Trajectories of loneliness and objective social isolation and associations between persistent loneliness and self-reported personal recovery in a cohort of secondary mental health service users in the UK

Participants were drawn from the Crisis Team Optimisation and Relapse Prevention (CORE) study. This large multi-site randomised controlled trial was primarily designed to test the effectiveness of a peer-provided self-management intervention versus treatment as usual supplemented by a self-management booklet on readmission rate for people with mental health problems, delivered following discharge from CRTs in six NHS trusts, including inner city, suburban and rural environments. Trial inclusion criteria and recruitment procedures are fully described in the published trial protocol [23]. For this paper, we pool participants in both arms of the trial to form a single cohort. A detailed baseline description of this cohort and the variations associated with loneliness at baseline has been previously published, however, in the original RCT, there was no evidence for an effect of the intervention on the primary outcome of the current study (i.e., self-rated personal recovery) [26].

Inclusion criteria for the CORE study required participants to have the capacity to provide written informed consent to participate and to have received support from one of the CRTs for a week or longer. Exclusion criteria included: 1) being clinically assessed as being too high risk to others to participate safely in the study or trial intervention; 2) lacking sufficient knowledge in English to complete the assessments; and 3) CRT discharge having occurred more than a month ago.

The CORE study was approved by the London Camden and Islington Research Ethics Committee (ref 12/LO/0988) before the commencement of data collection. Written informed consent was received from all participants prior participation. All methods were performed in accordance with the relevant guidelines and regulations.

All participants recruited for the CORE study (i.e., both experimental and control groups) were assessed at three timepoints using structured interviews: baseline, 4- and 18-month follow-up. Sociodemographic information was only collected at baseline. The measures used in this research are reported below.

Case mean substitution was implemented to replace missing data on continuous variables, including loneliness, social network size and personal recovery. Given the low level of missing data on all scales at both baseline and 18-month follow-up in the current study, case mean substitution should have equivalent effectiveness as other missing data techniques [33]. Following established guidance [34, 35], cases with over 25% of data missing on a single scale were excluded from the final analysis. Pearson’s correlation was used to check for potential collinearity between all baseline independent variables.

T-test and chi-square tests were used to check predictors of missingness for the outcomes. When baseline variables were found to be different between the participants who completed the ULS-8 and LSNS-6 at all three timepoints (i.e., completers) and those who did not (i.e., non-completers), these variables were controlled for in the final analysis. The primary outcome, that is the association between the three loneliness groups and personal recovery at 18-month follow-up was tested by multivariable linear regression analyses. Firstly, a univariate linear regression model was carried out. Then four blocks of baseline variables were added into the multivariable regression models, in the following steps: 1) baseline social network size; 2) baseline sociodemographic variables (i.e. age, gender, ethnicity, educational attainment, and whether employed or in education or not) and baseline social network size; 3) baseline sociodemographic variables, baseline psychiatric variables (i.e. number of years since first contact with mental health services, the Brief Psychiatric Rating Scale (BPRS) total score, number of psychiatric inpatient hospitalisation), and baseline social network size; and 4) baseline sociodemographic and clinical variables, baseline social network size, and baseline QPR score. Comparable five steps were also carried out to examine the association between the three objective social isolation groups and personal recovery at 18-month follow-up.

In total, 399 participants were included in the baseline analysis. Baseline characteristics of our sample are described in Additional file 1. A detailed description of participants at baseline has been previously published [24]. Of these 399 participants, only 224 participants completed measures at all three timepoints and could be included in the final analysis for research question 2. Please see Table 1 for the characteristics of participants at 18-month follow-up. The mean age of these participants was 40.0 (SD = 12.5), and about 40% were male. Approximately 30% of this sample was diagnosed with schizophrenia or other psychotic disorders. The mean of the QPR reported by these participants was 43.3 (SD = 10.7). Reported loneliness at 18-month follow-up had a mean score of 22.1 (SD = 4.5), and the mean of 18-month social network size was 5.0 (SD = 2.3).

Educational attainment and employment status were the only two variables identified as predictors of missingness at a p = 0.05 level of significance, so they were included in multivariable analyses for research question 2. Pearson’s correlation revealed that the correlation between baseline variables ranged from − 0.27 to 0.39. Based on the cut-off proposed by existing literature [36], collinearity was unlikely to be a problem.

The proportion of participants in each loneliness group is shown in Fig. 1. Of the 224 participants, 36 (16%) participants were in the persistently severely lonely group, 113 (50%) participants were never severely lonely, and the rest of the sample (n = 75, 34%) was intermittently severely lonely. Of these 75 participants in the intermittently severely lonely group, 32 participants were lonely at an earlier timepoint but not subsequently (Groups 3 and 4); 19 were not lonely at an earlier timepoint but reported being so later (Groups 6 and 7), and another 24 participants experienced fluctuating loneliness (Groups 5 and 8).

Of the same 224 participants, 28 participants (13%) met our criteria for persistent objective social isolation, 124 (55%) participants were not objectively socially isolated at any timepoint, and the rest of the sample (n = 72, 32%) met the criteria for the intermittent social isolation group. Of these 72 participants, a pathway out of objective social isolation was observed for 28 participants (Groups 3 and 4); a pathway into objective social isolation was observed for 19 participants (Groups 6 and 7), and another 25 participants experienced fluctuating objective social isolation (Groups 5 and 6) (Fig. 2).

The characteristics of participants in different loneliness groups are shown in Table 2. Not only were the participants in the persistent severe loneliness group more likely to be single, separated, divorced or widowed (p < 0.001), they were also less likely to be employed, or in education, or any full-time caring role (p = 0.002) than those who were never severely lonely. The persistent severe loneliness group had the smallest social network, followed by the intermittent severe loneliness group. There were significant differences in the QPR score at baseline between the three loneliness groups: persistent severe loneliness group scored the lowest on the QPR. This group was followed by the intermittent severe loneliness group. The group who was never severely lonely scored the highest on the QPR.

The characteristics of participants in different social isolation groups are described in Table 3. The comparisons between the three objective social isolation groups indicated that the participants in the persistent objective social isolation group were less likely to be born in the UK, were less likely to be employed, in education or any full-time caring role (p = 0.01), and they were less likely to have been admitted as a psychiatric inpatient previously (p = 0.001), compared to the participants who were never objectively socially isolated during the study period. They were also more likely to have a diagnosis of depression, or anxiety disorder or post-traumatic stress disorder (p = 0.03), rather than having a diagnosis of psychosis, bipolar disorders, personality disorders or other mental health diagnoses. Statistically significant differences were also reported for their baseline loneliness score: the persistent objective social isolation group scored the highest on the ULS-8, followed by the intermittent objective social isolation group. The persistent objective social isolation group also scored lowest on the QPR. The group who was not socially isolated at any timepoint had the highest QPR score. The detailed results are shown in Table 3.

Results from the multivariable linear regression analyses demonstrate a significant relationship between loneliness group and self-rated personal recovery at 18-month follow-up. As shown in model 1, with being never severely lonely as the reference category, intermittent severe loneliness was associated with a statistically significant 9.8-point lower score on the QPR (95% CI -13.59, − 6.02), persistent severe loneliness was associated with a 21.75-point lower score on the QPR (95% CI -26.58, − 16.93). These associations remained statistically significant in the rest of the models, after controlling for the three blocks of baseline variables (i.e., sociodemographic and psychiatric variables, and social network size) and baseline QPR. In the final model (i.e., model 5), persistent (coef. = − 12.8, 95% CI -18.83, − 6.83, p < 0.001) and intermittent severe loneliness (coef. = − 7.8, 95% CI -11.80, − 3.75, p < 0.001) were both associated with a lower QPR score at 18-month follow-up than for the group who did not report severe loneliness at any timepoint. One clinical variable ‘2-10 years since first contact with mental health services’ (i.e., compared with ‘less than 3 months since first contact with mental health services’ as a reference category) was also negatively associated with the QPR at 18-month follow-up in model 5. The detailed results are shown in Table 4.

Multivariable linear regression analyses demonstrate a significant relationship between the objective social isolation groups and self-rated personal recovery at 18-month follow-up. In model 1, persistent objective social isolation was the only group significantly associated with a lower QPR at 18-month follow-up (with being never objectively socially isolated as a reference category). This relationship remained statistically significant even after controlling for the three blocks of baseline variables and baseline QPR score (coef. = − 9.8, 95% CI -15.71, − 3.79, p = 0.001). In the final model (i.e., model 5), the QPR at 18-month follow-up was also negatively associated with ‘2-10 years and over 10 years since first contact mental health services’ (i.e., with ‘less than 3 months since first contact mental health services’ as a reference category). The detailed results are presented in Table 5.

Our study included a diagnostically diverse sample, and the median baseline loneliness was equivalent to a moderate level of loneliness. Baseline loneliness score for our cohort is comparable to people with diagnoses across the entire spectrum of mental disorders [37, 38]. Our sample also reported having seen or heard from approximately five family members and friends in the previous month, which is comparable to previous studies of people with mental health problems [39, 40].

Whereas much research in people with psychosis or in mixed groups of mental health service users has been cross-sectional in nature, the three timepoints in this study allowed a novel exploration of associations between personal recovery and loneliness trajectory. While around 16% were lonely at all three study timepoints and 13% socially isolated at all time points, around twice as many met such criteria at only some timepoints. This suggests that a longitudinal perspective on loneliness and social isolation is valuable, as there is considerable fluctuation: as in other populations [20], it may be that some are transiently lonely or less in contact with others in response to important transitions, economic situations or fluctuations in health, while for others loneliness and social isolation may be much more enduring difficulties that they lack resources to address themselves.

While age, sex and ethnic group did not vary by loneliness trajectory (which may highlight the fact that loneliness is a universal experience that everyone may experience, regardless of one’s age, gender or ethnic background), in exploratory analyses the persistently lonely group was also more likely to be single, separated, divorced or widowed, and to be unemployed, not in education or any full-time caring role. Both marriage or cohabiting status (11% of persistently lonely vs. 38% of never lonely) and being employed at baseline (31% of persistently lonely vs. 66% of never lonely) had large associations with subsequently loneliness trajectory. These findings are in line with previous research, in which being in a stable and supportive relationship can bring positive influence on one’s physical and emotional wellbeing, for example, a sense of belonging, the feeling of be cared for and loved [41], all of which may subsequently reduce the risk of loneliness [42]. In terms of employment status, being unemployed has been linked to more financial hardships [43], which may restrict one’s access to a wide range of social activities, and workplaces may also be a significant source of friendship and social contact. The persistently lonely group also had more severe symptoms at baseline, and a large difference in self-rated recovery was already present. The group whose trajectory showed relatively severe loneliness at some but not all time points were intermediate between the persistently and never lonely groups in their initial symptom and recovery scores, and in baseline loneliness scores and social network size. Thus, at baseline, the group whose subsequent trajectory was defined at persistently lonely, were already disconnected from others in several respects, with poorer scores for both recovery and symptom severity. Prolonged loneliness has been argued to be especially significant in across age groups in the general population [20] and in our study, those who were lonely at multiple timepoints appear a group warranting attention in future research and intervention development.

People who were socially isolated at each time point, were likewise markedly more likely to be unemployed and not cohabiting or married, had more severe symptoms and rated their recovery as poorer at baseline. There was likewise a gradient between those who appeared socially isolated at each time point, those for whom this fluctuated, and those who were not isolated at any point.

To the best of our knowledge, the current study is the first study examining the detrimental impact of persistent loneliness and objective social isolation on self-rated personal recovery among mental health service users. Our primary hypothesis, that people who were lonely at multiple timepoints would show less improvement in self-rated recovery was confirmed, with a persisting effect following adjustment for clinical and social variables and baseline recovery score (with recovery already poorer at baseline). This is compatible with a causal relationship between persistent loneliness and poor improvement in self-rated recovery, although the caveats below in the limitations must be noted. Social connection is conceptualised as an aspect of personal recovery [44, 45], making interpretation of their relationship over time more complex. However, our findings at least suggest that people who are and remain severely lonely and/or social isolated, with few connections with others, are a group worthy of a specific focus as they appear to recover poorly following a crisis. Despite a less prominent effect for objective social isolation on self-rated personal recovery than that for loneliness, we may speculate a shared effect of loneliness and objective social isolation on personal recovery. Loneliness and objective social isolation tend to co-occur. Not only there is a significant correlation between loneliness and objective social isolation [46], the two experiences also share some same contributing factors, including small social network and infrequent social connections with family and friends [47]. We may therefore expect that, in the current study, the majority of severely lonely participants were also socially isolated, which may further exacerbate their joint negative effect on personal recovery. The mechanisms through which persistent severe loneliness and persistent objective social isolation might impact one’s personal recovery remain unexplored. The interrelationships between a number of underlying factors, such as self-esteem, public and self-stigma, sociodemographic backgrounds, and social deprivation, may play a part.

The current study benefits from several main strengths: 1) we included a diagnostically varied sample which was recruited from standard secondary mental health services in the UK; 2) and they were at an illness stage that is of high clinical relevance; and 3) to the best of our knowledge, the current study is the first investigating the chronicity of loneliness and objective social isolation, and their impact on personal recovery among people with mental health problems. However, our results also subject to certain limitations: 1) our study only included mental health patients who were willing to participate in an RCT; 2) the current study had a large proportion of missing data on the diagnosis variable; 3) the attrition rate of this trial at 18-month follow-up was relatively high (i.e., 44%), therefore, the final sample of the current study only included 56% of those who completed baseline measures. Reasons for drop-out could include being unwell or having moved away, and those who dropped out may not be fully representative of those originally recruited. It is possible that this high attrition rate may compromise the generalisability of our findings. However, t-test and chi-square tests were used to compare the differences between the participants who completed measures at all three timepoints and those who did not. Educational attainment and employment status were found to be the two predictors of missingness and were subsequently controlled for in the final analysis; 4) despite the well-established scales included in the current study, the quality of our measures is subject to limitations. For example, despite being widely used in this population [5], the ULS-8 was not originally developed for people with mental health problems, and lacks clear thresholds for severe and moderate loneliness, so that we needed to specify these pragmatically for the study. In terms of the measure for social network size, given that the LSNS-6 measures both subjective and objective aspects of one’s social relationships, we had to make our own selection of items from the scale in order to focus only on objective social isolation, assessed by numbers of contracts. The measurement of loneliness at more than one timepoint is an advance over many previous studies: however, that loneliness and self-rated personal recovery were measured over the same time period and that they are relatively closely related concepts, we cannot make confident statements regarding the causal link between the two concepts.

Results of our study carry some important implications for future research and clinical practice. In the current study, 16% of our cohort were persistently severely lonely, and 13% suffered from persistent objective social isolation. Large-scale longitudinal cohort studies will be valuable in providing rigorous evidence concerning the extent and the impact of persistent severe loneliness and persistent objective social isolation among mental health service users. The significance of prolonged loneliness across people in different age groups in the general population has been acknowledged [20] and the current study suggests that people who are persistently lonely might also be a group of interest for further investigation and intervention development among people with longer-term mental health problems. People who are lonely or report being relatively isolated at times, but in whom improvements are observed may be a group who have more resources to remedy their own lack of connections: understanding this process better would also contribute to a greater understanding of how lack of connection may improve. Despite widespread acknowledgement of the social determinants of mental health, social interventions still tend to receive less attention in mental health care than pharmacological or psychological approaches.

Given the poor outcomes associated with persistently loneliness and social isolation, routine enquiry into loneliness and the extent of social networks as part of assessments seems desirable so that practitioners are at least aware of these potentially remediable factors. The Office for National Statistics [48] has recommended a quick screening tool, which involves three questions from the ULS-3 and a direct question as an efficient approach to identify and address loneliness in all public settings and health practices. Finally, there is a growing interest in the development of interventions targeting loneliness among mental health service users with various diagnoses [49, 50], our results suggest that targeting people who are lonely and/or socially isolated at multiple timepoints may be a fruitful approach. So far approaches to loneliness and social isolation among people with mental health problems [49] have tended to target a broad group identified as lonely and/or socially isolated on cross-sectional assessment. However, our finding that recovery from loneliness and/or social isolation at a single timepoint was frequent and outcomes were worst for those who do not recovery, fits with research in other populations suggesting that persistent loneliness is especially significant [20]. Thus, it may make sense to focus development and evaluation of interventions to reduce loneliness and/or social isolation in a mental health context on those who report being persistently lonely, with our results suggesting potential benefits for recovery. Currently there are not to our knowledge interventions with well-established effectiveness and cost-effectiveness among people with mental health problems [49], but interventions with a cognitive modification component for loneliness and supported socialisation for objective social isolation show some promise [49, 51]. Approaches found to be effective for loneliness or social isolation in other populations, such as digital interventions and support groups [52, 53], may also have potential to be adapted and tested for people with mental health problems. Our findings underscore the potential value of continuing intervention development and evaluation in this area, with people with persistent loneliness and/or social isolation a group for whom such intervention may have more potential value than in wider groups found to be lonely only at a single time point. However, although potential interventions with at least evidence, the implementation of promising interventions including interventions with a cognitive modification component and interventions that support socialisation for loneliness and objective social isolation [49], respectively, are warranted.