PRACTICAL 1: CONFOUNDING BY INDICATION
SOLUTIONS
(Based on Stattin P et al. Outcomes in Localized Prostate Cancer: National Prostate
Cancer Register of Sweden Follow-up Study. JNCI 2010;102:950-958)
Read the Introduction and methods of the paper
Briefly:
o This observational study aimed at using population-based data to compare
treatment with survival vs. treatment with curative intent (radical prostatectomy
and/or radiotherapy) in patients with localized prostate cancer.
o Out of the 6849 patients included in the study, 2686 had low-risk disease, while the
others has intermediate-risk disease (patients with high-risk disease were excluded)
o Treatment was assigned in clinical practice at the discretion of the treating
physician
Read the results and answer to the questions
The following Table reports the distributions of treatment by risk category:
Question 1. How does the treatment approach differ between the two risk categories?
What is the likely explanation for the difference?
Answer: Patients with low-risk disease are more often treated with surveillance only
(lower risk of death, same risk for side effects of treatment)
1
Overall mortality. The Figure compares Observed overall mortality among patients
treated with surveillance, prostatectomy or radiotherapy with the expected mortality in the
age-matched general population
Expected mortlality
Observed
Question 2. Why do patients treated with prostatectomy or radiotherapy have a lower
overall mortality than expected in the general population? Why does this do not apply for
patients treated with surveillance?
Answer: Overall patients diagnosed with prostate cancer have a lower risk of dying for all
causes. This indicates that in clinical practice there is selection of healthy men for PSA
testing and further work-up leading to a diagnosis of localized prostate cancer. Patients
with a short life expectancy (for whatever reason), however, were more often selected for
surveillance than curative treatment (confounding by indication). NOTE: because of
confounding by indication overall mortality cannot be used to compare the three
treatments
Prostate cancer specific mortality. Prostate cancer specific mortality after 10 years of
follow-up was 3.6% (95% CI: 2.7% - 4.8%) in the surveillance group, 2.7% (2.1%-2.4%) in
the prostatectomy group and 3.3% (2.5%-5.7%) in the therapy group.
Question 3. Does the cancer specific mortality vary between the three treatment groups?
Can these proportions be correctly interpreted (justify your answer)?
2
Answer. These estimates cannot be directly compared because surveillance is more
common among patients with low-risk disease (again confounding by indication).
Also, direct comparison is complex if competing risks have not been taken into account.
Further analyses. Further analyses are conducted stratifying by risk category (low-risk
disease and intermediate-risk disease).
o Among patients with low-risk disease prostate cancer specific mortality is 2.4% for
the surveillance group, 0.4% in the prostatectomy group and 1.8% in the
radiotherapy group (note: for sake of simplicity confidence intervals are omitted).
o Among patients with intermediate-risk disease prostate cancer specific mortality is
5.2% for the surveillance group, 3.4% in the prostatectomy group and 3.8% in the
radiotherapy group (note: for sake of simplicity confidence intervals are omitted)
Question 4. How and why do these results differ from the results reported above?
Answer. These estimates are more comparable because some of the effect from
confounding by indication is adjusted for. However we cannot exclude that some
confounding by indication is still present. NOTE(1) that, If anything the differences are
underestimated. NOTE(2) that the risk difference between surveillance and prostatectomy
is quite small (5.2%-3.4% = 1.8; NNT=55)
3