Reliabilities of intra-individual mean and intra-individual variability of self-reported pain derived from ecological momentary assessments: Results from the Einstein Aging Study

Abstract

An individual’s pain experiences vary substantially over time. Though variability in pain may be an important metric which usually predicts health consequences, research on the measurement of pain variability estimates is lacking among older adults. We aimed to examine the reliabilities of both intra-individual mean (IIM) and intra-individual variability (IIV) of pain assessed using ecological momentary assessments (EMA) among racially diverse, systematically recruited community dwelling cohort of older adults. Participants (N=311, age=70-91) completed a 14-day EMA protocol which included self-reports of pain intensity, pain interference with activities, and pain interference with concentration multiple times a day. Over a 2-week period, we found excellent reliabilities for both pain IIM (.99) and pain IIV (≥.90). We also found that we need 5 to 6 days to achieve good reliability (0.8) for pain IIV, suggesting that a shorter protocol may be used to reduce participants’ burden among the current sample, although caution is required when using this result to determine EMA study designs among different samples. Future studies are required to examine the associations of various EMA pain metrics with different health outcomes among older adults to facilitate the detection of underlying mechanisms linking pain to health as a prelude to interventions.

Introduction

More than 50% of adults 65 and older in the US experience bothersome pain²⁴ or meet a case definition for chronic pain¹⁹. Pain among older adult populations has been associated with decreased physical and cognitive function, dementia, brain changes, and mortality^4,6,7,38. However, older adults are often underrepresented in clinical trials²³ or in pain research using novel study designs²⁸. The aim of the current study is to examine the reliability of pain indices, assessed using ecological momentary assessments (EMA), among community dwelling older adults.

Traditional pain measures rely on retrospective assessments of average levels of pain intensity or interference, usually with recall intervals spanning periods of 1week to 3 months prior to assessment^2,3,19. These recall based measures may be biased such that they are influenced by the most intense or recent pain experiences³⁰. In addition, subjective pain experiences vary substantially over moments, hours, or days, even among individuals with chronic pain^9,15,21,31. As a consequence, focusing on pain levels over various recall intervals may not accurately reflect average pain and may miss important aspects of pain including metrics of pain variability.

Using EMA methods, pain experiences may be captured with increased accuracy and with characterization of within-person change over time. In EMA studies, participants are asked to report their pain experiences in real time multiple times a day in their natural environments³⁰. In addition to the increased ecological validity and reduction in recall bias, this method provides the opportunity to construct additional measures that reflect the dynamic patterns of everyday pain experience: intra-individual mean (IIM), and intra-individual variability (IIV) of pain^18,30.

Pain variability (IIV) indicates the magnitude of pain fluctuation within individuals. Prior studies have shown that pain variability over the short-term (days, weeks) is a meaningful individual trait and is associated with diverse health outcomes^21,28,29,37. However, this has not been assessed among community dwelling older adults. Further, little is known regarding the reliability of IIM or IIV of pain among community dwelling older adults as prior research has focused on clinical population with pain disorders such as fibromyalgia or arthritis^1,10,13 or simulation studies^5,33.

The Einstein Aging Study (EAS) provides the opportunity to evaluate the reliabilities of mean pain and pain variability in a community-based, racially diverse sample of adults over age 70. Pain was assessed multiple times a day for 14 days via EMA. We examine the mean and variability in three domains of pain – pain intensity, pain interference with activities, and pain interference with concentration. Although pain intensity has been the most common pain domain examined in pain research⁸, it is important to use various aspects of pain to better understand its subjective nature over time.

Methods

The current study is based on data from the EAS, a longitudinal study with annual collection of standard neurological and neuropsychological assessments and an EMA component. Data for the present study derived from individuals’ first EMA assessments collected between May 2017 and February 2020.

Participants

Participants included 311 older adults (mean age=77.5, range=70 to 91) recruited using systematic sampling of New York City Registered Voter Lists for Bronx, New York¹². Eligible participants were aged over 70, ambulatory, fluent in English, and residents of Bronx County, NY. Exclusion criteria included significant hearing or vision loss, current substance abuse, severe psychiatric symptoms that may interfere with testing, chronic medicinal use of opioids or glucocorticoids, treatment for cancer within the last 12 months, and a diagnosis of dementia at enrollment. The EAS cohort is a systematically recruited sample which is broadly representative of the community from which it was derived, and which has characteristics similar to the US Census demographic data for the Bronx at the time of recruitment¹¹.

Procedure

The Institutional Review Board of the Albert Einstein College of Medicine approved the study protocol. Following a phone-screening assessment, eligible participants completed the consent process and were invited to attend a visit to the research clinic. During this visit, they completed an extensive neuropsychological battery based on the Uniform Data Set³⁵ and questionnaires to assess demographic, medical and family histories and psychosocial characteristics¹². Written informed consent was obtained in this initial clinic visit. Participants returned for a second clinic visit when they were given a study smartphone and were trained on the use of EMA brain games study surveys. The day following the clinic visit, participants began a 2-day run-in practice phase, followed by a 14-day EMA protocol. The protocol included a self-initiated wake-up assessment, a self-initiated end-of-day assessment, and four quasi-randomly prompted assessments each day. An auditory beep was given to signal participants to complete prompted assessments, which were spaced at approximately 3.5 hour intervals. Participants received up to $160 as compensation based on completion of the tasks. It should be noted that the run-in data were not used in the current analyses.

Measures of EMA Pain

Pain intensity was assessed by asking “Right now, what is your level of pain?” using a visual analogue scale slider from no pain (0) to worst pain (100). Participants were asked to report their levels of pain intensity 6 times throughout each day (a wake-up assessment, four beeped assessments, and an end-of-day assessment) for 14 days, resulting in 84 total assessments. Pain interference with activities was assessed by asking “Since the last survey, how much did pain interfere with your daily activities?” from not at all (0) to very much (100). Pain interference with concentration was assessed by asking “Since the last survey, how much did pain interfere with your ability to concentrate?” from not at all (0) to very much (100). Because the interference measures assessed pain since last assessment, they were not included in the wake-up assessment and only included in four beeped assessments and the end-of-day assessment each day, for a total of 5 assessments per day resulting in 70 total assessments over 14 days.

Measures of Chronic Pain

Total pain index, which addressed pain location, frequency, severity and duration in the past 3 months, was used to indicate chronic pain. Data for the total pain index was collected during the phone screening assessment. For each of 8 body areas including head, face, neck and shoulder, back, arms and hands, legs and feet, chest, and abdomen and pelvis, participants were asked “in the past 3 months, how often did you have pain in the [insert 1 of 8 body areas]”. Response options included ‘none of the time’, ‘a slight bit of time’, ‘some of the time’, ‘most of the time’, and ‘all of the time’. Participants were then asked for each body area with pain to rate their worst pain over the last three months on a scale of 0 to 10. Participants were classified as having chronic pain if pain in at least one location fulfilled the following criteria: 1) Pain in the past three months ‘some’, ‘most’ or ‘all of the time’, 2) Pain of moderate or severe intensity (≥4/10)¹⁹.

Analytic plans

Stationarity check.

Measures of pain indices can be influenced by time trends in the EMA data. For example, if pain intensity increases or decreases over the period of intensive measurement within an individual, that trend could bias the estimate of pain variability for that individual relative to someone whose pain does not show a trend. Thus, the stationarity assumption (i.e., the mean and variance of each individual’s pain rating do not change over the entire period of EMA) needs to be met to investigate the measurement of pain indices derived from EMA³⁴. To test this assumption of stationarity, we conducted the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test¹⁶ separately for each person. The data were stationary for 68% of participants for pain intensity, 79% for pain interference with activities, and 83% for pain interference with concentration. The percentage was somewhat low for pain intensity, and we opted to remove systematic time-dependent structure before studying IIV ^25,33 for all pain domains. After removing linear time trends, stationarity was achieved in more than 96% of participants in all pain domains. Mean pain and pain variability were quantified with detrended raw values as described below.

Calculating IIM and IIV for each pain domain.

For each participant, mean pain was calculated as the intra-individual arithmetic mean of all pain ratings over 14 days, separately for each pain domain (intensity, interference with activities, interference with concentration). Pain variability was calculated as the intra-individual standard deviation (iSD) of pain ratings over 14 days for each domain. This measure is meant to capture fluctuations in pain. In addition, iSD is the most frequently used variability metric and has proved to be useful in pain research as well as in other areas including cognition, emotion, and personality^21,32.

Calculating intra-class correlation (ICC): the reliability of single assessment for each pain domain.

Reliability can be defined as the square of the correlation between observed scores and true scores. Here, the reliability of single assessment (i.e. ICC) also reflects the proportion of the total variance in the observed scores which is not attributable to measurement error. Because pain is dynamic in nature, calculating ICC based on the entire 14 days’ EMA protocol (e.g., across all 84 assessments for pain intensity) would underestimate the actual ICC. Thus we calculated ICC for each day then the average of all the 14 daily ICCs was used as a final estimate of ICCs. The ICC calculated for each day was based on an absolute agreement with other assessments that day, using a 2-way mixed effects model¹⁴ using R package psych²⁶. ICC was calculated for each pain domain separately.

Calculating reliability of IIM and IIV.

To calculate the reliability of intra-individual mean pain and variability, the approach proposed by Wang and Grimm³³ was applied and summarized as below. Briefly, we first estimated between-person variance, expected intra-individual variance and error variance based on our sample of 311 participants, then plugged those estimates into the formulas to calculate reliabilities of IIM and IIV while letting the number of assessments (denoted as T) vary to get the reliability for a given T. Therefore, the reliability only depends on the number of assessments but does not depend on the number of participants, although the latter (i.e., sample size) may affect the precision of estimation.

When there is no systematic trend in pain, the pain rating for subject i (i=1,2…N) at session t (t=1,2,..T_i) can be modeled as:

$$x_{it}={\mu }_i+v_{it}+e_{it}$$

where x_it is the pain rating for subject i at session t, μ_i is the true trait/level of individual i, v_it is the unsystematic (in relation to time) change at session t due to true intra-individual variability of individual i, corresponding to net IIV unstructured in relation to time, which is in contrast to time-structured IIV for systematic change (Ram and Gerstorf 2009), e_it is the measurement error of individual i at session t.

Additional assumptions are:

1. E(v_it) = 0: There is no systematic change over time, which can be met by the stationary requirement.

2. $Var(v_{it})={\sigma }_{vi}^2$: This is the true intra-individual variance for subject i and can vary among participants.

3. $e_{it}\sim N(0,{\sigma }_e^2)$: Measurement error follows normal distribution.

4. Cov(μ_i, e_it) = 0, Cov(v_it, e_it) = 0 : This usually can be met.

Further assumptions to calculate IIV reliability include:

1. $v_{it}\sim N(0,{\sigma }_{vi}^2)$: v_it follows normal distribution.

2. ${\sigma }_{vi}^2\sim Gamma(a,b)$: ${\sigma }_{vi}^2$ follows a Gamma distribution.

Reliability of IIM and IIV can be computed by the two formulas below.

$$\begin{gathered} \text{Pain IIM reliability}=\frac{1}{1+\frac{1}{T}\left(\frac{E\left({\sigma }_{vi}^2\right)+{\sigma }_e^2}{{\sigma }_{\mu }^2}\right)} \\ \text{Pain IIV reliability}=\frac{{\left(\sqrt{\frac{2}{T-1}}\frac{\Gamma \left(\frac{T}{2}\right)}{\Gamma \left(\frac{T-1}{2}\right)}\right)}^2{\left[E\left(\sqrt{{\sigma }_{vi}^2+{\sigma }_e^2}\sqrt{{\sigma }_{vi}^2}\right)-E\left(\sqrt{{\sigma }_{vi}^2+{\sigma }_e^2}\right)E\left(\sqrt{{\sigma }_{vi}^2}\right)\right]}^2}{\left[E({\sigma }_{vi}^2)+{\sigma }_e^2-{\left(\sqrt{\frac{2}{T-1}}\frac{\Gamma \left(\frac{T}{2}\right)}{\Gamma \left(\frac{T-1}{2}\right)}\right)}^2\right]{\left[E\left(\sqrt{{\sigma }_{vi}^2+{\sigma }_e^2}\right)\right]}^2\left(E({\sigma }_{vi}^2)-{\left(E\left(\sqrt{{\sigma }_{vi}^2}\right)\right)}^2\right)}, \end{gathered}$$

where T is the number of assessments (T varies to calculate the reliability for different number of assessments), ${\sigma }_{\mu }^2$ indicates inter-individual variance of true trait scores, $E({\sigma }_{vi}^2)$ is expected or average intra-individual variance, ${\sigma }_e^2$ is error variance, and Γ stands for Gamma function. Readers should refer to Wang and Grimm (2012) for the derivation of these formulas and Appendix 1 for the calculation of each component in the equations.

Sensitivity analysis.

We conducted three sets of additional analyses. First, for our relatively healthy older adult sample, reliabilities may be fundamentally high because they reported overall low and stable levels of pain over time. Thus we conducted additional reliability analyses on a subsample of individuals with chronic pain as these individuals may often experience fluctuating pain and show lower reliabilities on all pain indices. Second, we conducted reliability analyses excluding individuals who consistently reported low levels of EMA pain (i.e., very little variability). Third, we examined whether a briefer protocol may also produce acceptable levels of reliability. We calculated number of assessments required to achieve pre-specified levels of reliability (reliability=0.7, 0.8, and 0.9).

Results

Sample characteristics and descriptive statistics

Table 1 shows basic demographic characteristics and scores for each EMA pain domain aggregated across the 14-day assessment period. Mean age was 77.5 (SD=4.86) and women made up 67.2% of the sample. The sample was racially diverse, with 45.3% self-identified as non-Hispanic White, 40.2% as non-Hispanic Black, and 14.5% as Other. Mean years of education were 15 years; 34.1% were married, 1.3% separated, 26% widowed, 21.2% divorced, and 17.4% never married.

Table 1.

Descriptive information (Mean (SD) or %)

		All (N=311)	Male (N=102)	Female (N=209)	p
Demographic	Age	77.5 (4.9)	76.9(4.7)	77.8(4.9)	0.108
	% Female	67.2%	-	-	-
	Years of education	15.0(3.6)	15.1(3.8)	14.9(3.4)	0.676
	% Marital Status				<0.001
	Married	34.1%	65.7%	18.7%
	Separated	1.3%	0.0%	1.9%
	Widowed	26.0%	8.8%	34.4%
	Divorced	21.2%	10.8%	26.3%
	Never married	17.4%	14.7%	18.7%
	% Race/ethnicity				<0.001
	Non-Hispanic White	45.3%	60.8%	37.8%
	Non-Hispanic Black	40.2%	27.5%	46.4%
	Other	14.5%	11.8%	15.8%
IIM for EMA pain measures	Pain intensity	20.5(20.9)	19.3(21.0)	21.1(20.9)	0.310
	Pain interference with activities	19.5(21.0)	18.8(21.4)	19.9(20.9)	0.470
	Pain interference with concentration	14.7(16.3)	13.8(15.1)	15.0(16.8)	0.605
IIV for EMA pain measures	Pain intensity	10.8 (8.0)	9.7 (7.9)	11.4 (8.0)	0.051
	Pain interference with activities	9.8 (8.2)	8.9 (7.9)	10.3(8.3)	0.107
	Pain interference with concentration	8.3 (7.4)	7.9 (6.7)	8.5 (7.7)	0.613
Recall-based pain measure	Total Pain Index	30.8 (36.5)	21.2(25.1)	35.5(40.2)	0.002

Note. EMA: Ecological momentary assessment; IIM: Intra-individual mean; IIV: Intra-individual variability (i.e., iSD). p values were calculated using Wilcoxon rank sum tests for continuous variables and Pearson’s chi-squared tests or Fisher’s exact tests for categorical variables.

Intra-individual mean and intra-individual variability (i.e., iSD) of each pain domain were calculated. The average of IIM was 20.5 (range=0 to 79.2) for pain intensity, 19.5 (range=0 to 96.2) for pain interference with activities, and 14.7 (range=0 to 72.3) for pain interference with concentration The average of IIV was 10.8 (range=0 to 40.1) for pain intensity, 9.8 (range=0 to 39.6) for pain interference with activities, and 8.3 (range=0 to 40.9) for pain interference with concentration. Figure 1 shows box plots for median, upper and lower extreme IIM and IIV of each pain measure with individual data points. The average of Total Pain Index was 30.8 (range=0 to 181). Total Pain Index showed a correlation of 0.59 with pain intensity, a correlation of 0.61 with pain interference with activities, a correlation of 0.45 with pain interference with concentration, suggesting moderate to strong correlations. Males, compared to females, were more likely to be married and non-Hispanic Whites, and report lower levels of Total Pain Index. No significant gender differences were found in the IIM or IIV of the three EMA pain measures.

Figure 1.

Box plots showing median, upper and lower extreme for each pain measure.

Reliability analysis

The entire sample (N=311) were used to conduct reliability analyses. The reliabilities for a single assessment (i.e., ICCs) were 0.80 for pain intensity, 0.83 for pain interference with activities, and 0.79 for pain interference with concentration. Table 2 shows the reliabilities for pain IIM and IIV that our study design can achieve (84 assessments for pain intensity, 70 assessments for pain interference with activities and pain interference with concentration). For mean pain, reliabilities were >0.99 for all pain domains. For pain variability, reliabilities were 0.892 for pain intensity, 0.903 for pain interference with activities, and 0.884 for pain interference with concentration.

Table 2.

Pain reliability over 14 days

Pain domain	Mean pain (IIM)	Pain variability (IIV)
Pain intensity	0.996	0.892
Pain interference with activities	0.996	0.903
Pain interference with concentration	0.995	0.884

Note. Number of assessments for 14 days is 84 for pain intensity, and 70 for pain interference with activities and concentration (N=311).

We conducted several sensitivity analyses. First, we calculated reliabilities on the subsample of individuals having chronic pain (N=170). Reliabilities of pain IIM ranged from .995 to .996 and reliabilities for pain IIV ranged from .890 to .893. The reliabilities were very similar to those in the whole sample. This should not be surprising because one of the factors that can affect the reliabilities of IIM and IIV is the ratio of the expected intraindividual variance and the between-person variance ($\frac{E({\sigma }_v^2)}{{\sigma }_u^2}$)³⁴. In the subsample with chronic pain, both the expected intraindividual variance $(E({\sigma }_v^2))$ and the between person variance (${\sigma }_u^2$) were estimated to be larger than those in the whole sample. Thus, the ratio only had a slight increase in the chronic pain subsample and the reliabilities results did not change much.

Second, we conducted reliability analyses excluding individuals who consistently reported low levels of EMA pain (i.e., <5 out of 100 on each pain scale). Among 311 participants, 15, 31, and 30 participants consistently reported low levels of pain for pain intensity, pain interference with activities, and pain interference with concentration respectively. Reliabilities of pain IIM were 0.995 and reliabilities for pain IIV ranged from .893 to .901, which were similar to previous results.

Third, we estimated the number of assessments needed to achieve acceptable, good, and excellent reliability scores (reliabilities of 0.7, 0.8, and 0.9 respectively). For mean pain, it was easy to achieve good reliability because ICCs for single assessment approached or exceeded 0.8. The average of 3 to 4 assessments produced reliability values that exceed 0.9 across all pain domains (Table 3). For pain variability, 13 (for pain interference with activities) to 17 (for pain interference with concentration) assessments produced reliability values that exceed 0.7, and 23 (for pain interference with activities) to 31 (for pain interference with concentration) assessments produced reliability values that exceed 0.8. To reach reliability values that exceed 0.9, 102 (for pain intensity), 67 (for pain interference with activities), and 91 (for pain interference with concentration) assessments were required (Table 4).

Table 3.

Number of assessments required to achieve a certain reliability level for mean pain

	Number of assessments required to reach reliability of 0.8	Number of assessments required to reach reliability of 0.9
Pain intensity	2	4
Pain interference with activities	1	3
Pain interference with concentration	2	3

Table 4.

Number of assessments required to achieve a certain reliability level for pain variability

	Number of assessments required to reach reliability of 0.7	Number of assessments required to reach reliability of 0.8	Number of assessments required to reach reliability of 0.9
Pain intensity	16	29	102
Pain interference with activities	13	23	67
Pain interference with concentration	17	31	91

Lastly, Figure 2 shows the reliabilities of IIM and IIV for different numbers of assessments for each pain domain. With the number of assessments, the reliability increased nonlinearly. IIM was more reliable than IIV, reached reliability of 0.9 after just 3 to 4 assessments, and asymptotically approached perfect reliability in all pain domains.

Figure 2.

Reliabilities for mean pain and pain variability over the number of assessments.

Discussion

We examined the reliabilities of mean pain and pain variability in a systematically recruited, community-based, racially diverse older adult cohort. Pain was measured via EMA multiple times per day for 14 days. We measured three pain domains – pain intensity, pain interference with activities, and pain interference with concentration – to calculate mean pain and pain variability. Reliabilities for a single assessment (ICCs) in all pain domains were around 0.8. Over a 2-week period, we found excellent reliabilities for both mean pain (>.99) and pain variability (>.88). The high reliability of these measures suggest the utility of this approach for future studies investigating their relation to various health outcomes. Moreover, this study provides added value by expanding the domains of pain evaluated and demonstrating that measures of pain interference with activities and pain interference with concentration can be assessed reliably and used to predict health outcomes in future research.

Establishing reliability is a necessary step to guarantee rigor and reproducibility of research²⁷. There has been little research on the reliability of pain variability although more and more studies have used pain variability as an important index to investigate health outcomes^21,36. Without investigating the reliability of pain metrics, the statistical significance of mean and variability would come into question. Moreover, if both mean and variability are used in the same analytic model and compete with each other, the less reliable measure would be disadvantaged³³. Some studies might be able to find statistically significant associations between variability and variables of interest, but a substantial number of assessments (i.e., reliable measures) would still be required to assess the magnitude of the effect with precision⁵. Otherwise, the magnitude of the effect of variability would be under- or over-estimated.

Results from the current study indicated that mean pain was relatively reliable and asymptotically approaching perfect reliability in all pain domains over a 2-week period. We also found that reliability of mean pain approached excellent reliability (0.9) with the average of only 4 assessments for all pain domains. This result is in line with prior literature with chronic pain patients^1,10. However, while most prior studies were interested in reliability of mean pain among individuals with chronic pain^18,28, the present study contributes to the literature by investigating pain reliabilities among community dwelling older adults, whose pain experiences were relatively less examined.

The reliability of pain variability was lower than the reliability of mean pain with any fixed number of assessments but increased with more measurement occasions. Reliability of pain variability approached excellent reliability (0.9) with the average of 67 to 102 assessments depending on the pain domain. We also examined whether a briefer protocol may produce good (0.8) or acceptable (0.7) levels of reliability. To achieve the reliability of 0.8, 23 to 31 assessments were required depending upon the pain domain. To achieve the reliability of 0.7, 13 to 17 assessments were required. Because pain intensity was measured 6 times a day and pain interference was measured 5 times a day, we need 5 to 6 days to achieve the reliability of 0.8 and 3 to 4 days to achieve the reliability of 0.7, suggesting that a shorter protocol or fewer assessments per day may be used to reduce participants’ burden. The current study can also provide implications in designing future EMA study to investigate pain in this population. However, it should be noted that our results cannot be used as a guideline for determining the number of assessments for an EMA study among different samples. For example, conditions affecting the reliability of variability, such as mean IIV or variability in IIV⁵, may differ between clinical samples and community dwelling older adults. Given the importance of IIV metric for health consequences²¹, future research would be required to examine IIV reliability across diverse populations.

The current study has some limitations to be discussed and implications for future research, clinical, and intervention studies. First, although we investigated pain reliability among community dwelling older adults and older adults who met chronic pain definition, reliabilities of pain metrics and related components may differ across subgroups of the population. For example, some subgroups of older adults may show distinct within-day, diurnal patterns that may influence how to detrend raw data and calculate ICC. In the current study, we did not take diurnal trend into account in our relatively healthy older adult cohort. Future studies should identify groups that have different profiles of within- and between-day pain and investigate how these groups differ in pain reliability.

Second, studies should examine the associations of specific pain domains with emotional, social, cognitive, and physical health consequences²². In this paper, we found evidence that both pain intensity and pain interference can be reliably measured among older adults, providing a firm foundation for future work. Although pain intensity has been the primary focus of prior research and clinical settings, we need multifaceted pain measures to capture full breath of individuals’ pain experience: pain interference with activities and pain interference with concentration. It is possible that various pain domains are differently associated with various health outcomes.

Third, the use of EMA pain measures could enhance observational clinical studies and clinical trials. Traditional, recall-based one-time pain measures may not capture intervention effects properly because of recall biases³⁰ (e.g., participants with more fluctuation in pain experiences may report higher levels of pain). Evaluating pre-post intervention improvements in both mean pain and pain variability will allow us to capture more precise pain levels and pain magnitude. In addition, investigating pain variability will provide additional information regarding therapeutic mechanisms of intervention to reduce pain and enhance health outcomes²¹. With the widespread us of pain diaries in clinical practice and research, the nearly ubiquitous use of smartphones, and the availability of downloadable pain diary apps, it is highly feasible to obtain repeated measures of pain intensity and interference and to incorporate these methods into research and practice. Pain variability and consistency of pain response has emerged as an outcome measure in recent clinical trials^17,20, emphasizing the need for its deeper examination in a multitude of studies, and specifically in older adults, where significant lack of studies persists.

Fourth, the current variability metric (iSD) does not have temporal sensitivity and does not capture the frequency of fluctuations, and other variability metrics such as autocorrelation and Mean square of successive difference (MSSD) may address some of the limitations of iSD²¹. However, the usefulness and interpretation of these other variability metrics in pain research has not been established, which warrants further investigation.

Conclusion

Pain is prevalent among older adults and fluctuates over moments, hours, and days. Pain variability, derived from EMA, is emerging as an important metric that may be associated with health consequences. We showed that both mean pain and pain variability, derived from EMA, could be reliably measured among community dwelling older adults over a 2-week period. Given that there has been no prior research on evaluating the reliability of pain variability, this study is an important first step in conducting rigorous research using EMA pain variability measures among older adults. We also found that a shorter protocol (5 to 6 days) was required to achieve good reliabilities for pain IIV, although caution is required when using this result to determine EMA protocols among different samples (e.g., clinical populations). Future work should consider utilizing EMA pain questionnaires to capture pain variability, consider pain variability in study designs (e.g., to determine number of assessments) both as a predictor of other health outcomes and as a measure of treatment benefits, and examine IIV reliability across diverse populations.

We also investigated the reliabilities of mean and variability for different pain domains – pain intensity, pain interference with activities, and pain interference with concentration—to better capture individual’s subjective pain experiences, and found that they could be reliably assessed. Future research should investigate more fine grained and unique associations between each pain domain and different health domains such as physical, social, and cognitive functioning. By considering how different facets of pain experience influence individuals’ short- and long-term functioning, we may be able to better understand underlying mechanisms of pain and inform clinical implications.

Highlights.

• Using EMA, we can capture within-person mean (IIM) and variability (IIV) of pain.

• Assessing reliability of pain IIV is important but not well examined in prior research.

• High reliabilities of pain IIM and IIV could be achieved in community-dwelling older adults.

• Future work should consider utilizing EMA to capture pain IIV and link IIV to health outcomes.

Perspective:

Mean levels and variability in pain intensity, pain interference with activities, and pain interference with concentration can be reliably measured and linked with various health outcomes in older adults. Future studies including these pain metrics will assess the natural history, the consequences, and effects of intervention of pain.