Specific multimorbidity patterns modify the impact of an exercise intervention in older hospitalized adults

Background

Decreases in the fatality of most chronic conditions have led older individuals to live longer with multiple long-term conditions (i.e., multimorbidity). Multimorbidity is commonly defined as the co-occurrence of two or more chronic conditions in an individual. ¹ Up to 90% of older adults 60 years and over are multimorbid, ² and multimorbidity accounts for many of the years lived with functional decline and disability. ³ It is also related to numerous adverse outcomes such as disability, cognitive impairment, nursing home admissions, death and intense healthcare use,^3–5 and has implications beyond those generated by each of the diseases. ⁶

Unplanned hospital admissions triggered by older patients’ clinical complexity is one of the negative outcomes associated with multimorbidity. ⁷ Unplanned admissions represent over a third of all hospital admissions, lead to high care costs, create uncertainty for those responsible for planning and delivering services, and are distressing for patients and their families.^8,9 Unplanned admissions in older patients often represent sentinel episodes for accelerated health deterioration, and they often lead to iatrogenic disability and cognitive deterioration in the short and long-term following the hospitalization,^10,11 which in many cases could be prevented by means of individualized early interventions. Despite the acknowledged negative consequences of hospitalizations with the subsequent development of frailty and disability,^12–14 insufficient progress has been made on the implementation of intervention programs already during the hospitalization period. Yet, increasing evidence supports the effectiveness and feasibility of physical exercise interventions to prevent hospital-associated disability in older adults.^15–18 Multicomponent exercise programs targeting the four physical function domains (strength, balance, endurance and mobility) have been shown to be able to modify the functional and cognitive trajectories of hospitalized older adults. ¹⁹

It is well established that tailored interventions focusing on nutrition, exercise or management of polypharmacy are effective in the maintenance of good health in old age; however, recent studies have demonstrated that certain groups of participating individuals experience no significant improvements. Such individuals are commonly termed as non- or negative responders. ²⁰ Exercise non-response appears to be mitigated by adjustments in training volume, duration, and intensity.^21–23 It is therefore a challenge, but essential, to advance towards the personalization of exercise interventions through a better stratification and targeting of potentially eligible patients. Along these lines, the identification of negative as well as positive responders to previously evaluated interventions will be key to optimize available resources.

Patients affected by specific patterns of multimorbidity have been shown to differ in their risk of developing functional decline and disability over time^6,24 but, as far as we know, there is no study that has evaluated the influence of such patterns of chronic diseases in modulating the effectiveness of an individualized exercise program aimed at delaying or reversing functional decline in hospitalized older adults. One may hypothesize that multimorbid patients affected by cardiovascular and metabolic conditions could benefit more from this type of interventions, given the responsiveness of these diseases to aerobic exercise, endurance and weight loss. Thus, the aim of this study was to identify groups of geriatric patients characterized by their patterns of multimorbidity who could benefit most from an individual exercise intervention during hospitalization.

Study design and participants

This study is a secondary analysis of a randomized controlled trial (RCT; NCT02300896) ¹⁹ conducted in the Acute Care Unit (ACE) of the Department of Geriatrics in a tertiary public hospital (Hospital Universitario de Navarra, Spain). This department has 40 allocated beds with a staff of eight geriatricians (distributed in the ACE unit, orthogeriatrics and outpatient consultations). Admissions to the ACE unit are mainly from the Accident and Emergency Department, with heart failure, pulmonary, and infectious diseases being the main causes of admission.

Acutely hospitalized patients who met the inclusion criteria were randomly assigned to the intervention or control (usual care) group within the first 48 hours of admission. In total, 370 participants were recruited for the original study. Details about randomization and allocation procedures and sample size considerations have been published elsewhere. ²⁵ Inclusion criteria were age >75 years, Barthel Index score >59 points, being able to walk (with/without assistance), and being able to communicate and collaborate with the research team. Exclusion criteria included expected length of stay <6 days, very severe cognitive decline, terminal illness, uncontrolled arrhythmias, acute pulmonary embolism and myocardial infarction, and extremity bone fracture in the past three months. Out of the 370 participants, 64 with less than two chronic diseases at the initial assessment were excluded from the analyses.

Intervention

Briefly, the usual-care group received usual hospital care, which included physical rehabilitation when needed. In addition to the standard care, an exercise-training program was delivered in two daily sessions (morning and evening) of 20 minutes’ duration during 5-7 consecutive days for the intervention group. The morning session included individualized progressive resistance, balance, and walking training exercises and was supervised by a physiotherapist with a background in exercise physiology (>10 years of experience). The evening session consisted of functional unsupervised exercises using light loads (0.5-1 kg anklets and handgrip ball), such as knee extension/flexion, hip abduction, and daily walking in the corridor of the ACE unit with a duration based on the clinical physical exercise guide “Vivifrail” (http://vivifrail.com/). ²⁶

The study followed the principles of the Declaration of Helsinki and was approved by the Hospital Universitario de Navarra Research Ethics Committee (Pyto 23/2014). All patients or their legal representatives provided written consent.

Chronic diseases and multimorbidity assessment

Demographic, functional, cognitive and multimorbidity data were recorded during the acute hospital admission and retrospectively completed using the electronic medical records from primary care. Initially, chronic diseases were coded according to the International Classification of Diseases, 10^th revision (ICD10) and later classified according to a consensus list of 60 chronic disease categories developed by Calderón-Larrañaga et al in the Swedish National study of Aging and Care in Kungsholmen (SNAC-K). ²⁷ Each chronic disease was included as an individual binary variable. Only chronic diseases with a prevalence ≥2% in the study population were included for final analyses, i.e., 38 diseases in total.

Clinical outcomes

The main outcomes examined in this secondary analysis were functional, cognitive and affective status. More specifically, the following scales were included: the Barthel index assessing performance in activities of daily living (ranging from 0 [totally dependent] to 100 [completely independent], ²⁸ the Short Physical Performance Battery (SPPB) that combines balance, gait velocity, and leg strength in a single score on a 0 (worst) to 12 (best) scale, ²⁹ the handgrip strength (dominant hand), ³⁰ the Mini-Mental State Examination (30-point questionnaire; scale from 0 [worst] to 30 [best]), ³¹ mood status according to the 15-item Yesavage Geriatric Depression Scale (Spanish version; scale from 0 [best] to 15 [worst]), ³² and the visual analog scale of the EuroQol–5 Dimension (EQ-5D) questionnaire for quality of life (QoL) assessment (Spanish version; scale from 0 [worst health state imaginable] to 100 [best health state imaginable]). ³³

Statistical analysis

Homogeneous groups of patients were first identified according to their commonly coexisting chronic diseases. A dimension-reduction technique on the dataset of 38 chronic diseases was applied through multiple correspondence analysis (MCA) for binary variables. This method reduced the size of the dataset while maintaining the complexity of the original data. The Karlis-Saporta-Spinaki rule was applied in order to select the appropriate number of dimensions to preserve, considering the eigenvalues of the MCA and the number of variables and individuals in the dataset. This approach accounts for the sampling variability, which can lead to the selection of dimensions that are in fact not statistically significant. ³⁴

Using the reduced database, multimorbidity patterns were determined through a fuzzy c-means soft clustering algorithm. The fuzzy c-means algorithm estimates c cluster centers (similar to k-means) but with fuzziness, so that individuals may belong to more than one cluster. The use of a fuzzy cluster analysis over a hard cluster analysis helps to better handle the stochastic nature of some disease associations, the potential noise stemming from the measurement (e.g., disease assessment), and the variance due to between-individual differences.^35,36 Since clustering algorithms are unsupervised techniques, the model fitting is traditionally assessed through cost functions that depend on both the dataset and the clustering parameters, and are denoted as validation indices. In our analysis, the following indices were included (Appendix 1): Fukuyama index, Xie-Beni index, Partition coefficient index, Partition entropy index and Calinski-Harabasz index. ³⁷ The final number of patterns was determined by combining both the analytical and clinical approaches. The consistency and clinical relevance of the final solution was evaluated in the context of previous literature as well as within the research team (2 geriatricians, 3 physiotherapists, 1 epidemiologist and 2 statisticians).

The patterns were described using three metrics: 1) the observed prevalence of a chronic disease in a specific pattern; 2) observed/expected ratios (O/E ratios) calculated by dividing the prevalence of a chronic disease in a specific pattern by the prevalence of the same chronic disease in the entire study population; 3) exclusivity determined by dividing the number of individuals with a chronic disease in a specific pattern by the number of individuals with the same chronic disease in the entire study population. A threshold of 2 for the O/E ratio was set in order for a disease to be considered overrepresented in a pattern. An exclusivity threshold of 25% was established as a complementary metric to evaluate the association of a given chronic disease with a pattern.^35,36

The patterns were subsequently characterized by demographic and clinical variables not included in the clustering process. For these analyses, participants were forced into the pattern they were more likely to belong to. Demographic and clinical outcome data were summarized using means and standard deviations for continuous variables and frequencies and percentages for categorical variables, and were presented by multimorbidity pattern and intervention group (i.e., exercise and usual-care).

Next, we assessed whether the impact of the intervention differed across patient groups characterized by their patterns of multimorbidity. For each clinical outcome variable, pattern-stratified linear models were fitted (i.e., between-intervention group differences), with post-intervention values as the response variable, intervention group as the explanatory variable, and adjusting by sex, age and baseline outcome value. Moreover, a linear model was fitted using the whole sample (i.e., between-pattern differences), with post intervention values as the response variable, intervention group, pattern and their interaction as explanatory variables, and adjusting by sex, age and baseline outcome value.

All analyses were carried out using R version 4.1.0. The significance level was set at α=0.05.