Automated%20Selection%20of%20Patients%20for%20Clinical%20Trials

1
Automated Selection of Patients for Clinical
Trials
Eugene FinkLawrence O. HallDmitry B.
GoldgofBhavesh D. GoswamiMatthew
Boonstra Jeffrey P. Krischer
2
Clinical trials
A clinical trial is an evaluationof a new
treatment procedure.
When physicians conduct a trial, theyrecruit
patients with matching healthproblems and
medical histories.
3
Selection of patients
The selection of trial participants isa manual
procedure, and physiciansmay miss eligible
patients.

• Gotay 1991 demonstrated that physicians
  choose only 39 of the eligible patients
• Fallowfield et al. 1997 showed that they
  choose less than 50 of the eligible patients

4
Expert system
We have developed a system thatautomatically
selects prospectivetrial participants.
5
Outline

• Knowledge base
• Selection results
• Cost reduction

6
Knowledge base

• Tests and questions
• Eligibility criteria

7
Tests and questions
The knowledge base includes a list of medical
tests. A test descriptionincludes a dollar cost
and relatedquestions about a patients health.
Example Mammogram, 150What is the cancer
stage?Does the patient have invasive
cancer? Biopsy, 400How many lymph nodes have
tumor cells?What is the greatest tumor diameter?
8
Eligibility criteria
The knowledge base also includeslogical
expressions that representeligibility for the
available trials.
9
Selection process
The system collects data until it candetermine
whether the eligibilityexpressions are TRUE or
FALSE.
If patient records do not provideenough data,
the system identifiesthe required medical tests.
10
Outline

• Knowledge base
• Selection results
• Cost reduction

11
Experiments
We have used fifteen breast-cancertrials at the
Moffitt Cancer Center.

• Past data from 187 patients

• Current data from 169 patients

12
Finding eligible patients Past data for 187
patients
ClinicalTrial Parti-cipants OtherEligible
1082210840 11072 11378 119921210012101 10 . 0 . 48 . 4 .5 .8 . 20 . 5 . 19 . 26 . 19 .6 .20 .30 .
13
Finding eligible patients Current data for 169
patients
ClinicalTrial Parti-cipants OtherEligible
1113211931 11971 12100 121011238512601 12643 12757 12775 4 . 2 . 4 . 0 .11 .0 . 0 . 16 . 1 . 23 . 1 . 26 . 0 . 5 .52 .19 .1 . 36 . 3 . 17 .
14
Participation in other trials Current data for
169 patients
ClinicalTrial Incom-patible Compa-tilbe No OtherTrial
1113211931 11971 12100 121011238512601 12643 12757 12775 0 . 0 . 0 . 0 .13 .8 . 0 . 0 . 0 . 3 . 1 . 11 . 0 . 1 .6 .2 .0 . 10 . 1 . 3 . 0 . 15 . 0 . 4 .33 .9 .1 . 26 . 2 . 11 .
15
Summary
The results suggest that the systemcan increase
the number of trialparticipants by a factor of
three.
16
Outline

• Knowledge base
• Selection results
• Cost reduction

17
Medical tests
The selection of trial participantsmay require
medical tests.

• The total cost of tests may depend on their
  ordering

• Finding the right ordering is often a complex
  problem

18
Ordering of tests
The system chooses the ordering of teststhat
reduces their expected total cost.
After getting the results of the first test,
itrevises the ordering of the remaining tests.
The ordering is based on the test costs,number
of trials that require each test,and structure
of eligibility expressions.
19
Cost reduction Past data for 187
patients
ClinicalTrial Mean Cost Mean Cost
ClinicalTrial W/O TestReordering With TestReordering
10822 11072 11378 70 . 209 . 35 . 11 . 60 . 19 .
20
Cost reduction Past data for 187
patients
250
W/O Reordering
200
With Reordering
150
100
50
0
10822
11378
11072
21
Cost reduction Current data for 169
patients
ClinicalTrial Mean Cost Mean Cost
ClinicalTrial W/O TestReordering With TestReordering
11971 12601 12757 192 . 36 . 107 . 192 . 3 . 107 .
22
Cost reduction Current data for 169
patients
250
W/O Reordering
200
With Reordering
150
100
50
0
11971
12757
12601
23
Conclusions
We have developed a system thatselects
prospective trial participants.

• Helps physicians to identify eligible patients

• Can increase the number of trial participants

• Can reduce the cost of the selection process

24
Future work

• Deploy the system at the Moffitt Cancer Center

• Evaluate the satisfaction of physicians and
  nurses

• Add probabilistic reasoning