Impact of old age on resectable colorectal cancer outcomes

Introduction

Age-specific risk rises markedly in old age, and as mortality from heart disease and other non-cancer causes decrease, this leaves the elderly population at high risk for developing bowel cancer (Papamichael et al., 2009). The increasing incidence of colorectal cancer (CRC) in the segment of the population >70 years of age necessitates an examination of what type of treatment is most appropriate for these patients with CRC. Approximately 60% of CRC patients are >70 years of age at the time of diagnosis, and 43% are >75 years of age (Papamichael et al., 2009). It remains controversial whether so-called “elderly CRC” exhibits a differential prognosis compared to younger CRC and whether age (i.e., old) is an independent prognostic factor. There is now a general consensus that those in the old population possess a high frequency of frailty and comorbidities, exhibiting increased mortality from other causes among those with CRC. However, it remains unknown whether old patients have an increased incidence of colorectal cancer-specific death (CSD). A previous report demonstrated that older patients with CRC who survived the first year after surgery exhibited the same overall cancer-related survival as did younger patients (Dekker et al., 2011). Of note, chronological age is distinct from biological age (Odden et al., 2012; Robinson et al., 2009). The definition of “elderly” differs, being given as anything between >65 years and >80 years in different studies (Seymour, 2004; Twelves et al., 2005). As such, it is difficult to know whether 70 years is a reasonable cutoff age to safely extrapolate these results or whether the decision should depend on the physical and functional status of the patient rather than just on chronological age. Unlike younger patients with CRC, wherein more reliable evidence-based guideline based on clinical trials are available, older patients are still in need of increased evidence to guide clinical practices due to most trials excluding older patients with CRC. We used big data to explore the impact of old age on CRC prognoses to help guide clinical practice in the treatment of elderly CRC patients.

Materials and Methods

Results

Baseline characteristics and identification of the old age cutoff value

A total of 76,858 eligible patients were included in this cohort. The endpoint date for follow-up was November 2013, and the median follow-up time was 55.0 months (range 1.0–119.0 months).

The median age of patients was 64 years (IQR 20–80 years). With age as a continuous variable, the HR of CSS was 1.54 (95% CI 1.39–1.69, P < 0.001). The HR of CSS slowly increased before 70 years of age, and then increased significantly after 70 years according to the Cox model (Fig. 1).

Competing risk regression model was used to explore the impact of age on CSD and non-CSD

A total of 14,425 (18.77%) and 6,982 (9.08%) patients died of CSD and non-CSD, respectively. The corrected test cohort after PSM, a total of 47,616 patients analyzed, showed that 9,273 (19.47%) and 5,514 (11.58%) patients died of CSD and non-CSD, respectively. The 5-year CSD in the younger and older groups were 15.84% and 22.42%, respectively, and were significantly different (P < 0.001). The 5-year non-CSD in the younger and older groups were 5.21% and 14.21%, respectively, and were also significantly different (P < 0.001) (Fig. 2). Univariate and multivariate analyses demonstrated that old age was associated with CSD (Table 2).

Table 2:

Univariate and multivariate analysis of colorectal cancer-specific death of 47,616 patients after PSM.

Risk factors	Univariate analysis		Multivariate analysis
	SHR (95%CI)	Pb	SHR (95%CI)	Pb
Age
<70	1		1
≥70	1.47 (1.41–1.53)	<0.001	1.50 (1.44–1.56)	<0.001
Gender
Female	1		1
Male	1.06 (1.02–1.10)	0.005	1.11 (1.06–1.16)	<0.001
Marital status
Married	1		1
Unmarried	1.48 (1.39–1.58)	<0.001	1.27 (1.19–1.36)	<0.001
Divorced	1.35 (1.30–1.42)	<0.001	1.25 (1.19–1.31)	<0.001
Race
White	1		1
Black	1.31 (1.23–1.40)	<0.001	1.26 (1.18–1.35)	<0.001
Other	0.95 (0.88–1.02)	0.176	0.90 (0.84–0.97)	0.008
Location
Left	1		1
Right	0.93 (0.89–0.97)	<0.001	1.01 (0.96–1.06)	0.759
Histology
Adenocarcinoma	1		1
Mucinous adenocarcinoma	1.14 (1.07–1.21)	<0.001	1.08 (1.01–1.15)	0.018
Signet ring cell carcinoma	3.05 (2.62–3.54)	<0.001	1.48 (1.26–1.74)	<0.001
Differentiated grade
Grade I	1		1
Grade II	1.43 (1.31–1.56)	<0.001	1.27 (1.16–1.39)	<0.001
Grade III	2.45 (2.23–2.69)	<0.001	1.59 (1.45–1.76)	<0.001
T-classificationa
T1	1		1
T2	0.73 (0.65–0.83)	<0.001	0.96 (0.85–1.08)	0.472
T3	2.15 (1.95–2.37)	<0.001	2.19 (1.98–2.43)	<0.001
T4	5.46 (4.92–6.05)	<0.001	4.77 (4.27–5.33)	<0.001
N-classificationa
N0	1		1
N1	2.33 (2.22–2.44)	<0.001	2.16 (2.05–2.28)	<0.001
N2	4.48 (4.26–4.71)	<0.001	4.05 (3.81–4.30)	<0.001
nLN
0	1		1
0–2	0.49 (0.42–0.58)	<0.001	0.39 (0.33–0.46)	<0.001
3–5	0.61 (0.55–0.68)	<0.001	0.43 (0.39–0.48)	<0.001
6–11	0.56 (0.52–0.61)	<0.001	0.33 (0.30–0.36)	<0.001
≥12	0.47 (0.44–0.51)	<0.001	0.24 (0.22–0.26)	<0.001
CT
No	1		1
Yes	1.70 (1.64–1.77)	<0.001	0.84 (0.80–0.89)	<0.001
RT
No	1		1
Yes	1.35 (1.28–1.43)	<0.001	1.15 (1.08–1.23)	<0.001

DOI: 10.7717/peerj.6350/table-2

Notes:

Left includes rectum, rectosigmoid junction, sigmoid colon, descending colon, and splenic flexure.

Right includes transverse colon, hepatic flexure, ascending colon, cecum, and appendix.

PSM, propensity score matching; CSD, cancer-specific death; nLN, number of lymph nodes; CT, chemotherapy treatment; RT, radiotherapy treatment; SHR, subdistribution hazard ratio.

a T-classification according to seventh AJCC staging system.

b P-values obtained from the χ2 test. All statistical tests were two-sided.

Subgroup analysis by characteristics in raw data

Subgroup analyses were performed based on gender, race, differentiated type, pathological type, T and N classification, nLN, TNM stage, chemotherapy, and radiotherapy in the raw data. In all subsets (except for ring signet cell cancer), patients in the older group exhibited poorer prognosis compared with those in the younger group (Fig. 3). Furthermore, we performed an interaction analysis using Cox model between the therapy and age in stage III patients in raw data (Fig. S2). The results showed that the older group benefited more from chemotherapy than younger group (P-value = 0.001). However, the interaction between radiotherapy and old was not significant (P-value = 0.328).

Univariate and multivariate Cox regression analysis

A Kaplan–Meier survival curve for CSS is presented in Fig. 4. Univariate analysis showed a HR for CSS of older patients of 1.57 (95% CI 1.51–1.64, P < 0.001, reference to younger group) in the corrected cohort. Multivariate analysis showed that the HR for CSS in older patients was 1.64 (95% CI 1.57–1.70, P < 0.001, reference to younger group) (for detailed data see Table 3; raw data are shown in Table S1).

Table 3:

Univariate and multivariate analysis of colorectal cancer-specific survival of 47,616 patients after PSM.

Risk factors	Univariate analysis		Multivariate analysis
	HR (95%CI)	Pb	HR (95%CI)	Pb
Age
<70	1		1
≥70	1.57 (1.51–1.64)	<0.001	1.64 (1.57–1.70)	<0.001
Gender
Female	1		1
Male	1.08 (1.04–1.13)	<0.001	1.14 (1.10–1.19)	<0.001
Marital status
Married	1		1
Unmarried	1.54 (1.44–1.64)	<0.001	1.33 (1.25–1.42)	<0.001
Divorced	1.39 (1.33–1.46)	<0.001	1.29 (1.23–1.35)	<0.001
Race
White	1		1
Black	1.34 (1.26–1.43)	<0.001	1.29 (1.21–1.38)	<0.001
Other	0.93 (0.86–1.00)	0.06	0.87 (0.81–0.94)	<0.001
Location
Left	1		1
Right	0.93 (0.89–0.97)	0.001	1.01 (0.96–1.06)	0.697
Histology
Adenocarcinoma	1		1
Mucinous adenocarcinoma	1.13 (1.07–1.21)	<0.001	1.07 (1.01–1.14)	0.033
Signet ring cell carcinoma	3.10 (2.69–3.58)	<0.001	1.55 (1.34–1.79)	<0.001
Differentiated grade
Grade I	1		1
Grade II	1.43 (1.31–1.57)	<0.001	1.30 (1.19–1.42)	<0.001
Grade III	2.47 (2.25–2.72)	<0.001	1.64 (1.49–1.81)	<0.001
T-classificationa
T1	1		1
T2	0.73 (0.64–0.82)	<0.001	0.96 (0.85–1.08)	0.488
T3	2.15 (1.95–2.37)	<0.001	2.24 (2.02–2.49)	<0.001
T4	5.57 (5.02–6.17)	<0.001	5.01 (4.49–5.59)	<0.001
N-classificationa
N0	1		1
N1	2.33 (2.23–2.45)	<0.001	2.22 (2.10–2.34)	<0.001
N2	4.53 (4.31–4.77)	<0.001	4.24 (4.00–4.49)	<0.001
nLN
0	1		1
0–2	0.48 (0.40–0.56)	<0.001	0.37 (0.31–0.43)	<0.001
3–5	0.58 (0.53–0.65)	<0.001	0.41 (0.37–0.46)	<0.001
6–11	0.54 (0.49–0.58)	<0.001	0.30 (0.28–0.33)	<0.001
≥12	0.44 (0.41–0.48)	<0.001	0.22 (0.20–0.23)	<0.001
CT
No	1		1
Yes	1.65 (1.59–1.72)	<0.001	0.78 (0.75–0.82)	<0.001
RT
No	1		1
Yes	1.32 (1.25–1.40)	<0.001	1.14 (1.07–1.22)	<0.001

DOI: 10.7717/peerj.6350/table-3

Notes:

Left includes rectum, rectosigmoid junction, sigmoid colon, descending colon, and splenic flexure.

Right includes transverse colon, hepatic flexure, ascending colon, cecum, and appendix.

PSM, propensity score matching; CSD, cancer-specific death; nLN, number of lymph nodes; CT, chemotherapy treatment; RT, radiotherapy treatment; HR, hazard ratio.

a T-classification according to seventh AJCC staging system.

b P-values obtained from the χ2 test. All statistical tests were two-sided.

Discussion

The statistical methods “RCS” and “Chow test” were used to determine that 70 years is a reasonable cutoff age to define elderly CRC. Elderly CRC included a high frequency of male patients, right site CRC, mucinous carcinoma, more poorly differentiated grades and earlier stages. There were fewer than 12 lymph nodes sampled and earlier stages in the older group. After eliminating the distribution difference between older and younger groups by PSM, elderly CRC had worse outcomes (CSS), and age was shown to be an independent prognostic factor, while elderly CRC was related with increased non-CSD as well as CSD. In almost all subgroups, elderly CRC exhibited worse outcomes. The external cohort validated the reasonability of 70 years as a cutoff age to define elderly CRC and further confirmed that elderly CRC exhibited worse outcomes (CSS or CSD).

In clinical practice, the optimal cutoff age is anticipated to define elderly CRC. The screening program for CRC defined 65 years as a cutoff age (Papamichael et al., 2009). In other previously published studies (McCleary et al., 2013a; Merchant et al., 2017; Tournigand et al., 2012; Twelves et al., 2005), 70 years was adopted as a cutoff age. Updated SEER-Medicare analysis data and three population-based data sets conducted by Sanoff et al. (2012) showed that only 44% of the 5,941 patients evaluated received adjuvant chemotherapy within 3 months of surgical resection for stage III CRC. In their study, 65 years was used to define elderly CRC. In clinical trials, 75 years was more frequently set as the upper limit; therefore, a more real-world data analysis adopted 75 years as their cutoff age (Van Erning et al., 2013). We concluded that 70 years should be adopted as a suitable cutoff age.

Once the cutoff age was defined, we found that elderly CRC exhibited a significantly different outcome than young CRC. Several studies have shown that elderly CRC has comparable outcomes compared with younger CRC. Dekker et al.’s (2011) study showed that if one excluded death due to operation comorbidities (mostly death occurring 1 year after an operation), there were rather similar outcomes between the younger and older groups. Late period survival was similar between older and younger subgroups for resectable CRC (Dekker et al., 2011). One Canadian study (Merchant et al., 2017) showed that elderly CRC did exhibit worse prognosis and was associated with a high Charlson index. In their study, they did not differentiate between CSD and non-CSD; therefore, their study did not conclude that elderly CRC was related with increased CSD. In our study, we used a competing risk model to distinguish non-CSD and CSD. Furthermore, we report that elderly CRC is associated with increased CSD. In late period follow up, the Gray’s cumulative events curve on CSD separated more clearly, indicating that elderly CRC is associated with worse CSD, which might be due to different register periods. Therefore, we included patients from different periods as a validation cohort. The validation cohort also confirmed that elderly CRC was not only related with increased CSD but also with increased non-CSD. Our study confirmed that elderly CRC exhibits poorer prognosis from several aspects. Except for external validation and competing risk mode, we also confirm that elderly CRC is associated with worse outcomes using PSM to balance the differential distribution.

Worse outcome of elderly CRC is multi-faceted. Fewer numbers of lymph nodes were removed from those with elderly CRC, contributing to residual tumor matter in the anticipated dissection region (mean to R1 resection) and underestimated tumor stages. Moreover, elderly CRC exhibited less capacity to endure stronger treatments, resulting in lower intensity adjuvant chemotherapy and radiotherapy. Our correlation analysis indicated that elderly CRC was related with lower incidence of chemotherapy. The outcome of less frequent chemotherapy was similar to findings in previous studies (Abraham et al., 2013; Kahn et al., 2010; Van Erning et al., 2013, 2014). Finally, a CRC screening program resulted in the increased detection of precancerous lesions (such as polyps) that were treated in the old population. If the CRC diagnosis escaped screening, there was typically increased short-term carcinogenesis and more aggressive tumor behavior. Another study showed that elderly CRC had a greater index of genetic mutations and that the incidence of BRAF mutations was higher. Berg et al. (2010) indicated that CIMP tumors are more common in the older population, who also have a higher rate of KRAS and BRAF mutations.

Elderly CRC with worse outcomes might require stronger treatments; however, a previous study suggested that elderly CRC does not require enhanced treatment. Many elderly patients will benefit from radical treatment approaches, but others will not, and in some cases, non-operative “palliative” management should be offered, even though the cancer is “curable.” Guidelines from the International Society of Geriatric Oncology did not recommend that elderly CRC patients regularly receive adjuvant chemotherapy with limited evidence to support the benefit from such strategy (Papamichael et al., 2009). MOSAIC: lyses showed there to be no statistically significant benefit conferred by addition of oxaliplatin in terms of disease-free survival (DFS) or OS for older patients (70–75 years), although female patients 70–75 years of age exhibited the same oxaliplatin benefit as did younger patients (Tournigand et al., 2012). Interestingly, the DFS and OS benefits in patients 70–75 years were similar to those of younger patients for the first 3 years of follow-up but were lost later on due to deaths from other causes (Andre et al., 2009; Tournigand et al., 2012). NSABP-C-07: Patients ≥70 years failed to derive a statistically significant DFS or OS benefit from addition of oxaliplatin (Tournigand et al., 2012). Indeed, those patients receiving FLOX had poorer survival, which was attributed to toxicity. XELOXA (NO16968): the benefits observed for XELOX were maintained, although to a lesser degree in patients ≥65 and ≥70 years of age, in contrast to the results from MOSAIC and NSABP-C-07 trials (Twelves et al., 2012). Meta-analysis of ACCENT did not support that patients ≥70 years receive additional oxaliplatin chemotherapy (McCleary et al., 2013a, 2013b). In contrast, from an analysis using real-world data, almost all showed that elderly CRC can benefit from adjuvant chemotherapy with acceptable toxicity (Abraham et al., 2013; Kahn et al., 2010; Sanoff et al., 2012; Sanoff & Goldberg, 2007; Schrag et al., 2001; Van Erning et al., 2013, 2014). One study using SEER-Medicare data indicated that elderly CRC tolerated chemotherapy well, exhibiting worse prognosis due to reduced treatment (Sanoff et al., 2012). More recent studies showed that with the technological development of laparoscopic surgery and enhanced recovery programs after surgery (ERAS), elderly CRC could tolerate operation well. The literature suggests that elderly patients benefit from multimodal rehabilitation programs or ERAS in the same way as younger patients (Van Steenbergen et al., 2013). Our interaction analysis between the therapy and age in stage III patients showed that the older group benefited more from chemotherapy than younger group (P-value = 0.001). And compared with younger patients, older patients were more likely to have fewer lymph nodes sampled and were less likely to receive chemotherapy and radiotherapy. Therefore, current treatment paradigms in the older group may be insufficient. As life expectancy increases, more effective treatments are necessary for the old population.

To reduce bias as much as possible, we used PSM to balance clinicopathological characteristics and used a competing risk model to exclude impact from non-CSD. Finally, we confirmed that elderly CRC was related with more CSD and non-CSD. Age (old) is an independent factor to predict increased CSD. And the Cox model as a sensitivity analysis also had a similar result. Our analysis provides more evidence for elderly CRC receiving stronger adjuvant chemotherapy. The above conclusions can only be acquired from real world data analysis rather than from clinical trials alone.

As a retrospective study, it is impossible to avoid all bias for patient selection. There are several limitations inherent to the database used in the current study. The Charlson index is not available in SEER data. Though the SEER-Medicare can retrieve Charlson index, only patients older than 60 years are registered in their dataset. The Charlson index is strongly related with non-CSD in the old population. Competing risk models can effectively eliminate the impact from unavailable Charlson index. BMI is also an important influence factor in the CRC specific mortality (Kroenke et al., 2016). However, the information about BMI in the SEER data is not available. Furthermore, insurance status is an independent risk factor both for advanced disease in their cancer diagnosis and for cancer mortality (Rosenberg et al., 2015), and uninsured patients more often have higher T, N, and M stage (Amini et al., 2016). But the insurance recode variable is only available from 2007 in the SEER database. Additionally, comorbidities and detailed information about driver gene mutations (KRAS or BRAF) is not available. The variable chemotherapy is only classified as chemotherapy “yes” or “no/unknown” since SEER treatment information cannot accurately distinguish between “no treatment” and “unknown” (Noone et al., 2016). Furthermore, the sensitivity of SEER chemotherapy data is only 72.1% (Noone et al., 2016), as the detailed regime and duration of chemotherapy is not available in the SEER dataset.

Conclusions

In summary, 70 years is a suitable cutoff age to define elderly CRC. Elderly CRC is associated with not only increased non-CSD but also increased CSD. This SEER-based analysis provides further evidence that current chemotherapy in the elderly may be insufficient. Additional research is required to investigate whether elderly CRC will receive stronger treatment if possible.

Supplemental Information