Using the Veil of Ignorance to align AI systems with principles of justice.

The philosopher John Rawls proposed the Veil of Ignorance (VoI) as a thought experiment to identify fair principles for governing a society. Here, we apply the VoI to an important governance domain: artificial intelligence (AI). In five incentive-compatible studies (N = 2, 508), including two preregistered protocols, participants choose principles to govern an Artificial Intelligence (AI) assistant from behind the veil: that is, without knowledge of their own relative position in the group. Compared to participants who have this information, we find a consistent preference for a principle that instructs the AI assistant to prioritize the worst-off. Neither risk attitudes nor political preferences adequately explain these choices. Instead, they appear to be driven by elevated concerns about fairness: Without prompting, participants who reason behind the VoI more frequently explain their choice in terms of fairness, compared to those in the Control condition. Moreover, we find initial support for the ability of the VoI to elicit more robust preferences: In the studies presented here, the VoI increases the likelihood of participants continuing to endorse their initial choice in a subsequent round where they know how they will be affected by the AI intervention and have a self-interested motivation to change their mind. These results emerge in both a descriptive and an immersive game. Our findings suggest that the VoI may be a suitable mechanism for selecting distributive principles to govern AI.

Simulated nonlinear genetic and environmental dynamics of complex traits.

Genetic studies of complex traits often show disparities in estimated heritability depending on the method used, whether by genomic associations or twin and family studies. We present a simulation of individual genomes with dynamic environmental conditions to consider how linear and nonlinear effects, gene-by-environment interactions, and gene-by-environment correlations may work together to govern the long-term development of complex traits and affect estimates of heritability from common methods. Our simulation studies demonstrate that the genetic effects estimated by genome wide association studies in unrelated individuals are inadequate to characterize gene-by-environment interaction, while including related individuals in genome-wide complex trait analysis (GCTA) allows gene-by-environment interactions to be recovered in the heritability. These theoretical findings provide an explanation for the "missing heritability" problem and bridge the conceptual gap between the most common findings of GCTA and twin studies. Future studies may use the simulation model to test hypotheses about phenotypic complexity either in an exploratory way or by replicating well-established observations of specific phenotypes.

Prediction of conversion to dementia disorders based on timed up and go dual-task test verbal and motor outcomes: a five-year prospective memory-clinic-based study.

BACKGROUND: While assessment tools can increase the detection of cognitive impairment, there is currently insufficient evidence regarding clinical outcomes based on screening for cognitive impairment in older adults. METHODS: The study purpose was to investigate whether Timed Up and Go dual-task test (TUGdt) results, based on TUG combined with two different verbal tasks (name different animals, TUGdt-NA, and recite months in reverse order, TUGdt-MB), predicted dementia incidence over a period of five years among patients (N = 186, mean = 70.7 years; 45.7% female) diagnosed with Subjective Cognitive Impairment (SCI) and Mild Cognitive Impairment (MCI) following assessment at two memory clinics. Associations between TUG parameters and dementia incidence were examined in Cox regression models. RESULTS: During follow-up time (median (range) 3.7 (0.1-6.1) years) 98 participants converted to dementia. Novel findings indicated that the TUGdt parameter words/time, after adjustment for age, gender, and education, can be used for the prediction of conversion to dementia in participants with SCI or MCI over a period of five years. Among the TUG-related parameters investigated, words/time showed the best predictive capacity, while time scores of TUG and TUGdt as well as TUGdt cost did not produce significant predictive results. Results further showed that the step parameter step length during TUGdt predicts conversion to dementia before adjustment for age, gender, and education. Optimal TUGdt cutoffs for predicting dementia at 2- and 4-year follow-up based on words/time were calculated. The sensitivity of the TUGdt cutoffs was high at 2-year follow-up: TUGdt-NA words/time, 0.79; TUGdt-MB words/time, 0.71; reducing respectively to 0.64 and 0.65 at 4-year follow-up. CONCLUSIONS: TUGdt words/time parameters have potential as cost-efficient tools for conversion-to-dementia risk assessment, useful for research and clinical purposes. These parameters may be able to bridge the gap of insufficient evidence for such clinical outcomes. TRIAL REGISTRATION: ClinicalTrials.gov Identifier: NCT05893524: https://www. CLINICALTRIALS: gov/study/NCT05893524?id=NCT05893524&rank=1 .

A systematic review of longitudinal risk factors for loneliness in older adults.

OBJECTIVES: To effectively reduce loneliness in older adults, interventions should be based on firm evidence regarding risk factors for loneliness in that population. This systematic review aimed to identify, appraise and synthesise longitudinal studies of risk factors for loneliness in older adults. METHODS: Searches were performed in June 2018 in PsycINFO, Scopus, Sociology Collection and Web of Science. Inclusion criteria were: population of older adults (M = 60+ years at outcome); longitudinal design; study conducted in an OECD country; article published in English in a peer-review journal. Article relevance and quality assessments were made by at least two independent reviewers. RESULTS: The search found 967 unique articles, of which 34 met relevance and quality criteria. The Netherlands and the United States together contributed 19 articles; 17 analysed national samples while 7 studies provided the data for 19 articles. One of two validated scales was used to measure loneliness in 24 articles, although 10 used a single item. A total of 120 unique risk factors for loneliness were examined. Risk factors with relatively consistent associations with loneliness were: not being married/partnered and partner loss; a limited social network; a low level of social activity; poor self-perceived health; and depression/depressed mood and an increase in depression. CONCLUSION: Despite the range of factors examined in the reviewed articles, strong evidence for a longitudinal association with loneliness was found for relatively few, while there were surprising omissions from the factors investigated. Future research should explore longitudinal risk factors for emotional and social loneliness.

Hierarchical Biometrical Genetic Analysis of Longitudinal Dynamics.

For many phenotypes, individual scores are obtained as the parameter estimates of person-level models fit to intensive repeated measures from physiological sensors or experience sampling studies. Biometrical genetic analysis of such phenotypes is often done in a 2-step sequence: first the phenotypic parameters are estimated for each individual, then classical twin modeling is used to partition their variance. This study demonstrates deficiencies in accuracy and statistical power of the two-step approach to estimate genetic signals and advocates for the use of hierarchical models to overcome both problems. Simulations are used to demonstrate the benefits to accuracy and statistical power from a hierarchical modeling approach. A model of heart rate fluctuations was applied to experimental data from twin pairs recorded in independent trials. Results of the data application reveal moderate but uncorrelated heritabilities for two parameters of heart rate: oscillation frequency and damping ratio. By merging biometrical genetic analysis with process models, hierarchical mixed-effects modeling has potential to assist with discovery and extraction of novel phenotypes from within-person data and to validate theoretical models of within-person processes.

Plasma tau, neurofilament light chain and amyloid-ß levels and risk of dementia; a population-based cohort study.

CSF biomarkers, including total-tau, neurofilament light chain (NfL) and amyloid-ß, are increasingly being used to define and stage Alzheimer's disease. These biomarkers can be measured more quickly and less invasively in plasma and may provide important information for early diagnosis of Alzheimer's disease. We used stored plasma samples and clinical data obtained from 4444 non-demented participants in the Rotterdam study at baseline (between 2002 and 2005) and during follow-up until January 2016. Plasma concentrations of total-tau, NfL, amyloid-ß40 and amyloid-ß42 were measured using the Simoa NF-light® and N3PA assays. Associations between biomarker plasma levels and incident all-cause and Alzheimer's disease dementia during follow-up were assessed using Cox proportional-hazard regression models adjusted for age, sex, education, cardiovascular risk factors and APOE ε4 status. Moreover, biomarker plasma levels and rates of change over time of participants who developed Alzheimer's disease dementia during follow-up were compared with age and sex-matched dementia-free control subjects. During up to 14 years follow-up, 549 participants developed dementia, including 374 cases with Alzheimer's disease dementia. A log2 higher baseline amyloid-ß42 plasma level was associated with a lower risk of developing all-cause or Alzheimer's disease dementia, adjusted hazard ratio (HR) 0.61 [95% confidence interval (CI), 0.47-0.78; P < 0.0001] and 0.59 (95% CI, 0.43-0.79; P = 0.0006), respectively. Conversely, a log2 higher baseline plasma NfL level was associated with a higher risk of all-cause dementia [adjusted HR 1.59 (95% CI, 1.38-1.83); P < 0.0001] or Alzheimer's disease [adjusted HR 1.50 (95% CI, 1.26-1.78); P < 0.0001]. Combining the lowest quartile group of amyloid-ß42 with the highest of NfL resulted in a stronger association with all-cause dementia [adjusted HR 9.5 (95% CI, 2.3-40.4); P < 0.002] and with Alzheimer's disease [adjusted HR 15.7 (95% CI, 2.1-117.4); P < 0.0001], compared to the highest quartile group of amyloid-ß42 and lowest of NfL. Total-tau and amyloid-ß40 levels were not associated with all-cause or Alzheimer's disease dementia risk. Trajectory analyses of biomarkers revealed that mean NfL plasma levels increased 3.4 times faster in participants who developed Alzheimer's disease compared to those who remained dementia-free (P < 0.0001), plasma values for cases diverged from controls 9.6 years before Alzheimer's disease diagnosis. Amyloid-ß42 levels began to decrease in Alzheimer's disease cases a few years before diagnosis, although the decline did not reach significance compared to dementia-free participants. In conclusion, our study shows that low amyloid-ß42 and high NfL plasma levels are each independently and in combination strongly associated with risk of all-cause and Alzheimer's disease dementia. These data indicate that plasma NfL and amyloid-ß42 levels can be used to assess the risk of developing dementia in a non-demented population. Plasma NfL levels, although not specific, may also be useful in monitoring progression of Alzheimer's disease dementia.

A comparison of spouse and non-spouse carers of people with dementia: a descriptive analysis of Swedish national survey data.

BACKGROUND: Being an informal carer of a person with dementia (PwD) can have a negative effect on the carer's health and quality of life, and spouse carers have been found to be especially vulnerable. Yet relatively little is known about the care provided and support received by spouse carers. This study compares spouse carers to other informal carers of PwDs regarding their care provision, the support received and the psychosocial impact of care. METHODS: The study was a cross-sectional questionnaire-based survey of a stratified random sample of the Swedish population aged 18 or over. The questionnaire explored how much care the respondent provided, the support received, and the psychosocial impact of providing care. Of 30,009 people sampled, 11,168 (37.7 %) responded, of whom 330 (2.95 %) were informal carers of a PwD. RESULTS: In comparison to non-spouse carers, spouse carers provided more care more frequently, did so with less support from family or the local authority, while more frequently experiencing negative impacts on their social life and psychological and physical health. Spouse carers also received more carer support and more frequently experienced a closeness in their relationship with the care-recipient. CONCLUSIONS: Spouse carers of PwD differed from non-spouse carers on virtually all aspects of their care situation. Policy and practice must be more sensitive to how the carer-care-recipient relationship shapes the experience of care, so that support is based on an understanding of the individual carer's actual needs and preferences rather than on preconceptions drawn from a generalised support model.

Cost analysis of informal care: estimates from a national cross-sectional survey in Sweden.

BACKGROUND: Over the past decades, informal care has increased in most OECD-countries. Informal care is costly to caregivers and to society in the form of lost income and direct costs of providing care. Existing evidence suggests that providing informal care affects caregivers' overall health. However, estimates of the social costs of informal care based on national data on individuals are currently scarce. OBJECTIVE: This study contributes to the existing evidence on the costs of informal care by estimating the direct and indirect costs to caregivers using a purposive national household survey from Sweden. METHODS: Adopting a bottom-up, prevalence approach, the direct and indirect costs are estimated using the survey data and the value of working time and leisure time from existing sources. RESULTS: The results suggest that around 15% of the adult population of Sweden provide informal care and that such care costs around SEK 152 billion per year (around 3% of GDP; USD 16,3 billion; EUR 14,5 billion), or SEK 128000 per caregiver. Around 55% of costs are in the form of income loss to caregivers. The largest cost items are reduced work hours and direct costs of providing informal care. Replacing informal caregivers with professional care providers would be costly at around SEK 193,6 billion per year. CONCLUSIONS: Findings indicate that, even in a country with a relatively generous welfare system, significant resources are allocated toward providing informal care. The costing analysis suggests that effective support initiatives to ease the burden of informal caregivers may be cost-effective.

Dual-task tests discriminate between dementia, mild cognitive impairment, subjective cognitive impairment, and healthy controls - a cross-sectional cohort study.

BACKGROUND: Discrimination between early-stage dementia and other cognitive impairment diagnoses is central to enable appropriate interventions. Previous studies indicate that dual-task testing may be useful in such differentiation. The objective of this study was to investigate whether dual-task test outcomes discriminate between groups of individuals with dementia disorder, mild cognitive impairment, subjective cognitive impairment, and healthy controls. METHODS: A total of 464 individuals (mean age 71 years, 47% women) were included in the study, of which 298 were patients undergoing memory assessment and 166 were cognitively healthy controls. Patients were grouped according to the diagnosis received: dementia disorder, mild cognitive impairment, or subjective cognitive impairment. Data collection included participants' demographic characteristics. The patients' cognitive test results and diagnoses were collected from their medical records. Healthy controls underwent the same cognitive tests as the patients. The mobility test Timed Up-and-Go (TUG single-task) and two dual-task tests including TUG (TUGdt) were carried out: TUGdt naming animals and TUGdt months backwards. The outcomes registered were: time scores for TUG single-task and both TUGdt tests, TUGdt costs (relative time difference between TUG single-task and TUGdt), number of different animals named, number of months recited in correct order, number of animals per 10 s, and number of months per 10 s. Logistic regression models examined associations between TUG outcomes pairwise between groups. RESULTS: The TUGdt outcomes "animals/10 s" and "months/10 s" discriminated significantly (p < 0.001) between individuals with an early-stage dementia diagnosis, mild cognitive impairment, subjective cognitive impairment, and healthy controls. The TUGdt outcome "animals/10 s" showed an odds ratio of 3.3 (95% confidence interval 2.0-5.4) for the groups dementia disorders vs. mild cognitive impairment. TUGdt cost outcomes, however, did not discriminate between any of the groups. CONCLUSIONS: The novel TUGdt outcomes "words per time unit", i.e. "animals/10 s" and "months/10 s", demonstrate high levels of discrimination between all investigated groups. Thus, the TUGdt tests in the current study could be useful as complementary tools in diagnostic assessments. Future studies will be focused on the predictive value of TUGdt outcomes concerning dementia risk for individuals with mild cognitive impairment or subjective cognitive impairment.

A Method of Correcting Estimation Failure in Latent Differential Equations with Comparisons to Kalman Filtering.

Studies have used the latent differential equation (LDE) model to estimate the parameters of damped oscillation in various phenomena, but it has been shown that correct, non-zero parameter estimates are only obtained when the latent series exhibits little or no process noise. Consequently, LDEs are limited to modeling deterministic processes with measurement error rather than those with random behavior in the true latent state. The reasons for these limitations are considered, and a piecewise deterministic approximation (PDA) algorithm is proposed to treat process noise outliers as functional discontinuities and obtain correct estimates of the damping parameter. Comprehensive, random-effects simulations were used to compare results with those obtained using a state-space model (SSM) based on the Kalman filter. The LDE with the PDA algorithm (LDEPDA) successfully recovered the simulated damping parameter under a variety of conditions when process noise was present in the latent state. The LDEPDA had greater precision and accuracy than the SSM when estimating parameters from data with sparse jump discontinuities, but worse performance for diffusion processes overall. All three methods were applied to a sample of postural sway data. The basic LDE estimated zero damping, while the LDEPDA and SSM estimated moderate to high damping. The SSM estimated the smallest standard errors for both frequency and damping parameter estimates.

Perspective: Clinical relevance of the dichotomous classification of Alzheimer's disease biomarkers: Should there be a "gray zone"?

The 2018 National Institute on Aging and the Alzheimer's Association (NIA-AA) research framework recently redefined Alzheimer's disease (AD) as a biological construct, based on in vivo biomarkers reflecting key neuropathologic features. Combinations of normal/abnormal levels of three biomarker categories, based on single thresholds, form the AD signature profile that defines the biological disease state as a continuum, independent of clinical symptomatology. While single thresholds may be useful in defining the biological signature profile, we provide evidence that their use in studies with cognitive outcomes merits further consideration. Using data from the Alzheimer's Disease Neuroimaging Initiative with a focus on cortical amyloid binding, we discuss the limitations of applying the biological definition of disease status as a tool to define the increased likelihood of the onset of the Alzheimer's clinical syndrome and the effects that this may have on trial study design. We also suggest potential research objectives going forward and what the related data requirements would be.

Cross-Lagged Analysis of Interplay Between Differential Traits in Sibling Pairs: Validation and Application to Parenting Behavior and ADHD Symptomatology.

Understanding the factors that contribute to behavioral traits is a complex task, and partitioning variance into latent genetic and environmental components is a useful beginning, but it should not also be the end. Many constructs are influenced by their contextual milieu, and accounting for background effects (such as gene-environment correlation) is necessary to avoid bias. This study introduces a method for examining the interplay between traits, in a longitudinal design using differential items in sibling pairs. The model is validated via simulation and power analysis, and we conclude with an application to paternal praise and ADHD symptoms in a twin sample. The model can help identify what type of genetic and environmental interplay may contribute to the dynamic relationship between traits using a cross-lagged panel framework. Overall, it presents a way to estimate and explicate the developmental interplay between a set of traits, free from many common sources of bias.

The importance of endpoint selection: How effective does a drug need to be for success in a clinical trial of a possible Alzheimer's disease treatment?

To date, Alzheimer's disease (AD) clinical trials have been largely unsuccessful. Failures have been attributed to a number of factors including ineffective drugs, inadequate targets, and poor trial design, of which the choice of endpoint is crucial. Using data from the Alzheimer's Disease Neuroimaging Initiative, we have calculated the minimum detectable effect size (MDES) in change from baseline of a range of measures over time, and in different diagnostic groups along the AD development trajectory. The Functional Activities Questionnaire score had the smallest MDES for a single endpoint where an effect of 27% could be detected within 3 years in participants with Late Mild Cognitive Impairment (LMCI) at baseline, closely followed by the Clinical Dementia Rating Sum of Boxes (CDRSB) score at 28% after 2 years in the same group. Composite measures were even more successful than single endpoints with an MDES of 21% in 3 years. Using alternative cognitive, imaging, functional, or composite endpoints, and recruiting patients that have LMCI could improve the success rate of AD clinical trials.

A systematic review of the psychometric properties of instruments for assessing the quality of the physical environment in healthcare.

AIM: To identify instruments measuring the quality of the physical healthcare environment, describe their psychometric properties. BACKGROUND: The physical healthcare environment is regarded as a quality factor for health care. To facilitate evidence-based design there is a need for valid and usable instruments that can evaluate the design of the healthcare environment. DESIGN: Systematic psychometric review. DATA SOURCES: A systematic literature search in Medline, CINAHL, Psychinfo, Avery index and reference lists of eligible papers (1990-2016). REVIEW METHOD: Consensus based standards for selection of health measurement instruments guidelines were used to evaluate psychometric data reported. RESULTS: Twenty-three instruments were included. Most of the instruments are intended for healthcare environments related to the care of older people. Many of the instruments were old, lacked strong, contemporary theoretical foundations, varied in the extent to which they had been used in empirical studies and in the degree to which their validity and reliability had been evaluated. CONCLUSIONS: Although we found many instruments for measuring the quality of the physical healthcare environment, none met all of our criteria for robustness. Of the instruments, The Multiphasic environmental assessment procedure, The Professional environment assessment protocol and The therapeutic environment screening have been used and tested most frequently. The Perceived hospital quality indicators are user centred and combine aspects of the physical and social environment. The Sheffield care environment assessment matrix has potential as it is comprehensive developed using a theoretical framework that has the needs of older people at the centre. However, further psychometric and user-evaluation of the instrument is required.

The association between the physical environment and the well-being of older people in residential care facilities: A multilevel analysis.

AIMS: To investigate the associations between the quality of the physical environment and the psychological and social well-being of older people living in residential care facilities. BACKGROUND: Many older people in care facilities have cognitive and physical frailties and are at risk of experiencing low levels of well-being. High-quality physical environments can support older people as frailty increases and promote their well-being. Although the importance of the physical environment for residents' well-being is recognized, more research is needed. DESIGN: A cross-sectional survey of 20 care facilities from each of which 10 residents were sampled. As the individual resident data were nested in the facilities, a multilevel analysis was conducted. METHODS: Data were collected during 2013 and 2014. The care facilities were purposely sampled to ensure a high level of variation in their physical characteristics. Residents' demographic and health data were collected via medical records and interviews. Residents' well-being and perceived quality of care were assessed via questionnaires and interviews. Environmental quality was assessed with a structured observational instrument. RESULTS: Multilevel analysis indicated that cognitive support in the physical environment was associated with residents' social well-being, after controlling for independence and perceived care quality. However, no significant association was found between the physical environment and residents' psychological well-being. CONCLUSION: Our study demonstrates the role of the physical environment for enhancing the social well-being of frail older people. Professionals and practitioners involved in the design of care facilities have a responsibility to ensure that such facilities meet high-quality specifications.

The physical environment, activity and interaction in residential care facilities for older people: a comparative case study.

The physical environment is of particular importance for supporting activities and interactions among older people living in residential care facilities (RCFs) who spend most of their time inside the facility. More knowledge is needed regarding the complex relationships between older people and environmental aspects in long-term care. The present study aimed to explore how the physical environment influences resident activities and interactions at two RCFs by using a mixed-method approach. Environmental assessments were conducted via the Swedish version of the Sheffield Care Environment Assessment Matrix (S-SCEAM), and resident activities, interactions and locations were assessed through an adapted version of the Dementia Care Mapping (DCM). The Observed Emotion Rating Scale (OERS) was used to assess residents' affective states. Field notes and walk-along interviews were also used. Findings indicate that the design of the physical environment influenced the residents' activities and interactions. Private apartments and dining areas showed high environmental quality at both RCFs, whereas the overall layout had lower quality. Safety was highly supported. Despite high environmental quality in general, several factors restricted resident activities. To optimise care for older people, the design process must clearly focus on accessible environments that provide options for residents to use the facility independently.

Assessing the physical environment of older people's residential care facilities: development of the Swedish version of the Sheffield Care Environment Assessment Matrix (S-SCEAM).

BACKGROUND: There is emerging evidence that the physical environment is important for health, quality of life and care, but there is a lack of valid instruments to assess health care environments. The Sheffield Care Environment Assessment Matrix (SCEAM), developed in the United Kingdom, provides a comprehensive assessment of the physical environment of residential care facilities for older people. This paper reports on the translation and adaptation of SCEAM for use in Swedish residential care facilities for older people, including information on its validity and reliability. METHODS: SCEAM was translated into Swedish and back-translated into English, and assessed for its relevance by experts using content validity index (CVI) together with qualitative data. After modification, the validity assessments were repeated and followed by test-retest and inter-rater reliability tests in six units within a Swedish residential care facility that varied in terms of their environmental characteristics. RESULTS: Translation and back translation identified linguistic and semantic related issues. The results of the first content validity analysis showed that more than one third of the items had item-CVI (I-CVI) values less than the critical value of 0.78. After modifying the instrument, the second content validation analysis resulted in I-CVI scores above 0.78, the suggested criteria for excellent content validity. Test-retest reliability showed high stability (96% and 95% for two independent raters respectively), and inter-rater reliability demonstrated high levels of agreement (95% and 94% on two separate rating occasions). Kappa values were very good for test-retest (κ = 0.903 and 0.869) and inter-rater reliability (κ = 0.851 and 0.832). CONCLUSIONS: Adapting an instrument to a domestic context is a complex and time-consuming process, requiring an understanding of the culture where the instrument was developed and where it is to be used. A team, including the instrument's developers, translators, and researchers is necessary to ensure a valid translation and adaption. This study showed preliminary validity and reliability evidence for the Swedish version (S-SCEAM) when used in a Swedish context. Further, we believe that the S-SCEAM has improved compared to the original instrument and suggest that it can be used as a foundation for future developments of the SCEAM model.

Predictors of loneliness among older women and men in Sweden: A national longitudinal study.

OBJECTIVES: Longitudinal research on loneliness in old age has rarely considered loneliness separately for men and women, despite gender differences in life experiences. The objective of this study was to examine the extent to which older women and men (70+) report feelings of loneliness with a focus on: (a) changes in reported loneliness as people age, and (b) which factors predict loneliness. METHOD: Data from the 2004 and 2011 waves of SWEOLD, a longitudinal national survey, was used (n = 587). The prediction of loneliness in 2011 by variables measured in 2004 and 2004-2011 variable change scores was examined in three logistic regression models: total sample, women and men. Variables in the models included: gender, age, education, mobility problems, depression, widowhood and social contacts. RESULTS: Older people moved into and out of frequent loneliness over time, although there was a general increase in loneliness with age. Loneliness at baseline, depression increment and recent widowhood were significant predictors of loneliness in all three multivariable models. Widowhood, depression, mobility problems and mobility reduction predicted loneliness uniquely in the model for women; while low level of social contacts and social contact reduction predicted loneliness uniquely in the model for men. CONCLUSION: This study challenges the notion that feelings of loneliness in old age are stable. It also identifies important gender differences in prevalence and predictors of loneliness. Knowledge about such differences is crucial for the development of effective policy and interventions to combat loneliness in later life.