Web-Based Cognitive Health Coaching for Elders in a Home Environment Holly Jimison, PhD Misha Pavel, PhD ORCATECH Oregon Center for Aging & Technology What Can Be Done about Cognitive Decline? • Good news – Neural plasticity at all ages – New drugs to delay cognitive decline • Potentially good news – Monitoring computer interactions may detect decline earlier – Development of cognitive exercises to delay decline and possibly remediate ORCATECH 2 Project Objectives • Develop & test computer games for continuous cognitive monitoring of elders in a home environment – Sufficiently enjoyable to play frequently – Embedded assessment algorithms – Dynamic user models • Develop & test a cognitive health coaching system that provides opportunities for cognitive skill building – Assess cognitive deficits – Provide tailored cognitive exercises – Interactive cognitive health coaching ORCATECH 3 Problems with Current Cognitive Assessment Methods • Assessments done only after concern on the part of the patient or caregiver – usually late • Repeated assessments infrequent – yearly at best • Large day-to-day variability in cognitive performance with onset of dementia • Unwanted variability in assessments due to education, language, motivation, etc. ORCATECH 4 Traditional Cognitive Tests • • • • • • • ORCATECH Verbal Fluency Word List Learning Constructional Praxis Trail Making Test Symbol Digit Modalities Test Letter-Number Sequencing Finger Tap Test 5 Adaptive Cognitive Computer Games • 30 users (all cognitively healthy) – Average age 81.5 yrs (standard dev= 6.04 yrs) – 13% male, 87% female Adaptive Cognitive Computer Games Spry Learning / OHSU (NIST grant) • 9 Interactive games developed with Spry Learning as part of NIST grant • Being evaluated in local elders’ residences • Neuropsychological data was collected at baseline, 3 month, 6 month, and at a planned 1 year period ORCATECH 6 Elders’ Use of Computers • Growth in computer use by elders rose 47% since 2000 – fastest growing group • 70% use email • 59% Web browsing • 35% play computer games (similar to general population rate of 39%) ORCATECH 7 In-Home Monitoring of Computer Interactions • Frequent data trend detection – Individual as their own control – Less influenced by • Educational background • Language • Cultural differences • Dynamic algorithms to model user performance • Inexpensive screening, coaching ORCATECH 8 Game Data Analysis and Interpretation • Objective: Develop metrics that reflect cognitive capability and function of the players For Each Game • Ideal Player: Develop framework capturing the information processing resources and capabilities required by an ideal player • Cognitive Models and Limits: Describe ways that a human player may differ from an ideal one • Specific Metrics: Characterize the required cognitive processes by a small set of parameters ORCATECH 9 Letter Lotto – Word Game Word games: • Word length • Word complexity • Speed of word generation • Total number of words Compare to: • Verbal Fluency ORCATECH 10 Letter Generation Rate as a Game Metric Total Time to Guess Words (Each Game) vs Number Letters Guessed Correctly (User 1020) 250 200 Time (Sec) 150 100 50 Regression Equation: y= 2.31189 x + 13.2406 r = 0.866714 0 ORCATECH 0 10 20 30 40 50 60 70 11 # Letters Guessed Correctly 80 90 100 Letter Time vs. Neuropsychological Scores Neuropsych Average Total Fluency vs Letter Generation Time 85 Fluency vs Letter Generation y = -4.44616 x + 75.3999 Average Total Fluency (Letter + Animal) 80 r = -0.343154 75 70 65 60 55 50 0 ORCATECH 0.5 1 1.5 2 2.5 3 3.5 4 12 Letter Generation Time (Sec/Letter) 4.5 5 Frequency Generation of Words by Length Average Frequency of Words Correct by Length – Try harder because it’s the goal – Have to use all letters – Hence not necessarily most difficult 0.9 0.8 Average Frequency (Words Guessed/Total Possible) • Shorter words are guessed more frequently than longer ones • Exception is 7 letter word 1 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 6 Word Length ORCATECH 13 7 8 9 10 Neuropsychological Scores vs. Proportion of Words Neuropsych AverageScore Total vs Fluency vs Game Difficulty Neuropsych Proportion of Words 85 Average Total Fluency (Letter + Animal) 80 75 70 65 Fluency vs Game Difficulty 60 y = 0.522433 x + 40.2374 r = 0.403347 (with outliers) Outliers 55 y = 0.687054 x + 31.3083 r = 0.659844 (w/o outliers) 50 30 ORCATECH 35 40 45 50 55 60 65 14of Game Difficulty Proportion Generated Words 70 75 On the Flip Side – Memory Game Memory games: • Short-term memory • Working memory • Spatial memory • Abstract reasoning ORCATECH 15 Deterministic Computational Model of Working Memory • Characterize Memory Capacity as a Function of: – Intervening number of events – Intervening time – Memory load • Simple Memory Model: Discrete Buffer Time, Events B A C D A E B C F B G H D G E I B A C D D E B B F F G H D D D B A C C D E E B E F G H H H B A B C D D E D E F G F F D E F E B ORCATECH C D 16 E Probability of Correct Results Subject 1020, N = 8687 1 0.5 0 0 5 10 15 Probability of Correct Intervening Number of Events 1 0.5 ORCATECH 0 0 5 10 15 Intervening 17 Time [sec] 20 25 0.5 0 2 4 6 8 10 12 14 16 Probability of Correct 1 0.8 0.6 0.4 2 4 6 8 10 12 14 16 18 20 22 0.8 0.6 0.4 0.2 0 5 10 15 1 0.5 0 0 5 10 15 Intervening Time [sec] Intervening Time [sec] ORCATECH Subject 1024, N = 892 1 Intervening Number of Events Intervening Number of Events 0.2 Probability of Correct Subject 1021, N = 387 1 Probability of Correct Probability of Correct Examples of Other Fits 18 20 25 Weekly Tracking of Working Memory Buffer Size Subject 1021, N = 120 2.5 Memory Size BufferSize RelativeBuffer 2 1.5 1 0.5 0 10 ORCATECH 20 30 40 50 60 Time19[Days] 70 80 90 Estimation of Working Memory Capacity 3.5 Size Buffer Square BuffereRoot Sizeof[#of Items] 3 2.5 2 1.5 1 0.5 0 0 ORCATECH 5 10 15 Subjects 20 20 25 Assessment of Neuropsychological Test Scor 200 180 TRAILS-B Scores 160 140 120 100 80 60 40 20 0 60 ORCATECH 70 80 90 100 Trail-Making Test B R = 0.60 Digit Span R = 0.36 110 120 130 140 Predicted Scores from Flipside 21 150 160 21 Tally (2-D Black Jack) 00 8 18 0 ORCATECH 19 22 Dual Task games: • Divided attention • Working memory • Spatial memory Evaluation Metrics high Double Bust Cost Single Bust Accept low Single Win Multiple Win Bust-Win low ORCATECH Quality 23 high Response Time vs Quality of Move Value of subject move / value of Ideal move Subject ID:1020 HIGH-QUALITY LOW-QUALITY α=3 FAST in quality function SLOW Response Time per move (sec) ORCATECH 24 Reference Players • Ideal Player - uniform card distribution - look a single move ahead - choose ‘loc*’ where Valuec(loc*)= min(Valuec(loci)), i = 1 … # empty loc Valueq(loc*)= max(Valueq(loci)), i = 1 … # empty loc • Chance Player - place card randomly - all locations are equally likely ORCATECH 25 Prob Searching Occ Loc Modeling Subject Reaction Time a= 9.9886 b= 38.4971 To= 1.284 s Te= 0.683 s Avg RT per move Number of Occupied Locations Subject ID 1025 “Serial Processing of locations” Number of Occupied Locations ORCATECH RTmin= 2.9397 s 26 ORCATECH 27 Modeling Subject Reaction Time ORCATECH 28 Modeling Subject Reaction Time Subject ID a b T_occ T_emp T_min 1020 10.7166 13.3114 1.2583 0.4043 3.3855 1021 14.1551 1.1685 1.8104 0.2804 1.2226 1022 11.8667 3.534 0.9403 0.9704 2.9144 Took more21.4271 time 1024 10.8912 looking at… 0.9 1.2769 0 2.053 37.9464 1.2048 1.6135 0.9237 1025 9.9886 38.4971 1.2840 0.6830 2.9397 1026 18.0063 1.348 0.8088 0.2773 1.1533 0.7120 0.0282 1.6898 1.1771 0 2.5044 0.5736 0.8603 2.2749 1023 1027 1028 1029 Model shows these 18.0757 0.9 subjects don’t12.3714 spend 13.3458 time searching 13.7456 1.0617 empty locations 1030 10.9933 19.787 1.2249 9.9950 7.3894 1032 11.5338 4.6529 1.2456 0.1127 1.176 1033 8.7695 12.867 0.5714 0.0385 2.137 1034 12.8849 21.9481 0.6782 9.9996 2.9198 ORCATECH 29 Cost and Benefits of Divided Attention Forced Bust-Win Unforced Bust Forced Bust ORCATECH 30 Pyramid Builder – Color / Shape Matching Shape / color games: • Divided attention • Speed / accuracy • Visual search • Planning • Working memory ORCATECH 31 Response Time vs. Move Number File: tst3.mouse 8 • Response time (move duration) Increases with the Complexity of the Game – Average move duration increases with move number – Variance increases with move number Response Time [Sec] 7 6 5 4 3 2 1 0 0 20 40 60 80 Move Number ORCATECH 32 100 120 140 Example: Response Time Distribution (Data from a Single Game) File: tst2.mouse 0.16 • Fast responses to known or easy decision • Medium requiring some search and decision 0.14 0.12 Relative Frequnecy Multimodal distribution suggests 0.1 0.08 0.06 0.04 0.02 0 • Long – pondering and planning of execution ORCATECH 0 0.5 1 1.5 2 Response Time[Sec] 33 2.5 3 3.5 Resulting Search Efficiency tst3.move 4 3.8 slope = 24 ms/item Move Times [sec] 3.6 3.4 Correlation with Trail-making B 3.2 3 2.8 R = 0.49 2.6 2.4 2.2 2 0 5 10 15 20 Number of Distractors ORCATECH 34 25 30 Scavenger Hunt – Trail Making Test Rapid movement games: • Motor speed • Visual search • Set switching • Speed of processing • Divided attention Compare to: • Finger Tap Test • Trail Making Test ORCATECH 35 FreeCell – Solitaire Planning Game Strategy games: • Planning • Visual search User Performance Metrics t Dt f d d 1 , • Difficulty 1 dt = number of moves to solution at time t • User Performance – Reduction in difficulty t dt dt 1 • Outcome of game • Time to completion ORCATECH 37 FreeCell Performance Curve S ubject nb1.als 90 80 70 subject Difficulty Difficulty 60 50 40 solver 30 20 10 0 Actual Expected 0 10 20 30 40 50 Events 60 70 80 Number of moves ORCATECH 38 90 100 Performance Scores for Normals vs MCI Average (Individual Ave(Perf)) St Dev (Individual Ave(Perf)) Average (Individual SD(Perf)) Normals 0.58 0.12 0.38 Mild Cognitive Impairment 0.27 0.72 0.55 ORCATECH 39 Benefits of Cognitive Computer Games Improvements over standard techniques: – – – – – Daily monitoring (not yearly, after-the-fact) Individual as their own control Adaptive measurement – more sensitive measure Understanding of variability, trend detection Unobtrusive, more representative performance – Potential for remediation ORCATECH 40 Cognitive Health Coaching Project Objectives Remote Cognitive Health Coaching to the Home • Provide tailored cognitive exercises & advice with tailored feedback • General coaching protocols based on principles of health behavior change – Using health behavior change techniques (motivational interviewing) – Tailored action plans – Automated feedback – Facilitate care manager and caregiver communications • Extend the outreach of health coaches – Tailored messaging / automated dynamic protocols • Incorporate patient and family members into the care team ORCATECH 41 Overview of Methods • Needs Assessment – Interviews & focus groups with patients, family caregivers, home health workers, clinicians, nurse care managers, payers – Interview researchers/experts with an interest in deploying and evaluating a health intervention in a home environment • • • • Clinical / Behavioral Protocol Development (Cognitive) Design Specifications System Development Implementation / Usability Testing / Stakeholder Feedback • Test in Homes of Elders ORCATECH 42 Elder Focus Group Findings • “Brain before body” • Most had multiple chronic diseases, but were not as interested in working on the “medical” aspects of health • Highest priorities were quality-of-life and independence • Advice for others – Considered depression and socialization to be the big problems for their age group – Many elders recognized a downward spiral in older people who became increasingly embarrassed about their abilities and appearance, so that they tended to avoid leaving their room or socializing. ORCATECH 43 Expert Interviews • Researchers • Government – Legislators, policy makers • Industry – Health record developers, interoperability specialists, disease management system vendors • Elder Care Professionals – Geriatricians, home health nurses, senior activity coordinators, health coaches • Cognitive Health Intervention Experts (continuing) ORCATECH 44 Cognitive Health Coaching Design Elder-Specific HRA HomeCare Nurse / Coach Health Coaching System Shared Care Plan Medication List (PHR for Chronic Care) ORCATECH Diagnoses Lab Values Cognitive Protocols (Spry Games) Monitoring Modules Sensors Allergies Elder 45 Family Support Welcome Ellen! Your Cognitive Exercise Plan Ellen – Last week you did a great job with the number of FreeCell and Sudoku exercises played. This week, in addition let’s try adding some verbal exercises, such as the Spelling Bee game featured below. Click here for more information on the importance of verbal exercise. Reply to Coach Previous Messages Update My Information View My Action Plan Ellen’s Cognitive Exercise Progress Detail on Progress Return to PHMS Goal 2/day 2/day 2/day 2/day 2/day 2/day 2/day 2/day FreeCell Hunt LLotto Sp Bee FlipSide Pyr Bldr 21Tally Sudoku Mon *** ** * ** * ** *** Tue ** ** Wed **** * ** ** ** ** Th * * *** Fri ** * * ****** * * ** ** *** **** Sat Log Off Cognitive Coaching for Better Living ORCATECH Sun 46 Review of Methods • Needs Assessment – Interview researchers/experts with an interest in deploying and evaluating a health intervention in a home environment – Interviews & focus groups with patients, family caregivers, home health workers, clinicians, nurse care managers, payers • Clinical / Behavioral Protocol Development (Cognitive) • Design Specifications • Implementation / Usability Testing / Stakeholder Feedback • Test in Homes of Elders ORCATECH 47 Components of Cognitive Health Coaching • Adaptive Computer Games - Cognitive Exercise – – – – NIST grant with Spry Learning Adaptive cognitive games Embedded cognitive assessment algorithms Suite of 9 games measuring various aspects of cognition • Health Advice Related to Cognition – – – – – – Physical exercise Social Interactions Nutrition Sleep Novelty exercises Mood management (depression advice, humor) ORCATECH 48 Sample Message – Novelty Exercises Non-routine actions and thoughts encourage the growth of new connections in the brain. This week let’s work on novelty exercises that can exercise your brain. Here are some examples to choose from: □ Turn your desk clock or an illustrated calendar upside down. □ Go to a new park, or a new store. □ Try brushing your teeth with your non-dominant hand. □ Distinguish coins using only your sense of touch. □ Turn the volume off on television and invent the dialog based on what you see the characters doing. □ Find a magazine you normally wouldn't choose, and read it for a while. ORCATECH 49 Sample Message – Novelty Exercises ◙ Turn your desk clock & calendar upside down. When you look at a familiar image right side up, your left brain quickly labels it and diverts your attention to other things. When the picture is upside down, the quick-labeling strategy doesn't work and your right-brain kicks in, trying to interpret the shapes, color and relationships of a puzzling picture. This week, place at least one clock and your calendar upside down. Practice using them this way and see how it feels. Do you get better over time? Next week we’ll try something new. ORCATECH 50 User Model 1. Elder Assessment • Health Risk Appraisal • Health Goals • Motivation • Barriers • Readiness-to-Change • Contact Preferences •Timing / Media • Social Support Contacts • Data Sharing Preferences • Authentication (pswd?) Web forms – spread timing Elder User Model • Goal (Assume Cognitive Health) • Motivations • Barriers • Change State • Contact Prefs • Social Support • Display Prefs Coach User Model Predefined 2. Family / Friend Assessment • Contact Preferences •Timing / Media • Level of commitment • Authentication (pswd?) Fam/Fr User Model • Contact Info/Prefs • Data Permission • Authentication (pwd) Web form ORCATECH 51 3. Create / Update Tailored Action Plan Daily goals & activities Monitor continuous 4. Tailored Feedback & Recommendations Based on Spry Games Based on meeting other goals Sleep behaviors Diet Exercise … 5. Generic News and Articles of Interest Standard Web content Architecture for Tailored Messaging Shared Care Plan Elder Assessment Elder DB Monitoring User Model Raw data User State Sensors Computer Interactions Self Report State Dependent Active Methods (Coaching Protocols) Update Action Plan Tailored Content Delivery • reminders • encouragement • tailored feedback/education • prompt for assessment Content Database Preconfigured Messages Tutorials General Content Definitions Summary Reports Prioritized Alerts Recommended Messaging Elder’s Technology Interface Health CoachTechnology (computer, cell phone, exercise watch) (computer, cell phone) ORCATECH 52 Welcome Ellen! Your Cognitive Exercise Plan Ellen – Last week you did a great job with the number of FreeCell and Sudoku exercises played. This week, in addition let’s try adding some verbal exercises, such as the Spelling Bee game featured below. Click here for more information on the importance of verbal exercise. Reply to Coach Previous Messages Update My Information View My Action Plan Ellen’s Cognitive Exercise Progress Detail on Progress Return to PHMS Goal 2/day 2/day 2/day 2/day 2/day 2/day 2/day 2/day FreeCell Hunt LLotto Sp Bee FlipSide Pyr Bldr 21Tally Sudoku Mon *** ** * ** * ** *** Tue ** ** Wed **** * ** ** ** ** Th * * *** Fri ** * * ****** * * ** ** *** **** Sat Log Off Cognitive Coaching for Better Living ORCATECH Sun 53 Review of Methods • Needs Assessment – Interview researchers/experts with an interest in deploying and evaluating a health intervention in a home environment – Interviews & focus groups with patients, family caregivers, home health workers, clinicians, nurse care managers, payers • Clinical / Behavioral Protocol Development (Cognitive) • Design Specifications • Implementation / Usability Testing / Stakeholder Feedback • Test in Homes of Elders ORCATECH 54 Summary • Scalable approach to delivering health care to the home – – – – Extends outreach of lower cost professionals Can involve family members Timely and continuous Home health management is key to caring for chronic conditions • Cognitive monitoring and potential remediation – Most important aspect of health – Keeps people independent – (better quality of life & lower costs) • Cognitive Health Coaching System – Serves as a framework for testing the best delivery of cognitive interventions ORCATECH 55 Acknowledgements • ORCATECH – NIA funding • Intel funding for initial FreeCell Cognitive Monitoring • NIST Funding for Spry Learning Cognitive Computer Games • Health Coaching funding to OHSU from Intel as part of the Behavioral Intervention & Assessment Commons More info: jimisonh@ohsu.edu ORCATECH 56 Representative FreeCell Performance Data MCI Subject and Healthy Subject Over Time Subject with Mild Cognitive Impairment Cognitively Health Subject MCI Subject FreeCell Efficiency Over Time Healthy Elder FreeCell Efficiency Over Time 2.5 2.5 2 2 1.5 1.5 1 1 0.5 0.5 0 2/23/04 -0.5 3/14/04 4/3/04 4/23/04 5/13/04 6/2/04 6/22/04 0 2/3/04 -0.5 -1 -1 -1.5 -1.5 ORCATECH 57 2/23/04 3/14/04 4/3/04 4/23/04 5/13/04 6/2/04 Median Login Inter-keystroke Intervals Milliseconds Subject with mild cognitive impairment Two cognitively healthy elders Date ORCATECH 58 Framework for Data and Inferences • Data – Movements – trajectory and speed – Selection of objects – ID and response time – Placement of objects – position and speed Raw Data • Interpretation – – – – Characterization of movements Inference of tactics Inference of strategy Inference of abilities Tactics Strategies State Estimation ORCATECH 59 Duration of Move ↑ with Complexity of Game File: tst3.mouse 8 – Average move duration increases with move number – Variance increases with move number Response Time [Sec] 7 6 5 4 3 2 1 0 0 20 40 60 80 Move Number ORCATECH 60 100 120 140 Mouse Move: Trajectory and Dwell Time – Multimodal Trajectory of a mouse movement on a single move. – The size of the disks indicates the dwell time at a that location Move number 53 650 600 550 500 450 400 350 300 250 200 ORCATECH 61 0 100 200 300 400 500 Background on Shared Care Plan • Personal Health Record specifically designed for people with multiple chronic conditions • Marc Pierson, MD – Whatcom County, Washington • Pursuing Perfection Project funded by Robert Wood Johnson Foundation • Further refinement of the application was made possible through grants from the Agency for Healthcare Research and Quality (AHRQ), the Foundation for eHealth Initiative (HRSA OAT), and PeaceHealth • Commercial version called PHMS from the Congral Company – Nic Ivancic ORCATECH 62 Project Team Members • • • • • • • Holly Jimison – PI Misha Pavel – Co-PI Nicole Larimer – Senior Research Associate Michael Shapiro – Graduate Research Assistant Bobby Hagerty – Health Educator / OHSU Brain Institute Nic Ivancic – Programmer, Congral Company Lori Nichols - Consultant – Director, Whatcom Health Information Network – Shared Care Plan Personal Health Record • Marc Pierson – Consultant • Spry Learning Company ORCATECH 63 Correlation Amongst Standard Test (Animal and Let • Average Animals vs Average Letter Fluency (C,F,L) 35 Average Animal vs Average Letter • y = 0.36845 x + 7.78601 30 2 R =0.354482 (with outliers) Outliers y = 0.378927 x + 7.89503 25 2 Letter Fluency R =0.464943 (w/o outliers) • • 20 15 10 5 0 0 5 10 15 20 25 30 35 Animal Fluency ORCATECH 64 What is the inherent variability within animal and letter fluency? Within users averaged over baseline, 3 month, and 6 month test period R^2~=.46 Used sum of animal and letter score as total verbal fluency Newly Developed Cognitive Computer Games Spry Learning / OHSU (NIST grant) ORCATECH 65