Study design and participants
In order to reflect clinical practice and to ensure generalisability we conducted a pragmatic, randomised, multicentre parallel group trial with two arms between 1 November 2013 and 30 June 2015 . Four municipalities and four hospital emergency departments were involved, one in each municipality, covering a total of 150 GPs and 228,000 citizens in Southern Jutland, Denmark. All four municipalities had established HaH services, either in the patient’s home (Sønderborg and Haderslev) or in in the local nursing home (Tønder and Aabenraa).
The study was run through the research unit for Emergency Medicine at the University of Southern Denmark, Institute for Regional Research, Southern Centre, and was overseen by a trial steering committee. It was designed in accordance with the SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) 2013 Statement .
Patients with an acute medical condition that otherwise would require acute hospital in-patient care, were invited to participate in the study. They were identified by their GP or the municipal nurses who usually provided care for the patients.
We included patients aged 65 and over or 60 and over with significant comorbidity residing in one of the four municipalities. The patient’s own GP should be available the first two working days after inclusion. We excluded patients if they were assessed by someone other than their usual GP. Other exclusion criteria were if the patient was permanently residing in a nursing home, unable to give written consent, or had no relatives to give acting consent or if the capacity of the municipal care was fully utilised.
Referral to the study was made through the GPs in their weekday opening hours. The GP was consulted either by the patient or the municipal nurse on the patient’s behalf concerning an acute medical problem. The GP then assessed the situation. If the GP was unable to see the patient immediately, a municipal nurse could be asked to visit the patient within an hour, regardless of whether or not the patient was already receiving municipal care services. The nurse would then assess the patient and report back to the GP. If the GP found, that the patient was in a condition that would require acute hospital in-patient care the patient was eligible for inclusion in the study. All 150 GP’s were allowed to refer patients to the project.
The GP or nurse informed the patient about the study both orally and in writing through prepared information documents. If the patient (or a relative in case of a temporarily or permanently mentally incompetent patient) agreed to be included, the GP or nurse called the randomisation centre and secured written consent.
Randomisation and masking
The randomisation process was performed by an independent coordinating centre. The investigators, patients, GPs, and nurses could not influence to which group the patients were allocated. Participants were randomly assigned (1:1) to a hospital specialist HaH model or GP based HaH model with permuted blocks of 10, stratified in two groups, according to the municipal HaH model in the patient’s home (Sønderborg and Haderslev) or in local nursing homes (Tønder and Aabenraa). Allocations were prepared by an independent statistician and placed in serially numbered, opaque, sealed, tamper-evident envelopes by the independent coordinating centre. Treatment was determined by selecting the next randomisation envelope in sequence and was checked against a randomisation log. The correct and sequential use of envelopes as described in the protocol was strictly audited by the site research team and the independent coordinating centre. The result of the randomisation was communicated to the municipal HaH team, the GP or hospital specialist, and the patient. Treating clinicians could not be masked to the allocation. All analyses were done by investigators masked to treatment allocation.
For both groups, the HaH nursing care was performed by the municipally employed nurses. Before the recruitment of the patients, all municipal nurses completed a 1-day training programme and examination in assessment of the acute patient, including the use of the ADAPT triage system , in treatment and handling of the most frequently encountered acute conditions and in communication to other health professionals. An optional training programme in basic HaH service was offered to the all GP’s during two evening sessions, in which 94 GPs participated. Since the hospital doctors had not been involved in any former HaH programmes, eight physicians, all specialists in internal medicine, participated in the study and they all received similar training like the GPs.
All of the included patients were offered the same nursing care, regardless of whether or not the care took place in the patient’s home or at a nearby nursing home. In order to ensure equal treatment and care in the municipal HaH services, “minimum professional standard requirements” was defined, which included visits by the nurses up to eight times daily, administration of IV fluid and IV medicine, inhalation therapy, measurement of vital values, and triage.
The patients, who were randomised to the hospital specialist arm, were transported to the nearest hospital department and assessed by one of the eight participating specialists, who prescribed blood tests and imaging studies within 30 min after arrival. A detailed plan for the treatment and observation was decided within 4 h based on the results from the clinical examination and investigations. The plan was immediately sent electronically to the municipal HaH nurses. The hospital ensured that the treatment could be implemented immediately by transporting the patient to the HaH destination together with medical supplies, such as medicine and IV fluids, as prescribed and agreed with the HaH nurses. The specialist was responsible for the treatment, and one of the hospital specialists could be contacted directly by telephone within the next 48 h but was unable to visit the patient at home or at the nursing home. The GP was to be contacted for other medical issues not related to the HaH service.
The patients randomised to the GP arm were either seen in their own GP’s consultation or at home, depending on the situation. The GP examined the patient and performed a limited number of laboratory tests, such as C-reactive protein, B-glucose, hemoglobin concentration, and urine stix for urinary tract infection. Based on these findings, the GP prescribed a treatment and observation plan, which was immediately electronically available to the HaH nurses. The GP was responsible for the treatment prescribed and could be contacted by telephone during daytime. Outside working hours, the local GP on duty could be consulted. If the GP prescribed IV treatment, municipal services brought medicine and IV fluids to the patient.
The HaH service lasted up to 48 h after the inclusion. By this time, the municipal nurses and the physician caring for the patient had to decide whether the patient could return to usual GP care, including normal or increased municipal services, or should be admitted to hospital.
The participants were followed for 3 months after the inclusion time and subsequent hospital admissions and deaths were recorded.
The patients were tested with a range of performance tests at the time of inclusion and after 7 days: The 30-s chair-stand test as an indicator of lower body muscle strength and functional capacity (number of stands from chair in 30 s, range 0- stands/30 s) , the Morton Mobility Index (DEMMI) to measure mobility (15 items measures mobility and balance across the spectrum.
from bed-bound to independent mobility, score range 0–100) [23, 24], Orientation-Memory-Concentration (OMC) to evaluate cognitive performance (6 tests of memory, orientation and concentration, range 0–28) , and hand grip strength to provide a measure of the patient’s total strength (Jamar hand dynamometer, range 0–90 kg) [26, 27]. QALYs (quality-adjusted life years) were estimated using the self-completed EQ-5D instrument  and Danish preference weights (5 questions concerning mobility, self care, activity, pain and depression, weighted range 0–1). All of the tests were performed at the same time in the order: EQ-5D, DEMMI, OMC, chair-stand test, and grip strength. The completion of the tests was supervised by project assistants (nurses or physiotherapists) specifically trained for the assignment and not involved in the care of the patients. It was secured that the same assistant did not test the patient twice and there was no access to former test results. A high inter-rater reliability was secured by an experienced project physiotherapist, who instructed and trained all project assistants in the performance of the tests during a one-day course and controlled each assistant by supervision of their first performance and again during a control visit after 5 months. For the different tests (DEMMI, chair-stand and hand grip-test) good-excellent agreement was found.
The primary outcome was the number of patients with a hospital admission within 7 days after the inclusion. Secondary outcomes were number of patients with a hospital admission within 14,21, 30 and 90 days after the inclusion, median admission days among the patients admitted within 7 and 90 days after the inclusion, number of patients who died within 30 and 90 days after the inclusion. We also measured the mean changes in the mental and physical performance tests and patient perceived quality of life score from day 0 to day 7. Data concerning hospital admissions and death were obtained from the Danish National Patient Register and by reviews of the patient files.
All data collected for this study were entered directly into electronic predesigned forms by using computer tablets and stored in accordance with the Danish Data Protection Agency requirements.
The RCT was stratified into two layers, depending on the municipal HaH care. The analyses were performed at an accumulated level.
To calculate the required sample size, we estimated a 40% admission rate in the community arm on the basis of local audit data. We powered our study to detect at least a 25% absolute reduction in admissions in the hospital specialist arm compared to the GP arm, using the minimum improvement we thought would effect a change in practice. We estimated that this difference would require 82 patients per group, 164 patients total, which would allow the analysis of the study at an aggregated level assuming a two-tailed test of statistical significance with an α of 0.05 and power of 0.8.
The data was analysed according to intention-to-treat principles. We report baseline data descriptively by group and for simple comparison of categorical variables we used Fisher exact test and for continuous variables Kruskall-Wallis test. For the analysis of the outcomes we compared binary outcomes (including the primary outcome) using logistic regression and continuous outcome using linear regression with allocation group as a fixed effect and municipals as a random effect. All patients had complete data for the primary outcome and death. For the secondary outcomes multiple imputation method was using the variables age, gender, municipal, residence, home care and triage to replace the missing data for the following outcomes: DEMMI score, Grip Strength test, OMC and RSS test (33% missing) and EQD5 (44% missing), The statistical analyses were performed in STATA v. 14. The reporting of the results is in accordance with CONSORT 2010 statement.