theo georghiou and dr jessica sheringham: data and colorectal cancer, 30 june 2014
DESCRIPTION
In this slideshow, Dr Jessica Sheringham, Visiting Fellow, and Theo Georghiou, Senior Research Analyst, Nuffield Trust describe what linked data can tell us about the GPs role in diagnosing colorectal cancer. Dr Jessica Sheringham and Theo Georghiou spoke at the Nuffield Trust event: The future of the hospital, in June 2014.TRANSCRIPT
© Nuffield Trust June 2014
What can linked data tell us about GPs’
role in diagnosing colorectal cancer?
30 June 2014
Jessica Sheringham & Theo Georghiou
© Nuffield Trust
Outline
Background: Why colorectal cancer?
What we did
• Aims & setting
• Linkage
• Constructing & examining routes to diagnosis
Illustrative findings
Discussion
• Colorectal cancer
• Wider applications
© Nuffield Trust 16 July 2014 © Nuffield Trust
Background
© Nuffield Trust
Why colorectal cancer?
4th most common cancer in UK
Incidence increasing
Most common in older people
55% overall survive 5 years after diagnosis
Survival much better if diagnosed at an early stage:
• 5-year survival: early stage (“Dukes Stage A”) = 93%
• 5-year survival: late stage (“Dukes Stage D”) = 6.6%
References: CRUK, 2014; NCIN data briefing, 2009
© Nuffield Trust
Reference: Coleman et al. Lancet 2011
Colorectal cancer: Age-standardised 1-year and 5-year
relative survival trends 1995–2007, by cancer and country
© Nuffield Trust
Improving outcomes for colorectal cancer: points for
intervention
Screening
Symptom awareness
Patients & public Prevention
© Nuffield Trust
Improving outcomes for colorectal cancer: points for
intervention
Prevention
Screening
Symptom awareness
Patients & public Secondary care
Access to effective
treatment
Diagnosis
© Nuffield Trust
Improving outcomes for colorectal cancer: points for
intervention
Diagnostic referrals
Primary care
• 2-week wait referral pathway underpinned by NICE guidance
• Decision support tools e.g. RATs(Hamilton 2013), Qrisk (H-Cox 2012, Collins 2012)
BUT
• Only 24% diagnosed on 2-week wait (2WW) pathway, 24% diagnosed
as emergencies(Thorne et al. 2006)
• Existing monitoring strategies, e.g. audit, reliant on GP/practice
participation – could underestimate variation
Access to effective
treatment
Patients & public Secondary care Prevention
Screening
Symptom awareness
Diagnosis
© Nuffield Trust 16 July 2014 © Nuffield Trust
The project
© Nuffield Trust
Project
Aim: Explore the feasibility of examining quality of diagnostic process across the patient pathway using routinely available data
Objectives
1. Establish whether linkage of three datasets (primary care, secondary care and cancer registry) possible
2. Apply chosen candidate indicator(s) of quality to examine variations in diagnostic process to identify points for intervention at patient or population level
© Nuffield Trust
Time-based
• Patient interval: symptoms to
presentation
• Primary care interval:
presentation to diagnosis
• Secondary care interval:
diagnosis to treatment
Event-based
• Stage at diagnosis
• Route: emergency diagnosis
• Short-term survival
Candidate indicators: How measure the quality of the
diagnostic process?
Reference: Lyratzopoulos, 2014
© Nuffield Trust 16 July 2014 © Nuffield Trust
Methods development
© Nuffield Trust
Project setting: Outer North East London
1m population & 4 diverse boroughs
RB WF
B&D
HV
Havering
(HV)
Waltham
Forest (WF)
Reference: borough profiles, www.london.gov.uk
% Population over 65 (2011)
Income support claimants (2013)
Redbridge
(RB)
RB WF
B&D
HV
Barking &
Dagenham
(B&D)
© Nuffield Trust
Datasets and linkage
Key data:
Date of cancer diagnosis
Stage of cancer
Colorectal cancer
diagnoses
Four CCGs
2009 – 2011
N = 1,367
Cancer registry data from Public Health England
© Nuffield Trust
Datasets and linkage
Key data:
Date of cancer diagnosis
Stage of cancer
Colorectal cancer
diagnoses
Four CCGs
2009 – 2011
N = 1,367
All cancer
diagnoses
2005 – 2010
Cancer registry data from Public Health England
© Nuffield Trust
Datasets and linkage
Identify and remove prior
cancers
Colorectal cancer
diagnoses
N = 1,367
All cancer
diagnoses
Cancer registry data from Public Health England
© Nuffield Trust
Datasets and linkage
Colorectal cancer
Diagnoses
2009-2011
N = 1,150
Cancer registry data from Public Health England
Colorectal cancer diagnosis,
no prior cancer
© Nuffield Trust
Datasets and linkage
Colorectal cancer
diagnoses
2009-2011
N = 1,150
GP and Hospital data from CCGs
For population with recorded colorectal cancer
diagnosis during 2007-2012
GP data
Four CCGs (registered)
2007-2012
Hospital data:
inpatient, outpatient, A&E
Key data:
Socio demographic information (e.g. age, gender, deprivation)
Hospital contacts & procedures
GP contacts & Read codes (GP recorded symptoms and activities)
© Nuffield Trust
Datasets and linkage
Colorectal cancer
diagnoses
N = 1,150
GP data
Hospital data:
inpatient, outpatient, A&E
GP and Hospital data from CCGs
© Nuffield Trust
Datasets and linkage
Colorectal cancer
diagnoses
N = 1,150
GP data
Hospital data:
inpatient, outpatient, A&E
Not all individuals with diagnosis found in CCG data
© Nuffield Trust
Colorectal cancer
diagnoses
2009-2011
N = 943
Datasets and linkage
GP data
At least 21 months prior to diagnosis
Hospital data:
inpatient, outpatient, A&E 82% of Registry
records ‘matched’
local data
‘Unmatched’: high
% missing stage
and higher % of
patients over 90
years
© Nuffield Trust
Assigning a ‘route’ to diagnosis
1. Looked back at patient records 6 months
(starting from the hospital episode closest to date of diagnosis)
Reference: Elliss-Brookes et al, 2012
© Nuffield Trust
© Nuffield Trust
Assigning a ‘route’ to diagnosis
1. Looked back at patient records 6 months
(starting from the hospital episode closest to date of diagnosis)
2. Examined previous activity and referral source
(refined to exclude activity NOT connected with colorectal cancer)
Reference: Elliss-Brookes et al, 2012
© Nuffield Trust
© Nuffield Trust
Referral source
= “GP 2WW”
© Nuffield Trust
Assigning a ‘route’ to diagnosis
1. Looked back at patient records 6 months
starting from the hospital episode closest to date of diagnosis
2. Examined referral source and previous activity
Refined to exclude activity NOT connected with colorectal cancer
3. Assigned each patient to one of four routes to diagnosis:
Emergency
GP – urgent/2WW
GP – routine/unknown
Consultant, other, unknown
Reference: Elliss-Brookes et al, 2012
© Nuffield Trust
Analysis at population and individual levels
1. POPULATION: Logistic regression to identify factors associated with
increased chance of emergency presentations
• Cancer stage at diagnosis: early, vs late/missing
• Consultation characteristics:
• no. GP visits
• relevant symptoms (using Read Codes in GP records: anaemia, rectal
bleeding, diarrhoea, constipation, abdominal pain, other, incl. weight loss,
fatigue other altered bowel)
• Patient demographics: age, gender
• Area: borough, deprivation
2. INDIVIDUAL: Characteristics of pathways within each route
© Nuffield Trust 16 July 2014 © Nuffield Trust
Illustrative findings
1. Cohort
2. Population level
3. Individual level
© Nuffield Trust 16 July 2014 © Nuffield Trust
Illustrative findings
1. Cohort
2. Population level
3. Individual level
© Nuffield Trust
Diagnostic route in our cohort vs. other estimates
31
52
19
26 24
24 24
0%
20%
40%
60%
80%
100%
Cohort Thorne et al
Emergency
GP urgent/2WW
Alternative route(Consultant/other/unknown)
Alternative route (GProutine/unknown)
Cohort, n=943 Thorne et al (2006)
© Nuffield Trust 16 July 2014 © Nuffield Trust
Illustrative findings
1. Cohort
2. Population level
3. Individual level
© Nuffield Trust
Characteristics of emergency presentation vs. other routes
Symptoms
Ref:
no symptom
Stage
Ref: early
Age
Ref: 60-69y
Borough
Ref: “2”
Area
deprivation
Ref: Most
deprived 20%
Ad
juste
d o
dd
s r
atio
0.01
0.1
1
10
"Late
"/M
issin
g
Tota
l n
o. G
P v
isits (
12
m b
efo
redia
gnosis
) Abdo
min
al
Co
nstip
ation
Re
cta
l
20
-59 y
70
-79 y
80
+ y 1 3 4
20
-40%
40
-60%
60
-80%
20
% le
ast de
prived
Mis
sin
g
Logistic regression, adjusted for stage, symptoms, age, borough, deprivation and clustering between practices
© Nuffield Trust
Characteristics of emergency presentation (EP) vs. other
routes
Ad
juste
d o
dd
s r
atio
0.01
0.1
1
10
"Late
"/M
issin
g
Tota
l n
o. G
P v
isits (
12
m b
efo
redia
gnosis
) Abdo
min
al
Co
nstip
ation
Re
cta
l
20
-59 y
70
-79 y
80
+ y 1 3 4
20
-40%
40
-60%
60
-80%
20
% le
ast de
prived
Mis
sin
g
Symptoms
Ref:
no symptom
Age
Ref: 60-69y
Borough
Ref: “2”
Area
deprivation
Ref: Most
deprived 20%
Stage
Ref: early
Higher odds of emergency presentation for late stage
cancers is consistent with:
- theory of EP as a marker of diagnostic delay
- other literature (McPhail 2013, Downing 2012)
© Nuffield Trust
Characteristics of emergency presentation vs. other routes
Symptoms
Ref:
no symptom
Stage
Ref: early
Age
Ref: 60-69y
Borough
Ref: “2”
Area
deprivation
Ref: Most
deprived 20%
Ad
juste
d o
dd
s r
atio
0.01
0.1
1
10
"Late
"/M
issin
g
Tota
l n
o. G
P v
isits (
12
m b
efo
redia
gnosis
) Abdo
min
al
Co
nstip
ation
Re
cta
l
20
-59 y
70
-79 y
80
+ y 1 3 4
20
-40%
40
-60%
60
-80%
20
% le
ast de
prived
Mis
sin
g
Fewer GP visits → EP
Abdominal pain & constipation → EP
more common
Rectal bleeding → EP less likely
?clinical manifestation of emergency
cases different?
© Nuffield Trust
Characteristics of emergency presentation vs. other routes
Age
Ref: 60-69y
Borough
Ref: “2”
Area
deprivation
Ref: Most
deprived 20%
Adju
ste
d o
dds r
atio
0.01
0.1
1
10
"Late
"/M
issin
g
Tota
l n
o. G
P v
isits (
12
m b
efo
redia
gnosis
) Abdo
min
al
Co
nstip
ation
Re
cta
l
20
-59 y
70
-79 y
80
+ y 1 3 4
20
-40%
40
-60%
60
-80%
20
% le
ast de
prived
Mis
sin
g
Symptoms
Ref:
no symptom
Age
Ref: 60-69y
Borough
Ref: “2”
Area
deprivation
Ref: Most
deprived 20%
Significant differences by
borough
No significant deprivation
associations
?? Healthcare system
factors??
© Nuffield Trust 16 July 2014 © Nuffield Trust
Illustrative findings
1. Cohort
2. Population level
3. Individual level
© Nuffield Trust
Pathway examples: “Emergency” routes
© Nuffield Trust
Pathway examples: Emergency (2)
© Nuffield Trust
Pathway examples: GP 2WW referred (1)
© Nuffield Trust
Pathway examples: GP 2WW referred (2)
© Nuffield Trust 16 July 2014 © Nuffield Trust
Summary and discussion
points
© Nuffield Trust
Summary
1. Linkage:
• feasible (not quick – cancer data was rate limiting step)
• relatively complete set, (cf 82% cancer cases vs 17% audit participation) BUT
• important biases to consider
2. Routes to diagnosis:
• distinguishing activity from pathways
• POPULATION: important differences between patients, clinical characteristics and boroughs by route to diagnosis
• INDIVIDUAL: diversity of healthcare use can identify cases for indepth audit
© Nuffield Trust
Discussion points and next steps
Variations in colorectal cancer diagnostic pathways can be identified using routine data:
• Identifies a) local targets for intervention b) specific cases for indepth audit
Next steps
• Refine measures/criteria to identify cases for indepth audit
Transferable methods, approaches to other clinical areas
• Challenges of defining diagnostic interval
• Pros and cons of pathways analysis
© Nuffield Trust
www.nuffieldtrust.org.uk
Sign-up for our newsletter www.nuffieldtrust.org.uk/newsletter
Follow us on Twitter: Twitter.com/NuffieldTrust
© Nuffield Trust
Acknowledgements:
• Xavier Chitnis, The Royal Marsden NHS Foundation Trust
• Dr Martin Bardsley, Nuffield Trust
• Knowledge & Intelligence Team (London), Public Health England: Neil
Hanchett and Ashu Sehgal
• Rob Meaker, Phil Kozcan, Outer North East London CCGs
• Stuart Bond, Health Analytics
Acknowledgements
© Nuffield Trust
Pathway examples: GP 2WW referred (3)
© Nuffield Trust
Pathways: Consultant/other examples