Data

Data came from the Understanding Society UK Household Longitudinal Study [33], a longitudinal survey containing approximately 40,000 households. Understanding Society collects data from UK adults aged 16 and over through self-completion online surveys or through face-to-face interviews. These data are collected with the aim of studying the effect of social, economic, and policy change on wellbeing in the UK. Data are captured for all key variables of interest for this study, being loneliness, wellbeing, quality of life, and healthcare resource use, using validated scales where available. Loneliness has been measured in Understanding Society from wave 9 (2017−2019) onwards. This research utilises data from wave 13 (2021−2023) of the study, the most recent data available at the time of analysis with a post-COVID-19 focus. This avoids potential confounding from the pandemic in earlier waves (waves 10–12) and provides insight on the relationships in the post-pandemic ‘new normal’ [34]. Data are included from individuals with complete responses for all the variables of interest to this study.

Demographics

Demographic variables for age, sex, ethnicity, marital status, education, and employment status are reported. Age is reported as a continuous variable while sex (female/male), ethnicity (white/non-white), marital status (married/not married), education (higher/other education), and employment (employed/unemployed) are transformed and reported as binary.

Loneliness measures

Loneliness was assessed in two ways; using a single item question, and using the UCLA 3-item scale [35]. The single item question directly asks, ‘How often do you feel lonely?’ and has three response options: 1) hardly ever or never, 2) some of the time, 3) often. The UCLA 3-item scale consists of three questions each indirectly related to loneliness: ‘how often do you feel you lack companionship?’, ‘how often do you feel left out?’, and ‘how often do you feel isolated from others?’. Each of the three UCLA questions also have three response options: 1) hardly ever or never, 2) some of the time, 3) often. Thus, a higher score indicates a greater level of loneliness. For the UCLA 3-item scale the loneliness score is calculated as the sum of all items yielding a scale from three to nine. However, because the scale is constructed from categorical responses rather than continuous data, we interpret the UCLA loneliness score using the initial categories 1) UCLA loneliness score = 3–4 as ‘hardly ever or never’, 2) UCLA loneliness score = 5–7 as ‘some of the time’, 3) UCLA loneliness score = 8–9 as ‘often’ lonely [36]. This enables the analysis to retain the nuance inherent in the subjective nature of loneliness, recognising its varying levels of severity, commonly described as ‘some of the time’ or ‘often’, and is consistent with existing research which recognises the importance of the categorical classification [37]. Furthermore, this classification aligns with the format of the single item question of loneliness, allowing for consistency across instruments.

In our analysis we use these categorical definitions of loneliness to transform the responses to the UCLA score, and to the single item question, into binary variables. For primary analysis, we distinguish respondents that report being lonely ‘often’ and contrast these with respondents who report being lonely either ‘some of the time’ or ‘hardly ever or never’. This allows us to evidence the effect of experiencing more severe loneliness. For secondary analysis we seek to understand the impact of having any experience of loneliness and so group together respondents that report feeling lonely ‘some of the time’ or ‘often’, and contrast with respondents who report being ‘hardly ever or never’ lonely. This approach is consistent with the classifications used in existing research [3]. It aids interpretation of the loneliness score and facilitates consideration of any differences based on severity of experience.

Quality of life (QoL) measures

QoL measures were selected to cover various facets of quality of life and wellbeing. Measures were included for physical and mental health (12 item Short Form Health Survey, SF-12), psychological distress (12 item General Health Questionnaire, GHQ-12), and wellbeing (Short Warwick-Edinburgh Mental Well-being Scale, SWEMWBS).

Healthcare resource use and costs

Participants were asked about the number of GP appointments, outpatient and inpatients visits over a 12-month period. These services were selected given evidence of their potential association to loneliness in other settings [25,29,31,32]. To determine the treatment costs, unit costs were assigned to the number of healthcare encounters to understand the costs to the NHS associated to loneliness. NHS costs were estimated using unit costs from the 2023 Personal Social Services Research Unit [47]. As detailed in S1 Table, a GP visit (10-minutes consultation at surgery) would cost £49, an outpatient visit cost £217, and an inpatient day case costs £1,111 per day.

Costs were combined with mean resource use for analysis using standard methodology for economic evaluations [48,49]. Inpatient number of days were reported as a continuous free-text response, however, number of GP appointments and outpatient visits were reported on a predefined scale with response options grouped into ranges. Thus, GP and outpatient costs were generated for lower and upper bound estimates. Some participants reported to have attended more than 10 GP appointments or outpatient visits. In these cases, a lower bound of 11 visits was assigned, while the upper bound was based on existing literature. For GP visits an upper bound of 25 visits was assigned as identified in recent research on frequent attenders [50]. For outpatient visits data were taken from the World Health Organisation outpatient contacts per person per year [51] for the UK yielding an upper bound of 13 outpatient visits per year.

Statistical and econometric analysis

Descriptive statistics reporting mean, median, standard deviation and range are used to describe the data. Median costs are reported given the skewness of cost data. Unadjusted difference in mean QoL measures, and healthcare resource use outcomes for those reporting feeling lonely versus not lonely are reported to indicate how loneliness interacts with physical and mental health, psychological distress, well-being, behavioural attitudes and its costs to the NHS. Findings were then adjusted for age, sex, ethnicity, marital status, level of education and employment using a generalised linear model (GLM) with gamma distribution and log link [52]. Models included loneliness as a binary independent variable and each of the QoL and healthcare resource use outcomes as dependent variables in turn.

The UCLA score is used in the main analysis with the direct loneliness question included as sensitivity analysis to allow comparison between loneliness measures. Primary analysis defined loneliness as the experience of feeling lonely ‘often’. Secondary analysis expanded this definition to consider an individual as lonely when experiencing loneliness often or sometimes.

Adjusted and unadjusted models were run with dependent variables for upper and lower bound healthcare service use, and related cost estimates. Total costs were then estimated across six age groups [53] to consider changes across the life course. Analyses are conducted using Stata/SE version 18. Missing responses were excluded from analysis.

Ethical approval statement

This study has been approved by the University of Exeter Faculty of Health and Life Sciences Sports and Health Sciences Ethics Committee. The University of Essex Ethics Committee has approved all data collection on Understanding Society main study, COVID-19 surveys and innovation panel waves, including asking consent for all data linkages except to health records. Requesting consent for health record linkage was approved at Wave 1 by the National Research Ethics Service (NRES) Oxfordshire REC A (08/H0604/124), at BHPS Wave 18 by the NRES Royal Free Hospital & Medical School (08/H0720/60) and at Wave 4 by NRES Southampton REC A (11/SC/0274).

Descriptive statistics

Descriptive statistics for the sample are reported in Table 1. Complete data for all included measures provided observations for 23,071 respondents. Respondents had a mean age of 50 (SD = 19, range 16–101), and were 56% (12,800/23,071) female. Most of the sample were of white ethnicity (84%; 19,342/23,071). The sample was quite evenly split between married (54%; 12,343/23,071) and not currently married (47%; 10,728/23,071), and between having received higher education (43%; 9,976/23,071) or not (57%; 13,005/23,071). Two thirds of the respondents were currently employed or in education (65%; 14,928/23,071). For both the UCLA loneliness scale and the direct loneliness score, around 60% of respondents reported feeling lonely ‘hardly ever or never’, around 32% felt lonely some of the time, and around 8% indicated they felt lonely ‘often’. There is slight variation in these proportions for each of the UCLA dimensions as for example 8% of respondents reported ‘often’ feeling ‘lack of companionship’, ‘isolated from others’, while slightly fewer (6%) said they ‘often’ feel ‘left out’.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive statistics, n(%).

https://doi.org/10.1371/journal.pone.0327671.t001

The mean SF-12 PCS score for the sample was 49.8 (50 or less signifies a physical condition), and mean SF-12 MCS score was 47.4 (42 or less indicates clinical depression). The GHQ-12 mean score of 11.7 using the Likert scale, and 1.9 using the Caseness coding with both numbers indicating low mental distress. Mean SWEMWBS score for the sample was 24.3, similar to the population norm reported in previous studies [54], and suggesting middling wellbeing.

Most of the sample visited a GP no more than two times in the last 12 months (79%), of which 41% did not visit a GP at all in the last year. Meanwhile 2% of respondents visited a GP more than ten times in 12 months. Two thirds (64%) of the sample did not attend a hospital or clinic out-patient in the last 12 months while 2% attended more than 10 times. Most respondents (94%) had not attended a hospital or clinic as an in-patient in the last 12 months, 6% of the sample did attend, thus overall, the full sample reported a mean of 0.5 (SD = 4.4, range 0–200) inpatient days.

Difference in mean

Across all variables a greater association is observed between loneliness and the QoL related outcomes when defining loneliness as experiencing lonely often, than when including any experience of loneliness (often or sometimes) in the analysis. The average marginal effect (predicted mean difference) is slightly smaller for the adjusted than the unadjusted model, as would be expected when accounting for covariates. In both models however, for each variable a statistically significant (p < 0.05) average marginal effect is observed. Unadjusted scores for both loneliness measures are reported in S2 Table.

The main difference in mean analysis reported in this section is based on the UCLA loneliness scale, as reported in Table 2. The sensitivity analysis considering the single item question for loneliness is also reported in Table 2. There is little difference in key results when comparing the UCLA scale to the single item question for loneliness, though statistical significance of the difference between loneliness measures was not explored in this study.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Difference in mean (adjusted average marginal effects)ᶲ^Ϯ.

https://doi.org/10.1371/journal.pone.0327671.t002

Psychological distress, physical and mental health and wellbeing

The physical subscale of the SF-12 found lower functioning in individuals experiencing loneliness than no loneliness with adjusted average marginal effect (AME) −3.28 reducing to −2.61 when including sometimes lonely in the definition for loneliness. For the mental component score adjusted difference in mean determined that the AME fell between −9.92 (lonely often) and −14.75 (often or sometimes lonely) indicating lower mental functioning in those experiencing loneliness.

For the SF-12, a score of 50 or less on the PCS signifies a physical condition, and a score of 42 or less on the MCS indicates clinical depression [39,40]. For the ‘not lonely’ group, the mean (adjusted) SF-12 component scores for both the physical and mental subscales fall outside these thresholds, however for those experiencing loneliness, mean physical (PCS) and mental (MCS) scores indicate the presence of a physical condition, and clinical depression, respectively. Findings indicate loneliness is associated to greater psychological distress as measured by the GHQ-12 scale. Using the GHQ-12 Likert scale adjusted AME was 8.04 for those experiencing loneliness often and 5.08 when including feeling lonely some of the time in the definition for loneliness. For the Caseness scale difference in means were 4.10 and 2.45 respectively. Adjusted mean scores for the Caseness scale indicate that when loneliness is more extreme (lonely defined as experiencing loneliness often) individuals experiencing loneliness also experience poor mental health (GHQ-12 score more than 4), whereas the threshold is not reached for individuals not experiencing loneliness. The threshold is not reached for either group when including individuals experiencing loneliness ‘some of the time’ in the definition for loneliness.

Finally, lower positive mental wellbeing (SWEMWBS) with adjusted AME −6.39 (lonely often) and −4.69 (lonely often or sometimes) is observed. Again, the broader definition for loneliness indicates a score above the threshold for depression for both lonely and non-lonely individuals. However, when restricting loneliness to something experienced ‘often’, individuals who experience loneliness are also seen to experience mild depression with a SWEMWBS score in the threshold 18−20. Each of these results, for the SF-12, GHQ-12, and SWEMWBS achieve statistical significance of p < 0.001.

NHS service use and related costs

Table 2 shows that individuals experiencing loneliness have more GP appointments and more out-patient visits over a 12-month period than individuals not experiencing loneliness. For inpatient visits, AME for whether an inpatient visit occurred or not (binary variable) was twice as large when restricting the definition of loneliness to ‘often’ only (AME = 0.02 compared to 0.04). Additional detail on the number of in-patient days is also provided with adjusted AME 0.64 (often) and 0.24 for the wider definition of loneliness (often or sometimes) indicating a higher number of inpatient days associated to individuals experiencing loneliness than not experiencing loneliness. Median scores (unadjusted) are also reported for all variables, however no difference was observed between groups, as reported in S2 Table.

Table 3 reports the adjusted AME for GP and outpatient visits, based on the lower and upper bounds of the response options. The adjusted AME for GP visits is estimated to be between 0.74 and 1.47 where loneliness is defined as something experienced ‘often’, and between 0.48 and 0.93 when widening the definition of loneliness to include those experiencing it ‘some of the time’. In all cases more GP visits were observed amongst those experiencing loneliness. For outpatient hospital or clinic visits the adjusted difference in mean is estimated between 0.37 and 0.53 (often) or between 0.25 and 0.38 (often or sometimes) with individuals identified as lonely attending more frequently. In each case, for GP and for outpatient visits, the higher estimates are derived from the upper bound. Unadjusted median service use is reported in S3 Table with the same median use of GP and outpatient services reported in each group.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Adjusted cost of health care service use based on mean service use ᶲ^Ϯ.

https://doi.org/10.1371/journal.pone.0327671.t003

Total NHS service costs combine GP, outpatient, and inpatient visits and is consistently higher for individuals experiencing loneliness compared to individuals not experiencing loneliness (Table 3). The adjusted difference in mean cost is higher when defining loneliness as feeling lonely ‘often’ with difference in cost expected between £813.35 and £885.49 per person, whereas when defining loneliness as feeling loneliness ‘often or sometimes’ service cost is expected to be between £352.39 and £409.42. Median unadjusted costs are reported in S3 Table. When using the wider definition of loneliness median service use cost is the same in each group. When restricting the definition of loneliness to be only those experiencing it often, a greater median cost is observed amongst those experiencing loneliness for both the lower (£147 vs £49) and upper bound (£245 vs £98). Considering the cost components (Table 3), the greatest adjusted difference in mean cost arises from inpatient days with AME £706.23 and £262.01 for the restricted (often) and less restricted (often or sometimes) definitions of loneliness respectively. In both cases higher costs arise in the ‘lonely’ group.

When defining loneliness as something experienced sometimes or often, age group analysis reveals the difference in cost between lonely and non-lonely groups increases in magnitude and statistical significance as individuals get older (Table 4). The difference in cost is larger in magnitude for all age groups other than over 75 years when restricting the definition of loneliness to something experienced ‘often’, i.e., a more severe experience. Additionally, individuals aged 16–24 years have higher health related costs than those aged 25–49, suggesting a U-shaped health-related cost of loneliness by age. Furthermore, difference in mean cost is only statistically significant for the upper and lower bound at age 16–24.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Difference in health care service cost by ageᶲ^Ϯ.

https://doi.org/10.1371/journal.pone.0327671.t004