Background Owing to detrimental hazards and substantial healthcare burden and costs, hospitalisation of older people has become a major focus. Frailty has increasingly been recognised as an important predictor of hospitalisation. This study aims to identify studies on physical frailty as a predictor of hospitalisation risks and to pool the risk estimates among community-dwelling older people.
Methods A systematic literature search was performed in August 2015 using five databases: EMBASE, MEDLINE, CINAHL, PsycINFO and the Cochrane Library for prospective studies examining physical frailty as a predictor of hospitalisation published in 2000 or later. OR and HR were combined to synthesise pooled effect measures using fixed-effects models. The included studies were assessed for heterogeneity, methodological quality and publication bias. Subgroup analysis and meta-regression analysis were conducted to examine study characteristics in relation to the hospitalisation risks.
Results Of the 4620 studies identified by the systematic review, 13 studies with average follow-up period of 3.1 years were selected. Frailty and prefrailty were significantly associated with higher hospitalisation risks among 10 studies with OR (pooled OR=1.90, 95% CI 1.74–2.07, p<0.00001; pooled OR=1.26, 95% CI 1.18–1.33, p<0.00001, respectively) and 3 studies with HR (pooled HR=1.30, 95% CI 1.12–1.52, p=0.0007; pooled HR=1.13, 95% CI 1.04–1.24, p=0.005, respectively). Heterogeneity was low to moderate. No publication bias was detected. The studies with older populations and unadjusted outcome measures were associated with higher hospitalisation risks in the subgroup analysis.
Conclusions This systematic review and meta-analysis demonstrated physical frailty is a significant predictor of hospitalisation among community-dwelling older people. Hospitalisation can potentially be reduced by treating or preventing frailty.
- SYSTEMATIC REVIEWS
Statistics from Altmetric.com
Older people are at a high risk of hospitalisation and the risk becomes higher as they age.1 The number of hospitalisations of older people has been steadily increasing in many countries2 and can increase further as the number and proportion of older people increase. In the USA, the cost of hospitalisation among Medicare beneficiaries is substantial, accounts for approximately half of all Medicare fee-for-service expenditure.3 Some treatments and interventions are only available at hospitals, and hospitalisation is often necessary to treat acutely ill older people with complex medical problems. However, because of its high healthcare burden and costs as well as the hazards of the hospitalisation, including disruption of care, functional decline due to prolonged bed rest, iatrogenic infections, falls, delirium, adverse drug reactions and exposure to unfamiliar environments, it has been a major focus of interest for healthcare providers and policymakers to prevent the hospitalisation of older people.2 ,4
Frailty has increasingly been recognised as one of the important predictors of hospitalisation. Frailty, a geriatric syndrome which has been receiving recent research attention, is characterised by a decline in physiological reserves in multiple systems and an increased vulnerability to adverse health outcomes, such as falls, disability and death, due to age-related accumulated deficits.5–7 Frailty has been also shown to be associated with negative psychological consequences, including depression,8 cognitive impairment9 and poor quality of life.10 ,11
Although there has been an increasing volume of frailty research in the literature, there has been no international consensus reached on how to operationalise frailty. Among various definitions and criteria proposed,5 the one developed by Fried and colleagues from the Cardiovascular Health Study (CHS) is most frequently used.6 They defined frailty as having three or more, and prefrailty as having one or two, of five components of physical phenotypes: (1) unintentional weight loss, (2) self-reported exhaustion, (3) weakness, (4) slow walking speed and (5) low physical activity. Shortly afterwards, the Study of Osteoporotic Fractures (SOF) criteria were proposed as a simpler version of the CHS criteria, consisting of three physical components: (1) intentional or unintentional weight loss >5% in the past year, (2) inability to rise from a chair five consecutive times without using the arms and (3) self-perceived reduced energy level.12 It may well be expected that frail older people with these negative health conditions or traits are more prone to hospitalisation compared with non-frail individuals. One study found comorbidities, prior history of hospitalisation, six or more primary care visits, advanced age and unmarried status were independent risk factors for hospitalisation,13 some of which are associated with frailty.5 ,6 ,14 In fact, multiple prospective studies have shown significant associations between frailty and a higher risk of hospitalisation; however, some studies did not show this association.15 ,16
Only one systematic review was found in the literature examining frailty and hospitalisation.17 This paper reviewed studies from 1990 to 2010 on associations between various geriatric syndromes and the risk of hospitalisation.17 The authors identified six articles on frailty and hospitalisation risks among elderly populations in the community.6 ,15 ,18–21 However, it should be noted that some important studies were not included22 ,23 and that a meta-analysis was not reported.17 In addition, it is expected that more related studies have been published since 2010 given that this is a rapidly progressing research field. Therefore, the current study aims to conduct (1) a systematic review to identify studies investigating prospective associations between physical frailty and future hospitalisation risks and (2) a meta-analysis to synthesise pooled evidence of hospitalisation risk according to physical frailty among community-dwelling older people. Given the conflicting findings from previous studies, the current systematic review and meta-analysis study providing pooled risk estimates will further increase our understanding of frailty as an important risk factor of hospitalisation as well as a possible screening tool to identify the elderly at risk of hospitalisation.
Data sources and search strategy
A systematic literature search was performed in August 2015 by a US-trained clinician researcher (GK) board certified in Internal Medicine and Geriatric Medicine with experiences in inpatient and outpatient settings based on a protocol developed in accordance with Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA)24 and Meta-analysis of Observational Studies in Epidemiology (MOOSE) statements.25 Five electronic databases, EMBASE, MEDLINE, CINAHL Plus, PsycINFO and the Cochrane Library, were used without language restriction for studies published from 2000 through current, using an explosion function if available. The search strategy was ((Hospitalization (Medical Subject Heading (MeSH))) OR (Hospital Admission (MeSH)) OR (Hospital(s) (MeSH)) OR (Patient Admission (MeSH)) OR (Patient Readmission (MeSH)) OR (Inpatient(s) (MeSH)) OR (Hospital Patient (MeSH)) OR (Hospitalized Patient (MeSH)) OR (Hospital Utilization (MeSH)) OR (Health Resource Utilization (MeSH)) OR (Health Care Utilization (MeSH)) OR (Patient(s) (MeSH)) OR (Hospitalization*) OR (Hospitalisation*) OR (Admission*) OR (Readmission*) OR (Hospital*) OR (Utilization*) OR (Utilisation*) OR (Inpatient*)) AND ((Frailty Syndrome (MeSH)) OR (frailty)). Reference lists of included and relevant articles were also manually searched.
This systematic review and meta-analysis was limited to prospective studies since cross-sectional or retrospective studies are subject to some potential biases and not capable of detecting temporal associations, in this case, between frailty as a predictor and hospitalisation as an outcome. The inclusion criteria were as follows:
Prospective study design.
Mean age of 65 years or older.
Frailty defined by original or modified versions of validated frailty criteria based on physical components.
OR, risk ratio (RR) or HR provided as a risk measure or computable from available data.
The exclusion criteria were as follows:
Frailty defined by surrogate measures, such as walking speed, or only by frailty components.
Frailty defined by multidimensional criteria or definitions, such as ones including cognitive, psychological and social factors.
Selected samples with certain diseases, such as heart failure or Parkinson's disease or institutionalised populations.
Poster presentations, dissertations, randomised controlled trials or review articles.
When multiple eligible studies used data from the same sample or cohort, the study with the largest number of individuals was included. When multiple physical frailty criteria were used, the data according to CHS criteria, which have been used most frequently in the literature, or its modified versions were included.
Data were extracted from the eligible studies using a standardised data collection sheet which included first author, publication year, location (country), sample size, proportion of female individuals, age (mean, range or age criterion for inclusion), frailty criteria, effect measure and follow-up period. Some studies did not analyse entire cohorts but subsamples for associations between frailty and hospitalisation; therefore, it was attempted to obtain the sample size, female proportion and age from the sample actually used for the analysis of interest. If these data were not available, the data from the entire cohort were substituted. When one study used different types of physical frailty criteria, the data based on the CHS criteria were included, if available. When different lengths of follow-up periods were used, the data from the follow-up length closest to the mean of the rest of the included studies were used.
Methodological quality assessment
All eligible studies were assessed for their methodological quality using the Newcastle-Ottawa scale for cohort studies.26 This nine-item scale covers selection, compatibility and outcome domains of cohort studies. A study was considered to be of good quality to be included in the meta-analysis if five or more of the nine items were met.
All analyses were performed using Review Manager 5 (V.5.2, The Cochrane Collaboration, Copenhagen, Denmark), IBM SPSS Statistics (V.22, IBM Corporation, New York, USA), Comprehensive Meta-Analysis (V.3.3, Biostat, New Jersey, USA) and StatsDirect (V.2.8, StatsDirect, Cheshire, UK).
The OR, RR and HR with 95% CI of hospitalisation risk for frailty and prefrailty compared with non-frailty/robustness were extracted directly from the included studies, or the OR was calculated from the numbers presented in the studies using a univariate logistic regression model. No study reported a RR. Adjusted OR and HR were preferred for meta-analyses when available, otherwise unadjusted ones were extracted.
The OR and HR were transformed by calculating their natural logarithms. The SEs of the log-transformed OR and HR were calculated by dividing the difference between log-transformed upper and lower 95% CI limits by 3.92. These numbers were used to calculate pooled estimates of OR and HR, 95% CI and p values, using a random-effects model if high heterogeneity was detected by using I2 statistic, and using a fixed-effects model otherwise, with the generic inverse variance method. Heterogeneity across the included studies was examined using Cochran's Q statistic. The magnitude of the heterogeneity was examined using I2 statistic, and I2 value of 25%, 50% and 75% were considered as low, moderate and high heterogeneity, respectively.27 Publication bias was examined by a visual inspection of the funnel plots and using Begg-Mazumdar's and Egger's tests.28 ,29
Subgroup analyses were performed to explore potential factors affecting hospitalisation risks by frailty. The factors considered were location (Europe vs USA), sample size (≥4000 vs <4000), female proportion (≥60% vs <60%), mean age (≥75 vs <75), frailty criteria (CHS vs SOF), follow-up period (>2 years vs ≤2 years, >1 year vs ≤1 year) and adjustment for outcome (adjusted vs unadjusted). A random-effects meta-regression analysis was also conducted to examine the study characteristics (sample size, female proportion of a cohort, mean age, follow-up period and Newcastle-Ottawa scale score, all as a continuous variable) for potential moderator effects on the hospitalisation risks by frailty.
Figure 1 is a flow diagram presenting the literature search and study selection. The systematic review using five electronic databases identified 4619 studies and the manual search found 1 relevant study. Of these 4620 studies, 1514 duplicate studies were excluded, and then 3081 studies were excluded by title and abstract review, leaving 25 studies for full-text review. Additional 12 studies were excluded for the following reasons: no effect measures of hospitalisation risk for frailty category were shown (n=6), a hospitalisation risk was not examined (n=3), the same cohort used (n=1), a poster presentation (n=1) or a cross-sectional study design (n=1). Thirteen studies remained and were examined for methodological quality using the Newcastle-Ottawa quality assessment scale for cohort studies.26 All 13 studies were scored as five or greater and considered to have an adequate quality of methodology to be included in the meta-analysis (table 1).
Table 1 shows the characteristics of the 13 included studies involving 74 900 community-dwelling older people who were examined for hospitalisation risk according to frailty status.6 ,15 ,16 ,18 ,19 ,22 ,30–36 Five studies were from the USA,6 ,15 ,19 ,22 ,30 two were from Italy34 ,35 and one each from France,18 the UK,33 Spain,32 Mexico,36 Portugal16 and Korea.31 The cohort sizes ranged widely with the largest one (n=40 657) from the Women's Health Initiative22 and the smallest one including 95 individuals.16 Two studies involved only women15 ,22 and the rest used mixed cohorts. Although most of the studies presented a mean age ranging from 65.8 to 81.5 years, some did not, but just presented numbers of individuals in age groups. Most of the included studies (10/13) used original or modified CHS criteria,6 ,15 ,16 ,18 ,19 ,22 ,30–33 and the rest used modified SOF criteria.34–36 The ORs were presented or calculated in 10 studies,16 ,18 ,19 ,22 ,30 ,31 ,33–36 and 3 studies presented HRs.6 ,15 ,32 Follow-up periods ranged from 10 months16 to 8 years.30
Frailty as a predictor of hospitalisation
Meta-analysis of studies presenting OR
The ORs and 95% CIs were available for hospitalisation risk according to frailty and/or prefrailty from 10 studies encompassing 67 288 older people in the community.16 ,18 ,19 ,22 ,30 ,31 ,33–36 Fixed-effects models were used to synthesise pooled ORs of hospitalisation risk for frailty and prefrailty as heterogeneity was moderate (p=0.02, I2=54%) and low (p=0.26, I2=21%), respectively. Frailty and prefrailty were significantly associated with a higher risk of hospitalisation (pooled OR=1.90, 95% CI 1.74–2.07, p<0.00001; pooled OR=1.26, 95% CI 1.18–1.33, p<0.00001, respectively) compared with non-frail individuals (figure 2A).
Meta-analysis of studies presenting HR
Three studies with 7970 older people presented HRs as a risk measure of hospitalisation by frailty and prefrailty.6 ,15 ,32 Since heterogeneity was moderate for frailty (p=0.09, I2=58%) and low for prefrailty (p=0.69, I2=0%), fixed-effects models were employed. Frailty and prefrailty were significantly associated with a higher risk of hospitalisation (pooled HR=1.30, 95% CI 1.12–1.52, p=0.0007; pooled HR=1.13, 95% CI 1.04–1.24, p=0.005, respectively) compared with non-frail individuals (figure 2B).
Publication bias assessment
Visual inspection of funnels plots for studies presenting OR and HR for frailty and prefrailty (figure 3) did not show obvious asymmetry. No significant publication bias was observed among the studies presenting OR for frailty and prefrailty using Begg-Mazumdar's and Egger's tests (all p>0.05). It was not possible to use these tests for the studies with HR owing to the small number of included studies.
Subgroup analysis and meta-regression analysis
The 10 studies with OR of hospitalisation risks for frailty were further examined by subgroup analysis. The studies were divided into subgroups according to several study characteristics including location, sample size, female proportion, mean age, frailty criteria, follow-up period, methodological quality and outcome adjustment, and were compared for subgroup differences (table 2). Hospitalisation risks according to frailty were higher among two subgroups: three studies with mean age ≥ 75 years (pooled OR=3.09, 95% CI 2.00–4.77) compared with six studies with mean age <75 years (pooled OR=1.77, 95% CI 1.56–2.01) and four studies providing unadjusted ORs (pooled OR=2.46, 95% CI 1.87–3.24) compared with six studies providing adjusted ORs (pooled OR=1.84, 95% CI 1.68–2.02). P value for subgroup difference was 0.02 and 0.05, respectively.
The study characteristics of these 10 studies were further analysed using meta-regression analysis for potential moderator effects on the hospitalisation risks by frailty. The characteristics investigated were sample size, female proportion of a cohort, mean age, follow-up period and Newcastle-Ottawa scale score, and were individually entered into the meta-regression models. None of these characteristics were significant modulators in the associations between frailty and hospitalisation.
This systematic review and meta-analysis shows that community-dwelling older people classified as frail or prefrail had significantly higher risks of hospitalisation than those classified as robust. Despite the substantial diversity in the methodologies, such as study location, sample size, gender proportion and follow-up period, the hospitalisation risks according to frailty were fairly consistent among the included studies and the heterogeneity was low to moderate across the studies (I2=0–58%). It was suggested by subgroup analysis that studies with older populations and unadjusted outcome measures were associated with higher hospitalisation risks according to frailty.
Among the 13 studies included in this systematic review, the majority (76.9%, 10/13) used original or modified versions of CHS, while the rest of 3 studies used SOF criteria. Given SOF criterion is a shorter version of CHS and they share some frailty components (weight loss and exhaustion), the consistent use of CHS or SOF criteria by the included studies may have contributed to low-to-moderate heterogeneity across the studies.
Subgroup analysis suggested higher hospitalisation risks according to frailty were associated with two study characteristics: higher mean age (≥75 years) and unadjusted outcome measures. It is inevitable for us to age without health declines and deficits, and it may be natural that older people are more predisposed to various negative health outcomes and therefore to higher hospitalisation risks by frailty compared with younger people. When frail older people are being hospitalised, there should be multiple confounding factors directly and indirectly associated with the hospitalisation risks. Therefore, it is important to take it into consideration by adjusting these factors in statistical models to examine independent associations between frailty and hospitalisation risks. It can be expected that the pooled risk estimate was smaller among the studies properly adjusting for the confounders. Another subgroup with five studies with smaller sample sizes (n<4000) was found to have a tendency to have higher hospitalisation risks (pooled OR=2.57, 95% CI 1.85–3.57) compared with another five studies involving 4000 or more participants (pooled OR=1.85, 95% CI 1.69–2.03, p for subgroup difference=0.06). Although it is not clear why the studies with smaller sample sizes showed higher hospitalisation risks, four out of the five smaller studies provided unadjusted risk measures, which could be the cause of this subgroup difference. In general, adjustment for multiple confounding covariates may sometimes be difficult or may not always be possible especially when a sample size is very small due to lack of statistical power.
Exact mechanisms underlying the associations between frailty and subsequent higher hospitalisation risks are unknown. Although causes and reasons for older people's hospitalisations can be multifactorial, falling can possibly explain the associations at least to some degree. Approximately one-third of older people aged 65 and older fall every year.37 Falls are a leading cause of mortality and morbidity, including hip fracture or head injury, and can lead to hospitalisation.38 Physical components of frailty, such as weakness or gait abnormality, may potentially increase risks of falling in frail older people.5 In fact, frailty has been shown to be a significant predictor of future falls among community-dwelling older people.7 ,39 Despite the strong associations of falls with frailty and hospitalisation, fall-related factors, such as a history of falling or a fall as a reason for hospitalisation, were not investigated in the studies included in this review.
Although this review focused on physical frailty criteria, some experts advocated that frailty should be conceptualised as a multidimensional syndrome including not only physical but also cognitive, psychological and social factors.40–42 Frailty index (FI) describes frailty as a continuous score based on accumulation of age-related deficits and impairments in multidimensional domains.43 FI was used in previous studies and showed those with higher FI (worse frailty status) were at increased risks for hospitalisation among community-dwelling older people.20 ,44–46 The results of these studies were not able to be combined mainly because of different methodologies (ie, effect sizes per FI unit, per 0.01 of FI or per 7 groups by FI).
This study has multiple strengths and some limitations. First, this is the first systematic review and meta-analysis on associations between frailty and future hospitalisation risk. Second, the robust methodology, according to the PRISMA and MOOSE statements, was employed, including conducting a comprehensive systematic review using five electronic databases and assessing the methodological quality, the publication bias and the heterogeneity across the included studies. Third, the meta-analyses showed persistent and dose–response findings: higher degree of frailty status (frail>prefrail>robust) was associated with a higher risk of future hospitalisation in both study groups presenting OR and HR. These findings seem reliable because there was no evidence of a high heterogeneity or publication bias detected among the studies. Despite these strengths, it should be recognised as a potential limitation that all processes were conducted by one investigator. It would have been ideal to have two independent investigators involved in some stages of the process, such as data extraction or methodological quality assessment. The findings of this study should therefore be interpreted with caution since potentially important studies may have been missed or the extracted data may have been inaccurate. Another potential limitation is that none of the included studies took into consideration a fall as a potential cause of hospitalisation or a history of falls as an important confounder, which might have influenced the results.
It is still unknown which frailty criteria are most suitable to detect older people at high risk of hospitalisation and what factors or causes (ie, falls) are involved directly and indirectly in the associations between frailty and hospitalisation. Researchers can fill the gap by designing and conducting longitudinal cohort studies focusing on how frailty is related to specific causes of hospitalisation. Given frailty is a dynamic state47 and can possibly be reversed back to being prefrail or robust by appropriate interventions,5 the findings of this review are also valuable for clinicians because there is a possibility that they could screen older people for frailty as a risk factor of hospitalisation and could start the interventions if appropriate to prevent them from being hospitalised. Last, policymakers could make the most of this review's findings as well to conduct campaigns or create policy schemes, such as exercise promotions or public education regarding nutrition, for older people to support them to prevent onset of frailty or reverse it. All of these efforts could lead to reduced detrimental effects and related substantial healthcare costs of hospitalisation among older people.
This systematic review and meta-analysis demonstrates the pooled evidence that frailty and prefrailty are significant predictors of hospitalisation among community-dwelling older people. The findings are important for all related parties including clinicians, researchers and policymakers.
What is already known on this subject
Hospitalisation of older people has become a major focus because of detrimental hazards and substantial healthcare burden and costs.
Frailty has increasingly been recognised as an important predictor of hospitalisation.
The objectives of this study were to systematically review the literature for the associations between physical frailty and hospitalisation risks among community-dwelling older people and to conduct meta-analyses to synthesise pooled risk estimates.
What this study adds
This systematic review and meta-analysis has demonstrated that physical frailty was a significant predictor of hospitalisation among community-dwelling older people.
The hospitalisation risks according to frailty may be higher among those with advanced age.
Interventions targeted at reducing frailty may potentially reduce hospitalisation risks.
The author thanks Dr Kyoko Kashima for her support to confirm accuracy of systematic review screening process.
Contributors GK designed the study, collected, analysed and interpreted the data and drafted and revised the manuscript. GK has full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Funding This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.