Simulating Pharmaceutical Sales and Disease Outbreaks based on Actual Store Sales and Outbreak Data

1
Simulating Pharmaceutical Sales and Disease
Outbreaks based on Actual Store Sales and
Outbreak Data

• Glenn Guthrie, David Calvert, Deborah Stacey
  Computing Information Science
• Victoria Edge
• Foodborne, Waterborne and Zoonotic Infections
  Division

2
The Problem

• This work is part of study into the use of
  over-the-counter (OTC) pharmacy sales data as an
  indicator of infectious disease in the local
  population.
• But, there exists a relatively small collection
  of sales data for OTC products during known
  outbreaks of infectious disease.

3
The Data

• Historical pharmaceutical sales data are
  available for the waterborne outbreaks of E. coli
  O157 in Walkerton, Ontario and Cryptosporidium in
  North Battleford, Saskatchewan.
• OTC data is available from one pharmacy chain for
  a number of stores over a few years.

4
(No Transcript)
5
(No Transcript)
6
OTC Sales Data format
7
Aberration Detection

• Our goal is to develop and test various
  aberration detection algorithms but in order to
  do this effectively, it is necessary to have more
  OTC sales data that relate to outbreak periods.
• So if we cant find real data then we will have
  to simulate it thus the data generator is born.

8
The Data Generator

• Step 1. Generate hundreds of years of baseline
  OTC sales based upon a pharmacy's statistical
  profile.
• Step 2. Outbreaks are created with various
  parameters that follow the patterns seen in
  Walkerton and North Battleford.
• Step 3. The baseline and outbreaks are melded in
  different proportions to arrive at the final OTC
  sales simulation.

9
The Store Footprint

• A database of known store parameters (overall,
  seasonal and daily) is used to create the
  baseline sales data for a store.
• The OTC sales simulator is designed to maintain
  the general noisiness seen in day to day
  pharmaceutical sales including spikes that are
  not indicative of outbreaks.

10
Parameters

• The simulator has a number of configurable
  parameters.
• Simulated outbreaks can be altered in duration,
  magnitude, frequency and daily randomness.

11
./baseline
Store sales statistics

• Input Stats File Name stats.conf
  ../statistics/stats.conf
• Output Baseline File Name baseline.txt
• Years of Data to Generate 100
• Stats Profile - (O)verall, (S)easonal or (D)aily
  D
• Random Seed - Positive Integer or System Clock
  
• Using Random Seed lt1098125118gt

Sales characteristics
Reproducibility
12
baseline.txt

• Sat-01-01-2000 1024 ANTI-DIARRHEALS 0
• Sun-02-01-2000 1024 ANTI-DIARRHEALS 0
• Mon-03-01-2000 1024 ANTI-DIARRHEALS 1
• Tue-04-01-2000 1024 ANTI-DIARRHEALS 0
• Wed-05-01-2000 1024 ANTI-DIARRHEALS 3
• Thu-06-01-2000 1024 ANTI-DIARRHEALS 8
• Fri-07-01-2000 1024 ANTI-DIARRHEALS 10

13
(No Transcript)
14
./outbreak
Generated Baseline

• Baseline File Name baseline.txt
• Output Outbreak File Name outbreak.txt
• Outbreak Scale Max 1.20
• Outbreak Scale Min 0.80
• Outbreak Duration Max 1.20
• Outbreak Duration Min 0.80
• Baseline to Outbreak Proportion 0.80
• Daily Randomness 0.20
• Frequency - Average Days Between Outbreaks 300
  
• Max Number of Outbreaks - Positive Integer or
  NOT SPECIFIED
• Type of Outbreak - (S)low, (R)apid Onset or
  (B)oth B R

Range of outbreak amplitudes

Outbreak length range

Additive Factors
Walkerton or Battleford
15
./outbreak (contd)
Aberration detection via ANN

• Output Additional Moving Window Formatted File
  (Y/N) N Y
• Size of Moving Window 7
• Moving Window Reference Point 4
• Random Seed - Positive Integer or System Clock
  
• Using Random Seed lt1098125728gt
• File ltoutbreak.txtgt Created Successfully - Hit
  ENTER To Continue

Reproducibility
16
outbreak.txt

• Thu-03-05-2001 1024 ANTI-DIARRHEALS 2
• Fri-04-05-2001 1024 ANTI-DIARRHEALS 4
• Sat-05-05-2001 1024 ANTI-DIARRHEALS 0
  OUTBREAK
• Sun-06-05-2001 1024 ANTI-DIARRHEALS 1
  OUTBREAK
• Mon-07-05-2001 1024 ANTI-DIARRHEALS 2
  OUTBREAK
• Tue-08-05-2001 1024 ANTI-DIARRHEALS 2
  OUTBREAK
• Wed-09-05-2001 1024 ANTI-DIARRHEALS 1
  OUTBREAK
• Thu-10-05-2001 1024 ANTI-DIARRHEALS 2
  OUTBREAK
• Fri-11-05-2001 1024 ANTI-DIARRHEALS 1
  OUTBREAK
• Sat-12-05-2001 1024 ANTI-DIARRHEALS 3
  OUTBREAK

17
Start
End
18
Research Issues

• Are the generator parameters meaningful?
• Are the outbreaks integrated properly into the
  baseline?
• Will each store have its own footprint or can we
  sample multiple stores to create one profile?
• Can spatially related stores be merged to produce
  a super sales profile?

19
Availability

• The data generator will be made available as open
  source software and researchers are encouraged to
  use the data generator to generate their own
  training and testing scenarios.

20
The Test Harness

• A test harness is being developed to
  automatically evaluate aberration detection
  algorithms used on standardized datasets created
  by the data generator.
• The test harness itself will be made available to
  the research community so that aberration
  detection algorithms can be effectively compared
  and a website will be made available to
  distribute results.

21
Contact Information

• http//hebb.cis.uoguelph.ca/asap
• dastacey_at_uoguelph.ca