Examples of Automation in Practice Trent Cancer Registry

1
Examples of Automation in PracticeTrent Cancer
Registry

• Alan Waterhouse
• Assistant Director (IMT)
• Coventry - 4th December 2002

Trent Cancer Registry
2
Once upon a time...

• second generation system not coping
• very expensive
• 70000 backlog (almost 3 years)
• only PAS received electronically
• requirement to develop multiple sources
• poor analytical tools
• registry re-inventing itself

Trent Cancer Registry
3
Transaction Volumes (2001-2)

• Patient Administration 46088
• Pathology 28528
• Extra Regionals 2960
• Cancer Deaths 14391
• Non-Cancer Deaths 6371
• Trace 34778
• Other 246
• (NHS Strategic Tracing Service 30000)

Trent Cancer Registry
4
Overall Process - All Data Sources

• gt99 of transactions electronic
• 60 pass load/validation
• 11 auto create new patient/tumour
• 2 auto create existing patient/ new tumour
• 10 auto amend existing patient/tumour
• gt99.9 of decisions at update automatic
• so around 14 of transactions untouched by hand

Trent Cancer Registry
5
Load/Validation by Data Source
Trent Cancer Registry
6
Load/Validation -PAS
Trent Cancer Registry
7
Load/Validation - Pathology
Trent Cancer Registry
8
Load/Validation - Death
Trent Cancer Registry
9
Tumour Text to ICD Morphology Code
7542 PAS Transactions
Tumour Text to Code Algorithm
78.5 Manual Algorithm First Choice
16.4 No Derivation -gt Manual Coding
4.3 Manual Algorithm Second Choice
0.7 Manual Algorithm Third Choice
94.9 NOT WRONG
5.1 WRONG
Trent Cancer Registry
10
Patient Matching (PAS)
36492 PAS Transactions
Patient Match Algorithm
38 New Patient 11
15 No Decision
47 Existing Patient 3
Manual Search
1 No Decision
Trent Cancer Registry
11
Tumour Matching (PAS) - Then
17177 PAS Transactions
Tumour Match Algorithm
0 New Tumour 19
100 No Decision
0 Existing Tumour 80
Manual Search
lt1 No Decision
Trent Cancer Registry
12
Tumour Matching (PAS) - And Now
20422 PAS Transactions
Tumour Match Algorithm
4 New Tumour 12
40 No Decision
56 Existing Tumour 27
Manual Search
lt1 No Decision
Trent Cancer Registry
13
Cancer Deaths / Pathology

• still all matched against database manually
• because site/morphology not trusted in coded
  form, it is coded manually from text
• recent investigations show that the first cancer
  ICD site mentioned (exc C80) on DC matches at 3
  digit level with manual decision in 84 of cases
• look at disagreement - improve algorithm
• similar situation, but worse, for Pathology

Trent Cancer Registry
14
Other Problem Areas

• Lack of standardisation (codes, defns, rules)
• Code set changes over time
• Mappings between code sets (ICD/Snomed)
• Initial blind faith in the system rules as
  delivered
• Must be constantly vigilant to changes in data
  sources

Trent Cancer Registry
15
Conclusion

• six years experience shows that ..
• a rules based system works and we can extend
  automation but ...
• the rules are difficult to tease out from ...
• our most valuable asset, the Tumour Registrars
• rules vary within E W, and internationally
• registries need to share more

Trent Cancer Registry
16
(No Transcript)
17
Process Steps

• Extract of data from source system
• Receipt and high level quality check
• Filter out previously received transactions
• Load and translate data to standard form
• Validation and correction
• Patient matching
• Tumour matching
• Update/Merge

Trent Cancer Registry