It is necessary to measure pain for a variety of clinical and research purposes, including etiology, diagnosis, monitoring disease state, and measuring and understanding intervention effect. Pain measures are part of the diagnostic criteria and classification criteria for various conditions, including fibromyalgia [1], chronic migraine [2], osteoarthritis [3], and rheumatoid arthritis [4]. Pain is also a common symptom for cancer, and the location of the pain is associated with the type and stage of the cancer [5].

Digital pain manikins, also known as pain drawings, pain charts, or pain body maps, are an increasingly common tool used to gather self-report pain data. They are an outline diagram of a human body, typically with a front and back view. Newer digital versions may provide more detail by using shading to indicate the breasts and chest cavity or the knee shape and structure [6]. People self-report pain by marking or coloring the location of their pain using a touch screen or mouse [7]. A key feature of pain manikins compared to other pain instruments is that people can self-report pain spatially. This gives pain manikins unique potential as a pain measurement tool. Throughout this review, we use the term “digital pain manikin” to refer to the tool and “pain drawing” to refer to an instance of a report created using a digital pain manikin.

We categorized the summary and analysis of digital pain drawings as an individual drawing–level summary measure, a multidrawing summary method, or a multidrawing analysis method. Individual drawing–level summary measures quantify an aspect of an individual’s pain experience at a specific moment in time. For example, pain extent (also referred to as pain area) quantifies the area of pain as marked on a single pain drawing [8]. Multidrawing summary methods give information about pain across a population, across time, or both. For example, heat maps (images visualizing an average of multiple pain drawings) can show the most common locations for an individual’s pain over time or the average pain profile for a specific condition across a population [9]. Multidrawing analysis methods provide direct interpretation of digital pain drawings rather than only compressing them, for example, by using machine learning clustering methods to group similar pain drawings together and characterize the distribution of pain [10].

Previous systematic reviews have noted a lack of standardization both in pain manikins and in summary measures derived from them [7,11], which may introduce problems with reproducibility of results. It also limits the ability to compare and synthesize results meaningfully across studies. For example, drawing from measurement theory [12], a lack of comparable conceptualizations of the constructs that are to be measured hampers the assessment of measurement properties (such as reliability and validity) of manikin-derived summary measures. Understanding the current state of how digital pain drawings derived from digital manikins are summarized and analyzed in the field is, therefore, a crucial step toward building more robust, reproducible, and scalable methods.

A 2019 systematic review of methodological milestones in pain manikins divided manikin-derived measures into “topographic” and “simple” measures, with topographic measures being those incorporating anatomical knowledge [7]. They found the most common simple measures to be those quantifying the size of the painful area and widespreadness to be the most widely used topographic measure. However, mapping digital pain manikin summary measures and analysis methods was not a focus of the review. This means that the full picture of which manikin-derived summary measures and analysis methods are being used and which pain constructs these relate to is not currently established [7].

Therefore, this review maps the ways digital pain drawings are summarized and analyzed, including the pain constructs measured using digital pain manikins.

The specific objectives were to (1) identify and characterize studies that used digital pain manikins for data collection, (2) identify (a) which individual drawing–level summary measures they reported and (b) the methods by which these summaries were calculated, and (3) identify if and how multidrawing (eg, time series) summary and analysis methods were applied.

We expect this review to inform the direction of future work on developing more advanced manikin-derived summary measures and analysis methods that make best use of the spatial information manikins provide. Ultimately, this will contribute to harnessing the potential of digital manikins to support pain outcome measurement in both research and clinical care.

We reported this review in line with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) reporting guidelines for scoping reviews [13]. The completed PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) checklist is available in Multimedia Appendix 1.

We limited the scope of this review to digital pain manikins, excluding paper pain manikins. The main reasons were that the field is shifting away from paper manikins and toward digital manikins [7] and that we anticipated that any summary or analysis methods applied to paper manikins were likely to also be represented in studies using digital manikins.

We used the same search strategy as for a related review, which was registered on PROSPERO (an international database of prospectively registered systematic reviews in health and social care) [14].

We searched MEDLINE, CINAHL and Embase via Ovid, Scopus, IEEE Xplore, and the ACM Digital Library using search terms related to pain and manikin, including a range of common synonyms such as pain drawing and pain body chart. The full search strategy is included in Multimedia Appendix 2. The search was not restricted based on publication date and was complemented by hand searching reference lists of included studies. We did not search additional sources for gray literature. We originally ran the search in November 2019 and updated the search in August 2023. The search strategy was developed by researchers with experience of conducting systematic reviews and supported by a qualified librarian.

Papers were included if they were published in English and met the criteria outlined in Textbox 1.

After deduplication, we performed title and abstract screening to identify potentially relevant papers, followed by full-text screening to confirm eligibility. Deduplication was performed by a single author (DAM). For both screening stages, all papers were screened independently by pairs of reviewers (SMA, DAM, Dr Rebecca Lee and Danielle Mountain), with disagreements resolved by discussion with a third author (SvdV).

We recorded whether a study was a clinical study, where a manikin was used for data collection to answer a clinical research question, or a development or validation study, where the primary aim of the study was one or both of the development and testing of a digital pain manikin. We also recorded whether a manikin was 2D, 3D, or pseudo-3D. We defined a 3D manikin to be a manikin with a rotatable model, as opposed to a pseudo-3D manikin with a fixed 2D perspective but with additional visual detail and shading that gave it a 3D appearance.

We developed a data charting form and pilot-tested it on 10 papers before starting full data charting. Data charting was performed for all included papers by one author (DAM), with 25% (23/92) in duplicate by a second author (SMA). Missing data on study, setting and population, and manikin characteristics was noted as “not reported” during data charting or extracted from references to previous studies using the same manikin or dataset.

We extracted data items related to study characteristics, setting and population characteristics, manikin characteristics, individual drawing–level summary measures and the methods used to produce them, multidrawing summary measures (ie, cross-sectional pain drawings across multiple individuals, multiple pain drawings of an individual over time, or multiple pain drawings of multiple individuals over time), and other analysis methods. Data items on manikin characteristics included which location-specific pain aspects could be recorded on the manikin, including location-specific pain quality (eg, burning or tingling) and location-specific intensity (typically on a scale of 1-10). This is distinct from additional nonmanikin measures collected at the same time (eg, an overall pain intensity score or participant ratings of the usability of the manikin). The full list of data extraction items is available in Multimedia Appendix 3.

Individual drawing–level manikin summary measures were defined as any variable extracted directly from a single pain drawing, compressing the high-dimensional pain drawing data into a single measurement, such as pain extent.

Multidrawing summary methods were defined as any method of combining (or compressing) data from multiple individual pain drawings without first summarizing the individual manikins, such as heat maps showing the average of multiple pain drawings.

Multidrawing analysis methods were defined as a method that produced new information about the data (eg, the use of principal component analysis to assess the knee pain distribution), as opposed to multidrawing summary methods that only compressed the data.

When a study reported descriptive summary statistics of individual summary measures, we extracted this as an individual drawing–level summary measure and not a multidrawing summary method. For example, if a study calculated the pain extent for each individual pain drawing and then reported the average pain extent across participants at baseline and follow-up, we recorded pain extent as an individual drawing–level manikin summary measure and did not record mean pain extent as a multidrawing summary measure.

We defined automated measures as those that were extracted without human intervention on an individual drawing level. For example, if the calculation of pain extent required manual tracing of the pain area, this was counted as manual even if part of the process was performed automatically. If a measure was not explicitly stated to be manual or automated but we could derive it from contextual information, we recorded this as manual or automated (assumed). For example, in studies involving thousands of manikins, manual processing was unlikely, so measures were assumed to be automated unless there was evidence to the contrary.

Guided by our objectives, we performed a narrative synthesis of the extracted data. For the synthesis of individual drawing–level summary measures (objective 2), we named and defined pain constructs after performing data extraction. A construct is an abstract concept that cannot be directly measured.

We defined pain constructs descriptively according to the following principles:

We recorded unique individual drawing–level summary measures within constructs when there were significant methodological variations in how that construct was measured (eg, pain extent with region-based and pixel-based measures) or where there were minor conceptual variations within the construct (eg, pain presence or absence inside a specific anatomical location versus pain presence or absence outside a specific anatomical location). The level of conceptual variation at which we defined a new construct instead of a summary measure within a construct was a subjective distinction. It would, for example, also be possible to consider pain presence inside versus outside a specific anatomical location as separate constructs, rather than different measures within the same construct.

Figure 1 shows that our search identified 5981 papers after deduplication, with another 208 identified via our hand search, resulting in a set of 6189 papers to be screened. Finally, we included 92 papers. The main reasons for excluding full papers were that they used paper-based manikins (637/6189, 10.29%) or did not use a manikin at all (132/6189, 2.13%). Of the 92 included studies, 87 (95%) reported at least 1 individual drawing–level summary measure, 42 (46%) reported at least 1 multidrawing summary, and 16 (17%) reported direct analysis of multiple pain drawings.

Table 1 shows the characteristics of the included studies. Most studies were conducted in the United States (31/92, 34%) or Denmark (18/92, 20%), were clinical studies (51/92, 55%), and collected data cross-sectionally (64/92, 70%).

Across all studies, we identified 27 unique, named manikins. Almost a quarter of included studies used the Navigate Pain manikin (21/92, 23%), and 23 (25%) studies did not report details on which manikin they used. Most studies used a 2D (61/92, 66%) or pseudo-3D (21/92, 23%) manikin. A total of 3 studies reported using a 3D manikin and 2 compared manikins of different dimensions; these were all developmental or validation studies. Manikins were used across a variety of conditions, primarily for chronic pain (52/92, 57%).

Manikins were most commonly either pixel-based manikins where participants could draw freely (56/92, 61%), analogous to paper-based manikins (such as the Navigate Pain manikin [15]), or manikins with nonoverlapping predefined regions that participants can select (21/92, 23%; eg, the CHOIR (Collaborative Health Outcomes Information Registry) manikin [16]). Exceptions to this include the Manchester Digital Pain Manikin, which used a grid [17]; the manikin used by Zuhdi et al [18] with overlapping predefined joints and areas; an early iteration of the Iconic Pain Assessment Tool, which uses drag-and-drop icons [19]; and the manikin used by Miękisiak et al [20], where users clicked with a mouse to make individual marks rather than shading areas. Example manikin images are shown in Figure 2 [20-22].

Table 3 shows the 5 multidrawing summary methods we identified. Of the 42 papers summarizing data from multiple pain drawings, 38 reported cross-sectional summaries of populations, 2 reported a summary of 1 individual over time, and 3 reported summaries of populations over time. A total of 4 studies incorporated location-specific pain aspects in their multidrawing summary methods: 2 used site- and symptom-specific methods and 2 used maximum symptom methods. A total of 35 studies presented a heat map. This included pixel-wise averages of multiple individual manikins (n=23; Figure 3 [9,100]); the region-based equivalent showing the proportion of reports selecting each individual region (n=7); and pixel-wise averages with some type of additional processing, such as a minimum threshold for the number of participants reporting pain in a particular location (n=4). Most heat maps (n=32) were cross-sectional summaries of populations.

Studies that reported the number of participants with pain in a particular location (n=14) for pixel-based manikins (n=9) calculated this by overlaying predefined regions and counting the number of participants who marked pain within each region; the other 5 studies reporting this measure used region-based manikins. A total of 2 studies reported site- and symptom-specific summaries; one study provided the number of participants who reported a specific pain quality at a specific site (eg, throbbing and pulsing) [98], while the other study reported the average Numeric Rating Scale score for specific symptoms (frequency of interference, intensity, and influence on playing) at a specific site [18].

The most common analysis performed directly on pain drawings was an assessment of similarity between linked pairs of pain drawings, including pairs generated by researchers copying an example drawing [17] and patient-clinician pairs where the clinician completed a pain drawing based on the patient’s verbal description of their pain [50] (n=6), as shown in Table 4. Of the 6 studies that assessed similarity, 3 calculated the Jaccard index, 2 counted the number of pixels colored in both pairs, and 1 performed manual assessment. Other studies used the Jaccard index as a measure of similarity as part of other analysis methods, for example, Galve Villa et al [42] used it to assess change over time and Alter et al [95] used it in a machine learning clustering technique to identify subgroups in a population. Van der Veer et al [17] used the Jaccard index as part of their assessment of test-retest reliability.

Of the 15 papers reporting a multidrawing analysis method, 9 were developmental or validation papers, 5 were clinical, and 1 was both clinical and developmental or validation. All similarity analysis methods (n=6) were part of development or validation studies for the purpose of evaluating the validity of the derived scores.

Studies also used multidrawing analysis methods to assess change over time (n=3), cluster similar drawings together (n=2), categorize drawings by diagnosis (n=2), or correlate pain location information with other data (n=2). Clustering is a machine learning technique for grouping similar examples together and requires a measure of similarity to be successful. Of the clustering studies, the study by Boudreau et al [10] used principal component analysis for dimensionality reduction followed by k-means clustering, and the study by Alter et al [95] performed hierarchical clustering using the Jaccard index as a measure of similarity. The study by Boudreau et al [10] was the only included study that characterized different patterns of pain distribution within an otherwise homogenous diagnosis, in comparison to other studies that simply summarized pain extent or widespreadness. Preserving the spatial information allowed them to identify 3 subgroups within patellofemoral pain that would not have otherwise been distinguishable. Of the studies that analyzed change over time, 1 study investigated the difference between consecutive pairs of pain drawings [42], and 2 studies reported the area under the pain area-time curve, to quantify the change in pain extent over time [68,74].

Ellingsen et al [45] performed pixel-wise correlation with a pain catastrophizing score, which they defined as “a pain-targeted psychosocial construct comprised of helplessness, pessimism, and magnification of pain-related symptoms and complaints.”

This scoping review identified 92 studies that used digital pain manikins for data collection. Most studies were cross-sectional (64/92, 70%) clinical studies (51/92, 55%) using manikin-derived summary measures to answer a clinical question, although a large minority of studies (51/92, 55%) were methodological work on the development or validation of digital pain manikins. Most studies (84/92, 91%) used 2D or pseudo-3D manikins, and manikins were most commonly pixel based. We identified 10 pain constructs expressed by individual drawing–level summary measures, with significant methodological variation between summary measures for the same pain construct.

Pain extent was the most commonly measured pain construct. It does not make as much use of the available spatial information when compared to other measures such as widespreadness and symmetry. For example, a hypothetical manikin report where alternate pixels were reported as painful would have the same pain extent score as one where every pixel in the lower half of the body is marked as painful, but widespreadness measures would distinguish between these two pain drawings. All studies reporting pain location either used region-based manikins or overlaid predefined areas onto pixel-based manikins to calculate these measures.

While pain is sometimes reported as a single entity, it can be a complex system made up of different pathophysiological causes from different anatomical sites with different characteristics (eg, a constant ache from musculoskeletal pain and a shooting neuropathic pain). A small proportion of manikins captured location-specific pain quality, but most studies (62/92, 67%) only collected pain location. Asking people to report on pain in general may miss nuances regarding the specific pain components, but it has the advantage of avoiding the difficult challenge for patients in disentangling the sources of their pain, which may not be apparent to them.

Heat maps and location frequency were the most common multidrawing summary methods. Multidrawing analysis methods were less commonly used than summary methods and were mostly concerned with different ways to quantify the similarity between pairs of manikins (often for development or validation studies), the change over time in one person, or grouping similar manikins together.

There was a general lack of clarity around methodology in the literature, with many studies missing basic information such as whether a manikin was pixel or region based. It was often difficult to determine what construct was intended to be measured or whether two measures used equivalent methods. We observed many studies reporting the number they calculated for a summary measure without explaining the process in sufficient detail for it to be reproduced. The issue of missing information was present in the descriptions of the acquisition methods, as well as of the summary and analysis methods.

We are not aware of any previous work that classified manikin-derived measures by pain construct, but 2 previous reviews looked at manikin-derived summary measures. The systematic review of digital pain manikin smartphone apps by Ali et al [11] found 9 manikin-derived summary measures, substantially fewer than the 31 found by our review. Although their review only included manikins available in app stores, all measures they reported were also found by our review. They suggested that there is a need to assess the measurement properties of smartphone-based pain manikins. Our mapping of pain constructs may help future work in this area as assessing the validity of measure requires an understanding of the construct it purports to measure.

One aspect of the systematic review of methodological milestones for the development of pain manikins by Shaballout et al [7] was manikin-derived measures, which they split into “simple measures” and “topographic measures.” They highlighted the need for standardization and the difficulty of comparing results between studies, advocating for the adoption of a common body template. They noted that digital pain manikins have the potential to record more pain attributes (eg, intensity or depth) when compared to paper manikins and that they expect further development in manikin-derived measures and analysis methods in future. Our study extends their work describing manikin-derived measures with an updated search that included more recent studies and a more detailed mapping of manikin-derived measures. On the basis of our findings, we concur with their suggestion that there is a need for standardization.

A limitation of our review is that we aimed to extract detailed data on methodological aspects of included studies, whereas manikin methods were often not our focus. Consequently, many of the included studies lacked detailed information on the extracted data items. However, including a broader range of studies made our review more comprehensive than if we had restricted it to those with a methodological focus.

Another limitation is that we restricted our search to published literature and did not include gray literature or apps and software not reported in literature, so we may have missed relevant measures and analysis methods from those sources. However, as all measures found by the app review by Ali et al [11] were also identified in our review, this suggests that we managed to identify a comprehensive set of manikin-derived pain constructs. Similarly, we restricted our search to digital pain manikins and excluded paper-based manikins. There is a possibility that by excluding paper manikins, we have missed summary measures or pain constructs that were not represented in the digital manikin set, but we believe that this had limited impact on our findings and conclusion.

Finally, we were unable to investigate measurement properties of manikin-derived summary measures as originally planned in our registered protocol because the high level of heterogeneity among methods of measuring constructs did not allow a meaningful synthesis. This review focused on identifying manikin-derived pain constructs (part 2 of our registered protocol’s objective 3). All remaining objectives were addressed elsewhere [103,104].

Manikin summary and analysis methods that make better use of location-specific pain aspects (such as location-specific pain intensity) should be developed. While 33 of the 92 included studies used manikins that captured location-specific pain aspects, only 12 (13%) of the reported individual-level summary measures, 4 (4%) of the multimanikin summary methods, and 0 (0%) of the multimanikin analysis methods used these data. For clinical studies with no plan to use location-specific pain aspects, the benefit of having these data available should be weighed against the additional burden to the patient in reporting them. This is in line with the consensus statement of recommendations to address respondent burden associated with patient-reported outcome assessment by Aiyegbusi et al [105], who also highlighted the need to consider the complexity and completion time of patient-reported outcome measures, which is particularly relevant when selecting a digital pain manikin for data collection.

Summary measures should be chosen with consideration of the underlying pain construct they attempt to measure and its relevance to the disease or clinical area being studied. Summary measures inherently lose spatial information, and very few studies used analysis methods that made use of this information. Different summary measures lose different parts of the spatial information; pain extent preserves the area but not the location of the pain, and widespreadness attempts to preserve an aspect of the location but not the area. This links to our previous recommendation to only capture location-specific information when there is a specific reason to do so. We recommend first identifying the pain constructs to be measured, then selecting appropriate summary measures for that construct, and finally selecting a pain manikin from which those summary measures can be derived with minimum participant burden. Through our review, it becomes evident that new methods may need to be developed to summarize multiple constructs simultaneously.

Our results suggest a need for standardization in pain manikin measures, whether this means settling on a single manikin or developing measures that are comparable between manikins. Due to the variety of methods and manikins used, even widely reported measures were not generally comparable between studies, meaning it would not be possible to assess measurement properties or validity across digital pain manikins as a whole. The lack of standardization may also lead to confusion in the context of clinical care. For example, a pain extent of 57% on one manikin is not necessarily the same as 57% on a different manikin, which could cause issues when translating results from research to clinical practice or when seeing patients who are using various different manikins. We contrast this with the standard methods of validating and developing questionnaires, where it is widely accepted that new questionnaires should not be developed if there is an existing validated questionnaire available. We recommend that researchers consider whether there is a suitable existing manikin before developing a new manikin, with future research focusing on providing insights into the practical benefits and limitations of different manikins as tools for various clinical and research applications (eg, the differences between the ability of 2D and 3D manikins to accurately capture pain in different contexts). Efforts should also be made to standardize the reporting of manikin studies and to agree on consistent terminology, ensuring terms such as “pain extent” and “widespreadness” are used consistently within the field; our review provides a strong foundation for this standardization.

We also suggest that it is not realistic that the field will settle on one manikin, particularly as different manikins may be appropriate for different research questions, and that efforts should be focused on developing methods to compare findings across manikins. One approach to this could be defining translations of multiple different manikins to one underlying representation, so that data collected on different manikins can then be summarized and analyzed in a consistent way. This is analogous to the problem of magnetic resonance imaging (MRI) dataset harmonization, where the specific machine used to collect MRI scans makes applying machine learning techniques across datasets challenging [106]. One approach to analyzing MRI scans is building a graph representation where the links between anatomical areas are an explicit part of the data format. A graph is a mathematical concept consisting of nodes that are linked by edges. For example, in a map of a social network, nodes would represent individual user accounts and edges would represent whether those users are connected. In a graph-based manikin, the nodes would represent individual anatomical locations and would be linked by edges only if they are anatomically adjacent. Future work could use a similar strategy of developing a graph-based manikin representation to solve the problem of standardization by defining translations of multiple manikins to the same graph representation. This would require a 2-way mapping between each manikin and the graph representation, allowing conversion between any manikins for which this mapping exists. To convert data collected from manikin A to manikin B, the data would be converted from manikin A to the graph-based representation and then from the graph-based representation to manikin B. Validation of this conversion would need to carefully account for the different shapes and sizes of different manikins to ensure that measures such as pain extent are preserved. A graph-based representation would also open the door to novel summary and analysis methods making use of the additional anatomical structure encoded in the format.

In addition to an evaluation of the practical benefits and limitations of different manikins as tools for various clinical and research applications, future work could also address questions that were outside the scope of this study, including an exploration of the potential applications of artificial intelligence, computer vision, and machine learning techniques in this domain.

Our review identified a substantial number of studies that used digital pain manikins for data collection, with the majority reporting relatively simple measures and methods of summarizing pain drawings. Only a few studies went beyond summarizing to perform a direct analysis of the spatial data. The fact that information on pain location and other location-specific pain aspects (such as pain intensity or pain quality) collected through digital pain manikins was often not used in summary measures and methods suggests that the rich information available from pain drawings is currently not being fully harnessed. Future work should focus on developing more advanced summary and analysis methods that harness the spatial nature of pain drawings by better incorporating anatomical and clinical knowledge, while also improving reporting and standardization of pain constructs and methods through which they are measured. Together, this will contribute to expediting the use of digital manikins to support pain outcome measurement in both research and clinical care.