1. Study design and participants
The Chinese Longitudinal Healthy Longevity Survey (CLHLS) is an ongoing, prospective cohort study of community-dwelling Chinese older adults [19-20]. It covers the majority of the provinces in China and aims to investigate the factors associated with healthy longevity of Chinese. Started in 1998, the follow-ups have been conducted every 2 to 3 years. To reduce attrition, new participants are continually enrolled as death and lost-to-follow-up are inevitable. Trained interviewers with a structured questionnaire conduct the survey from door to door. A weight of age-sex-residence in the sample with the distribution of the total population was employed to reflect the unique sampling design .
In the 2008/2009 wave (baseline), 16948 older adults were interviewed in total, the number of which was the most among different waves. We excluded 385 participants younger than 65 years and 308 participants living in an institution, and finally included 16255 community-dwelling older adults for cross-sectional analyses. Among these older adults, 50.3% survived, 34.0% died, and 15.7% were lost in the 3-year follow-up (the 2011/2012 wave, see Table S1 for detailed information). Generally, the sociodemographic factors, socioeconomic status, and physical and cognitive functions of older adults who were lost in follow-up fell in between those who survived and died. The flow chart of our study was shown in Figure S1.
We used the data of living arrangements and feelings of loneliness at baseline, and assessed the association of loneliness with adverse health outcomes, including cognitive impairment, functional limitation and frailty at baseline and the 3-year follow-up, as well as 3-year mortality from 2008/2009 to 2011/2012, in the total sample and stratified by living arrangements.
Assessment of living arrangements and loneliness
Living arrangements were assessed using the question “Who do you live with?” with responses including ‘living with family (including house maid)’, ‘living alone (LA)’, or ‘living in an institution’. The 1st was defined as “not living alone/living with others (NLA, 84.3%)”. As only 308 (1.9%) older adults who were much older (93.1±9.1 years) lived in an institution, and institution living was different from ‘living with family’, ‘living alone’ and “community-dwelling”, we excluded them from analysis.
Loneliness was assessed via the question “Do you feel lonely or isolated?” with answers ‘always’, ‘often’, ‘sometimes’, ‘seldom’, ‘never’ and ‘not able to answer’, which has been demonstrated to be feasible for loneliness assessment by previous studies [22-23]. For the purpose of statistical analysis, we recoded the responses into a trichotomous variable: ‘always’, ‘often’, and ‘sometimes’ were defined as “lonely (FL, 28.6%)”, ‘seldom’ and ‘never’ as “not lonely (NFL, 56.8%)”, and ‘not able to answer’ as “NA (14.6%)”.
Adverse health outcomes
(1) Cognitive impairment
The CLHLS used the Chinese version of the Mini-mental State Examination (MMSE), validity and reliability of which have been verified [19-20], as a measure of cognitive function at each wave. The total scores range from 0 to 30, with higher scores representing better cognitive function. We used education-adjusted criteria to define “cognitive impairment”: for participants without formal education, MMSE score ≤17 was defined as CI; for those with 1-6-year education, MMSE score ≤20 was defined as CI; for those with more than 6-year education, MMSE score ≤24 was defined as CI [24-25].
(2) Functional limitation
The Katz Basic Activities of Daily Living (ADL) Scale and Lawton Instrumental Activities of Daily Living (IADL) Scale were used to assess participants’ physical function. Having difficulty in performing any one or more of the ADL tasks (6 items: bathing, dressing, toileting, transfers, continence and eating) was defined as having ADL limitation; having difficulty in performing any one or more of the IADL tasks (8 items: be able to go outside to visit neighbors, shop by oneself, make food by oneself, wash clothes by oneself, walk one kilometer, carry 5 kg weight, crouch and stand 3 times, take public transportion) was defined as having IADL limitation. Participants with either ADL or IADL limitation were defined as functional limitation.
(3) Frailty assessment
Our Frailty Index (FI) was the same as the previous CLHLS studies [26-27]. FI included 39 self-reported items, including functional limitations, cognitive function, self-reported health status, interviewer-rated health status, mental health, auditory and visual ability, heart rhythm, and chronic diseases. We scored each term as 0 (absence of deficit) or 1 (presence of deficit) for 38 of 39 terms, and scored 1 term as 2 if the participants reported 2 or more serious illnesses that caused hospitalization or being bedridden in the past 2 years. FI score was equal to the number of reported deficits divided by the total number of included deficits. It was a continuous variable ranging from 0 to 1, with a higher value indicating severer frailty. The continuous FI score was classified into non-frailty (FI≤0.21) and frailty (FI>0.21) following a previous report [26-27].
Mortality was measured by survival status and duration of exposure to death. The survival status was measured by whether a respondent interviewed in the 2008/2009 wave died or survived at the 2011/2012 wave. The exposure duration for a survivor was measured by number of months between the interview date in the 2008/2009 and 2011/2012 waves. For those who died before the 2011/2012 wave, the exposure time was measured by the time interval between date of death and the interview date in the 2008/2009 wave. The date of death was collected from officially issued death certificates whenever available, otherwise the next-of-kin and local residential committees were consulted. The average follow-up period of all participants was 3.2 (±1.5) years, with 1.5 (±0.9) years for deceased participants, and 4.4 (±0.2) years for survived participants. The data quality of mortality in the CLHLS has been proved to be high .
Measures of sociodemographic characteristics at baseline included age, gender, race, marital status, residence, occupation, education, BMI, smoking, alcohol drinking, and socioeconomic status including sufficient financial support, economic independence, adequate medical service and public medical payment, dietary habits including fruit and vegetable eating, and tea drinking. Living preference was assessed via the question “What kind of living arrangement do you like best?”, with answers ‘living alone (or only with spouse) regardless of proximity to children’, ‘living alone (or only with spouse) with close proximity to children’, ‘live with children’. The former two were comined as “prefer LA”, and the 3rd as “prefer NLA”. Social/leisure activity score was calculated in the way same as a previous study , with a high score representing a high frequency of social and leisure activities. Physical exercise was assessed via the question “Do you take exercise or not at present?” with answers ‘yes’ or ‘no’. Self-reported health was assessed via the question “How do you rate your health status?” with answers ‘bad’ and ‘very bad’ defined as “poor self-reported health”. Interviewer-rated health was assessed by interviewers, with ‘moderately ill’ and ‘very ill’ defined as “poor interviewer-rated health”. Comorbidities was assessed via whether suffering from 24 common chronic diseases including hypertension, diabetes, heart disease and stroke. Serious illness in the past 2 years was defined as “illness that causes hospitalization or being bedridden all the year around”. Hearing and visual ability were also assessed.
3. Statistical analysis
Categorical variables were presented as numbers (percentages), and continuous variables were presented as means (SD). Differences in the distribution of categorical variables among groups were tested by c2 test. For continuous variables, the F test or Kruskall-Wallis test was used for comparison between different groups. Logistic regression models were performed to estimate the odds ratios (ORs) and 95% confidence intervals (CIs) of associations between loneliness and cognitive impairment, functional limitation and frailty (in the total sample or stratified by living arrangements), as well as the interaction terms between living arrangements and loneliness. Cox proportional hazards regression model was performed to calculate the hazard ratios (HRs) and 95% CIs between loneliness and 3-year mortality (in the total sample or stratified by living arrangements). The impacts of loneliness on adverse health outcomes were also measured within strata of age groups (<80 or ≥80 years) and genders. OR estimates for prevalent cognitive impairment, functional limitation and frailty, and HR for 3-year mortality were adjusted for baseline values of age, gender, race, marital status, residence, occupation, education, BMI, smoking, alcohol drinking, living preferences, socioeconomic status, dietary habits, social/leisure activity score, physical exercise, poor self-rated health, poor interviewer-rated health, comorbidities (≥2), hypertension, diabetes, heart disease, stroke, serious illness in the past 2 years, hearing problem, and visual impairment. As many variables changed from 2008/2009 to 2011/2012, OR estimates for incident cognitive impairment, functional limitation and frailty were adjusted for age, gender, race, education, occupation, hypertension, diabetes, heart disease, stroke, and changes of other variables from 2008/2009 to 2011/2012. The acceptable level of significance was set as two-sided P<0.05. Stata version 14.0 (StataCorp LP, College Station, TX, USA) was used for data analysis.