Increasing Access to State Psychiatric Hospital Beds: Exploring Supply-Side Solutions

The hospital under study is one of three state psychiatric hospitals in North Carolina. At the time of this research, it had 398 beds and served a 25-county region with a 2012 total population of approximately 3.4 million (7). Its ratio of 11.7 beds per 100,000 population placed it below the national average of 14.1 beds per 100,000 (2). Most of the study hospital’s adult patients were male and self-pay or uninsured, had a diagnosis of severe mental illness, and were admitted under involuntary commitments (accounts receivable data on file). Community-based crisis services in the region included 494 adult psychiatric beds in 14 general hospitals or private psychiatric hospitals (19.1 beds per 100,000 population) and 66 nonhospital crisis beds in five facilities (2.6 beds per 100,000 population) (7–9).

These community-based services are the first tier of response for individuals in psychiatric crisis in the region, but they also serve as a major source of referrals to the state hospital. The study hospital provides a safety net when local inpatient capacity is exceeded and for patients whose illness is too acute or too difficult to manage in community-based settings. During the last six months of 2012, a mean±SD of 520±63 people waited in emergency departments across the state for admission to a state psychiatric hospital (10). These people waited an average of 3.1±.3 days, although not all patients placed on waitlists were admitted.

At the time of this research, the study hospital’s clinical service units were organized into screening and admissions, medical, child and adolescent, adult acute (short-term), adult community transition (longer-term rehabilitative), and geriatric services. The study reported here focused on patients served in the adult acute and community transition units, because this subpopulation constituted the majority of the monthly waitlist. Each of these units contained male- and female-specific subunits, as well as a high-management subunit for patients requiring more intensive services or increased supervision.

We held nine mental health system stakeholder meetings between October 2008 and February 2012 to communicate the study team’s goals and elicit written and unwritten “rules” governing how people flowed through the hospital. The number of attendees ranged from four to 30. The first four meetings were held with stakeholders from the hospital and from two of the hospital’s local community service areas, including representatives from local mental health agencies, psychiatric crisis care and outpatient care providers, community general hospitals, and law enforcement, all of whom were affected by delays in state psychiatric hospital admission. The last five meetings were held with the hospital’s clinical management staff, with three of the last five meetings also including executives from local mental health agencies. These later meetings were conducted to gain specific information about hospital unit capacities and the flow of patients through the hospital, as well as to seek feedback on preliminary findings.

Hospital administrators indicated that the total number of staffed beds was fixed at 398, but overtime and temporary staffing could be used to accommodate fluctuations in patient case-mix severity. Our research team initially considered cost-neutral scenarios that involved shifting beds between treatment units to explore efficiencies. However, preliminary simulation results indicated that most units were already operating at or near full capacity and that substantially decreasing wait times through supply-side interventions would require increasing bed capacity. Increasing staffed-bed capacity was potentially feasible at the study hospital, because 15% of its licensed beds were not in use at the time of this research. Therefore, our hypothetical scenarios focused on how many additional beds would be needed to reduce wait times by targeted amounts.

We built the simulation model using North Carolina’s Healthcare Enterprise Accounts Receivable Tracking System (HEARTS), state psychiatric hospital waitlist data, information shared at mental health system stakeholder meetings, and personal communications with hospital stakeholders. HEARTS is an administrative claims database and billing system maintained for all visits to the state’s psychiatric hospitals. We used admissions data for all adult acute and community transition unit admissions between July 1, 2010, and July 31, 2012, to determine rates of patient arrivals and gain insights about the rules of patient flow through the hospital. We also used state psychiatric hospital waitlist data to confirm that the baseline model results were consistent with available data.

Discrete-event simulations (6,11) model the operation of a system, such as a hospital, as a sequence of separate events in time (for example, patient admissions, treatment unit transfers, and discharges). In this study, these events changed two system states: the number of patients on the waitlist (a whole number) and treatment bed capacity (full or not). The random variables that needed to be characterized to model this system stochastically were patient waitlist arrival times and treatment unit length of stay.

Our baseline discrete-event simulation model captured the flow of patients through the state psychiatric hospital during fiscal year 2012 (July 1, 2011, through June 30, 2012). We began with a simple model based on preliminary conversations with stakeholders, iteratively incorporating additional factors suggested by stakeholders to approximate the true complexity of patient flows. All scenarios started with an empty hospital, using a one-year warm-up period to allow the simulated hospital and waitlist to reach realistic census levels. The model was built using ProModel, version 8.6 (12). The main components and operationalizations are described below. [Additional technical details of the simulation modeling are included in an online supplement to this article and can be obtained from the first author.]

The model included patients ages 18–64 (the only patients eligible for care in the units being studied). These patients were randomly selected, with replacement from a pool of actual patient admissions to the hospital during fiscal year 2012, thus reflecting actual patient population diversity in age, sex, and diagnoses. Several additional attributes were assigned to patients in the model on the basis of HEARTS data, including length of stay (estimated using a Cox proportional hazards model) and eligibility for the high-management and community transition units.

We modeled patient arrivals to reflect all patients who entered the waitlist, regardless of whether they were ultimately admitted, by increasing observed hospital admission rates by a factor of 1 divided by the mean probability of admission (.38, from waitlist data). In practice, we were told, certain patients may be prioritized for admission from the waitlist for acuity and management concerns. However, stakeholders indicated that these patients could not be identified from the utilization data available to us; thus, in our simulations, hospital admission rates were increased uniformly for all patients.

Hospital clinicians and managers noted that upon referral, if an appropriate adult acute unit bed was available, the patient was admitted to that bed. Otherwise, the patient was placed on the waitlist and retained at the referral source until a bed at the hospital became available. When a regular adult acute unit bed for a male or female became available, a patient was admitted from the male or female waitlist on a first-in, first-out basis. We used person-level waitlist data from one local service area between January 1 and June 30, 2010, to estimate the amount of time people waited before they were admitted to the state hospital, found care elsewhere, or no longer needed inpatient care.

According to clinical managers at the study hospital, admitted patients followed a number of different pathways through the hospital. Specific rules were elicited to describe how and when patients flowed between units. These rules varied by patient sex, eligibility for high-management and community transition units, and availability of beds. The rules were translated into the care pathways outlined in Figure 1.

Figure 1 depicts the two main treatment locations included in the model: the adult acute unit and the community transition unit. To account for fluctuations in staffed beds, we used HEARTS data to determine the daily average number of beds in use in each unit during the year. In the adult acute unit, we modeled 86 beds designated for men, 44 beds for women, and ten high-management beds for use by either men or women. Similarly, in the community transition unit, we modeled 44 beds for men, 22 for women, and ten high-management beds.

We took several steps to verify that the model behaved as intended. For example, we examined the model’s behavior with only one patient arriving and varied the patient attributes to confirm that the patient’s pathway through the hospital was consistent with the pathways depicted in Figure 1 and with the associated rules. Similarly, we varied length-of-stay estimates and probabilities of high-management and community transition unit eligibility to ensure that the model reacted as expected. The first two authors independently reviewed the model’s code and conducted independent scenario analyses. Finally, technical consultants from ProModel verified our simulation model and improved its computational efficiency.

After verifying the model’s structure, we compared baseline simulated patient flow to fiscal year 2012 HEARTS data. We adjusted rules of patient flow so that the simulated outcomes better reflected actual patterns of hospital use (for example, matching the annual expected number of community transition unit discharges). Model results related to patient queuing and unit utilization were also compared with waitlist data and stakeholder input. We used four hypothetical scenarios to simulate alternative capacity expansion targets. In the first expansion scenario, we added 24 beds, distributed equally across the four regular treatment units. In each of the other three scenarios, we repeatedly added a bed to the treatment unit with the highest utilization rate (that is, the bottleneck unit) until a wait time threshold (three, two, or one days) was reached.

The study was approved by the Institutional Review Board at the University of North Carolina at Chapel Hill.

Base case simulation results (Table 2) indicate that the model approximated actual fiscal year 2012 utilization data. During one year, a mean±SD of approximately 1,251±65 patients were admitted to the simulated hospital. Patient lengths of stay (among patients discharged in the simulated year) ranged from one day to 655 days, with a mean of approximately 44±2 days across 50 simulations, compared with an average of 41±67 days for a comparable set of patients from 2011 or 2012 in HEARTS data. The average wait time of 3.3±.1 days for admission was considered sufficiently close to the observed averages of 3.5±.3 days for all patients waiting between July and September 2010 (13) and 3.1±.6 days for patients waiting in general hospital emergency departments during fiscal year 2012 (14). The male and female adult acute units were the most highly utilized units, at nearly 100% each, although all units had utilization rates over 90% (not shown).

Findings from scenarios 1–4 (Table 2) quantify the expected increases in admissions and decreases in average wait times following targeted hospital capacity expansions. Adding 24 beds increased annual admissions by 9% (115.2 patients) and decreased average wait time by 6% (.2) to 3.1±.1 days. In scenario 2, adding 48 beds increased annual admissions by 17% (213.0 patients) and decreased average wait time by 9% (.3) to 3.0±.3 days. Even with the alternate inpatient psychiatric resources available in the region, reducing simulated average wait times below two days in scenario 3 and below one day in scenario 4 required respective capacity increases of 84% (182 beds, for a 73% increase in admissions) and 165% (356 beds, for a 131% increase in admissions), which greatly exceeded the 70 licensed beds that could have become operational at the study hospital if additional staffing had been made available.