Survival endpoints in colorectal cancer and the effect of second primary other cancer on disease free survival

In cancer research the selection and definition of survival endpoints are important and among the central issues discussed in the Consolidated Standards of Reporting Trials (CONSORT) statement [1, 2]. The definitions of the endpoints are often missing in clinical trials, even in the majority of randomised controlled cancer trials [3] making it difficult to compare studies. Meta-analyses will not be reliable when studies with different endpoint definitions are being compared.

It is central in the quest for improved cancer treatment that we minimize problems related to study analyses. One way to do this is always to use endpoints that have the same definitions.

The selection of survival endpoints depends on the study question and the information available. It is important to predefine the endpoints that should be collected during the study design phase since prospectively sampled information is more reliable and complete than information sampled retrospectively. Second primary other cancer is probably more often thought of as a late adverse effect or a competing risk than as a part of a survival endpoint. It can therefore be forgotten in treatment trials when the treatment is not thought to induce second primary other cancers and may not be available for the survival calculations. The use of disease-free survival (DFS) rather than overall survival (OS) has become more frequent in adjuvant cancer trials [5]. DFS offers earlier presentation of data as events due to disease recurrence by nature occur earlier than death from the disease. Besides, there are more events in DFS than in OS, as events, such as disease recurrence, and second primary other cancers that do not necessarily lead to death are included in DFS but not in OS. Despite the increased use of DFS as an endpoint, its definition varies widely between studies, in particular regarding the occurrence of a second primary other cancer [6]. Research on how DFS is affected by the inclusion or exclusion of second primary other cancers as an endpoint in DFS calculations is lacking. Therefore this study was initiated using a population-based cohort with well-documented follow-up data including information on second primary other cancers.

The aim was to compare different survival endpoints in patients with primary CRC according to the work by Punt et al (Table 1) and to better understand the relevance of inclusion or exclusion of second primary other cancers as an endpoint in the DFS calculations.

The study cohort included 415 patients consecutively diagnosed and treated for CRC in the county of Västmanland, Sweden, which has a population of 260 000 inhabitants. The time period of diagnosis was between August 2000 and December 2003. The patients were identified through the Regional Cancer Registry which forms the regional report to the Swedish Cancer Registry, to which the reporting is mandated by law and which has a high completeness [7, 8]. It is estimated that more than 99% of all incident colorectal cancers in the county are included in the cohort. The majority of the patients were treated at the Central District Hospital in Västerås. During the time-period for this study patients undergoing surgery for CRC were invited to donate tumour tissue and blood for future studies (n = 322) [9]. Information on family history was prospectively collected from 318 patients [10]. The information on tumour stage and grade of tumour differentiation was gained from pathology reports. Information on second cancer, cancer recurrence, death and causes of death was obtained by matching with the Clinical Database for Colorectal Cancer held at the Regional Oncologic Center in the Uppsala/Örebro region [11] and from hospital records at the Departments of Surgery, Oncology and Pathology at the hospital in Västerås.

Treatment-related death and date of death were available for both colon and rectal cancer patients, but information on locoregional recurrence and distant metastases was available only for rectal cancer patients. This information was double-checked during retrieval of data from hospital records. Other survival endpoints not available in the clinical database were retrieved from hospital records. If patients had moved from the county of Västmanland to another part of Sweden copies of their hospital records were obtained from their local hospital. One patient was lost to follow-up due to emigration.

Of the 415 patients, the following were omitted from the analyses: 68 patients with metastatic disease at presentation, 4 patients with stage II and III disease that were non-radically operated, and 11 patients with unknown disease stage of whom 7 were not operated. Remaining for participation in the analyses were 332 patients treated with curative intention with complete information on disease stage. Four patients treated with local excision of T1 tumours and polyp cancers were classified as curatively treated. Although the screening for metastases proved negative for the four patients treated with local excision, their lymph node status and, hence, disease stages were unknown.

The guidelines for observational studies in epidemiology (STROBE) were followed during the preparation of the manuscript [12]. Ethical approval was obtained from the Ethics committee at Uppsala University, Uppsala, Sweden.

The median (range) follow-up time for surviving patients was 8 (6-10) years. The median age of the cohort (n = 415) was 73 (34-97) years, males represented 203 (49%) patients and the disease location was colon in 284 (68%) patients and rectum in 131 (32%). The distribution of disease stage (TNM) was 57 (14%) stage I; 150 (36%) stage II; 125 (30%) stage III; 68 (16%) stage IV and 15 (4%) unknown. The median number of lymph nodes analysed was 15 (0-55) and a 30 day mortality was seen in 13 (3%) patients.

When comparing the survival curves of different types of endpoint in patients treated with curative intention it was observed that TTR initially has a steeper slope than CSS, but that the curves merge with time (Figure 1). Similar observations were seen with RFS and OS (Figure 1).

When analysing the same endpoints in individual disease stages, the curves for TTR and CSS have similar patterns for both stages I and II due to few recurrences in stage I (Figure 2) and to high mortality from CRC recurrence in stage II (Figure 3). For stage III the curves for TTR and CSS continue to be separated by more events in TTR during the first five years, representing patients surviving after CRC recurrence (Figure 4). An additional observation is that proportionally more patients with stage II develop late recurrences after three years compared with stage III patients. In stage II, 11 (55%) out of 20 recurrences occurred after three years of follow-up compared with 8 (17%) out of 47 recurrences in stage III patients (P = 0.002).

When continuing the comparison between stages II and III, a different progression of the survival curves for OS and DFS between three and five years is observed; for stage II the curves diverge from an 8 to a 14% difference and the curves merge for stage III from 12 to 7% (Figure 3 and 4). This observation is explained by a trend for more second primary other cancers in patients with stage II (12%) compared with stage III (6%) after five years of follow-up (P = 0.073). The differences in the cumulative proportion of patients without second primary other cancers between stages II and III are greatest during three to five years of follow-up. The cumulative proportions, with standard error, of patients without second primary other cancer after three years were 92.5 ± 2.5 for stage II and 98.2 ± 1.2 for stage III and after 5 years 87.5 ± 2.9 for stage II and 92.9 ± 2.9 for stage III.

Furthermore, when two different definitions of DFS were compared, one including only second primary CRC and the other including all second primary cancers, a worse DFS was observed if second primary other cancers were included. DFS for patients with stage I-III disease was 62% after five years if second primary other cancer was not included as an event, compared with 58% if second primary other cancer was included (Figure 5). However, the difference was larger for stage II (68 vs 60%) than for stage III (49 vs 47%)(Figure 6). In multivariate analyses, emergency operation became a statistically significant parameter of poor prognosis if second primary other cancers were included as an event in DFS (Table 3).

The differences between stages II and III result from more CRC deaths and fewer second primary other cancers in patients with stage III (n = 7) compared to patients with stage II disease (n = 18). Eleven (8%) out of 142 patients receiving pre- or postoperative (neo-) adjuvant treatment developed a second primary other cancer compared to 17 (9%) out of 188 patients not receiving this treatment. The most common types of second primary other cancers were breast (n = 7), lung (n = 3) and prostate (n = 3) cancer with no differences in type between stages. Stage III patients were younger (P = 0.037) compared with stage II patients. For 283 patients information on heredity was available. No differences were seen between stages in the proportion of patients with first degree relatives with CRC, numbering for stage II 12 (10%) out of 122 and for stage III 9 (10%) out of 91. No differences were seen in the age at diagnosis for patients with and without second primary other cancer.

The present study reveals that inclusion of second primary other cancers affects the results of DFS in patients with CRC. Second primary other cancers generate a significant number of events during follow-up of patients with CRC, which causes worse DFS when the second primary other cancers are included as an event in the calculations. Almost half as many second primary other cancers were observed as endpoints as death from the same cancer or non-cancer-related deaths. The most common events to occur were distant metastases closely followed by death from CRC and non-cancer related deaths, while second primary CRC was rarely seen.

The inclusion of second primary other cancer as an event in DFS did also influence the results of multivariate Cox models, in our example causing significant changes in the HR for emergency operation. Interestingly, the inclusion of second primary other cancers had a more detrimental effect on DFS of stage II patients than stage III patients, and this effect became greater with increasing time from diagnosis. The observation of more second primary other cancers in stage II patients than other disease stages has been reported by others [16]. The reason for this is not known. In the present study second primary other cancers were more common in those with first degree relatives with CRC, however, the proportion was similar in stages II and III, and no association was seen with adjuvant treatment. This indicates that the second other cancers are likely not therapy-induced. The reason for the greater effect of second primary other cancer on DFS survival in stage II is therefore largely explained by fewer events from other endpoints such as distant metastasis and death from CRC.

Second primary other cancers are significantly more common in patients with CRC than in the normal population [17, 18]. One reason for this can be related to radiation therapy for rectal cancer [19], although this theory has been debated [20]. Patients with family history of CRC are also more likely to develop second primary other cancers than patients without heredity [10, 21].

When studying the efficacy of adjuvant cancer treatments, the primary aim is to see whether the treatment reduces the risk of recurrence and subsequent death in the cancer of interest. In this situation TTR and CSS are the most specific endpoints because they only depend on events directly related to CRC and the effect of age is small. However, it is important to determine whether the investigated treatment is safe or whether it has serious adverse effects. Analyses that include events such as treatment-related death, non-cancer related death and second primary other cancer may then be necessary, thus OS and DFS, which include these events, serve as the primary endpoints in most randomised adjuvant trials.

Three year DFS has been suggested as a surrogate endpoint for five year OS in the setting of adjuvant treatment CRC trials [5]. In the study second primary other cancer was not included as an endpoint in DFS [5]. A recent publication has suggested that three year DFS is superior to five year OS and comparable with six and seven year OS, since extended survival due to more effective cancer treatments after disease recurrence is frequently seen [22]. Unfortunately it is not possible to determine whether or not second primary other cancers are included as a DFS endpoint in these analyses. This is important as after three years second primary other cancers already generate a significant number of events that can affect the results of the DFS.

It is essential that endpoints in studies of CRC are clearly defined as this will increase the comparability of studies. The events selected should be simple to collect to minimise the risk of missing events, and the definitions should be pre-specified at the time of study design. An example of a recent study lacking definition of the endpoints is the study on the prognostic value of KRAS and BRAF [23]. RFS and OS were analyzed according to stage in patients included in three adjuvant chemotherapy trials. The study is the largest of its type and published in a high impact journal and should therefore serve as a reliable reference for others, but it is not known if RFS refers to time to recurrence only or time to recurrence and death, nor is it clear if second primary other cancer is included as an event. This makes it difficult to interpret the results and it is impossible to compare the RFS to other studies.

A good example of endpoint definitions is found in a meta-analysis of 18 adjuvant treatment trials in colon cancer [24]. In this study OS, DFS and TTR were clearly defined: "OS is defined as time to death from any cause. DFS is defined as the time to recurrence or death, whichever occurs first. TTR is defined as the time to disease recurrence, where deaths without recurrence were censored at the time of death. Recurrence was defined only by a reappearance of primary colon cancer; second primary colon cancers or other non-colon cancers were not classified as recurrences."

The present study is a population-based observational study, with some of the endpoints retrospectively collected. This results in an older population with a broader range of stage at diagnosis and less reliable information on disease recurrence, secondary cancers and causes of death than an adjuvant treatment trial. Yet a population-based cohort, as in the present study, has the advantage of limited patient selection, frequently being very large in clinical trials [25]. However, our study setting represents a real life situation and a population that would be the recruitment base for clinical studies. Furthermore, the proportion of patients with second other cancers is, in CRC, largely independent of age and disease stage. Therefore we could expect an at least similar effect of second primary other cancers on DFS survival calculations in an adjuvant treatment trial compared with the present study.

It is debatable whether second primary other cancer should be regarded as a primary endpoint or as an adverse effect and therefore not included as an event in the main analysis of DFS. To increase clarity Punt [4] recommends that if second primary other cancers are ignored as an event, the survival endpoint should be named RFS. Choice of survival endpoints is an important topic and, to the best of our knowledge, this is the first study to address the use of second primary other cancers in DFS calculations in CRC.

Different definitions of survival endpoints have a significant effect on the survival analyses. Inclusion of second primary other cancers as an endpoint in DFS analysis significantly alters the survival for patients with CRC. Researchers and journals must clearly define survival endpoints in all trial protocols and published manuscripts. To minimise the differences in survival calculations between studies and to enable more precise comparisons of studies we recommend the general use of the definition of endpoints published in the consensus document of Punt et al [4].