Stage-specific survival of screen-detected versus clinically diagnosed colorectal cancer

- evidence from the FOBT screening trials-

Iris Lansdorp-Vogelaar

Janneke Wilschut

Marjolein van Ballegooijen

Methods and Applications for Population Based Survival

Frascati, 20 September 2010

Outline

Background

Microsimulation modeling and length and lead-time bias

Analysis

Results

Future work

Colorectal Cancer

Colorectal cancer (CRC) 2nd leading cause of cancer death worldwide

CRC develops through adenoma-carcinoma pathway:

NormalColorectum

SmallAdenoma

ColorectalCancer

AdvancedAdenoma

FOBT Screening

Three randomized trials showed 15-33% reduction in CRC mortality

from fecal occult blood testing (FOBT)

Mortality reduction assumed to be result of more favorable stage

distribution with screening

Mapp et al. found different survival between screendetected CRC and

CRC in control group after correcting for stage (Mapp et al, Br J Surg; 1999)

Lead-time and length bias

Lead-time bias: longer survival of screendetected CRC because of

earlier detection and not by later death

Length bias: longer survival of screendetected CRC because slower-

growing tumors are detected by screening

Correcting for lead-time and length bias

Kafadar & Prorok: Compare survival of cases in screen and control

groups of randomized trial using time since entry of the trial (Kafadar & Prorok, Stat Med; 1994)

Key assumption:

Cases in two groups are comparable

Limitations:

No correction for overdiagnosis

Comparison stage-specific survival not possible

Research objective

To test hypothesis that stage-specific survival of screen-detected CRC

is the same as of clinically diagnosed CRC.

Microsimulation modeling of colorectal cancer

Simulation of a life-history

Simulating the effect of screening

Microsimulation modeling & lead-time bias

Microsimulation modeling & length bias

Screening Intervention

Validation of MISCAN-Colon model

Used model to try and reproduce results of randomized trials of

Minnesota, Nottingham and Funen simultaneously

Model was adjusted to account for differences in demography,

background incidence, and trial design

The model with a higher sensitivity shortly before clinical diagnosis

gave the best fit

This model reproduced CRC incidence and detection rates by stage

well

Analysis

Use validated MISCAN-colon model that reproduces observed

incidence and CRC detection by stage for three trials

Assume same stage-specific survival for screendetected and clinically

diagnosed CRC

Compare simulated mortality reduction with observed for three trials

Results

Observed mortality

reduction

Simulated mortality

reduction

Minnesota, annual

screening

32.5% 20.6%

Minnesota, biennial

screening

17.3% 11.3%

Nottingham 13.4% 5.1%

Funen 17.8% 8.9%

First approach to modeling within stage shift

Model validation suggested higher screendetection in the stage in

which the cancer would have been diagnosed in the absence of

screening than in earlier stages

Of the screendetected cancers, the cases that are detected in the same

stage as they would have become clinical, are the most likely

candidates for better survival because of within stage-shift

Survival assumptions for within stage shift

Assumed following survival for these screendetected cancers:

Survival in stage I is 100%

Survival in stage II = Survival in stage I of clinical cases

Survival in stage III = Survival in stage II of clinical cases

Survival in stage IV of these cases was not improved

Results with within-stage shift

Observed mortality

reduction

Simulated

mortality reduction

(no within shift)

Simulated

mortality reduction

(within shift)

Minnesota, annual

screening

33% 20.6% 33.5%

Minnesota, biennial

screening

21% 11.3% 21.2%

Future work

Current approach for effect of within-stage shift quite arbitrary

Estimate effect of within-stage shift through hazard ratio for survival of

cancers screendetected in same stage as clinical diagnosis

Explore alternative approaches to obtain estimate for improvement that

is independent of screening intensity

Conclusions

The improvement in stage-distribution from FOBT screening is

insufficient to explain the observed mortality reduction

Even after correcting for lead-time en length bias, stage-specific

survival of screendetected cases needs to be better than of clinically

diagnosed cases