Table 1.
Technology themes, study types, and main findings.
| Reference and project or intervention name |
Health topic / study population | Technology themes | Type of study | Main findings | Evidence summary (if applicable)a |
| Ananthanarayan & Siek 2010 [27] (HealthSense) | Obesity / children | Wearable computing, “6th sense;” actionable feedback. | Design of prototype to support children’s motivation for exercise and for self-monitoring. | Not an empirical study. | Not applicable |
| Arteaga et al 2009 [28] | Obesity / teenagers | Motivational agent (mobile phone games). | Design of prototype to motivate exercise based on personality type. | Not an empirical study. | Not applicable |
| Bickmore & Picard 2005 [29] (FitTrack) | Physical activity / healthy adults | ECAb: relational agent. | RCTc (n = 101; 30 days). Measures: acceptability (self-report) + PAd (pedometer). Groups: relational agent, nonrelational agent and control. | Positive acceptance; increased PA during intervention but reduced PA after follow-up. Relational agent more liked. Dialog repetitiveness annoying. | 4: +/– |
| Bickmore et al 2005 [30] (FitTrack) | PA / older adults | ECA: relational agent. | RCT (n = 21; 2 months), to test acceptability (usage history) + PA (pedometer) + loneliness (self-report). Groups: relational vs control (usual care). | Positive acceptance and significant increase in PA during intervention. No significant decrease in loneliness. | 4: +/– |
| Bickmore & Sidner 2006 [31] | General behavior change / adults | Making dialog more robust by linking with ontologies for behavior change theories (TTMe, MIf). | Prototype. | Not an empirical study. | Not applicable |
| Bickmore et al 2009 [32] | Physical activity / adults (male students) | Context awareness of mobile PA monitor + ECA (relational agent). | Pilot study (n = 8): test whether agent context awareness promotes social bonding (acceptance). Effectiveness: does it promote walking? | Some positive acceptance but less actual walking in context-aware condition. | 1: +/– |
| Bickmore et al 2009 [33] | Compliance / low health-literacy patients (hospital discharge) | ECA: virtual nurse with relational behaviors and empathy. | Self-report usability tests: 2 tests: nonpatients (n = 30) + patients (n = 19) with 47% low literacy. Both groups tested with relational vs nonrelational agent. | Both tests: relational preferred. Overall ECA acceptance. ECA allows more time and sense of control than human face-to-face communication. | 1: + |
| Bickmore et al 2010 [34] | Medication adherence, PA / schizophrenia patients | ECA: simple concrete communication. Authors counter ethical criticism of ECA for mental health. | Pilot evaluation (n = 20; 31 days) to test acceptability (self-report) + adherence + PA (no control). | Positive acceptance. Adherence + PA high. ECA provides simplified conversation, less confusing than human face-to-face. | 1: + |
| Bickmore et al 2011 [35] | 2 domains: exercise and diet / adults | Semantic ontology for behaviors and theories (TTM, MI); semantic models of user, data, and intervention. | Qualitative study (n = 8) on acceptability of ECA health counselor based on reusable ontology. | Positive acceptance, but limited evaluation. | 1: + |
| Bickmore et al 2010 [36] | Physical activity / adults | ECA: promoting long-term use; avoid repetitive dialog. Introduce variability + storytelling. | 2 RCTs: 1. Variability (n = 24, 100 days); variable vs nonvariable; 2. Story(n = 26, 30 days): first-person story vs third-person story. Measures: usage + step count + self-reported satisfaction. | 1. Variability: more system usage, but less exercise. 2. Story: first person had more usage than third person, but less exercise. Self-reported satisfaction high for test conditions. | 4: +/– |
| Bieber et al 2009 [37] (DiaTrace) | Physical activity / adults | Mobile phone as sensor for activities and calorie estimate. | Prototype. | Not an empirical study. | Not applicable |
| Buttussi & Chittaro 2008 [38] (MOPET) | Physical activity / adults | ECA; context-aware sensing; user model. | Prototype. | Initial qualitative evaluation positive (n = 12). | 1: + |
| Consolvo et al 2008 [39] (UbiFit) | Physical activity / adults | Graphic display with “garden” metaphor; mobile sensing device with inference; interactive app (edit or add to journal). | RCT: 3-month field experiment (n = 28): full system (10) vs no mobile sensing device (9) vs no display (9). Measures: (1) sensed activities + self-report; (2) qualitative analysis on user experience. | System with display led to more exercise than without display. User experience positive: more self-awareness, which motivated exercise. | 4: + |
| De Rosis et al 2006 [40] | Diet / adults | ECA: recognize user’s emotional state, social attitude, and TTM stage during dialog; dynamically update user model during dialog. | Prototype: raters label emotional states, TTM stages, and social attitudes in test dialogs (WOZg and corpus). | Labeling of emotions by raters used to guide design of dialog system. | Not an intervention evaluation |
| Farzanfar et al 2007 [41] | Treatment adherence, suicide prevention / depressed adults | Telephone agent: monitoring + self-care management. | Preliminary qualitative trial (n = 15), 4 weeks. Modifications made in response. | Dialog was helpful for adherence, but sounded artificial and insensitive, particularly in suicide risk. Users prefer more human-like agent with empathy and understanding of serious issues. (For suicide, hotline preferred). Authors’ conclusion: anthropomorphism is not valid (people do not attribute human qualities to machines—only in metaphor). | 1: +/– |
| Hakulinen et al 2008 [42] (COMPANIONS project) | PA / adults | Mobile companion; semantic ontology of user environment for PA planning. | Prototype. | Not an empirical study. | Not applicable |
| Hartmann et al 2007 [43] | Improve patient questions to physician / adults with asthma | Educational website to suggest questions, encourage patient involvement in care, prevent more serious illness. | Pilot study: (n = 37) record usage experiences. | Positive self-report: (1) improved relations, (2) more active involvement. | 1: + |
| Hayes et al 2009 [44] | Medication adherence / older adults | Instrumented pillbox, home sensors. | Pilot study (n = 10): effectiveness of context awareness on adherence. Test phases (same group): no-prompt, time-based, context-aware prompt. | Initial evaluation: positive for context-aware phase. | 1: + |
| Jin 2010 [45] | Stress management / college students | Education-entertainment / health belief, self-efficacy; educational interactive test (game) for responses to stress scenarios. Agent gives educational messages. | RCT: (n = 60). Effectiveness of virtual agent on student’s intent/mood. Interactive test with agent (test) vs no agent (control) vs no test (true control). | Positive self-report on enjoyment and educational value for agent condition. Interactive test improves stress management self-efficacy (over true control, without test). | 2: + |
| Kaipainen et al 2011 [46] (HealthPGS) | General health decisions / adults | Health Personal Guidance System: guide user through day-to-day choices in ecosystem. Virtual individual: maintains user profile and context; HealthGuide: planning, context-aware messages. Personal Guidance System Mall: services all in one place. | Prototype. | Not an empirical study. | Not applicable |
| Klein et al 2011 [47] (eMate) | Adherence / diabetic patients | Automated reasoning based on COMBIh model (combines different theories) ensures dynamic tailored messages depending on user’s context and state of mind. | Prototype: computational model of behavior change (mobile + website). | Not an empirical study. | Not applicable |
| Konovalov et al 2010 [48] (GATE) | Mental health promotion / military service personnel | Blog analysis to understand moods and emotions (combat experience): GATE algorithm + ontology. | Design and pilot study for technology: compare algorithm with expert opinion. | Precision of algorithm: 0.9, recall: 0.75; F score: 0.82. | Not an intervention evaluation. |
| Lee et al 2010 [49] | Health promotion / older adults | Telehealth: action-based behavior model (1) increase user’s awareness of health, (2) set goals, (3) educate user in how to achieve goal, (4) remind, (5) reward + assess. | Design: overcome limits of sensing only; need high-level assessment information with models of persuasion to determine whether behavior changed. | Not an empirical study. | Not applicable |
| Levin & Levin 2006 [50] | Pain management / adults | Ecological momentary assessment, detect unexpected errors in dialog. | Feasibility study: evaluate interactive voice response system dialog for health and behavior monitoring. Feasibility study for pain monitoring voice diary (n = 24). 171 dialog sessions. | Accuracy of voice recognition: 98%. Dialog efficiency increased with user experience. | 1: + |
| Lisetti & Wagner 2008 [51] (ABLE) | Mental health promotion / adolescent | ECA companions. | Design: ECA companion to act as MI counselor. | Not an empirical study. | Not applicable |
| Looije et al 2010 [52] (SuperAssist) | Adherence / older adults | ECA (robot cat), MI, persuasion. | Pilot study (n = 24): physical ECA (n = 12) vs virtual (n = 12). Each group experienced text, social ECA, and nonsocial ECA. | 90% acceptance. Social ECA preferred over nonsocial ECA; half preferred text interface over social ECA (“conscientious” personality type). Virtual ECA more “empathic” than physical. | 1: + |
| Maier et al 2010 [53] (SEMPER) | Work-related disorders and alcohol / adults | Semantic Web portal; semantic search. | Prototype. | Not an empirical study. | Not applicable |
| Mazzotta et al 2007 [54] (PORTIA) | Healthy eating / adults | Persuasion agent: tailoring of messages based on inferred personality traits and likely motivations of user. | Prototype of dialog design, based on corpus analysis of persuasive dialogs produced by participants in role-playing scenarios. | Corpus analysis found that persuasion is most often based on nonrational arguments and positive framing. | Not an intervention evaluation |
| Munguia Tapia 2008 [55] | Obesity / adults | Sensors and algorithms for activity recognition and energy expenditure estimate. | Prototype. | Activity recognition most accurate if simple examples are given; high variability is difficult (eg, housework). Energy estimate more accurate for simple activities and with multiple body sensors. | Not an intervention evaluation |
| Nguyen & Masthoff 2008 [56] | General behavior change / adults | Persuasive dialog, MI. | Acceptability test (n = 41): is MI dialog more persuasive than argumentation? Questionnaire + qualitative analysis in comments. | Self-report positive: persuasiveness, likeability scores higher for MI than for 2 types of argumentation. | 1: + |
| Oddsson et al 2009 [57] (SKOTEE) | Adherence / adults | Robotic assistance for intelligent reminding and companionship. | Design. | Not an empirical study. | Not applicable |
| Op den Akker et al 2011 [58] | PA / adults | Software agent for smart phone: use machine learning to develop user model. Tailor messages to user history and current context. | Prototype. | Not an empirical study. | Not applicable |
| Rojas-Barahona & Giorgino 2009 [59] (AdaRTE) | General behavior change / adults | Framework for health dialog. | Design. | Not an empirical study. | Not applicable |
| Smith et al 2008 [60] | Healthy lifestyle / adults | ECA; collaborative planning, update planned activities through ongoing dialog. | Prototype with technical evaluation. | Approach is feasible, although dialog error rate is still high. | Not applicable |
| Sorbi et al 2007 [61] | Migraine attack prevention / adults | PDAi + coaching. Response behaviors to precursors of migraine. Ecological momentary intervention experience sampling: randomized calls overcome memory bias. Tailored messages depending on current experience. | Pilot study (n = 5): feasibility and user acceptance. | Positive acceptance but too many calls are annoying. Technical problems: data loss due to buildings. | 1: +/– |
| Spring et al 2010 [62] (Make Better Choices–MBC) | Obesity / adults | PDA: find optimal advice for exercise; goal thermometers; “in the moment” decision support/multiple theories, including self-regulation.Study design. | Study design. | Design of a trial only. | Not applicable |
| Tiwari et al 2011 [63] | Adherence / older adults | Robotic assistance, dialog. | Prototype development using grounded theory participatory design. | Emerging themes: usability, empowerment, collaboration, and safety: used as requirements for dialog design. | Not applicable |
| Turunen et al 2011 [64] (COMPANIONS project) | Health and fitness / adults | Home and mobile health and fitness companion. | Pilot study (n = 20): feasibility of complex dialogues in home and mobile scenarios. | System behaves robustly in realistic experimental scenarios, but word error rates are still high. | 1 + |
| Uribe et al 2011 [65] | Adherence general | Reminders based on inferred mental state; user modeling using ontologies. | Prototype. | Not an empirical study. | Not applicable |
| van der Putten et al 2011 [66] (SERA projet–Social Engagement with Robots and Agents) | PA / older adults | Social robot; health advisor. | Pilot study(n = 6). Video recording of interactions in homes. Iteratively modify setup based on results of previous session. | 3 iterations, variable interactions, and satisfaction reports. Positive for motivation but some frustration over lack of control of dialog and too much time taken up. | 1: +/– |
| Watson et al 2012 [67] | PA / overweight adults | ECA: virtual coach | RCT (n = 70; 12 weeks); primary measure: step count; secondary: weight + self-reported satisfaction, self-efficacy, PA recall, and PA stage of change. Groups: virtual coach vs control (no coach: website + pedometer only). | Average step count for intervention group remained constant over 12 weeks while control group dropped. Repeated measures analysis of variance showed significant difference in step count change between intervention and control. No significant difference in secondary measures; acceptance mixed. | 4: +/– |
a <weight of study>: <effect>: weight of study was scored as 5 (randomized controlled trial [RCT] with at least one objective measure, long-term), 4 (RCT with at least one objective measure, short-term), 3 (RCT with self-report only, long-term), 2 (RCT with self-report only, short-term), or 1 (qualitative or pilot study). Effect was scored as + (positive), (negative), or +/– (mixed or insignificant).
b Embodied conversational agent.
c Randomized controlled trial.
d Physical activity.
e Transtheoretical model.
f Motivational interviewing.
g Wizard of Oz study, where humans pretend to be dialog agents to understand the likely responses to an automated agent.
h Computerized behavior intervention.
i Personal digital assistant.