Automation in Registry Practice

1
Automation in Registry Practice

• Thames Cancer Registry

Jason Hiscox, Stephen Richards, Pam
Acworth Automated Registration Workshop 4th
December 2002
2
Registry Background

• Established 1958 as South Metropolitan CR
• Population based since 1960
• Merged with North Thames 1985
• Database of 2 million registered tumours
• approximately 70,000 new incident cases per year

3
Total Processing Volume
4
Processing Volume by Data Source
5
Savings on Manual Collection
Example Tertiary referral centre with a caseload
of approx. 6000 incident cases per year.
Manual Collection
Electronic Processing
240 wte days (25 records abstracted by tumour
registrar per day)
Abstraction
4 wte days (1 day per quarter)
18 wte days (1 day pre-processing, 8 days
validation correction, 9 days matching and batch
resolution)
80-100 wte days (60-75 registrations per
operator per day)
Entry
6
Achieving Full Automation

• Historically progress has been limited by the
  limited availability to the Registry of good
  quality data from NHS Trusts.
• Would require a minimum fourfold increase in
  batch processing volume. (Approximately
  400,000-500,000 transactions per year as a
  conservative estimate - but could easily be
  double that.)
• Relies heavily on the Registry systems ability
  to effectively scale up to those volumes.
• Requires robust quality assurance and monitoring
  of processes and data quality.

7
Scalability - Pre-requisites
The Key Factors for Successful Scalability

• The Availability of Data
• The Quality of the Data
• Confidence in Processing technology

8
Proportion of records processed without manual
intervention of any kind
9
Quality variation over time for a data source -
approximate equilibrium
10
Quality variation over time for a data source -
quality degradation
11
Supplier specific confidence levels for patient
and tumour matching
12
Validation
You cant have too much validation!

• 120 Single field validations
• 120 Cross field validations
• 40 Post merge nightly QA validation runs
• 100 other ad hoc and periodic QA routines
• Modular reusable validation code designed to
  provide
• consistent support for both automated
  validation and
• manual entry

13
Drill down functionality provides access to
automated data to facilitate QA and build user
confidence through transparency.
14
Lessons Learned

• Automation can be a gradual and cautious process
  - building confidence in the process through a
  series of small steps.
• Where the process needs to be scaled up for
  larger volumes a proactive approach to data
  quality needs to be adopted to ensure that
  problems are picked up as early in the process as
  possible.
• The quality of the data received can
  significantly effect the efficiency and viability
  of automated registration and should be monitored
  carefully.

15
Future development
You cant have too much validation!

• More pre-processing record level validation
• More post processing record level validation
• Pre-commit record level validation
• Standard data quality reports to suppliers
• Full update roll-back (and re-apply)