Study design and population
- Sudlow C
- Gallacher J
- Allen N
- et al.
and replicated this analysis using data from an independent cohort, the nationwide population-based Finnish Health and Social Support (HeSSup) study.
- Korkeila K
- Suominen S
- Ahvenainen J
- et al.
Baseline data for the UK Biobank were collected between 2006 and 2010 in 22 research assessment centres across the UK. We included participants aged 38–73 years, who were linked to national health registries, had no history of hospital-treated infections at or before baseline, and had complete data on loneliness or social isolation. The HeSSup study comprised a random sample of individuals in Finland aged 20–54 years.
- Korkeila K
- Suominen S
- Ahvenainen J
- et al.
We included individuals from the HeSSup study with available data on loneliness or social isolation who were linked to national health registries. The HeSSup study included repeated assessments of loneliness and social isolation (in 1998 and 2003), which allowed evaluation of reverse causality. We excluded participants with missing data on hospital-treated infections, loneliness, and social isolation from both cohorts. The outcome of interest was defined as hospital admissions with a primary diagnosis of infection, ascertained via linkage to electronic health records.
All participants provided written informed consent for the baseline assessments and for registry linkage. The UK Biobank was approved by the National Health Service National Research Ethics Service (11/NW/0382), and the HeSSup study by the ethics committee of Turku University Central Hospital and the Finnish Population Register Centre (VRK 2605/410/14).
Procedures
- Hughes ME
- Waite LJ
- Hawkley LC
- Cacioppo JT
In the UK Biobank, social isolation was assessed by asking three questions:
- Sudlow C
- Gallacher J
- Allen N
- et al.
(1) “Including yourself, how many people live in your household? Include those who usually live in the house such as students living away from home during term time, and partners in the armed forces or in professions such as pilots” (1 point for living alone); (2) “How often do you visit friends or family or have them visit you?” (1 point for less than one friend or family visit per month); and (3) “Which of the following (leisure or social activities) do you engage in once a week or more often? You may select more than one” (1 point for not participating in any social activities at least weekly). The sum of the responses to these three questions resulted in a scale ranging from 0 to 3. We classified respondents with 2 or 3 points as socially isolated. The loneliness and social isolation measures were dichotomised with no weighting of responses.
- Sarason IG
- Sarason BR
- Potter EH
- Antoni MH
The respondents could choose one or more of six alternatives (husband, wife, or partner; some other relative; close friend; close co-worker; close neighbour; or someone else close). The responses to the items were combined so that each source of support contributed one point to the final social support score (range 0–20).
- Sarason IG
- Sarason BR
- Potter EH
- Antoni MH
We used dichotomised scores in our analyses (0–6, socially isolated; 7–20, not isolated). Additional references and a description of validity issues are provided in the appendix (p 2).
- Eisenberger NI
- Moieni M
- Inagaki TK
- Muscatell KA
- Irwin MR
(METs) and was dichotomised on the basis of median split (high, 3·6 or more; low,
- Mendes MA
- da Silva I
- Ramires V
- et al.
- Sipila PN
- Heikkila N
- Lindbohm JV
- et al.
We classified hospital-treated infectious diseases according to 925 ICD-10 codes (appendix pp 5–9). For comparison, we examined the associations of loneliness and social isolation with other broad disease categories including cancers; diseases of the endocrine, circulatory, respiratory, digestive, musculoskeletal, genitourinary, and nervous systems; diseases of the blood, eye, ear, and skin; and mental and behavioural disorders.
Statistical analysis
- Elovainio M
- Hakulinen C
- Pulkki-Raback L
- et al.
,
- Elovainio M
- Lahti J
- Pirinen M
- et al.
The interaction effect was tested by adding interaction terms into each model.
Second, we performed stepwise multivariable analyses in the UK BioBank cohort to test the extent to which the associations were independent of baseline covariates, and whether the multivariable-adjusted results were replicable in the first 3 years of follow-up and from year 3 onwards. All models included loneliness or social isolation as the exposure and covariates were added as follows. Model 1 included age and sex. In addition to age and sex, other models included ethnicity (Model 2); education and the Townsend deprivation index (Model 3); smoking, alcohol consumption, physical activity, and BMI (Model 4); long-term illness (Model 5); C-reactive protein (Model 6); depressed mood (Model 7); and all the aforementioned covariates (Model 8). Given that the covariates can act as both as confounders and mediators, we interpreted the results cautiously and considered the association between loneliness or social isolation and infectious diseases independent of other factors only if the association remained significant after adjustment for the covariates. We calculated the percentage of excess risk attributable to covariates (PERM) for the associations of social isolation and loneliness with infections using the following formula:
PERM% = ([HR(age and sex) – HR(age, sex, and covariates adjusted)]/[HR(age and sex adjusted) – 1]) × 100.
Third, in sensitivity analyses of the UK BioBank cohort, we tested whether the associations were robust to the exclusion of participants with physical conditions that increase the risk of infectious diseases. To examine reverse causation in the HeSSup study, we tested whether infectious diseases at baseline were associated with loneliness or social isolation at follow-up among those who did not report these exposures at baseline. The exposure was a hospital-treated infectious disease and the outcome loneliness or social isolation. We included those with and without an infectious disease at baseline (the exposure) but excluded those who reported being lonely or isolated. Incident cases were those who had become lonely or socially isolated at follow-up. To investigate disease specificity, we examined associations between loneliness or social isolation and other disease categories. In each step, participants with missing data on covariates were excluded from the analysis.
A two-sided p value of less than 0·05 was considered to indicate statistical significance. Because this was a hypothesis-testing study with multiple sensitivity analyses rather than an exploratory study with multiple independent tests, we did not correct for multiple testing.
In the HeSSup study, the analyses were done in two steps, first adjusted for age and sex and second adjusted for age, sex, education, alcohol consumption, smoking status, physical activity, and depressive symptoms.