From Fable to Reality at Parkland Hospital: The Impact of Evidence-Based Design Strategies on Patient Safety,Healing,and Satisfaction in an Adult Inpatient Environment

Abstract

Objective:

This research aimed to evaluate the quantitative effects of new hospital design on adult inpatient outcomes.

Background:

Tenets of evidence-based healthcare design, notably single-patient acuity-adaptable and same-handed rooms, decentralized nursing stations, onstage offstage layout, and access to nature were expected to promote patient healing and increase patient satisfaction, while decreasing adverse events.

Methods:

Patient healing was operationalized through length of stay (LOS) and patient safety through three adverse events: falls, hospital-acquired infections (HAI), and medication-related events. Standard patient surveys captured patient satisfaction. Patient records from 2013 through 2017 allowed for equivalent time periods surrounding the move to the new hospital in August 2015. Stratified by hospital division where significant, pre/post comparisons utilized proportional hazards or logistic regression models as appropriate; interrupted time series analyses afforded longitudinal interpretations.

Results:

Observed higher postmove LOS was due to previously increasing trends, not increases after the move. In surgical and trauma units, a constant increase in falls was unaffected by the move. Medication events decreased consistently over time; medication events with harm dropped significantly after the move. No change in HAI was found. Significant improvement on most relevant patient satisfaction items occurred after the move. Call button response decreased immediately after the move but subsequently improved.

Conclusion:

Results did not clearly indicate a net change in adult inpatient outcomes of healing and safety due to the hospital design. There was evidence that the new hospital improved patient satisfaction outcomes related to the environment, including comfort, noise, temperature, and aesthetics.

Keywords

inpatient healthcare design length of stay adverse events patient satisfaction single-patient rooms acuity-adaptable patient rooms same-handed patient rooms decentralized nursing stations onstage offstage unit layout

The new Parkland Hospital (NPH) in Dallas, TX, USA, was designed to support staff workflow and optimize patient healing. As a safety net hospital that serves a largely underinsured and uninsured population, a primary goal of Parkland Health and Hospital System (Parkland) in planning the new facility was that “The new hospital will be a safe, welcoming, patient-centered healing environment that serves as a sustainable community resource for Dallas County” (Harper et al., 2011). A Center for Health Design Pebble Project during the initial stages of conception (Berry et al., 2004; Joseph & Hamilton, 2008), the design of NPH was thoughtfully planned to consider the best current evidence and further contribute to the field through rigorous evaluation of the new facility’s impact on outcomes for patients and staff. The salient design strategies planned for NPH—acuity-adaptable and same-handed single-patient rooms, a decentralized nurse station model, an onstage offstage inpatient unit layout, and access to nature—are often promoted as beneficial to patients and are similar to the frequently cited recommendations for new hospital designs outlined in the Fable Hospital 2.0 business case. (Sadler et al., 2011). Research evidence around these design strategies, however, varies in strength and applicability.

The salient design strategies planned for NPH are often promoted as beneficial to patients and are similar to the frequently cited recommendations for new hospital designs outlined in the Fable Hospital 2.0 business case.

Two recent reviews summarizing the literature since the 2006 Facility Guidelines Institute requirement for single-patient rooms in the United States (The Facility Guidelines Institute & the American Institute of Architects Academy of Architecture for Health, 2006) reported evidence of improved patient satisfaction in single-patient rooms (Taylor et al., 2018; Voigt et al., 2018). Taylor et al. also found evidence of reduced infections and noise. Voigt et al. concluded strong evidence of increased falls in single patient compared to multioccupancy rooms; Taylor et al. determined the impact of single-patient rooms on falls as inconclusive. Within single-patient rooms, acuity-adaptable and same-handed configurations have been suggested to reduce medical errors, falls, and length of stay (LOS), partly due to a decrease in patient transfers in the former and staff cognitive load in the latter (Bonuel & Cesario, 2013; Stichler & McCullough, 2012).

At the unit level, decentralized nurse station models have been the focus of several studies, summarized in two recent systematic reviews (Fay et al., 2019; Jimenez et al., 2019). Fay et al. concluded sufficient evidence to support decentralized nursing for decreased falls and improved patient satisfaction, but Jimenez et al. revealed that no study has reported a statistically significant effect on patient outcomes. Onstage offstage designs that separate staff-only areas (often in the central core of a racetrack floor plate, as at NPH) from patient and visitor areas have recently become popular with the intent to reduce noise in patient rooms and improve patient satisfaction (Pati et al., 2015), but no research has been conducted to measure the effects of this approach on patients in inpatient units.

Increasing connection to natural elements in healthcare settings has received much attention in the literature, stemming from Ulrich’s (1984) seminal work that showed reduced LOS for patients with a view of trees with foliage compared to those with a view to a brick wall. Yet in a recent literature review on patient satisfaction, no other studies were cited in connection to the effects of nature and views on health outcomes (MacAllister et al., 2016). Another review of studies relating nature and views to health only cited the Ulrich study as an example of views affecting a patient population; although several other studies linked improved patient healing to daylight exposure often through a mediating effect of depression or mood (Beute & de Kort, 2014).

The aim of this research was to evaluate the quantitative effects of the holistic NPH design, as compared to the previous hospital environment, on the adult inpatient outcomes of safety, healing, and satisfaction. These outcomes are of particular importance to health systems as judgments of the quality of care and proxies measuring the success of doctors and hospitals. Patient healing is often expressed through LOS, which serves as a common indicator for timeliness and efficiency of care (Brasel et al., 2007; Thomas et al., 1997). Measures of adverse events—falls, medication-related events, and hospital-acquired infections (HAI)—reflect a level of safety in healthcare settings (Chen et al., 2005; De Vries et al., 2008; Institute of Medicine, 2000; Morgan et al., 1985; World Health Organization, 2002). Patient satisfaction, especially as measured by the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey, quantifies patient-centeredness, timeliness, and other quality indicators and is tied to healthcare cost reimbursement (Jha et al., 2008; Zusman, 2012). As the overall goal of NPH is to achieve excellence in healthcare quality, it was hypothesized that implemented design strategies would promote patient healing and increase patient satisfaction, while decreasing the occurrence of negative safety incidents. More detailed study aims can be found in Brittin et al. (2019). Specific goals of the design were previously outlined in an article in Healthcare Design Magazine (Harper et al. 2011).

Method

Study Design

Changes in patient outcomes between the old and new hospitals were assessed through pre/post and longitudinal comparison of adult inpatient encounter data from electronic medical records and other hospital data sources. Patient healing time was operationalized as LOS and patient safety through the occurrence of three types of adverse events: falls, HAI, and medication-related events. Patient satisfaction was captured through standard patient surveys routinely conducted at Parkland.

Setting

The move into NPH occurred on August 20, 2015. The typical adult inpatient unit in the new hospital consists of 36 single-patient conversion-ready (without gas or electrical for complete acuity-adaptability) and same-handed 310 ft² rooms (Figure 1). Outboard toilets have large angled doorways and no-curb showers. A large in-room family area accommodates overnight visitors and includes a substantial window with exterior views. Charting alcoves outside each pair of patient rooms have windows into adjacent rooms for unobtrusive patient observation. Patient rooms line the exterior length of the units in a racetrack configuration and departmental gross square feet (DGSF) per unit total 31,112 with building gross square feet (BGSF) of 35,945. In the core a dedicated 40-ft wide off-stage area gives access to clean and soiled materials and houses supply and medication rooms, staff lockers and break areas, restrooms, and dedicated elevators for staff and materials. Medicine units in the new hospital each have four airborne infection isolation rooms (AIIR) and two anterooms, with additional rooms in the unit that specializes in infectious diseases (20 AIIR and four anterooms) and the medical intensive care unit (12 AIIR). Three rooms on each unit have installed patient lifts. A high-efficiency particulate air filtration system is monitored from a central location, and local indicators alert staff to elevated noise levels. Slip-resistant flooring and antimicrobial finishes were used throughout the hospital, and a large public garden is lon the ground floor.

Units varied considerably in the old Parkland hospital but typically consisted of 20–28 patient beds with about 75% in shared rooms (Figure 1). Mirrored-layout 218 ft² patient rooms had a small in-board toilet but no in-room shower and were configured in a double-loaded corridor linear unit design with one centrally located multiseat nurse station. Unit size also varied but averaged about 7,650 DGSF and 9,500 BGSF. Hallways were narrow with little storage for equipment. Space was unavailable for visitors in patient rooms except for possibly a single chair. Additional explanation of the old and new unit designs, along with sample floor plans can be found in Brittin et al. (2019).

Figure 1.

Old and new patient room sizes and configurations.

Operationally, Parkland consistently focused on providing quality patient care and preventing adverse events throughout the study period, with occasional campaigns to improve results, such as a system-wide focus on two patient identifiers to reduce errors from December 2016 through September 2019 and an ongoing Glycemic Committee started in May 2014, of which one goal was to reduce medication errors among diabetic patients. With the move, medication label scanners and printers were available in patient rooms, although in varying functioning condition. Engineering controls, such as the air filtration system, are now monitored centrally rather than locally on the units as in the old hospital. Due to a lack of space, overnight visitors were prohibited in the old hospital but are allowed and have accommodations in the new patient rooms.

Data Collection

Medical and safety records, for adult inpatient encounters with admission dates between January 1, 2013, and December 31, 2017, were obtained from Parkland. Initial analyses included all data, and sensitivity analyses were performed excluding admissions 90 days both before and after the move to control for any inconsistent effects due to the preparation for and recovery from the move. Responses to inpatient surveys for patients discharged between January 1, 2013, and December 31, 2017, were included, with the exception of responses collected during the month immediately following the move. This study was submitted to Western Institutional Review Board and determined exempt from review as all data existed at the time of data collection.

Statistical Analysis

Descriptive statistics

Differences in demographic and outcome variables between the pre- and postperiods were assessed using univariate tests: the Wilcoxon rank-sum test for continuous variables and χ² analysis for categorical variables. An index of socioeconomic status (SES) links patient zip codes to U.S. Census data: percent unemployment, percent below U.S. poverty line, median income, property values, education level, and household crowding (Bonito et al., 2008). Given the relatively low SES within our sample (55.1% in the lowest quartile of national block group data) and interest in differentiating between SES groups within the studied patient population, we created our own SES quartiles following the method outlined by Bonito et al.

Length of stay

Comparisons between the pre and postperiods were performed using a proportional hazards regression model, which is appropriate for skewed and nonnormally distributed data. Continuous variables were assessed for functional form and if continuous form was inappropriate, the variable was categorized. If LOS was negative or equal to zero, it was counted as a data entry error and set to missing. Deaths and transfers to other hospitals were included in the model as competing risks. In addition to time period (pre, post), confounding variables included age, race/ethnicity, gender, FY2015 Medicare Severity-Diagnosis Related Group (MS-DRG) weights as a measure of acuity (U.S. Department of Health and Human Services: Centers for Medicare and Medicaid Services, 2014; Lu et al., 2015), SES index quartile, and the hospital division where a patient spent the longest time—Women & Infants Specialty Health (WISH), surgical & trauma (S&T), or Medicine Services (MED). As typical LOS varies for different types of patients, an interaction between time period and division was explored as well.

An interrupted time series model—using median LOS aggregated into one month periods—examined trends over time. Controlling for patient acuity, the model assumed a linear trend and allowed for changes in the intercept and slope at the time of the move. Separate models were used for interaction groups as appropriate.

Adverse events

Three types of adverse events were each analyzed for changes from pre to post using logistic regression and interrupted time series models: falls, medication-related events, and HAI. A zero-inflated Poisson model was considered to assess differences in the count of events per encounter over time. Due to the low number of events and few encounters with multiple events, the Poisson models fit poorly and an indicator of at least one event within an encounter was used instead. Medication events were reported at Parkland with a harm score between 1 and 9. Medication event analyses were performed including all medication event harm levels and for only medication events resulting in some degree of harm (i.e., harm score of 4 or higher). HAI analyses were performed for two organism types combined and separately: Clostridium difficile (C. diff) and Methicillin-resistant staphylococcus aureus (MRSA).

The logistic regression model determined differences in the odds of an event during an encounter adjusted for LOS. Confounding variables included age, MS-DRG weights, SES index, race/ethnicity, gender, and division where the patient stayed the longest. The analysis of falls also included the initial Morse fall risk score (the standard risk score used at Parkland) for each encounter as a confounder (Morse et al., 1989).

The time series model analyzed the rate of events per patient day. Days were counted as each date the patient was in the hospital excluding the day of discharge. The division for all encounter days was set equal to where the patient stayed the longest. The division for each event was the first patient location on the day of the event. The model assumed a linear change over time and explored changes in intercept and slope at the time of the move.

Patient satisfaction

Twelve HCAHPS (www.hcahpsonline.org) and Press Ganey (PG; www.pressganey.com) inpatient patient experience questions were selected for analysis based on their plausible connection to the built environment and the design changes implemented in the new inpatient units (Table 1). Monthly aggregated patient survey responses were analyzed using an interrupted time series model that examined differences in the trend of the percentage of positive responses for HCAHPS questions and mean scores, from a scale of 0 to 100, for PG questions.

Table 1.

Patient Satisfaction Questions.

Item Number	Question Wording	Positive Response (HCAHPS)
HCAHPS 1	Using any number from 0 to 10, where 0 is the worst hospital possible and 10 is the best hospital possible, what number would you use to rate this hospital during your stay?	9 or 10
HCAHPS 2	Would you recommend this hospital to your friends and family?	Definitely yes
HCAHPS 3	During this hospital stay, how often was the area around your room quiet at night?	Always
HCAHPS 4	During this hospital stay, after you pressed the call button, how often did you get help as soon as you wanted it?	Always
PG 1	Overall rating of care given at hospital
PG 2	Likelihood of your recommending this hospital to others
PG 3	Noise level in and around room
PG 4	Promptness in responding to the call button
PG 5	Pleasantness of room decor
PG 6	Room temperature
PG 7	Accommodations and comfort for visitors
PG 8	How well staff worked together to care for you

Note. HCAHPS = Hospital Consumer Assessment of Healthcare Providers and Systems; PG = Press Ganey.

All regression models included sensitivity analysis to assess the impact of missing data on the model using multiple imputations through the fully conditional specification method (Lee & Carlin, 2010). Fit of all models was assessed, and periodic adjustments in time series models assessed the possibility of seasonal effects. Analysis was performed using SAS software, Version 9.4, of the SAS System for Windows (SAS Institute Inc., Cary, NC, USA).

Results

Descriptive Medical Records Data

Overall, 192,066 encounters—100,186 premove and 91,880 postmove—were included in the study. All demographic factors indicated a statistically significant change from the pre to postperiod (Table 2); however, most changes were small and may not be clinically meaningful.

Table 2.

Patient Descriptive Statistics.

Continuous Variables	Premove (January 1, 2013–August 20, 2015)		Postmove (August 21, 2015–December 31, 2017)		Wilcoxon p
Continuous Variables	N	Median (Q1, Q3)	N	Median (Q1, Q3)	Wilcoxon p
Patient age	100,186	42.0 (30.0, 56.0)	91,880	41.0 (29.0, 57.0)	<.001
MS-DRG weights	100,109	1.00 (0.74, 1.51)	91,810	1.02 (0.74, 1.54)	<.001
Length of stay (days)	100,160	3.30 (2.27, 5.83)	91,801	3.56 (2.42, 6.33)	<.001
WISH	33,819	2.86 (2.27, 3.90)	33,676	2.93 (2.31, 3.84)	<.001
S&T	24,788	3.70 (2.12, 7.19)	24,576	4.24 (2.43, 7.92)	<.001
MED	41,394	4.02 (2.40, 7.01)	33,379	4.75 (2.94, 8.19)	<.001
Categorical Variables	Frequency (%)		Frequency (%)		χ² p
Division (longest)					<.001
WISH	33,819 (33.8)		33,676 (36.8)
S&T	24,794 (24.8)		24,584 (26.8)
MED	41,395 (41.4)		33,387 (36.4)
Gender					<.001
Male	38,166 (38.1)		34,182 (37.2)
Female	62,020 (61.9)		57,698 (62.8)
Ethnicity					<.001
Hispanic	52,037 (52.3)		48,849 (53.8)
Non-Hispanic	47,497 (47.7)		41,975 (46.2)
Race					<.001
White	69,537 (70.0)		62,855 (69.4)
Black	26,820 (27.0)		25,021 (27.6)
Asian	2,511 (2.5)		2,086 (2.3)
Other	464 (0.5)		600 (0.7)
SES Index					.117
Quartile 1 (0 to <44.41)	25,106 (25.33)		23,420 (25.73)
Quartile 2 (44.41 to <47.02)	25,166 (25.39)		23,004 (25.27)
Quartile 3 (47.02 to <51.072)	24,181 (24.40)		22,286 (24.48)
Quartile 4 (51.072 to 100)	24,662 (24.88)		22,319 (24.52)
Discharge					.002
Home/AMA/custody	93,243 (93.2)		85,148 (92.8)
Heath facility/hospice	5,541 (5.5)		5,262 (5.7)
Death	1,321 (1.3)		1,358 (1.5)
Insurance type					<.001
Commercial	5,762 (5.8)		6,538 (7.1)
Government	59,782 (59.7)		55,699 (60.6)
Charity/self-pay	34,168 (34.1)		20,970 (31.6)
Other	474 (0.5)		573 (0.6)
At least one fall during stay					.017
Yes	1,026 (1.0)		1,044 (1.1)
No	99,160 (99.0)		90,836 (98.9)
At least one medication event during stay					<.001
Yes	939 (0.9)		684 (0.7)
No	99,247 (99.1)		91,196 (99.3)
Medication event harm score (n = 1,731)					.115
1 = Unsafe condition	37 (3.7)		29 (4.0)
2 = Near miss	43 (4.3)		18 (2.5)
3 = No harm evident, physical, or otherwise	600 (60.2)		436 (59.4)
4 = Emotional distress or inconvenience	201 (20.2)		180 (24.5)
5 = Additional treatment	102 (10.2)		62 (8.4)
6 = Temporary harm	13 (1.3)		8 (1.1)
7 = Permanent harm	1 (0.1)		0 (0.0)
8 = Severe permanent harm	0 (0.0)		0 (0.0)
9 = Death	0 (0.0)		0 (0.0)
Missing harm score	0 (0.0)		1 (0.1)
At least one HAI during stay					.822
Yes	411 (0.41)		383 (0.42)
No	99,775 (99.59)		91,497 (99.58)
HAI organism type (n = 834)					.015
C. diff	390 (91.33)		350 (95.99)
MRSA	37 (8.66)		57 (14.00)

Note. HAI = hospital-acquired infections; SES = socioeconomic status; MS-DRG = Medicare Severity-Diagnosis Related Group; WISH = Women & Infant Speciality Health; S&T = Surgical & Trauma; MED = Medicine Services; MRSA = Methicillin-resistant staphylococcus aureus.

Length of Stay

There was a significant interaction in LOS between time and division. (The results of the regression model by division are presented in Table 3). A hazards ratio value less than one indicates a lower risk of discharge (i.e., a longer LOS) in the postperiod compared to the preperiod. There was significantly higher LOS in the postperiod in all divisions, but more notably for MED and S&T than for WISH.

Table 3.

Proportional Hazards Regression of LOS.

Unit	Hazard Ratio	95% CI	p Value
WISH (post vs. pre)	.98	[.96, .99]	.02
S&T (post vs. pre)	.87	[.86, .89]	<.001
MED (post vs. pre)	.88	[.87, .90]	<.001

Note. n = 187,061. WISH = Women & Infant Speciality Health; S&T = Surgical & Trauma; MED = Medicine Services.

A linear time series model fit the data well; no seasonal effect was found. Models were fitted separately for each division, given a significant interaction. Acuity generally increased over time and was included as a covariate in the model for MED and S&T where it was a significant predictor of LOS. For MED, there was an upward trend in LOS prior to the move (p < .001), an increase at the move (from 4.5 days to 4.9 days, p = .005), and a downward trend after the move (p < .001). For S&T, there was an upward trend prior to the move (p < .001) that changed significantly after the move (p < .001) to a relatively constant rate (4.2–4.3 days). For WISH, there was a slight upward trend prior to the move (p < .001), which changed significantly after the move (p = .005) to a constant rate (2.9 days). In Figure 2, dashed lines represent the model estimates, with S&T and MED adjusted for median MS-DRG weight.

Figure 2.

Controlled time series analysis of median monthly length of stay by division.

Falls

A total of 2,324 falls occurred in 2,070 encounters. A significant interaction was found between time and division. The results of the regression model are presented in Table 4. For S&T, there were significantly higher odds of an encounter with a fall in the postperiod compared to the preperiod with no significant difference in the odds of falls for MED and WISH.

Table 4.

Logistic Regression of Falls by Division.

Division	Odds Ratio	95% CI	p
WISH (post vs. pre)	0.84	[0.61, 1.14]	.267
S&T (post vs. pre)	1.29	[1.10, 1.52]	.002
MED (post vs. pre)	0.94	[0.83, 1.06]	.308

Note. n = 184,337. WISH = Women & Infant Speciality Health; S&T = Surgical & Trauma; MED = Medicine Services.

The time series analysis was performed separately for each division given the significant interaction. For WISH, falls were relatively rare. The model indicated a slight upward trend in the preperiod and a slight downward trend in the postperiod (change in slope: p = .02), but no change in fall rate (intercept) at the time of the move. Falls within the MED division had a significant change in slope between time periods (p = .01) with a slight upward trend in falls before the move and a slight downward trend after the move. For S&T, an increasing rate of falls in the preperiod (p = .02) then remained consistent after the move (p = .98). In Figure 3, dots represent the median monthly fall rate, and dashed lines are the model estimates for each division.

Figure 3.

Controlled time series analysis of monthly fall rate by division.

Medication-Related Events

A total of 1,731 medication events occurred in 1,623 encounters. A total of 567 medication events with at least some evidence of harm occurred in 552 encounters. Table 2 lists the frequency of medication events by harm score in both time periods.

The logistic regression model found significantly lower odds of a medication event during an encounter in the postperiod compared to the preperiod (odds ratio [OR] = 0.71, 95% confidence interval [CI] = [0.64, 0.79], p < .0001). There was a significant interaction between time period and division (p = .03). All ORs within divisions were in the same direction as the overall model with only differences in the magnitude of effect. WISH had the biggest difference in medication events followed by MED and then by S&T; division results are provided in Table 5. Similar results were found for events with at least some harm, with lower odds of an event in the postperiod compared to the preperiod (OR = 0.77, 95% CI [0.65, 0.92], p = .004), but the interaction between division and time was not significant (p = .34) and was removed from the model.

Table 5.

Logistic Regression of Medication Events by Division.

Division	Odds Ratio	95% CI	p
WISH (post vs. pre)	.52	[.39, .69]	<.001
S&T (post vs. pre)	.82	[.69, .99]	.03
MED (post vs. pre)	.71	[.62, .82]	<.001

Note. n = 187,099. WISH = Women & Infant Speciality Health; S&T = Surgical & Trauma; MED = Medicine Services.

In the time series analysis of all reported medication-related events, there was a slight downward trend of medication events over time (p < .001) which was consistent after the move. In Figure 4, dots represent the median monthly medication event rate, and dashed lines are the model estimates. For medication events that caused harm, the model estimated a constant rate before the move and a constant but significantly lower rate after the move (p = .02; Figure 5).

Figure 4.

Controlled time series analysis of monthly medication event rate.

Figure 5.

Controlled time series analysis of monthly medication event rate for events causing harm.

Hospital-Acquired Infections

A total of 834 HAI occurred in 794 encounters. By organism type, 740 C. diff HAI occurred in 714 encounters and 94 MRSA HAI in 84 encounters. The logistic regression model found no difference in the odds of an HAI during a hospital encounter in the postperiod compared to the preperiod (OR = 0.92, 95% CI [0.80, 1.07], p = .27). There was not a significant interaction between time period and division and no significant differences in the odds of an HAI specific to organism type: C. diff (OR = 0.88, 95% CI [0.76, 1.03], p = .11) or MRSA (OR = 1.35, 95% CI [0.85, 2.15], p = .20).

The time series model showed a significant difference in the slope between the pre and postperiods (p = .03) but not the intercept, and no seasonal effect was found. The rate of HAI was slightly increasing before the move (p = .07) and slightly decreasing after the move (p = .15). Neither slope was different from zero at a p = .05 level, but the difference between the two time period slopes was statistically significant. In Figure 6, dots represent the median monthly HAI event rate, and dashed lines are the model estimates. A majority of the HAI were C. diff (88.7%), and the time series analysis for only these cases was very similar as for all HAI. Rates of MRSA HAI were extremely low, in only 0.04% of encounters and with no events in 31.7% of the months measured. Thus, a time series analysis of MRSA HAI was not attempted.

Figure 6.

Controlled time series analysis of monthly hospital-acquired infection rate.

Patient Satisfaction

Model results by patient satisfaction question are shown in Table 6. Model estimates include the preperiod slope when the rate or change in slope was significant, the estimated change in intercept at the time of the move, and the change in slope after the move if significant. Figures 7 and 8 display the monthly percentage of positive responses for HCAHPS questions and mean response scores for PG questions, respectively, with dashed lines of the model estimates.

Figure 7.

Hospital Consumer Assessment of Healthcare Providers and Systems questions: Controlled time series analysis of percentage of positive responses.

Figure 8.

Press Ganey questions: Controlled time series analysis of mean scores.

Table 6.

Time Series Parameter Estimates for Inpatient Survey Questions.

Question	Initial Slope		Change at Move		Slope Change
Question	Estimate (SE)	p	Estimate (SE)	p	Estimate (SE)	p
HCAHPS 1: Hospital rating	—	—	9.70 (0.74)	<.001	—	—
HCAHPS 2: Recommend hospital	.09 (.04)	.02	6.45 (1.31)	<.001	—	—
HCAHPS 3: Quiet at night	—	—	12.15 (1.01)	<.001	—	—
HCAHPS 4: Call button response	−.09 (.06)	.17	−6.11 (1.75)	.001	.37 (.11)	<.001
PG 1: Overall rating of care given	—	—	3.04 (0.32)	<.001	—	—
PG 2: Likelihood to recommend	—	—	4.09 (0.31)	<.001	—	—
PG 3: Noise level in and around room	—	—	9.52 (0.50)	<.001	—	—
PG 4: Call button response promptness	−.01 (.03)	.80	−1.34 (0.70)	.06	.16 (.04)	<.001
PG 5: Pleasantness of room decor	−.05 (.02)	.04	15.41 (0.81)	<.001	—	—
PG 6: Room temperature	—	—	9.38 (0.44)	<.001	—	—
PG 7: Visitor accommodation/comfort	—	—	13.90 (0.32)	<.001	—	—
PG 8: Staff worked together to care	−.01 (.02)	.45	1.83 (0.50)	<.001	.07 (.03)	.03

Note. HCAHPS = Hospital Consumer Assessment of Healthcare Providers and Systems; PG = Press Ganey.

For seven questions, there were relatively constant responses before and after the move with a significant positive shift (p < .001 for all) at the time of the move (HCAHPS 1 and 3 and PG 1, 2, 3, 6, and 7). PG 8 followed a similar pattern, but with a small yet significant increase (p < .001) at the time of the move followed by a marginally significant upward trend in the postperiod (p = .03). Two questions had a slightly positive or negative slope in both time periods (HCAHPS 2: p = .02, PG 5: p = .04), but a significant increase in positive responses at the time of the move (p < .001 for both). Both questions about call button response (HCAHPS 4 and PG 4) had a nonsignificant slight downward trend prior to the move (HCAHPS 4: p = .17, PG 4: p = .80), and a significant or marginally significant drop in positive responses at the time of the move (HCAHPS 2: p = .001, PG 5: p = .06), followed by an upward trend after the move that was significantly different from the premove trend (p < .001 for both). Results of all sensitivity analyses were nearly identical to the original models.

Discussion

In 2011, the well-known Fable Hospital 2.0 business case outlined design recommendations for long-term cost benefits (Sadler et al., 2011). Many of the economic gains were explained through hypothesized improvements in patient outcomes in healing, safety, and satisfaction. In 2020, one author of the Fable Hospital papers called for outcomes results to support design hypotheses (Hamilton, 2020). NPH was designed with many of the design principles recommended in the Fable 2.0 case, and this study used patient records data to statistically assess the connection of the new hospital design to hypothesized benefits of reduced LOS, fewer adverse events, and improved patient satisfaction.

Employing an interrupted time series analysis, in addition to more commonly used pre/postregression models, revealed subtle changes in patient outcomes that allow for potential causal interpretations. A comparison between cross-sectional and longitudinal analysis methods revealed consistent results, but sometimes with very different interpretations. The time series method allows us to expand upon a comparison between two points in time in order to attribute changes in outcomes to the change in design at the specific time of the move and also note the difference between an immediate effect and a longer term trend.

Employing an interrupted time series analysis, in addition to more commonly used pre/postregression models, revealed subtle changes in patient outcomes that allow for potential causal interpretations

Results showed some indication of higher LOS after the move, especially in S&T and MED. MED LOS decreased postmove, but not below premove levels during the dates of the study. Despite previous studies connecting acuity-adaptable rooms and access to nature with reduced LOS in specific patient conditions (Bonuel & Cesario, 2013; Ulrich, 1984), there is no evidence the design using these strategies at NPH decreased LOS across a wider population. We saw similar trends in fall rates in WISH and MED of increasing rates prior to the move followed by decreasing trends after the move. As of this study, the rates remained above premove levels and it is not clear that the inclusion of single-patient rooms or a decentralized nursing layout caused fewer incidences of falls overall. Fall rates in S&T were higher postmove, but the trend had begun before the move with no indication that it was impacted by the new hospital design. There is some indication from the time series analysis that the trend in HAI changed after the move and may be declining slightly, but there is no evidence that the number of cases has been reduced beyond premove levels because of the hypothesized influence of single-patient rooms. There was some evidence of a lower occurrence of medication events in the new hospital, especially for events causing harm. The trend including events at all harm levels, however, had been decreasing before the move, so could be attributed to other factors such as organizational efforts and programs to prevent such events, possibly in addition to same-handed rooms and updated technology in the new hospital.

In contrast to patient healing and safety outcomes, there was a noticeable and significant improvement in nearly all patient experience results. The immediate shift in question response corresponding to the month of the move makes it likely that the design of the new hospital had a substantial positive impact on patient satisfaction. These results diverge somewhat from a previous study of satisfaction in a renovated hospital where not many differences were found (Siddiqui et al., 2015). While the holistic design and newness of the hospital most certainly impacted the satisfaction scores, the spacious single-patient rooms with dedicated family space and outdoor views are thought to have a great impact on the change in experience. The initial drop in satisfaction to call button response time may have been due to the larger physical size of the units, creating longer distances for nurses to travel that required adaptations to previous nursing practices, leading to subsequent improvement.

The immediate shift in question response corresponding to the month of the move makes it likely that the design of the new hospital had a substantial positive impact on patient satisfaction.

Limitations

Due to the observational nature of the study, the new hospital design must be considered holistically, without separating the effects of each design strategy. A 2009 IHI Innovation Series paper expressed that improvements will be seen only through an integrated set of design changes implemented together rather than as single interventions (Sadler et al., 2009). This is indeed a common challenge in design research, where many changes are implemented simultaneously in a new building or renovation, as was the case in this study.

The study of a single site with no control did not allow for comparison of trends across time between other hospitals that either implemented similar changes or remained constant throughout the period of study. Institutional efforts to improve outcomes were unaccounted for in the analyses. In addition, errors and missing data are inherent in medical records data. Due to the large sample in this study and the consistent results using imputation methods, however, we believe the impact of missing data on results was minimal. While MS-DRG weights were used as a measure of acuity across all patient diagnoses, there may be aspects of acuity not captured in that value.

Conclusions

The hospital design change did not have significant effects on the adult inpatient outcomes of healing and safety. Most results varied over time, but the change in trends at the time of the move did not clearly indicate either a positive or negative net change. By contrast, there was strong evidence that the new hospital improved patient satisfaction related to overall experience, and outcomes closely related to the environment, such as comfort, noise, temperature, and aesthetics. Patient healing and safety are paramount in providing quality care, but due to the complex nature of healthcare delivery may be more difficult to directly influence through hospital design alone. This study provides strong evidence, however, that patient satisfaction can be significantly improved through thoughtful and intentional design.

Recommendations for Future Results

For future studies, more sophisticated longitudinal analysis methods, such as interrupted time series, should be routinely employed to assess and understand the effects of a design change. This study should be replicated in hospitals that transition to similar new facility designs, along with control settings, to compare results and trends over time. Whenever possible, an experimental research design should be utilized to isolate the impact of individual design strategies.

Implications for Practice

Exercise caution in promising substantial changes in patient healing and safety, and resulting cost benefits, attributed to standard hospital design recommendations.

Recognize that patient satisfaction may be more directly and immediately influenced by a change in design than health outcomes.

Restraint should be applied in projecting specific patient results of recommended design changes as most patient outcomes have multiple determinants cannot be strongly predicted using current scientific evidence.

Plan to assess the effects of the design change on patient outcomes of interest.

Use appropriate longitudinal analysis methods to understand trends in order to attribute differences to a specific point in time.

Footnotes

Acknowledgments

The authors are grateful for the invaluable input and support of the following individuals at Parkland Health & Hospital System: Susan Partridge, RN, BSN, MBA, CCRC; Leticia A. Blea, MPA; Karen Watts, MSN, RN, NEA-BC; David Lopez, FACHE; and Frederick P. Cerise, MD, MPH. In addition, Jackline G. Opollo, PhD, RN, MSN, MPH; Debra Gilbert, MBA, MIS; Jerry Nickerson, PE, CxA, CMVP, FSR; and Jessica Martinez, CMRP, no longer employed at Parkland, previously played key roles in supporting the evaluation research efforts. We also acknowledge the following individuals for their commitment to and participation in the New Parkland Hospital Research Coalition: Kathy Okland, RN, MPH, EDAC, Independent Consultant; Juliet L. Rogers, PhD, MPH, Blue Cottage of CannonDesign; Tina Larsen, AIA, EDAC, LEED AP, Corgan; Doug Bazuin, MS, EDAC, Community Action House; Leslie Echols, IIDA, EDAC, Nora Systems; Cyndi McCullough, RN, MSN, EDAC, HDR; Lori McGilberry, Kilter LLC; Robert Agosta, Mitchell Design Inc.; and Gena English, CHID, EDAC, RAS, University of Texas Southwestern.

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Renae Rich, Francesqca Jimenez, Susan Puumala, and Jeri Brittin were employed by HDR, an architecture and engineering firm, during the research. HDR provided salary support for the authors but did not have any additional role in the design, decision to publish, or preparation of this manuscript. Lonnie Roy, Sheila DePaola, and Kathy Harper declare no conflict of interest.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by HDR, as part of a larger collaborative study also supported by firms participating in the New Parkland Hospital Research Coalition, with in-kind support from Parkland Health & Hospital System.

ORCID iDs

Renae K. Rich, MS, EDAC

Jeri Brittin, PhD, Assoc. AIA, Allied ASID, EDAC

References

Berry

L. L.

Parker

Coile

R. C.

O’Neill

D. D.

Sadler

B. L.

(2004). The business case for better buildings. Frontiers of Health Services Management, 21(1), 3–24. http://www.ncbi.nlm.nih.gov/pubmed/15469120

Beute

de Kort

Y. A. W.

(2014). Salutogenic effects of the environment: Review of health protective effects of nature and daylight. Applied Psychology: Health and Well-Being, 6(1), 67–95. https://doi.org/10.1111/aphw.12019

Bonito

A. J.

Bann

Eicheldinger

Carpenter

(2008). Creation of new race-ethnicity codes and socioeconomic status (SES) indicators for Medicare beneficiaries (Final report, Sub-Task 2). RTI International.

Bonuel

Cesario

(2013). Review of the literature: Acuity-adaptable patient room. Critical Care Nursing Quarterly, 36(2), 251–273. https://doi.org/10.1097/CNQ.0b013e31828d63c7

Brasel

K. J.

Lim

H. J.

Nirula

Weigelt

J. A.

(2007). Length of stay: An appropriate quality measure? Archives of Surgery, 142(5), 461–466. http://www.ncbi.nlm.nih.gov/pubmed/3557348

Brittin

Okland

Rogers

J. L.

Rich

R. K.

Bazuin

Harper

Roy

(2019). From competition to collaboration: How a multi-firm research coalition is realizing rigorous facility evaluation at parkland hospital. Health Environments Research & Design Journal. https://doi.org/10.1177/1937586719879717

Chen

Y. Y.

Chou

Y.-C.

Chou

(2005). Impact of nosocomial infection on cost of illness and length of stay in intensive care units. Infection Control & Hospital Epidemiology, 26(3), 281–287.

De Vries

E. N.

Ramrattan

M. A.

Smorenburg

S. M.

Gouma

D. J.

Boermeester

M. A.

(2008). The incidence and nature of in-hospital adverse events: A systematic review. Quality and Safety in Health Care, 17(3), 216–223. https://doi.org/10.1136/qshc.2007.023622

The Facility Guidelines Institute & the American Institute of Architects Academy of Architecture for Health. (2006). Guidelines for design and construction of healthcare facilities. American Institute of Architects.

10.

Fay

Cai

Real

(2019). A systematic literature review of empirical studies on decentralized nursing stations. Health Environments Research & Design Journal, 12(1), 44–68. https://doi.org/10.1177/1937586718805222

11.

Hamilton

D. K.

(2020). Return to fable hospital: The frustrating search for elusive design quality. Health Environments Research & Design Journal, 13(1), 24–29. https://doi.org/10.1177/1937586719895338

12.

Harper

Adams

Thomas

(2011). Applying evidence-based design at the new Parkland Hospital. Healthcare Design, 11(7), 18–23.

13.

Institute of Medicine. (2000). To err is human. Building a safer health system (Vol. 6). In Kohn

L. T.

Corrigan

J. M.

Donaldson

M. S.

(Eds.). The National Academies Press. https://doi.org/10.17226/9728

14.

Jha

A. K.

Orav

E. J.

Zheng

Epstein

A. M.

(2008). Patients’ perception of hospital care in the United States. New England Journal of Medicine, 359(18), 1921–1931. https://doi.org/10.1056/NEJMsa0804116

15.

Jimenez

F. E.

Puumala

S. E.

Apple

Bunker-Hellmich

L. A.

Rich

R. K.

Brittin

(2019). Associations of patient and staff outcomes with inpatient unit designs incorporating decentralized caregiver workstations: A systematic review of empirical evidence. Health Environments Research & Design Journal, 12(1), 26–43. https://doi.org/10.1177/1937586718796590

16.

Joseph

Hamilton

D. K.

(2008). The Pebble projects: Coordinated evidence-based case studies. Building Research and Information, 36(2), 129–145. https://doi.org/10.1080/09613210701652344

17.

Lee

K. J.

Carlin

J. B.

(2010). Multiple imputation for missing data: Fully conditional specification versus multivariate normal imputation. American Journal of Epidemiology, 171(5), 624–632. https://doi.org/10.1093/aje/kwp425

18.

Sajobi

Lucyk

Lorenzetti

Quan

(2015). Systematic review of risk adjustment models of hospital length of stay (LOS). Medical Care, 53(4), 355–365. https://doi.org/10.1097/MLR.0000000000000317

19.

MacAllister

Zimring

Ryherd

(2016). Environmental variables that influence patient satisfaction: A review of the literature. Health Environments Research & Design Journal, 10(1), 155–169. https://doi.org/10.1177/1937586716660825

20.

Morgan

V. R.

Mathison

J. H.

Rice

J. C.

Clemmer

D. I.

(1985). Hospital falls: A persistent problem. American Journal of Public Health, 75(7), 775–777. https://doi.org/10.2105/AJPH.75.7.775

21.

Morse

J. M.

Morse

R. M.

Tylko

S. J.

(1989). Development of a scale to identify the fall-prone patient. Canadian Journal on Aging, 8(4), 366–377.

22.

Pati

Harvey

T. E.

Redden

Summers

Pati

Sipra

(2015). An empirical examination of the impacts of decentralized nursing unit design. Health Environments Research & Design Journal, 8(2), 56–70. https://doi.org/10.1177/1937586715568986

23.

Sadler

B. L.

Berry

L. L.

Guenther

Hamilton

D. K.

Hessler

F. A.

Merritt

Parker

(2011). Essays: Fable hospital 2.0: The business case for building better health care facilities. Hastings Center Report, 41(1), 13–23. https://doi.org/10.1002/j.1552-146X.2011.tb00093.x

24.

Sadler

B. L.

Joseph

Keller

Rostenberg

Keller

Rostenberg

Keller

Rostenberg

(2009). Using evidence-based environmental design to enhance safety and quality: The center for health design. Healthcare Improvement, 28. http://www.ihi.org/knowledge/Pages/IHIWhitePapers/UsingEvidenceBasedEnvironmentalDesignWhitePaper.aspx

25.

Siddiqui

Z. K.

Zuccarelli

Durkin

A. W.

Brotman

D. J.

(2015). Changes in patient satisfaction related to hospital renovation: Experience with a new clinical building. Journal of Hospital Medicine, 10(3), 165–171.

26.

Stichler

J. F.

McCullough

(2012). Same-handed patient room configurations: Anecdotal and empirical evidence. Journal of Nursing Administration, 42(3), 125–130. https://doi.org/10.1097/NNA.0b013e318248073d

27.

Taylor

Card

A. J.

Piatkowski

(2018). Single-occupancy patient rooms: A systematic review of the literature since 2006. Health Environments Research & Design Journal, 11(1), 85–100. https://doi.org/10.1177/1937586718755110

28.

Thomas

J. W.

Guire

K. E.

Horvat

G. G.

(1997). Is patient length of stay related to quality of care? Hospital & Health Services Administration, 42(4), 489–507. https://search-proquest-com-s.web.bisu.edu.cn/docview/206682891?accountid=13771

29.

Ulrich

(1984). View through a window may influence recovery from surgery. Science, 224(4647), 420–421. https://doi.org/10.1126/science.6143402

30.

U.S. Department of Health and Human Services: Centers for Medicare and Medicaid Services. (2014). Medicare program; hospital inpatient prospective payment systems for acute care hospitals and the long-term care hospital prospective payment system and fiscal year 2015 rates; quality reporting requirements for specific providers; reasonable compensation. Federal Register, 79(163), 49835–50536. http://www.gpo.gov/fdsys/pkg/FR-2014-08-22/pdf/2014-18545.pdf

31.

Voigt

Mosier

Darouiche

(2018). Private rooms in low acuity settings: A systematic review of the literature. Health Environments Research & Design Journal, 11(1), 57–74. https://doi.org/10.1177/1937586717702597

32.

World Health Organization. (2002). Prevention of hospital-acquired infections: A practical guide. In Ducel

Fabry

Nicolle

(2nd ed.). https://apps.who.int/iris/handle/10665/67350

33.

Zusman

E. E.

(2012). HCAHPS replaces Press Ganey survey as quality measure for patient hospital experience. Neurosurgery, 71(2), 21–24. https://doi.org/10.1227/01.neu.0000417536.07871.ed