Study design

We applied a cross-sectional study design using data from The Older Persons and Informal Caregivers Survey Minimum DataSet (TOPICS-MDS). The development and design of TOPICS-MDS has been described in more detail.23 TOPICS-MDS is a database designed to capture information on the well-being of older persons in the Netherlands. TOPICS-MDS was developed to collect uniform information from studies funded under the National Care for older citizens programme.24 Included survey items were based on the recommendations of experts from eight medical research centres in the Netherlands who identified key outcomes in older persons’ health using validated instruments. Then, an independent multidisciplinary panel with expertise in gerontology, epidemiology, biostatistics and health services research evaluated and revised TOPICS-MDS. Dutch TOPICS-MDS questionnaires were piloted in four regions throughout the Netherlands. Misinterpretation of questions were identified and minor changes were made by a plain language expert. The Dutch version of TOPICS-MDS was translated into Turkish, the two Moroccan languages and the two Surinamese languages. Two independent translators, who were native speakers of the target language, translated all items and differences were reconciled in a consensus meeting. Another translator back translated the forward translation and discrepancies were discussed, which resulted in a final translation. Questionnaires were then cross-culturally validated by the three-step test-interview.25 The translation and cross-cultural validation of TOPICS-MDS was done as part of the SYMBOL study and has been described in more detail.26 Between 2010 and 2013, 50 studies were conducted in the Netherlands which applied the TOPICS-MDS questionnaire.23 TOPICS-MDS consists of pooled data of these studies which differ across study design, sampling framework and inclusion criteria. TOPICS-MDS is a fully anonymised dataset.

Patient and public involvement

The research proposal of this study was presented to a forum of older persons and informal caregivers who are part of the geriatric network of Southwest Netherlands and was adapted according to their suggestions. The study design was presented at the ‘TOPICS Special Interest Group Diversity’, who advised on design and subgroup analyses. This interest group consisted of researchers, older persons and a representative of the Network of Organisations of Older Migrants in the Netherlands. The results of this study will be disseminated to participants on the TOPICS-MDS website https://topics-mds.eu/ and on the website of the Organisation of Health Research and Development (ZonMw) https://www.zonmw.nl/nl/.

Ethnic background

We defined ethnic background according to the classification of Statistics Netherlands.27 The participant had a non-Dutch ethnic background if at least one of the parents was born in another country than the Netherlands. To determine the country of origin, the following was applied.27 If the participant was also (as well as parent/parents) born outside of the Netherlands, the participants’ country of birth determined the ethnic background. If the participant was born in the Netherlands but their mother was not, the country of birth of the mother determined the ethnic background. If the mother was born in the Netherlands but the father was not, the country of birth of the father determined the ethnic background. We categorised participants according to ethnic background: Dutch, Indonesian, Surinamese, Moroccan or Turkish ethnic background.

Population for analysis

We restricted our analysis to TOPICS-MDS data from studies on independently living persons aged 55 years and older with a Dutch, Indonesian, Surinamese, Moroccan or Turkish ethnic background. We included studies that recruited participants from the general population and from general practitioners’ registries. We excluded studies among persons living in a nursing home, studies that recruited participants in a hospital setting (n=6136) and studies for persons with dementia (n=507). We excluded other ethnic groups due to small numbers (n=1718). We also excluded persons with more than 15 missing items for the TOPICS-Frailty Index (n=3662) and with missing country of birth (n=752). The final sample comprised data from 29 studies with data of 23 371 persons (see figure 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Population of analysis.

TOPICS-Frailty Index and components

Frailty was measured by the TOPICS-Frailty Index, which was developed and validated using data from the TOPICS-MDS questionnaire by Lutomski et al.28 As described previously,11 in our study we included the 45 item TOPICS-Frailty Index, after exclusion of the item measuring prostatism. Searle et al showed that a Frailty Index with 30–40 items is accurate for predicting adverse outcomes.9 29 The TOPICS-Frailty Index was calculated when at least 30 items were available. Binary items were scored 0 when the deficit was absent and 1 when the deficit was present. For items with multiple response categories, responses were coded so that they could be mapped on the interval 0–1, as proposed by Searle et al.9 For example, when the response categories were no/some/extreme, responses were coded 0/0.5/1, respectively. Then, the health deficits a person reported were added up and divided by the total health deficits measured for this person. This resulted in a score between 0 and 1. Persons were also grouped according to their TOPICS-Frailty Index score, we used a cut-off score ≥0.25 to indicate frailty as suggested by Rockwood et al.30

The TOPICS-Frailty Index consists of 45 items of the TOPICS-MDS questionnaire that belong to six components, each measured by validated instruments: morbidities, ADL, IADL, HRQoL, psychosocial health and SRH.9 The component ‘morbidities’ was measured by 16 items regarding the self-reported presence (yes/no) of diabetes, stroke, heart failure, cancer, respiratory condition (asthma, chronic bronchitis, lung emphysema or Chronic obstructive pulmonary disease), incontinence, joint damage of hips or knees, osteoporosis, hip fracture, fractures other than hip, dizziness with falling, depression, anxiety/panic disorder, dementia, hearing problems, vision problems. The component ‘ADL limitations’ was measured by six items using a modified version of the Katz et al instrument.31 32 Persons could indicate whether they needed help (yes/no) with the following activities: bathing, dressing, toileting, incontinence, sitting down, eating. The component ‘IADL limitations’ was measured by nine items using a modified version of the Katz et al instrument.31 32 Persons could indicate whether they needed help (yes/no) with the following activities: using the telephone, travelling, shopping, preparing a meal, cleaning, taking medications, handling finance, brushing hair and walking. The component ‘HRQoL’ was measured by using the six items of the EuroQol 5D+C.33 Persons could indicate whether they had problems (no/some/extreme) with the following: mobility, self-care, usual activities, pain/discomfort, anxiety/depression and cognition. The component ‘psychosocial health’ was measured with five items of the RAND-36.34 Persons could indicate how much of the time in the past month (none/a little/some/a good bit/most/all) they had been the following: nervous, calm, downhearted, happy and down in the dumps, and how much time (none/a little/some/most/all) health problems had interfered with social activities. The component ‘SRH’ was measured with two items of the RAND-36,34 one regarding perceived current health status (poor/fair/good/very good/excellent) and one regarding perceived changes in health in the past year (much worse/slightly worse/about the same/a little better/much better). The score for distinct components of the TOPICS-Frailty Index was calculated analogous to the TOPICS-Frailty Index, by adding up the health deficits within the component that a person had, divided by the total of possible health deficits included in the component.9 Higher scores represent a higher proportion of health deficits for that component. We accepted no missing variables for the SRH component score and a maximum of one of three missing variables for the other frailty component scores.

Potential confounders

We incorporated the following potential confounders in this study: gender, age, living arrangement, marital status, generation of immigration, education level and socioeconomic status of the neighbourhood. Age in years was assessed by asking year of birth. Living arrangement was assessed by asking whether participants were living: independent alone or independent with others. We categorised answers into ‘not alone’ and ‘alone’. Marital status was assessed by asking whether participants were: married, divorced, widowed, unmarried, long-term cohabitation unmarried. We categorised answers into ‘married/cohabitant partners’, ‘divorced’, ‘widowed’ and ‘single’. Generation of immigration was assessed by country of birth of participant and his/her parents. We categorised answers into ‘first generation’ when the person was born in a country other than the Netherlands and ‘second generation’ if the person was born in the Netherlands and (one of) the parents were born in another country. Socioeconomic status was assessed with both education level and neighbourhood socioeconomic status. TOPICS-MDS uses the 1997 International Standard Classification of Education35 to assess education level; participants were asked whether they had completed: fewer than 6 years of primary school; 6 years of primary school; primary school without further completed education; vocational school; secondary professional education or university entrance level or tertiary education. We categorised the level of education into ‘primary education or less’ and ‘secondary education or higher’. For the neighbourhood socioeconomic status, the 2006 reference scores for area codes were used, as calculated by the Netherlands Institute for Social Research36 based on the education level, income and labour market position of persons living in each area code. We categorised scores into quartiles: quartile 1 is the least deprived quartile and quartile 4 is the most deprived.

Statistical analysis

We calculated the statistical significance of differences in sociodemographic characteristics, frailty and frailty components (morbidities, ADL limitations, IADL limitations, psychosocial health, HRQoL and SRH) by ethnic background using χ² tests for categorical variables and one-way analysis of variance for continues variables.

To examine the association of ethnic background with frailty and with distinct frailty components, we estimated multilevel random-intercept models because data were clustered in studies.37 As such dependency between the observations of participants of a study (because of sampling design and or inclusion criteria) was taken into account. We built one multilevel linear regression model for the continues TOPICS-Frailty Index outcome and one multilevel logistic regression model for the dichotomous outcome ‘frailty’, with ethnic background as independent variable. We furthermore built six separate multilevel linear regression models for the six frailty component outcomes with ethnic background as independent variable. Only potential confounders that led to a substantial change in effect estimates in the association between ethnic background and frailty (ie, ≥10% change) were included in all models: age, sex, living arrangement (alone/not alone) and education level.38 Marital status was excluded due to multicollinearity with living arrangement. Interaction effects of ethnic background with age, sex, living arrangement and education level on frailty and the six frailty components were assessed.

Percentages of missing values in the covariates were 2% or less (table 1). Because the missing values were not completely at random, multiple imputation was used to deal with the missing values in the covariates. Five imputed datasets were created using a fully conditional specified model, thus taking into account the uncertainty of imputed values.39 We used pooled estimates from these five imputed datasets to report regression coefficients and their 95% CIs. We considered a p value of. 05 or lower to be statistically significant and used Bonferroni correction for testing explorative interactions (p value=0.05/4).40 Descriptive analyses were performed using SPSS V.23.0 (IBM SPSS. Multilevel linear regression analyses were performed using R- V.3.3.2 using lme4 package.

Non-response analysis

A comparison of persons included in the study (n=23 371) with persons not included due to missing values for the TOPICS-Frailty Index and/or country of birth (n=4414) did not indicate significant differences in terms of age (p=0.872), sex (p=0.779), living arrangement (p=0.444) and marital status (p=0.166). However, excluded persons were more often living in rural areas and in deprived neighbourhoods (p<0.001) than persons included in the study.