GroupNotesInnerx

beadkennelΤεχνίτη Νοημοσύνη και Ρομποτική

15 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

106 εμφανίσεις

Problems/Needs


Need to identify within
-
person variability to predict decline in health status

and use data for monitoring progress of interventions


Lack of customization to the individual


Lack of ability to determine if perturbations over time are
normal variances
or early signs of incident disease


Too much data


Lack of understanding of relationships between and among the many
influences on health


genetic, biologic, behavioral, social, environmental


Inability to factor in context for any given
individual


Imprecise and even misleading analytic methods


Dealing with signal/noise issues


what is important and what isn’t?

Is it true inherent variability in the data
or

is it something important?


Need to solve current, pressing problems


applied r
esearch

Need to step back and look for the unexpected


basic research


Health is (may be) modular; how can person
-
level health be viewed as
scalable to population health?


Lack of closing the loop on data
-
application
-
knowledge
-
new data, etc…


How are dat
a analyzed? Hierarchical data analysis


multilevel/multidimensional


Potential reactivity to monitoring devices


What is important to measure at the person
-
level?






Solutions


Need to hypothesize what is signal and what is noise and test to see


Take
the laboratory to the person


De
velop de
vices

that

can produce better resolution of constructs


over time
and within time


Make data actionable


useful for a present health need


Better methods of data visualization to improve comprehension


For patients


For individuals


For clinicians


For public health professionals


Build time
-
series model
s

of multiple levels of data that may be important for
health, cast a wide net for and collect these data to begin to better
comprehend influences on health

Example:

To predict who is most at risk we are now using EMR a
nd other
limited sets of data. It’s t
ime to expand upon this.



To Do:


Large, long
-
term studies from longitudinal cohorts


--
Different things can be learned from diverse vs. homogenous populations


Increased interactions with

the

machine learning community to mash up and
analyze sensor data with EMR data with other forms of data


Leverage current observational cohorts


add
-
ons of new person
-
level
technologies


Ensure that we have repeated measures i
n cohorts that are frequent and
powerful enough to drive new inferences and insights


Develop a “navigator function” for centralized data sets



propose novel
methods of doing this for any current data set


Enroll cohorts of volunteers willing to share dee
p levels of data to improve
understandings of contributors to health and wellbeing



“crowdsource
things”




We have NHANES, BRFSS,
other ongoing large
-
scale population studies
that focus on assessing behaviors and disease states. Why not stand up a
large
scale, prospective cohort of individuals who are willing to have
monitored data collected and analyzed
?