Exploring the relationship between pain, cognition, and chronic conditions: insights from the HAALSI study in rural South Africa

Abstract

Background

Research on cognition and pain is limited in Low and Middle-income Countries (LMICs) and understanding how chronic conditions and pain treatment may moderate this association is underexplored. This study aimed to explore the relationship between pain and cognition and the moderating effect of hypertension, diabetes, HIV, pain treatment, and depressive symptoms.

Methods

We analyzed data from 3803 individuals enrolled in the HAALSI study, a longitudinal population study of older adults in Agincourt, South Africa. Pain was measured with the Brief Pain Inventory. Cognition was assessed using a composite of orientation questions, a memory test, and the Trails Making Test B. Chronic conditions were assessed using biological measures, and depressive symptoms were measured using the CES-D scale. Linear regression models were used to investigate the relationship.

Results

Baseline and longitudinal pain were significantly associated with poorer episodic memory (ß = −0.17 [P < .001]; ß = −0.18 [P < .001]). Hypertension amplified the negative effect of pain on episodic memory, while diabetes and HIV did not moderate the relationship between pain and cognition (ß = −0.10 [.006]). Pain treatment was associated with poorer cognitive performance. Depressive symptoms moderated the relationship between pain and both cognition and executive function (EF) (P = .02). The negative effect of pain on episodic memory was observed in individuals with both acute and persisting pain, while it only affected EF in those with acute pain.

Conclusions

These findings highlight the importance of examining factors that may moderate the relationship between pain and cognition and strategies to mitigate the effect pain has on cognition, particularly in LMICs.

Introduction

In 2011, it was estimated that 20% of the global population suffers from pain, with 10% being diagnosed with persistent pain annually.^1,2 Despite pain being a significant international public health issue, there is limited information on its prevalence in low- and middle-income countries (LMICs).^2,3 Persistent pain is defined as any pain that is experienced beyond three months.⁴ In South Africa, research on persistent pain prevalence is sparse, with most studies focused on clinical samples^5,6 and only a handful examining the prevalence and impact on quality of life within communities.^7,8 Approximately one in five South African adults and one in three older adults suffer from chronic pain.⁸ As South Africa’s population is rapidly aging, the prevalence of chronic pain is expected to rise.⁹

Pain, when not effectively treated, can disrupt all aspects of quality of life. Individuals with persistent pain often experience limitations in work and activities of daily living, as well as negative changes to their mental health and cognition.^10,11 Studies have shown that individuals who experience persistent pain perform poorly on multiple cognitive measures, including attention, learning, executive function (EF), and memory.¹² Persistent pain signals can interfere with other sensory inputs, impairing cognitive performance.¹³ Moreover, a longitudinal study found persistent pain to be linked to a more rapid decline in memory and increased risk of dementia. Understanding the relationship between cognition and pain is crucial for identifying how pain contributes to both short-term and long-term cognitive decline.¹⁴ This understanding can inform the development of comprehensive care strategies that address both pain management and cognitive health, ultimately improving the quality of life for individuals suffering from chronic pain.

Research on the relationship between pain and cognition has been largely limited to clinical samples and community-dwelling populations, with only a few studies examining the relationship in LMICs. One study examined the relationship between pain and Mild Cognitive Impairment (MCI) in a sample of 32 712 older adults from Ghana, China, Mexico, India, and South Africa and found that compared to no pain, severe/extreme pain was significantly associated with a threefold increase in the odds for MCI across all countries.¹⁵ Similarly, only a few longitudinal studies investigated the long-term effect of pain on cognitive domains such as memory, learning and processing speed, and EF. These studies have only been restricted to high-income countries.^14,16 Findings on the relationship between chronic pain and specific cognitive domains, mainly EF, have been inconsistent. Some studies have found a significant association between EF and pain, while others have found no association.¹⁷ Inconsistent study samples (e.g., clinical versus national data, concentrated in wealthier countries), variations in cognitive measures and outcomes, and limited longitudinal data in LMICs contribute to inconsistent findings on the relationship between pain and cognition.¹² Further research is needed to clarify the relationship between pain and cognitive health across diverse populations to better understand the mechanisms underlying this association over time, and to inform the development of targeted interventions in these contexts.^2,12

One of the largest pain studies, which followed 10 605 older adults in the United States over 10 years, found that persistent pain was associated with a more rapid decline in memory and an increased risk of dementia.¹⁴ Longitudinal studies are valuable for tracking the progression of pain and cognition over time, providing insights into how these conditions develop and influence each other at different stages.¹⁴ This approach can help identify critical time windows when interventions might be most effective and can distinguish between the short- and long-term impacts of persistent pain on cognitive function.

Along with inconsistencies in relationships between pain and certain cognitive domains, and limited information on the relationship between pain and cognition in LMICs, studies generally have not explored how chronic conditions and treatment for persistent pain may influence this relationship.¹⁸ Some studies have controlled for chronic conditions such as hypertension and diabetes or have examined pain in patients with these conditions. However, they have not directly assessed the potential moderating role of these comorbidities. Understanding how chronic conditions moderate the relationship between pain and cognition is essential, as the majority of older adults suffer from multiple comorbidities alongside cognitive decline and increased susceptibility to pain. LMICs, in particular, carry the “double burden” of treating and managing both communicable and non-communicable diseases while also facing the most rapidly aging populations.⁸ Investigating the potential moderating role that chronic conditions may play in the pain/cognition relationship could improve strategies for mitigating the effects of pain on cognition.

This study aims to explore the relationship between pain and cognition in a diverse context and to examine the moderating role of chronic conditions by investigating the association between pain and cognition among middle-aged and older adults in rural South Africa—a population where this relationship remains understudied. The first aim was to examine the cross-sectional relationship between episodic memory, EF, and pain. The second aim was to determine whether diabetes, Human Deficiency Immuno-Virus (HIV), depressive symptoms, as well as pain treatment moderate the relationship between cognition and pain. This is particularly relevant given the high prevalence of these conditions in Agincourt, South Africa.⁶ Last, the third aim was to assess the longitudinal effect of acute vs persistent pain on cognition. We hypothesized a negative relationship between pain and cognition that persists over time, with conditions like hypertension, diabetes, HIV, and depressive symptoms exacerbating this effect, while pain treatment potentially mitigating the negative impact of pain on cognitive function.

Methods

Data were analyzed from the Health and Aging in Africa: A Longitudinal Study of an INDEPTH Community in South Africa (HAALSI), a longitudinal population-based study examining the social, economic, and biological risks for chronic health conditions, with a focus on dementia, cognitive functioning, and cardiovascular disease.¹⁹ HAALSI was conducted in the Agincourt sub-district in Mpumalanga Province, South Africa, and is an international partner study to the U.S. Health and Retirement Study (HRS).²⁰ The HAALSI cohort is nested within the Agincourt Health and Demographic Surveillance System (HDSS), where long-term health and sociodemographic surveillance is conducted.^19,21 The study area spanned 32 villages, consisting of 1 16 000 people, with high levels of illiteracy and unemployment post-apartheid.^19,22,23 Details of the study sampling and methodology have been published in the original paper.¹⁹ Briefly, participants had to be at least 40 years old as of July 2014 and have lived in the study area for at least 12 months to be included. A sampling frame of 12 875 individuals was identified and 6 281 community-dwelling adults were randomly selected to participate in HAALSI; gender-­specific sampling fractions were developed to ensure a gender­-balanced cohort. From the selected 6 281 men and women, 5 059 completed home interviews at the baseline wave (Wave 1), comprising 2 345 men and 2 714 women and a response rate of 85.9%. A total of 391 (6%) of the sample were ineligible due to death or out-migration before the interview, 353 (6%) were not found, 48 (1%) were unable to participate, and 430 (7%) refused. Data were collected through in-person interviews conducted by trained field workers in the native language (Shangaan) and captured on tablets using a Computer Assisted Personal Interview (CAPI) program.

Participants responded to questions on sociodemographic, health, and household conditions and underwent biomarker collection and physical assessments. HAALSI has completed three waves of data collection between 2014 and 2022. Ethical approval for HAALSI was obtained from the University of the Witwatersrand Human Research Ethics Committee (#M141159), the Harvard T.H. Chan School of Public Health Office of Human Research Administration (#13–1608), and the Mpumalanga Provincial Research and Ethics Committee. The data are publicly available at https://www.haalsi.org.

This study used data from Wave 2 (2018–2019) and Wave 3 (2021–2022) of HAALSI, since these waves had a larger sample of those who answered the pain questions compared to Wave 1. For the cross-sectional analysis, participants who had missing data for pain questions or did not have an episodic memory score or EF score were excluded, resulting in a final baseline sample of 3 803 participants analyzed for episodic memory and 3 369 for EF. Of note, the difference in random missingness between episodic memory and EF is expected due to the fact that EF tasks are more difficult to execute, especially in this population due to low literacy. For the longitudinal sample, participants with missing data on pain, episodic memory or EF for Wave 3 were excluded, resulting in a final longitudinal sample of 2 975 participants (see eFigure S1).

Measurements

Pain

Pain was assessed using a modified version of the Brief Pain Inventory.¹² Specifically, the analysis included binary responses (Yes/No) to the following two questions: “Have you had pain—other than these everyday kinds of pain—today?” and “Are you currently receiving any treatment or currently taking any medications for your pain?” The analysis additionally included an ordinal question “On a scale of 0 to 10, where 0 is ‘No Pain,’ and 10 is ‘Pain as bad as you can imagine,’ please rate how much pain you have right now for the area that hurts the most.” To assess pain by body part most affected, we also included a nominal question “Please tell me which areas hurt the most.” Pain treatment was assessed using yes/no responses to whether the participant used pain treatment. In the longitudinal analysis, we named groups that presented pain in both waves as the “persistent pain” group, to differentiate those that only presented with pain once or none of the times. We do not have information on how long the pain has been felt at each timepoint.

Cognition

A brief cognitive battery was employed to assess participants’ episodic memory.^23,24 It included a time orientation task (ability to state the correct date, month, year, and South African President; maximum score of 4 points) and a first-trial immediate and delayed recall task involving a 10-word list presented verbally (1 point per correctly recalled word; maximum score of 20 points) .This battery of selected tests, adapted from the US HRS brief screening tool for dementia,²⁵ yielded a total episodic memory score ranging from 0 to 24 points, based on the combined time orientation and recall tasks. This approach followed the methodology used by Kobayashi and colleagues.²³ The combination of orientation and memory tasks was used to extend the range of variability in the cognitive outcome while preserving its conceptual alignment with the memory domain. Furthermore, it allows harmonization with other sister studies using the same measure and ensures consistency with our previous work in HAALSI, demonstrating its effectiveness in reflecting cognitive performance.²⁰ Given the large-scale nature of the HAALSI study, the use of a set of brief measures of cognition was essential to ensure feasibility and minimize participant burden. This efficient design enables reliable data collection across large populations while maintaining a robust cognitive assessment. All scores were subsequently z-standardized to allow results from this composite score to be shown on the same scale across the cohort sample and within subjects.²⁴ In this transformation, the composite score was centered so that the sample mean corresponds to 0, with each unit increase or decrease representing one standard deviation. For interpretability, a scaled score of 0 corresponds to an average raw score of 12.8 (out of a maximum of 24). A ± 1 standard deviation shift equates to a raw score difference of 4.4 points in the cross-sectional sample and 13.4 (±3.6) in the longitudinal sample.

Additionally, EF was assessed using only the Trails Making test part B. This task required participants to connect a series of circles and squares presented on a tablet screen. In the first part, participants connected circles in ascending order of size [0–7]. The second part involved connecting squares in descending size [0–7]. Finally, participants had to alternate between connecting circles and squares while following the respective size rules [0–14]. Only the scores from the switching task were included for a more specific metric for EF. Raw scores were used, which could range from 0–14 points. This culturally adapted trails task, and the scoring approach used in this work was successfully used in this population before^24,26 to accommodate the high rate of low education/illiteracy in the sample, difficulties with motor coordination, and unfamiliarity with testing procedures.^27,28

We assessed normality using skewness, kurtosis, and visual inspection of residual plots. Episodic memory (skew = −0.40, kurtosis = 3.10) and EF (skew = 0.24, kurtosis = 1.67) showed acceptable distributions. The Trails Test showed a flatter peak and lighter tails compared to the episodic memory variable. Residuals met the necessary assumptions to support the use of linear regression.

Chronic conditions and depressive symptoms

Hypertension was defined as either (1) untreated hypertension, indicated by blood pressure measures of diastolic ≧90 or systolic ≧ 140 or (2) treated hypertension, indicated by self-­reported hypertension medication use.¹⁹ Diabetes was defined as (1) glucose levels ≧ of 11.1 mmol/L with no fasting or glucose ≧7 mmol/L fasting or (2) glucose <7 mmol/L on diabetes medication.¹⁹ Dried blood spots were collected and assessed for HIV, viral load, and the presence of Antiretroviral Therapy (ART) drugs.²⁹ In instances where data on chronic conditions were missing in waves 2 and 3, we employed a last observation carried forward (LOCF) approach, where a positive diagnosis from any previous wave was carried forward to subsequent waves to account for the missing data. This method assumes that once a chronic condition is diagnosed, it remains present in the individual, which aligns with the chronic nature of the conditions under study. Depressive symptoms were measured using the Center for Epidemiological Studies (CES-D) scale. Scores ranged from 0 to 60, with a higher score indicating higher symptomology.³⁰ In the longitudinal moderation analysis, a cutoff point of 20 was used to categorize individuals with low depressive symptomology or high depressive symptomology instead of the commonly recommended value of 16.³¹ A meta-analysis of 22 studies found that a cutoff of 16 had a sensitivity of 0.87 and specificity of 0.70, while a cutoff of 20 (examined in 12 studies) had a slightly lower sensitivity (0.83) but higher specificity (0.78), with a diagnostic odds ratio (DOR) of 16.6. Overall, the study concluded that a cutoff of 20 offers a better balance between sensitivity and specificity, reducing false positives while maintaining a reasonable ability to detect depression.³²

Covariates

Self-reported data on years of education, age, wealth, and sex were collected during participant interviews. Years of education were categorized into groups, including no formal education, some primary (1–7 years), some secondary (8–11 years), and secondary or more (12+ years). The wealth index was calculated using principal components of household characteristics, ownership of household items, vehicles, and livestock, and divided into quintiles. A higher wealth index corresponds with great financial means.³³

Statistical analysis

Analyses were conducted using R version 4.4.0. Descriptive statistics were used to summarize the characteristics of the full study sample and to provide further detail on the subsample of participants who reported experiencing pain and those who have received pain treatment.

For the cross-sectional analysis, data from Wave 2 were treated as the baseline. Linear regression models were employed to examine the association between cognitive outcomes and pain at baseline. The cognitive outcomes included episodic memory, measured via a cognitive composite score, and EF, measured via the Trails Making Test. The primary predictor was the presence of pain, defined as participants reporting pain beyond everyday aches. Models were then adjusted for age, education, hypertension, diabetes, depressive symptoms, sex and wealth index. The same regression model was used to examine the association between pain severity (rated on a 0–10 scale) and pain treatment (yes or no) as predictors of episodic memory and EF.

Secondary analyses were conducted to evaluate the moderating effect of hypertension, diabetes, HIV, or depressive symptoms on the association between pain presence and severity with cognitive outcomes. Specifically, the moderation effect was tested by including an interaction term between each chronic condition and depressive symptoms and the pain variable in the adjusted models. A significant interaction term indicated that the chronic condition or depressive symptoms served as a moderator. Further analysis was then performed within each chronic condition subgroup to identify unique patterns in the association based on chronic condition status.

Data from both Waves 2 and 3 were used for the longitudinal analysis. A linear mixed model examined the association between pain and cognitive outcomes. These models included a timepoint indicator as a covariate to account for repeated measures over time (1 = Wave 2; 2 = Wave 3). Repeated measures within the same participant were accounted for by including the participant’s identification number as a random term in the model to address the correlation of measures within the same individual over time.³⁴ Both unadjusted and adjusted models were performed, with episodic memory and EF as the outcome and pain (presence, severity, treatment) as the exposure. The models were adjusted for age, education, hypertension, diabetes, depressive symptoms, sex, wealth index, timepoint indicator, and the within-individual repeated adjustment in the random term. To better understand the longitudinal trajectory of pain and its relationship with cognition, the sample was categorized into three subgroups based on their longitudinal pain status: no pain (no pain reported in either wave), acute pain (pain reported in only one wave), and persistent pain (pain reported in both waves).

Results

Baseline descriptives and the sample distribution are presented in Table 1. Among the 3 803 respondents, 972 reported (26%) experiencing pain beyond typical everyday pains. Among those reporting pain, 50% indicated their legs, knees, and upper/lower back as most affected. The sample was comprised of 55% females, with an average age of 64 years (SD = 12). Educational attainment was generally low, with 43% of participants reporting no formal education. Hypertension was prevalent, affecting 65% of respondents, while 15% had diabetes and 29% were HIV positive. Respondents had an average CES-D score of 15 (SD = 9) and 30% had scores above 20, which indicated high depressive symptoms on the CES-D scale. The mean episodic memory score for the high depressive symptomology group (n = 1 153) was −0.04, and the mean EF score was 5.38. The mean episodic memory score for the low depressive symptomology group (n = 2 640) was 0.018 and the mean EF score was 6.69. This corresponds to an approximate 0.26-point difference on the raw episodic memory score (range 0–24), and a 1.31-point difference in the EF score, indicating slightly lower memory and EF performance among individuals with high depressive symptoms when compared to those with low depressive symptoms.

Table 1.

Baseline demographic and health characteristics of the HAALSI sample by pain status (n = 3 803).

Characteristic	Overall, N = 3 803	Participants without reported pain  N = 2 831	Participants with reported pain  N = 972
Education
No formal education	1 616 (43%)	1 136 (40%)	480 (49%)
Some primary (1–7 years)	1 347 (35%)	988 (35%)	359 (37%)
Some secondary (8–11 years)	470 (12%)	387 (14%)	83 (8.6%)
Secondary or more (12+ years)	362 (9.5%)	314 (11%)	48 (4.9%)
Missing	8	6	2
Sex
Female	2 107 (55%)	1 489 (53%)	618 (64%)
Male	1 696 (45%)	1 342 (47%)	354 (36%)
Age (years)	64	63	70
Wealth indexa
1 (poorest)	747 (20%)	558 (20%)	189 (19%)
2	731 (19%)	530 (19%)	201 (21%)
3	747 (20%)	560 (20%)	187 (19%)
4	763 (20%)	577 (20%)	186 (19%)
5 (wealthiest)	815 (21%)	606 (21%)	209 (22%)
Diabetes
Yes	556 (15%)	387 (14%)	169 (18%)
No	3 128 (85%)	2 335 (86%)	793 (82%)
Missing	119	109	10
Hypertension
Yes	2 447 (65%)	1 734 (62%)	713 (74%)
No	1 334 (35%)	1 077 (38%)	257 (26%)
Missing	22	20	2
HIV
Negative	2 177 (71%)	1 531 (68%)	646 (80%)
Positive	897 (29%)	734 (32%)	163 (20%)
Missing	729	566	163
Depressive symptoms (CES-Db)	15 (9)	14 (9)	16 (9)
Missing	10	9	1
No Depressive symptoms group	2 640 (70%)	1 976 (70%)	664 (68%)
Depressive symptoms group	1 153 (30%)	846 (30%)	307 (32%)
Missing	10	9	1
Episodic memory	0.00 (1.00)	0.12 (0.96)	−0.34 (1.03)
Episodic memory (raw scores)c	12.8 (4.4)	13.4 (4.2)	11.4 (4.5)
EF	6 (5)	6 (5)	6 (5)

N (%) or Mean (SD)

^aThe wealth index was calculated using principal components of household characteristics, ownership of household items, vehicles, and livestock and divided into quintiles. A higher wealth index corresponds with great financial means.

^bCenter for Epidemiologic Studies Depression Scale.

^cSum of correct items on words list (immediate + delayed recall) and four orientation questions. Ranges from 0 to 24.

Cross-sectional analysis

Self-reported pain was significantly associated with episodic memory in both unadjusted and adjusted models (ß = −0.46 [P < .001], ß = −0.17 [P < .001], respectively).

For EF, the association with pain was significant only in the unadjusted model, indicating that the initial reduction in EF among those reporting pain was attenuated when accounting for covariates. Among participants reporting pain (n = 972), higher pain severity was significantly associated with episodic memory (ß = −0.05 [P = .004]) and EF (ß = −0.28 [P = .002) in unadjusted model, and was also significantly associated with episodic memory (ß = −0.04 [P = .005]) and EF (ß = −0.30 [P < .01) in adjusted model. For every unit increase in pain severity on a scale of 1 to 10, there was a 0.04 lower score in episodic memory and a 0.30 reduction in EF, when adjusting for covariates. Since taking pain medication could alter the relationship between pain severity and cognitive outcomes, we tested whether pain treatment moderated this relationship. The results indicated that pain treatment did not significantly moderate the association with either episodic memory (β  =  0.02, P = .43) or EF (β = −0.16, P = .42) (Table 2). Hypertension was the only chronic condition that was a significant predictor of pain, (ß = −0.10[.006]). On a 1–10 severity scale, individuals with hypertension reported pain levels that were 0.10 units lower than those without hypertension.

Table 2.

Cross-sectional associations between pain, cognitive outcomes, and chronic conditions.

Variables	Cognitive outcomes	Model 1a: B (SE)	P-value	Model 2b: B (SE)	P-value
Pain (yes/no)	Episodic memory	−0.46 (0.04)	< .01	−0.17 (0.04)	< .001
	EF	−0.63 (0.20)	.0021	0.08 (0.22)	.72
Pain severity (n = 972)	Episodic memory	−0.05 (0.11)	.004	−0.04 (0.02)	.005
	EF	−0.28 (0.09)	.002	−0.30 (0.10)	< .01
Moderators
Treatment	Episodic memory	–	–	0.02 (0.03)	.43
	EF	–	–	−0.16 (0.20)	.42
Hypertension	Episodic memory	–	–	0.07 (0.08)	.95
	EF	–	–	−0.64 (0.48)	.17
Diabetes	Episodic memory	–	–	0.01 (0.1)	.91
	EF	–	–	0.46 (0.60)	.44
HIV	Episodic memory	–	–	0.12 (0.08)	.16
	EF	–	–	−0.26 (0.52)	.61
Depressive symptoms	Episodic memory	–	–	−0.01 (0.004)	.01
	EF	–	–	0.07 (0.02)	.001
No depressive symptoms	Episodic memory	–	–	−0.12 (0.04)	< .01
Group (CES-Dc <20)	EF	–	–	−0.47 (0.28)	.09
Depressive symptoms	Episodic memory	–	–	−0.26 (0.07)	< .001
Group (CES-D ≧20)	EF	–	–	0.68 (0.39)	.08

^aModel 1 is an unadjusted model.

^bModel 2: Model 1+ adjusts for age (as continuous); gender; educational level (as categorical variable); Wealth index; CES-D score, presence at baseline of diabetes, high blood pressure, HIV (all yes vs no).

^cCenter for Epidemiologic Studies Depression Scale.

Further analysis revealed that only depressive symptoms, but not diabetes, hypertension, or HIV, moderated the relationship between pain and episodic memory (ß = −0.01 [P = .01]) and EF (ß = 0.07 [P = .001]). The negative association between pain and episodic memory was exacerbated in individuals in the high depressive symptoms group—indicated by a CES-D score of ≥20 (ß = −0.26 [P < .001]), compared to those in the non-­depressed group—indicated by a CES-D of <20 (ß = −0.12 [P < .01]). This distinct pattern of more accentuated impaired performance in individuals with high depressive symptoms compared to those without was observed only for episodic memory and not for EF (CES-D < 20 group [P = .09]; CES-D ≥ 20 group [P = .08]) (Table 2).

Among pain-reporting participants (n = 972), depressive symptoms moderated the relationship between pain severity and episodic memory (P = .02), with a more substantial negative impact of pain severity on cognition in individuals with high depressive symptoms. However, pain severity did not predict cognition and EF in either participants with or without depression symptoms (CES-D < 20 group [P = .5]; CES-D ≥ 20 group [P = .5]). There was no moderating effect of diabetes, hypertension, HIV, or depressive symptoms on the relationship between pain severity and EF (Table 2).

Among the 972 participants who reported experiencing pain, 415 have received treatment. Compared to those who did not receive treatment, those with treatment had a lower mean CES-D score and had lower rates of hypertension. Those who received treatment had a mean episodic memory score of −0.431 compared to −0.274 in the non-treatment group and a mean score of 5.35 for EF compared to 6.15 in the non-treatment group. Pain treatment was a significant predictor of episodic memory and EF in unadjusted models (ß = −0.16 [P = .02], ß = −0.80 [P < .03], respectively), indicating that those receiving treatment had lower episodic memory and EF scores than those not receiving treatment. This effect was attenuated after adjusting for covariates for both episodic memory [P = .6] and EF [P = .33].

Longitudinal analysis

Table 3 presents the descriptive characteristics of the longitudinal sample, categorized into three groups based on participants’ pain responses across waves 2 and 3: no pain (n = 2 140), acute pain (n = 751), and persistent pain (n = 84). Participants who reported no pain across the waves had a mean episodic memory score of 14 (95% CI [12, 16]), compared to 13 (95% CI [10, 15]) among those with acute pain and 12 (95% CI [7, 14]) among those who reported pain in both waves (persistent pain). This corresponds to an approximate 1–2-point decline in episodic memory scores associated with pain exposure. In contrast, EF scores showed less variation, whereby those with no pain reported across waves had a mean score of 6 (95% CI [2, 10]), while both the acute and persistent pain groups had a score of 5 (acute: 95% CI [2, 9]; persistent: 95% CI [3, 8]). This suggests a weaker longitudinal relationship between EF and pain compared to episodic pain.

Table 3.

Descriptive characteristics of the longitudinal sample by response to pain at each wave.

Characteristic	Overall, N = 2 975	No pain, N = 2140	Acute pain, N = 751	Persistent pain, N = 84
Education
No formal education	1 217 (41%)	841 (39%)	332 (44%)	44 (52%)
Some primary (1–7 years)	1 091 (37%)	760 (36%)	298 (40%)	33 (39%)
Some secondary (8–11 years)	379 (13%)	301 (14%)	72 (9.6%)	6 (7.1%)
Secondary or more (12+ years)	281 (9.5%)	234 (11%)	46 (6.1%)	1 (1.2%)
Missing	7	4	3	0
Female (%)	1 714 (58%)	1 168 (55%)	485 (65%)	61 (73%)
Age	65 (57, 74)	63 (57, 72)	70 (61, 79)	75 (69, 82)
Wealth index
1 (poorest)	549 (19%)	413 (20%)	127 (17%)	9 (11%)
2	568 (20%)	382 (18%)	166 (22%)	20 (24%)
3	573 (20%)	407 (20%)	149 (20%)	17 (20%)
4	600 (21%)	441 (21%)	137 (19%)	12 (27%)
5 (wealthiest)	617 (21%)	441 (21%)	161 (22%)	15 (18%)
Missing	68	56	11	1
Diabetes
Yes	537 (18%)	358 (17%)	157 (21%)	22 (26%)
No	2 438 (82%)	1 782 (83%	594 (79%)	62 (74%)
Hypertension
Yes	2 061 (70%)	1 410 (66%)	579 (77%)	72 (86%)
No	902 (30%)	719 (34%)	171 (23%)	12 (14%)
Missing	12	11	1	0
HIV
Negative	1 676 (69%)	1 138 (66%)	482 (75%)	56 (84%)
Positive	760 (31%)	592 (34%)	157 (25%)	11 (16%)
Missing	539	410	112	17
Depressive symptoms	11 (5, 22)	10 (5, 22)	12 (6, 22)	17 (11, 20)
Non-depressed group	2056 (71%)	1 495 (70%)	504 (67%)	57 (68%)
Depressed group	919 (31%)	645 (30%)	247 (33%)	27 (32%)
Episodic memory	0.02 0.99)	0.11 (0.94)	−0.18 (1.04)	−0.62 (1.07)
Episodic memory (Raw)a	13.4 (3.6)	13.8 (3.5)	12.7 (3.8)	11.1 (3.9)
EF	6.0 (4.2)	6.2 (4.2)	5.5 (4.1)	5.6 (4.0)

Note. N (%) or Mean (SD)

^aSum of correct items on words list (immediate + delayed recall) and four orientation questions. Ranges from 0 to 24.

The relationship between pain and episodic memory was significant in both unadjusted (ß = −0.29, [P < .001]) and adjusted models (ß = −0.18 [P < .001]), indicating that pain negatively affects memory longitudinally, thereby reinforcing the findings from the cross-sectional analysis. For EF, the relationship with pain was significant in the unadjusted model (ß = −0.49 [P = .0002]) but not after adjustments (ß = −0.22 [P = .26]).

Consistent with the cross-sectional results, chronic conditions, such as HIV, hypertension, and diabetes, did not significantly moderate the relationship between pain and cognitive outcomes (P > .05). However, higher depressive symptoms were associated with lower EF (ß = −0.05 [P < .001]), although depressive symptoms did not moderate the pain-cognition relationship.

In unadjusted and adjusted models, episodic memory scores were significantly lower for both the “acute pain” and “persistent pain” groups compared to the “no pain” group, highlighting the detrimental effect of persistent or recurrent pain on overall cognitive function (Table 4). For EF, only the “acute pain” group showed significantly poorer EF compared to the “no pain” group in both unadjusted and adjusted models, suggesting even occasional pain can negatively impact EF. However, after adjustments, this effect was not observed in the “persistent pain” group (Table 4).

Table 4.

Adjusted linear mixed models on the association between pain severity and cognition between Waves 2 and 4.

	No pain (n = 2140)	Acute pain (n = 751)	P-value	Persistent pain (n = 84)	P-value
Episodic memory (n = 3803)	Reference	−0.13 (0.03)	< .001	−0.30 (0.08)	< .001
EF (n = 3369)	Reference	−0.33 (0.12)	.04	−0.046 (0.44)	.92

Note. Fully adjusted model includes age (as continuous); gender; educational level (as categorical variable); wealth index; CES-D score, presence at baseline of diabetes, high blood pressure, HIV (all yes vs no)

Figure 1 illustrates that both acute and persistent pain was associated with lower episodic memory scores and, to a lesser extent, poorer EF.

Figure 1.

Comparison of cognitive changes in pain groups between Waves 2 and 3.

Point estimates representing means and standard errors.

Discussion

The present study aimed to investigate the cross-sectional and longitudinal associations between cognition and pain in a cohort of middle-aged and older adults in rural South Africa. Additionally, this study explored the moderating effects of chronic conditions and pain treatment on these relationships. A cross-sectional analysis revealed a significant relationship between pain presence and both episodic memory and EF, with the presence of pain linked to poorer cognitive health. There was also a notable association between pain severity and cognitive outcomes, where individuals with higher pain severity exhibited poorer cognitive performance. Furthermore, individuals who reported experiencing pain and received treatment had lower cognitive scores compared to those who did not. Depressive symptoms were found to moderate the relationship between pain severity and episodic memory, with a more substantial negative effect of pain severity on cognition observed in individuals with high depressive symptoms compared to those without. However, this moderating effect of depressive symptoms was not observed on the relationship between pain severity and EF. In the longitudinal analysis, the presence of pain had a negative effect on episodic memory in both the acute and persisting pain subgroups; however, the negative effect of pain on EF was only significant in the acute pain group.

To our knowledge, this is the first study to investigate the relationship between pain and cognition in an African population. Previous research on pain and cognition typically includes smaller clinical samples of patients with chronic pain or community-dwelling older adults in the US and UK.^16,18,35 Findings from this study contribute to the understanding of longitudinal associations between pain and cognition and address gaps in the literature by exploring potential moderators and the role of treatment in shaping cognitive outcomes. In a longitudinal study using data from the English Longitudinal Study (ELSA), a community-based sample of 6 515 older adults in England, researchers found no association between pain (yes vs no) and verbal fluency, and between pain and processing speed, at the four-year follow-up.¹⁶ The only significant association reported was between severe pain and memory (ß = −0.36 [P = .004]). However, researchers noted high attrition, differences in cognitive measures, and cofounders as reasons for contrasting results.¹⁴ Similarly, a longitudinal analysis using data from the Health and Retirement Study (HRS), following a population-based sample of 10 065 older adults across ten years, found significant associations between incident and persistent pain with cognition.¹⁴ Participants with persistent pain experienced a mean of 9.2% (95% CI, 2.8%–15.0%) more rapid decline in memory scores than participants without persistent pain.¹³ Our finding that pain negatively impacted episodic memory and EF within just two years, compared to most studies that examine longer periods of 4–15 years, highlights how early these cognitive effects can be observed.^14,16,17

Although we expected that the longitudinal analysis would reveal a negative association between EF and both acute and persisting pain, our findings indicated no association between persisting pain and EF. One possible explanation is the differences in the mechanisms that drive the relationship between these cognitive processes and pain. Working memory, attention, and pain share neural networks and consequently compete for limited resources, causing negative changes to cognition.³⁶ The brain has a limited capacity to focus, process information, and manage tasks. When pain demands a portion of these resources, it reduces the brain’s ability to allocate them to other tasks, leading to cognitive difficulties.³⁷ Research suggests that for EF, it is the stress from experiences of pain that drives EF impairment. Chronic exposure to psychosocial stress can dysregulate the neural pathways activated by dopamine, adrenaline, and cortisol, contributing to impaired EF.²⁵ It is possible that over time, those with persisting pain develop coping strategies or other factors to reduce the stress that impairs cognitive function. In the case of chronic pain, research has found that the body can become used to pain and adapt to it.³⁸ In addition, we found no significant moderation effect of pain treatment on the relationship between pain severity and cognition. This suggests that simply receiving pain treatment does not necessarily buffer the cognitive consequences of pain. Future work should explore the type, duration, and effectiveness of treatment in modifying the cognitive impact of chronic pain. Future studies should expand on the differences in mechanisms that underly relationships between pain and cognition.

We also expected diabetes, HIV, and hypertension to exacerbate the relationship between pain presence and cognition. However, we only found hypertension to be a predictor of pain presence but not a moderator. This finding aligns with previous research.³⁹ A US study on a clinical sample comparing the prevalence of hypertension in individuals with chronic pain and those without found that chronic pain intensity was a significant predictor of hypertensive status.⁴⁰ Although diabetes and HIV have also been associated with increased pain severity, our analysis showed no significant relationship.^41,42 One possible explanation is that participants with HIV and diabetes might have more effective treatment regimens that mitigate pain.⁴³ Another possibility is that differing mechanisms may drive the relationship between these chronic conditions and pain. For example, HIV is often associated with increased inflammation, while diabetes can lead to nerve damage, both of which may contribute to pain.^41,42 Hypertension, on the other hand, does not directly cause pain but may affect pain sensitivity by altering how pain is processed or perceived.^39,40 The condition instead moderates pain sensitivity, which may explain our findings.³⁹ Additionally, we observed higher rates of hypertension in our sample compared to the other two conditions, which may have affected our findings.

Similarly, we hypothesized that individuals receiving pain treatment would exhibit better cognitive scores than those without; we observed the opposite effect in the unadjusted model and an attenuated effect, marginally significant, in the presence of adjustments. This may be attributed to the known association between the long-term use of opioids and cognitive impairment.^44,45 Opioids block pain through the disruption of normal neurotransmitter activity in the brain, which can lead to cognitive impairment.⁴⁶ The association between opioids and cognition has greater implications for older adult populations since cognitive decline is associated with difficulties in performing everyday tasks, decision-making, and repeated risk of falls.^47,48 Future research should investigate alternative interventions to mitigate the effect of pain management medication on cognition.

In contrast to our unexpected findings regarding pain treatment and chronic conditions, our hypothesis concerning the moderating role of depressive symptoms was supported by our results. Depression has been consistently associated with increased vulnerability to pain and impairments in various cognitive domains, including EF and working memory.^39,49 Previous research has explored how both pain and depression independently predict cognitive decline but has not explicitly examined how depression moderates the relationship between pain and cognition. Our findings demonstrate that depressive symptoms exacerbate the negative impact of pain severity on cognitive function. Our findings of depression as a moderator in the pain/cognition relationship align with other studies that have identified depression as a mediator in the pain-cognition relationship. While our analysis focused on the moderating role of depressive symptoms in the relationship between pain and cognitive function, other studies have explored depression as a potential mediator in this pathway. For instance, pain may contribute to the development of depressive symptoms, which in turn could negatively impact cognitive performance. Although we did not formally test mediation in this study, prior work has taken this approach. A longitudinal study of older adults in England found that co-occurring pain and depressive symptoms were associated with greater declines in global cognition and memory. Similarly, a national study of older adults in China identified depression as a moderator of the pain–dementia relationship. Together, these findings highlight the complex interplay between pain, depression, and cognitive outcomes, and suggest that future work should consider both mediating and moderating mechanisms.^50,51 However, it is worth noting that our analysis did not find depressive symptoms to be a significant moderator of EF, which may be due to the smaller subgroup sizes for EF compared to overall cognition, potentially affecting statistical significance. Overall, our results emphasize the need for future research to explore further the complex interplay between pain, depressive symptoms, and cognitive function.

A key strength of this study is its expansion of the understanding of pain-cognition relationships to a more diverse cultural and socioeconomic setting. Our analysis also uniquely considers the role of chronic conditions in shaping this relationship, an area that has been underexplored in previous research. However, there are also a few limitations worth noting. The measure for pain was limited to only three questions, limiting our ability to create a more robust measurement of pain and chronic pain. We also did not have information on the timing of pain, such as for incident pain, whether the pain severity was within the last 24 hours or at that moment. Additionally, the smaller sample size for EF measures compared to episodic memory may have limited the statistical power to detect associations, potentially explaining why some associations were observed for episodic memory but not for EF. Importantly, differences in measurement strategies and potential selection bias due to missing data, refusals, or loss to follow-up may limit the generalizability of our findings. We were unable to apply survey weights or principled techniques to address missingness, which further restricts the representativeness of the analytic sample. Participants included in the final sample may systematically differ from those excluded, potentially being healthier and less cognitively or physically impaired by pain, while those excluded may have higher levels of morbidity, lower socioeconomic status, or greater cognitive impairment. Additionally, assumptions underlying the linearization of outcome measures may not be fully met, influencing results. Given the small effect sizes observed, our findings may be particularly sensitive to these methodological limitations. Despite these limitations, the analytic sample is drawn from a population-based cohort that broadly reflects the regional population, supporting the relevance of the findings within this context. These limitations warrant careful consideration when interpreting our results.

An inherent limitation of our cross-sectional analyses is the potential for reverse causality, where cognitive impairment may lead to disability or inactivity, subsequently causing pain rather than pain leading to cognitive decline. While our study includes longitudinal analyses, causal directionality cannot be definitively established. Moreover, given the sociodemographic and healthcare context of our rural South African population, this reverse causal pathway may be less common in our sample. These considerations highlight the need for caution in interpreting the relationship between pain and cognition in this population.

Future research should explore potential moderators of the cognition-pain relationship, including the role of physical comorbidities. Previous studies have suggested that the observed relationship between cognition and pain may reflect the influence of comorbid conditions linked to pain, rather than pain itself.¹⁶ Research should also examine whether the relationship between pain and cognition varies depending on the location of the pain. This study was limited in addressing this question due to the low prevalence of participants reporting pain in multiple areas. An additional potential new direction for future work is to examine positive psychosocial factors as possible mediators of the relationship between cognition and pain. Social relations, religious networks, and perceived control have been identified as protective factors against cognitive decline.⁵² Future studies should explore whether such factors can also mitigate the impact of pain on cognitive function.

Author contributions

Camryn Dixon (Conceptualization, Formal Analysis, Writing Original Draft [equal]), Tamara P Taporoski (Conceptualization [equal], Formal Analysis-Lead, Methodology [equal], Supervision [equal], Visualization-Lead, Writing Original Draft-Supporting, Writing Review Editing [equal]), F Xavier Gomez-Olive Casas (Writing Review Editing [equal]), Stephen M Tollman (Writing Review Editing [equal]), Lisa F Berkman (Writing Review Editing [equal]), and Darina T Bassil (Conceptualization, Methodology, Supervision, Validation, Writing Review Editing [equal])

Supplementary material

Supplementary material is available at The Journals of Gerontology, Series A: Biological Sciences and Medical Sciences online.

Funding

This work was supported by the National Institute on Aging at the National Institutes of Health (grant number P01 AG041710); the University of the Witwatersrand; the South African Medical Research Council; and the Welcome Trust, UK (grant numbers 058893/A/99A, 069683/Z/02/Z, 085477/Z08/Z). The funding agencies had no role in the design, analysis, interpretation of data, or production of the manuscript.

Conflicts of interest

None declared.