Abstract
Background
Understanding how pain manifests in people living with dementia can facilitate pain management but evidence from large real-world studies is limited. Through digitally compiled large number of pain assessments, we explore how behavioural indicators manifest across all widely used domains of pain assessment.
Methods
This is a population based observational study with data from aged care homes across Australia, New Zealand and the United Kingdom. Pain assessments were conducted over a period of 5.3 years in 89 230 residents living with advanced dementia. The primary outcome was to determine pain indicator occurrence across pain levels and explore their predictive contributions on pain intensity. This included pain outcomes related to assessment at rest and movement analysed using binary logistic generalized estimating equation.
Results
A total of 2 125 022 pain assessments were conducted on 89 230 residents, and just over half (i.e. 52.8%) exhibited indicators consistent with the presence of pain. Indicators suggestive of higher pain intensity were more likely when assessed post-movement (OR = 2.5, CI 2.4–2.7, χ2 = 1129.6, P < .001). All pain indicators across various assessment domains were predictive of pain, confirming the value of a multidimensional pain assessment. Higher pain levels were manifested through horizontal mouth stretch (OR = 4.6), noisy sounds (OR = 9.0), restlessness (OR = 18.7), distress (OR = 27.9) resisting care (OR = 13.5) and rapid breathing (OR = 20.1).
Discussion
Our findings suggest that pain indicators occur across multiple pain assessment domains and provide insights into how pain manifests across different pain levels. We also confirm the importance of movement-based pain assessment during which pain levels are higher compared to rest.
Keywords: pain experience, dementia, pain assessment, digital health, pain levels, older people
Key points
We demonstrate the presence of pain indicators across all widely used domains of pain behaviour in people living with dementia.
There is an increase in the presence of pain indicators across various domains with an increase in pain intensity.
Pain levels are higher when assessed during movement, hence supporting movement-based assessment of pain.
There are likely distinct phenotypes of pain expression in people living with dementia.
Background
Identification of pain in people living with moderate to advanced dementia is difficult due to diminished ability to self-report pain [1, 2]. Pain often goes undetected and therefore undertreated, despite availability of various pain assessment scales which are either underutilised or not implemented [1, 2]. This can trigger or exacerbate neuropsychiatric symptoms and distress and these symptoms may be misinterpreted as a manifestation of dementia without considering the presence of pain [3–6].
Pharmacologic management of pain in this population is also challenging with evidence of over [7, 8] or under [9–11] use of analgesics and sub-optimal use of non-pharmacological interventions [6]. These gaps point towards the need for a more individualised approach to both pain identification and a better understanding of individual variations related to how pain manifests in people living with dementia. In order to achieve this, pain assessment process should identify ‘pain phenotypes’ which can enable health care providers design individual interventions [6]. Clinical ‘phenotyping’ of pain, in addition to describing its presence and intensity, also implies characterisation of how pain manifests in terms of non-verbal expressions, behavioural symptoms and functional and emotional impact in people living with dementia [6, 12, 13].
There are currently limited available data to facilitate clinical phenotyping of pain in people living with dementia. In those with moderate to advanced dementia and unable to self-report, pain should be assessed multidimensionally [14, 15]. In this regard, the American Geriatrics Society (AGS) recommended dimensions of pain assessment include facial expressions, vocalisations, body movements as well as changes in mental status, activity patterns and interpersonal interactions such as aggressiveness and resisting care [15]. While pain behaviours such as facial expressions and non-verbal vocalisations have been individually studied [16–18], there is currently a lack of real-world data from large cohorts that have combined multidimensional pain assessment including behavioural symptoms, body movements and other activities that directly affect people living with dementia during display of indicators suggestive of pain. It is worth noting that the AGS recommendations have been developed based on clinical experience and panel consensus rather than scientific evidence [15]. Additionally, pain evoked by movement has drawn significant attention in the way pain is understood considering interaction of psychological, sensory and motor factors and that movement and pain affect each other reciprocally [19, 20]. In this regard, movement-based assessments have been proposed to better understand interference with a person’s function [19]. However, more evidence is needed to support this in people living with dementia and hence better understand manifestation of pain in this vulnerable population group.
In this data-driven population-based study, we aim to address these gaps in the existing evidence. Specifically, by exploring a large database of pain assessments conducted in clinical practice we aim to describe how pain manifests across all widely used domains of pain assessment and explore the relationship between pain intensity and movement-based assessment of pain in people living with dementia unable to self-report pain.
Methods
In this data-driven population-based observational retrospective study, we reviewed pain assessments conducted over 5.3 years from the period 15 February 2018 to 14 June 2023 in Australia, New Zealand and the UK. These assessments were conducted in people living with dementia unable to self-report pain by clinical staff who used a digital pain assessment system, namely the PainChek system which also integrates the PainChek instrument that has been previously validated and its pain categories calibrated against the Abbey Pain Scale [21–23]. Additionally, the PainChek instrument is a regulatory cleared medical device in all jurisdictions where data were collected and has recently been granted Food and Drugs Administration De Novo status (6 October 2025).
The digital pain assessment system
The PainChek pain assessment system is a digital solution comprised of three key components: the PainChek pain assessment tool (available as point-of-care App), education for users and PainChek® analytics [22]. PainChek instrument is a hybrid point-of-care mobile application which uses an AI enabled algorithm trained to detect 9 facial action unit (AU) codes indicative of pain (i.e. muscle features producing facial expressions) in real time. In addition to the face domain analysis, the PainChek instrument has another five domains each containing specific pain indicators, i.e. Voice (nine indicators), Movement (seven indicators), Behaviour (seven indicators), Activity (four indicators) and Body (six indicators). Each of these pain assessment dimensions includes distinctive pain indicators such as confusion and distress indicators included in the Behaviour domain. A description of each indicator across PainChek instrument’s domains has been published elsewhere [22]. These domains are based around those pain behaviour categories identified by the American Geriatric Society (AGS) as being associated with pain in people with cognitive impairment [15]. These additional five domains are scored by user through observation and/or monitoring the subject. In total the PainChek instrument contains 42 pain indicators with each having a binary option of scoring 0 (i.e. absent) or 1 (i.e. present) [21–23]. A definition of each indicator is available within the app for the user to refer to. The app automatically calculates a total pain score and assigns a pain intensity category: no pain (0–6), mild pain (7–11), moderate pain (12–15) and severe pain (16–42). More information about the conceptualisation, development and validation of the tool has been published elsewhere [21–23].
Data sourcing
We explored a digitally compiled database containing a total of 2 125 022 pain assessments. Staff involved in conducting pain assessments received a 2-h training on challenges and triggers of pain assessment in people living with dementia as well as on the use of the PainChek pain assessment system. This training also has components aimed at ensuring consistency of scoring between assessors. Staff were instructed to not use the PainChek instrument in people with dementia who are able to self-report, considering that self-report is the current gold standard [24]. PainChek Ltd granted permission to access the deidentified dataset used in this study. The data obtained from PainChek Ltd comes from pain assessments conducted at point-of-care and then synchronised into web administration portal (PainChek portal). This is deidentified data for the purposes of research and analysis according to conditions in the service agreement with each aged care provider. Data accessed and analysed included logs of pain assessments conducted chronologically (which includes pain presence, intensity and scoring across each domain), and deidentified demographics to allow linking of repeat and follow-up assessment analysis.
The study received ethics approval from the Human Research Ethics Committee of X University (HR10/2014) and Ethics Committee of the Faculty of Medicine, University of X (10941/2023).
Data analysis
Description of pain presence and occurrence of pain indicators across pain level categories and domains
To analyse the data, we used IBM SPSS version 27. We explored key demographic details with gender described in frequency (f) and percent (%). Pain score (non-normal distribution Kolmogorov–Smirnov P < .001), age, number of assessments, were described as mean (M), standard deviation (SD), 95% confidence interval (CI) of the M, median (Md) and inter quartile range (IQR) (25th–75th percentile) for total and for the presence and absence of each indicator.
We analysed different pain categories at rest and post-movement. These categories were organised in Group 1 (i.e. low = no or mild pain and high = moderate or severe pain) and Group 2 (i.e. no pain, mild, moderate and severe). The dichotomous grouping (i.e. Group 1) was done considering that moderate to severe pain usually requires intervention whereas mild pain may only necessitate observation or non-pharmacological intervention [25]. Hence the decision was to also investigate these categories separately. We further analysed presence of each PainChek pain indicator across different pain domains. These are described using frequency (f) and percentage (%).
Analysing the predictive value of pain indicators on presence of pain and how pain indicators manifest across pain level categories
The generalized estimating equation (GEE) model, a population-level approach, was used as it is a robust and flexible method for analysing unbalanced data in longitudinal studies and is known for its ability to handle missing data, varying observation counts, complex data structures and ability to incorporate covariates [25, 26]. A binary logistic GEE was used to examine pain outcome (low/high) with at rest versus post-movement treated as a fixed effect. Additional models for each domain investigated the pain outcomes defined above (low/high) and pain (no pain/pain) with domain indicators treated as fixed effects. Negative binomial (log link) GEE examined the predictive contributions of each domain’s indicators (factors) on pain score. For all models, confounders age and sex were included as fixed effects, and patients treated as repeated. Model fixed effect Wald χ2 and P-values, beta coefficient (β) and standard error (SE) for age, odds ratio (Exp (β)) for sex and indicators, and 95% CIs are reported.
Results
A total of 2 125 022 pain assessments were conducted on 89 230 residents living with dementia in Australia (1 981 745; 93.3%), New Zealand (8 470; 0.4%) and the UK (134 807, 6.3%). These pain assessments were conducted by a total of 14 987 trained assessors across multiple clinical sites which were part of 434 different aged care providers. Over the 5.3 years period of time, residents received between 1 and 2932 pain assessments (M = 23.8; 95 CI 23.3–24.4; SD = 81.3; Md = 3; IQR 1–16) with most assessments being conducted while residents were at rest (1 700 937 80.0%). Of the 89,230 residents who had pain assessments conducted, a total of 47 158 (i.e. 52.8%) indicated the presence of pain during this period (i.e. PainChek pain score > 6). Pain categories and pain scores are described for the total sample in Table 1 with the majority of assessments reporting no pain (88.7%, pain score ≤ 6). The high versus low pain GEE model results indicated that patients had a significantly higher risk of exhibiting indicators suggestive of a high pain level post-movement (OR = 2.5 CI 2.4–2.7, χ2 = 1129.6 P < .001) compared to when assessed at rest, with no significant effect for sex (χ2 = 2.2 P = .139) or age (χ2 = 2.7 P = .101).
Table 1.
Description of pain categories and pain scores for assessments completed at rest or post- movement.
| Resident activity | |||||
|---|---|---|---|---|---|
| Pain group | Pain level | n | Total 2 125 022 | At rest 1 700 937 | During movement 424 085 |
| Group 1 | Low pain | f (%) | 2 057 981 (96.8) | 1 659 424 (97.6) | 398 557 (94.0) |
| High pain | f (%) | 67 041 (3.2) | 41 513 (2.4) | 25 528 (6.0) | |
| Group 2 | No pain | f (%) | 1 885 305 (88.7) | 1 540 183 (90.5) | 345 122 (81.4) |
| Mild pain | f (%) | 172 676 (8.1) | 119 241 (7.0) | 53 435 (12.6) | |
| Moderate pain | f (%) | 47 843 (2.3) | 30 563 (1.8) | 17 280 (4.1) | |
| Severe pain | f (%) | 19 198 (0.9) | 10 950 (0.6) | 8248 (1.9) | |
| Pain score | M (SD) | 3.1 (3.2) | 2.8 (3.0) | 4.1 (3.8) | |
| Md (IQR) | 2.0 (1.0–4.0) | 2.0 (1.0–4.0) | 3.0 (1.0–5.0) | ||
Low pain comprises no and mild pain, high pain comprises moderate and severe pain categories.
Frequency of pain indicator occurrence across different pain level categories
The number of separate pain indicator was associated with a higher pain score. The categorisation of pain as present is related to increases in the number of observed pain indicators across or within domains. This is illustrated by the mean value increase throughout all pain indicators. Additionally, increases in the number of observed pain indicators across or within domains corresponded to increased pain level categories (i.e. from no vs mild vs moderate vs severe pain), as illustrated in Figure 1.
Figure 1.
Illustration of multidimensional pain indicators across different pain levels. AU4 = brow lowering; AU = cheek raising; AU7 = tightening of eyelids; AU9 = wrinkling of nose; AU10 = raising of upper lip; AU12 = pulling at corner lip; AU20 = horizonal mouth stretch; AU25 = parting lips; AU43-closing eyes.
Predictive value of pain indicators on categorising the presence of pain
According to our GEE model investigations using no pain versus pain across three pain categories (mild, moderate and severe), all pain indicators in the pain assessment instrument were predictive of pain-level categorisation as illustrated by significant association with increased odds of pain presence (mild, moderate and severe). Pain indicators across domains that most predicted the high pain-level category were: Horizontal mouth stretch (OR = 4.1) and brow lowering (OR = 4.1); noisy pain sounds (OR = 15.9) and groaning (OR = 11.7); guarding touch (OR = 20.7) and restlessness (OR = 17.1); distress (OR = 21.6); resisting care (OR = 10.9) and rapid breathing (OR = 23.0). On the other hand, pain indicators which were associated with the lowest odds of pain presence across individual domains were: Pulling at corner lip (OR = 1.6); screaming (OR = 5.9); wandering (OR = 2.4); confusion (OR = 3.0); prolonged resting (OR = 3.5) and painful injuries (OR = 2.2). Across domains the confounding effect of age and sex differed. For further details see Supplementary Table 1 in Appendix 1.
Manifestation of pain indicators contributing to a high pain-level category
The binary logistic GEE odds of reporting high pain (i.e. moderate or severe) are reported in Table 2. For all domains, the presence of individual indicators was associated with significantly increased odds of high pain. In terms of how residents were impacted by pain intensity across PainChek’s domains, our findings reported that presence of horizontal mouth stretch (i.e. AU20) and brow lowering (i.e. AU4) was associated with residents having a higher risk of experiencing high pain with OR = 4.6 and 4.4 respectively (i.e. indicators of the Face domain). Residents were more likely to vocalise noisy pain sounds (OR = 9.0), sighing (OR = 8.2), express loudtalk (OR = 6.7) and request help (OR = 6.1) during high pain respectively (i.e. indicators of the Voice domain). Furthermore, on the Movement domain, residents were at highest risk of experiencing high pain in presence of restlessness (OR = 18.7) and were more inclined to guard or protect body part (OR = 11.8) and experience freezing (OR = 10.8). Experiencing distress (OR = 27.9) followed by fear or dislike of being touched (OR = 12.0) were the top two indicators of pain in the Behaviour domain which were associated more with residents experiencing high pain levels in our sample. The resisting care (OR = 13.5) indicator in the Activity domain was most strongly associated high pain while rapid breathing (OR = 20.1) followed by appearing pale or flushed (OR = 12.4) were indicators most associated with high pain in the Body domain. Further details, are provided in Table 2.
Table 2.
Generalized estimating equation predictability of high pain scores when indicator item was present, separated for each domain.
| 95% CI exp (β) | 95% CI exp (β) | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Face indicators | Exp (β) | Lower | Upper | χ2 | P-value | Voice indicators | Exp (β) | Lower | Upper | χ2 | P-value |
| Age (years) | 1.0 | 1.0 | 1.0 | 3.3 | .068 | Age (years) | 1.0 | 1.0 | 1.0 | 13.4 | <.001 |
| Sex femalea | 0.9 | 0.8 | 1.0 | 1.6 | .198 | Sex femalea | 0.6 | 0.6 | 0.7 | 73.6 | <.001* |
| AU4b | 4.4 | 4.2 | 4.6 | 4847.1 | <.001* | Noisy soundsb | 9.0 | 8.4 | 9.6 | 3835.0 | <.001* |
| AU6b | 2.1 | 2.0 | 2.2 | 1095.0 | <.001* | Requesting helpb | 6.1 | 5.5 | 6.8 | 1084.3 | <.001* |
| AU7b | 4.1 | 3.9 | 4.3 | 4829.0 | <.001* | Groaningb | 5.6 | 5.3 | 5.9 | 3798.0 | <.001* |
| AU9b | 4.0 | 3.8 | 4.2 | 3491.3 | <.001* | Moaningb | 4.8 | 4.5 | 5.1 | 3076.3 | <.001* |
| AU10b | 3.4 | 3.3 | 3.6 | 2712.8 | <.001* | Cryingb | 5.9 | 5.4 | 6.4 | 1423.8 | <.001* |
| AU12b | 1.4 | 1.3 | 1.5 | 189.0 | <.001* | Screamingb | 4.5 | 4.1 | 4.9 | 1150.3 | <.001* |
| AU20b | 4.6 | 4.4 | 4.8 | 3386.1 | <.001* | Loudtalkb | 6.7 | 6.2 | 7.2 | 2199.0 | <.001* |
| AU25b | 3.7 | 3.5 | 3.8 | 3782.3 | <.001* | Howlingb | 4.1 | 3.7 | 4.6 | 690.6 | <.001* |
| AU43b | 3.1 | 2.9 | 3.2 | 1854.4 | <.001* | Sighingb | 8.2 | 7.7 | 8.6 | 5195.9 | <.001* |
| Movement indicator | Behaviour indicators | ||||||||||
| Age (years) | 1.0 | 1.0 | 1.0 | 12.6 | <.001 | Age (years) | 1.0 | 1.0 | 1.0 | 0.6 | .420 |
| Sex femalea | 1.0 | 0.9 | 1.1 | 0.0 | .873 | Sex femalea | 0.9 | 0.8 | 1.0 | 2.4 | .121 |
| Altered random movementb | 7.4 | 6.8 | 8.0 | 2278.1 | <.001* | Introvertb | 4.4 | 4.1 | 4.8 | 1320.6 | <.001* |
| Restlessnessb | 18.7 | 17.7 | 19.7 | 12012.1 | <.001* | Verbally offensiveb | 3.9 | 3.6 | 4.3 | 1192.8 | <.001* |
| Freezingb | 10.8 | 9.9 | 11.9 | 2551.7 | <.001* | Aggressiveb | 4.1 | 3.8 | 4.4 | 1370.1 | <.001* |
| Guarding touchb | 11.8 | 11.2 | 12.4 | 8844.8 | <.001* | Fear or dislikeb | 12.0 | 11.2 | 12.9 | 4721.8 | <.001* |
| Moving awayb | 6.5 | 6.2 | 6.9 | 4333.4 | <.001* | Inappropriate behaviourb | 4.3 | 3.9 | 4.6 | 1201.7 | <.001* |
| Abnormal movementb | 6.1 | 5.8 | 6.4 | 4860.1 | <.001* | Confusedb | 2.5 | 2.3 | 2.6 | 745.0 | <.001* |
| Pacing wandering b | 2.1 | 1.9 | 2.3 | 328.3 | <.001* | Distressedb | 27.9 | 26.5 | 29.3 | 16134.6 | <.001* |
| Activity indicators | Body indicators | ||||||||||
| Age (years) | 1.0 | 1.0 | 1.0 | 7.2 | .007 | Age (years) | 1.0 | 1.0 | 1.0 | 6.0 | .014 |
| Sex femalea | 1.0 | 0.9 | 1.1 | 0.1 | .811 | sex femalea | 0.9 | 0.4 | 1.0 | 1.8 | .181 |
| Resisting careb | 13.5 | 12.8 | 14.3 | 7951.4 | <.001* | Profuse sweatingb | 7.5 | 6.1 | 9.4 | 329.6 | <.001* |
| Prolonged restingb | 2.4 | 2.2 | 2.5 | 685.1 | <.001* | Pale flushedb | 12.4 | 11.6 | 13.3 | 5261.4 | <.001* |
| Altered sleepb | 6.2 | 5.8 | 6.7 | 2765.2 | <.001* | Feverish coldb | 6.9 | 5.5 | 8.5 | 311.6 | <.001* |
| Altered routineb | 6.7 | 6.3 | 7.2 | 2812.9 | <.001* | Rapid breathingb | 20.1 | 18.1 | 22.2 | 3214.187 | <.001* |
| Painful injuriesb | 3.0 | 2.8 | 3.2 | 1339.136 | <.001* | ||||||
| Painful conditionsb | 3.7 | 3.5 | 3.9 | 1940.949 | <.001* | ||||||
AU4 = brow lowering; AU = cheek raising; AU7 = tightening of eyelids; AU9 = wrinkling of nose; AU10 = raising of upper lip; AU12 = pulling at corner lip; AU20 = horizonal mouth stretch; AU25 = parting lips; AU43 = closing eyes. Pain(low/high) model reports a binary logistic with Odds Ratio (Exp[β]) and 95% confidence intervals (CI) presented. Category compared to amale babsent. *Statistically significant at P < .000001.
Discussion
In this study, we mined a large database of pain assessments conducted in real world clinical practice by trained care staff. To our knowledge, this is the largest database of its kind describing the manifestation of pain-related features in people living with dementia. Our findings shed light on how this vulnerable population group could manifest pain, as captured by a range of pain indicators across multiple domains of pain assessment.
Our findings reveal that there may be distinct phenotypes of pain expression among persons with dementia. For instance, we demonstrate that people living with dementia who exhibit indicators suggestive of moderate or severe levels of pain can generally be described as manifesting their higher levels of pain (i.e. moderate to severe pain) through facial expressions such as horizontal mouth stretch and brow lowering and vocalising noisy pain sounds, sighing and talking loudly. Furthermore, they demonstrate restlessness and tendency to guard or protect their body parts. They also experience distress, dislike being touched, resist care and breathe rapidly. While association between pain and individual behaviours such as facial expressions, vocalisations, neuropsychiatric symptoms and body movement related indicators including restlessness and guarding have been reported before [3, 16–18], this is the first time that it has been done so comprehensively while considering these behaviours together and therefore analysing all widely used domains of pain assessment.
As expected, the number of pain indicators present is higher in the presence rather than the absence of pain level categorisation and with higher severity of pain overall. Furthermore, our GEE model also provides insight into predictive function of pain indicators vis-à-vis the presence of pain. These findings affirm the validity of a multidimensional approach to assessment of pain in people with dementia unable to self-report pain and provide additional evidence supporting AGS’s recommendations [15].
In just over half of the participants pain indicators consistent with pain were identified during this study period, however the overall mean of pain assessment measurements for all data was in the no pain category. This can be explained in part by the fact that the majority (i.e. 80%) of pain assessments were conducted at rest, which are generally associated with lower levels of pain [19]. In this regard, our study suggests participants had a significantly greater risk of exhibiting indicators consistent with higher pain levels following movement compared to rest and this was not affected by gender, supporting the importance of movement-based assessments of pain in order to better understand how pain affects function [19, 27]. Nonetheless, the fact that higher presence of at rest assessments in our analysis may highlight potential challenges and barriers that nursing home staff face in performing movement-based assessments. While training provided to PainChek users highlights the relevance of movement-based assessments, more emphasis should be placed on supporting staff to monitor specific better outcomes for residents. The decline from pain to no pain within participants’ assessments is consistent with the fact that having identified pain through validated assessment, typically clinicians would likely intervene to reduce or resolve pain, in accordance with good clinical practice.
A key strength of this study is that our findings are derived from a large database of pain assessments which enhances its generalizability. This dataset was compiled digitally through automatic synchronisation of point-of-care assessments via cloud transmission hence supporting documentation processes and reporting accuracy. An additional strength is that data comes from pain assessments that were standardised using a validated pain assessment tool and purposely trained assessors therefore minimising inconsistency.
A number of limitations also bear consideration. Firstly, we do not account for types or severity of dementia as the database used does not collect this data. The sub-type of dementia is important considering that pain processing can be affected by different neural mechanisms affected by various types of dementia [3, 16, 28]. While all trained staff who administered the PainChek tool determined subjects’ ability to self-report pain and hence only administered the tool in those unable to reliably self-report, our findings do not take into account dementia severity levels. This does not allow us to map pain expression across the continuum of dementia severity. Additionally, our findings are limited by the fact that we did not account for potential effects of race, ethnicity, other medical conditions and medication use which may have affected the results. Lastly, it should also be taken into consideration that our findings are based on pain assessment that indicate the likely presence of pain. While our findings indicate a decline from pain to no pain in study participants (likely due to pain resolution following intervention), future studies should focus on exploring pain assessment and monitoring following intervention.
Recent studies point out the role of digital solutions in contributing towards phenotyping of pain which can lead to important clinical implications in terms of improving patient care and increasing pain assessment objectivity [12, 29]. These authors primarily link adding value to digital phenotyping through observable patient characteristics collected in digital format by using sensing devices. In this study we describe how behaviours suggestive of pain manifest across different pain levels using a comprehensive set of pain indicators collected through a real-time digital documentation process at point of care. Our findings suggest that assessment and monitoring of these pain indicators could mitigate against a potential risk of not detecting pain and hence missing certain pain phenotypes. This supports other studies suggesting that assessing pain through different categories of pain behaviours is considered a strength [30]. Our findings also suggest variability in terms of how age and gender predict occurrence of indicators suggestive of pain presence and high-level pain categorisation. This was evidenced in a number of widely used domains of pain assessment. As an example, vocalisation and movement indicators were impacted by both gender and age in predicting pain presence categorisation. On the other hand, vocalisation indicators were impacted by gender and age in terms of their predictability of high-pain level categorisation but movement indicators were only influenced by age. In this regard, behavioural indicators were not significantly influenced by age or gender. These are important considerations that could provide useful insights towards phenotyping of pain. Further research is needed to help us advance how to leverage more from a digitally assembled large datasets like this and hence make additional contributions towards clinical phenotyping and a better understanding of pain intensity. For example, pain indicators can be clustered, hence confirming clinically suspected distinct patterns of pain manifestation making it easier for clinicians to provide specific pain management recommendations related to pharmacological and non-pharmacological interventions [31]. This clustering would be based on inter-dependency and relationships between various individual pain indicators within and across pain assessment domains.
Conclusions
This study provides empirical evidence from real world clinical practice on manifestation of pain indicators across different levels of pain. We validated the predictive value of a comprehensive set of indicators used in the assessment of pain. Furthermore, our results suggest that people living with dementia exhibit indicators consistent with greater levels of pain following movement and are impacted by pain multidimensionally. Cumulatively, our findings contribute towards the better understanding of how pain could be experienced by this vulnerable population group and provide valuable foundation information that could contribute towards clinical phenotyping of pain.
Supplementary Material
Acknowledgements
We thank everyone involved in the study. We thank Scott Robertson from PainChek Ltd for facilitating compilation of the dataset for analysis in this study.
Contributor Information
Kreshnik Hoti, Faculty of Medicine, University of Prishtina, 31 George Bush st Prishtina, Prishtina 10000, Kosovo.
Ipsit V Vahia, Department of Psychiatry, Harvard Medical School, Boston, MA, USA; Division of Geriatric Psychiatry, McLean Hospital, Belmont, MA, USA.
Paola Chivers, Institute for Health Research, The University of Notre Dame Australia, Fremantle, Western Australia, Australia; School of Medical and Health Sciences, Edith Cowan University, Joondalup, Western Australia, Australia; Research Department, Child and Adolescent Health Service, Nedlands, Western Australia, Australia.
Jeffery David Hughes, Curtin University – School of Diagnostic and Therapeutic Sciences, Perth, Western Australia, Australia.
Declaration of Conflicts of Interest
K.H. and J.D.H. are co-inventors of the original PainChek® instrument (branded ePAT at the time), which was acquired and subsequently commercialised by PainChek Ltd. They are shareholders of PainChek Ltd. They are also named as co-inventors with Mustafa Atee on the patent entitled ‘A pain assessment method and system’, which has been granted in the United States, China, Japan and Europe. K.H. acts as an independent consultant in the capacity of a Senior Research Scientist by PainChek Ltd, while also serving as a Professor at the University of Prishtina. J.D.H. is employed as the Chief Scientific Officer of PainChek Ltd and holds an Emeritus Professor appointment at the School of Diagnostic and Therapeutic Sciences, Curtin University. P.C. was engaged through DATaR Consulting and was paid as an independent private consultant to undertake the biostatistical analysis for the project by PainChek Ltd. P.C. also serves as Manager Research Support and Development, Child and Adolescent Health Service and holds an Adjunct appointment at the School of Medical and Health Sciences, Edith Cowan University.
Declaration of Sources of Funding
This work was supported by PainChek Ltd; PC was paid as an independent consultant for biostatistical analysis. The funding or sponsor had no other role in the study.
Data Availability
Data which are deidentified and do not violate confidentiality can be made available following approval from PainChek Ltd, upon reasonable request. These data cannot be shared with a third party and can be used for research purposes only, not for product-related work.
References
- 1. Hadjistavropoulos T, Herr K, Prkachin KM et al. Pain assessment in elderly adults with dementia. Lancet Neurol 2014;13:1216–27. 10.1016/S1474-4422(14)70103-6. [DOI] [PubMed] [Google Scholar]
- 2. van Kooten J, Smalbrugge M, van der Wouden JC et al. Prevalence of pain in nursing home residents: the role of dementia stage and dementia subtypes. J Am Med Dir Assoc 2017;18:522–7. 10.1016/j.jamda.2016.12.078. [DOI] [PubMed] [Google Scholar]
- 3. Atee M, Morris T, Macfarlane S et al. Pain in dementia: prevalence and association with neuropsychiatric behaviors. J Pain Symptom Manage 2021;61:1215–26. 10.1016/j.jpainsymman.2020.10.011. [DOI] [PubMed] [Google Scholar]
- 4. Sampson EL, White N, Lord K et al. Pain, agitation, and behavioural problems in people with dementia admitted to general hospital wards: a longitudinal cohort study. Pain 2015;156:675–83. 10.1097/j.pain.0000000000000095. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Flo E, Gulla C, Husebo BS. Effective pain management in patients with dementia: benefits beyond pain? Drugs Aging 2014;31:863–71. 10.1007/s40266-014-0222-0. [DOI] [PubMed] [Google Scholar]
- 6. Pringle J, Mellado AS, Haraldsdottir E et al. Pain assessment and management in care homes: understanding the context through a scoping review. BMC Geriatr 2021;21:1–3. 10.1186/s12877-021-02333-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Collins JT, Harwood RH, Cowley A et al. Chronic pain in people living with dementia: challenges to recognising and managing pain, and personalising intervention by phenotype. Age Ageing 2023;52:1–4. 10.1093/ageing/afac306. [DOI] [PubMed] [Google Scholar]
- 8. Haasum Y, Fastbom J, Fratiglioni L et al. Pain treatment in elderly persons with and without dementia: a population-based study of institutionalized and home-dwelling elderly. Drugs Aging 2011;28:283–93. 10.2165/11587040-000000000-00000. [DOI] [PubMed] [Google Scholar]
- 9. Rodger KT, Greasley-Adams C, Hodge Z et al. Expert opinion on the management of pain in hospitalised older patients with cognitive impairment: a mixed methods analysis of a national survey. BMC Geriatr 2015;15:1–5. 10.1186/s12877-015-0056-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Morrison RS, Siu AL. Survival in end-stage dementia following acute illness. JAMA 2000;284:47–52. 10.1001/jama.284.1.47. [DOI] [PubMed] [Google Scholar]
- 11. Achterberg WP, Erdal A, Husebo BS et al. Are chronic pain patients with dementia being undermedicated? J Pain Res 2021;14:431–9. 10.2147/JPR.S239321. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Tan ECK, Jokanovic N, Koponen MPH et al. Prevalence of analgesic use and pain in people with and without dementia or cognitive impairment in aged care facilities: a systematic review and meta-analysis. Curr Clin Pharmacol 2015;10:194–203. 10.2174/157488471003150820144958. [DOI] [PubMed] [Google Scholar]
- 13. Collins JT, Walsh DA, Gladman JR et al. The difficulties of managing pain in people living with frailty: the potential for digital phenotyping. Drugs Aging 2024;41:199–208. 10.1007/s40266-024-01101-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Edwards RR, Dworkin RH, Turk DC et al. Patient phenotyping in clinical trials of chronic pain treatments: IMMPACT recommendations. Pain Reports 2021;6:e896. 10.1097/PR9.0000000000000896. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. American Geriatrics Society (AGS) . Panel on persistent pain in older persons: the management of persistent pain in older persons. J Am Geriatr Soc 2002;50:205–S224. 10.1046/j.1532-5415.50.6s.1.x. [DOI] [PubMed] [Google Scholar]
- 16. Atee M, Hoti K, Chivers P et al. Faces of pain in dementia: learnings from a real-world study using a technology-enabled pain assessment tool. Front Pain Res 2022;3:827551. 10.3389/fpain.2022.827551. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Hoti K, Atee M, Chivers P et al. Technology-guided assessment of vocalisations and their diagnostic value as pain indicators for people living with dementia. Age Ageing 2023;52:1–9. 10.1093/ageing/afad088. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Kunz M, Meixner D, Lautenbacher S. Facial muscle movements encoding pain—a systematic review. Pain 2019;160:535–49. 10.1097/j.pain.0000000000001424. [DOI] [PubMed] [Google Scholar]
- 19. Booker SQ, Herr KA, Horgas AL. A paradigm shift for movement-based pain assessment in older adults: practice, policy and regulatory drivers. Pain Manag Nurs 2021;22:21–7. 10.1016/j.pmn.2020.08.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Corbett DB, Simon CB, Manini TM et al. Movement-evoked pain: transforming the way we understand and measure pain. Pain 2019;160:757–61. 10.1097/j.pain.0000000000001431. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Atee M, Hoti K, Parsons R et al. Pain assessment in dementia: evaluation of a point-of-care technological solution. J Alzheimer's Dis 2017;60:137–50. 10.3233/JAD-170375. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Atee M, Hoti K, Hughes JD. A technical note on the PainChek™ system: a web portal and mobile medical device for assessing pain in people with dementia. Front Aging Neurosci 2018;10:117. 10.3389/fnagi.2018.00117. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Hoti K, Atee M, Hughes JD. Clinimetric properties of the electronic pain assessment tool (ePAT) for aged-care residents with moderate to severe dementia. J Pain Res 2018;11:1037–44. 10.2147/JPR.S158793. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Achterberg W, Lautenbacher S, Husebo B et al. Pain in dementia. Pain reports 2020;5:e803. 10.1097/PR9.0000000000000803. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Colosimo EA, Fausto MA, Freitas MA et al. Practical modeling strategies for unbalanced longitudinal data analysis. J Appl Stat 2012;39:2005–13. 10.1080/02664763.2012.699954. [DOI] [Google Scholar]
- 26. Lipsitz SR, Fitzmaurice GM, Orav EJ et al. Performance of generalized estimating equations in practical situations. Biometrics 1994;50:270. 10.2307/2533218. [DOI] [PubMed] [Google Scholar]
- 27. Kovach CR, Weissman DE, Griffie J et al. Assessment and treatment of discomfort for people with late-stage dementia. J Pain Sympt Manage 1999;18:412–9. 10.1016/S0885-3924(99)00094-9. [DOI] [PubMed] [Google Scholar]
- 28. Scherder E, Schwaab DF, Pain in dementia. In: Cervero F, Jensen TS, eds. Handbook of Clinical Neurology. Amsterdam: Elsevier, 2006, 817–35. 10.1016/S0072-9752(06)80059-X [DOI] [PubMed] [Google Scholar]
- 29. Patrascu M, Berge LI, Vahia IV et al. The story of pain in people with dementia: a rationale for digital measures. BMC Med 2025;23:227. 10.1186/s12916-025-04057-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Herr K, Coyne PJ, Ely E et al. Pain assessment in the patient unable to self-report: clinical practice recommendations in support of the ASPMN 2019 position statement. Pain Manag Nurs 2019;20:404–17. 10.1016/j.pmn.2019.07.005. [DOI] [PubMed] [Google Scholar]
- 31. Close J. Big data—big opportunity. Age Ageing 2023;52:afac262. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data which are deidentified and do not violate confidentiality can be made available following approval from PainChek Ltd, upon reasonable request. These data cannot be shared with a third party and can be used for research purposes only, not for product-related work.