Loneliness prevalence of community-dwelling older adults and the impact of the mode of measurement, data collection, and country: A systematic review and meta-analysis

Objectives: The aim of this systematic review and meta-analysis is to assess the prevalence of loneliness in many countries worldwide which have different ways of assessing it. Design: Systematic review and meta-analysis. Setting: We searched seven electronic databases for English peer-reviewed studies published between 1992 and 2021. Participants: We selected English-language peer-reviewed articles, with data from non-clinical populations of community-dwelling older adults ( > 60 years), and with “ loneliness ” or “ lonely ” in the title. Measurements: A multilevel random-effects meta-analysis was used to estimate the prevalence of loneliness across studies and to pool prevalence rates for different measurement instruments, data collection methods, and countries. Results: Our initial search identi ﬁ ed 2,021 studies of which 45 ( k = 101 prevalence rates) were included in the ﬁ nal meta-analysis. The estimated pooled prevalence rate was 31.6% ( n = 168,473). Measurement instrument was a statistically signi ﬁ cant moderator of the overall prevalence of loneliness. Loneliness prevalence was lowest for single-item questions and highest for the 20-item University of California-Los Angeles Loneliness Scale. Also, differences between modes of data collection were signi ﬁ cant: the loneliness prevalence was signi ﬁ cantly the highest for face-to-face data collection and the lowest for telephone and CATI data collection. Our moderator analysis to look at the country effect indicated that four of the six dimensions of Hofstede also caused a signi ﬁ cant increase (Power Distance Index, Uncertainty Avoidance Index, Indulgence) or decrease (Individualism) in loneliness prevalence. Conclusions: This study suggests that there is high variability in loneliness prevalence rates among community-dwelling older adults, in ﬂ uenced by measurement instrument used, mode of data collection, and country.


Introduction
Lonely people can be at increased risk of death (Elovainio et al., 2017).In many countries, loneliness has arisen in the last few years on the policy agenda as an important societal challenge, which was amplified by the COVID pandemic (Lampraki et al., 2022).Moreover, socially and emotionally satisfying contacts can form a buffer against loneliness in later life when negative life events may occur (Switsers et al., 2021).Although loneliness in older adults is sometimes called a "silent epidemic," estimates of loneliness prevalence differ widely across nations and across different assessment scales.
Perlman and Peplau defined loneliness in 1981 as "the unpleasant experience that occurs when a person's network of social relations is deficient in some important way, either quantitatively or qualitatively" (Perlman & Peplau, 1981, 31).De Jong Gierveld's definition from 1987 adds that "this includes situations, in which the number of existing relationships is smaller than is considered desirable or admissible, as well as situations where the intimacy one wishes for has not been realized" (de Jong Gierveld, 1987, 120).Both definitions describe loneliness as a negative and subjective feeling, which is in contrast to, e.g.social isolation, which refers to the objective situation and the absence of relationships with other people (De Jong Gierveld et al., 2006).
The WHO reports that there are no global assessments of the proportion of communitydwelling older people who are experiencing loneliness, but estimates that between 20% and 34% of older people in China, Europe, Latin America and the United States are lonely (World Health Organization, 2021).A recent meta-analysis based on prevalence data from 106 countries in 24 studies suggests that older adults ( ≥ 60 years; not explicitly community dwelling) in general have a higher prevalence of loneliness compared with their younger counterparts (i.e., young adults [18-29 years] and middle-aged adults [30-59 years]) (Surkalim et al., 2022).
Today, however, there is a high variability in loneliness prevalence (Surkalim et al., 2022).Possible explanations are differences in culture (Jylhä & Jokela, 1990), demography (Fokkema et al., 2012), socioeconomic status (Hansen & Slagsvold, 2016) or trust (Rapolienė & Aartsen, 2021).Another recent systematic review and meta-analysis on the prevalence of loneliness among older people in high-income countries (not explicitly community dwelling) hypothesizes that high variability between different prevalence studies could be influenced by differences in used measurement instruments and different modes of data collection (e.g., face-to-face, written questionnaires, etc.) (Chawla et al., 2021).Today, the use of different measurement instruments is increasing (e.g., Awad et al., 2023) using the De Jong Gierveld Loneliness Scale and Ost-Mor et al. (2023) using the (University of California-Los Angeles (UCLA) Loneliness Scale, so the multidimensionality of loneliness is already widely recognized.
However, current research suggests that loneliness measures should be considered carefully in relation to the opposed research question(s) of a study and encourages researchers to include multiple measures in their studies to ensure robustness and to identify potential discrepancies among measures in existing and future research (Mund et al., 2022).Su et al. (2023) published a systematic review on the prevalence of loneliness and social isolation among older adults during the COVID-19 pandemic, but the influence of measurement instrument and mode of data collection were not treated.Through a systematic literature review and meta-analysis, this study reviews the prevalence of loneliness among communitydwelling older adults in countries worldwide and examines the study characteristics of these loneliness prevalence studies with specific attention to the influence of measurement instruments, mode of data collection, and the country where the study was conducted.

Search strategy and selection criteria
This study follows the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Page et al., 2021).We screened seven electronic databases, i.e., Web of Science, PubMed, Sociological Abstracts, Social Services Abstracts, Embase, PsycINFO, and Cochrane Library, for eligible studies.The literature search included studies published between January 1, 1992 and October 31, 2021.We used "loneliness" as a search term in the title, "community-dwelling older adults" and "prevalence" as search terms in title and abstract, as well as possible variations, keywords and MeSH headings, if applicable for the database.The detailed search strategy can be found in Appendix 1 (published as supplementary material online attached to the electronic version of this paper at https:// www.cambridge.org/core/journals/internationalpsychogeriatrics).
Authors HS and HC selected the studies derived from Web of Science and PubMed.HS and DD selected studies from Sociological Abstracts, Social Services Abstracts, PsycINFO, Embase, and Cochrane Library.After removing duplicates for both selection processes, a random sample of 10% was assessed by HS and HC for Web of Science and PubMed, and by HS and DD for the five other databases, to make sure the different 2 Hannelore Stegen et al.
authors selected studies based on the same benchmarks.To decide upon inclusion, the title, the abstract and eventually the full text of the study (if necessary) were screened.When one of the selection criteria was not met, the study was excluded without evaluating the other selection criteria.In case of doubt, HP, LDD, and ED decided together upon in-or exclusion.Reference lists from the included studies and studies citing our included studies were screened in the final stage to assure no further studies would be left unnoticed.
Studies were eligible if "loneliness" or "lonely" was mentioned in the title of English-language peerreviewed studies and if data was reported on a nonclinical population of community-dwelling older adults where a minimum age of 60 years was specified.The definition of community-dwelling older people by Steultjens et al. (2004) was followed, stating that community-dwelling older people are "people aged 60 years or older living independently," and therefore not living in institutionalized settings such as nursing homes, care homes or other types of residential care (Steultjens et al., 2004).The final inclusion criterion was that studies should have as an explicit aim to estimate the loneliness prevalence, since clearly outlining the explicit purpose of the study contributes to a paper of better quality (Mack, 2015).The primary objective of prevalence studies is to produce frequency estimates for the overall population, and sometimes population subgroups (Boyle, 1998).Altogether, prevalence studies about loneliness among communitydwelling older adults were selected for this study.Studies from all countries and world regions were included to get a complete image of existing prevalence studies and the corresponding loneliness measurement instruments and modes of data collection, since there are known differences between countries and cultures in terms of loneliness prevalence.

Data analysis
The following data were extracted: year of publication, year of data collection, was the study conducted pre-or during-COVID, sample size (of loneliness questions), percentage of women, type of sample, country (reclustered into region), level on which the study was conducted (national or regional), mode of data collection, data source (own or existing dataset), and used measurement instruments.
As part of the meta-analysis, the quality of the studies was appraised by HS and DD using the Joanna Briggs Institute (JBI) Critical Appraisal Checklist (Munn et al., 2020).HS and DD first appraised all studies separately, and when no initial consensus was found, they decided together upon inclusion.Following JBI Checklist guidelines, studies included for review were given a quality cutoff score (Munn et al., 2020), whereby studies with a "lowquality" score (0-3) were excluded and studies with moderate (4-6) and high (7-9) quality scores were included for the meta-analysis.We also excluded papers in the meta-analysis if they used data that had already been used in another paper or data that originated from the same wave in the same database, and included the most complete or recent studies.
For the meta-analysis, carried out by PS and HS, supervised by LS, uniform response options were needed, and therefore we dichotomized the loneliness answers of all the studies to include them in the meta-analysis; this means that studies with more than two categories were also dichotomized.Furthermore, for articles using the De Jong Gierveld loneliness scale, when other cut-off scores than the proposed scores of De Jong Gierveld & van Tilburg (De Jong Gierveld & van Tilburg, 1999) were used, we recalculated the prevalence percentages, using the original dataset received upon request from the original authors.For the UCLA loneliness scales, we did not do this, since the authors did not propose any cutoff scores and since they indicated that there are no diagnostic criteria for being lonely (Russell, 1996).We therefore followed the cutoff that each of the studies proposed since we then had some clarity in who is considered as "lonely" in each of the studies.
In this review, all measurement instruments capture momentary loneliness, meaning that they measure loneliness as it is "now," at the moment of measuring (Compernolle et al., 2021).The answers of the participants, both on the De Jong Gierveld Loneliness Scale as well as on the UCLA Loneliness Scale and the single-item questions, are subjective to how people feel at the moment of answering the question(s), even if they ask about loneliness, e.g., in the past week.Of all prevalence studies, none included a measurement tool that measured lifetime prevalence.This means that in this study, specifically point prevalences of loneliness are being studied.Therefore, it was appropriate to compare all the different prevalence percentages, since they all cover this momentary loneliness, mentioning a point prevalence percentage of loneliness.
Further information on the quality appraisal including the completed JBI Critical Appraisal Checklist for each study can be found in Appendix 2, as well as an overview of the classification (not/mildly lonely vs. lonely) that can be found in Appendix 3 (both published as supplementary material online attached to the electronic version of this paper at https:// www.cambridge.org/core/journals/internationalpsychogeriatrics).
Following this, two steps were undertaken: the calculation of the pooled prevalence and a moderator analysis.First, a generalized linear mixed model (GLMM) was constructed.Such a model can directly model event counts with binomial likelihoods and fully account for within-study uncertainties (Lin & Xu, 2020).This approach has several advantages over the two-step meta-analysis which typically uses the Freeman-Tukey double arcsine transformation (Lin & Xu, 2020).In particular, we used a random intercept logistic regression model with a logit link function for the calculation of pooled prevalence rates (van Den Noortgate & Onghena, 2003).The outcome thus was the prevalence of loneliness (individual proportions) measured as the number of lonely older adults among the sample.A three-level meta-analytic model was used to analyze the data (Assink & Wibbelink, 2016), modeling three sources of variance: sampling variance of the observed prevalence rates (Level 1), the variance between prevalence rates from the same study (Level 2), and variance between studies (Level 3) (Cheung, 2014;Van den Noortgate et al., 2013).Results were backtransformed for easier interpretation.
Second, a multilevel random effects model was used for the moderator analyses to evaluate the impact of the measurement scale, the mode of data collection, and the country where a study was conducted on loneliness prevalence: the F-distribution was utilized to determine whether the pooled prevalence of loneliness was significantly affected by the moderators.Two separate one-tailed log-likelihood-ratio tests were conducted, comparing the deviance of the full model to the deviance of a model that excluded one of the variance parameters, to determine whether respectively the variance between prevalence rates within studies (Level 2) and the variance between studies (Level 3) was significant.All model parameters were estimated using the maximum likelihood estimation method.We considered p-values < 0.05 as statistically significant.The statistical analyses were carried out using the dmetar and metafor-packages (Viechtbauer, 2010) in R (version 4.2.1).
To look at the effect of country, we used the six dimensions of Hofstede (Hofstede, 2011), i.e., the Power Distance Index, Individualism, Motivation towards Achievement and Success, the Uncertainty Avoidance Index, Long-Term Orientation, and Indulgence.Despite the fact that there are some critiques on these dimensions now because of the idea of oversimplification and the static nature of cultures that these dimensions entangle (Chun et al., 2021;Minkov, 2018), the Hofstede dimensions were used because they do provide a standardized way to compare cultures and they increase the awareness and sensitivity to cultural norms (Hofstede Insights, 2023).Moreover, this study is not necessarily about the precise meaning and labeling of the dimensions, but about comparing cultural aspects measured with the same scale in each individual country in the first place.
The Power Distance Index signifies a society's acceptance of hierarchical power distributiona higher score indicates a greater acceptance of inequality.Second, there is the spectrum of Individualism versus Collectivism, where higher scores suggest weaker interpersonal connections beyond the core "family," and less responsibility for others' actions.The dimension of Motivation towards Achievement and Success is about what motivates people: wanting to be the best (Decisive; high score) or liking what you do (Consensusoriented; low score).A high score means that the society is driven by achievement, success, and competition, while a low score indicates a society that is driven by quality of life as a sign of success.The Uncertainty Avoidance Index measures a society's inclination to control unpredictability.A higher score indicates a preference for predictability and control in life.Long-Term Orientation versus Short-Term Orientation reflects a society's inclination toward pragmatism, modesty, and thriftiness with higher scores indicating a longterm focus.Finally, Indulgence versus Restraint explores how freely people gratify their desires and emotionshigher scores indicate a more permissive approach to enjoying life and expressing emotions.
Using the Country Comparison Tool of Hofstede (Hofstede Insights, 2023), we obtained a score for every included country for each of these dimensions between 0 and 100.These scores were gathered from survey responses over time, starting between 1967 and 1973 but still going on until today (Hofstede Insights, 2023).The dimension identification happens through factor analysis or other scaling methods, and next, normalization of factor scores is done to fit data from previous studies.The validity comes from correlations with dimensions of previous studies and national indices such as educational achievement or crime rates (Hofstede Insights, 2023).
We centered the continuous scores around the grand mean and used a multilevel approach, in which effect sizes are nested within studies (van Den Noortgate & Onghena, 2003) and which enables using all effect sizes in the primary studies so that maximum statistical power is achieved (Assink et al., 2015).
The protocol of this review was registered at the International Prospective Register of Systematic Reviews (PROSPERO) (registration number: CRD 42021230197).
In our systematic review, most prevalence data (k = 125, 70.6%) spanned from 2006 to 2015, and a majority (k = 127, 71.8%) came from European countries.The majority of the data collection was done face-to-face (k = 114, 64.4%), and through single-item questions (k = 139, 78.5%).Table 1 shows an overview of the study characteristics of the included studies in both the systematic review and meta-analysis.Table S1 specifically shows an overview of the study characteristics related to the loneliness prevalence found in the studies.While we included 62 studies in our systematic review, several studies included prevalence percentages of different countries, with corresponding differences in, e.g., sample size and percentage of women (compared with men), leading to separate prevalence rates designated as "k" (k = 177 for the systematic review and k = 101 for the meta-analysis).Appendix 4 (published as supplementary material online attached to the electronic version of this paper at https://www.cambridge.org/core/journals/internationalpsychogeriatrics)gives an overview of the study characteristics per study.
Calculation of the pooled prevalence of loneliness A total of 45 studies were included in the meta-analysis reporting on n = 168 473 participants with valid prevalence percentages; n = 107 267 using singleitem questions, n = 9795 using the UCLA 20-item scale, n = 13 668 using a shortened version of the UCLA scale, n = 37, 339 using the De Jong Gierveld (DJG) scale and n = 404 using a combination of different measures.Within these 45 studies, a total of 101 prevalence percentages were extracted.Descriptive information on the demographic and methodological characteristics is summarized in Table 1.The median of the included prevalence percentages was 26.0% (IQR 14.0% to 45.0%).Appendix 5 (published as supplementary material online attached to the electronic version of this paper at https:// www.cambridge.org/core/journals/internationalpsychogeriatrics)presents four forest plots showing the prevalences of all the included studies in the metaanalysis, for each measurement instrument separately.
Table 2 presents the estimated pooled prevalence of loneliness among community-dwelling older adults based on the random-effects model.The pooled prevalence was 31.6% (95% CI 24.4-39.9)and it was statistically significant (p < .001).The results of the likelihood-ratio test showed there was significant within-study variance (at level 2, X ² (1) = 57.06,p < .001) as well as significant between-study variance (at level 3, X ² (1) = 6221.89,p < .001).From Table 2, 0.23% of the total variance could be attributed to variance at level 1 (i.e., sampling error variance), 28.03% of the total variance to differences between the prevalence of loneliness within studies at level 2 (i.e., within-study variance) and 71.74% of the total variance could be attributed to differences between studies at level 3 (i.e., between-study variance).

Moderator analysis
We performed moderator analyses to assess the effect of measurement instruments, the mode of Loneliness in older adults: measurement and context 5 data collection, and the country where the study was conducted on the pooled loneliness prevalence.The results of all univariate moderator analyses are presented in Table 3.
We also found moderating effects of the mode of data collection on the overall pooled prevalence (F (3, 96) = 3.23, p = .008).This implied there were significant differences between the pooled prevalence from the four data collection methods.The loneliness prevalence for face-to-face data collection was 39.4% (95% CI 30.0-49.6), being significantly higher than telephone and CATI ) and self-report ).However, this moderator explained the variability between studies only modestly as the level 3 variance decreased by only 8% (from 1.091 to 1.002).
Regarding the effect of country, four of the six dimensions of Hofstede were significant (p < .001).The prevalence of loneliness among communitydwelling older adults was significantly higher (compared to the initial 31.6%we found) in a country with the mean score of our sample on the Power Distance Index (32.5% [95% CI 25.5-40.4]),the Uncertainty Avoidance Index (35.9% (95% CI 27.4-45.4))and the Indulgence index (34.0%[95% CI 26.9-41.9]).Countries with a mean score of our sample on the Individualism index had a significantly lower pooled prevalence of loneliness ).The dimension of Long-Term Orientation was not significant (p = .073),as well as the dimension of Motivation towards Achievement and Success (p = .152).
To check for residual heterogeneity, which is the remaining variability between the studies not accounted for by the moderators, we fitted a model with all the significant moderator variables.After adjusting for these variables, 0.34% of the total variance was attributed to the sampling error variance (level 1), 15.97% to differences within studies (level 2); and 83.69% of the total variance could be attributed to differences between studies (level 3).

Discussion
This systematic review and meta-analysis reports on the prevalence of community-dwelling older adults, as well as the impact of the used measurement instrument, mode of data collection, and country on reported prevalence percentages.Using 101 prevalence percentages from 45 studies, our study demonstrates that the pooled prevalence of loneliness among community-dwelling older adults is 31.6%.This percentage corresponds greatly to the percentage of a previous systematic review and meta-analysis on the prevalence of loneliness among older people in high-income countries (not explicitly community-dwelling), which was 28.5% (Chawla et al., 2021).Our results show that the (level 3) variance of pooled loneliness prevalence that can be explained was 63% by differences in the used measurement instrument and 8% by data collection method.
The prevalence of loneliness is lower for singleitem questions (21.2%) and shortened UCLA scales (25.0%), compared to the 20-item UCLA (59.3%) and the De Jong Gierveld loneliness scale (55.4%),where the loneliness rates are significantly higher.This might be a result of the fact that singleitem questions, and by extension short measurement scales, may be more vulnerable to certain biases in interpretation and meaning as well as on social desirability, and that multiple-item scales are more prone to cover the whole range of a complex construct, such as, in this case, loneliness (Hoeppner et al., 2011).Looking at the used measurement instruments, single-item questions are indeed more often used despite the existence of validated instruments and despite the critiques on single-item questions mentioning that these cannot capture a construct in all its complexity (Mund et al., 2022).
For the mode of data collection, loneliness prevalence rates vary from 14.6% for telephone interviews (including CATI) to 39.8% for CAPI.
A study specifically about the De Jong Gierveld loneliness scale suggests that data collection procedures indeed can have an impact on the motivation, accuracy, and self-disclosure of the participants while being subject to the data collection (van Tilburg & de Leeuw, 1991), and this is thus also visible in our review.
Regarding the country, four of the six dimensions of Hofstede (Hofstede, 2011) caused a significant increase (Power Distance Index, Uncertainty Avoidance Index, Indulgence) or a decrease (Individualism) in loneliness prevalence.Also here, we see that country, and more broadly, culture (Jylhä & Jokela, 1990), should be taken into account when making statements about loneliness prevalence among community-dwelling older adults.
The main strengths of this study are that the search strategy and the analyses were thoughtfully carried out and the choice of prevalence studies specifically on community-dwelling older people was made consciously, as we assessed the risk of bias very thoughtfully through our selection process.Moreover, despite the high heterogeneity of our pooled prevalence percentages, we assessed the quality of our studies carefully utilizing the JBI Critical Appraisal checklist (Munn et al., 2020), so that a high quality of the included studies and their data collection methods and measures was ensured.However, results from this study should also be viewed with caution in light of its limitations.First, although a comprehensive search is seen as a potential mechanism for minimizing bias (Cooper et al., 2018), our selection criteria were rather strict.It is possible that because of this, certain percentages were excluded while they would have been included if the criteria were less rigorous.Another limitation is that the field of loneliness research is a rapidly evolving research area, certainly as a result of the COVID pandemic (Lampraki et al., 2022).This means that we could have missed certain studies published since our search was conducted.Third, not all world regions were equally represented in our study: in our systematic review, a low number of prevalence percentages obtained in Africa (k = 2) and Oceania (k = 5) were included, mainly due to the lack of loneliness prevalence studies from these regions, in contrast to prevalence percentages originating in Europe (k = 127) or Asia (k = 31).To capture the diversity in the included countries, however, we used Hofstede's dimensions (Hofstede, 2011).Although we were aware of the prevailing criticisms surrounding this model (Chun et al., 2021;Minkov, 2018), the standardized scores and the possibility of comparing countries were decisive to incorporate them in this study.Fourth, we split the UCLA scale into two groups in our analyses (i.e., the original and the shortened scale separately), but this was not done for the DJG due to an insufficient number of prevalence percentages in the two subcategories (i.e., the original vs. the shortened version) to be allowed to conduct separate statistical analyses.Possibly, more studies with the 11-item DJG could provide additional information on the differences between the original and the abbreviated scale.Furthermore, in our meta-analysis, we could not include several possible moderators because they were not consistently mentioned, such as the year of data collection (k = 11) or the percentage of men or women (k = 46), or because the sample size was relatively small (only k = 34 had a sample size of > 1000).Also, age was not included as a moderator because information on age in the studies was incomplete or too heterogeneous.For example, in some studies, the age classes of 60-69, 70-79, and 80 + were used, while 60-74 and 75 + were used in other studies.In addition, numerous studies simply give little or no information on age: several mention a general age range of their participants (e.g., 60-85), but there was no further information on the difference in loneliness prevalence for different ages or age groups.
Future prevalence studies are therefore recommended to comprehensively capture participants' characteristics, including potential loneliness risk factors such as education, marital status, percentage of people living alone, etc., which were frequently absent in the current studies.Additionally, while existing studies differentiate types of loneliness (social, emotional, and existential), specific prevalence percentages for these types of loneliness are often lacking.
This study reviewed the prevalence of loneliness among community-dwelling older adults.Our results show that measurement instruments, mode of data collection, and country acted as moderator variables, leading to varying loneliness prevalence percentages.Nevertheless, considerable variation within and between studies suggests the influence of other factors, such as participant age and gender.Future prevalence studies should consider the contextual impact, including respondents' personal and cultural characteristics, as well as study design, on reported loneliness prevalence rates.