Gynecological malignancy risk in colorectal cancer survivors:
A population-based cohort study
Wei-Chun Chang
a, b, 1, Chih-Hsin Muo
c, Ji-An Liang
a, d, Fung-Chang Sung
a, c, 1,
Chia-Hung Kao
a, e, *Introduction
Colorectal cancer is one of most common cancers in Taiwan, with a total of 14,040 new cases being reported in 2010. (http://
www.bhp.doh.gov.tw) The risk of gynecological malignancy
(ovarian, endometrial, and breast cancers) could be higher in women diagnosed previously with colorectal cancer because of diverse reasons including shared etiological factors, hereditary and environmental infuences, and potential carcinogenic effects of prior treatments with chemotherapy or radiotherapy (Ng and
Travis, 2008). With survival rates after diagnosis of numerous
cancers continuing to improve, follow-up examinations of cancer survivors should not only consist of monitoring for local-regional and distant cancer recurrence, but should also include evaluations of health maintenance, life quality, and adverse effects of
treatments, and surveillance for subsequent primary malignancy. Fortunately, for gynecological cancers, several identifed etiological factors are amenable to preventive approaches.
To date, no studies have evaluated the risk of gynecological malignancy after colorectal cancer diagnosis. We aimed to assess whether women with prior colorectal cancer had an increased risk
of developing gynecological malignancy, because if the risk of gynecological malignancy is higher in colorectal cancer survivors,
early cancer detection and prevention measures should be advocated strongly. Materials and methods
Data sources
This retrospective cohort study used the inpatient database and the registry for catastrophic illness database (RCID), a part of the National Health Insurance Research Database (NHIRD), established by the Bureau of National Health Insurance (NHI) on March 1, 1995. NHI program has a coverage rate over 99% for 23 million people in Taiwan and feature contracts with 90% of all health care facilities science 1996 (Lu et al., 2003 MayeJun). These databases included medical claims from 1996 to 2010 and information on benefciaries.
Diseases were defned based on the International Classifcation of Diseases, 9th Revision, Clinical Modifcation (ICD-9-CM). All insurance claims should be scrutinized by medical reimbursement
specialists and peer review. The cancer diagnoses were based on the ICD-9 code determined by pathologic fndings, therefore the cancer diagnoses in this study should be accurate and reliable. We confrm that all data were de-identifed and analyzed anonymously. In addition, this study was also approved by the Ethics
Review Committee at China Medical University (CMU-REC-101-012).
Study subjects
Fig. 1 shows the fow chart of selecting study cohorts. We
selected 44,397 women with colorectal cancer (ICD-9-CM 153 and 154) from 1998 to 2009 from RCID. Exclusions were 1940 patients with a history of cancer (ICD-9-CM 140e152 and 155e208), 34 patients aged less than 20 years, and 5247 patients with a follow-up duration of <0.5 years. The remaining 37176 patients were included in the colorectal cancer group. The date with colorectal cancer diagnosed was used as entry date for estimating the follow-up time. Comparisons were selected from women without cancer history, including metastases (ICD-9-CM 196e199), solid cancer (ICD-9-CM 140e195) and hematopoietic cancer (ICD-9-CM
200e208) at the baseline. In order to increase the statistical power, 4-fold of controls were randomly selected, frequency-matched by age (stratifed in 5-year durations) and entry date of the colorectal cancer case. To exclude certain control group members,
we used the same criteria as those used for the colorectal cancer group.
Variables of interest
Study subjects were evaluated, from the entry date until the
gynecologic cancers occurred, end of 2010, withdrew from the insurance or death. The female-specifc cancers including breast
(ICD-9-CM 174), cervical (ICD-9-CM 179), endometrial (ICD-9-CM 182), and ovarian (ICD-9-CM 183) cancers were evaluated. Comorbidity evaluation included diabetes (ICD-9-CM 250), hypertension
(ICD-9-CM 401e405), dyslipidemia (ICD-9-CM 272) and hereditary colorectal polyposis (IC-9-CM 211.3), which were defned before the entry date.
Statistical analysis
All analyses were performed using SAS statistical software version 9.2 for Windows (SAS Institute Inc., Cary, NC, USA) and 2-sided P < 0.05 was considered signifcant. Demographic differences between the 2 cohorts were compared using X2 (Ng and
Travis, 2008) for categorical variables. Because the distribution of
age was not ft normally distribution, we used Wilcoxon rank sum test between colorectal cancer and control groups. We also calculated the incidence for gynecological cancers per 1000 personyears for both groups. Used Cox proportional hazards regression
analysis to estimate the hazard ratio (HR) and 95% confdence intervals (CIs) for gynecological cancers in the colorectal cancer
group, compared with control group after controlling continuous age, comorbidity (including diabetes, hypertension, dyslipidemia, and hereditary colorectal polyposis) and multiplicative interaction variable. According to previous reports, those comorbidities were potential risk factors for gynecological cancer. We tested the possible interactions between variables. When the interaction test p < 0.05, we added the multiplicative interaction for adjustment in measuring HR of gynecological cancers and specifc-type cancers. Because of the signifcant interaction between age and colorectal cancer, we assessed age-specifc HR in multivariable Cox proportional hazards regression. Accounting for the competing risks of
death and other types of cancer, we used the Fine and Gray model
(Fine and Gray, 1999) to estimate the cumulative incidence of gynecological
cancer. The identifcation of death events was based on
hospital discharge for death or withdrawal from the NHI. Subhazard ratio (SHR) for gynecological cancer were estimated by
multivariate competing-risks regression models after adjusting for
Fig. 1. Flow chart for selecting study cohorts.
2 W.-C. Chang et al. / European Journal of Oncology Nursing xxx (2015) 1e6
Please cite continuous age, comorbidity and multiplicative interaction variables. To assess age-specifc SHR, data analysis was performed after
controlling for comorbidity and multiplicative interaction variables, including death and other types of cancer. KaplaneMeier analysis was used to plot the cumulative incidence of overall gynecological cancers, breast cancer, endometrial cancer and ovarian cancer, and the log-rank test was used to evaluate the difference between the 2
groups for corresponding cancer. Results
There were 37,176 patients in the colorectal cancer cohort and 148,700 women in the comparison cohort (Table 1). There was no signifcant difference in the age distribution between these 2 cohorts, with a median age of 65.8 years (interquartile range ? 20.4
years) in the colorectal cancer group. Compared with the comparisons, the colorectal cancer group had higher prevalence of diabetes
(15.3% vs. 8.43%), hypertension (26.3% vs. 14.8%), dyslipidemia (4.27% vs. 3.64%) and hereditary colorectal polyposis (4.54% vs. 0.26%) (all p < 0.0001 by Chi-square test).
The cumulative incidences of breast, endometrial and ovarian cancers were all higher in colorectal cancer survivors than in comparisons after the 13 years of follow-up (Fig. 2, all log-rank test p < 0.0001). The mean durationwas 3.51 years (SD ? 2.87) from the colorectal cancer diagnosis to the other types cancer diagnosis (data not shown).
Based on interaction tests for potential variables in different cancer, the signifcant effects in overall cancer was diabetes with hypertension (interaction p ? 0.003); in cervix cancer was colorectal cancer with dyslipidemia (interaction p ? 0.02); in endometrial cancer were age with diabetes (interaction p ? 0.0001), and age with hypertension for (interaction p ? 0.003); and in ovarian cancer was age with diabetes for (interaction p ? 0.009, data not shown). Those signifcant multiplicative interaction effects were considered for adjustment in multivariable Cox proportional hazard
regression. Overall, the incidence of gynecological cancers was 1.39-time higher in colorectal patients (2.99/2.14 per 1000 personyears)
with an adjusted HR of 1.46 (95% CI ? 1.31e1.62) (Table 2).
The HRs of breast, endometrial, and ovarian cancers were all signifcant for the survivals. In multivariate competing-risks regression
models, colorectal cancer women had a SHR of 1.56 (95% CI ? 1.40e1.74) for the 4 gynecological cancers than control women. The site specifc data show that the SHRwas the highest for endometrial cancer (3.40, 95% CI ? 2.59e4.47), followed by ovarian cancer (2.77, 95% CI ? 2.07e3.70) and breast cancer (1.30, 95% CI ? 1.12e1.50). The SHR was 0.68 (95% CI ? 0.52e0.89) for cervical cancer. The analysis by follow-up years in Table 2 shows that the
incidence of all gynecological cancers consistently higher in later years than the earlier 5 years, and higher for colorectal cancer patients than comparisons.
Table 3 shows age-specifc risk for gynecological cancers.
Colorectal cancer women were at a signifcantly greater risk than comparison women for gynecological cancers in each age group, except the cancer of cervix. Among the colorectal cancer survivals, younger women were at higher risk than older women. Most of endometrial cancer occurred in the survivors <50 years old, with an incidence 11.2-fold greater than the comparisons of same ages, a SHR of 6.64 (95% CI ? 4.18e10.6) estimated by the competing risk model. Similar contrast appeared for ovarian cancer with lower incidence in the survivals and a SHR of 4.55 (95% CI ? 2.76e7.50).
Discussion
Since Billroth frst described multiple primary cancers
developing in a single patient in 1889, numerous investigators have explored the association between multiple cancers by using data from cancer registries (Billroth, 1889; Schoenberg et al., 1969). Association of multiple primary cancers suggests a common environmental or genetic etiology, but other possible explanations include heightened medical surveillance after
diagnosis of a frst primary cancer, and cancer resulting from the treatment of the frst primary cancer. Any possible association raises concerns about the requirement of more intensive monitoring of women with a history of neoplasia for second
malignancy.
Current evidence from epidemiological, experimental, and clinical studies supports the emerging hypothesis that metabolic syndrome may be a key etiological factor for the development and progression of certain types of cancer (Zhou et al., 2007). Relative to our comparison group, women with colorectal cancer had higher prevalence of diabetes, hypertension, dyslipidemia and hereditary colorectal polyposis (Table 1). The results of a systematic review also support the view that compared to non-diabetic people, patients with diabetes have an increased risk of colorectal cancer
(Jiang et al., 2011). Main fndings
The cumulative incidences for all female-specifc cancers,
breast, endometrial, and ovarian cancers, were signifcant higher in colorectal cancer survivors after 13 years of follow-up (Fig. 2). Multiple primary malignancies may arise by any combination of 3 etiological categories: (1) shared syndromic or genetic susceptibilities, (2) common environmental or lifestyle factors, and (3)
treatment-related effects (Travis et al., 2006). Women with colorectal cancer had 46% and 56% higher hazard of gynecological
cancer than controls in the multivariable Cox proportional hazards model and competing risk model (Table 2). Specifcally, the hazard ratio was 2.72 for all gynecological cancers when the age of the colorectal cancer survivors was _50 years (Table 3), meaning that the women were affected when they were in a fnancially productive and reproductive stage of their lives. Thus, additional resources should be devoted toward developing affordable methods
for detecting and preventing the subsequent gynecological cancers for colorectal cancer survivals. Strengths and limitations
This study has the strength of using a very large populationbased data set. Taiwan launched the national health insurance in 1995, operated by a single-buyer, the government. All insurance claims should be scrutinized by medical reimbursement specialists and peer review. The cancer diagnoses in the NHIRD were based on the ICD-9 code determined by pathologic fndings. Therefore the cancer diagnoses in this study should be accurate and reliable. However, misclassifcation of cases' stages based on ICD9 codes alone without other more defnitive criteria for potential early diagnostic bias is still possible. In addition, several limitations inherent to our database should be mentioned. First, the diagnoses recorded in the NHIRD primarily serve the purpose of administrative billing and not all have undergone verifcation for scientifc
purposes. Second, the database does not contain information regarding the daily lives or behaviors of the patients, including information on smoking habits, alcohol consumption, body mass index, socioeconomic status, and family history. Third, there were lacking of personal laboratory and imaging data in NHIRD. Finally, the cohort study is not exactly like the randomized controlled trial. There are still some uncontrolled confounding factors could have infuenced our fndings.
Interpretation
In summary, colorectal cancer survivors had a signifcantly higher risk of developing gynecological cancers than agematched women without a history of colorectal cancer. The
converse may also be true, because women diagnosed previously with gynecological malignancy have been estimated to have an increased risk of developing colorectal cancer (Radhika et al., 2007). The association between breast, endometrial, ovarian, and colorectal cancers suggests common etiological factors, which may be either genetic or environmental. Given this moderately increased risk of gynecological malignancy in women with colorectal cancer, early cancer detection and prevention measures must be employed by health-care workers for optimal use of available resources.
Conclusion
This increased risk of gynecological malignancy in women with colorectal cancer should be addressed using early detection and prevention measures by health-care workers for optimal use of available resources.
