Transcript Temporal Reasoning

Temporal Reasoning
A Riddle:
“Time Flies.
You Can’t, They Go By Too Fast.”
Temporal Reasoning ?
“Time Flies. You Can’t, They Go
By Too Fast.”
The Problem



~75 - 80% of medical applications require
reasoning about time
Time is a complex and mutable phenomenon
Humans live in a biologically and culturally
determined temporal matrix...
– so pervasive that normal people rarely perceive the
extreme conceptual complexity of “time” .


We are all “expert” temporal reasoners.
Computers are novices.
Temporal Reasoning 101
Specific Examples
Point
(“Now”)
 Interval (March 1, 1999:0800 to
March 2, 1999:0800)
 Rate (Miles per Hour, CCs per Minute)

Temporal Reasoning 201
Abstract/Uncertainty






Before
After
Following
During
Since
Schedule






Trend
Synchronize
Speed
Plan
Simultaneously
…..etc……...
Advanced Temporal
Reasoning
Implicit in the Medical Domain








Worse
Better
The Same
Changing
Stable
Unstable
On Call
Trade Call








Bolus
Piggyback
Input
Output
Wait
Stat
Emergent
Shift (night, day, split)
Advanced Temporal Reasoning
Examples: Y2K-Type
CODE ..”blue”…”red”…”1”…”2”…”yellow”
 Triage
 Strategic Plan
 “Time Pressure”
 “Dr Drummond, please go to Baby x’s
bedside.” (tight, formal voice on the phone)
 Scramble…...

Clinical Databases for Critical
Care
Clinical Use Case:
“A Comparison of the Temporal
Expressiveness of Three Database
Query Methods”
Amar K Das & Mark A Musen
JAMIA: Proc 19th Symposium,
1995, p331 -337
Stanford University
Needs Temporal Expressiveness:
Decision Logic for HIV Protocols
Multiple HIV clinical trial protocols, 2
hospitals
 256 encoded eligibility criteria assign
patients to individualized trials
 All necessary data were in database !!!
 20 criteria (8%) could not be
programmed for Automatic Logic
Processes (ALP)

Eight (40%) of the algorithm
failures were due to a
complex temporal pattern:
“No combination therapy for greater
than two weeks with 2 or more agents
active against MAC more than one
month before enrollment.”
Decision Logic for HIV Protocols
Problematic criteria had
4 Temporal Patterns:
 Temporal Duration (“..how long since..?”)
 Temporal Window (“..since last visit..?”)
 Prior Presence (“..must have [ ] for [ ] days before

[time expression]...”)

Temporal Concatenation
(“…add all fluids and
express the answer in cc/kg/d...)”
Temporal Duration (4%)

‘The temporal duration pattern finds the
length of time between a reference time t and
the time u of a clinical datum.’

or…”How long since the last time this
kid had a set of LFTs?”
Temporal Window (39%)

‘The temporal window pattern selects the last
occurrence of a instant-stamped patient datum
within a time span d prior to a reference time t.’

or...“What’s the last K+ this baby has
had since you added the aldactone?”
Prior Presence (33%)


‘The prior presence pattern determines whether the
interval stamp of a datum overlaps a span of time d prior
to a reference time t. Three [4] separate interval
comparisons (T1,T2, T3 can instantiate this pattern).’
or…”Did you start feeding this baby before
the surgeons said OK, or not?”
Temporal Concatenation (25%)


‘The temporal-concatenation pattern requires the union
of adjoining or overlapping intervals, T and U. This
pattern returns the duration of the resultant interval.’
or..”How long has this baby been on any
systemic antifungal since the first LP?”
Temporal Reasoning is
Important For:
Intelligent analysis of Time-oriented
Clinical Data (~80% for NICU care)
 Decision Support, Alerts
 Clinical Guidelines & Counseling
 Quality Assessment
 Data Visualization
 Data Mining

Temporal Reasoning “Standard” ?

How do we communicate temporal complexities
across computer systems ?
Example - Temporal “Standards” 101:
User-Oriented Grading Schema
“Document” = Database, Software Module, Object, Report…….

F: Document is undated
– (example: U of U grade/class status report sheets)



E+: Document has some date, somewhere, but is
incomplete and/or year is in “YY” format
D: Document has a data-related date/time stamp in
addition to the local download stamp
C: Document can express at least one additional
temporally-related function in contextually
appropriate, computerized units (interval, rate,…)
– e.g. Differentiates interval from total (“over whole student
career” vs “this quarter”)
Example - Temporal “Standards” 101:
User-Oriented Grading Schema
User Oriented Grading Schema – ‘User’ is a Grad Student

B: Document supports operational functionality for at
least 10 temporal concepts as defined by local
institutional needs (of the ## ISO defined temporal
expressivity standards [ …….. ].
– Must support date/time arithmetic accurately and consistently

A: Document allows full temporal expressivness, logical
time-related data-crunching and full support for the
ISO-Standard list [~32+…..] of :
“Temporal Reasoning Concepts”
The Beginning

As yet, There are NO Temporal Reasoning Standards for Healthcare.
Acknowledgements
Amar K. Das (Stanford)
 Mark A. Musen (Stanford)
 Carlo Combi, PhD (Universita ‘degli
Studi di Undine, Italy)
 Yuval Shahar, MD, PhD (Stanford)
 For teaching and publishing about time and
 ... for thinking like Salvador Dali : - )
