Trajectories of pain and cognitive function: 22 years of evidence in mid-to-later life

Abstract

Chronic pain has been identified as a risk factor for cognitive decline in later life. However, most studies measure pain at a single time point and none have investigated whether variations in pain severity are associated with changes in cognitive function over time. This project aimed to assess the relationship between individual-level change in pain severity and decline in cognitive function over time. We used data from the English Longitudinal Study of Ageing (ELSA), a cohort of nationally representative middle aged and older adults. Pain severity was measured at each wave using a 4-point scale (none, mild, moderate and severe) and cognitive function was assessed using 3 objective tests. We applied latent growth curve modelling, a method for longitudinal analysis, to 19,376 ELSA participants' data collected over 11 waves, spanning more than 20 years, to examine the relationship between initial level and change of both pain and cognitive function. Adjusting for age and sex, worsening chronic pain severity was associated with accelerated decline in a general measure of cognitive function (β = -0.053, p = 0.039). However, when additionally adjusting for ethnicity, socioeconomic status and comorbid chronic conditions, this association was attenuated to non-significance (β = -0.025, p = 0.365). Greater initial pain severity was associated with steeper decline in cognitive function even in the fully adjusted model (β = -0.104, p < 0.001). Our study suggests that baseline level of pain severity but not worsening pain severity is associated with steeper decline in cognitive function over time.

Keywords: Pain severity; cognitive decline; latent growth curve modelling; longitudinal study; older adults.

Figure 1:. Simplified path diagram of the parallel process latent growth curve model of pain severity and general cognitive function.

Circles are latent factors, squares represent observed variables, single-headed arrows are regression paths and double-headed arrows are correlation paths. The red arrows are the focus of our hypothesis. Dashed grey lines are regression paths for covariates. I = Intercept, S = Slope, G = general cognitive function, Animal = Animal naming test, Cancell = Letter cancellation test, Recall = Word recall test. For illustrative purposes, the loadings for animal naming and letter cancellation are not displayed; however, these are the same as the loading for recall and pain. Following the marker variable approach, the word recall test’s loadings on general cognitive function intercept and slopes was set to 1 and the means of intercept and slope of the word recall test were set to 0.

Figure 2:. Trajectories of pain severity.

A: Sankey diagram of pain severity with category change over time in the entire sample. B: Nine participants’ pain reports over 11 waves and their model-estimated trajectories. These nine participants were selected for illustrative purposes and are not systematically representative of the study sample. The green lines labelled “Model” are the individual-level model-estimated trajectories. C: The unstandardised estimates of pain severity intercept and slope from a random sample of 150 participants estimated by the measurement model are shown in grey, and the mean unstandardised estimate is shown in dark red (intercept β = −1.381, slope β = 0.024).

Figure 3:. Trajectories of general cognitive function.

The unstandardised estimates of general cognitive function intercept and slope from a random sample of 150 participants estimated by measurement model are shown in grey, and the mean unstandardised estimate is shown in dark red (intercept β = 10.364, slope β = −0.076).

Figure 4.. Simplified path diagram of parallel process LGCM of pain severity and cognitive function.

Model 1 (M1) has not been adjusted for any covariates, model 2 (M2) was adjusted for age and sex, and model 3 (M3) was adjusted for age, sex, ethnicity, socioeconomic status and comorbidities. I = Intercept, S = Slope, G = General cognitive function. The values inside the circles are the mean estimates. Single-headed arrows are regression paths and double-headed arrows are correlation paths. All estimates are standardised. Standard errors are in brackets and † indicates a statistically significant value at α = 0.05 after false discovery rate correction following the Benjamini–Hochberg procedure.