Eating disorders (EDs) are difficult to treat with high non-response, dropout, and relapse rates (Fassino et al., 2009; Keel and Brown, 2010; Abbate-Daga et al., 2013). As such, there is a need to better understand the factors that impact both treatment response and long-term outcomes. Identifying predictors of these factors is necessary to improve therapy protocols, inform treatment planning, and identify patients at risk for unfavorable prognoses (Keel and Brown, 2010; Vall and Wade, 2015). An important consideration when examining treatment predictors is the level of care the patient is receiving.

Treatment at partial hospitalization programs (PHP) is most often intended for individuals who have made insufficient progress in outpatient ED treatment or have ED-related behavioral or medical instability that requires regular monitoring. Research has demonstrated that the PHP level of care is effective for many patients (Brown et al., 2018; Reilly et al., 2020). PHPs represent attractive treatment choices to patients, clinicians, and insurance providers, as they demonstrate improved cost-effectiveness and comparable outcomes to inpatient and residential treatment (Anderson et al., 2017). However, not all patients achieve recovery in PHP treatment settings. Thus, some of these individuals later seek admission to a higher level of care (HLOC), including inpatient hospitalization and residential treatment (Abbate-Daga et al., 2015). Although PHP settings are an increasingly common treatment choice for individuals with EDs, research concerning outcome predictors in these settings remains limited. In a recent meta-analysis of outcome predictors, 67% of included studies were from randomized controlled trials and reflected specific treatment settings: 51.5% inpatient, 32.5% outpatient, 1.6% PHP and 0.8% residential settings (Vall and Wade, 2015). As randomized control trial findings might not reflect outcomes in more naturalistic settings, such as outpatient or stepped-level of care settings (e.g., PHP), additional research is warranted to replicate results and identify other variables that might impact ED treatment outcomes in these environments (Vall and Wade, 2015; Walker et al., 2020).

Previous research, not necessarily pertaining to PHP settings as mentioned above, has identified numerous baseline variables that predict outcomes and mediators that help explain favorable treatment response, including: higher body mass index (BMI), fewer binge/purge episodes, increased motivation to recover, lower shape/weight concern, fewer comorbidities, and better interpersonal functioning (Vall and Wade, 2015; Linardon et al., 2016). Several studies have also evaluated whether age is a predictor of outcome, with inconsistent results. For example, some studies have found younger age predicts more favorable outcome in outpatient settings (e.g., Agras et al., 2014), while others have demonstrated the opposite pattern of older age predicting better outcome (Grilo et al., 2012), or found no association between age and treatment outcome (e.g., Lammers et al., 2015). Temperament represents a relatively new area of focus associated with ED treatment outcome (Kaye et al., 2015); five-year follow-up from outpatient treatment suggests that temperamental traits such as low novelty seeking, high harm avoidance, and high reward dependence predict clinical improvement in ED symptoms (Segura-García et al., 2013). Similarly, emotion dysregulation has recently been identified as a critical mechanism in the development and maintenance of EDs (Lavender, 2015) with greater emotion regulation skills predicting favorable ED outpatient treatment outcomes (MacDonald et al., 2017).

Although identifying factors that predict favorable treatment outcomes is essential, it is equally—if not more important—to identify variables that predict poor prognosis. A systematic and meta-analytic review by Vall and Wade (2015), looking mostly at data from inpatient and outpatient settings, demonstrated that higher eating pathology at baseline predicted worse outcomes. Another systematic review evidenced a consistent link between anxiety and depression and worse ED treatment outcomes (Berkman et al., 2007). Smith et al. (2018) identified baseline general anxiety, as well as social anxiety, as significant predictors of poor end of treatment ED psychopathology in residential settings. Accurso et al. (2016) found that higher depression scores predicted more ED psychopathology at short-term follow-up from outpatient treatment, and two studies examining outpatient outcomes at 12-month follow-up found that higher depression scores and the presence of major depression were associated with more episodes of binge eating and purging (Fahy and Russell, 1993; Bulik et al., 1998). Further, research has consistently revealed that longer duration of illness and lower body mass index (BMI) at baseline are poor prognostic factors (Howard et al., 1999; Reas et al., 2000; Pinter et al., 2004).

Although research has identified general predictors for unfavorable treatment outcomes, a critical yet understudied subset of patients with poor outcomes are those that enter PHP for an ED but later require a HLOC due to needing more support and intensive treatment. As treatment in PHP levels are an increasingly common and attractive choice for ED patients, it is crucial to understand which types of patients will have favorable outcomes in these settings. Moreover, it is important to identify patients early on in treatment who often need or could benefit from a HLOC. Further, elucidating PHP patient characteristics that necessitate a HLOC is advantageous in identifying and implementing supplemental programming in PHP settings that might benefit individuals experiencing severe eating pathology and potentially prevent having to step up to a HLOC (Fewell et al., 2017).

As noted, there is a lack of literature examining outcome predictors at the PHP level, and there is limited research that evaluates the clinical indicators that predict unfavorable outcomes. As such, the purpose of the current study was to better characterize PHP patients who require a HLOC and identify predictors of those who discharged to a HLOC (i.e., residential or inpatient) after an initial PHP admission. Specifically, we aimed to characterize patients who require a HLOC by examining their demographic and baseline clinical characteristics and differences on these variables between patients who discharged to a HLOC and patients appropriate for PHP. Although the current study was exploratory in nature, it also aimed to assess previously identified predictors of treatment outcome (i.e., anxiety, depression, eating pathology, duration of illness, age, BMI) as predictors of patients requiring a HLOC.

Patients with missing reasons for discharge data (n = 8) were significantly younger [t₍₇₉₆₎ = 5.98, p < 0.001] and had a shorter duration of illness [t₍₇₇₇₎ = −6.55, p < 0.001] when compared to patients whose reasons for discharge were documented. There were no group differences on missing/non-missing reasons for discharge data on any of the following continuous and categorical variables: eating disorder psychopathology, age of onset, admit or discharge BMI, trait anxiety, depressive symptoms, emotion dysregulation, sensitivity to punishment, race, ethnicity, or diagnoses of comorbid depressive, anxiety, alcohol use, and substance user disorders (ps > 0.05). Individuals with missing data on anxiety [χ²_{(1, N = 793)} = 6.33, p = 0.02] and depressive disorders [χ²_{(1, N = 793)} = 5.66, p = 0.02] were more likely to discharge to a higher level of care, while missing data on all other baseline variables were not related to reason for discharge (ps > 0.05).

When compared to patients who were appropriate for continued PHP, a significantly greater proportion of patients who discharged to a HLOC had previously received ED treatment at residential treatment centers (ϕ_c = 0.21) and were diagnosed with comorbid anxiety disorders (ϕ_c = 01). Moreover, patients who discharged to a HLOC had a lower BMI (d = 0.58) and reported greater anxious (d = 0.71) and depressive (d = 0.67) symptomatology than those who were appropriate for continued PHP. See Table 1 for details.

When all patients were included in the first logistic regression, the overall model was significant in predicting the dependent variable [i.e., admission to a HLOC following PHP admission; Likelihood Ratio χ²₍₈₎ = 22.94, p = 0.003]. Only BMI significantly predicted discharging to a HLOC. Specifically, for every one unit decrease in BMI, participants were 15% more likely to admit to a HLOC (Table 2). Upon excluding patients who were hospitalized for suicidality and/or suicide attempts following PHP admission, the overall model remained statistically significant [Likelihood Ratio χ²₍₈₎ = 21.57, p = 0.005]. BMI remained the only significant predictor (χ² = 10.91, OR = 0.78, p = 0.001), such that every one unit decrease in BMI resulted in a 22% greater likelihood of admitting to a HLOC. This model's restricted log likelihood value (−113.27) was closer to zero than in the original model (−143.17), suggesting model fit is improved when patients who were hospitalized for suicidality and/or suicide attempts were excluded from the analysis.

The current study sought to identify predictors of patients first admitted into a PHP level of care who later required a step up to residential or inpatient treatment settings. Findings demonstrate that patients with lower BMIs were more likely to admit to a HLOC from PHP. Data indicate that this finding remains statistically significant when psychiatric hospitalizations for suicidality and/or suicide attempts were excluded. Results highlight the importance of considering BMI when developing treatment plans for newly admitted patients to PHP settings.

Previous research demonstrates that lower BMI at pretreatment is a poor prognostic factor (Fahy and Russell, 1993; Howard et al., 1999; Agras et al., 2000; Pinter et al., 2004; Berkman et al., 2007). One study identified lower BMI at the time of inpatient admission predicts PHP failure and inpatient readmission for patients with anorexia nervosa (Howard et al., 1999). Another study conducted in an inpatient setting highlighted that patients with anorexia nervosa with BMIs below 15 kg/m² were significantly more likely to develop a lower BMI at follow-up (Pinter et al., 2004). Two studies examining bulimia nervosa in outpatient settings identified that lower BMI pretreatment was associated with worse outcome in terms of binge/purge frequency and eating disorder psychopathology at end of treatment and short-term follow-up (Fahy and Russell, 1993; Agras et al., 2000). The current study adds to the previous literature in revealing that transdiagnostic patients with EDs with lower BMIs at admission, specifically in PHP settings, were more at risk for stepping up to HLOC. As such, patients across ED diagnoses with lower BMIs might demonstrate greater need for more comprehensive, targeted interventions, specifically in regards to nutrition, when admitting to PHP settings.

Exploration of group differences revealed that when compared to patients appropriate for continued PHP, patients who discharged to a HLOC were significantly more likely to have previously received ED residential treatment. This highlights the importance of acknowledging and exploring previous treatment history at admission. Certainly, communication between treatment facilities is necessary to supplement success (Anderson et al., 2017). Further, knowledge of previous treatment can inform the implementation of expectations outlined at admission to a PHP. Setting clear behavioral contingences based on previous treatment that reinforce functional rather than dysfunctional behaviors might engender early change (Wisniewski and Ben-Porath, 2015; Ziser et al., 2018). Establishing an explicit treatment contract collaboratively with patients to align with their personal goals might enhance patients' motivation, compliance, and autonomy (Wisniewski and Ben-Porath, 2015). As such, initiating collaborative contingency contracts at the outset of PHP treatment for patients with a history of residential treatment might augment outcomes.

Additionally, group differences demonstrated that patients who discharged to a HLOC were more likely to be diagnosed with comorbid anxiety disorders and endorsed greater anxious and depressive symptomatology.

Previous research consistently demonstrates that higher levels of anxiety and depression are related to poorer prognosis (Berkman et al., 2007; Vall and Wade, 2015). Indeed, neurobiological research indicates that anxiety inhibits motivation to eat, which maintains the cycle of restriction and weight loss (Frank et al., 2019). As such, combining nutritional rehabilitation with specific biological interventions might supplement treatment outcome (Frank et al., 2019).

To our knowledge, this was the first study to explore clinical predictors of patients who require a HLOC after initial admission to a PHP. Findings indicated targeted early interventions in ED PHP settings might benefit patients with lower BMIs. Additionally, current findings suggested that patients exhibiting higher levels of anxiety and depression and/or reporting previous ED residential treatment experience might also benefit from supplemental intervention strategies. Specialized treatment for these cases is both practically and economically advantageous as it might reduce the risk of re-hospitalization and at-risk patients needing to step up to a HLOC.

The data the support the findings of this study can be made available from the corresponding author upon request.

The studies involving human participants were reviewed and approved by University of California, San Diego. Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.