Data Source and Study Population
Data were obtained state central cancer registries participating in the Enhancing Cancer Registry Data for Comparative Effectiveness Research (CER) Project, a special project funded by the Centers for Disease Control and Prevention (CDC) National Program of Cancer Registries (NPCR). Details of the CER study have been described elsewhere . Of specific interest and valuable information collected by the CER project were detailed information on the first course of chemotherapy, including: names of chemotherapy agents, numbers of planned and received cycles, cumulative planned and received doses, units planned and received, starting and ending dates of chemotherapy, and whether chemotherapy was completed as planned .
Men and women diagnosed with colon cancer (the International Classification of Diseases for Oncology 3rd edition codes of C18.0, C18.2-C18.9) in 2011 at American Joint Committee on Cancer (AJCC) 7th edition  stage III who received colon resection and chemotherapy of FOLFOX4 or FOLFOX6 were eligible for this study. To maintain the sample size, we included all colon cancer cases whether that was the first and only cancer present or the patient had multiple cancers. Out of the 10 states participating in the NPCR CER study, only eight states (i.e. Alaska, Colorado, Florida, Idaho, Louisiana, North Carolina, New Hampshire, and Rhode Island) were included based on having complete chemotherapy data. Additional exclusion criteria included: cases with histology codes of 9050-9055, 9140, or 9590-9992 which were hematopoietic and lymphoid colon cancers, died within 30 days of colon resection, received neoadjuvant chemotherapy; and had missing data on vital status or date of last contact. Among 509 cases, 102 of them had an unknown value in RDI due to missing data on the related variables of received cycles (n=20), received doses (n=76), chemotherapy starting date (n=2), chemotherapy ending date (n=25), height (n=22) and weight (n=23), which results in 407 cases for final analyses.
Two outcomes were defined in this study: all-cause and cause-specific mortality. State cancer registries linked their data with state death certificate data files and National Death Index (NDI) database to obtain data on vital status and causes of death. Not all state cancer registries do NDI linkages every year. Death data obtained for Alaska and Florida were up to December 2014; North Carolina was up to June 2013. However, the remaining 5 states obtained death information up to November 2016 from the NDI. We used the Surveillance, Epidemiology, and End Results Program (SEER) cause-specific death classification to define the death due to colon cancer, which took into account causes of death in conjunction with tumor sequence, the primary site of the cancer diagnosis, and comorbidities (e.g., AIDS or site-related diseases) . For patients with colon cancer as the second or later primary cancer, we defined the colon cancer death as the death that was attributed to the particular cancer of C18.0, C18.2-C18.9.
The survival time was measured in months from colon cancer diagnosis to the date of death or last follow-up if a participant was still alive. The observations were censored if patients were alive during the time of clinical follow up.
Chemotherapy dose intensity was defined as the amount of drug administered per unit time (week or day) per square meter of body surface area (BSA) [12, 13, 19]. Chemotherapy RDI was the ratio of DDI to the SDI . The details of RDI calculation are described in Supplemental Material 2.
Based on previous research that optimal RDI of adjuvant chemotherapy was 70%  for patients with stage III colon cancer, and clinical trials suggesting that a high dose of chemotherapy (6 months of FOLFOX other than 3 months of therapy) was needed for high-risk stage III colon cancer patients , we defined the low and high RDI groups by using the cut-off points of 55%, 60%, 65%, 70%, 75%, and 80% for high-risk patients. For low-risk patients, we used the cut-off points of 45%, 50%, 55%, 60%, 65%, and 70%. For each cut-off point, patients receiving RDI greater than or equal to the predefined cut-off point were compared with patients receiving RDI lower than the predefined cut-off point, on their overall and cause-specific mortality. The optimal RDI was defined as the lowest RDI administered without significant increase in either overall mortality or cause-specific mortality.
Covariates included age at colon cancer diagnosis (<50, 50-59, 60-69, and >70 years), race/ethnicity (Non-Hispanic white, Non-Hispanic black, Hispanics and others) , health insurance (private insurance including Medicare with private supplement, Medicare/other public, Medicaid, and not insured), census-tract residence (100% urban, 100% rural, and urban/rural mixed), census-tract population below the federal poverty level (<20% and >20%), census-tract adults with less than high school education (<25% and >25%) [21, 22], census-tract population percentage of married (<50% and >50%) , tumor size (< 4 cm and > 4 cm) [24, 25], lymph nodes retrieved (<12 and >12) , tumor grade (well/moderately differentiated, and poor/undifferentiated), Charlson comorbidity index (0, >1) , anatomic subsites (proximal, distal, and others) , colon cancer classification (only with colon cancer, multiple cancers with colon as the first primary cancer, multiple cancers with colon as the non-first cancer) , number of positive lymph nodes (continuous variable), and delayed chemotherapy (receiving chemotherapy > 8 weeks after surgery: yes and no) .
Socio-demographic and clinical characteristics of the patients were summarized using descriptive statistics.
The overall survival probability was estimated using the Kaplan-Meier method. The differences of all-cause mortality between various levels of chemotherapy RDI groups were compared using the log rank test. The competing risk approach was applied to analyze RDI level impact on cause-specific mortality. The probabilities of cause-specific mortalities were estimated using the cumulative incidence function (CIF), which were compared among the various RDI groups using the Gray’s test . Cox proportional hazards model and the competing risks model developed by Fine and Gray were applied to estimate the hazard ratio (HR) for overall mortality and cause-specific mortality, respectively.
In sensitivity analyses, we restricted our sample to patients with colon cancer as the only or the first primary tumor, to exclude the effects of previously diagnosed cancers on survival or on chemotherapy use.
All analyses were conducted using SAS, version 9.4 (SAS Institute, Inc., Cary, North Carolina). A P-value<0.05 was considered statistically significant.