Study design and setting

The study had a single-centre, retrospective cohort design (Fig. 1). Different health outcomes were compared among patients who were admitted before (control group) and after the implementation of a geriatric co-management project on 1 February 2018 (intervention group). We retrospectively collected the data of patients admitted between 1 January 2016 and 1 August 2020. The study was performed in the Cardiac Department of the Jeroen Bosch Hospital, a large teaching hospital in the Netherlands. The STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines were used in the design and reporting of this study.

Fig. 1

Study design and patient identification

Inclusion criteria

All patients aged 85 and over who had an unplanned hospital admission to the cardiac ward for more than 24 h were eligible for inclusion. As the prevalence of frailty increases with age [14], an age cut-off of 85 years was used to include the highest number of frail patients and still make the project practically feasible.

Intervention: standard geriatric co-management

All eligible patients in the intervention group underwent a CGA (Fig. 2) within 48 h of admission performed by a geriatric physician and a geriatric nurse (the regular geriatric consultation team in our hospital). After the initial assessment, the geriatrician and treating cardiologist discussed tailored care for that patient. The geriatric team was involved in the patient’s further treatment, based on that patient’s needs. Cardiac nurses and a geriatric nurse met weekly to discuss admitted patients and their care.

Fig. 2

Comprehensive geriatric assessment

The control group received usual care. If their treating physician requested a geriatric consultation during their admission, patients underwent the same CGA.

Data collection

Electronic patient records were searched for eligible patients using the software program CTcue version 2.1.10 [15]. This software collects clinical data, using the previously mentioned inclusion criteria and time frame as search criteria.

Two researchers (R.R. and J.S.) manually searched medical records for additional data that could not be obtained with the software (e.g. residency). Patients who were falsely identified by the software were also excluded (see Fig. 1).

Baseline characteristics

The following baseline characteristics were collected: general demographics, medical history, admission diagnosis, previous hospital admission in the last 12 months, and information obtained with the CGA (e.g. the number of medications used, number of falls over the last 6 months).

Outcomes

The primary outcome was all-cause hospital readmission rate at 1, 6, and 12 months after the initial admission. Multiple time points were used to account for differences in follow-up time and multiple hospitalisations. The secondary outcomes were LOS, all-cause in-hospital mortality, all-cause mortality within 3 months of admission, number of interprofessional consultations, change in residency, or discharge to a rehabilitation facility. The number of complications, such as falls or delirium, were also recorded.

Statistical analysis

Baseline characteristics were compared between groups using statistical tests (χ2-test, Fisher’s exact test, two-sample t-test and Mann-Whitney U test, where appropriate). All analyses were performed with an intention-to-treat approach to prevent selection bias. The primary outcome readmission rates and the incidence rate ratio were calculated using a Poisson regression model. These models were adjusted for baseline differences and risk factors for readmission.

For normally distributed secondary outcome variables odds ratios were calculated using logistic regression. The non-normal outcome variables, number of consultations and LOS, were first analysed using a Mann-Whitney U test. For the multivariate analysis of the number of consultations, a Poisson regression model was chosen. For the LOS data, a normal distribution was achieved through a square root transformation so a multiple linear regression model could be used. Furthermore, we attempted to reduce the influence of extreme outliers by truncating the data at 30 days. This means that the duration of hospitalisation for all patients hospitalised for longer than 30 days was changed to 30 days. All the multivariate models contained the variables age, number of comorbidities, and any baseline differences.

The data were analysed with the statistical package for social sciences (SPSS) version 25 [16]. Statistical inference was based on a p-value of 0.05.

Sample size calculation

A power calculation was performed based on the incidence of our most important outcomes rehospitalisation, mortality, and LOS (β = 15%, α = 5%, two-sided test) [12, 17, 18]. For a clinically significant reduction in any of these outcomes, 1150 patients were needed.

Ethical considerations

The regional Ethics Review Board (METC Brabant/20.435, #NW2020-78) deemed this study to fall outside the scope of the Dutch Law on Medical Research (WMO). The study was conducted in accordance with the Dutch Medical Treatment Contracts Act (WGBO) article 458 and the principles of the World Medical Association Declaration of Helsinki (2013). The physical and psychological integrity of the patients were not harmed in any way.