Colorectal Cancer Survival in Multiple Sclerosis

Setting and Design

Forty-two percent of the Canadian population resides in the provinces of Manitoba and Ontario. We included these 2 provinces because of the combined population size of ≈14 million people, as well as their differing sizes (Manitoba's population is 1.4 million) and organization of health services, to improve generalizability of our findings. Each province is responsible for delivery of universal, publicly funded medically necessary services. As part of health systems management, these provinces prospectively record the use of health care services by nearly all residents in administrative databases. Because privacy regulations do not permit person-level data to leave the province of origin, we applied the same protocol in each province.

In both provinces, the databases used included the respective provincial population registries: the Discharge Abstract Database, medical/physician services (in Ontario, the Ontario Health Insurance Plan [OHIP]), and the Cancer Registry. In Manitoba, we accessed the administrative databases through the Manitoba Population Research Data Repository at the Manitoba Centre for Health Policy (MCHP). These databases were linked with the use of an encrypted unique provincial identifier. In Ontario, we accessed administrative databases at ICES. ICES is an independent, nonprofit research institute with a legal status under Ontario's health information privacy law that allows it to collect and analyze health care and demographic data, without consent, for health system evaluation and improvement. These databases were linked through the use of unique encoded identifiers and analyzed at ICES. The population registries provided sex, region of residence (postal code), dates of birth, dates of death, and dates of health care coverage. The Discharge Abstract Database captures hospital admission and separation dates and discharge diagnoses. These diagnoses are recorded using the ICD system, either ICD-9-Clinical Modification or ICD-10-Canada, depending on the year. Medical/physician services (OHIP) capture the date of service and physician-assigned diagnoses using 3-digit ICD-9 codes. In Ontario only, we accessed the Ontario Drug Benefit (ODB) database, which captures prescription drug dispensations for all those ≥65 years of age and receiving home care or long-term care (i.e., nursing home). Use of the ODB allowed us to identify persons residing in long-term care settings. In Ontario, we also accessed the Home Care Database, which captures individuals who have received home care services since 2005. Since 1965, the Manitoba Cancer Registry has captured cancer cases in Manitoba, with a completeness of 95% to 98%.¹¹ Cancer stage has been captured since 2004. Completeness of the Ontario Cancer Registry is 98%, and cancer cases have been recorded since 1964.¹² Cancer stage data have been captured since 2007. We linked postal code to area-level census data from Statistics Canada to determine the average household income of each individual's enumeration area.¹³

Standard Protocol, Approvals, Registrations, and Patient Consents

The University of Manitoba Health Research Ethics Board and Queen's Research Ethics Boards approved the study. The Manitoba Health Information Privacy Committee approved data access. Administrative data access (via ICES) was authorized under section 45 of Ontario's Personal Health Information Protection Act.

Study Populations

The study period was January 1, 1994, to December 31, 2016. We applied a validated case definition that identified persons with MS as those who ever had ≥3 hospital or physician claims for MS based on ICD-9 or ICD-10-Canada diagnostic codes (340/G35).¹⁴ For each person who met the MS case definition, we sought the earliest ICD-9 or ICD-10 claim for CNS demyelinating disease to determine the MS diagnosis date. This prevalent MS cohort was linked to the cancer registries. We then identified members of the MS cohort with incident colorectal cancer after MS diagnosis using ICD-O-3 topography codes (C18.0, C18.2–C18.9, C19.9, C20.9, C26.0). We designated the date of colorectal cancer diagnosis as the study index date. Next, we excluded anyone who (1) did not have a valid provincial registration number; (2) was not resident in the province with provincial health insurance for ≥12 months before the index date (eligibility is lost due to death or emigration); (3) was <18 years of age at the index date; (4) had any cancer before the index date; (5) had concurrent cancer, defined as subsequent cancer diagnosed within 6 months after the index date; or (6) had carcinoma in situ at diagnosis (Figure 1). This constituted the MS–colorectal cancer cohort. To identify a matched cohort, we selected all persons diagnosed with colorectal cancer without MS. We excluded anyone with diagnostic codes for any CNS demyelinating disease, including MS, and then applied the same exclusion criteria as for the MS cohort. From this pool of potential matches, we chose 4 persons matched on sex, year of birth ± 3 years, year of colorectal cancer diagnosis ± 2 years, and region of residence (first 2 digits of postal code) at the index date. Cases with <4 matches were excluded.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Directed Acyclic Graph of the Association Between MS and Death After Colorectal Cancer Diagnosis

MS = multiple sclerosis.

Outcomes

The primary outcome was overall survival (considering all causes of death, including MS) after colorectal cancer diagnosis, truncated at 5 years, reflecting that this is the typical benchmark in colorectal cancer.¹⁵ The secondary outcome was cancer-specific survival, defined as any cancer death and also truncated at 5 years. We also describe cancer stage at diagnosis to characterize the cohorts. The stage of cancers that originate from solid tumors is described with the Union for International Cancer Control TNM classification of malignant tumors, which considers the size of the primary tumor, whether it has invaded local tissue, regional lymph node involvement, and presence of metastases. This information can be combined into an overall stage grouping that ranges from 0 (in situ) to IV, or unknown, for each cancer.¹⁶ Cancer stage was grouped as I/II and III/IV to allow consistent reporting across provinces while meeting privacy requirements not to report cell sizes <6.

Covariates

We included several covariates that are associated with cancer survival, including age at cancer diagnosis (continuous), region of residence (urban [reference group] vs rural), socioeconomic status, cancer diagnosis year (1994–2004 [reference group], 2005–2009, 2010–2016), and comorbidity.¹⁰ In Ontario, we classified urban regions as communities with populations ≥100,000 based on the census metropolitan areas or census agglomeration. In Manitoba, we classified the only 2 cities (populations of 47,000 and 600,000) as urban. We categorized socioeconomic status, based on quintiles of average household income, as lowest quintile, quintiles 2 to 4, and highest quintile (reference). We included cancer diagnosis year to account for temporal trends in colorectal cancer incidence and survival.¹⁷ We used the Elixhauser comorbidity index (continuous variable) with a 1-year lookback period because it adjusts better for comorbidity in cancer studies than other indices.¹⁸ The index was modified to exclude cancer diagnosis codes so that we would not misclassify metastases or secondary cancers as comorbidity. Similarly, we excluded paralysis and demyelinating disease codes to avoid misclassifying MS-related symptoms as comorbidity. Notably, the index captures obesity and drug and alcohol abuse in addition to chronic health conditions. We updated socioeconomic status, region (urban/rural), and comorbidity status at the study midpoint to account for changes in status and to address nonproportionality identified during testing of Cox regression model assumptions. In Ontario, we also measured disability status (as of 2007) based on receipt of home care services or admission to a long-term care facility. Home care services are provided only to individuals with persistent impairments in functional status. Nearly half of individuals with MS who receive home care in Canada use a wheelchair and have some degree of cognitive impairment; one-third report falling; and one-third are incontinent. When health care needs are too great to be met in the community, long-term care (i.e., nursing home) is provided. In Canada, 8 of 10 persons with MS residing in long-term care require a wheelchair and have some degree of cognitive impairment, making long-term care admission a highly specific marker of severe disability.¹⁹ We used ≥1 OHIP fee code for services delivered in long-term care or the long-term care code in the ODB to identify long-term care admissions^20,21; this definition has a positive predictive value of 93.2% and negative predictive value of 99.9%.²² The earliest long-term care code identified constituted the date of disability in that setting. Thus, use of either home care or long-term care indicated the presence of moderate to severe disability.

Analysis

We characterized the cohorts using descriptive statistics including frequency (percent), mean (SD), and median (interquartile range). The cohorts were compared by the use of standardized differences (values of 0.20–0.49 represent small effect sizes, values of 0.50–0.79 represent medium effect sizes, and values ≥0.80 represent large differences between cohorts in baseline characteristics).²³ We reported crude case fatality rates (deaths per 100 person-years) in the 2 cohorts.

We used Cox proportional hazards models to examine the association between MS and all-cause survival after colorectal cancer diagnosis in which zero time was the colorectal cancer diagnosis date. Each participant was followed up until emigration/loss of provincial health care eligibility, death due to any cause, or the end of the study period (December 31, 2016), whichever came first. Because follow-up varied between the MS and matched cohorts due to differential survival, the matching factors were included as covariates,²⁴ along with the other covariates described above. In colorectal cancer, comorbid diseases such as MS may influence the stage of illness at diagnosis, and cancer stage is associated with survival.²⁵ In addition, a study in British Columbia, Canada, found that compared to persons without MS, persons with MS had larger tumors at diagnosis with colorectal, breast, lung, or prostate cancers.²⁶ Therefore, we did not adjust for cancer stage in the primary model because we considered it a causal intermediate, that is, a mediator of the association between MS and survival (Figure 1). We tested the proportional hazard assumption using graphical methods and time-dependent covariates.²⁷ This assumption was not met for the MS vs non-MS variable; the violation of the assumption was more pronounced in Manitoba than in Ontario. Therefore, we included an interaction term between MS and the natural log of time, and we report hazard ratios at 6 months and 1, 2, and 5 years after diagnosis.

To test the association between MS and cancer-specific survival, we used a cause-specific hazard model that accounted for the competing risk of death to other (noncancer) causes²⁸ and adjusted for the same covariates as for the all-cause survival model. In each regression model, we tested for interactions between cohort (MS vs controls) and age and cancer diagnosis year. We used random-effects meta-analysis to pool adjusted hazard ratios (HRs) and 95% confidence intervals [CIs] from the regression models.²⁹ We reported I² as a measure of heterogeneity of the findings between provinces,³⁰ where I² ≤ 25% indicates low heterogeneity and I² ≥ 75% indicates high heterogeneity.

Complementary Analysis

We conducted several complementary analyses. First, for our primary model, we calculated the E-value to determine the sensitivity of our findings to unmeasured confounders, where the E-value is a function of the strength of association (HR) of the exposure (i.e., MS) and the outcome (i.e., death). It describes the strength of association needed between an unmeasured confounder and both the exposure and the outcome to explain away the observed association of interest.^31,32 Second, among individuals in Ontario diagnosed in 2007 or later for whom cancer stage data were available, we incorporated stage (in 4 separate groups) as a covariate, viewing stage as a potential mediator of the association between MS and survival (Figure 1). Notably, the proportional hazards assumption for the MS vs non-MS variable was not violated for the subcohort, but there was an interaction between stage and age. This analysis was not possible in Manitoba due to the small size of the cohorts. Third, we explored the association between disability status and survival in Ontario, where the study population was large enough, to understand how much of the potential survival differences was mediated by differences in underlying disability between those with MS and those without (Figure 1). We expected that including this covariate would attenuate the association between MS and survival.

Statistical analyses were performed with SAS version 9.4 (SAS Institute Inc, Cary, NC).

Data Availability

The source data for Manitoba are not owned by the researchers or MCHP and thus cannot be provided to a public repository. With the appropriate approvals, the Manitoba data can be accessed through the MCHP. The Ontario dataset from this study is held securely in coded form at ICES. While data-sharing agreements prohibit ICES from making the dataset publicly available, access may be granted to those who meet prespecified criteria for confidential access.