Is It Time for the Mental Health Field to Consider Unplanned Discharge a Key Metric of Patient Safety?

Abstract

Mental health lacks robust measures to assess patient safety. Unplanned discharge is common in mental health populations and associated with poor outcomes. Clarifying whether unplanned discharge varies across settings may highlight the need to develop measures to reduce harms associated with this event. Unplanned discharge rates were compared across the Department of Veterans Affairs’ acute inpatient and residential mental health treatment settings from 2009 to 2019. Logistic regression was used to create facility-level, adjusted unplanned discharge rates stratified by setting. Results were described using central tendency. Among 847,661 acute inpatient discharges, the mean unplanned discharge rate was 3.3% (range, 0%–18%). Among 358,117 residential discharges, the mean unplanned discharge rate was 17.9% (range, 1%–48.3%). Unplanned discharge is a marked problem in mental health, with large variation across treatment settings. Unplanned discharge should be measured as part of patient safety efforts.

The health care system plays a central role in the problem of adverse events and medical errors. Despite strides toward safer health care (Donaldson, 2021), there has been little focus on patient safety in mental health treatment settings (Donaldson, 2021; Shields et al., 2018; Thibaut et al., 2019). The mental health field lacks a robust set of metrics to assess adverse events and patient harm. Shields et al. (2018) underscore that “inpatient psychiatric care has been left on the sidelines of efforts to measure and improve patient safety, despite glaring need (p 1859).” Of particular note, unplanned discharge is currently not an established measure of patient safety in mental health settings (Centers for Medicare and Medicaid Services [CMS], 2020; Thibaut et al., 2019). Yet, unplanned discharge is common among mental health populations and predicts a number of poor health outcomes including all-cause mortality, suicide mortality, and readmission after discharge (Alfandre, 2019; Holmes et al., 2021; Riblet et al., 2018). Therefore, operationalizing unplanned discharge as a patient safety measure in mental health should be a chief consideration for providers, researchers, and policy makers.

In the field of mental health, providers use related but differing terms to denote an unplanned discharge. For example, the CMS uses the term against medical advice (AMA) discharge to indicate that a patient “left against medical advice or discontinued care (CMS, 2021).” The National Health Service uses the term self-discharge (National Health Service, 2019) to indicate that there was a “self-discharge, patient request despite medical advice, or request after breach of ward rules” (Bickley et al., 2013, p 655). Addiction programs use the term “disciplinary discharge” to indicate that there was an “adversarial termination of services due to a client’s failure to comply with program rules and expectations” (White et al., 2005, p 12). The Veterans Affairs (VA) uses the term irregular discharge to indicate the “release of a competent patient from a VA or VA-authorized hospital, nursing home, or domiciliary care due to refusal, neglect, or obstruction of examination or treatment; leaving without the approval of the treating health care clinician; or disorderly conduct and discharge is the appropriate disciplinary action” (United States Government Publishing Office, 2009, p 1177). None of the available coding methods distinguish between different unplanned discharge scenarios. Yet, collectively unplanned discharge is associated with poor health outcomes (Bickley et al., 2013; Kuo et al., 2010; Riblet et al., 2018; White et al., 2005).

Experts emphasize that unplanned discharge is a system-level problem that cannot be solely attributed to behaviors or characteristics of individual patients (Alfandre, 2019; White et al., 2005). This concern has been borne out in the literature where studies have found that rates of AMA discharge vary based on hospital characteristics such as urban location, medium hospital size, and for-profit status (Onukwugha et al., 2021). Available studies, however, have generally focused their attention on acute medical wards and have not considered whether unplanned discharge rates also vary based on mental health treatment settings including acute inpatient and residential settings. Gaining insight into whether there is variation in unplanned discharge rates across mental health settings offers an opportunity to evaluate potential drivers of this variation and to develop strategies to address this patient safety event.

We aimed to address this concern by examining whether there is variation in unplanned discharge rates across acute inpatient and residential mental health treatment settings in the US Department of Veterans Affairs, one of the largest, integrated health care systems. Evidence of substantial variation in unplanned discharge rates would begin to build the case for the need for the mental health field to treat unplanned discharge as a patient safety measure and to develop approaches to prevent or mitigate related harms.

METHODS

Using the VA Corporate Data Warehouse (CDW), we conducted a population-based cohort study comparing unplanned discharge rates across VA acute inpatient or residential mental health treatment settings (herein after referred to as acute inpatient and residential). The study cohort consisted of patients who accessed VA care (i.e., VA users) and had an admission to a VA acute inpatient or residential setting between 2009 and 2019. In the VA, residential programs treat a range of mental health conditions such as substance use disorders (SUDs) and posttraumatic stress disorder (PTSD) (Department of Veterans Affairs, 2017). In an analysis of individual-level predictors of irregular discharge in VA residential and acute inpatient settings, Riblet et al. (2021) observed that primary discharge diagnoses in residential programs were typically related to SUD (72%) or PTSD (17%).

The study was approved by the Veteran’s institutional review board of Northern New England.

Study Variables

From the CDW, we abstracted the following covariates: age, sex, race/ethnicity (Black, non-Black Hispanic, White, other), marital status at time of admission, primary discharge diagnosis, year of admission, and preadmission mental and physical health diagnoses over the 2 years before admission coded using a previously developed index (Shiner et al., 2021). We categorized primary discharge diagnosis using the following groupings: alcohol use disorders, bipolar disorders, depressive disorders, drug use disorders, psychotic disorders, trauma-related disorders, and other mental health disorders. We coded the number of mental health diagnoses using the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition categories (0–1, 2–3, and 4+) (American Psychiatric Association, 2013). We coded the number of physical health diagnoses using non–mental health Elixhauser conditions (0, 1, and 2+) (Quan et al., 2005). We excluded hypertension without complications. We collapsed diabetes with and without complications into a single condition. We also collapsed cancer diagnoses into a single measure. For all diagnostic groupings, we considered a single occurrence from an inpatient facility or two or more outpatient encounters 7 to 365 days apart.

We identified eligible facilities using station numbers as described in the VA inpatient file. We required that residential facilities reported at least 200 discharges during the entire study period and that acute inpatient facilities reported at least 1200 discharges during the entire study period. We chose different thresholds for these two settings because we noted in our study population that the unplanned discharge rate was roughly one sixth for acute inpatient stays. Therefore, we required different thresholds to ensure adequate confidence in the point estimates.

Analysis

We performed our analysis at the level of the facility. We calculated unplanned discharge rates using the following approach: 1) the numerator included the total number of discharges that were coded as irregular within the first 182 days of admission; and 2) the denominator included the sum total of discharges that were coded as irregular plus those that were coded as regular. We coded any discharges that were associated with stays longer than 182 days as “regular.” We chose 182 days as a cutoff because residential stays in the VA typically range from 30 to 90 days. Therefore, length of stay (LOS) beyond 182 days represents several times the typical (Cook et al., 2014; Shiner et al., 2018; Substance Abuse and Mental Health Services Administration, 2015). We excluded transfers and deaths in hospital when calculating the unplanned discharge rate.

We used logistic regression to create facility-level, unplanned discharge rates that were adjusted for all covariates described previously. We saved residuals for each individual facility and averaged residuals by facility. The facility adjusted value is simply the grand mean (% irregular) plus the mean residual for all admissions from the facility.

We describe the observed variation in unplanned discharge rates across facilities both graphically using a Turnip Graph (Woloshin, 2001) as well as numerically using a central tendency and variation.

We performed data management and analysis using SAS Version 9.4 (SAS Institute, Cary, NC).

RESULTS

The study population included a total of 436,234 unique patients experiencing 1,205,778 discharges. Among the 112 acute inpatient facilities that met our inclusion criteria during the study period, there were 847,661 discharges (343,411 patients). The mean distribution of discharges per facility over the study period was 7,568 (SD, 3,967). Among the 181 residential facilities that met our inclusion criteria during the study period, there were 358,117 discharges (187,859 patients). The mean distribution of discharges per facility was 1,979 (SD, 2,157). Less than 1% of all residential and acute admissions were excluded due to being associated with facilities that had fewer admissions then met our inclusion criteria. Many of these claims likely represented data errors as some had extreme and implausibly low counts.

We stratified our results by setting (acute inpatient and residential) because we found in our data that unlike acute inpatient settings (median LOS, 6.0; interquartile region [IQR], 8.0), residential settings had substantially longer LOS (median LOS, 37.0; IQR, 57.0). Residential stays also had considerably higher rates of unplanned discharge. Across acute inpatient facilities, we found that the mean adjusted unplanned discharge rate was 3.3% with rates ranging from 0% to 18% (IQR, 3). Conversely, the mean adjusted unplanned discharge rate in residential settings was 17.9% with rates ranging from 1.0% to 48.3% (IQR, 11.1).

Aligned with these trends, Figure 1 showed marked variation in unplanned discharge rates across facilities and especially among residential facilities.

FIGURE 1.

Turnip graph describing variation in adjusted, unplanned discharge rates among acute inpatient, and residential settings, the Department of Veterans Affairs, 2009–2019.

DISCUSSION

The rates of unplanned discharge are fivefold higher among residential versus acute stays, and there is marked variation in rates across facilities. Most striking is the fact that there are residential facilities with rates approaching 50%, whereas others have rates as low as 1%. This degree of variation is striking and highlights a potential patient safety problem at sites with such high rates. Future work should examine the underlying causes of high rates of unplanned discharge at certain sites. Sites with very low rates of unplanned discharge may also have important lessons to share with those with very high rates.

Unplanned discharge is a serious problem in mental health treatment settings and especially among residential settings. It may be useful to operationalize unplanned discharge as a patient safety measure and incorporate this measure into mental health patient safety dashboards. These data could yield critical information to inform the design and implementation of effective interventions to prevent harms associated with unplanned discharge. For example, evidence-based management of withdrawal symptoms has been associated with large reductions in AMA rates (Holmes et al., 2021). More work, however, needs to be done to translate this health care event into a reliable patient safety measure. We observed that health care systems use variable terms and definitions to denote an unplanned discharge. This may serve as a critical barrier to developing effective strategies to prevent associated adverse outcomes. To the best of our knowledge, there is no mechanism by which VA administrators and researchers (or CMS) can readily distinguish between types of unplanned discharge. CMS does not include disciplinary discharge in its coding schema (CMS, 2021). At the local level, it may be appropriate to use a broad term such as irregular discharge to inform quality improvement (QI) efforts because the event rate may be low and clinical teams presumably have the resources to study each case. At the system level, however, catch-all phrases such as AMA discharge or irregular discharge may hamper QI efforts because the system cannot discern the type of unplanned discharge or respond with a fitting, system-level intervention. For example, facilities with a high prevalence of self-initiated discharges due to inadequate management of withdrawal symptoms may require different interventions than facilities with a high prevalence of disciplinary discharges due to noncompliance with unit policies.

Related to these concerns, large health care databases such as the VA are not designed to provide granular data about the reasons for unplanned discharge or related contextual factors. This is a notable barrier to studying the problem of unplanned discharge as researchers have difficulty precisely distinguishing system-level and patient-level factors. For example, it is entirely possible that different interventions are required to address self-initiated discharges due to dissatisfaction with symptom management (Holmes et al., 2021) versus self-initiated discharges due to distress over unit policies that penalize noncompliance with treatment (Williams and Bonner, 2020). In our analysis, we aggregated data over 10 years to generate facility-level estimates. Process improvement work, however, necessitates that researchers develop reliable measures that can provide more timely and actionable feedback on system performance. Ultimately, researchers need to demonstrate that lowering the rates of unplanned discharge results in improved outcomes and safer care. To solve the problem of unplanned discharge in mental health settings, it may be necessary that health care systems not only develop a robust set of measures that delineate the type of unplanned discharge but also supplement these data with qualitative surveys (or interviews) of high- and low-performing sites.

Our study has several strengths. We report on unplanned discharge rates in the largest, integrated health care system, and our data spans from 2009 to 2019. We are the first study to report on variation in unplanned discharge rates across acute inpatient and residential mental health settings. Our results, however, may not be generalizable to other populations because our analysis focuses exclusively on VA-provided care. In comparison to the general US population, VA users tend to be male and have more health comorbidities (Eibner et al., 2016). These factors may hinder (or facilitate) treatment engagement. Furthermore, VA residential programs treat a broad range of mental health problems such as PTSD. It is possible that the system-level drivers of unplanned discharge in PTSD residential programs may differ from those seen in SUD residential programs. VA facilities also lack financial incentives experienced by civilian hospitals and private insurers. Depending on payment structure, hospitals may be adversely incentivized to discharge a patient AMA. The Inpatient Psychiatric Facility Quality Reporting Program specifically excludes AMA discharges from the denominator of measures that assess the quality-of-care transitions (CMS, 2020). Similarly, the CMS’ Hospital Readmission Reduction Program excludes AMA from the denominator when calculating hospital readmissions for targeted conditions including heart failure, acute myocardial infraction, and pneumonia (Onukwugha and Alfandre, 2019). There is emerging evidence to suggest that AMA discharge rates have risen in the United States since the implementation of the Hospital Readmission Reduction Program (Alfandre et al., 2022; Onukwugha and Alfandre, 2019). Although these financial incentives do not apply to VA facilities, it is noteworthy that the VA Strategic Analytics for Improvement and Learning program excludes irregular discharges in its calculation of facility-level, all-cause, and disease-specific readmission rates. We also combined facility-level data across many years. It is possible that during this timeframe some units either closed or opened. A prior study found that irregular discharges rates have fallen in the past 10 years in acute inpatient settings but have risen in residential settings (Riblet et al., 2021). To account for this concern, our model adjusts for calendar year. Although our study adjusts for salient factors associated with unplanned hospital discharge, unmeasured confounding could have influenced some facility rates, although the huge variation found is unlikely to be explained by factors not included in our adjustment model (Riblet et al., 2021). Finally, as stated earlier, we did not explore the system-level factors that may explain the observed variation between high- and low-performing facilities. It has been suggested that high-quality hospitals may be less likely to discharge patients AMA (Onukwugha et al., 2009). Thus, our findings highlight the need for future research to determine the complex set of factors that contribute to variation in unplanned discharge rates across mental health settings and use these data to develop interventions that can benefit low performers.

CONCLUSIONS

In summary, unplanned discharge is a marked problem in mental health populations, with large variation across mental health treatment settings. The mental health field should consider incorporating unplanned discharge into its patient safety measures and using these data as part of its efforts to improve patient safety.