Background Understanding the use of invasive procedures (IPs) at the end-of-life (EoL) is important to avoid under- and overtreatment, but epidemiologic analysis is hampered by limited methods to define treatment intent and EoL phase. This study applied novel methods to report IPs at the EoL using a colorectal cancer (CRC) case study. Methods An English population-based cohort of adult patients diagnosed between 2013 and 2015 was used with follow-up to 2018. Procedure intent (curative, non-curative, diagnostic) by cancer site and stage at diagnosis was classified by two surgeons independently. Joinpoint regression modelled weekly rates of IPs for 36 sub-cohorts of patients with incremental survival of 0-36 months. EoL phase was defined by a significant IP rate change before death. Zero-inflated Poisson regression explored associations between IP rates and clinical/sociodemographic variables. Results Of 87,731 patients included, 41,972 (48%) died. 9,492 procedures were classified by intent (interrater agreement 99.8%). Patients received 502,895 IPs (1.39 and 3.36 per person year for survivors and decedents). Joinpoint regression identified significant increases in IPs four weeks before death in those living 3-6 months, and eight weeks before death in those living 7–36 months from diagnosis. 7,908 (18.8%) patients underwent IPs at the EoL, with stoma formation the most common major procedure. Younger age, early-stage disease, men, lower comorbidity, those receiving chemotherapy and living longer from diagnosis were associated with IPs. Conclusions Methods to identify and classify IPs at the EoL were developed and tested within a CRC population. This approach can be now extended and validated to identify potential under- and overtreatment. 

# Data from: Novel methods to define invasive procedures at the end-of-life were developed to improve quality of end of life care research: A population-based cohort study in colorectal cancer [https://doi.org/10.5061/dryad.hmgqnk9jq](https://doi.org/10.5061/dryad.hmgqnk9jq) This dataset categorizes OPCS codes according to treatment intent (potentially curative, non-curative, diagnostic and metastectomy) ## Description of the data and file structure This data is structured with in rows corresponding to OPCS codes (version 4) by cancer site (ICD10) and stage (UICC TNM). Each permutation is categorized by two independent colorectal surgical specialists with discrepancies reconciled into a final categorization. ## * ##