Study participants
Older adults aged 60 years and over (n=701) who agreed to use their National Registration Identity Card (NRIC) number to link with administrative database to retrieve healthcare utilisation data were sampled from the Population Health Index (PHI) study, a population-based health survey conducted in the Central Region of Singapore. The baseline data of PHI which were collected during November 2015 to November 2016 were used for this study. The sampling procedure and survey methodology of the baseline PHI study has been described elsewhere [19–21]. In brief, eligible participants (Singapore citizens or permanent residents, aged 21 years and above and lived in the selected housing unit for the past six months) staying in randomly selected household units in the Central Region of Singapore were identified via door-to-door visits by trained surveyors and one eligible household member was randomly selected using Kish grid [22]. There were 1942 eligible community-dwelling adults recruited and underwent detailed structured interviews for the baseline PHI survey.
The PHI study was approved by the ethics review committee of the National Healthcare Group (NHG) Domain Specific Review Board (Reference Number: 2015/00269). Written informed consent was obtained from all individual participants after they were fully informed of the study objectives and procedures.
Frailty assessment
Frailty was determined using the revised five-item FRAIL scale (Fatigue, Resistance, Ambulation, Illnesses, & Malnutrition) with the “Malnutrition” replacing the “Loss of weight” in the original FRAIL scale (Fatigue, Resistance, Ambulation, Illnesses, & Loss of weight) [23, 24]. Each item is scored either 0 or 1. The revised FRAIL scale is scored from 0 (best) to 5 (worst) and is translated into three categories: robust (0), pre-frail (1–2), and frail (3–5). Similar to other studies [23, 24], we operationalised the FRAIL scale based on information obtained from specific questions included in the PHI survey questionnaire. “Fatigue” was measured by asking how often they felt tired with responses of “more than half the days” or “nearly every day” scored 1. “Resistance” was assessed by asking their difficulty in walking up and down one flight of stairs without using handrail, and “Ambulation” was measured by asking their difficulty in walking around one floor of home or several blocks without aids; “quite a lot” or “cannot do” responses were each scored as 1. “Illness” was scored 1 for those who reported 5 or more illnesses out of 14 illnesses. “Malnutrition” was scored 1 if Body Mass Index (BMI)<18.5 or MNA screening score<8 or Mini Nutritional Assessment (MNA) total score <17. A complete description of the revised FRAIL scale items’ scoring criteria is provided in Table 1.
Table 1. FRAIL scale items.
Item
|
Criteria
|
Fatigue
|
1. “Over the last 2 weeks, how often have you been bothered by feeling tired or having little energy?” 0=Not at all, 1=Several days, 2=More than half the days, 3=Nearly every day
Responses of “2” or “3” are scored as 1 and all others as 0.
or
2. “Over the last 4 weeks, how often have you been bothered by getting tired very easily?”
0=Not at all, 1=Several days, 2=More than half the days, 3=Nearly every day
Responses of “2” or “3” are scored as 1 and all others as 0.
|
Resistance
|
1. Stairs in Activities of Daily Living
1=Unable to climb stairs, 2=Assistance is required in all aspects of stair climbing, 3=Able to ascent/descend but is unable to carry walking aids, and needs supervision and assistance, 4=Generally no assistance is required, 5=Able to go up and down a flight of stairs safely without help or supervision
Responses of “1”, “2” or “3” are scored as 1 and all others as 0.
or
2. How much difficulty do you have in going up & down a flight of stairs without using handrail?”
5=None, 4=A little, 3=Some, 2=Quite a lot, 1=Cannot do
Responses of “1” or “2” are scored as 1 and all others as 0.
|
Ambulation
|
1. Ambulation in Activities of Daily Living
1=Dependent in ambulation, 2=Constant presence of one or more assistants is required during ambulation, 3=Assistance is required with reaching aids and/or their manipulation. One person is required to offer assistance, 4=Independent in ambulation but unable to walk 50 yards/metres without help, or supervision is needed for confidence or safety in hazardous situations, 5=Must be able to use crutches, canes, or a walker, and walk 50 metres/yards without help or supervision.
Responses of “1”, “2”, “3” or “4” are scored as 1.
or
2. “How much difficulty do you have in walking around one floor of your home, taking into consideration thresholds, doors, furniture, and a variety of floor coverings?”
5=None, 4=A little, 3=Some, 2=Quite a lot, 1=Cannot do
Responses of “1” or “2” are scored as 1 and all others as 0.
or
3. “How much difficulty do you have waling several blocks?”
5=None, 4=A little, 3=Some, 2=Quite a lot, 1=Cannot do
Responses of “1” or “2” are scored as 1 and all others as 0.
|
Illnesses
|
“Have you ever been told to have any of these conditions by a Western-trained doctor?” The conditions include diabetes, high blood pressure, high blood cholesterol, heart failure, stroke / transient ischaemic attacks, asthma, chronic bronchitis/ emphysema/COPD, chronic kidney disease, cancer, osteoarthritis/gout/rheumatoid arthritis, osteoporosis, dementia/Alzheimer’s, schizophrenia, Parkinson)
1=Yes, 0=No
Responses of “1” are scored as 1.
|
Malnutrition
|
or
2. Screening score of the Mini Nutritional Assessment <8 or total score <17
|
The revised FRAIL scale score ranges from 0 (best) to 5 (worst).
0: Robust, 1-2:Pre-frail, 3-5: Frail
Healthcare utilisation
The healthcare utilisation data during the retrospective 6-month and prospective 6-month periods were obtained from RHS database [25]. The RHS database contains linked NHG polyclinic visit records, specialist outpatient clinic (SOC) visit records, emergency department (ED) attendance records, day surgery (DS) attendance records and hospital discharge records from three government hospitals - Tan Tock Seng Hospital, Khoo Teck Puat Hospital and Institute of Mental Health, chronic disease management system records and mortality records from local registries. The healthcare utilisation data were categorised according to the main healthcare settings into polyclinic visits, SOC visits, ED visits, DS attendances and hospitalisations. Polyclinic visits refer to doctor consultation and technical visits made by the individual to any of the nine linked NHG polyclinics. SOC visits and ED visits refer to visits to the specialists in outpatient clinics and the emergency rooms located within three government hospitals, respectively. Similarly, DS attendances refer to surgical procedures performed in day surgery rooms where patients were discharged on the same day without admitting to inpatient wards; and hospitalisations refer to inpatient episodes with at least one overnight stay at these hospitals. The survey data and healthcare utilisation data were linked using NRIC numbers which were removed thereafter for data analysis.
Other variables
We controlled for the confounding effects of covariates to examine the independent effect of frailty on the rates of healthcare utilisation in different care settings. These covariates included demographic factors (age, gender (male / female), Chinese (yes / no), marital status (single / married / widowed or divorced), living arrangement (alone / with others)) [8, 26] and smoking status (non-smoker / past smoker / current smoker) [17]. Highest education level (no formal education / primary / secondary or above) and self-perceived money sufficiency for basic living needs (sufficient / insufficient) were also included as control variables as they are enabling factors which influence individuals’ health seeking behaviours and healthcare utilisation [8, 27, 28].
Multimorbidity and disability, which are related to but also distinct from frailty [29, 30], were commonly adjusted in studies examining the association between frailty and healthcare utilisation [8, 14]. We controlled for multimorbidity as a dichotomous variable (yes / no) which was defined as the presence of two or more of the following 17 chronic conditions: dyslipidemia, high blood pressure, diabetes, chronic kidney disease (CKD), heart attack / ischemic heart disease, heart failure, stroke / transient ischemic attack, asthma, chronic bronchitis / emphysema / chronic obstructive pulmonary disease (COPD), cancer, osteoarthritis / gout / rheumatoid arthritis, osteoporosis, depression, anxiety disorder, schizophrenia, dementia / Alzheimer’s, and Parkinson’s disease [20] (The prevalence of individual chronic diseases among the study participants is presented in the supplementary table S1). Disability, which was determined based on whether assistance was required in any of the ten activities of daily living (ADLs) (yes / no) measured using the Modified Barthel Index [31], was also controlled in the models.
Statistical analysis
Characteristics of the study population were described using mean and standard deviation (SD) for continuous variables, and frequency and percentages for categorical variables. Mean and SD were used to describe healthcare utilisation in every frailty group. To examine the differences in characteristics and utilisation across frailty groups, one-way analysis of covariance (ANOVA) tests (normally distributed) or Kruskal-Wallis H tests (non-normally distributed) were performed for continuous variables, and chi-squared tests were conducted for categorical variables.
Healthcare utilisation by settings are count variables characterised by a point mass at zero followed by a right-skewed, discrete distribution, and non-negative values [32, 33]. Given the over-dispersion of data (the conditional variance is larger than the conditional mean), a negative binomial distribution was chosen over a Poisson regression [34]. Healthcare utilisation in five settings formed five different dependent variables and were analysed independently; and the three-level frailty category (robust, pre-frail, frail) was the independent variable of interest. We further adjusted for control variables including demographic factors (including age, gender, ethnic group, marital status, and living arrangement), socioeconomic status (highest education level, self-perceived money insufficiency), smoking status, multimorbidity, and disability. The results were presented as incidence-rate ratios (IRRs) and their corresponding 95% confidence intervals (CIs). All analyses were performed using Stata/SE 16.1. A p value of less than 0.05 was set as the level of significance.