Diapositive 1

1
How Statistics Explain What Cancer Is Bernard
Junod, MD, MPH, MS in Epi Ecole des Hautes
Etudes en Sante Publique EHESP School of Public
Health, France Senior lecturer Visiting Scholar
at UCLA Colloquium, University of Colorado
Colorado Springs March 19th 2009
2

• Conflict of interests of Bernard Junod
• None
• Salary from EHESP School of Public Health
• Professional purpose 
• To raise debate for improving peoples health
• Signatory of the Chart of FORMINDEP acting for
• independent training of health professionals

3
Objective

• To identify how to use statistics for providing
  better understanding of cancer
• To clarify when statistics tell us about the
  nature of cancer
• (what cancer is and what cancer is not)

4
What is cancer ?

• A cause of death
• A disease
• A public health issue
• Understanding what cancer is aims to avoid it

5
How statistics explain what cancer is
Statistical study Examples provided
should be

• Cause of lung cancer
• Questioning screening
• Smoking and lung cancer
• Breast cancer surgery and
• breast cancer screening
• Use of cancer statistics for improving peoples
  health

Ambitious Valid Correctly interpreted
6
Use of statistics in epidemiology
Epidemiology Study of the distribution and
determinants of disease
frequency in man (B. McMahon)
Real world
4. Interprete for use
1. Observe
Experimental data
Descriptive data
Comparison observational data
2. Formulate hypotheses
3. Formulate a model
7
Lung CancerDescriptive Data

• 1950 Doll and Hill observation in UK
• Increasing lung cancer mortality
• More than 90 smokers in hospitals
• Hypothesis
• Link between tobacco smoking and lung cancer

8
Comparison of observed facts

• Design Follow-up of male medical doctors in UK
• Risk factor Tobacco smoking (by questionnaire)
• Outcome Lung cancer death (by death certificate)
• Result
• Non smokers 7 deaths for 100000 person-years
• Smokers 130 death for 100000 person-years
• Dose-effect relationship
• Risk reduction among those who quit smoking

9
Public health use of the results obtained by Doll
and Hill in 1956 (UK)

• 1991, US Smoking bans in hospitals under
• the pressure of Smoking or
  Health
• 2008, France Smoking bans in public institutions

10
How statistics explain that lung cancer isa
smokers cancer
11
Breast Cancer SurgeryDescriptive Data

• 120 years ago
• 2 years survival for breast cancer 40-50
• 1892
• Halsted introduces a new surgical intervention
• for breast cancer by including
• Extended removal of breast
• Removal of a large muscle behind the breast
• Removal of lymph nodes in armpit

12
Halsteds publication in the American Journal of
Surgery - 1894

• Title
•  THE RESULTS OF OPERATIONS FOR THE CURE OF
• CANCER OF THE BREAST PERFORMED
• AT THE JOHNS HOPKINS HOSPITAL
• FROM JUNE,1889, TO JANUARY,1894 
• First sentence of Halsteds paper
•  In fifty cases operated upon by what we call
  the complete method we have been able to trace
  only three local recurrences. 

13
Breast Cancer SurgeryComparison data

• Study design
• Outcome measured occurrence of death
• Control group life table of 651 women with
  breast cancer without surgical treatment
• Intervention group women in Halsteds series
• with a potential follow-up of at least 2 years
• Analysis compare observed death in Halsted
  series with expected death from life table

14
Untreated group - life table651 women described
in 1926 by J. GreenwoodNatural Duration of
Untreated Breast Cancer From Onset of Symptom to
Death
15
Comparison of Halsteds study with an untreated
group

• According to case description in Halsteds
  paper
• From 1889 to 1892, 25 women underwent surgery
• Follow-up available for 24/25 women (mean 26
  months)
• Results
• Observed death after surgical intervention 15
• Expected death without surgery (Greenwood) 10.0
• Conclusion
• Survival is not improved by Halsteds surgery
• (B. Junod, 2009)

16
Comparison studies for breast cancer treatment

• 1975-85 results of Fisher and Veronesi
• Randomized controlled trials comparing
• types of surgical intervention
• Use of radiotherapy

17
No advantage of radical mastectomy
Lower survival with node involvement
18
Use of published results on breast cancer
surgery Due to Halsteds belief and/or to
surgeons pride

• Up to 1980
• Extensive use of Halsteds radical mastectomy
• Up to 2008
• Belief in Halstedian model on cancer
• cancer spreads out through lymph vessels
• Screening policy supported by better survival if
  cancer detected without node involvement

19
Experimental data on breast cancer screening

• Study design
• Outcome measured breast cancer death due to
  metastases in vital organs
• Randomized allocation of women in either
• A control group usual care without special
  invitation
• An Intervention group invitation every second
  year to mammography screening
• Identical follow-up in both groups for comparing
  breast cancer death occurrence

20
Mammography screening trials according to
methodological quality
21
Cumulative breast cancer death after 7 years in
the best randomized controlled trials
Malmö n 42283 Canada I n 50430 Canada
II (ref palpation) n 39405 Total about
900000 person-years 7 years after beginning of
screening

• Study (date of publication) Invited
  to mammography screening
• yes no
• Malmö (1988) 44 38
• Canada I (1992) 38
  28
• Canada II (1992) 38 39
• Total, 7 years 120 105

An audit confirmed that no bias explained
larger cancer mortality when screened
22
A problem of logic

• Evidence of better survival if patients are
  detected early or without node involvement
  favors screening
• Evidence of similar occurrence of breast cancer
  death among women screened and not screened does
  not favor screening
• What is wrong ?

23
Zahl, Welch et al Arch Int Med 2008 Evidence of
overdiagnosis

• Outcome occurrence of breast cancer diagnosis
• NB Groups compared should have identical rate
• 119472 women invited 3 times to mammography
  screening. Rate after 6 years 1909 per 100000
• 109784 women invited only once at the end of
  6 years of follow-up. Rate 1564 per 100000
• Diagnoses in excess when 3 invitations, hence

Overdiagnosis of breast cancer exists
24
Interpretation of Zahls result
Examination by histology may provide diagnoses of
breast cancer that regress spontaneously
25
What was wrong? Diagnosis validity!

• Comparison of patients survival is subject to
  bias if cancer is not correctly diagnosed
• When node involvement does not exist, breast
  cancer diagnosis is less reliable and all false
  positive cases improve survival
• Comparison of populations in RCT trials do not
  suffer from this bias.

26
Factors leading to overdiagnosis

• Large scale intervention aiming to cure
• Faith in the validity of examination by histology
• Wrong beliefs based on Halstedian model
• If cancer is to be cured, an early diagnosis must
  be made
• Surgery may cure invasive breast cancer
• Cure is more likely for early than for late
  surgery

27
Scientific Conclusion

• Cancer diagnosis based on tumor sample examined
  by histology is not fully reliable
• If true progressive cancer exists, it is reliable
• If there is no true progressive disease, it may
  give a false cancer diagnosis overdiagnosis
• Overdiagnosis increases with screening
• Early treatment does not improve survival up to 7
  years following the first screening

28
Cancer death and diagnosis for lung, breast and
prostate France 1980-2005
Death
Diagnosis
Cases per year
Lung (no screening)
Breast (increasing screening)
Prostate (Increasing screening)
60000
50000
40000
30000
20000
10000
Calendar year
Calendar year
Calendar year
Epidemic of overdiagnosis many times the AIDS
epidemic, just for breast cancer
29
Consequences of overdiagnosis on cancer research

• Example
• Diagnosed lung cancer at screening
• and tobacco consumption
• Screening by spiral tomography followed by
  bronchoscopy was proposed to thousands of
  Japanese volunteers from 1995 to 1997

30
Results of lung  cancer  screening
Expected number of lung cancer death in the
screened population 10.8 cases Lung cancer
confirmed by histology diagnosed from 1995 to
1997 84 cases
31
Lung cancer screeningInterpretation of the
results

• Screening by spiral tomography provides 8 times
  more lung cancer diagnoses than expected lung
  cancer deaths
• Overdiagnosis conceals the link between true
  progressive lung cancer and smoking
• Cancer research is obscured by overdiagnosis

32
Consequences of screening

• Epidemic of overdiagnosed cancer
• Overdiagnosis impedes research on cancer
• Probable stimulation of metastases due to early
  biopsy in true progressive cancer
• 3rd consequence is hidden by overdiagnosis

33
Statistics explain what cancer is when health
experts make right use of them!

• Ambitious questions require humble experts It
  is acceptable to say we do not know
• Validity of a study requires genuine experts
• Use of research for public policy on screening
  requires that decision makers be independent of
  financial and theoretical conflicts of interest

34
Coming back to everyday life

• Good use of statistics on cancer requires
• patients choosing doctors who prefer
• First do not harm! versus Act now!
• Patients health versus dollars
• To help patient make informed decisions versus
  cancer war