Elderly people face higher pain risks due to age-related conditions like osteoarthritis, etc. Dementia intensifies this risk due to under-detection and under-treatment, stemming from cognitive impairment and communication limitations [2]. Formerly, it was thought that people with dementia didn’t feel pain due to brain cell injury. However, a 2006 study in Melbourne using fMRI found pain-related brain activity in Alzheimer’s patients, revealing they do feel pain but struggle to express it [3]. In recent years, the inadequacy of pain assessment and management in dementia has earned attention [4][5]. Pain, crucial for avoiding harm, is defined by the International Association for the Study of Pain as an unpleasant sensory and emotional experience [6]. Assessing pain in non-communicative demented elderly patients is challenging and depends heavily on observational tools due to the progression of cognitive decline [7].

For effective pain management, the presence of appropriate pain assessment tools is crucial [8]. In the past two decades, several tools, including self-reporting and behavioural-observational ones, have been designed, and classified as one-dimensional or multi-dimensional [9]. However, self-reporting remains a paradigm; it may not always be possible, particularly in cases of advanced dementia, where behavioural/observational tools are needed.

1.2.1. Commonly Used Self-Reporting Tools

Coloured Analogue Scale: Like a thermometer, it assesses pain intensity by using colours, with white denoting no pain and red showing intense pain. Patients slide a bar along the scale to specify their pain level which correlates with a numerical rating on the back (ranging from 0 to 10). In a study on older adults, 80% with moderate dementia correctly understood the CAS, but tool reliability reduces with progression in dementia severity [10].

Verbal Rating Scale: offers patients options to specify their pain severity level, ranging from no pain to extreme pain, each analogous to a number on the back of the scale.

Numeric rating scale: evaluate pain intensity across different ranges, like 0 - 5 or 0 - 10, where 0 shows no pain and the highest number signifies severe pain.

Visual Analogue Scale: includes lines with verbal descriptors like “no pain” and “severe pain.” Patients indicate their pain level with a cross on the line, and the distance from the start of the line to the cross specifies the score, suggesting pain severity [11].

1.2.2. Observational-Behavioural Tools

In advanced dementia, where verbal communication is impaired, behavioural observation and alternative pain-reporting methods are crucial for pain detection. The American Geriatrics Society (AGS) Panel on Persistent Pain in the Elderly has drawn a comprehensive framework based on behavioural cues to help in nonverbal elderly pain assessment. The framework recognizes 6 behavioural domains for observational pain assessment: facial expressions, verbalizations, body movements, changes in interpersonal interactions, changes in activity patterns, and mental status changes [12]. Many tools have been developed to enhance accuracy in evaluating pain in dementia.

Commonly used tools include the APS, Algo plus, CPAT, and PAINAD [13]. However, various existing tools lack advanced technology and innovative design. Lichtner et al. in a meta-review on pain assessment tools in dementia, highlighted the need for such tools emphasizing the importance of incorporating advanced technology to enhance pain assessment in non-verbal demented patients [14]. In 2017, the electronic Pain Check tool appeared and was approved for assessing pain in non-vocal elderly patients, including those with dementia [15].

APS:

Utilized in advanced dementia patients. It assesses vocalizations, facial expressions, body language, behavior changes, psychological well-being, and physical aspects. Each part is rated on a 4-point scale, with scores ranging from 0 - 18. Qualitative and quantitative evidence support its worth, and it can be completed in one minute [16].

CNPI:

A checklist-based tool that assesses pain using 6 elements: vocalization, facial expression, rubbing, restlessness, response to stimuli, and verbal expressions. Each element is noted as either “present” (1) or “absent” (0) while the patient is active or at rest. Scores range from 0 to 12, with the highest possible score of 12 points [17][18].

CPAT:

Assesses pain-related behaviors by focusing on 5 domains: facial expressions, behavior, mood, body language, and activity level. Each domain is marked as either pain-related (1) or indicating no pain (0), with a maximum total score of 5. When the collective score reaches 1 or more, a follow-up assessment is conducted, prompting caregivers to note their actions. However, none of the reviewed studies on toll offer interpretation guidelines for the total score [19].

DOLOPLUS-2:

An upgraded version of the doloplus, designed for non-verbal elderly patients, with 10 elements grouped into somatic, psychometric, and psychosocial domains and covers 5 of the 6 pain behavior categories outlined in the AGS guidelines [20]. Each item in doloplus-2 is graded with 4 escalating behavioral depictions indicating pain intensity, rated from 0 to 3. Scores are added to calculate a total score ranging from 0 to 30 points, with a threshold of 5 points indicating pain presence. The tool shares similarities with other commonly used pain assessments like PAINAD, PACSLAC, PADE, and APS as they all assess facial expression, body posture/movement changes, and verbal expression [21].

MOBID:

Assesses pain during active body movements, with health personnel performing 5 movements involving the trunk and limbs individually. The assessment is paused instantly and marked if pain behaviors such as utterances, facial expressions, or defense are observed by the assessor [22].

MOBID-2:

The tool has demonstrated its clinical suitability in multiple studies [22][23] and entails 2 segments: the first involves 5 active movements from the MOBID, while the second involves the caregiver reporting pain, starting from the head, and progressing through various body regions [22].

MPS:

Measures pain by assessing facial expressions, body language, vocalization, breathing, agitation cues, sleep/appetite changes, vital signs fluctuations, and pain history. It differentiates between pain and agitation and allows pain to be indicated on a diagram. Patients are ideally monitored at rest, and trained assessors gently tap 22 body regions and mark an “x” on a paper sketch where behavioral reactions or pathology evidence are observed.

NOPPAIN:

Mainly administered by nursing assistants during caregiving activities and identifies pain-related behaviors in people with dementia both at rest and in motion. The tool comprises 4 main divisions as documented by Lints-Martindale et al. [18]:

1) Observation of pain behaviors during care tasks like dressing and bathing.

2) Measurement of pain behaviors’ presence through 6 items such as pain noises and facial expressions etc.

3) Rating pain behavior intensity using a 6-point Likert scale.

4) Using a pain thermometer for overall pain intensity assessment [11].

PACSLAC:

Evaluates both obvious and subtle pain behaviors’ and entails 4 subscales: facial expressions (13 items), activity/body movements (20 items), social/personality/mood (12 items), and physiological indicators/eating and sleeping changes/vocal behaviours’ (15 items), totalling 60 items [18][20]. Each item is assessed on a binary scale (present or absent), yielding a total score ranging from 0 to 60. However, there is currently no established interpretation for this total score.

PADE:

Aims at assessing pain in advanced dementia patients and allows healthcare professionals to observe patient behaviors suggest ing pain. It involves 24 items divided into 3 parts:

Part 1: Physical assessment, focusing on facial expressions, breathing patterns and posture.

Part 2: Global assessment, where caregivers assign an overall pain rating for their patients.

Part 3: Assessment of activities of daily living, including feeding, dressing, and transferring from bed to wheelchair [18].

PAINAD:

Developed for people with advanced dementia, it offers a straightforward and clinically relevant pain assessment by covering 5 behavior categories: respiration, negative vocalizations, facial expressions, bodily movements, and comfort-seeking actions. Each category has 3 components with specific indicators, for identifying pain presence or absence [17][18][24]. Items are rated on a 3-point scale (0 to 2) to reflect severity. Notably, this tool is a modified version of the DS-DAT and the FLACC.

PAINE:

Consists of 22 items divided into 2 sections. The first 15 items comprise certain motor repetitive behaviors, certain vocal repetitive behaviors, unusual behaviors, and activity-related behaviors. The remaining 7 items assess clinical markers such as falls, trembling, vital sign variations, blood stains, edema, and broken bones [13].

ADD:

Developed to identify and manage physical discomfort and uses a checklist with 5 categories of pain-related behaviors: facial expressions (8 items), body language (9 items), vocalizations (9 items), emotional state (5 items), and behavioral responses (11 items) [25]. Even though ADD may not be a conventional tool for pain assessment [11], when potential pain behaviors are noted, the protocol involves 5 steps:

1) Evaluate physical signs and symptoms.

2) Assess the individual’s current and past pain history.

3) If Steps 1 and 2 yield negative outcomes, assess environmental factors, and try nonpharmacological interventions accordingly.

4) If proven ineffective, administer nonnarcotic analgesics as prescribed in a written order.

5) If symptoms persist, seek consultation with a medical professional or another qualified healthcare provider.

DS-DAT:

Initially designed for research purposes, it assesses discomfort in elderly with advanced dementia who rely on caregivers for pain management. It has also been applied in research to measure pain in dementia [11]. It consists of 9 behavioral items and evaluates discomfort in patients with Alzheimer-type dementia through behaviors like noisy breathing, negative vocalizations, and facial expressions. Each item is scored based on presence, frequency, duration, and severity, resulting in a total score ranging from 0 - 27 (indicating the highest level of detected discomfort).

REPOS:

A potential tool for patients at varying cognitive levels (van Herk et al., 2009) and comprises 10 behavioral items. Observers mark these items as present or absent after observing the non-communicative patient for 2 minutes during a possible painful care moment [26].

PAIC:

It includes 15 pain-related indicators that help in assessing patients who may struggle to verbalize their discomfort.

EPCA:

Assesses pain severity in non-verbal elderly patients through 8 behavioral items [5].

PIMD:

It assesses pain severity in moderate-severe dementia using 7 items. Each item is rated on a scale of 0 - 3, with 0 indicating ‘Absence’ and 3 indicating ‘Severe’. Total scores range from 0 - 21, with higher scores signifying the highest pain intensity [27].

The ePAT, or PainChek, is a smart device application designed to support healthcare providers in assessing pain in patients with moderate to severe dementia. It utilizes automated facial recognition and clinical markers to assess pain. The PainChek software system comprises 2 main components:

1) Mobile application;

2) Web Admin Portal [28].

The mobile app consists of 6 domains: FACE, VOICE, MOVEMENT, BEHAVIOUR, ACTIVITY, and BODY, totaling 42 items. Domain 1 evaluates 9 micro facial expressions, automatically detected by the app using AI algorithms. Domains 2 - 6 involve clinical observations completed by the user. The final pain assessment score is automatically calculated for each domain, classifying pain as no pain (0 - 6), mild pain (7 - 11), moderate pain (12 - 15), or severe pain (≥16). The second component, WAP, is a protected website for managing patient data and user access [28][29].

For this study, a literature review methodology was selected due to its value in examining connections across studies, identifying research gaps, and guiding future inquiry. In addition, Kable, Pich, and Maslin-Prothero’s 12-step documentation framework was followed, together with the PRISMA 2020 checklist to ensure clarity and methodological rigor.

A systematic literature search was conducted across PubMed, Google Scholar, Academia, and the University of South Wales Library. The search strategy combined MeSH terms and free-text keywords related to dementia, pain assessment, and psychometric properties.

Example PubMed search strategy:

(“Dementia” [MeSH] OR dementia OR “Alzheimer Disease” [MeSH])

AND (“Pain Measurement” [MeSH] OR “pain assessment”)

AND (validity OR reliability OR psychometric*).

Limits applied:

English language; publication date January 2003 - January 2023; human subjects; aged ≥ 65 years.

Google Scholar searches used the string:

“dementia” AND “pain assessment” AND “psychometric properties”,

with results restricted to 2003 - 2023.

The study selection process comprised three phases. Phase 1 involved database searching, yielding 25,045 records. Phase 2 included duplicate removal and title/abstract screening, reducing the dataset to 179 studies. Phase 3 applied predefined inclusion and exclusion criteria, resulting in 55 studies eligible for further review. An additional 10 studies were identified through citation tracking, producing 65 articles for full-text assessment. Following full-text evaluation, 23 studies and two guidelines met the eligibility criteria (Table 1). Guidelines discussed narratively but not counted as included records (see Figure 1).

Table 1.Criteria for study selection.

Figure 1.Strategy & study selection.

Characteristics of the included studies :

See Appendix 1: Summary of the characteristics of the included studies.

The quality of the included studies underwent assessment using various tools. Observational studies and RCTs were evaluated using the Newcastle Ottawa Scale, typically scoring 6 out of 9, suggesting medium quality. Reviews were assessed with the AMSTAR tool and Oxford Centre for Evidence-Based Medicine (CEBM) Criteria. Although most reviews demonstrated thorough searches, they lacked specific details. Among them, two studies were rated as low quality, four as medium, and two as high, resulting in an overall assessment of medium quality for the included studies.

Screening of titles, abstracts, and full-text articles was conducted by a single reviewer using predefined inclusion and exclusion criteria. Due to the study being undertaken as part of an MSc dissertation, duplicate screening was not feasible.

Data extraction was performed by the same reviewer using a standardized data extraction form. Structured procedures were applied to ensure consistency. While single-reviewer processes may introduce bias, predefined criteria and systematic methods were used to minimize this risk.

The following data was derived from the selected studies:

Given the diverse range of studies, a narrative synthesis approach was employed to draw meaningful conclusions from the included literature. The review compared tools based on criteria such as construct, concurrent/criterion validity, internal consistency (homogeneity), inter-rater, intra-rater reliability, and applicability/clinical utility, while also evaluating the impact of ePAT on health services. Inter-rater reliability was measured using correlation, intra-class correlation, kappa and Spearman’s correlation coefficients, and percentage agreement. Intra-rater reliability was reported using correlation and intra-class correlation coefficients, while internal consistency was assessed using Cronbach’s alpha.

The methodological quality of the included studies was considered during the evidence synthesis process. Given that most studies were appraised as medium quality, the strength of the conclusions was framed with appropriate caution. Findings originating from low-quality studies were interpreted conservatively and were not relied upon as primary evidence when formulating comparative judgments regarding psychometric performance. Accordingly, any statements indicating the relative superiority of one pain assessment tool over another should be understood as reflecting moderate rather than conclusive evidence, in recognition of the methodological constraints identified across the reviewed literature.

APS: Across included studies conducted predominantly in nursing homes/residential aged-care contexts, APS demonstrated internal consistency α= 0.65 - 0.81, inter-rater reliability 0.75 - 0.88, intra-rater reliability 0.66 - 0.88, and construct validity 0.49 - 0.91. Where criterion/concurrent validity was reported (typically against broader “holistic” clinical judgement approaches in long-term care), the correlation was 0.586.

CNPI: In the cross-sectional descriptive work explicitly reporting rest vs movement, CNPI’s internal consistency differed by condition: α= 0.92 and 0.97 at rest versus α= 0.74 and 0.90 during movement (i.e., the same residents were scored in two conditions, producing separate psychometric estimates). Inter-rater reliability also varied by condition, with ICC 0.70 (rest) vs 0.65 (movement) and kappa 0.25 (rest) vs 0.43 (movement). Construct validity correlations with agitation (PAS) were low at rest (0.16 - 0.17) and higher during movement (0.33 - 0.41), again reflecting the same cohort assessed under two observation contexts. Other included studies (typically in long-term care/nursing settings, often nurse-completed) reported broader ranges: internal consistency 0.60 - 0.90, inter-rater 0.45 - 0.59, intra-rater 0.23 - 0.65, and construct validity 0.46 - 0.88; concurrent/criterion validity against self-report anchors remained low-moderate (e.g., VDS r ≈ 0.372 rest; 0.428 movement; VAS ~0.30 - 0.50).

CPAT: Certified Nursing Assistant Pain Assessment Tool (CPAT)—Evidence comes primarily from nursing home settings in residents with cognitive impairment, with assessments aligned to care staff (CNA) use. Reported internal consistency was α= 0.72 - 0.84. Reliability indices included ICC 0.55 - 0.57, inter-rater reliability 0.71, intra-rater reliability 0.67, and construct validity 0.25. Validity evidence was generally reported as comparisons with other observational tools (e.g., DS-DAT), but the synthesis should note that these studies reflect institutional long-term care workflow rather than acute settings.

DOLOPLUS-2: Validation evidence derives mainly from geriatric facilities and palliative care settings, where the tool is intended to capture progressive/ongoing pain rather than momentary procedural pain. Across studies, internal consistency ranged α= 0.67 - 0.95; inter-rater reliability 0.35 - 0.86; ICC 0.77 - 0.90; intra-rater reliability 0.71; and construct validity 0.33 - 0.70. Concurrent/criterion validity varied depending on the comparator: PAINAD ~0.34, PACSLAC ~0.29 - 0.38, and self-report ~0.31 - 0.65 (where self-report was feasible).

DS-DAT: Used largely in Alzheimer-type dementia/severe cognitive impairment cohorts, typically in care environments where observation is feasible. Across studies, internal consistency was consistently high (α= 0.86 - 0.89). Inter-rater reliability was reported either as % agreement (84% - 94%) or correlations (0.61 - 0.98), and intra-rater reliability was reported around 0.60. Concurrent/criterion validity varied by comparator (e.g., CMAI 0.25; PAS 0.51; self-report VAS 0.31 - 0.65).

MOBID and MOBID-2:

These tools are explicitly movement-evoked pain assessments, with pain judged during standardized mobilizations performed by staff; evidence is therefore anchored to movement-based observation rather than rest-only scoring. MOBID showed internal consistency α= 0.82 - 0.90, inter-rater reliability 0.86 - 0.97, ICC 0.70 - 0.96, kappa 0.05 - 0.90, and intra-rater reliability 0.79 - 0.92; construct validity was 0.51 - 0.54 with concurrent validity against proxy reports 0.41 - 0.64. MOBID-2 studies reported internal consistency α= 0.82 - 0.94, inter-rater ICC 0.80 - 0.94, intra-rater 0.85 - 0.92, with good construct/concurrent validity in clinical settings.

NOPPAIN:

Designed for nursing assistants during caregiving tasks (i.e., pain observation embedded in care activity), and therefore best interpreted as care-activity/movement-linked observation rather than quiet rest observation. Across studies/reviews, internal consistency was α= 0.65 - 0.84; inter-rater reliability 0.79 - 0.94; agreement 82% - 100%; kappa 0.70 - 0.87; intra-rater reliability 0.71 - 0.89; and construct validity 0.48 - 0.88.

PAINAD:

Evidence includes observation during contrasting activity conditions (e.g., pleasant vs unpleasant activity), which should be stated explicitly because some reliability coefficients are condition-specific rather than different samples. Internal consistency ranged α= 0.50 - 0.88; inter-rater reliability 0.72 - 0.97; ICC 0.76. One study reported Pearson correlations 0.97 during pleasant activity vs 0.82 during unpleasant activity—these are best read as the same cohort rated under two activity contexts rather than separate cohorts. Intra-rater reliability ranged 0.71 - 0.89, construct validity 0.48 - 0.88, and concurrent/criterion validity varied by comparator (DS-DAT 0.56 - 0.76, proxy report 0.84, self-report Pain VAS 0.75, discomfort VAS 0.76).

PACSLAC I & II:

PACSLAC I evidence is largely drawn from dementia care settings where repeated observation is feasible; internal consistency ranged α= 0.74 - 0.92 with inter-rater reliability 0.52 - 0.96, ICC 0.77 - 0.96, agreement 94%, intra-rater reliability 0.86 (and intra-rater ICC 0.72 - 0.96). Construct validity ranged 0.54 - 0.72, with concurrent/criterion validity against proxy pain report 0.35 - 0.54. For PACSLAC II, reported internal consistency was α= 0.74 - 0.77 and inter-rater reliability 0.63 - 0.86; note that reported construct validity values should be explicitly labelled by statistic type (e.g., correlation/ICC) to avoid confusion.

PADE:

Evidence is primarily from long-term care facilities, including a study in four LTC facilities (n = 25) and an experimental reliability design using two raters over 10 days (784 observations). Internal consistency was wide (α= 0.54 - 0.96) and differs by subscale/part, so the synthesis should state that Part I was stronger (α= 0.76 - 0.88) than Part III (α= 0.23 - 0.63). Inter-rater reliability (ICC) ranged 0.54 - 0.96, with very high ICCs reported by part (I 0.93, II 0.81, III 0.96) in the repeated-observation design; intra-rater reliability ranged 0.70 - 0.98.

ePAT/PainChek (novel tool):

The key point for synthesis is that psychometric estimates are frequently derived from paired assessments on the same residents, split into rest vs movement, and that assessor roles differ by tool. In one prospective observational study across three Australian residential aged-care facilities (n = 40; mean age ~79.7), APS was administered by facility staff, while ePAT was administered mainly by the researcher (with some staff assistance). The dataset comprised 353 paired assessments (209 at rest; 144 during movement)—these are the same cohort measured twice under different conditions, not separate samples. Reported ePAT properties were internal consistency α= 0.925, inter-rater reliability vs APS weighted kappa = 0.74 (rest 0.71; movement 0.78), and concurrent validity r = 0.882. A second prospective observational study in two residential aged-care facilities (n = 34) also used paired rest/movement assessments (400 paired assessments: 204 rest; 196 movement). It reported overall internal consistency α= 0.950, with condition-specific alphas α= 0.766 (rest) and α= 0.797 (movement)—again, same cohort, different conditions. Inter-rater reliability was weighted kappa = 0.857; ICC 0.902 (rest) and 0.879 (post-movement); concurrent validity vs APS was r = 0.911 overall (rest 0.897; movement 0.904).

Clinical utility is defined as the usefulness of the measure for making decisions [26] and helps in the management of the patient. Assessing clinical utility involves considerations like the availability of cut-off scores, item numbers, and scoring interpretation for decision-making. However, evidence regarding clinical utility is often lacking, with scoring criteria and cut-off scores frequently scarce. Numerous pain assessment tools mentioned were mostly evaluated in nursing homes or residential care facilities, with limited data on their application in other healthcare settings. Most studies on clinical utility lack detailed evidence, except Lichtner et al. [14] which reviewed the utility of pain assessment tools but stressed the need for further research. Abbey’s pain scale, with clear scoring and extensive use in residential facilities, is included in Australian Pain Guidelines. PAINAD shows potential in emergency departments for elderly patients with cognitive impairment, with studies revealing significant pain score improvements post-analgesia. CNPI is ideal for its ease in clinical settings, but its binary scoring may limit sensitivity to pain changes. NOPPAIN requires minimal training and is valued by nursing assistants. DS-DAT shows potential across facilities with proper training, but lack of training leads to incorrect use. CPAT’s clinical utility is renowned but needs further validation. ePAT’s cutoff score (≥7) and high clinical utility (0.95) are validated, with strong sensitivity (96.1%), specificity (91.4%), and accuracy (95.0%) in dementia pain assessment. Its user-friendly app design and comprehensive training materials enhance accessibility and usefulness in various settings.

ePAT shows positive impact on Health Services by simplifying pain assessment with objective scoring [30]. Its structured methodology aids in assessing pain in non-verbal dementia patients, improving interdisciplinary communication for better pain management [29]. The electronic data collection streamlines pain profiling, allowing researchers to identify patterns and enhance treatment outcomes over time, aligning with recommendations for dementia patient management [14].