Study design and population
An exploratory cross-sectional study based on the Minimum Basic Data Set from Acute-care Hospitals (MBDSHA) was performed. The MBDSHA included 24 public hospitals in the Catalonian territory and HaH contacts for 2014. A contact was every time a patient received any kind of treatment from commencement to finalization. The same individual could present more than one contact during the study period. Programmed contacts with a specific diagnosis according to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) were included (n=9,805)[15]. Those who belonged to a diagnostic group with an insufficient number of contacts for robustness (n=95, 1.0%), lacked an identification number (n=49, 0.5%), had no established age (n=2, <0.1%), and whose dates of admission and discharge were wrongly codified (58, 0.6%) were excluded. Finally, a total sample of 9,601 HaH contacts was considered for analysis: 7,214 (75.1%) admission avoidance and 2,387 (24.9%) early assisted discharge.
Main outcomes
Three indicators were established as the main outcomes for the exploratory analysis. These indicators were selected both based on their relevance to assess the performance of the HaH modalities, as well as based on the recommendations for the assessment of HaH from previous systematic reviews [6, 7].
Readmission: consecutive HaH contacts, either HaH or conventional hospitalization (CH), in a period ≤ 30 days provided they were related to the first contact.
Mortality prior to discharge: HaH contacts in which the patient status at discharge was death.
Mean length of stay: for the admission avoidance modality, duration (days) of HaH contact from date of program admission to finalization. For early assisted discharge, duration was a combination of the immediately preceding contact of CH and the HaH one, taking it from the CH contact date of admission to HaH finalization.
Contact characteristics
Sex: male and female.
Age (years): considered a continuous variable.
Diagnosis: categorized according to the ICD-9-CM chapters [15].
Comorbidity according to the Charlson Comorbidity Index (CCI) [16, 17]. The CCI, considered to be an objective measurement of an individual’s general state of health, is employed to predict mortality in terms of the patient’s comorbidity. General comorbidity is calculated through the weight assigned to the presence of each of the 19 conditions making up the index. The results are classified as: 0 or 1, 2, and ≥3.
Type of hospital (according to the portfolio of services offered in the hospital itself, irrespective of the patient’s territorial assignment): reference hospital, district hospital, high-technology general hospital, and high-resolution hospital.
Number of contacts per patient (number of HaH episodes): 1 or more than 1.
Data analysis
A descriptive analysis of the characteristics of the contacts according to HaH modality was performed and results were evaluated at the bivariable level. To compare possible differences in the explicative variables between the two modalities, chi-square and Fisher’s exact tests were employed for the categorical variables, and the Mann Whitney U test for age and mean length of stay due to the lack of normality of their distributions. The selected outcomes were then calculated for each modality, and the association between the contacts’ characteristics and each of these indicators was assessed with multivariable models. Due to the characteristics of the variables considered as outcomes, logistic regression models were fitted for readmission and mortality, and Poisson models for mean length of stay. From these results, the β coefficients and their respective 95% confidence intervals (95% CI) were obtained, and their exponential was presented to aid interpretation (the Odds Ratio in logistic regression models and ratios for Poisson models). All multivariable models were done individually for the two HaH modalities and adjusted for sex, age, comorbidity (CCI), and type of hospital. The absence of relevant interactions between explanatory variables was tested using a Chunk test. All analyses were carried out with STATA v.14® [18] software and statistical significance set at α=0.05.