So, What If Something Happens (SWISH): an International Multi-centre Retrospective Cohort Study of Escalations Back to Hospital during Hospital at Home Admissions

Key points

Question

What is the rate, cause, and risk factors of unplanned escalation in Hospital-at-Home (HaH) programs across three international sites, and can a standardised definition of unplanned escalation serve as a core safety metric?

Findings

Unplanned escalation rates were consistent across sites (6.1–6.5%), mostly due to deterioration of existing conditions. Age <65 compared to those 85 and older was associated with significantly lower odds of unplanned escalation (OR 0.61, 95% CI 0.38-0.96). Comorbidities of diabetes (OR 1.39, 95% CI 1.02-1.9) and stroke (OR 1.76, 95% CI 1.1-2.6) were associated with higher odds of escalation.

Meaning

Consistent escalation rates (6–7%) across three countries suggest a potential internal benchmark for HaH safety performance. The low rates of ICU admission and mortality reinforce that HaH can safely manage acute patients, with appropriate systems for escalation. Unplanned escalation is a viable and standardisable safety metric for HaH programmes internationally.

Introduction

Hospital-at-home (HaH) is an evolving and expanding acute, hospital-level care model internationally that delivers acute hospital services and treatment to patients at home¹ who would otherwise require admissions to a brick-and-mortar (BAM) hospital. In many trials and meta-analyses,^2–13 HaH has been demonstrated to be acceptable to patients, clinically effective and safe, resulting in the expansion of this model of care around the world. As HaH programs expand, there is a need to standardise quality metrics, specifically regarding timely response to deterioration resulting in hospital escalation.

In HaH settings, unplanned escalation events serve as key indicators of clinical risk detection, response, and outcomes. Safety measures in published data on HaH have largely addressed mortality and readmissions after discharge.^14–18 However, mortality is a very low frequency occurrence in HaH, and readmission after discharge does not address clinical deterioration during the episode of HaH care. Despite its importance as a key quality indicator, unplanned escalation in HaH has not been clearly defined in the literature.¹⁹ Existing studies of escalation from HaH to BAM hospitals, have provided minimal detail on escalation reasons and outcomes,²⁰ focused primarily on COVID-19 patients²¹ or are single-centre studies. Definitions remain heterogenous and may include clinical, logistical, patient preference and operational reasons rather focused on patient safety outcomes.

In this study, we propose a standardised definition of an unplanned escalation event as a patient safety outcome measure for HaH care, defined as an unplanned return to hospital for BAM ward admission during a HaH episode, requiring at least an overnight stay. The aim of this study was to apply this standard definition across three HaH units in different countries to describe the rate, reasons, and characteristics of unplanned escalations, and to identify their risk factors.

Methods

Setting

We conducted a multi-center, retrospective study from 1 January 2023 to 31 December 2023. The three sites were: Epworth Hospital (EPW) in Australia, Mass General Brigham (MGB) in the United States of America and National University Health System (NUHS) in Singapore. EPW is a private not-for-profit hospital system in Melbourne, Australia, comprising 3 acute hospitals. The HaH unit was established in 2000 and at the time of the study had a maximum capacity of 30 beds. MGB is an academic health system in Boston, MA, United States of America. The HaH unit was established in 2016, and at the time of the study had a maximum capacity of about 50 beds. NUHS is an academic health system in Singapore comprising 3 acute hospitals. The HaH unit was established in 2020, and at the time of the study had a maximum capacity of 30 beds. The admission criteria for each program is included in eAppendix table 1.

The models of HaH in the three services have broad similarities. Patients are admitted inpatients and managed by a specialist hospital-based HaH clinical unit. The units provide all medical, nursing, pharmacy, therapy, and other care, and deliver face to face and virtual visits. All radiology, pathology and labs are serviced by the hospital. The hospital retains clinical and legal governance. The units provide 24 hour, 7 day cover for their patients. However, some treatment capabilities of each unit differed, e.g. only MGB had access to continuous remote monitoring.

Ethics approval was sought from each hospital institution: EPW 307/24 , MGB’s IRB (2024P000412), and NUHS (2024-3199). Waiver of informed consent was granted as this was a descriptive retrospective record based study of events occurring within a HaH service in the normal course of its operation.

Patients

We included all patients admitted to HaH in each unit during this period. We excluded patients admitted to HaH for day admissions lasting <24 hours, example includes for a single dose of intravenous therapy or a procedural change such as suprapubic catheter at home. Patients were included regardless of admission source (including primary care, emergency department, and ward transfers). Patients were identified through the electronic health record.

Patient characteristics were extracted from the electronic health record and included: age, sex assigned at birth, race/ethnicity, nursing home residency status, socioeconomic status, point of entry into HaH, conditions treated in HaH and major comorbidities, treatment in HaH and length of stay before and during HaH admission. For socioeconomic status, both MGB and EPW used postal-code associated SES. In NUHS, as it was a government hospital, we used a SES-proxy of ward class, where subsidised ward class was determined to reflect low-SES and private ward class to reflect high-SES.

Outcomes

The primary outcome was an unplanned escalation episode, defined as an unplanned return to the hospital for BAM ward admission during a HaH admission, for at least an overnight stay. Unplanned escalations were categorised as: deterioration of existing condition for which the patient was admitted, deterioration due to new condition, social/environmental, patient/caregiver preference, nonadherence and others. The secondary outcomes included: planned escalations – which included return to the hospital for investigations or consultations which cannot be undertaken at home and flagged as possibilities at the time of transfer, including surgery, chemotherapy, transfusion, other infusion, or other; mortality prior to escalation and mortality following escalation. The rate of mortality was defined as the number of patients who died after escalation divided by total number of patients who escalated.

Classification of planned or unplanned escalations was conducted through chart review and a standardised data collection form by the individual hospital investigators. Any events that were deemed uncertain by an individual hospital investigator were reviewed and adjudicated by the authors (DML, MM, SQK).

Statistical Analysis

We used summary statistics to describe patient demographics, using numbers and proportions for categorical data, mean for normally distributed data and medians for non-normally distributed data. We conducted a logistic regression analysis of several key, pre-defined risk factors to determine if there were significant predictors of unplanned escalations. Authors (DML, MM, SQK) pre-selected the variables based on likely clinical association with the outcome. p-values of <0.05 was considered to be statistically significant. R Studio for MacOS (R 4.4.3) statistical software was used.

Results

Patient and site characteristics

A total of 3114 participants were included in the analysis, 558 from EPW, 1515 from MGB and 1041 from NUHS (Table 1). The median age was 62.5 in EPW, 70 in MGB and 61 in NUHS. The proportion of males overall was 44.8%. Race was classified differently in every site and included in eAppendix Table 2. Majority of patients from all sites were transferred to HaH from hospital wards (64% EPW, 69% MGB, 57% NUHS).

The primary condition treated in HaH was infection in all sites (66% EPW, 53% MGB and 79% NUHS), though EPW treated more surgical conditions (27% EPW, 2% MGB, 1% NUHS). MGB also had more cases of chronic obstructive pulmonary disease (COPD) and asthma (0.18% EPW, 16.3% MGB, 0.19% NUHS). Most of the patients came from medical specialty, up to 91.93% and 99.6% in NUHS and MGB, respectively. The most common comorbidities were heart failure, COPD/asthma, diabetes and cancer. Most patients received more than 1 different type of treatment, with the most common treatment being intravenous antibiotics. The median number of nights in the hospital prior to HaH was 1 day while the median number of nights treated in HaH was 5 days. MGB and NUHS had largely similar lengths of stay (4 days), while EPW had a clinically meaningfully longer length of stay (7 days) under HaH.

Unplanned Escalations

The rate of unplanned escalations was similar between all units – 6.1% in EPW, 6.5% in MGB and 6.5% in NUHS (Table 2). The most common reason was deterioration of the existing condition for which the patient was admitted followed by deterioration due to a new condition. The rate of planned escalations varied between units from 4.3% at EPW to 0.2% at MGB and 2.4% at NUHS. The main reason for planned escalation across the 3 units was surgery (eAppendix 3). The rate of mortality after unplanned escalation was highest in EPW at 11.7% followed by 5.8% in NUHS and 1.0% in MGB (eAppendix 4).

Haemodynamic instability emerged as the most common reason for unplanned escalation of care at both MGB and NUHS (Table 2). In contrast, worsening infection without haemodynamic instability was the leading cause at EPW, accounting for 33.3% of cases. Shortness of breath was consistently the second most common reason across all 3 units. Although biochemical abnormalities and altered mental status were among the top causes of unplanned escalation at MGB and NUHS, they were notably less common at EPW, reported at only 3.33% and 0%, respectively. Other emergent symptoms such as bleeding, chest pain, and arrhythmias were observed at lower frequencies across all units.

The majority of unplanned escalations, were directed to emergency departments at NUHS (67.65%) and MGB (89.9%), whereas at EPW, most patients were admitted directly to the ward (55.88%). A small proportion of patients were escalated to the ICU within 24 hours of returning to the hospital – 0% at EPW, 3.0% at MGB, and 1.4% at NUHS. 4.3% had subsequent death in the hospital, with respiratory failure as the leading cause of death (45.5%). Most deteriorations occurred at a median of 4 days following admission to HaH. 23.8% of patients returned to HaH after escalation to BAM care, with patients typically spending a median of 2 days in hospital before returning. Only a small minority (3.4%) required a second escalation to BAM care.

Predictors of Escalation

Regression analysis revealed that age <65 compared to those 85 and older was associated with significantly lower risk of unplanned escalation (OR 0.61, 95% CI 0.38-0.96). Comorbidities of diabetes (OR 1.39, 95% CI 1.02-1.9) and stroke (OR 1.76, 95% CI 1.1-2.6) were associated with higher risk of escalation (Table 3). Sex, socioeconomic status, point of entry, condition treated in HaH, number of different treatments provided by HaH, nights treated in HaH in total and number of nights in the hospital prior to HaH were not associated with risk of unplanned escalations.

Discussion

In our study, applying a standardised definition of unplanned escalation across three international centres, the rate of unplanned escalations were remarkably similar at 6-7%, mostly due to deterioration of the acute condition. Comorbidities of stroke and diabetes were associated with higher odds of unplanned escalation, and age <65 was associated with lower odds of unplanned escalation.

Our findings build on existing research around escalations in HaH. Escalation was examined in early HaH studies, first reported at 2% in a 2005 study by Leff et al.²² A study from Israel’s largest HaH provider reported unplanned escalations as 8.6%, but did not include reasons for escalation or patient outcomes following escalation.²⁰ Another study focusing primarily on COVID-19 patients reported unplanned escalation of 18%, but has limited generalisability due to its dependence on regional pandemic policies and the inclusion of a heterogeneous cohort from both urban and rural areas.²¹ A final single-centre study from Australia reported unplanned escalation at 4.2%.²³ Our study adds to the literature by applying a clear and consistent definition to large case numbers, uniform HaH models, a long study period, and multi-centre data.

To our knowledge, this is the first study to propose a standardised definition of unplanned escalations as a patient safety outcome measure to offer prospective HaH planners a benchmark of such events in a properly functioning, high quality HaH service. Our findings suggest that this definition may be applicable across diverse HaH units worldwide. Despite differences in patient demographics, diagnoses, unit experience, financing and geography amongst the participating units, the similarity in unplanned escalation rate suggests that each unit may operate with an internal benchmark of what is considered an acceptable level of risk. Although international comparisons are challenging due to the local tailoring of HaH programmes to local constraints and admission thresholds shaped by clinical norms and cultural expectations, clinical processes are continually reviewed to maintain a manageable risk profile.²⁴ Elevated escalation rates would prompt evaluation of intake criteria and workflow processes, while unusually low rates may signal that the admissions are unnecessarily low acuity which may not require hospital-level care. This dynamic highlights how HaH evolves within the unique clinical culture and risk tolerances of each institution.

The pattern and reasons for escalation elicited in this study can reassure clinicians that are concerned about patient safety in HaH and help HaH systems plan more effectively for unanticipated events. First, the reasons for unplanned escalation, such as haemodynamic instability and shortness of breath, are largely consistent across all three units, and protocols should be in place for these events. The low rate of escalation for patient factors including non-adherence, environmental or caregiver factors suggests that social and environmental factors are not an inhibitor to HaH care. Second, the unplanned escalation rate did not differ between HaH admissions from the ED compared to BAM, which may indicate the standardised triage and intake assessments tailored for each setting. Third, the number of deaths exceeded the number of ICU admissions, possibly reflecting the inclusion of patients nearing end-of-life or those with limitations on resuscitative efforts amongst HaH patient profiles. This highlights the importance for HaH programmes to build end-of-life capabilities to support dying at home if preferred. Fourth, unplanned escalations were not disproportionately higher on weekends, which may reflect that HaH teams were adequately staffed irrespective of day of the week. Finally, between-site variations may reveal contextual differences – for example, the higher proportion of patient- or caregiver-initiated escalations at NUHS may reflect the higher prevalence of family caregivers and involvement in medical decision making in Singapore; these caregivers may have lower thresholds to request hospital transfer when uncertainty arises.

Age below 65 years appears to be a protective factor against escalation to hospital care, whereas there was higher incidence of escalation of older adults attributed to non-adherence, environmental barriers, and caregiving challenges.²⁵ For older adults, HaH programs must therefore proactively address caregiver burden and frailty to enable them to benefit from HaH including reduced risk of hospital-acquired infections and lower incidence of delirium through care in a familiar environment.²⁶

Unplanned escalations are an expected component of every HaH service and should be anticipated in service design, staffing and supervision. They signal events that interrupt a HaH episode and require unexpected and time-sensitive interventions. Prompt detection and response will reflect a high functioning responsive HaH unit.²⁷ Escalation rates can be utilised in internal audits, piloting of new clinical applications or service improvement, as a proxy for effectiveness and impact on patient safety. Interventions such as HaH rapid response teams, which provide timely response without ED transfer²⁸ may further influence escalation rates. The standardisation of outcome measures can help in evaluation of such interventions and serve as a benchmark for performance across HaH units both locally and internationally. Furthermore, that 20-40% of unplanned escalations returned to HaH may reflect an operational feature of mature HaH units that maintain continuity of care, allowing re-transfer once the acute issue is stabilised. This also reinforces the flexibility of HaH as part of a continuum of hospital care rather than a one-way transfer.

This study has several limitations. First, we were not able to measure patient acuity at admission due to the lack of standardised admission acuity scoring systems internationally. Future research should seek to define measures of acuity and capture them more systematically to enable more meaningful cross-site benchmarking and to determine whether escalation rates reflect service performance or case-mix. We were also not able to report a measure of multimorbidity due to the variation in how comorbidities are coded in the electronic medical record. Instead, we report the presence or absence of important comorbidities as part of baseline characteristics. Second, the mode of detection of deterioration (e.g. remote monitoring signals, nurse assessments or patient initiated alerts) were not consistently recorded across sites. For example, while MGB deployed continuous remote monitoring for all patients, it is not clear whether this enabled earlier identification of hemodynamic instability.²⁹ Future works should examine such mechanisms and timeliness for deterioration detection. Third, the documentation of advance care plans or code statuses were not consistently captured across all sites, limiting the differentiation of palliative deaths. Fourth, the units selected in this study may not be representative of HaH units worldwide and did not include units in rural or low and middle income countries.³⁰ Fifth, as our approach was retrospective, the classification of the events was based on chart review rather than documented by the acting clinical team. However, the rates of events are unbiased as they are clearly documented by a change in care setting. Finally, for this outcome to be applied consistently across HaH programs requires the ability to link the HaH admission to an escalation outcome reliably. For HaH units that are not part of a hospital system, it may be more challenging to collect and retain such information.

Our findings suggest the use of unplanned escalations as a patient safety outcome that should be reported in all studies that describe HaH clinical outcomes. It also suggests that a reasonable benchmark may range from 6-7%. Further research should explore applicability of this measure across other HaH settings in other countries or rural settings. These findings prompt consideration of targeted safety nets, such as enhanced caregiving support and environmental modifications, to mitigate social and environmental risks, thereby making HaH a more viable option for older patients.³¹ Additional work has to be done to reduce escalation of older adults.

Acknowledgments

MGB: Internal departmental funds

No other sources of financial, material, and other support for the investigation, including salary support for each of the investigators.

Data Sharing

All authors had full access to all of the data (including statistical reports and tables) related to the study.

Conflicts of Interest

The authors have no conflict of interest.

DML: Biofourmis, royalties; Feminai, scientific advisor

Author Contributions

• All authors had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

• Study concept and design: MM, DML, SQK

• Acquisition, analysis, or interpretation of data: ET, CS, NS, TN, SW, CD, CEC, SS, LSY, MM, DML, SQK, CAY

• Drafting of the manuscript: MM, DML, SQK, CEC

• Critical revision of the manuscript for important intellectual content: MM,DML,SQK,CEC

• Statistical analysis: SS, MM, DML, SQK, CAY

• Administrative, technical, or material support: ET, CS, NS, TN, SW, CD, CEC, LSY

• Study supervision: MM, DML, SQK

Corresponding Author

David M. Levine: dmlevine@bwh.harvard.edu