infusing design into mHealth development
Predrag “Pedja” KlasnjaGroup Health Research Institute &University of Michigan
let’s start with an exercise…
• imagine that you are designing a new mobile app to help with weight loss
• you want to implement self-monitoring in your app
• figure out how the diet tracking part of the app should work
some things to think about
• what other information would you need to complete this task successfully?
• how would you know if you got it right?• how generalizable would a solution be?
this problem
• is ill-defined / underspecified• is context-dependent (and, so, unique)• doesn’t have a right or wrong solution• doesn’t have a clear test for solutions
most (mHealth) design problems are “wicked”
Rittel HW, Webber MM. Dilemmas in a general theory of planning. Policy sciences. 1973;4(2):155-69.
design is a process aimed at creatingartifacts, policies, and processes that fulfill theirpurpose well, given the intended users, contextof use, and other constraints.
key steps in the design process
• formative work to understand user needs potential approaches, and constraints
• formulating the problem• concretizing the context of use: scenarios,
personas, storyboards• ideation: sketching, brainstorming• analyzing tradeoffs of different alternatives• prototyping• user testing
mix, match, and repeat!
key steps in the design process
• formative work to understand user needs potential approaches, and constraints
• formulating the problem• concretizing the context of use: scenarios,
personas, storyboards• ideation: sketching, brainstorming• analyzing tradeoffs of different alternatives• prototyping• user testing
mix, match, and repeat!
Theory
TargetUsers
Context
YourawesomemHealth
app
Healthoutcome
Theory
TargetUsers
Context
YourawesomemHealth
app
Healthoutcome
Abraham, C., & Michie, S. (2008). A taxonomy of behavior change techniques used in interventions.Health Psychol, 27(3), 379-87.
problem: any theoretical construct can beimplemented in many different ways.
Very little A lot
challenge: figure out how to implement each component of the system to end up with anoptimized intervention.
Linda M. CollinsThe Methodology Center
Penn State
methodology.psu.edu
designing an optimized mHealth system
• generate– intervention components– decision rules
• optimize– intervention components– decisions rules– system as a whole (set of components)
How do you develop a good intervention component?
You come up with—and then choose among—a bunch of different versions
You then optimize the selected version
generate phase
Very little A lot
how granularly do we track?
• Exact calories• Courser amounts• Servings of individual ingredient types• Servings of courser food groups• Simple amount ratings• We don’t care about the amount
hunch: variability in daily goals can help individuals walk more
ideation: construct variations
selection phase
Design is compromise.- Bill Buxton
food tracking constraints
• can be maintained long-term
• doesn’t require deep nutritional knowledge
• fast to use• easy to use from the
beginning• supports graphing and
correlations
sources of constraints
• behavioral theory• formative work with target users• usability broadly construed
– e.g., integration with daily routines, privacy, social acceptability, user expectations, etc.
• previous studies • requirements from other parts of the
design
efficacy constraint: which of the options is likelyto work well?
evaluation for version selection
• goal: select among several different ways a feature can be implemented
• criteria:– efficacy for proximal outcomes– user experience
• methods– single-case studies– micro-randomized trials– randomized between-subject studies (when
possible)
goal variability study
• versions tested– epic goals– user choice goals– high amplitude goals
• methods– 10 fitbit users for 4 weeks
• step goal from one of the conditions sent by SMS each day
– latin square single-case design, one week per condition + control
• outcome: average daily step count per condition
evaluation requirements
• clear proximal outcome• way to do quick and dirty prototyping
proximal outcomes
most immediate intended outcomes of an intervention component
You have been sitting for 40 minutes. Get up and stretch your legs!i
Studies have shown that prolonged sitting has harmful physiological effects. Getting up regularly, even just for a minute or two, helps prevent those effects.
XSitting Reminder
OK
Proximal outcome: whether the user got up after the reminder
proximal & distal outcomes
• proximal outcomes are presumed mediators of desired distal outcomes– micro versions of distal outcome
e.g., reaching step goal for a single day– part of causal pathway to distal outcome
interactions with abstinence-supportingfriends for remaining drug-free
• different intervention components can target different proximal outcomes
potential proximal outcomes
• health behavior of interest (e.g., steps)• mediators of that behavior or distal
outcome• engagement• interest
[a prototype is] a representation of a design, made before the final solution exists.
-Bill Moggridge
building a prototype to test
prototypes for design evaluations
• quick• cheap• minimal representation of the
construct
BULK SMS
http://www.bulksms.com/countries/u/united-kingdom
what is bulk SMS? • service provider for sending and receiving SMS text
messages• 2-Way SMS communication • works for target users with iPhone, android, or any
phone capable of receiving SMS messages• allows researcher to pre-schedule SMS messages
using an excel sheet that is uploaded as a csv file• everything stored and processed through an online
account
www.pacoapp.com
what is paco?• a tool originally designed to do ecological
momentary assessment
• being reworked as a tool to support Just in Time Adaptive Interventions
IFTTT
https://ifttt.com/
what is IFTTT? • a service that connects applications (channels)
using recipes: if this, then that• works with nearly 300 channels including:
Facebook, Twitter, LinkedIn, Dropbox, Google Drive, Wordpress, Slack, Fitbit, Nike+, Jawbone UP, etc
• Bbowse channels and recipes for intervention ideas• works with android: battery, location, sms, photos,
and calls• works with iOS: contacts, location, photos,
reminders
early design evaluation studies • provide preliminary data to support the
design process• do not need to provide robust efficacy
results• emphasize efficiency and “good enough”
results to guide decision making
things to consider for selection
• there might not be one clear answer• it’s important to know the most
important criteria• empirical data (from focus groups, N-of-
1 studies) are essential for understanding tradeoffs
• a design decision can affect many other aspects of the project
optimization phase
component optimization
• goals– refine decision rules– refine user interaction– abandon ineffective components
• methods– micro-randomized trials– open loop system ID experiments– fractional factorial designs
summary
• mHealth design is a “wicked” problem without definitive solutions
• design is a process that can be used to “tame” such problems
• ideation and user testing can help develop promising components that are worth optimizing and testing