Uncovering age-related differences in communication by people with persistent pain when interacting with a pain history assessment chatbot in Australia: an exploratory mixed-methods study using a comparative analysis Authors: Elizabeth Hay, David Ireland, Andrew Claus, Tanya Rose, Jenny Strong, Daria Westerman, Mikaela Schlumpf, Pranavie Vijayakumar, Lyza Burvill, Nicole Andrews Published: 2025 ABSTRACT AbstractObjectivesThere is limited research exploring the age-related difference in communication when describing pain experiences. This project aimed to identify key differences between adolescents’, young adults’ and adults’ (i) preferred communication method, (ii) language content and (iii) lexical amount and variety when discussing their persistent pain experience using chatbot technology.DesignAn exploratory concurrent nested mixed-methods design using a comparative analysis.SettingParticipants were recruited through a convenience sampling strategy from two tertiary multidisciplinary pain centres in Australia.Participants20 adolescents, 20 young adults and 20 adults completed a pain history assessment using the Dolores application. The inclusion criteria for this study were (i) persistent non-cancer pain for at least 3 months, (ii) 10 years of age and over and (iii) adequate expressive and receptive language skills to complete the required tasks in English, as determined by clinicians in the treating team.ResultsDrawn, voice and typed responses provided by participants during the chatbot interaction were analysed using content analysis. Fisher’s exact tests and χ2 tests were used to examine differences between age groups. Strong effect size estimates generated from comparative analyses suggested that adolescents were the most likely age group to utilise drawing (p=0.153, Cramer’s V=0.268), preferred typing over speaking (p=<0.001, Cramer’s V=0.470) and used non-specific language to describe pain (p=0.007, Cramer’s V=0.433). Young adults used the greatest number and variety of words in response to chatbot questions including evaluative language (p=0.097, Cramer’s V=0.296). Adults tended to use more metaphors (p=0.085, Cramer’s V=0.305) and had a strong preference for speaking over typing (p=<0.001, Cramer’s V=0.470).ConclusionsThe results of this study provide insights into age-related differences in communication and preferences when using technology to communicate about persistent pain. Future research exploring individualised age-related approaches to pain assessment, supported by the findings of this study, in comparison to current standardised assessments administered by clinicians are warranted.Keywords: Adolescents, PAIN MANAGEMENT, Artificial Intelligence, Chronic Pain, eHealth, Natural Language Processing STRENGTHS AND LIMITATIONS OF THIS STUDY By utilising chatbot technology, we were able to investigate language and communication in a way that removed some known barriers to communication for young people. By utilising chatbot technology, we were able to investigate language and communication in a way that removed some known barriers to communication for young people. The study design allowed us to examine data generated from speech, text and drawings to gain a deeper understanding and appreciation of language and communication differences when discussing a persistent pain experience. The study design allowed us to examine data generated from speech, text and drawings to gain a deeper understanding and appreciation of language and communication differences when discussing a persistent pain experience. Obtaining effect size estimates was useful to determine potentially important associations that can inform future research directions. Obtaining effect size estimates was useful to determine potentially important associations that can inform future research directions. Due to the exploratory nature of this study, potential confounding variables were not measured or controlled for. Due to the exploratory nature of this study, potential confounding variables were not measured or controlled for. External validity is limited due to the sampling method and sample size. External validity is limited due to the sampling method and sample size. INTRODUCTION Pain that persists over 3 months or more has been reported in around 25% of children and adolescents1 and has been estimated to affect around 30% of the adult population globally.2 Persistent pain is complex, with effective assessment and management requiring understanding of the whole person and their experiences.3 While many assessments rely on rating scales to measure pain, Melzack’s research has demonstrated that “pain refers to an endless variety of qualities…not a specific, single sensation”. 4, p. 278 Age-related language differences may adversely impact the appropriateness of client-clinician interactions and understanding of pain. Young people’s pain is often assessed through observation and parent-proxy measures.5 However, an exploratory study comparing pain vocabularies of children with the language adults believed they would use discovered decreased convergence with increased child age.6 A recent study indicated that young people aged 8–12 years focus on concrete pain concepts, such as physical damage, without identifying abstract elements that are more often recognised by adults.7 Other studies have proposed that art facilitates more detailed self-expression regarding pain experiences than verbal interviews when used with people aged 8–12 years.7 8 Inherent biases that clinicians bring to communication, particularly during interview-based face-to-face assessments, can hinder the quality of the clinical encounter, diagnostic decision making and treatment planning.9,11 Biases can be particularly pronounced when a clinician has different characteristics to a client, such as age.9 In these circumstances, it is more likely that the clinician will not take into consideration the client’s preferences.9 Technology-based solutions have been proposed as one way to overcome biases and improve assessment practices.11 While technology-based healthcare solutions often offer benefits like increased access to healthcare and efficiency, there is the potential for new barriers to be introduced for subsets of the population based on factors such as socioeconomic status and previous exposure to technology.12 The use of artificial intelligence chatbots is one technology solution that is currently being propelled to the forefront of many healthcare approaches.13 Chatbots are computer systems that enable interactive two-way communication through a variety of modalities such as written language, spoken language, videos and images.14 15 Chatbots have been used for both screening and treatment in a variety of fields including mental health and chronic disease management.13 Systematic reviews have identified that health consumers find chatbots easy to use and report high levels of satisfaction with chatbot encounters.16 17 In addition, people perceive chatbots as offering a non-judgmental safe space to provide information,18 which has been found to improve the self-expression of young people.7 19 Chatbots can be used by clients of various ages, language capacity and health literacy levels.14 20 Thus, chatbots offer an innovative research tool to investigate communication differences, by removing some known barriers to communication, and could serve as a less biased way to gather information and improve self-expression in young people. As a foundation to improve clinician-client shared understanding of pain, this exploratory study aimed to identify key differences between adolescents’, young adults’ and adults’ (i) preferred communication method, (ii) language content and (iii) lexical amount and variety, when discussing their persistent pain experience using chatbot technology. In addition, the study aimed to compare the language content generated from the chatbot interaction to a language list used in an established pain assessment, the McGill Pain Questionnaire (MPQ), for convergence. METHODS Study Design This exploratory study uses a concurrent nested mixed-methods design21 involving a comparative analysis. This study design allowed the simultaneous collection of quantitative (app interface selections, responses to feedback questionnaires) and qualitative data (text, spoken language and drawings) to provide a more comprehensive understanding of pain language and communication preferences. An exploratory approach provided the flexibility to gather and analyse a rich dataset to direct more focused future research. This study is a component of a larger project that was approved by The Children’s Health Queensland Hospital and Health Service Human Research Ethics Committee (No: HREC/21/QCHQ/73157), the CSIRO Health and Medical Human Research Ethics Committee (No: 2021_049_RR), and The University of Queensland Human Research Ethics Committee (2021/HE001056). The analyses presented in this paper were planned during project conceptualisation and are not a secondary analysis of the data. The protocol for this study was not preregistered but can be provided on reasonable request to the corresponding author. Patient Involvement Patients were not involved in the design and conduct of this research beyond being active research participants. Chatbot Technology and Pain History Interview The Dolores application was used for the purpose of this project.22 23 Dolores is a pain history and education chatbot that was programmed in 2021 using an in-house framework developed at the CSIRO in Australia. This framework has been used in other health-related chatbots, including Harlie24 25 and Edna.26 Dolores utilises rule-based processing, a small evidence-based knowledge database and sentiment analysis. Rule-based processing, while now considered inferior compared to Large Language Models, provides safe responses constrained to the internal knowledge database and can run ubiquitously on any platform with no specialised hardware and requiring no internet access. Details of the development, pilot testing and acceptability of the Dolores app have been previously reported.22 23 This study reports on data obtained using the app’s Pain History Interview. The Pain History Interview is a structured interview that involved gathering details about the individual’s pain history and pain experience. Key constructs for this interview-based assessment were drawn from existing assessment proformas used by the local tertiary pain centres and the Electronic Persistent Pain Outcomes Collaboration referral questionnaire which is widely used across Australia and New Zealand.27 These constructs and the questions used to assess each construct are detailed in table 1. Table 1Chatbot interview script and relevant constructs assessed QuestionLabelConstructCould you tell me where on your body your pain is located?*Pain location1. Pain locationCould you describe or draw what your pain physically feels like?*Pain physical description2. Pain descriptionCould you tell me or draw how your pain makes you feel? Physically and mentally?*Pain feeling2. Pain descriptionAnd how long have you been in pain for?Pain duration3. Pain historyCan you tell me or draw how your pain began?Pain beginning3. Pain historyHow does your pain affect your day-to-day life?Impact4. Impact on daily lifeWhat do you believe is the cause of your pain?*Cause of persistent pain5. Pain and pain neuroscience understandingDo you think your pain will go away one day?Permanence5. Pain and pain neuroscience understandingWhat do you believe are the best treatments for managing your pain?Treatment5. Pain and pain neuroscience understandingDo you think your pain affects how you make and keep friends?†Friendship6. Impact on friendship*Drawing interface offered for responses.†Question only posed to adolescents. Chatbot interview script and relevant constructs assessed Drawing interface offered for responses. Question only posed to adolescents. The Dolores application allows the user to either speak, type or draw to communicate with the chatbot. The chatbot generates responses using a combination of speech and written outputs with a ‘chat bubble’ produced. Each question posed by the chatbot is open-ended and allows for full response freedom through typing, speaking or drawing (on a body diagram for Pain Location or free hand for Pain Physical Description, Pain Feeling and Pain Beginning). For this study, the type, quality and length of response to one question did not alter the next question posed. Participants Participants were recruited through a convenience sampling strategy from two tertiary multidisciplinary pain centres. The inclusion criteria for this study were (i) persistent non-cancer pain for at least 3 months, (ii) 10 years of age and over and (iii) adequate expressive and receptive language skills to complete the required tasks in English, as determined by clinicians in the treating team. Participants with co-existing medical conditions were not excluded from the study. All participants were receiving active treatment at one of the pain clinics and hence had some exposure to biopsychosocial treatment models and pain education. The target sample size (n=60) for this pioneering exploratory study took into consideration current guidelines proposing that appropriate qualitative interview sample size is dependent on when data saturation is reached28 29 and that previous exploratory quantitative studies in pain have found significant results with samples consisting of eight to thirty-one participants.30,32 Thus, the sample size in this mixed-methods study was predicted to allow for data saturation and provide enough data to obtain effect size estimates that could influence future research directions. Procedure Two physiotherapy research students facilitated the participant recruitment and data collection process. All participants were provided with written and verbal information about the study and written informed consent was required before participation. For younger participants (under 16 years of age), consent was obtained from both the participant and their parent or guardian. Participants were recruited from outpatient waiting rooms and could elect to participate prior to or following the conclusion of their routine appointment. In the adolescent cohort, the presence and assistance of parents or guardians was allowed throughout study engagement, as parent presence has been found to provide support and improve recall33; however, only the adolescent’s responses were recorded. Seventeen parents/guardians provided support throughout the study. Participants completed a 3 min paper-based demographic questionnaire, after which they were able to converse and interact with the chatbot for approximately 20–30 min, in a distraction-reduced environment (ie, a private clinic room). On completion of the chatbot interactive experience, participants completed a paper-based custom-made evaluation questionnaire22 which included a question about their preferred communication method (ie,“Did you prefer typing or speaking when communicating with Dolores?”). Participants received a AUD 25 voucher for use in popular retail stores in Australia as a gesture of appreciation. The chatbot software recorded all typed, spoken and drawn communication provided by the participants and the chatbot session was also recorded using a handheld audio recorder in the room. The transcript produced by the chatbot was cross-checked against the audio recording to ensure accuracy, then the recording was deleted. Analysis Preferred Communication Method Coding Interface selection, between drawing or word (typed/spoken) responses, for Pain Location, Pain Physical Description, Pain Feeling and Pain Beginning was recorded and coded as a dichotomous variable (0=no response or word response; 1=drawing). Coding of the responses to the communication preference item of the custom-made feedback questionnaire generated three codes: ‘speaking’, ‘typing’ and ‘combination of speaking and typing’. Pain Language and Content Analysis The content analysis process used to examine pain language is outlined in figure 1, with an example of the analysis process depicted in figure 2. The principles of content analysis34 were used to analyse the drawings and the word responses generated by participants for questions regarding Pain Physical Description and Pain Feeling. This involved both manifest and latent analysis.35 36 Each question and associated responses were analysed separately from the other questions. The trustworthiness of findings was established through: (i) use of a chatbot delivered structured interview script during data collection which facilitated consistent probing across participants, (ii) cross-checking chatbot generated transcripts versus audio recordings, to ensure accuracy prior to coding, (iii) clearly defining codes and themes through detailed record keeping and (iv) multiple-analyst triangulation.37 The two researchers who analysed transcripts were from different disciplines (occupational therapy and physiotherapy) and had different levels of experience, with EH being a research student and NA an experienced health professional, clinician and qualitative researcher. This allowed a greater level of critical reflexivity throughout the coding process through discussions about both professional and personal biases.37 Content analysis process. Content analysis example for drawing and word responses. Content Analysis of Drawn Responses Each drawing feature was identified as a meaning unit. Two researchers (EH, NA) independently familiarised themselves with each meaning unit, then generated initial codes for manifest content analysis. Aspects of the drawings such as colours and written words within the drawing were considered when generating codes. However, a participant’s responses to other questions posed by the chatbot were not considered when interpreting their drawing. These codes were classified as subcategories, which exactly described the meaning unit. Differences in interpretation of drawing subcategories were resolved by a third independent researcher (AC). Content Analysis of Spoken/Typed Word Responses Key words and phrases in responses generated by participants were identified as meaningful units for analysis. These units were coded into subcategories, grouped by the exact word or by synonyms if an alternative phrasing was used. Subcategories were grouped into categories that shared manifest level commonalities. Categories incorporated subcategories from both drawn and word responses. These groupings were checked by two researchers (EH, NA). Categories were further grouped into themes using a latent analysis process. This involved interpreting the overall underlying meaning behind the participant responses and was performed collaboratively by the two researchers. Quantitative Variable Formation to Establish Age-Related Differences Themes and categories were organised into dichotomous variables to measure whether the participant used language from that specific theme or category to communicate about their pain. A lack of response was coded to indicate that the participant did not use that language theme or category. To compare the lexical amount across age cohorts, analyses identified the number of words uttered by participants for responses to questions for Pain Physical Description and Pain Feeling. Four dichotomous variables were created for further analysis. Due to limited range and variance in the data, the four lexical amount variables were (1) response containing one word, (2) response containing >2 words, (3) drawn response and (4) no response. Comparison to the MPQ Subcategories in responses to identifying Pain Physical Description and Pain Feeling were also examined for similarities to the MPQ.38 The MPQ has been widely used clinically and in pain research for over 40 years.39 This standardised verbal pain assessment identifies 78 key pain descriptors. The identification of the qualitative aspects of pain, captured by the MPQ, remains an important aspect of diagnostic assessment with the pain field.39 Statistical Analysis A quantitative data analysis programme, the Statistical Package for Social Sciences (SPSS) (V.18) (SPSS Inc, Chicago, IL, USA), was used to statistically compare differences in responses across age cohorts. Arbitrary cut-off points were used for age grouping, based on what was achievable for recruitment at the study sites, as there are no universally accepted age groupings for ‘adolescent’ and ‘young adult’. Demographic data and theme/category level responses to each question were first examined using descriptive statistics. The frequency of pain descriptors used by participants that aligned with descriptors included in the MPQ was also examined. Fisher’s exact test was used to examine differences across the three age categories for preferred communication method (ie, interface selection), lexical amount, comparison to the MPQ and themes/categories identified in responses to the two established key questions. A χ2 test was used to examine the association between age group and communication preferences from the custom-made feedback questionnaire. Missing questionnaire data resulted in that case being excluded from the analysis. Missing data were minimal and were considered missing completely at random. Hence, a sensitivity analysis was not applied. Statistical significance was identified as p<0.05 for all analyses. Phi and Cramer’s V tests were produced to assess the strength of the association between two discrete variables. Cramer’s V>0.25 indicated that one variable was very strongly determined by another, while Cramer’s V>0.15 indicated a strong association.40 Both very strong and strong associations were considered to be potentially clinically significant. RESULTS Results are presented in tables 2–6 and explained below. All analyses presented met required assumptions for Fischer’s exact test and χ2 test. Table 2Characteristics of the study populationsDemographic variableTotal% (n=59)Adults% (n=20)Young adults% (n=19)Adolescents% (n=20)Gender Male44.1 (26)50 (10)52.6 (10)30 (6) Female55.9 (33)50 (10)47.4 (9)70 (14)Age (years)*32.8854.628.7915.05(10–76)(38–76)(21–35)(10–18)Country of birth Australia91.5 (54)80 (16)100 (19)95 (19) Egypt1.7 (1)5 (1)00 England1.7 (1)5 (1)00 New Zealand1.7 (1)5 (1)00 The Netherlands1.7 (1)5 (1)00 Germany1.7 (1)005 (1)Languages spoken English only91.5 (54)80 (16)100 (19)95 (19) More than one language (including English)8.5 (5)20 (4)05 (1)Main language English98.3 (59)95 (19)100 (19)100 (20) Arabic1.7 (1)5 (1)00Education level University18.6 (11)30 (6)45.5 (5)0 Diploma10.2 (6)20 (4)10.5 (2)0 Certificate22 (13)25 (5)31.6 (6)10 (2) Secondary education33.9 (20)25 (5)31.6 (6)45 (9) Primary education15.3 (9)0045 (9)Co-existing medical conditions† Dyslexia5.1 (3)5 (1)010 (2) Vision impairment3.4 (2)10 (2)05 (1) Behaviour/attention disorder6.8 (4)010.5 (2)10 (2) Autism spectrum disorder3.4 (2)05.3 (1)5 (1)Pain cause Injury at home5.1 (3)5 (1)5.3 (1)5 (1) Injury at work/school15.3 (9)15 (3)21.1 (4)10 (2) Injury in another setting13.6 (8)15 (3)21.1 (4)5 (1) Road traffic crash8.5 (5)15 (3)5.3 (1)5 (1) Medical condition30.5 (18)40 (8)26.3 (5)25 (5) After surgery8.5 (5)5 (1)10.5 (2)10 (2) No obvious cause18.6 (11)5 (1)10.5 (2)40 (8)Pain duration (years)*8.09(0.67–40)15.00(0.67–40)6.05(1–20.5)3.12(0.67–14.5)Pain location Head/face20.3 (12)25 (5)10.5 (2)25 (5) Lower back50.8 (30)60 (12)52.6 (10)40 (8) Upper limb42.4 (25)45 (9)31.6 (6)50 (10) Abdomen/groin25.4 (15)20 (4)21.1 (4)35 (7) Hips28.8 (17)40 (8)26.3 (5)20 (4) Knees25.4 (15)35 (7)15.8 (3)25 (5) Neck39 (23)50 (10)26.3 (5)40 (8) Upper back27.1 (16)15 (3)31.6 (6)35 (7) Chest11.9 (7)10 (2)10.5 (2)15 (3) Buttocks6.8 (4)15 (3)5.3 (1)0 Thighs11.9 (7)15 (3)10.5 (2)10 (2) Calves/ankle/feet20.3 (12)25 (5)15.8 (3)20 (4) Total body3.4 (2)5 (1)5.3 (1)0Confidence using appsNot at all confident0000 Slightly confident1.7 (1)5 (1)00 Moderately confident15.3 (9)40 (8)5.3 (1)0 Confident33.9 (20)40 (8)36.8 (7)25 (5) Very confident49.2 (29)15 (3)57.9 (11)75 (15)Hours using smartphone*25.77(0–70)‡20.58(2–56)§28.68(3–70)27.93(0–70)Note: values = % (n).*Values=mean (range).†Co-existing medical conditions that were response options but not selected by any of the participants included language delay/disorder (eg, aphasia) /difficulty, a speech difficulty (eg, dysarthria/dyspraxia), intellectual impairment, cerebral palsy, a global developmental delay, a genetic syndrome (eg, down syndrome) and a significant hearing impairment.‡One adult non-respondent, thus n=58.§One adult non-respondent, thus n=19. Characteristics of the study populations Note: values = % (n). Values=mean (range). Co-existing medical conditions that were response options but not selected by any of the participants included language delay/disorder (eg, aphasia) /difficulty, a speech difficulty (eg, dysarthria/dyspraxia), intellectual impairment, cerebral palsy, a global developmental delay, a genetic syndrome (eg, down syndrome) and a significant hearing impairment. One adult non-respondent, thus n=58. One adult non-respondent, thus n=19. Participant Characteristics and Chatbot Engagement There were 20 participants recruited in each age cohort, with 60 participants in total. Demographic information was available for 59 of the participants and is reported in table 2. The mean (range) age in the adult cohort was 54.6 (38–76) years, in young adults was 28.8 (21–35) years and in adolescents was 15.1 (10–18) years. Young adults tended to have the highest level of education, with 45.5% holding university level qualifications. The most common pain location was the lower back for adults (60%) and young adults (52.6%). Upper limb was the most common pain location for adolescents (50%). Differences in pain duration corresponded to participant age. Adolescents had the most co-existing medical conditions affecting language and communication. Confidence with technology was high across age groups, with all but one participant reporting being at least ‘moderately confident’ with using smartphone apps. Young adults spent the longest duration of time on their smartphones, at a mean (range) of 28.68 hours (3-70) per week compared with 20.58 hours (2-56) for adults and 27.93 hours (0–70) for adolescents. There were three participants who did not respond to specific questions during the chatbot pain history interview. Two participants, one adult and one adolescent, did not respond to Pain Physical Description, and one adult did not provide a response to Pain Feeling. All young adults responded to all questions posed. Preferred Communication Method Of the 60 participants, 68.3% chose to draw in response to at least one question where the drawing interface was offered (see table 3). There was a very strong association that did not reach statistical significance (p=0.153, Cramer’s V=0.268) between age and using the drawing widget, as adolescents used drawing the most to respond to at least one question, with 85% using the drawing interface, compared with 65% of young adults and 55% of adults. Table 3Preferred communication method for questions that provided optional drawing widget responsesInterface selection variableTotal% (n=60)Adults% (n=20)Young adults% (n=20)Adolescents% (n=20)Pain location Drew68.3 (41)55 (11)65 (13)85 (17) Word31.7 (19)45 (9)35 (7)15 (3)Pain physical description Drew5 (3)5 (1)010 (2) Word91.7 (55)90 (18)085 (17) No response3.3 (2)5 (1)05 (1)Pain feeling Drew11.7 (7)5 (1)15 (3)15 (3) Word86.7 (52)90 (18)85 (17)85 (17) No response1.7 (1)5 (1)00Pain beginning Drew5 (3)0015 (3) Word95 (57)100 (20)100 (20)85 (17)≥One drawing Drew68.3 (41)55 (11)65 (13)85 (17) Did not draw31.7 (19)45 (9)35 (7)15 (3)≥One drawing excluding location Drew13.3 (8)5 (1)15 (3)20 (4) Did not draw86.7 (52)95 (19)85 (17)80 (16)Note: values = % (n). Preferred communication method for questions that provided optional drawing widget responses Note: values = % (n). Pain Location was the question where the drawing interface was most used. This interface was most used by adolescents compared with other age cohorts, and only adolescents chose to draw in response to Pain Beginning. A very strong and statistically significant association was found between age group and preferred communication method based on the 54 responses in the custom-made feedback questionnaire (p=<0.001, Cramer’s V=0.470). As displayed in table 4, adolescent participants had a strong preference for typing to communicate with the chatbot, with 88.2% of adolescents choosing this communication method over speaking. In contrast, the majority of adults (78.9%) and young adults (55.5%) preferred to speak to the chatbot to communicate. Table 4Preferred communication method based on responses to custom-made feedback questionnaireCommunication preferenceTotal% (n=54)Adults% (n=19)Young adults% (n=18)Adolescents% (n=17)Speaking48.1 (26)78.9 (15)55.5 (10)5.9 (1)Typing42.6 (23)10.5 (2)33.3 (6)88.2 (15)Combination of speaking and typing9.2 (5)10.5 (2)11.1 (2)5.9 (1)Note: values = % (n). Preferred communication method based on responses to custom-made feedback questionnaire Note: values = % (n). Pain Language Content Pain Physical Description Four themes were identified in participant responses when describing the physical sensation of pain: Non-specific—use of non-specific descriptors that do not establish differences between pain experiences such as “painful” provided by adult participant 12 (A12) or “tender” (adolescent participant 12 (T12)). Non-specific—use of non-specific descriptors that do not establish differences between pain experiences such as “painful” provided by adult participant 12 (A12) or “tender” (adolescent participant 12 (T12)). Physical/sensory—recognition of specific qualities that separate the current pain from other sensations or experiences. This involved identifying aspects like pressure, temperature, sensation quality and sharpness, such as “burning” (A3), ‘“stabbing” (young adult participant 2 (YA2)) or “itching” (A14). Physical/sensory—recognition of specific qualities that separate the current pain from other sensations or experiences. This involved identifying aspects like pressure, temperature, sensation quality and sharpness, such as “burning” (A3), ‘“stabbing” (young adult participant 2 (YA2)) or “itching” (A14). Evaluative—use of language providing a value to the pain experience and involving a judgement of the pain. This included reference to the duration of the pain, such as “constant” (YA1) or “variable” (YA15), alongside identifying the location of the pain, and the pain intensity using words such as “dull” (A1) or “strong” (A6). Evaluative—use of language providing a value to the pain experience and involving a judgement of the pain. This included reference to the duration of the pain, such as “constant” (YA1) or “variable” (YA15), alongside identifying the location of the pain, and the pain intensity using words such as “dull” (A1) or “strong” (A6). Metaphor—use of figurative language, such as referring to the pain as “like my brain has been set on fire” (T14) or “Chinese burns” (YA9). Metaphor—use of figurative language, such as referring to the pain as “like my brain has been set on fire” (T14) or “Chinese burns” (YA9). There was a very strong and statistically significant relationship between age and using language included in the Non-specific theme (p=0.007, Cramer’s V=0.433). This theme was referenced by 60% of adolescents, with only 15% of young adults and 20% of adults referring to this theme (see table 5). Table 5Frequency counts for themes across the sampleQuestionThemeTotal%(n=60)Adults%(n=20)Young adults% (n=20)Adolescents% (n=20)Pain physical descriptionNon-specific*31.7 (19)20 (4)15 (3)60 (12)Physical/sensory66.7 (40)65 (13)70 (14)65 (13)Evaluative45 (27)40 (8)65 (13)30 (6)Metaphor18.3 (11)35 (7)10 (2)10 (2)Pain feelingNegative emotions and emotional reaction73.3 (44)65 (13)70 (14)85 (17)Tired and fatigued48.3 (29)40 (8)50 (10)55 (11)Non-specific28.3 (17)30 (6)20 (4)35 (7)Reduced function26.7 (16)40 (8)15 (3)25 (5)Minimal impact3.3 (2)05 (1)5 (1)*Significant at p<0.05 level. Frequency counts for themes across the sample Significant at p<0.05 level. Metaphor use showed a very strong association with age but failed to reach statistical significance (p=0.085, Cramer’s V=0.305). Responses within the Metaphor theme were provided by 35% of adults compared with 10% of both young adults and adolescents. The most common motif, expressed by seven of the eleven participants who used a metaphor, involved describing the pain as an external force inflicted on the participant, such as “like somebody twisting a knife” (A7). Three others provided metaphors describing pain as a weight, such as “like a heavy brick” (A11). While the relationship between age and recognition of the Evaluative concept was very strong, it was not statistically significant (p=0.097, Cramer’s V=0.296). The Evaluative theme was most identified by young adults (65%) and was least referenced by adolescents (30%). Sharp (eg, “stabbing” (A8)) was the most popular language category overall, referenced in 51.7% of responses across all age cohorts. Consequently, this was the most popular category for adults (45%) and young adults (60%), though not for adolescents. Adolescents more often used a descriptor in the Non-specific category (eg, “ache” (T3); 60%). Pain Feeling Five themes were identified in responses to express the mental and physical impact of pain (ie, feeling): Negative emotions and emotional reaction—referencing emotions surrounding discomfort and sadness associated with the pain, such as “confused” (T4), “defeated” (YA10) and “depressed” (YA4), alongside expressions of anger or annoyance because of the pain, such as “ready to snap” (A15) and “frustrating” (A5). Negative emotions and emotional reaction—referencing emotions surrounding discomfort and sadness associated with the pain, such as “confused” (T4), “defeated” (YA10) and “depressed” (YA4), alongside expressions of anger or annoyance because of the pain, such as “ready to snap” (A15) and “frustrating” (A5). Tired and fatigued—identifying a feeling of exhaustion as a consequence of the pain experience, such as “draining” (YA1) and “lethargic” (A1). Tired and fatigued—identifying a feeling of exhaustion as a consequence of the pain experience, such as “draining” (YA1) and “lethargic” (A1). Non-specific—use of generalised descriptors associated with pain that do not refer to a specific emotion or consequence, such as “uncomfortable” (YA10) or “sore” (T6). Non-specific—use of generalised descriptors associated with pain that do not refer to a specific emotion or consequence, such as “uncomfortable” (YA10) or “sore” (T6). Reduced function—referring to limitations on a participant’s life or perceptions of oneself as being impaired, such as “unable to complete activities” (A7) and “not like a whole person” (YA6). Reduced function—referring to limitations on a participant’s life or perceptions of oneself as being impaired, such as “unable to complete activities” (A7) and “not like a whole person” (YA6). Minimal impact—establishing that the pain experience did not impact on the participant’s life significantly, such as “coping” (YA17) or “mentally not really affected” (T9). Minimal impact—establishing that the pain experience did not impact on the participant’s life significantly, such as “coping” (YA17) or “mentally not really affected” (T9). Negative Emotions and Emotional Reactions was the most prominent theme for each age cohort, referenced in 65%, 70% and 85% of adult, young adult and adolescent responses respectively. This led to a strong but statistically insignificant relationship between the theme of Negative Emotions and Emotional Reactions and age (p=0.434 Cramer’s V=0.192). There was a strong association between age and responses acknowledging the theme of Reduced Function, that failed to reach statistical significance (p=0.237, Cramer’s V=0.232). Reduced Function was referred to the most by the adult cohort. No adult responses were included in the category of being Over it (eg, “don’t want to have to deal with it” (T11)) and the category of Unhappy (eg, “upset” (T12)) was referenced by adolescent (45%) and younger adults (35%) more than adults (15%). Lexical Amount and Variety The differences regarding number of words used to describe pain (ie, Pain Physical Description and Pain Feeling) across the age cohorts are displayed in table 6. Young adults provided the most words to physically describe pain, with 75% of the cohort using more than one word compared with 65% of adults and 50% of adolescents. This was a statistically insignificant, but a strong association (p=0.545, Cramer’s V=0.163). Table 6Number of words used when responding to questions about pain descriptionLexical amountTotal% (n=60)Adults% n=20)Young adults% (n=20)Adolescents% (n=20)Pain physical description One word16.7 (10)15 (3)15 (3)20 (4) >One word63.3 (38)65 (13)75 (15)50 (10) Drew16.7 (10)15 (3)10 (2)25 (5) No response3.3 (2)5 (1)0 (0)5 (1)Pain feeling One word18.3 (11)20 (4)15 (3)20 (4) >One word61.7 (37)65 (13)70 (14)50 (10) Drew18.3 (11)15 (3)15 (3)30 (6) No response5 (1)5 (1)0 (0)0 (0)Note: values = % (n). Number of words used when responding to questions about pain description Note: values = % (n). Young adults also incorporated the greatest variety of language categories for Pain Feeling. Young adults’ language covered all 12 identified categories for Pain Feeling compared with only nine categories that captured the language used by adolescents and adults. Comparison to the MPQ There were 114 total utterances that matched MPQ descriptors. A total of 80% of participants used at least one of the MPQ descriptors; however, only 33.33% of the 78 potential descriptors were used. Adolescents were most likely to use at least one descriptor, with 85% of adolescents referencing at least one MPQ descriptor in a response. There were 108 words provided by 88.33% of participants in response to Pain Feeling that were not represented in the MPQ. In their responses, young adults uttered the most words (40) that were not represented by the MPQ. Words used in response to Pain Physical Description were more effectively captured by the MPQ, with only 65 words uttered by participants not captured. Adolescents demonstrated the least language variation, providing only 17 words not captured by the MPQ when responding to Pain Physical Description, compared with 25 words for adults and 23 words for young adults. While also not statistically significant, there was a very strong relationship between age and use of at least one descriptor not included in the MPQ for Pain Physical Description responses (p=0.133, Cramer’s V=0.277). 85% of the young adults used at least one word not captured by the MPQ compared with 75% of adults and 55% of adolescents. DISCUSSION This mixed-methods study investigating pain communication using chatbot technology identified key age-related variations in communication response preferences and pain language when discussing persistent pain experiences. Recognising and understanding these differences provides new insight toward age-specific communication approaches for more effective persistent pain assessment and management. Adolescent Communication Preference for drawing was strongly associated with younger age, though this was not statistically significant. Adolescents were the most likely to draw, and some used this interface where drawing is not typically considered during current assessments (eg, Pain Beginning).41 42 Adolescents also showed a strong and statistically significant preference for typing over speaking compared with the other age groups. These results may reflect how adolescents prefer to use technology. As previously reported, adolescents were more likely to want mute settings as an added feature of the Dolores app.22 Alternatively, results may be attributed to the drawing and writing process providing more time to produce responses, recall memories and organise ideas to form a practical persistent pain representation.7 This complex thought process is more difficult for paediatric cohorts compared with more cognitively developed older cohorts.43 In this study, there was a significant association between age and use of non-specific pain descriptors, with adolescents the most likely to use non-specific language. This supports the notion that adolescents may have more difficulties describing their persistent pain experience compared with older cohorts. Previous systematic reviews have established that from approximately 4 years old, even young children have sufficient language capacity for adequate clinical pain expression.44 A previous Canadian qualitative descriptive study involving 10–18 year-olds with neuropathic pain found that these adolescents mostly used more specific pain descriptors (eg, ‘pressure’, ‘sharp’).33 While some adolescents in the current study did use specific pain descriptors, the language used was more comparable to the non-specific language adopted by 4–11 year-olds during pain interviews.45 Differences in data collection methods may account for the differences between the current study and the Canadian research findings, with the Canadian qualitative interview guide including more probing questions (eg,“How would you describe these feelings to someone who hasn’t had them before?”). Taken together, findings suggest that individualised, uncomplicated, focused questions with adequate probing, combined with drawings and writing/typing, may more effectively yield clinically useful, rich expressions in adolescent pain experiences. Young Adult Communication Although findings were not statistically significant, young adults tended to use more words to respond to questions, use more evaluative language when describing their pain and use a greater variety of language categories compared with other age cohorts. Young adults were also the only cohort to respond to all questions. Thus, it could be argued that young adults were more expressive and engaged better with the chatbot compared with the other age groups. Minimal literature has investigated young adult language as studies focusing on younger populations commonly combine young adults and adolescents.46 47 The young adults who participated in this study spent more time using smartphones compared with other age groups, and prior literature has determined they regularly use technology for self-expression and to investigate health concerns.46,48 In this study, young adults tended to be the most educated, thus lexical variety could also be related to greater literacy with education a practical determinant of vocabulary.49 Adult Communication Adults provided the most metaphors resulting in a very strong association between age and metaphor use. Adults tended to provide vivid descriptions that conveyed the severity of their pain such as “boiling water on the top of my skin” and “someone shoving a pencil into my abdomen”. Previous studies have suggested that adults may use these powerful descriptions as a way of both coming to terms with their own pain experience and helping others to understand their personal experience.50 51 In the current study, the adult cohort had been experiencing pain for the longest period of time. It is possible that they may have had greater exposure to a biomedical model of care and negative prior experiences with health professionals, due to their longer pain durations, impacting on the results of this study. Adults were the least likely to draw to communicate their pain experience and had a strong preference for speaking over typing. Communication preference may have been influenced by the adult cohort’s generally longstanding pain durations contributing to greater experience and consequent comfort with current verbal pain assessment methods. The chatbot context may have also influenced adult responses, as practical determinants, including dexterity and technological confidence, regulate chatbot interface choices,14 and smartphone confidence was lowest for the adults in this study. Overall Communication The variation in language used in responses indicates there are limitations to current standardised pain assessment that rely on language lists. The MPQ language list is still widely used within the pain field, but the assessment was unable to adequately capture language across all cohorts. A total of 82 descriptors used by participants in the current study were not included in the MPQ. These findings extend previous research that revealed the MPQ did not capture 31 descriptors used by adults when describing a past pain event.52 This is an important consideration when formulating new pain assessments including technology-based assessments delivered by chatbots. Limitations The results of this study should be interpreted with several caveats in mind. The findings would be most appropriately applied to Australian persistent pain cohorts seeking treatment at metropolitan tertiary pain centres. The convenience sampling method did reduce sampling control, increasing risk of selection bias and sampling error for a non-representative sample. Additionally, the study was underpowered for statistical comparisons, so that although there were findings with clinically relevant Cramer’s V and strong effect sizes, many did not reach statistical significance. However, obtaining effect size estimates was useful to determine potentially important associations to inform future research directions. Due to the exploratory nature of this study, potential confounding variables were not measured or controlled for. Factors such as pain diagnoses and exposure to certain treatments may be important variables to consider in future studies. Future Research Additional empirical tests to determine if the individualised age-related approaches, supported by the findings of this study and prior research, have merit for age cohorts in comparison to current standardised assessments administered by clinicians are warranted. For example, the responses provided during a digital assessment based on the draw, write, tell method53 could be compared with responses provided using a standardised interview script delivered by a clinician for adolescents. In addition, this study did not include any participants with language difficulties that required communication supports to participate. The utility of chatbot technology to aid with communicating with people with pain and language difficulties is an avenue for future research. Conclusions and Clinical Implications This study found key differences in communication that suggest different approaches to pain assessment and treatment are required. Results support incorporating drawing options into pain interviews, especially for adolescents. Metaphors and spoken narratives should be considered when communicating with adults. While it is currently unclear how chatbots can best be used within healthcare settings, technology-assisted approaches to pain assessments may best facilitate young adult and adolescent communication and engagement. For example, gathering pertinent information about pain using chatbot technology, prior to an assessment with a clinician, may enhance communication during the clinical encounter and the clinician-client shared understanding of pain. Acknowledgements The authors thank the staff and patients of Tess Cramond Pain and Research Centre and the Queensland Children’s Hospital Persistent Pain Outpatient Service for their contribution to data collection. Footnotes Funding: The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors. Prepublication history for this paper is available online. To view these files, please visit the journal online (https://doi.org/10.1136/bmjopen-2025-104984). Data availability free text: Data related to the present work will be available upon reasonable request to the corresponding author. Patient consent for publication: Not applicable. Ethics approval: This study involves human participants and was approved by The Children’s Health Queensland Hospital and Health Service Human Research Ethics Committee (No: HREC/21/QCHQ/73157), the CSIRO Health and Medical Human Research Ethics Committee (No: 2021_049_RR) and The University of Queensland Human Research Ethics Committee (2021/HE001056). Participants gave informed consent to participate in the study before taking part. Provenance and peer review: Not commissioned; externally peer reviewed. Patient and public involvement: Patients and/or the public were not involved in the design, or conduct, or reporting or dissemination plans of this research. Data availability statement Data are available upon reasonable request. References Data Availability Statement Data are available upon reasonable request.