Transcript Dianne O`Connell - Record Linkage

Using linked data for assessing
patterns of cancer care
Dianne O’Connell
David Goldsbury
POC Study Teams
Cancer Epidemiology Research Unit
Cancer Council, NSW
Overview
• Patterns of cancer care studies
• Use of linked records
• Validation of linked data sets for patterns
of care
• Possible analyses of linked data sets
• Other research using linked records
Patterns of Cancer Care Studies
•
•
•
•
Describe treatment patterns
Compare management with guidelines
Assess accessibility to care
Identify inequities in care
Methods
• Ad hoc data collections
• Clinical Cancer Registries
• Record linkage of routinely collected
(administrative) data sets
Patterns of cancer care studies
• Lung, colorectal, prostate cancers in
NSW
• Methods
– Identify patients through population-based
NSW Central Cancer Registry (CCR)
– Consent from doctors and patients to pass
details on to researchers
– Informed consent from patients
Patterns of care studies –
methods
• Questionnaires from treating doctors
– Find correct doctor
– Obtain treatment and referral information
– Field collection where necessary
– Response rates
• colorectal 88-97%
• lung 62%
• prostate 64%
• Participants diagnosed 1999-2002
Patterns of care studies
- limitations
•
•
•
•
•
Resource intensive
Consent rates from patients
Response rates from doctors
Relies on doctors’ clinical notes
Data represents a snapshot and soon out
of date
Is there an easier way?
• Cancer registry information in Australia
does not include treatment
• Hospital discharge records will not capture
or identify all relevant cases of cancer
• Combined, they may be more useful
Use of administrative datasets (1)
• NSW Central Cancer Registry (CCR)
• Population-based
– Cancer is notifiable due to Public Health Act
– Hospitals, pathology labs, radiation
oncologists, nursing homes, deaths
Use of administrative datasets (2)
• NSW Admitted Patient Data Collection
(APDC)
– All hospital separations (discharges,
transfers, deaths)
– All NSW public and private hospitals and
day procedure centres
Record linkage
NSW Central
Cancer Registry
1993 - 2002
Admitted Patient
Data Collection
July 1992 - Jun 2003
• Linked by NSW Health in 2005
• 86% of CCR cases linked to APDC
• Procedures & comorbidities identified in APDC records
Variables
• CCR
– sex, age at diagnosis, health area and SLA
of residence, date of diagnosis, best method
of diagnosis, spread of disease at diagnosis,
cause of death (if dead), survival time
• APDC
– sex, age, health area and SLA of residence,
health area of treatment, type of hospital,
date of admission, date of separation,
procedures, principal diagnosis, additional
diagnoses, health insurance status on
admission
Analysis issues
• One record per hospital episode, each
with multiple procedure/diagnosis codes
• Assign relevant formats, SES and
accessibility/remoteness categories
• Identify procedure/diagnosis codes for
each type of treatment
• Summarise!
Validation of CCR-APDC data
• Cancer Council patient surveys were linked
to CCR-APDC data set
• Linked, routinely collected data validated at
individual patient level
• Usefulness of these data for patterns of
cancer care studies assessed
Validation data
Prostate, colorectal and
lung cases for linkage
(n=7425)
No CCR link
(n=206)
Link to CCR
(n=7219)
Survey treatment info
(n=6643)
No survey
treatment info
(n=576)
No APDC link
(n=516)
Link to APDC
(n=6127)
Validation data
• Overall: 6127 cases with data from patient
surveys and administrative records
– Prostate: 1591 cases
– Lung: 1580 cases
– Colorectal: 2956 cases
Validity of surgical data
Percentage of linked cases
100%
Survey
CCR-APDC
80%
84%
Sensitivity Specificity
63%
60%
92%
60%
40%
17% 16%
20%
0%
Radical
prostatectomy
Lung resection
Colorectal
resection
Prostate
96%
100%
Lung
92%
99%
Colorectal
90%
86%
Surgery records missed
Percentage of shortfall
100%
80%
No surgery or matching admission
Matching admission, surgery not recorded
After end of APDC follow-up
79%
60%
60%
40%
40%
20%
14%
7%
0%
0%
Radical prostatectomy (n=42)
Lung resection (n=20)
• Extra cases in admin data: 2 for prostate, 15 for lung
Validity of chemotherapy data
Percentage of linked cases
100%
80%
Sensitivity Specificity
Survey
CCR-APDC
60%
40%
N/A
N/A
Lung
39%
96%
Colorectal
40%
97%
35%
30%
20%
Prostate
15%
16%
N/A
0%
Prostate
Lung
Colorectal
• Generally an outpatient procedure, no admission recorded
Validity of radiotherapy data
Percentage of linked cases
100%
80%
Sensitivity Specificity
Survey
CCR-APDC
60%
39%
40%
Prostate
32%
99%
Lung
30%
96%
Colorectal
14%
98%
30%
20%
10%
14%
11%
3%
0%
Prostate
Lung
Colorectal
• As with chemotherapy, often an outpatient procedure with no
admission
Prostate radiotherapy data
Percentage of linked cases
50%
Survey
CCR-APDC
40%
30%
Sensitivity Specificity
30%
27%
20%
10%
All XRT
32%
99%
EBRT
18%
99%
Brachytherapy
87%
100%
10%
8%
7%
3%
0%
Radiotherapy
EBRT
Brachytherapy
• Brachytherapy involves general anaesthetic, often a hospital admission
Investigations: Lung
Percentage of linked cases
100%
98%
Survey
CCR-APDC
80%
Sensitivity Specificity
60%
51%
36%
40%
42%
43%
27%
20%
0%
Chest scan
Bronchoscopy
Biopsy
Chest scan
27%
60%
Bronchoscopy
65%
95%
Biopsy
64%
72%
Investigations: Colorectal
Percentage of linked cases
100%
80%
79%
83%
Survey
CCR-APDC
Sensitivity Specificity
60%
55%
52%
40%
28%
20%
6%
0%
Colonoscopy
Abdo-pelvic scan
Chest scan
Colonoscopy
92%
51%
Abdo-pelvic scan
30%
79%
Chest scan
5%
93%
Validation conclusions
• Linked routinely collected data useful for:
– Major surgical procedures
– Other inpatient procedures (e.g. brachytherapy)
• Complementary data sources required for:
–
–
–
–
Chemotherapy
Radiotherapy
Investigative procedures
Comorbidities
• Medicare Australia information to improve data coverage
– MBS: referrals, diagnostic and therapeutic procedures
– PBS: medicines
What else can we analyse using
these CCR-APDC data?
• Treatment trends over time
• Time from diagnosis to treatment
• Distance travelled for treatment (approx.)
• Changed address at treatment
• Factors associated with treatment
– Rural/urban, socioeconomic, age, insurance
• Survival after treatment
Patterns of surgical care for prostate
cancer in NSW, 1993-2002: rural/urban
and socio-economic variation
Andrew Hayen et al
ANZ J Pub Health 2008;32:417-420
15
10
5
Age-standardised percent
Crude percent
0
Per cent
20
25
Radical prostatectomies in NSW
1994
1996
1998
Year of diagnosis
2000
2002
Radical prostatectomy
Area of residence – urban
rural
Socioeconomic group
1 (least disadvantaged)
2
3
4
5 (most disadvantaged)
RR*
1
0.69
1
0.89
0.93
0.87
0.83
95% CI
0.65-0.73
0.85-0.94
0.89-0.98
0.82-0.92
0.78-0.89
* Adjusted for age, spread of disease and year of diagnosis
Time from diagnosis to treatment
Time to surgery for lung cancer
Percentage of lung surgery cases
60%
CCR-APDC
52%
50%
40%
34%
30%
20%
10%
10%
4%
0%
Within 1 month of 1-3 months after 4-12 months after >12 months after
diagnosis
diagnosis
diagnosis
diagnosis
Using month of diagnosis from CCR
CCR-APDC dataset - advantages
•
•
•
•
Availability of data
Relatively cheap
Includes large numbers of individuals
Ongoing data collection – monitoring
CCR-APDC dataset - limitations
• Incomplete coverage of chemotherapy
and radiotherapy
• Doesn’t cover pathways to diagnosis and
referral patterns or outcomes
• Lack of disease clinical detail (NSW –
crude disease staging)
• Incomplete matching – no hospital record
 no treatment
• Cross-border patient flows
Other research with linked records
• Descriptive patterns of care studies for other
cancer types
• Medicare and pharmaceutical benefits data to
improve treatment coverage
– Colorectal cancer referral pathways study
• Survival follow-up analysis for previous POC
studies
• Hepatitis B and C linked with Cancer Registry
(National Centre in HIV Epidemiology and Clinical
Research)
• Cancer Registry linked to Midwives Data
Collection (Cancer Institute NSW)