Evaluation of Algorithms to Identify Incident Cancer Cases by Using French Health Administrative Databases

Overview

"This study is a population based observational study conducted on the French administrative databases to estimate cancer incidence in 2012 by using the "ECHANTILLON GENERALISTE DES BENEFICIAIRES" (EGB, a 1/97th dynamic random sample of the SNIIRAM). The EGB database contains anonymous and prospectively recorded data about all beneficiaries' medical reimbursements. Many algorithm definitions are defined to estimate the incident rate of cancer in 2012 from the EGB database. The incidence rates obtained by each algorithm definition are compared to national incidence rates by indirect age and sex standardization. National incidence rates are obtained from "FRANCE CANCER INCIDENCE ET MORTALITE" (FRANCIM): the French network of cancer registries."

"This study is a population based observational study conducted on the French administrative databases to estimate cancer incidence in 2012 by using the "ECHANTILLON GENERALISTE DES BENEFICIAIRES" (EGB, a 1/97th dynamic random sample of the SNIIRAM). The EGB database contains anonymous and prospectively recorded data about all beneficiaries' medical reimbursements including age, gender, long-term chronic disease (LTD), date of death, all out-hospital health-spending reimbursements and all patients' hospitalizations. Many algorithm definitions are defined to estimate the incident rate of cancer in 2012 from the EGB database and are applied separately for men and women. These algorithms use information from either out-hospital care only (LTD status, anticancer specific drugs, outpatient radiotherapy sessions), or inpatient stays only (primary or related diagnosis of cancer, cancer-related procedures) or combine both information. The incidence rates obtained by each algorithm definition are compared to national incidence rates by indirect age and sex standardization. National incidence rates are obtained from FRANCE CANCER INCIDENCE ET MORTALITE (FRANCIM): the French network of cancer registries that has collected cancer cases since 1975 from 21 French registries (general or specific) covering 17 of the 95 French metropolitan departments. The most recent estimation of the FRANCIM network was published for 2012 and included all cancer locations (C00-C97) excluding non-melanoma skin cancer (C44). Follow up All patients included are followed from January 1, 2012 until the occurrence of the first of death, cancer occurrence, moving out of the general insurance scheme or January 1, 2013. To allow for comparison with data from cancer registries, only malignant neoplasms are considered: ICD-10 codes (C00-C97), excluding non-melanoma skin cancer (C44). Statistical analysis The number of incident cancer cases obtained with each algorithm is compared to the expected number of cancer cases calculated by using national estimation for the same age and sex stratum. The standardized incidence ratio (SIR) with 95% confidence intervals is calculated by indirect age and sex standardization. Age- and sex-specific incident rates are compared to the incident rates in 2012 estimated by FRANCIM. The investigators also apply the most accurate algorithm separately for men and women to the 3 most common cancers in the corresponding gender by restricting the involved cancer codes, procedures and drugs to those related to the cancer of interest. The investigators also restrict our population to the most-studied age groups in cancer etiological studies (40 to 75 years). All the analyses are performed using SAS Enterprise Guide, version 4.3."