Study protocol, final version, 12.04.2018.
Wojciech Michalski and Krzysztof Bujko
Evaluation of cancer treatment efficacy is hampered by mortality from non-cancer causes, especially
in elderly patients. To account for non-cancer mortality, competing risk (CR) method is recommended.
However, the Kaplan-Meier (KM) method, where death from non-cancer causes is censored (i.e.
ignored), is still being used. The survival estimators calculated in this way is usually called “cancerspecific
survival” (CSS) or disease free interval. When the local treatment efficacy is of interest and
distant metastases can be ignored (like in our example where distant metastases occur along or after
local failure), local recurrence free interval (LRFI) is being used. These kind of estimators are biased
because they do not take into account competing events, i.e. non cancer deaths. This bias is larger in
elderly patients than in their younger counterparts due to the higher incidence of non-cancer deaths.
Such bias might be considered clinically irrelevant until it does not change the interpretation of results.
We hypothesize that for elderly patients, quantitative evaluation of efficacy of cancer treatment
obtained by using the KM method might be meaningfully different from that obtained by using CR
method. The aim of this study is to evaluate two issues: i) to assess the quantitative difference
between the two statistical methods in the evaluation of a risk of cancer death and a risk of local
recurrence in the cohort of elderly patients and in the cohort of their younger counterparts; ii) to assess
whether an interpretation of the results evaluated by the KM method is biased for the elderly patients.
Material and methods
The material includes patients with rectal cancer treated with preoperative radiotherapy and fullthickness
local excision in the frame of the randomized trial comparing two radiotherapy schedules:
short-course radiotherapy versus chemoradiation (waw). The protocol of this trial required registration
of patients who met the entry criteria but could not be randomized due to age and co-morbidity. These
patients were treated using short-course radiotherapy and full-thickness local excision, i.e. with the
same method that was used in the control arm of the randomized cohort. Thus, there were two cohorts
to compare: 31 elderly patients (non-randomized) and 33 younger patients (randomized), which
differed in age and incidence of co-morbidities. We assumed that differences in age and incidence of
co-morbidities do not influence cancer biology and hence cancer response to treatment. Because of
the good tolerance of short-course radiotherapy, adherence to the treatment schedule was not
different in the two groups. Consequently, we assume that the risk of local recurrence and the risk of
death from cancer should not differ between these two groups. Because of bias inherent to the KM
method, we might expect, that by using it, quantitative evaluation of treatment efficacy in the elderly
patients and in their younger counterparts are different. A question arises whether such difference is
clinically relevant. If our assumption of similar cancer biology in our two cohorts is correct, we expect
that quantitative evaluation of treatment efficacy in the elderly patients and in their younger
counterparts is not much different by using the CR method.
The following analyses are planned:
1. Comparison of the cumulative incidence function (CIF) of local recurrence between the elderly
group and the younger group by using the CR method with the Gray’s test and estimation of
the hazard ratio (HR) and the CIF at 3, 5 and 10 years.
2. Comparison of the LRFI between the elderly group and the younger group by using the KM
method, long-rank test and estimation of the HR and the 1 – LRFI at 3, 5 and 10 years.
3. Comparison of the CIF of cancer death between the elderly group and the younger group by
using the CR method with Gray’s test and estimation of the HR and the CIF at 3, 5 and 10
4. Comparison of the CSS between the elderly group and the younger group by using KM
method, log-rank test and estimation of the HR and 1 – CSS at 3, 5 and 10 years.
All analyses will be illustrated by the curves showing probability of event occurrence as a function of
follow-up. The quantitative differences between the CR and KM methods (as stated above) will be
display in a table with the estimations and 95% confidence intervals. Discussion will be carried out as
to whether differences in results evaluation obtained by the CR method in comparison to that obtained
by the KM method might lead to differences in clinical data interpretation. In this discussion, a
difference between the elderly group and the younger group in the number of patients having
conversion to radical surgery (1 patient vs. 5) will be taken into account. Such discussion will suggest
(or will not suggest) whether, from the clinical point of view, CR method should replace KM method in
the evaluation of cancer specific survival and cumulative incidence of local recurrence.
Additionally, in the elderly patients group, a graphical display of cumulative incidence of cancer death
and cumulative incidence of non-cancer death will be depicted within one figure. Discussion will be
carried out as to whether such figure might be helpful in data interpretation. For comparative purpose
the same figure will be created for the younger patients group.
Wawok P, Polkowski W, Richter P et al. Preoperative radiotherapy and local excision of rectal cancer:
Long-term results of a randomised study. Radiother Oncol 2018, accepted for publication