Jarrett Interaction Design

using Topology and Node Value Differences

• example: compare national budgets across two years; what agencies disappear? which ones grow large absolutely? grow by large percentage? what other relationships
• put data in a tree, then encode absolute amount of change, color by percentage of change, outline based on removal and creation
• filter by layers in the tree, from top level departments, down to initiatives, down to accounts; quickly see where the changes have been made, what was created and what was eliminated
• synchronize across views for data painting/filtering and highlighting
• tested by letting participants try to perform comparisons without any training in the tool
• types of tree comparison: (type 0) topology but not node values; (type 1) values on leaf nodes only; (type 2) values on leaf and interior nodes; (type 3) type 1 + topology; (type 4) type 2 + topology; previous work has only done 0 and 1 successfully; this work focuses on type 2-4; only research known to address all these areas
• interesting analysis of actual 2012 and 2013 US federal budgets; another on airline passengers;

• also uses temporal events; looks for temporal associations, considering both order and relative time difference
• across many records, align by one focal event, as in EventFlow; aggregate into a single timeline; summarize in a histogram on the timeline
• to compare across event pairs, there are many histogram combinations; need to organize the most interesting/relevant histogram comparisons
• used graduate student dataset as example set, e.g. compare when defense occurred relative to proposals; no defenses before proposal
• designed multiple measures of interestingness, e.g. which events occurred most often after focal event; other examples, periodic occurrences
• work is still in early stages; looking for other data sets to analyze

Display Methods and Interactive Visual Query

• Big Data -> LifeFlow -> EventFlow
• tools can apply to any domain with events across time
• traditional medical research model: hypothesis, recruit, observe, analyze, report (takes years)
• big data model: hypothesis, mine, pull records, analyze, report (takes only months); large, messy data sets
• project focus on analyzing temporal event sequences: point event = item ID, event type, time; interval adds a time, for start and end times
• LifeFlow aggregation: align all flows by first event, group each event into vertical bar positioned by average time; effective for population level observations
• US Army Pharmacovigilance Center: all our data is interval based events, but this only works on point data; so, make it work with interval data
• dealing with intervals adds complexity in input, data structure, display, and especially querying
• existing query engine can’t express interval relationships

Absolutely awesome presentation style. You go girl! Love it.

• patients are harmed by missed steps in complex workflows; most common cause in malpractice litigation
• some current solutions show alerts in chronological order and difficult to find, no pending ones shown, no notion of lateness, no hierarchy of data
• model the workflow, provide an improved tabular display for interaction, and a visualization of overall state
• tabular data is not sparse, it’s broken into sections (current vs pending), and color coded for state, no truncation of data – fit on one screen; recommended actions popped up when a result is selected, with next step or completion as the terminal action in the popup
• provide different table views for different roles, such as physician (across patients), resident (a single patient), assistant (ensure result flow)
• manager gets a process completion diagram to understand overall workflow, issues, and bottlenecks; can drill in to each step of the process to see where most problems arise; drill-down on a step and you can compare individual performance in the step to identify individual performance problems

• IOM Report: Health IT and Patient Safety: recommendation – user-centered design and human factors applied to health IT
• we spend more than any other country on healthcare, but we have poorer outcomes
• 98,000 unnecessary deaths each year from medical mistakes
• technology makes complex delivery even more complex and hazardous; e.g. dosing errors, failures in detection, and treatment delays
• SHARP – Strategic Health IT Advanced Research Projects; involved in Patient-Centered Cognitive Support; clinician driven, better decisions, more safely, effectively, rapidly
• Health Care Reform: one goal – by 2014, every citizen has an Electronic Health Record
• HCIL research initiatives: medical lab tests (tracking completion in complex workflows, guidelines for design of tables and actions, retrospective analysis); medication reconciliation; patient histories – analysis and search (temporal even sequences, point and interval events); wrong patient errors
• up to 10% of test results don’t get back to the ordering physician; 40+ reasons where the workflow breaks down
• most medical IT software companies prohibit, via agreements, from writing and publishing about their systems
• set of guidelines for good table design; starting to incorporate elements from visual designers as well
• series of projects for patient histories: LifeLines (about 1/2 systems out there have this type of display embedded for individual patients); LifeLines2 (used in Washington Hospital System); LifeFlow (up to 203,000 cases summarized);  EventFlow (Army medical operations; medication histories); PairFinder
• data is incomplete and messy, so analysis works better and is easier on point events; now moving back to interval events
• overview, zoom and filter, details on demand (2,000 citations); useful guiding principal
• reducing wrong patient errors: 1 in 1000 error in selecting the wrong patient and take next steps without noticing; animated transitions (500ms dimming of others) and photos (of patient on admitting) – from 7% recognition of error to 63% if combined animation and photos)

• example: how to choose a new smartphone; a bunch of published physical characteristics + a bunch of personal criteria that are harder to discover; using a few attributes, filter to a smaller group of choices, then dig into each on the other attributes to decide
• other similar problems: find a specialist physician, location for company HQ, grad school to apply to, vacation itinerary, meeting time
• Amazon Mechanical Turk: pay people small amounts to gather information; good for collecting specs
• example: which store has the cheapest total price for a shopping list? use mechanical turk to get people to look up a price in each store’s special flier for a small set of items (meaningful chunks); one store at a time; order the tasks so that the highest price, highest variation prices are first because they impact total most
• AskSheet: list all items, approximated spec ranges, and goal cell; custom ASK() function to tell system how to create the task forms for Mechanical Turk; prioritizes things so not all possible tasks need to be performed; can expect ~30% task savings with the optimization
• not all problems appropriate for the optimization

Medication reconciliation and product comparison

• medication reconciliation: what is the patient really taking? different sources across times; dosages and periods, brand names, chemical compounds
• levels of equivalence: form, functional, partial, none
• Twinlist: spatial layout and animation for reconciliation; very well received (code is available); applied in different areas, e.g. medical problem list reconciliation;
• does animation help learn? without animation, many people needed clarification, especially in “similar” case; task completion is slower objectively, but not necessarily subjectively, and maybe fewer errors; participants prefer animation for learning, but would prefer not for regular work
• people forget to scroll long lists; hidden reject operation
• ManyLists: use in product comparison (code will be available)
• spacial layout for comparison, multi-step animation for explanation/learning

Motivating volunteers to participate in citizen science projects.

• scientists who study species are becoming extinct
• citizen science: involve amateur scientists and use crowd-sourcing for species discovery (and other scientific projects)
• principles of social participation: egoism, altruism, collectivism, principalism
• volunteers believe more in collectivism; scientists are slightly less motivated by egoism than volunteers
• scientists need the volunteers, feet on the ground; want to educate the public
• volunteers are personally interested, expect attribution, and want feedback from scientists; also want to see impact on community
• demotivating factors: lack of trust across volunteers and scientists;
• process model to focus on motivators and build trust and collaboration

[140 character limit on Twitter sucks; been losing the #HCIL hashtag when this gets cross-posted]

• CDC has lots of disaster preparedness; the Zombie Apocalypse is coming soon, need to prepare; social network analysis can help with this
• taxonomy of zombies: slow, dumb zombies (Night of the Living Dead); medium-speed, slightly-aware (Walking Dead); fast, angry zombies (28 days later; we’re doomed)
• social network analysis: spread things through social networks, like zombinism and information
• tie strength: strong ties (close friends and family), weak ties (acquaintances); in 70’s, thought strong ties most important; now understood that most things accomplished through weak ties; you get all the weak and strong ties of all your weak ties too; weak ties are less trustworthy, and more likely to infect you
• cut off physical contact with weak ties and avoid most active protects from infection, but information spreads best on weak ties too; need long-distance communication; keep information channels open
• PSA video for preparation for Zombie Apocalypse

• children are interesting and important; often overlooked in design process; children under 12 experience differently than adults (e.g. keep visual and auditory streams separate, concrete instead of abstract, visual instead of verbal), kids are spending more of “their own money” and influence 47% of household spending; it’s fun!
• participatory design research with kids; co-design; cooperative inquiry
• Kidsteam: twice a week for 16 years; snack-time (non-structured discussion), circle-time (semi-structured, discuss question of the day), design-time (low-tech prototyping), big-idea-session (synthesize ideas and how to move forward)
• Online kidsteam: distributed design
• limitations of current methods: travel/logistics, time delays, limited co-design; online is synchronous, adult-focused, hard to manage iterations
• research by design: build prototypes to answer research questions; create the right thing; include children as partners
• 12 geographically distributed children, 8 adults
• mimics kidsteam cycle; drupal-based tool (disco – asynchronous whiteboard); develop iteratively throughout; snack-time mimicked using messaging; circle-time is forum; embedded whiteboard software for design-time; iteratively added avatars, audio recording, moved from Flash to HTML5, co-located multi-user logon, iOS browser-based; points for generating designs; library of projects
• working with children: early tasks build to later tasks, including caregivers, have multiple entry points (didn’t like drawing with mouse, but liked drawing with iPad)