LPN)/licensed vocational nurse (LVN) FTEs, and calculated these measures as ratios of the total inpatient days for each hospital. We calculated the financial resources directed to patient care as the labor expense per total FTEs (i.e. the sum of hospital total payroll expenses and employee benefit expenses, divided by total FTEs). Because of significant missingness in unit-level (service line) variables in the AHA data, we used total facility-level rather than unit-level AHA variables for nurse staffing and other volume-related variables. We also used an indicator for “whether a hospital maintained a separate nursing home (no hyphen before home) or not” as a control variable.

Community-level characteristics: We considered the following community-level independent variables: hospital competition, teaching status, nurse staffing levels, patient resources directed to patient care, health maintenance organizations (HMO) penetration rate, payer mix, excess capacity, per-capita income, and the percentage of minority population. We derived these variables from ARF, except hospital competition, which we calculated using the AHA data. The unit of analysis was the county for most variables,^16–18 except HMO penetration, which was measured in ARF at the metropolitan statistical area (MSA) level.

We calculated the Herfindahl-Hirschman index (HHI) as the sum of the squares of the individual hospital market (community) shares of total inpatient days per county to measure hospital competition.¹⁹ We measured teaching status as the number of short-term general hospitals in a given hospital’s county that were members of the Council of Teaching Hospitals. While the ARF data included total payroll expenses, they did not include any information about employee benefit expenses; therefore, the resources directed to patient care variable was measured as the ratio of total payroll expense for short-term general hospitals in thousands of dollars divided by the total short-term general inpatient days (in 1000s) per county. HMO penetration rate²⁰ was measured in ARF as the total HMO enrollment divided by the total population in the county. We measured nurse staffing intensity at the community level as the number of hospital RN FTEs and LPN/LVN FTEs in a hospital’s county per 1000 populations. To measure payer-mix, we calculated the total numbers for each of the Medicare and Medicaid inpatient days for all short-term general hospitals in the county per 1000 population. We measured excess capacity²⁰ for short-term general hospital beds by subtracting the county’s occupancy rate from 100% (we measured occupancy rate as the total number of inpatient days for short-term general hospitals in a county divided by the product of the total number of the county’s short-term general hospital beds and 365). We also used mean per-capita income per county and the percentage of racial/ethnic minorities (calculated as one minus the percentage of the white population (one race alone)) as measures of the sociodemographic characteristics. We did not include rural versus urban status measures because they were highly collinear with hospital HHI and HMO penetration.

Analytic approach

We calculated descriptive statistics (means, standard deviations, and ranges) and Pearson’s correlations for the study variables. We fitted a null hierarchical linear model (HLM) with a random intercept for the overall inpatient experience with care composite to estimate within-hospital, between-hospital, and between-community variations. Next, we estimated conditional HLM models.²¹ In determining our models, we dropped independent variables with beta coefficients less than 1.5 times their estimated standard errors: we used this criterion because we were more concerned about the bias that might arise from failing to specify independent variables, rather than about the lack of efficiency that arises when the model is slightly overfit.²¹ We first entered patient-level independent variables only to the HLM model, then dropped independent variables that did not meet our retention criterion (Model 1). Second, we added hospital-level independent variables to Model 1 then dropped independent variables that did not meet our retention criterion (Model 2). Lastly, we added community-level independent variables to Model 2, then dropped independent variables that did not meet our retention criterion (Model 3). We grand-mean centered all independent variables in order to make the intercept more meaningful. The equation for the final fully conditional model (Model 3) is

Y_{ijk} = γ_{000} + \sum γ_{00 x} \times (community-levelpredictors)_{k} + \sum γ_{0 x 0} \times (hospital-level predictors)_{jk} + \sum γ_{x 00} \times (patient-levelpredictors)_{ijk} + u_{00 k} + r_{0 jk} + e_{ijk}

where γ₀₀₀ was interpreted as the adjusted average hospital mean overall inpatient experience with care; the terms γ_00x, γ_0x0, and γ_x00 were the regression coefficients for community-, hospital-, and patient-level independent variables, respectively. γ_00x was the average degree to which a given community-level variable predicted the mean outcome; γ_0x0 was the average degree to which a given hospital-level variable predicted the mean outcome; and γ_x00 was the average degree to which a patient-level variable predicted the mean outcome. Also, u_00k, r_0jk, and e_ijk represented the residual between-community, between-hospital, and within-hospital variances, respectively.

We manipulated the data in SAS (SAS Institute, Inc.), fitted CFA models in Amos (IBM SPSS Amos), and fitted multilevel models in HLM (Scientific Software International, Inc.).

Results

Descriptive results

Panel 1 of Table 2 presents the descriptive results for the patient-level variables. The overall mean score for this overall inpatient experience with care composite was 21.49 (SD = 5.12). Slightly more than half (53%) received surgical services, and the remaining 47% received medical services during hospitalization. Females made up 57% of the sample, and the mean age was 63 (SD = 17.04) years. Only 16% had less than high school education. Nearly a third (31%) described their health status as poor or fair. Non-whites accounted for 32% of the respondents. While 36% reported having Medicare insurance, 13% reported having Medicaid insurance, and the remaining 51% reported having private health insurance.

Table 2.

Descriptive statistics.

Variables	Coding and range	Mean	SD
Patient-level variables (n = 24,887)
Overall experience with care composite score (6 items)	(1, 28)	21.49	5.12
Female	0 = no; 1 = yes	0.57	0.49
Age	(18, 102)	63.00	17.04
<High school	0 = no; 1 = yes	0.16	0.36
Non-white	0 = no; 1 = yes	0.32	0.47
Low health status	0 = no; 1 = yes	0.31	0.46
Medicare insured	0 = no; 1 = yes	0.36	0.48
Medicaid insured	0 = no; 1 = yes	0.13	0.33
Surgical service	0 = no; 1 = yes	0.53	0.50
Hospital-level variables (n = 173)
Government hospital	0 = no; 1 = yes	0.19	0.39
Teaching status	(0, 3)	0.62	1.02
Occupancy rate	(23.81, 178.26)	65.58	18.71
Percent Medicare INP days	(0.00, 88.40)	43.40	15.43
Percent Medicaid INP days	(0.00, 93.31)	20.51	17.90
RN staffing intensity	(0.58, 24.38)	5.53	2.91
LPN/LVN staffing intensity	(0.00, 7.73)	0.73	0.76
Resources directed to patient care	($4,776, 121,099)	$61,044	$17,917
Nursing home unit	0 = no; 1 = yes	0.39	0.49
Community-level variables (n = 46)
Hospital competition	(0.06, 1.00)	0.52	0.27
Teaching status	(0.00, 9.00)	0.52	1.53
RN staffing intensity	(1.00, 4.93)	2.34	0.87
LPN/LVN staffing intensity	(0.07, 1.73)	0.42	0.30
Resources directed to patient care	($482.24, $5,376)	$1096	$734.35
Medicaid days per 1000 population	(15.87, 680.52)	132.94	141.84
HMO penetration	(0.00, 0.74)	0.33	0.22
Excess capacity	(−15.23, 76.95)	37.60	15.78
Per-capita income	($16,112, $60,618)	$29,207	$10,710
Percent minority population	(6.60, 51.30)	29.07	14.13

HMO: Health maintenance organization; INP: inpatient; LPN: licensed practical nurse; LVN: licensed vocational nurse; RN: registered nurse.

Hospital characteristics are presented in Panel 2 of Table 2. Less than a quarter (19%) were government-owned (federal/non-federal), and the remaining 81% were non-government-owned (7% were for-profit and 74% were not-for-profit) hospitals. The mean teaching status score was 0.62 (SD = 1.02). Overall, the mean occupancy rate across hospitals was 65.58% (SD = 18.71%), and the mean percentages of Medicare and Medicaid inpatient days were 43.40% (SD = 15.43%) and 20.51% (SD = 17.90%), respectively. The mean RN FTEs per thousand inpatient days ratio was about 5.53 (SD = 2.91), and the mean FTE LPNs per thousand inpatient days was 0.73 (SD = 0.76). The mean financial resources committed to patient care ratio was $61,044 (SD = $17,917). Overall, 39% of the hospitals reported maintaining a separate nursing-home unit.

Community characteristics of the 46 California counties represented in this study are presented in Panel 3 of Table 2. The mean RN FTEs and LPN/LVN FTEs per thousand population were 2.34 (SD = 0.87) and 0.42 (SD = 0.30), respectively. The mean teaching status was 0.52 (SD = 1.53), while the mean Medicaid inpatient days per thousand populations was 132.94 (SD = 141.84). The mean resources directed to patient care was $1096 (SD = $734.35) per 1000 inpatient days. The mean hospital HHI was 0.52 (SD = 0.27). The mean HMO penetration rate was 0.33 (SD = 0.22). The mean excess capacity was 37.60% (SD = 15.78%). The mean of per-capita income was $29,207 (SD = $10,710), and the mean percentage of non-white population was 29.07% (SD = 14.13%). The correlations between the study measures are presented in Table 3.

Table 3.

Correlations of independent variables^a.

	Female	Age	<High school	Non-white	Low health status	Medicare insured	Medicaid insured	Surgical service
Correlations of patient-level variables
Female	1.00
Age	−0.06	1.00
<High school		0.11	1.00
Non-white	0.02	−0.21	0.26	1.00
Lower health status		0.18	0.15	0.06	1.00
Medicare	−0.07	0.58	0.02	−0.17	0.11	1.00
Medicaid	0.04	−0.07	0.25	0.24	0.15	−0.29	1.00
Surgical service	0.04	−0.17	−0.07		−0.23	−0.10	−0.08	1.00
	Government hospital	Teaching status	Occupancy rate	Percent Medicare	Percent Medicaid	RN staffing intensity	LPN staffing intensity	Resources patient care	Nursing home unit
Correlations of hospital-level variables
Government hospital	1.00
Teaching status		1.00
Occupancy rate		0.20	1.00
Percent Medicare INP days	−0.38	−0.20	−0.16	1.00
Percent Medicaid INP days	0.42			−0.51	1.00
RN staffing intensity			−0.31		−0.29	1.00
LPN/LVN staffing intensity		−0.18	−0.27			0.52	1.00
Resources directed to patient care						−0.17	−0.20	1.00
Nursing home unit	0.18				0.39	−0.33			1.00
	Hospital competition	Teaching status	RN staffing intensity	LPN staffing intensity	Resources patient care	Medicaid days	HMO penetration	Excess capacity	Per-capita income	Percent minority population
Correlations of community-level variables
Hospital competition	1.00
Teaching status	−0.41	1.00
RN staffing intensity		0.29	1.00
LPN/LVN staffing intensity	0.34		0.41	1.00
Resources patient care					1.00
Medicaid days per 1,000 population			0.37	0.67	−0.31	1.00
HMO penetration	−0.64			−0.47			1.00
Excess capacity					0.53		−0.29	1.00
Per-capita income	−0.47						0.56	−0.25	1.00
Percept minority population	−0.49	0.41		−0.33			0.38			1.00

HMO: health maintenance organization; INP: inpatient; LPN: licensed practical nurse; LVN: licensed vocational nurse; RN: registered nurse.

Only significant coefficients are shown (P < 0.05).

HLM results

The null model showed that the within-hospital, between-hospital, and between-community variances in the overall inpatient experience with care composite were 25.31, 0.85, and 0.15, respectively (Table 4). Hence, 96.17% of the total variance was within hospitals, and 3.24% was between hospitals, with a scant 0.59% of the variance between communities.

Table 4.

Variances and variances accounted for at the patient, hospital, and community levels.

	Variance estimates
	No variables (Null model)	After entering patient variables (Model 1)	After entering patient and hospital variables (Model 2)	After entering patient, hospital, and community variables (Model 3)	Percentage of variance accounted for in the fully conditional model (Model 3)
Within-hospitals	25.31	22.53	22.53	22.53	10.98%
Between-hospitals	0.85	0.75	0.68	0.56	34.12%
Between-communities	0.15	0.15	0.04	0.001	99.33%

Table 5 shows the results of the significant independent variables in the conditional models. For Model 3, three level-1 independent variables were favorably associated with the overall inpatient experience with care composite: surgical service (β = 0.55; standard error (SE) = 0.06), less than high school education (β = 0.39; SE = 0.10), and age (β = 0.01; SE = 0.003). Conversely, lower self-reported health status (β = −1.59; SE = 0.07) and female sex (β = −0.45; SE = 0.07) were both unfavorably associated with the dependent variable. At the hospital level, RN staffing and resources directed to patient care were favorably associated with the overall inpatient experience with care composite: (β = 0.11; SE = 0.03) and (β = 0.00001; SE = 0.000004), respectively. Conversely, teaching status had an unfavorable relationship (β = −0.11; SE = 0.06) with the overall inpatient experience with care composite. At the community level, four independent variables were associated with the overall inpatient experience with care composite. Specifically, the lesser the degree of HHI of competition among hospitals, the higher the overall inpatient experience with care (β = 0.99; SE = 0.53). Likewise, the percentage of minority population had an unfavorable relationship with overall inpatient experience with care (β = −0.03; SE = 0.01). Conversely, teaching status and the log of per capita income had favorable relationships with the overall inpatient experience with care composite: (β = 0.04; SE = 0.02) and (β = 0.34; SE = 0.18), respectively.

Table 5.

Multilevel regression results for overall inpatient experience with care composite.

Regression coefficient estimates (standard error estimates in parentheses)^a
	Model 1	Model 2	Model 3
Intercept	21.90 (0.10)	21.80 (0.08)	22.06 (0.09)
Patient-level variables
Female	−0.45 (0.07)	−0.45 (0.07)	−0.45 (0.07)
Age	0.01 (0.003)	0.01 (0.003)	0.01 (0.003)
<High school	0.38 (0.10)	0.38 (0.10)	0.39 (0.10)
Low health status	−1.60 (0.07)	−1.59 (0.07)	−1.59 (0.07)
Surgical service	0.54 (0.06)	0.54 (0.06)	0.55 (0.06)
Hospital-level variables
Teaching status		−0.22 (0.05)	−0.11 (0.06)
RN staffing intensity		0.14 (0.03)	0.11 (0.03)
Resources directed to patient care		0.00001 (0.000004)	0.00001 (0.000004)
Community-level variables
Hospital competition			0.99 (0.53)
Teaching status			0.04 (0.02)
Per-capita income (log)			0.34(0.18)
Percent minority population			−0.03 (0.01)

HMO: health maintenance organization; INP: inpatient; LPN: licensed practical nurse; LVN: licensed vocational nurse; RN: registered nurse.

Independent variables with beta coefficients less than 1.5 times their estimated standard errors were dropped.

Discussion

We found that the majority of the variations in overall inpatient experience with care exist within hospitals, whereas variations between hospitals and between communities represented less than 4% of the total variations. At the patient level, less than high school education, older age, and surgical service were positively associated with overall inpatient experience with care, whereas female sex and lower health status were negatively related to this patient experience measure. Altogether, patient characteristics accounted for only 10.98% of within-hospital variations in overall inpatient experience with care. This finding is consistent with a previous study that found patient-level demographic and socioeconomic variables accounted for as little as 2% to 3% of the total variations in patient assessment of care.²²

At the hospital level, RN staffing and resources directed to patient care were positively associated with overall inpatient experience with care, while teaching status was negatively associated with this dependent variable. The positive association between RN staffing and overall inpatient experience with care is consistent with previous research that links high-skill nurse staffing levels to positive hospital outcomes.²³ Our finding regarding the positive role of resources directed to patient care supports the general notion that increasing labor resources will enable hospitals to improve patient care and improve overall inpatient experience with care. The negative association between teaching status and overall inpatient experience with care found in this study is consistent with a previous study.²⁴ A possible explanation for this finding is that teaching hospitals may have more problems with care coordination and other processes of care than non-teaching hospitals, resulting in lower overall inpatient experience with care.¹⁴ Altogether, hospital characteristics accounted for a sizable amount (34.12%) of the between-hospital variations.

Contrary to economic theory, we found higher competition among hospitals was negatively associated with overall inpatient experience with care. This finding may be explained by the assumption that when hospital administrators are faced with intense competition, they may attempt to cut costs by making efficiency-oriented interventions,²⁵ which could adversely impact patient care and lower patient experience with care. Unlike the negative association of teaching status found at the hospital level, the higher the teaching-hospital score at the community level, the more favorable the overall inpatient experience with care. A potential explanation for this finding is that a larger number of teaching hospitals in a given community may increase the focus on inpatient experience.

Our finding that per-capita income is positively associated with overall inpatient experience with care appears intuitive, as communities with lower income levels tend to receive lower health-care quality and therefore, would likely report less favorable experience with care. While previous studies did not examine the role of minority population at the community level, the negative role of this measure found in our study is supported by previous patient-level studies that demonstrated that minority populations rate their care lower than non-minority groups.²⁶ Overall, competition and socioeconomic factors accounted for over 99% of the variations at the community level.

Despite the associations between overall experience of care and several of the patient, hospital, and community factors considered in this study, the majority of the within-hospital and between-hospital variations remain unexplained. This finding suggests that factors not included in this study (such as those relating to certain patient subgroups and their interactions with health-care providers) may account for these variations. While the literature is scant in this area, a recent study found that substantial heterogeneity exists in inpatient experience ratings across a wide variety of patient subgroups including race/ethnicity, primary language, education, age, and overall health status.²⁷ Therefore, the authors of that study implied that what may be considered the best hospitals for certain patient subgroups may not be the best hospitals for others.

Limitations

Our study makes an original contribution to the literature related to understanding multilevel variations in overall inpatient experience with care. However, it has limitations that need to be noted. First, the number of explanatory variables examined in this study is limited, posing a risk for omitted variable bias. For example, measures of cultural competence were not available in our data, but have recently been found to be associated with patient experience.²⁸ Similarly, factors related to the communication between the patient and the health care provider, especially as it relates to various patient subgroups, were not feasible in this study. Other limitations include the fact that our data were drawn from a single state and date back to the last decade. It is likely that the hospital industry has encountered a number of changes since our data were collected. Therefore, more recent data representing a larger number of hospitals and communities are needed to replicate our findings.

Implications for policy and research

Notwithstanding its limitations, our study demonstrates that overall inpatient experience with care varies within hospitals, between hospitals, and between communities; these variations can be explained by variables at these data levels. Measures of inpatient experience with care can now be derived from the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS).²⁹

Pending replication in future studies, our findings suggest that certain patient, hospital, and community characteristics need to be taken into account when comparisons are drawn across hospitals in relation to overall inpatient experience with care. Alternatively, policymakers can consider implementing changes targeting hospital and market characteristics to improve overall patient experience with care. For example, if intense competition among hospitals proves to result in lower health-care quality, then regulations to reduce competition among hospitals, such as Certificate of Need laws, can be pursued.³⁰ Similarly, if RN nurse-staffing levels prove to be critical for improving patient experience with care, then policies that dictate acceptable staffing levels need to be established and enforced. Notably, future research is needed to understand how patient experience with care varies across patient subgroups based on health-care provider communication and interaction. Such knowledge seems to be a natural prerequisite to patient-centered quality measurement and reporting.

Footnotes

Acknowledgements

The authors thank the California Health Care Foundation for providing them with access to the de-identified PEP-C data without any charges. AA thanks Jennifer Rubio and Dr. Katie Stringer Andersen for reviewing the manuscript.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The AHA data were purchased with funds from the College of Health and Human Development, Pennsylvania State University. The authors received no additional funding.

Ethical approval

Exempt.

Guarantor

AA.

Contributorship

RWM and AA conceptualized and designed the study. AA researched the literature. AA analyzed the data and wrote the first draft of the manuscript. RWM critically revised the manuscript. Both authors reviewed, edited, and approved the final version of the manuscript.

References

Donabedian

. The quality of care: how can it be assessed? JAMA 1988; 260: 1743–1748.

Institute of Medicine (U.S.), Committee on Quality of Health Care in America. Crossing the quality chasm: a new health system for the 21st century, Washington, D.C.: National Academy Press, 2001.

Goldstein

Elliott

Lehrman

. Racial/ethnic differences in patients’ perceptions of inpatient care using the HCAHPS survey. Med Care Res Rev 2010; 67: 74–92.

Lehrman

Elliott

Goldstein

. Characteristics of hospitals demonstrating superior performance in patient experiences and clinical process measures. Med Care Res Rev 2010; 67: 38–55.

Jha

Orav

Zheng

. Patients’ perception of hospital care in the United States. N Engl J Med 2008; 359: 1921–1931.

Sixma

Spreeuwenberg

PMM

van der Pasch

MAA

. Patient satisfaction with the general practitioner: a two-level analysis. Med Care 1998; 36: 212–229.

Gerteis

Edgman-Levitan

Daley

. Through the patient’s eyes: understanding and promoting patient-centered care, San Francisco: Jossey-Bass, 1993.

Brown

Sandoval

Murray

. Comparing patient reports about hospital care across a Canadian-US border. Int J Qual Health Care 2008; 20: 95–104.

California Institute for Health Systems Performance. Results from the patients’ evaluation of performance (PEP-C) survey. Technical Report. Oakland, CA, 2003.

10.

Arbuckle

Wothke

. Amos 4.0 user’s guide, Chicago, IL: SPSS Inc. and SmallWaters Corporation, 1999.

11.

Bentler

. Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives. Struct Equat Model 1998; 6: 1–55.

12.

Cronbach

. Coefficient alpha and the internal structure of tests. Psychometrika 1951; 16: 297–334.

13.

Jaipaul

Rosenthal

. Are older patients more satisfied with hospital care than younger patients? J Gen Intern Med 2003; 18: 23–30.

14.

Hargraves

Wilson

Zaslavsky

. Adjusting for patient characteristics when analyzing reports from patients about hospital care. Med Care 2001; 39: 635–641.

15.

Westaway

Rheeder

Van Zyl

. Interpersonal and organizational dimensions of patient satisfaction: the moderating effects of health status. Int J Qual Health Care 2003; 15: 337–344.

16.

Proenca

Rosko

Zinn

. Community orientation in hospitals: an institutional and resource dependence perspective. Health Serv Res 2000; 35: 1011–1035.

17.

Iwashyna

Chang

Zhang

. The lack of effect of market structure on hospice use. Health Serv Res 2002; 37: 1531–1551.

18.

Zinn

Mor

Castle

. Organizational and environmental factors associated with nursing home participation in managed care. Health Serv Res 1999; 33: 1753–1767.

19.

Rhoades

. The Herfindahl-Hirschman Index. Fed Reserve Bull 1993; 79: 188–189.

20.

Magnussen

Mobley

. The impact of market environment on excess capacity and the cost of an empty hospital bed. Int J Econ Bus 1999; 6: 383–398.

21.

Raudenbush

Rowan

Cheong

. Higher order instructional goals in secondary schools: class, teacher, and school influences. Am Educ Res J 1993; 30: 523–553.

22.

Finkelstein

Singh

Silvers

. Patient and hospital characteristics associated with patient assessments of hospital obstetrical care. Med Care 1998; 36: AS68–AS78.

23.

Harless

Mark

. Nurse staffing and quality of care with direct measurement of inpatient staffing. Med Care 2010; 48: 659–663.

24.

Young

Meterko

Desai

. Patient satisfaction with hospital care: effects of demographic and institutional characteristics. Med Care 2000; 38: 325–334.

25.

Shortell

Hughes

. The effects of regulation, competition, and ownership on mortality rates among hospital inpatients. N Engl J Med 1988; 318: 1100–1107.

26.

Weech-Maldonado

Morales

Elliott

. Race/ethnicity, language, and patients’ assessments of care in Medicaid managed care. Health Serv Res 2003; 38: 789–808.

27.

Elliott

Lehrman

Goldstein

. Do hospitals rank differently on HCAHPS for different patient subgroups? Med Care Res Rev 2010; 67: 56–73.

28.

Weech-Maldonado

Dreachslin

Brown

. Cultural competency assessment tool for hospitals: evaluating hospitals’ adherence to the culturally and linguistically appropriate services standards. Health Care Manag Rev 2012; 37: 54–66.

29.

Goldstein

Farquhar

Crofton

. Measuring hospital care from the patients’ perspective: an overview of the CAHPS Hospital Survey development process. Health Serv Res 2005; 40: 1977–1995.

30.

Starkey

Weech-Maldonado

Mor

. Market competition and quality of care in the nursing home industry. J Health Care Finance 2005; 32: 67–81.