Brain predicted age in chronic pelvic pain: A study by the Multidisciplinary Approach to the Study of Chronic Pelvic Pain (MAPP) Research Network

1. Introduction

The biological age of the brain can be estimated from T1-weighted structural MRI of the brain (i.e., brain-predicted age). From brain-predicted age, brain-predicted age gap estimation (brainAGE; brain-predicted – chronological age) can be calculated and used as a biomarker of chronological and pathophysiological changes in brain health[12]. Larger brainAGE values have been associated with increased risk of mortality[13] and reduced cognitive function[4] in healthy adults, and poorer outcomes post-stroke[37]. This approach also shows promise for early brain pathology detection [15,17] and measurement of disease progression [16,18].

Various chronic pain conditions are associated with modifications in brain structure and function[5,6,32,54] similar to that in healthy chronological aging[53]. This opens the possibility that chronic pain may accelerate aging-related changes in the brain. However, several cross-sectional studies of the impact of chronic pain conditions on brainAGE report conflicting results[14,27,50,59]. Cruz-Almeida, et al. found community-dwelling older adults with various chronic pain conditions exhibited higher brainAGE than older adults without chronic pain[14]. However, Sörös and Bantel found no difference in the brainAGE of older adults with and without chronic pain using the same brain-age prediction algorithm[50]. The conflicting results may be due to the relatively small sample sizes (n<60), the age of the population, or the heterogeneity of pain conditions in each sample, given that the effect of chronic pain on the brain is not uniform across conditions[24,26]. More recent studies of brainAGE and chronic pain focus on homogenous chronic musculoskeletal pain cohorts (e.g. knee osteoarthritis, chronic low back pain)[27,55,59]. These studies suggest that the effect of chronic pain on brainAGE may depend on the pain condition – with the largest study reporting that people with chronic knee osteoarthritis have older brains than people without or people with chronic low back pain[55].

Here we aim to address some limitations of prior work. Urologic chronic pelvic pain syndrome (UCPPS) includes chronic prostatitis and interstitial cystitis/bladder pain syndrome, which impact millions of people in the United States spanning a large age range [3,51]. The structural, functional, and chemical changes in the brain in people with UCPPS collectively suggest alterations of pain-processing pathways and central processes[22,23,25,28,33,43,58]. However, brainAGE of people with UCPPS has not been previously examined, nor has its relationship with several established stratifying factors in the UCPPS population, including inflammation[47–49], widespread pain[31,33], and psychological comorbidities[40,41].

We analyzed data from 564 participants (492 with UCPPS, 72 pain-free controls) from the Multidisciplinary Approach to the Study of Chronic Pelvic Pain (MAPP) database to examine the effect of UCPPS on brainAGE values relative to pain-free controls. We hypothesized that people with UCPPS would have larger brainAGE values than pain-free controls. We also completed secondary analyses to understand the association between established stratifying factors in UCPPS and brainAGE.

2. Materials and Methods

2.1. Study Design and MAPP Network Organization

This was a secondary analysis of data collected by the MAPP Research Network as part of a longitudinal observational cohort study that was registered at Clinicaltrials.gov ( NCT02514265). The MAPP Research Network consists of six recruiting sites (Los Angeles, CA; Chicago, IL; St Louis, MO; Iowa City, IA, Seattle, WA, and Ann Arbor, MI), a Data Coordinating Core (DCC; Philadelphia, PA), a Tissue Analysis and Technology Core (TATC; Aurora, CO); and a Neuroimaging scan repository and Reading Center (UCLA/USC; Los Angeles, CA). Data collection procedures have been previously published[1,9]. All procedures were approved by institutional review boards at the participating institutions, and all subjects provided informed consent. This analysis used retrospective clinical and neuroimaging data collected during the first study visit across these six study sites.

2.2. Participants

This study analyzed the structural T1-weighted magnetic resonance images (MRI) that were obtained at the baseline assessment of the longitudinal study from 492 individuals with UCPPS and 72 pain-free controls. Full inclusion and exclusion criteria for participants in the original study have been published [9]. Most participants with UCPPS in the study also provided whole blood samples that were analyzed for inflammatory markers as described below in a subset of these participants (n=250) who were selected to reflect the broader age and sex distribution of the whole cohort.

2.3. Data Analysis

2.3.1. Brain-Predicted Age Estimation

To estimate brain-predicted age, structural MRI scans were analyzed with the brainageR software v2.1[10]. This software employs a Gaussian Processes regression model trained on structural MRI scans from a cohort of n=3377 healthy individuals (age range 18–92 years; mean±SD age 40.6 ± 21.4) from seven publicly available neuroimaging datasets. The brainageR model was validated on held out test data as well as an independent data set (n=857) of healthy adults of similar ages. This general approach to brain age prediction has been previously described[12,13]. Briefly, brainageR uses SPM12 in MATLAB to segment and normalize raw T1-weighted images and FSL’s slicedir script to create images of slices in the sagittal, coronal, and axial direction for visual inspection. The individual segmented and normalized images were then vectorized in R using the RNfiti package[8]. Brain-predicted age was estimated using the kernlab package[29] and the pre-trained regression model. This brain age prediction model was selected over others[21,56] for this study because 1) it is publicly available – promoting reproducibility, 2) it was trained on MRI scans from seven study sites – a similar site composition to our own dataset, and 3) two previous studies of chronic pain and brain age used a similar model[14,50]. Mean absolute error of the brain age prediction model was calculated using the metrics package [19] comparing the brain-predicted age to the actual chronological age. We then used a Pearson’s correlation and the coefficient of determination to assess the relationship between chronological and brain-predicted ages.

2.3.2. Brain-Predicted Age Difference

We calculated brainAGE by subtracting each participant’s chronological age from their brain-predicted age. Negative values of brainAGE indicated brains that were younger than the participants’ chronological age; whereas positive values indicated brains that were older than chronological age. BrainAGE value of zero indicated a perfect chronological age prediction.

2.3.3. Inflammatory Markers

A detailed description of the inflammatory phenotyping protocol for the MAPP Symptom Pattern Study has been reported[48]. In brief, the Truculture system (Rules Based Medicine) was used to test whole blood ex-vivo immune responses to lipopolysaccharide (LPS), fibroblast-stimulating lipopeptide (FSL), and media after 24-hours of incubation. Seven cytokines/chemokines were measured using Luminex Xmap technology. These were monocyte chemoattractant protein-1 (MCP-1), macrophage inflammatory protein 1-α (MIP-1α), IL-1β, IL- 6, IL-8, IL-10, and tumor necrosis factor-α (TNF-α). Ranges of each assay and their intra- and inter-assay coefficients of variation are provided in the supplementary material.

2.3.4. Widespread Pain

Widespread body pain is a common comorbidity in UCPPS[33]. Pain widespreadness was assessed using the collaborative health outcomes information registry body map[46]: patients were asked to select any of 76 body sites where they felt pain and then rate their pain 0–10 at all selected sites. These data were further reduced to 12 nonpelvic regions, and the pain rating averaged across body sites included in each region[48].

2.3.5. Psychological comorbidities

Anxiety and depression were both assessed at the time of MRI with the Hospital Anxiety and Depression scale[61]: possible scores for both anxiety and depression were 0–21.

2.3.6. Statistical Analysis

All analyses were conducted in RStudio (R v.4.2.2)[45]. To assess differences in age and sex between the UCPPS and control groups, t-tests and chi-squared tests were used. To determine the association of UCPPS with brainAGE, we first fit a linear model that included fixed effects for group, sex, and chronological age. Sex and age were included in the model as covariates consistent with prior studies[14,37,56]. To account for possible clustering due to study site, we then fit a linear mixed-effect model with the lme4 package[2] that included a random intercept for study site. Using the Akaike information criterion (AIC) we performed a model comparison and that with the lowest AIC was selected as the final base model. We then included terms to test for interactions between age*sex, group*age, and group*sex on brainAGE. Interaction terms for age*sex and group*age were included to evaluate if the known age bias in braineageR [10] differed by group and sex.

To understand the impact of UCPPS symptom duration on brainAGE, we also fit a model with fixed effects for symptom duration, sex, and chronological age and a random intercept for study site. We did not include a term for group because it covaried with symptom duration (as controls had a symptom duration of zero). For all analyses, we grand-mean centered age to allow for more interpretable estimates of age interactions. We verified linear regression assumptions as well as the presence of influential points using graphical evaluations of normality and homogeneity of the residuals. Statistical significance was set a priori at 0.05 (two-sided).

To examine the relationship between three known stratifying factors within UCPPS (inflammatory markers, psychological comorbidities, and widespread pain) and brainAGE, we completed permutation analyses[57]. We fit three separate linear models using either standardized inflammatory markers (mean=0, standard deviation=1; Equation 1 where LPS indicates lipopolysaccharide, FLS indicates fibroblast-stimulating lipopeptide, and null indicates incubation in media), individual item scores on the HADS (Equation 2; where HADS_Q1 stands for response to question 1 on the HADS), or region-specific pain scores for calculating widespread pain (Equation 3; where bpi stands for Brief Pain Inventory) to predict the original brainAGE values. To control for the known effect of chronological age on brainAGE, each model also included standardized age (mean=0, standard deviation=1) as an independent variable.

$$\begin{gathered} \mathrm{I}\mathrm{n}\mathrm{f}\mathrm{l}\mathrm{a}\mathrm{m}\mathrm{m}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r}\mathrm{y}\mathrm{M}\mathrm{a}\mathrm{r}\mathrm{k}\mathrm{e}\mathrm{r}\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{c}\mathrm{t}\mathrm{e}\mathrm{d}\mathrm{B}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{A}\mathrm{G}\mathrm{E}~{\mathrm{\beta }}_1\mathrm{m}\mathrm{c}\mathrm{p}1\_\mathrm{l}\mathrm{p}\mathrm{s}+{\mathrm{\beta }}_2\mathrm{m}\mathrm{i}\mathrm{p}1\mathrm{a}\_\mathrm{l}\mathrm{p}\mathrm{s}+{\mathrm{\beta }}_3\mathrm{i}\mathrm{l}1\mathrm{b}\_\mathrm{l}\mathrm{p}\mathrm{s}+{\mathrm{\beta }}_4\mathrm{i}\mathrm{l}6\_\mathrm{l}\mathrm{p}\mathrm{s} \\ +{\mathrm{\beta }}_5\mathrm{i}\mathrm{l}8\_\mathrm{l}\mathrm{p}\mathrm{s}+{\mathrm{\beta }}_6\mathrm{i}\mathrm{l}10\_\mathrm{l}\mathrm{p}\mathrm{s}+{\mathrm{\beta }}_7\mathrm{t}\mathrm{n}\mathrm{f}\_\mathrm{l}\mathrm{p}\mathrm{s}+{\mathrm{\beta }}_8\mathrm{m}\mathrm{c}\mathrm{p}1\_\mathrm{f}\mathrm{l}\mathrm{s}1+{\mathrm{\beta }}_9\mathrm{m}\mathrm{i}\mathrm{p}1\mathrm{a}\_\mathrm{f}\mathrm{l}\mathrm{s}1+{\mathrm{\beta }}_{10}\mathrm{i}\mathrm{l}1\mathrm{b}\_\mathrm{f}\mathrm{l}\mathrm{s}1+{\mathrm{\beta }}_{11}\mathrm{i}\mathrm{l}6\_\mathrm{f}\mathrm{l}\mathrm{s}1+ \\ {\mathrm{\beta }}_{12}\mathrm{i}\mathrm{l}8\_\mathrm{f}\mathrm{l}\mathrm{s}1+{\mathrm{\beta }}_{13}\mathrm{t}\mathrm{n}\mathrm{f}\mathrm{a}\_\mathrm{f}\mathrm{l}\mathrm{s}1+{\mathrm{\beta }}_{14}\mathrm{m}\mathrm{c}\mathrm{p}1\_\mathrm{n}\mathrm{u}\mathrm{l}\mathrm{l}+{\mathrm{\beta }}_{16}\mathrm{m}\mathrm{i}\mathrm{p}1\mathrm{a}\_\mathrm{n}\mathrm{u}\mathrm{l}\mathrm{l}+{\mathrm{\beta }}_{17}\mathrm{i}\mathrm{l}1\mathrm{b}\_\mathrm{n}\mathrm{u}\mathrm{l}\mathrm{l}+{\mathrm{\beta }}_{18}\mathrm{i}\mathrm{l}6\_\mathrm{n}\mathrm{u}\mathrm{l}\mathrm{l}+{\mathrm{\beta }}_{19}\mathrm{i}\mathrm{l}8\_\mathrm{n}\mathrm{u}\mathrm{l}\mathrm{l} \\ +{\mathrm{\beta }}_{20}\mathrm{i}\mathrm{l}10\_\mathrm{n}\mathrm{u}\mathrm{l}\mathrm{l}+{\mathrm{\beta }}_{21}\mathrm{t}\mathrm{n}\mathrm{f}\mathrm{a}\_\mathrm{n}\mathrm{u}\mathrm{l}\mathrm{l}+{\mathrm{\beta }}_{22}\mathrm{c}\mathrm{h}\mathrm{r}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{a}\mathrm{g}\mathrm{e}+\mathrm{\epsilon } \end{gathered}$$ (Eq 1)

$$\begin{gathered} \mathrm{P}\mathrm{s}\mathrm{y}\mathrm{c}\mathrm{h}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{C}\mathrm{o}\mathrm{m}\mathrm{o}\mathrm{r}\mathrm{b}\mathrm{i}\mathrm{d}\mathrm{i}\mathrm{t}\mathrm{y}\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{c}\mathrm{t}\mathrm{e}\mathrm{d}\mathrm{B}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{A}\mathrm{G}\mathrm{E}~{\mathrm{\beta }}_1\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}1+{\mathrm{\beta }}_2\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}2+{\mathrm{\beta }}_3\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}3+ \\ {\mathrm{\beta }}_4\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}4+{\mathrm{\beta }}_5\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}5+{\mathrm{\beta }}_7\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}6+{\mathrm{\beta }}_7\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}7+{\mathrm{\beta }}_8\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}8+{\mathrm{\beta }}_9\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}9+ \\ {\mathrm{\beta }}_{10}\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}10+{\mathrm{\beta }}_{11}\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}11+{\mathrm{\beta }}_{12}\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}12+{\mathrm{\beta }}_{13}\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}13+{\mathrm{\beta }}_{14}\mathrm{H}\mathrm{A}\mathrm{D}\mathrm{S}\_\mathrm{Q}14+{\mathrm{\beta }}_{15}\mathrm{c}\mathrm{h}\mathrm{r}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{l} \\ \mathrm{a}\mathrm{g}\mathrm{e}+\mathrm{\epsilon } \end{gathered}$$ (Eq 2)

$$\begin{gathered} \mathrm{W}\mathrm{i}\mathrm{d}\mathrm{e}\mathrm{s}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{d}\mathrm{P}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{c}\mathrm{t}\mathrm{e}\mathrm{d}\mathrm{B}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{A}\mathrm{G}\mathrm{E}~{\mathrm{\beta }}_1\mathrm{b}\mathrm{p}\mathrm{i}\_\mathrm{p}\mathrm{e}\mathrm{l}\mathrm{v}\mathrm{i}\mathrm{s}+{\mathrm{\beta }}_2\mathrm{b}\mathrm{p}\mathrm{i}\_\mathrm{h}\mathrm{e}\mathrm{a}\mathrm{d}+{\mathrm{\beta }}_3\mathrm{b}\mathrm{p}\mathrm{i}\_\mathrm{r}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t}\mathrm{a}\mathrm{r}\mathrm{m}+ \\ {\mathrm{\beta }}_4\mathrm{b}\mathrm{p}\mathrm{i}\_\mathrm{l}\mathrm{e}\mathrm{f}\mathrm{t}\mathrm{a}\mathrm{r}\mathrm{m}+{\mathrm{\beta }}_5\mathrm{b}\mathrm{p}\mathrm{i}\_\mathrm{r}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t}\mathrm{l}\mathrm{e}\mathrm{g}+{\mathrm{\beta }}_6\mathrm{b}\mathrm{p}\mathrm{i}\_\mathrm{l}\mathrm{e}\mathrm{f}\mathrm{t}\mathrm{l}\mathrm{e}\mathrm{g}+{\mathrm{\beta }}_7\mathrm{b}\mathrm{p}\mathrm{i}\_\mathrm{t}\mathrm{r}\mathrm{u}\mathrm{n}\mathrm{k}+{\mathrm{\beta }}_8\mathrm{b}\mathrm{p}\mathrm{i}\_\mathrm{b}\mathrm{a}\mathrm{c}\mathrm{k}+{\mathrm{\beta }}_9\mathrm{c}\mathrm{h}\mathrm{r}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{a}\mathrm{g}\mathrm{e}+ \\ \mathrm{\epsilon } \end{gathered}$$ (Eq 3)

Then, we fit linear models on the three separate sets of stratifying factor-predicted brainAGE values (dependent) versus actual brainAGE values (independent) (Equations 4–6).

$$\mathrm{I}\mathrm{n}\mathrm{f}\mathrm{l}\mathrm{a}\mathrm{m}\mathrm{m}\mathrm{a}\mathrm{t}\mathrm{o}\mathrm{r}\mathrm{y}\mathrm{M}\mathrm{a}\mathrm{r}\mathrm{k}\mathrm{e}\mathrm{r}\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{c}\mathrm{t}\mathrm{e}\mathrm{d}\mathrm{B}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{A}\mathrm{G}\mathrm{E}~{\mathrm{\beta }}_1\mathrm{A}\mathrm{c}\mathrm{t}\mathrm{u}\mathrm{a}\mathrm{l}\mathrm{B}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{A}\mathrm{G}\mathrm{E}+\mathrm{\epsilon }$$ (Eq 4)

$$\mathrm{P}\mathrm{s}\mathrm{y}\mathrm{c}\mathrm{h}\mathrm{o}\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{c}\mathrm{a}\mathrm{l}\mathrm{C}\mathrm{o}\mathrm{m}\mathrm{o}\mathrm{r}\mathrm{b}\mathrm{i}\mathrm{d}\mathrm{i}\mathrm{t}\mathrm{y}\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{c}\mathrm{t}\mathrm{e}\mathrm{d}\mathrm{B}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{A}\mathrm{G}\mathrm{E}~{\mathrm{\beta }}_1\mathrm{A}\mathrm{c}\mathrm{t}\mathrm{u}\mathrm{a}\mathrm{l}\mathrm{B}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{A}\mathrm{G}\mathrm{E}+\mathrm{\epsilon }$$ (Eq 5)

$$\mathrm{W}\mathrm{i}\mathrm{d}\mathrm{e}\mathrm{s}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{d}\mathrm{P}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{c}\mathrm{t}\mathrm{e}\mathrm{d}\mathrm{B}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{A}\mathrm{G}\mathrm{E}~{\mathrm{\beta }}_1\mathrm{A}\mathrm{c}\mathrm{t}\mathrm{u}\mathrm{a}\mathrm{l}\mathrm{B}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{A}\mathrm{G}\mathrm{E}+\mathrm{\epsilon }$$ (Eq 6)

The coefficients of these models estimate the association of predicted BrainAGE (predicted by sets of inflammatory markers, anxiety/depression items, regional pain scores) with BrainAGE – i.e. how well the three sets of markers contribute to predicting BrainAge. The resulting regression coefficients were used for comparison between the permutations described in the following steps.

For the permutation analyses, actual brainAGE values were randomly re-assigned to each participant 5,000 times. This was done to ensure that factors associated with brainAGE were unlikely to arise randomly under the non-parametrically estimated null distribution. Using the permuted brainAGE data, Equations 1–6 were repeated to obtain regression coefficients under the null distribution. To determine if any relationship between brainAGE and sets of inflammatory markers, individual HADS item scores, or region-specific pain scores was due to chance, the number of times that the coefficient estimate after random re-assignment of the residuals exceeded the original coefficient value was determined (5,000 coefficient values compared to the original coefficient value). If the permutated coefficients from Equations 4–6 were equal to or larger than the original coefficients value less than 250 out of 5,000 times, then the probability of the original coefficient occurring by chance was less than 5%.

3. Results

3.1. Demographic and clinical data

Table 1 presents the demographic and clinical characteristics of the participants. All participants with at least one MRI scan at the longitudinal study’s baseline assessment were included. The only statistically significant difference between the groups was that the UCPPS group contained a higher proportion of women relative to the control group (p=0.02).

Table 1:

Demographic and clinical characteristics of the UCPPS and control groups

	UCPPS (N=492)	Control (N=72)	Total (N=564)
Age
Mean (SD)	44.2 (15.5)	41.1 (14.8)	43.8 (15.4)
Range	18.5 – 78.9	19.9 – 73.4	18.5 – 78.9
Participants by Age Group
< 30	109 (22.2%)	20 (27.8%)	129 (22.9%)
> 70	24 (4.9%)	1 (1.4%)	25 (4.4%)
30 – 39	118 (24.0%)	18 (25.0%)	136 (24.1%)
40 – 49	71 (14.4%)	11 (15.3%)	82 (14.5%)
50 – 59	100 (20.3%)	10 (13.9%)	110 (19.5%)
60 – 69	70 (14.2%)	12 (16.7%)	82 (14.5%)
Sex
Men	177 (36.0%)	37 (51.4%)	214 (37.9%)
Women	315 (64.0%)	35 (48.6%)	350 (62.1%)
Race
White	433 (88.0%)	47 (65.3%)	480 (85.1%)
Black	29 (5.9%)	9 (12.5%)	38 (6.7%)
American Indian	3 (0.6%)	0 (0.0%)	3 (0.5%)
Native Hawaiian	1 (0.2%)	0 (0.0%)	1 (0.2%)
Asian	4 (0.8%)	7 (9.7%)	11 (2.0%)
Multi-Race	16 (3.3%)	6 (8.3%)	22 (3.9%)
Other	6 (1.2%)	3 (4.2%)	9 (1.6%)
Ethnicity
N-Miss	1	0	1
Hispanic	32 (6.5%)	8 (11.1%)	40 (7.1%)
Non-Hispanic	459 (93.5%)	64 (88.9%)	523 (92.9%)
Education
< High School	0 (0.0%)	0 (0.0%)	0 (0.0%)
High School/GED	32 (6.5%)	4 (5.6%)	36 (6.4%)
Some College	118 (24.0%)	15 (20.8%)	133 (23.6%)
College/University Grad	221 (44.9%)	29 (40.3%)	250 (44.3%)
Professional/Grad Degree	121 (24.6%)	24 (33.3%)	145 (25.7%)
Site
NU	74 (15.0%)	12 (16.7%)	86 (15.2%)
UCLA	84 (17.1%)	13 (18.1%)	97 (17.2%)
U of Iowa	87 (17.7%)	10 (13.9%)	97 (17.2%)
U of Michigan	77 (15.7%)	12 (16.7%)	89 (15.8%)
U of Washington	83 (16.9%)	13 (18.1%)	96 (17.0%)
Wash U St Louis	87 (17.7%)	12 (16.7%)	99 (17.6%)
Symptom Duration
Mean (SD)	11.671 (11.548)	0.000 (0.000)	10.181 (11.467)
Range	0.000 – 59.000	0.000 – 0.000	0.000 – 59.000

3.2. Brain-predicted age estimation

The correlation between chronological age and brain-predicted age was R=0.90 (R²=0.81) with a mean absolute error of 5.6 years (Figure 1). These values are similar to the prediction performance in previous reports of brain-predicted ages in healthy people[11,13,35,56].

Figure 1: Relationship between chronological and brain-predicted age at baseline assessment for all participants.

The chronological age of each participant is plotted against their brain-predicted age based on their neuroimaging scan on their first visit (n=564). Data from all participants across study sites (color) and groups (symbol) are included. Data points on the unity line (solid black line) indicate a perfect prediction (chronological age = brain-predicted age). Data points below the unity line represent brains that were predicted to be younger than the participant’s chronological age (negative brainAGE), whereas data points above the unity line represent brains that were predicted to be older than the participant’s chronological age (positive brainAGE).

3.3. UCPPS and brainAGE

The model comparison resulted in a base model with fixed effects for mean-centered chronological age, sex, and group, as well as a random intercept for study site (Equation 7; where $i$ denotes subject, $j$ denotes site, ${\mathrm{\alpha }}_{\mathrm{j}}$ is a site random effect, and ${\mathrm{\epsilon }}_{\mathrm{I},\mathrm{j}}$ is the model residual).

$$\mathrm{B}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{A}\mathrm{G}{\mathrm{E}}_{\mathrm{i},\mathrm{j}}~{\mathrm{\beta }}_1\mathrm{U}\mathrm{C}\mathrm{P}\mathrm{P}{\mathrm{S}}_{\mathrm{i}}+{\mathrm{\beta }}_2\mathrm{F}\mathrm{E}\mathrm{M}\mathrm{A}\mathrm{L}{\mathrm{E}}_{\mathrm{i}}+{\mathrm{\beta }}_3\mathrm{M}\mathrm{E}\mathrm{A}\mathrm{N}-\mathrm{C}\mathrm{E}\mathrm{N}\mathrm{T}\mathrm{E}\mathrm{R}\mathrm{E}\mathrm{D}\mathrm{A}\mathrm{G}{\mathrm{E}}_{\mathrm{i}}+{\mathrm{\alpha }}_{\mathrm{j}}+{\mathrm{\epsilon }}_{\mathrm{I},\mathrm{j}}$$ (Eq 7)

Interactions of age*sex, group*sex, and age*group were then simultaneously added to this base model (Equation 8). Pain-free control and Male were coded as the reference for the categorical factors of group and sex, respectively.

$$\begin{gathered} \mathrm{B}\mathrm{r}\mathrm{a}\mathrm{i}\mathrm{n}\mathrm{A}\mathrm{G}{\mathrm{E}}_{\mathrm{i},\mathrm{j}}~{\mathrm{\beta }}_1\mathrm{U}\mathrm{C}\mathrm{P}\mathrm{P}{\mathrm{S}}_{\mathrm{i}}+{\mathrm{\beta }}_2\mathrm{F}\mathrm{E}\mathrm{M}\mathrm{A}\mathrm{L}{\mathrm{E}}_{\mathrm{i}}+{\mathrm{\beta }}_3\mathrm{M}\mathrm{E}\mathrm{A}\mathrm{N}-\mathrm{C}\mathrm{E}\mathrm{N}\mathrm{T}\mathrm{E}\mathrm{R}\mathrm{E}\mathrm{D}\mathrm{A}\mathrm{G}{\mathrm{E}}_{\mathrm{i}}+{\mathrm{\beta }}_4\mathrm{U}\mathrm{C}\mathrm{P}\mathrm{P}\mathrm{S}*\mathrm{F}\mathrm{E}\mathrm{M}\mathrm{A}\mathrm{L}{\mathrm{E}}_{\mathrm{i}}+ \\ {\mathrm{\beta }}_5\mathrm{F}\mathrm{E}\mathrm{M}\mathrm{A}\mathrm{L}\mathrm{E}*\mathrm{M}\mathrm{E}\mathrm{A}\mathrm{N}-\mathrm{C}\mathrm{E}\mathrm{N}\mathrm{T}\mathrm{E}\mathrm{R}\mathrm{E}\mathrm{D}\mathrm{A}\mathrm{G}{\mathrm{E}}_{\mathrm{i}}+{\mathrm{\beta }}_6\mathrm{U}\mathrm{C}\mathrm{P}\mathrm{P}\mathrm{S}*\mathrm{M}\mathrm{E}\mathrm{A}\mathrm{N}-\mathrm{C}\mathrm{E}\mathrm{N}\mathrm{T}\mathrm{E}\mathrm{R}\mathrm{E}\mathrm{D}\mathrm{A}\mathrm{G}{\mathrm{E}}_{\mathrm{i}}+{\mathrm{\alpha }}_{\mathrm{j}}+{\mathrm{\epsilon }}_{\mathrm{I},\mathrm{j}} \end{gathered}$$ (Eq 8)

Figure 2A displays the brainAGE values for both groups. Table 2 presents the results of our cross-sectional analysis of brainAGE. We found a statistically significant inverse association of age with brainAGE that differed by sex, with a stronger age association in females (β (SE)= −0.16 (0.02); calculated from the sum of the coefficients for mean-centered age term and the Female*mean-centered age interaction term) than males (β (SE)= −0.07 (0.05)). This inverse association is consistent with the known age bias in this algorithm [10] that, in our data, is even stronger for females than males. The association of age with brainAGE (which represents the association of chronological age with brainAGE in males, the reference value for sex) did not differ by group (UCPPS*mean-centered age interaction β (SE)= −0.005 (0.05), p=0.92). With adjustment for mean-centered age, we found a statistically significant effect of group that differed by sex (age-adjusted UCPPS*Female interaction β (SE)= 3.6 (1.6), p=0.02; Figure 2B). Female participants exhibited an effect of UCPPS in the hypothesized direction, with higher brainAGE values observed on average in females with UCPPS than female controls (β (SE)=1.6 (1.1); calculated from the sum of the coefficients of the UCPPS and the UCPPS*Female interaction terms)), while males with UCPPS exhibited lower brainAGE values than male control participants on average (β (SE)= −2.0 (1.1)). However, this effect of UCPPS on brainAGE was not statistically significant in either sex subgroup when considered independently (females (p=0.15) or males (p=0.07)). We found no significant effect of chronic pain symptom duration on brainAGE (symptom duration β= −0.002 (0.03), p=0.93). To confirm that the inclusion of pain-free controls in this analysis (with a symptom duration = 0) did not confound the result, we also examined the effect of symptom duration on brainAGE in only the people with UCPPS; symptom duration was not associated with brainAGE in this UCPPS group (β= −0.004 (0.03), p=0.87).

Figure 2: BrainAGE in UCPPS and pain-free controls.

A) Individual brainAGE values (symbols) are displayed over the mean values (bars) for men and women within the UCPPS (n=492) and control groups (n=72). The individual data are color-mapped by mean-centered age in UCPPS and pain-free controls. B) BrainAGE in men and women with UCPPS and controls adjusted for mean-centered age (estimated marginal means displayed with 95% confidence intervals) demonstrating that the presence of UCPPS had the opposite effect on brainAGE in males and females. (brainAGE, brain-predicted age gap estimation)

Table 2:

Results of a cross-sectional analysis of brainAGE at baseline in participants with UCPPS and pain-free controls.

	Estimate	Standard error	p-value
Intercept	−0.4	1.4	0.79
UCPPS (compared to Control)	−2.0	1.1	0.07
Female (compared to male)	−3.1	1.5	0.04
Mean-centered age	−0.07	0.05	0.21
UCPPS*Female	3.6	1.6	0.02
Female*mean-centered age	−0.09	0.04	0.01
UCPPS*mean-centered age	−0.005	0.05	0.92

3.4. Secondary analysis of stratifying factors in people with UCPPS and brainAGE

The demographic and clinical characteristics of the subset of participants with UCPPS (n=250; on whom both MRI and inflammatory data were collected) who were included in this secondary analysis are presented in Supplemental Table 1. Our permutation analysis found that the original coefficients (Equations 4–6) were larger than the permuted coefficients for all analyses (p<0.0001). To further explore the association of these factors with brainAGE, we compared the beta coefficient magnitudes for each inflammatory marker, HADS question, and pain area score between the actual data and the permuted data. For the inflammatory marker analysis, only age, TNF-α and MCP-1 (both under the LPS condition) had beta coefficients that were significantly higher than the permuted coefficients (Figure 3). For the HADS and widespread pain analyses, only age had a beta coefficient significantly higher than the permuted coefficients. This preliminarily suggests that biomarkers of inflammatory load, and not widespread pain or psychological comorbidities, may be related to brainAGE. Inflammatory markers of immune system reactivity to stimulation may have a more robust association with brain age compared to other factors that have previously been shown to subtype chronic pelvic pain patients.

Figure 3: Results of the permutation analysis with inflammatory markers.

Distribution of beta value magnitude from permutation analysis (n=250). True beta values are displayed in orange and the average permuted beta values are in green. Inflammatory markers are ascending along the x-axis by the significance.

4. Discussion

We performed cross-sectional analyses of brain age in a large and well-phenotyped cohort in chronic pain (UCPPS) and smaller sample of controls. Our primary objective was to evaluate the relationship between brain-predicted age and chronic pain in this population. Our cross-sectional analysis of baseline data revealed that the presence of UCPPS on brainAGE has the opposite effect in males and females – though the effect of UCPPS on brainAGE was not statistically significant in either males or females when considered independently. We also performed a secondary exploratory analysis to understand if stratifying factors of UCPPS may impact brainAGE and found preliminary evidence that brainAGE values are related to markers of inflammatory load in participants with UCPPS.

In contrast to musculoskeletal chronic pain conditions, we found that the presence of UCPPS alone did not have a significant positive association with brainAGE. The discrepancy between our results and those reported for chronic high-impact knee pain and chronic low back pain[27,55,59] may be explained in several ways. First, our results suggest that the association between the presence of UCCPS and larger brainAGE (older brains) may be related to sex. Specifically, the presence of UCPPS had the opposite effect on brainAGE in males and females – with females with UCPPS exhibited larger brainAGE values than female controls. This result is consistent with recent evidence from Hung et al., 2022[26], who found a larger effect of chronic pain on brainAGE in women across three different chronic pain conditions (trigeminal neuralgia, osteoarthritis, and chronic low back pain). Our results, in combination with the results from Hung et al., 2022[26], suggest that future studies of brainAGE in chronic pain conditions should evaluate the interaction between pain group and sex, and not just adjust for sex as a covariate[14,27].

However, there is an important detail to consider when interpreting these results. Of the participants in MAPP Symptom Pattern Study that were used for this analysis, males with UCPPS were largely diagnosed with chronic prostatitis, while females were diagnosed with interstitial cystitis/bladder pain syndrome. This may be relevant to the differences seen in brain age patterns compared to controls because male participants generally show fewer indications of nociplastic pain/central sensitization. As an example, women in the MAPP Symptom Pattern Study were about 77% more likely to have a comorbid pain condition than men [31], and more likely to have a widespread pain manifestation[34]. Additionally, female participants were more likely to self-report traumatic experiences in childhood[42]. Central sensitization and childhood trauma are both associated with increased immunoreactivity in female interstitial cystitis/bladder pain syndrome patients[39,48], suggesting that MAPP enrollment criteria may have biased the female participant sample toward factors associated with greater inflammatory activity and subsequent impact on the central nervous system. The factors that drive the differences between men and women reported here require further investigation.

Second, associations between chronic pain and brain changes[24], as well as accelerated brainAGE[26], are known to vary across chronic pain conditions. This is likely due to the presence or absence of mechanisms that mediate accelerated brain aging in a given chronic pain condition. However, the physiological processes that accelerate brain-predicted age are poorly understood; this is an emerging area of research. Several association studies in healthy aging populations have highlighted the potential role of factors like allostatic load (defined as a composite measure of different biological parameters)[13] and polygenic risk scores[56] on the brainAGE. Yet, the mechanism for accelerated brain aging in different pathological conditions remains unclear.

UCPPS patients show heightened ex vivo inflammatory responses to stimulation with common bacterial elements that suggest altered immune function with potential consequences for the health of the central nervous system[47–49]. In our exploratory analyses of the relationship between brainAGE and well-known stratifying factors of UCPPS, we showed that measures of immunoreactivity from whole blood were associated with brainAGE, specifically under LPS-stimulation. This provides preliminary evidence that brainAGE is related to inflammatory load in people with chronic pain, but future hypothesis-driven work is necessary to delineate the strength and direction of the relationship. It should be noted here that a number of measured analytes were not significantly associated with brainAGE, suggesting that the potential link between brainAGE and inflammation is specific to particular cytokines/chemokines or signaling pathways. While this finding is preliminary and exploratory, LPS’ action at Toll-Like Receptor 4 on circulating immune cells should be explored further in the context of brainAGE[38]. Advanced brain age has recently been linked to peripheral inflammation (TNF-α) in schizophrenia[30] and polygenic risk scores for elevated C-reactive protein in depression[20], suggesting that inflammatory markers measured in peripheral blood reflect changes in the central nervous system primarily associated with aging. These associations may also parallel well-established relationships between markers of inflammation in blood as well as cerebrospinal fluid in conditions characterized by neurodegeneration[7,52]. The current preliminary results require replication and further hypothesis-driven analyses but suggest that measures of immunoreactivity could be useful for understanding accelerated aging processes in the central nervous system.

Current evidence from cross-sectional studies suggests that chronic pain may be related to accelerated aging processes in the brain, and this association may be stronger in women. However, it is important to acknowledge that recent work suggests that brain-predicted age may be more reflective of an invariant trait than a modifiable acceleration of brain aging processes[56]. Vidal-Pineiro et al., 2021 found that cross-sectional brainAGE values were not significantly related to changes in brainAGE measured over time in a large cohort of healthy adults. Rather, brainAGE was associated with congenital factors (e.g., birth weight), which they assume to indicate an early, static influence on brain structure and health. While the most appropriate interpretation of brain-predicted age estimates is currently debated, this work from Vial-Pineiro, et al. highlights the importance of examining both cross-sectional and longitudinal data to understand the influence of chronic pain on brain aging.

This paper has a few limitations. First, we chose to use the validated algorithm, brainageR[10], to be consistent with previous papers on brain age in chronic pain [14,50]. However, other studies suggest that multimodal imaging data may provide a more accurate prediction of brain age[36,44]. Additionally, machine learning-based brain-predicted age estimations have a known age bias. While we and others who study brain-predicted age control for this through various statistical methods, recent work has highlighted that existing bias correction methods may not be sufficient[60]. Yet, there is no current consensus on an effective approach. This may limit the interpretability of our results.

Here, we found that the presence of UCPPS has the opposite relationship with brainAGE in males and females, and preliminary analyses that indicate brainAGE values are related to markers of inflammatory load in participants with UCPPS. These results demonstrate that future studies of brainAGE should specifically examine sex differences and further investigations are warranted to understand the direction of the relationship between inflammatory load and brainAGE in people with chronic pain.

Data availability statement:

Original datasets were not generated over the course of this research. This is a secondary analysis of a dataset collected by the MAPP research network. MAPP anticipates the original data will be publicly available in 2024. Additional details are provided in the cover letter.