Transcript Hurry Up and Wait: The Effect of Delayed Treatment on

Hurry Up and Wait:
The Effect of Delayed Diagnosis and
Treatment on Survival in Patients with
Non-Small-Cell Lung Cancer
Michael K. Gould, MD, MS
VA Palo Alto Health Care System
Stanford School of Medicine
Lung Cancer



175,000 new cases in U.S. in 2004
160,000 deaths in U.S. in 2004
More deaths than breast, prostate and
colon cancer combined
Jemal et al. CA Cancer J Clin 2004;54:8-29

Common in veterans

6,600 cases in 2003 (~20% of all cancers)
VA Central Cancer Registry: http://www1.va.gov/cancer/index.cfm
Lung Cancer Histology
9%
19%
45%
27%
SEER: http://seer.cancer.gov
adenocarcinoma
squamous
small cell
large cell
Evaluation in Suspected Lung Cancer

Diagnosis



Staging





Imaging tests (e.g. CXR, chest CT, PET)
Biopsy (e.g. bronchoscopy, TTNA)
Imaging tests (e.g. brain CT or MR)
Biopsy (e.g. mediastinoscopy, adrenal Bx)
Pre-operative assessment (PFTs, cardiac eval)
Consultations
Tumor Board
Research Agenda: Lung Cancer

Defining Best Practices:










Examining Current Practices:
Cost-effectiveness of low-dose
 Quality of practices for lung
CT for lung cancer screening
cancer diagnosis and staging
(with CanCORS)
Accuracy of FDG-PET for SPN
diagnosis
Cost of FDG-PET
 Aligning Current and Best
Cost-effectiveness of tests for
Practices:
SPN management
 Development, validation and
Predictors of mediastinal
evaluation of a computer-based
metastasis
decision support system for
managing SPN
Accuracy of CT and FDG-PET
for staging in NSCLC
Accuracy of TBNA for staging in  Eliciting preferences for
NSCLC
shared decision making in
Accuracy of mediastinoscopy
patients with lung nodules
for staging in NSCLC
Cost-effectiveness of tests for
staging in NSCLC
CanCORS



NCI-funded collaboration
Population based, prospective cohort
study of practices and outcomes in
patients with lung and colorectal cancer in
diverse geographic regions of U.S.
8,000 lung cancer patients, including
1,000 U.S. veterans with lung cancer
enrolled at 13 sites
Specific Aims: Wait Times



Describe variation in time to diagnosis and
treatment in U.S. veterans with non-small cell
lung cancer (NSCLC)
Identify facilitators and barriers to timely
diagnosis and treatment in VA
Examine the effect of delayed diagnosis and
treatment on stage distribution and survival
Why Measure Wait Times?



Longer wait times contribute to emotional
distress of patients and family members
Longer wait times may lead to missed
opportunities for cure and/or effective
palliation
Longer wait times may (arguably) result in
increased health care costs
Guidelines for Wait Times

RAND Quality Indicators


Diagnosis within 2 months of presentation
Treatment within 6 weeks of diagnosis
http://www.rand.org/publications/MR/MR1281/

BTS



Referral & evaluation by respiratory specialist within 2-7
days
Results of diagnostic test communicated within 2 weeks
Thoracotomy within 8 weeks, palliative XRT within 4
weeks, radical XRT within 2 weeks, chemotherapy within
2 weeks
Thorax 1998;53(Suppl 1):S1-8.

ATS, ACCP, CCO: No recommendations
Prior Research

Type and length of delay




n=17 studies between 1989 to 2004
Heterogeneous patient populations
Most studies from Europe, 3 from North America, 1
from Japan
Effect of delay on lung cancer outcomes




n=11 studies between 1993 and 2004
4 studies of surgical patients (1 from U.S.)
2 studies of delays following screen-detection of lung
cancer in Japan
1 European study of patients referred for curative XRT
Prior Research: Length of Delay
Interval
# of Studies
Median Time
Symptom to first contact
5
~3 weeks
First contact to diagnosis
6
First contact to treatment
5
2-6 weeks
( 1 study >12 weeks)
~3 months
Diagnosis to radiation
2
5 to 6 weeks
Diagnosis to surgery
1
7 weeks
Waiting for Cancer Surgery
Simunovic et al. CMAJ 2001;165:421-5.
Waiting for Cancer Surgery




One U.S. study from SFVA (retrospective)
83 veterans with stage I or II lung cancer
Underwent surgical resection between
1989-99
Median time from initial contact to
resection was 82 days
Quarterman et al. J Thorac Cardiovasc Surg 2003;125:108-14.
Median Wait Times for Radiation
and Chemotherapy

Ontario, Canada
1 to 4.1 weeks from referral to radiation
 1.9 to 6.3 weeks from referral to
chemotherapy

http://www.cancercare.on.ca/access_waitTimes.htm

No data from U.S.
Predictors of Delay

Longer symptom delay in patients <45 years old
Bourke et al. Chest 1992;102:1723-9.

Age not related to diagnostic or treatment delay
Deegan et al. J Royal Coll Phys London 1998;32:339-43.
Simunovic et al. CMAJ 2001;165:421-5.
Pita-Fernandez et al. J Clin Epidemiol 2003;56:820-5.
Kanashiki et al. Onc Reports 2003;10:649-52.

Gender not related to symptom or treatment delay
Pita-Fernandez et al. J Clin Epidemiol 2003;56:820-5.
Kanashiki et al. Onc Reports 2003;10:649-52.

No data for race/ethnicity, SES, education,
physician or institutional factors
Length of Delay and Outcomes


Delays of 18 to 131 days between diagnostic CT
and XRT planning CT associated with 19% increase
in tumor X-sectional area (range 0% to 373%)
6/29 patients (21%) progressed to incurable
disease while waiting
O’Rourke & Edwards. Clin Oncol 2000;12:141-4.

Delays in patients with screen-detected lung cancer
associated with 2-fold reduction in survival time
Kanashiki et al. Onc Reports 2003;10:649-52.
Kashiwabara et al. Lung Cancer 2003;40:67-72.
Length of Delay and Outcomes

No association between different types of
delay and survival in 4 studies of surgical
patients
Quarterman et al. J Thorac Cardiovasc Surg 2003;125:108-14.
Pita-Fernandez et al. J Clin Epidemiol 2003;56:820-5.
Aragoneses et al. Lung Cancer 2002;36:59-63.
Billing and Wells. Thorax 1996;51:903-6.
Length of Delay and Outcomes:
Stage Distribution
N=103
N=69
P=0.04
N=103
N=69
P=0.02
Christensen et al. Eur J Cardio-thorac Surg 1997;12:880-4.
Research Methods




Retrospective cohort study
129 U.S. veterans with NSCLC
Consecutive patients diagnosed and
treated at VAPAHCS between 1/1/02 and
12/31/03
Median follow-up:
270 days from 1st x-ray abnormality
 194 days from histologic diagnosis
 147 days from treatment

Statistical Methods

Associations between length of delay and
potential predictors of delay
Non-parametric correlations for continuous
predictors
 Pearson chi-square for categorical predictors
 Multiple logistic regression


Associations between length of delay and
survival

Kaplan-Meier, Cox proportional hazards
Patient Characteristics
Characteristic
Age (years)
n=129
67.2 ± 9.5
Gender (Male), %
97.7
White, %
82.4
Tumor size, cm
3.9 ± 2.4
Adenocarcinoma, %
50.0
Squamous cell, %
28.8
Central location, %
55.6
Any symptom, %
58.3
Any CXR finding, %
25.0
SPN, %
18.2
Pre-treatment Imaging Tests
X-ray chest
CT chest
PET
CT abdomen/pelvis
CT brain/spinal cord
MRI head
X-ray bone
MRI spinal cord
MRI chest
N
128
126
107
51
29
23
19
15
10
%
99
98
83
40
22
18
15
12
8
>1 test
30%
11%
3%
PET imaging more common in patients without symptoms
(p=0.02), and those with centrally located tumors (p=0.02) or
malignant solitary nodules (p=0.07)
Pre-treatment Staging Procedures
Bronchoscopy/TBNA
Mediastinoscopy
Endoscopic ultrasound
N
15
7
1
%
12
5
1
>1 test
4%
Mediastinal biopsy more common in patients with primary tumors
that were centrally located (p=0.02) or spiculated (p<0.05)
Treatment Received
Characteristic
%, n=129
Surgery
27.3
Radiation
35.6
Chemotherapy
40.2
No treatment
19.7
Admit within 7 days
33.3
Length of Delay (Days)
Type and Length of Delay
42d
11-117
84d
38-153
22d
8-41
Predictors of Delay <90 days
Characteristic
Delay<90 d (n=67)
Delay>90d (n=62)
66.5 ± 9.8
67.9 ± 9.2
Gender (Male), %
98.5
96.9
White, %
77.4
87.8
4.7 ± 2.8
3.1 ± 1.8
Adenocarcinoma, %
57.4
42.2
Squamous cell, %
25.0
32.8
Central location, %
54.7
56.5
Any symptom, %*
72.1
43.8
Any CXR finding, % †
32.4
17.2
SPN, %*
7.4
29.7
Age (years)
Tumor size, cm*
*p=0.001; † p=0.04
Treatment and Delay
Characteristic
All, %
(n=129)
Delay<90 d, %
(n=67)
Delay>90d, %
(n=62)
Surgery *
27.3
13.2
42.2
Radiation
35.6
41.2
29.7
Chemotherapy
40.2
45.6
34.4
No treatment †
19.7
26.5
12.5
Admit within 7 days *
33.3
48.5
17.2
*p<0.0001; † p=0.04
Longer Treatment Delays in SPN
N=106
116 days
N=23
222 days
P=0.002
Longer Delays in Surgical Patients
N=93
106 days
N=36
208 days
P<0.0001
MV Predictors of Treatment Delay
Predictor
OR
95% CI
Admit within 7 days of 1st abnormal CXR
6.0
2.2 – 16.2
Tumor Size > 3.0 cm
5.4
2.1 – 14.1
Any additional abnormality on CXR
2.6
0.9 – 7.5
Any symptom
2.5
1.0 – 6.0
R2= 0.37; p= 0.82 for Hosmer-Lemeshow test; all correlations< 0.35
ROC Curve for Predictors of Rx Delay
AUC= 0.80;
(0.73 to 0.87);
P<0.0001
Model included admission within 7 days, presence of any
symptom, presence of any additional CXR abnormality, tumor
size, age, sex and race/ethnicity
Predictors of Diagnostic Delay

Independent predictors of diagnosis within
42 days included hospitalization within 7
days (OR 10.3, 95% CI 3.5 to 30), tumor
size greater than 3 cm (OR 5.5, 95% CI
2.0 to 15), and white race (OR 3.0, 95% CI
1.1 to 8.0)
Outcomes: Stage Distribution
Stage
Delay<90 d,
% (n=67)
Delay>90d,
% (n=62)
Stage I
Stage II
All, %
(n=129)
15.9
15.0
9.7
11.3
23.5
19.6
Stage III
Stage IV
32.7
36.3
29.0
50.0
37.3
19.6
P=0.006
Outcomes: Survival

Treatment within 90 days of presentation
associated with an increased risk of death
RR=1.45 (95% CI 79.4% vs. 54.7%)
 P=0.002

Effect of Delay on Survival
Med survival = 321 vs. 122 days,
P=0.001
Med survival = 570 vs. 161 days,
P<0.0001
Multivariable Predictors of Survival



In Cox proportional hazards models, TNM stage
III (HR 11.4, P=0.01) and TNM stage IV (HR
24.0, P=0.001) were the only statistically
significant predictors of survival
Trend towards worse survival in patients with
symptoms (HR 3.1, P=0.08) and patients with
shorter treatment delays (HR 1.5, P=0.09)
Age, ethnicity, tumor size, histology not
associated with survival
Longer Delay=Better Survival
Symptom Delay
Hospital Delay
After adjusting for age, sex, stage & surgery, longer
symptom delay (HR 0.79) and hospital delay (HR 0.87)
were associated with better survival.
Myrdal et al. Thorax 2004;59:45-9.
Sources of Bias and Variation

Sources of Bias
Selection bias
 Confounding by severity of disease
 Lead-time bias


Sources of Variation
Heterogeneous patient populations
 Heterogeneous health care systems

Strategies for Dealing with
Selection Bias

Stratification


Should be performed according to baseline
characteristics
Propensity score methods

Adjust, match or stratify by propensity or likelihood
of receiving intervention/exposure
Connors et al. JAMA 1996;276:889-97.

Instrumental variable methods
Newhouse & McClellan. Ann Rev Pub Health 1998;19:17-34.
McClellan et al. JAMA 1994;272:859-866.
Stratification by SPN
Med survival = 467 vs. 142 days,
P=0.001
P=0.19
Stratification by Surgery
Med survival =478 vs. 142 days,
P=0.001
P=0.08
Propensity Scores

Used to control for selection bias in observational studies
of valve surgery for endocarditis, chemotherapy for
advanced lung cancer, coronary angiography following
acute myocardial infarction and right heart
catheterization for critical illness

Controls for observed differences between groups
Typically use logistic regression to predict use of
intervention
Adjust, match or stratify by propensity to receive
intervention/exposure
5 strata usually sufficient to remove over 90% of bias
due to selection



Effect of chemotherapy on survival
Method
Hazard Ratio
Cox PH
Propensity score
1st
2nd
3rd
4th
5th
0.81
0.78
0.81
0.85
0.80
0.78
Earle et al. J Clin Oncol 2001; 19:10641070.
Stratification by Propensity
P=0.06
P=0.43
Improving Propensity Model in
CanCORS

Patient characteristics



Institutional characteristics





Age, sex, race/ethnicity, education, marital status, SES
Measures of disease severity, sypmtoms and co-morbidity
Lung cancer volume; frequency of thoracic tumor board meetings
Presence of dedicated thoracic surgeon, number of other specialists
Availability of PET scanner, number of CT scanners
Availability of OR time for thoracic surgeons
Other non-clinical factors



Distance of residence to VA
Means test category
Other insurance
Instrumental Variables



Can control for unobserved characteristics
Instrument” should be associated with use
of intervention, but not with health status
or outcome
Example: Heart catheterization following
acute MI—differential distance from home
to hospital with/without cardiac
catheterization lab.
Strengths & Limitations

Strengths




Study sample captured full spectrum of NSCLC
Objective measurement of time intervals avoided faulty
recall
Measurement of survival from time of 1st abnormal CXR
minimized lead time bias
Limitations





Small sample size
Stratification limited statistical power further
Single center limited variability in practices
Retrospective design—unable to assess symptom delay
Not able to fully control for severity at presentation
Conclusions





Important biases complicate the interpretation of
previous studies of delayed treatment in NSCLC
Delays in diagnosis and treatment are longer than
is currently recommended
Patients with aggressive tumors tend to experience
the shortest delays
Reducing delays in patients with malignant SPNs
and other potentially resectable tumors may yield
greatest benefits
Future studies should be large & prospective,
avoid selection & lead time biases, and use
sophisticated methods to account for confounding
by severity of disease at presentation
Acknowledgements

Funding


Collaborators




Advanced RCDA, VA HSR&D Service
David Au, MD, MS
Dawn Provenzale, MD, MS
Sharfun Ghaus
CanCORS Ancillary Study Investigators



Jay Bhattacharya, PhD
Todd Wagner, PhD
Doug Owens, MD, MS
Specific Aims: Staging Practices



Describe variation in use of FDG-PET imaging
and invasive mediastinal biopsy procedures for
staging in U.S. veterans with NSCLC
Examine the effect of PET imaging and
mediastinal biopsy on survival and rate of
thoracotomy without cure in VA
Measure pre-treatment resource utilization and
evaluate the cost-effectiveness of selected
imaging tests and biopsy procedures for lung
cancer staging
Correlations

Age not correlated with time to treatment


Spearman’s rho= 0.10, P=0.26
Tumor size negatively correlated with time
to treatment

Spearman’s rho= -0.32, P<0.0001
Effect of Delay on Survival
Med survival = 321 vs. 122 days,
P=0.001
Med survival = 570 vs. 161 days,
P<0.0001