Overview of study design
The current study combined quantitative and qualitative research methodologies. The quantitative study was a retrospective observational analysis of the characteristics of HaH recipients and health results of hospitalizations occurred between January 1, 2015, and December 31, 2019. To select HaH episodes in the admission avoidance modality, we selected patients with unplanned hospitalizations and less than 24 hours between hospital entry and HaH registration. For the qualitative assessment, we conducted focus groups and surveys16 with a panel of experts in HaH to interpret the results of the quantitative analysis and generate recommendations for CQI.
The quantitative study was reported according to the STROBE17 guidelines for observational studies, and the qualitative analysis was reported according to the SRQR18 guidelines.
Population and data sources
All data used in the quantitative analysis were retrieved from the Catalan Health Surveillance System (CHSS).19 Since 2011, the CHSS has collected detailed information on the utilization of healthcare resources by the entire population of Catalonia. The CHSS assembles information on the use of healthcare resources across healthcare tiers, drugs, and other billable healthcare costs, such as non-urgent medical transportation, outpatient rehabilitation, respiratory therapies, and dialysis. We screened the CHSS for all episodes of HaH reported in Catalonia during the study period.
The same database was used to create a retrospectively-matched control group of contemporary conventional hospitalizations. The control group was created using a 1-to-1 propensity score matching (PSM)20,21 and Genetic Matching22 technique based on GENetic Optimization Using Derivatives (GENOUD)23 algorithm to check and improve covariate balance iteratively. To ensure the comparability of the matched episodes, we screened contemporary admissions within the same hospital with identical Medicare Diagnosis Related Group12 category. In addition, the patient’s baseline characteristics were characterized and matched using data on demographics (i.e., age and gender), utilization of healthcare resources during the previous year (i.e., hospital admissions, emergency room visits, number of pharmacological prescriptions and the total healthcare expenditure), clinical and social risk factors (i.e., the morbidity burden, using the Adjusted Morbidity Groups24,25 (AMG) score, and the presence of active diagnoses related with health-related social needs26).
The overall comparability of the matched group was assessed using the Mahalanobis distance,27 and Rubin’s B and Rubin’s R metrics. Comparability after PSM was considered acceptable if Rubin’s B was less than 0.25 and Rubin’s R was between 0.5 and 2.28
Variables and Outcomes
The baseline characteristics of study patients (i.e., before admission) included age, sex, morbidity burden measured using the AMG score, hospitalizations, emergency room admissions, and expenditure within the past year. Information regarding the HaH episode included the length of stay (LoS) and the complexity of hospitalization, measured using two case-mix tools: the Case Mix Index- APR-DRG v35 (CMI),29 broadly used for payment purposes, and the Queralt index,30,31 recently developed by the Catalan Institute of Health and showing higher performance for predicting general hospitalization endpoints. Readmissions in in-hospital settings and visits to the emergency room without the need to discontinue HaH were also considered clinical outcomes of the HaH episode.
Besides the baseline and episode characteristics, we gathered information regarding healthcare expenditure, hospitalizations, and visits to the emergency room within the 30-days after discharge. Expenditure information was obtained from reimbursements by the Catalan Health Service,32 since no operational costs33,34 were available for the entire study group. HaH delivery is reimbursed as a specific healthcare service, with case costs estimated based on the APR-DRG categories of the main diagnostic. Other relevant outcomes included mortality, during the hospitalization and 30 days after discharge.
Statistical analysis
Before the analysis, we removed from the databases all the incomplete records, duplicate entries, and outliers with unrepresentative baseline characteristics or anomalous LoS with a Z-value greater than |3|.
Since p-values tend to drop in large population-based samples, yielding significant differences in most comparisons,35 we used effect size measures to compare the baseline characteristics of the matched HaH individuals with their respective controls to establish the impact of the intervention. Cohen’s D test was used to determine the effect size in numerical variables; the magnitude of the difference was assessed according to the following ranges: weak (< 0.20), small (0.2–0.5), moderate (0.5–0.8), large (0.8–1.3), and very large (> 1.3). Cohen’s W test was used for categorical variables, with the following ranges used to assess the magnitude of difference: weak (< 0.10), small (0.1–0.3), moderate (0.3–0.5), and large (> 50). We computed 1,000 bootstrap replicates in both scenarios to generate the 95% CI.
Categorical variables were summarized as absolute values and percentages, whereas continuous variables were described by the mean and the standard deviation or the median and the interquartile range as appropriate.
To analyze heterogeneity among hospitals regarding the patient profile, we described the age and the morbidity burden (measured using the AMG index) of HaH patients in each center. We also assessed the inclusion bias of each center by measuring the difference in mean age and AMG index between HaH and conventional hospitalizations admitted for the exact cause within the same hospital. Heterogeneity was also assessed regarding the LoS, the complexity of the episode and the repetition rate among HaH patients. In addition to the descriptive analysis, we addressed heterogeneity by conducting an ancillary cluster analysis using the K-means36 algorithm, incorporating information on the category of the hospital based on the number of hospital beds and their role in their corresponding health district. The average silhouette37 method was used to determine the optimal number of clusters.
All the data analyses were performed using R,38 version 4.1.1 (R Foundation for Statistical Computing, Vienna, Austria).
Qualitative assessment
The qualitative study, which followed a grounded theory approach, included two focus group sessions with HaH experts. The first session aimed at interpreting the results obtained in the quantitative analysis described above, whereas the second session sought to discuss the efficiency and value generation of HaH (considering the heterogeneities and challenges of the service) and providing recommendations of core KPIs for CQI of HaH delivery after service adoption. The second session was preceded by the administration of a questionnaire (Supplementary material S1) for assessing the consensus strength. Experts were also provided with the 2020 consensus document aiming at regional standardization of HaH.13
The panel of 7 experts included 1-to-2 representatives of the most relevant organizations in implementing or assessing HaH services in Catalonia: two members from the Catalan-Balearic Society of Hospital at Home,39 two staff members from the Catalan Health Service,40 one staff member from the Health Quality and Assessment Agency of Catalonia (AQuAS),41 and two HaH experts from the local JADECARE team. Four out of the seven experts were clinical leaders of different HaH programs. A qualitative research and service design specialist was recruited as a facilitator for planning and leading the expert panel discussions. An extended description of the methodological details is provided in the online Supplementary material S1.