USING A SPOKEN DIARY AND HEART RATE MONITOR IN MODELING HUMAN EXPOSURE TO AIRBORNE POLLUTANTS FOR EP

1
USING A SPOKEN DIARY AND HEART RATE MONITOR IN
MODELING HUMAN EXPOSURE TO AIRBORNE POLLUTANTS
FOR EPAS CONSOLIDATED HUMAN ACTIVITY DATABASE

• Curry I. Guinn, UNC Wilmington
• Daniel J. Rayburn Reeves, UNC Wilmington

2
Collecting Human Activity Data

• Purpose To develop a method of generating an
  activity/location/time/energy expenditure
  database of sufficient detail to accurately
  predict human exposures and dose.

3
Goals of our study

• To Evaluate
• the use of digital voice recordings
• the use of the ambulatory heart rate monitor
• participant/instrumentation interactions
• To Develop
• a protocol for automating the processing of voice
  recordings
• an autocoding program that will be able to map
  the text of the diary entries to CHAD

4
Problems with Collecting Human Activity Data

• Recall Data
• Failure to recollect many daily activities
• Lack of detail
• Real-Time Paper Diaries
• Increased number of reports/better detail
• Burdensome
• Direct Observation
• Greatest number of reports/most detail
• Inefficient and expensive

5
The Experimental Platform

• Data Collection
• Audio diary using a digital voice recorder
• Ambulatory Monitoring System that monitors heart
  rate and prompts subjects to provide diary
  entries when heart rate increases by a specified
  criterion level.

6
Digital Voice Recorder

• Allows Real-Time Activity/Location Data to be
  Collected Easily
• Reduce burden of paper or computerized diary
  entries
• Relies on efficient, simple naturally spoken
  reports
• Potentially richer, more detailed reports
• No restrictive diary format
• Electronic format

7
Ambulatory Monitoring System

• Provides an objective measure of exertion that is
  more reliable than self-reported respiratory
  rates
• Prompts subjects to report activity when heart
  rate variation exceeds criterion levels

8
Natural Language Processing Application

• Applies contextual language constraints to
  facilitate speech-to-database conversion
• Speech ? Text ? Database Encoding
• Processes and codes the diary reports using the
  CHAD code scheme
• Reduce need for manual transcription and coding

9
Spoken Diary

• From an utterance like I am on the bus on my way
  to South Square Mall, map that utterance into
• 18400 Travel for goods and services
• 31140 Travel by bus)
• Text abstraction
• Technique
• Statistical language processing using n-grams and
  Bayesian statistics

10
Subjects
11
Voice Diaries

• Average 29 entries/ day
• With average monitoring time of 8.56 hours, 3.39
  recordings/hour
• First 3 days of trial 34.44/ day
• Last 2 days of trial 20.65/ day
• 1 out of 63 reporting periods data lost (1.6)

12
Recordings Per Day
13
Quality of Diary Entries

• Advantages
• High Entry Rate
• Timed correlation with heart rate data
• Disadvantages
• Little prompting
• Unformatted data
• Variable reporting of subjects

14
Quality of Diary Entries

• Entry Length
• 9.39 words average
• Some entries invalid because of length (subject
  failed to turn off recording)
• 1/30 recordings (3)

15
Heart Rate Change Indicator Tones and Subject
Compliance
16
Statistical Processing Accuracy of
Hand-Transcribed Data with Threshold of 0.3
17
Threshold values affect the precision and recall
The higher the threshold, the greater the
precision but the lower the recall
18
Time, Activity, Location, Exertion Data Gathering
Platform
19
Research Topics

• How do we fuse data from other sources (gps,
  beacons, heart rate monitor, etc.)?
• How do we provide interactive prompts to the
  subject to improve reporting?