Study groups and design
This was a retrospective CCA of all first-time HaH-HA admissions issued from the emergency room department (ER) in the Hospital Clínic of Barcelona (HCB) among non-surgical patients between October 31, 2017, and November 1, 2018. The direct costs and outcomes of HaH-HA patients were compared with a 1:1 matched comparator group of conventional hospitalizations in our center. Patients under the modality HaH-ED were excluded from the analysis.
Candidates to HaH-HA were screened in the ER by trained professionals of the HaH team. Individuals were eligible for HaH-HA if: they were aged 18 years or older, lived in their house within the catchment area, had a formal or informal caretaker (including relatives) available 24 hours per day, had a phone at home and signed the informed consent to be hospitalized at home. We considered all medical conditions.
The comparator group (controls) was built from non-surgical patients admitted for conventional hospitalization from the ER within the same period. We paired HaH-HA patients with control patients 1:1 using a propensity score matching (PSM) [20, 21] and genetic-matching technique [22]. For matching purposes, we took into account two sets of matching variables to ensure patients’ comparability regarding both baseline characteristics (i.e., before admission) and hospitalization characteristics.
The first set of matching variables included age, gender, number of admissions in the previous year, patient’s healthcare costs across the health system in the previous year, and health risk based on the adjusted morbidity groups (AMG) index [23]. The AMG is a summary measure of morbidity that considers a weighted sum of all chronic and relevant acute conditions from all diagnostic group codes of the International Classification of Diseases, clinical modification (ICD-10-CM). The AMG can be used as a numerical index or as population-based risk groups, defined according to percentile thresholds for the distribution of the AMG index across the entire population of Catalonia. Both the index and the risk groups have shown a good correlation with relevant health outcomes and the use of healthcare resources [24, 25].
The second set of variables for paring HaH-HA and control patients included relevant characteristics of the hospitalization episode, such as the main diagnosis at discharge based on the ICD-10-CM categories and the case mix index (CMI). The CMI summarizes the severity and complexity of the main diagnosis and health events occurring during the hospital stay.
Characteristics of home and conventional hospitalizations
The HaH-HA group followed the standard of care for HaH at HCB, which has been extensively reported elsewhere [7]. Briefly, a patient admitted to HaH-HA is assessed in person daily by the HaH team, which consists of either a nurse or a nurse and physician (at physician’s discretion) with remote access to the patient’s electronic record. Interventions available at home include regular tests (e.g., blood and microbiology tests, clinical ultrasound, electrocardiogram), most of the intravenous and nebulized treatments, and oxygen therapy. A pathway for elective transfer back to the hospital (e.g., for additional tests not available at home) and ER transfer in case of clinical deterioration are also available.
The control group followed the usual care for in-house hospitalizations; patients were assigned to a hospital bed within the corresponding service according to the primary diagnosis and followed up by the medical and nurse staff of the corresponding ward or service.
Upon discharge, patients in the two groups were transferred to the corresponding primary care teams, with access to electronic health records. However, the HaH team shares responsibilities with the primary care team during the transitional care period until 30 days after discharge.
Outcomes and costs
The CCA included health outcomes and direct costs [26]. Health outcomes included length of hospital stay, 30-day mortality, and all-cause hospital admissions and visits to the ER within the 30 days following discharge. In patients admitted to HaH-HA, we also collected the patient experience by administering a 9-item satisfaction questionnaire [7] on discharge.
Costs were estimated using an analytical accounting approach [27]. Direct costs included honoraria of staff professionals, pharmacological and non-pharmacological therapy, consumables, testing and procedures, transportation, catering, and structural costs. We also considered healthcare expenditure associated with any resource use of the healthcare system during the 30 days following discharge.
The two data sources used for the study were: the SAP Health Information System at HCB and the Catalan Health Surveillance System (CHSS) for analysis of the acute episode and calculations after discharge, respectively.
Data analysis
Health outcomes and costs were described by the number and percentage over available data for categorical variables and mean and standard deviation (SD), or median and interquartile range (IQR, defined by the 25th and 75th percentiles), as appropriate. The matching parameters were tuned to enhance the covariate balancing, as follows: caliper: 0.2, function: logit, replace: FALSE, ratio: 1:1, matching method: Genetic Matching. Genetic Matching uses an optimization algorithm based on “GENetic Optimization Using Derivatives (GENOUD)” [28] to check and improve covariate balance iteratively, and it is a generalization of propensity score and Mahalanobis distance [29]. The matching was assessed by the Mahalanobis distance, Rubin’s B (the absolute standardized difference of the means of the linear index of the propensity score in the HaH-HA and Controls) and Rubin’s R [30] (the ratio of HaH-HA to Controls variances of the propensity score index) metrics. Quality of comparability between HaH-HA and Controls after PSM was considered acceptable if Rubin’s B was less than 0.25 and Rubin’s R was between 0.5 and 2. Unpaired Student T tests, Mann-Whitney, and Chi-squared tests comparing HaH-HA with Controls were used to assess changes in the costs and clinical outcomes. Data analyses were conducted using R [31], version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria). The threshold for significance was set at a two-sided alpha value of 0.05.