Study population
Data were taken from waves 4 (2011/2012) (26, 27), 5 (2013) (28, 29), 6 (2015) (30, 31) and 7 (2019) (32, 33) of the SHARE study, a multidisciplinary, cross-national, and longitudinal research project focusing on community dwelling adults aged 50 or older (13). Detailed information about the entire SHARE project is available atwww.share-project.org. SHARE respondents were included in our sub sample if: (a) they did not report a diagnosis of dementia, Alzheimer’s, senility or Parkinson’s; (b) they had participated in wave 5; (c) they reported a diagnosis of osteoarthritis between wave 5 and 6 (2013-2015); (d) they did not report a diagnosis of osteoarthritis at wave 5; (e) they reported a diagnosis at wave 7 (Figure 1). Osteoarthritis diagnosis was assessed at all waves with the following question: “Has a doctor ever told you that you had/Do you currently have: Osteoarthritis? (With this we mean that a doctor has told you that you have this condition, and that you are either currently being treated for or bothered by this condition.)”(29).
Variables
Cognitive function
Cognitive function was assessed at all waves and was based on multiple items: (1) immediate recall (participants were presented a list of 10 words and asked to repeat the words immediately; range = 0‐10), (2) delayed recall (participants were asked for the list of 10 words after a delay; range = 0‐10), (3) subtraction (participants were asked to mentally solve a subtraction task; range = 0‐5), and (4) verbal fluency (participants were asked to produce as many animal names as possible within a given period of time; range = 0-100). A joint scale was created based on all items with a total score range of 0 to 125 (Cronbach’s alpha = 0.80, for wave 5 and Cronbach’s alpha = 0.76, for wave 6). The higher the score, the higher the participant’s cognitive function (34). We decided not use a standardised score of cognitive function due to the risks associated with their use in the analysis of longitudinal data (35, 36).
Pain
Pain was constructed from two questions asked at all waves of the survey. Participants were asked whether they had been troubled by pain (yes/no). Those who replied positively were then asked to rate how bad their pain was most of the time (either mild, moderate or severe). The two variables were added to create a single score ranging from 1 (not troubled by pain) to 4 (troubled by severe pain), representing whether respondents were troubled by pain and how severe it was. This verbal rating scale have been used widely in the pain literature (37).
Anxiety
In SHARE wave 5, five items were used to measure severity of anxiety that were taken from the Beck Anxiety Inventory (38). The respondents were asked about anxiety symptoms (“I had fear of the worst happening”, “I was nervous”, and “I had a fear of dying”, “I felt my hands trembling”, “I felt faint”) they experienced in the last 7 days and answer on a four point Likert scale (“never”, “hardly ever”, “some of the time”, “most of the time”). We created a single anxiety scale by adding the scores of all five items to obtain an overall score, with higher scores indicating higher anxiety (Cronbach’s alpha = 0.69).
Social deprivation index
A social deprivation index was provided in wave 5 of SHARE and was constructed using a battery of 15 questions related to participation in everyday life, social activities and the quality of the neighbourhood following Chakravarty and D'Ambrosio (39) and Levitas, Pantazis (11). In order to combine different social deprivation items into a single index, the authors computed the weight of each item based on a regression of the chosen items on the reported values of life satisfaction (40). The most important elements of the index, i.e. those with the highest weight are: feeling left out of things, not feeling part of the neighbourhood, having no helpful people in the local area and waiting too long to see a doctor (40).
Instrumental activities of daily living (IADL)
A modified version of IADL (41) was used in SHARE (42). IADL included seven activities in wave 5: “using a map to get around in a strange place”, “preparing a hot meal”, “shopping for groceries”, “making telephone calls”, “taking medications”, “doing work around the house or garden” and “managing money” with a total score ranging from 0 to 7. Two more items were added in waves 6 and 7: “leaving the house independently and accessing transportation services”, and “doing personal laundry”, resulting in nine items in total (score: 0-9). A higher index score indicates more difficulty with these activities and lower mobility of the respondent (Cronbach’s alpha = 0.71, 0.80 and 0.86, for waves 5, 6 and 7, respectively).
Additional variables
Other variables included age (>40 y.o.), gender, education measured with the International Standard Classification of Education (ISCED‐97) (43), body mass index (BMI), smoking status (Currently smoking, Ex-smoker, Never smoked, and No response), alcohol consumption (How many drinks in 3 months), physical inactivity (Never moderate or vigorous activity and Other), number of chronic diseases (0-9), and marital status (Married and living together, Divorced, Widowed, and Other).
Statistical analysis
Modelling was performed using R version 4.0.1. For path analysis the ‘lavaan’ package was used (44). The missing mechanism of the SHARE data is assumed to be missing at random and the level of dropout in this subsample is small (17.1%), thus, missing data were handled using full information maximum likelihood estimation (FIML). We conducted sensitivity analyses using multiple imputations (m = 40). No significant differences between the two methods for handling missing data were found (see Additional File 1).
Crossed-lagged path analysis was run using the pain and cognitive function measurements at waves 5 and 6 and IADL at waves 5, 6 and 7. The model was constructed such that pain levels, cognitive function, and IADL each predicted each at the following wave. Path model was adjusted for age, sex, education level, number of chronic diseases, BMI and alcohol consumption at baseline (See Supplementary Table 7, Additional File 1 for correlational analyses).
Model fit was assessed using the Comparative Fit Index (CFI), the Tucker Lewis Index (TLI), the Root Mean Square Error of Approximation (RMSEA) and the Standardised Root Mean Square Residual (SRMR). To determine acceptable fit we used the cut-off criteria proposed by (45), who recommended that an RMSEA lower than .06 and CFI and TLI greater than .95.