e18810 Background: Breast, stomach and colorectal cancers have high incidence in Colombia. Official cancer incidence registries depend upon sentinel site reports as there is no nationwide surveillance system. We aimed to identify administrative database algorithms for breast, stomach and colorectal cancer case selection and to compare their cancer incidence estimates to official registries. Methods: We conducted a systematic review to identify algorithms with high positive predictive values (VPP) for breast, stomach and colorectal cancer case identification in administrative databases. For each cancer we selected two algorithms: a sensitive algorithm, based on cancer-specific ICD-10 codes, and a specific algorithm, combining cancer-specific ICD-10 codes with at least one code for oncological procedures. We varied the number of months a cancer-specific ICD-10 code was registered within each algorithm to test for algorithm stability. We conducted a cohort study to estimate incident cancer cases for 2013 in four cancer sentinel cities in Colombia and one cancer reference center using both algorithms. We defined incident cases as cases lacking a cancer-specific ICD-10 code in the preceding two years and adjusted incident cases for type of regimen affiliation. Algorithms with results closest to official sources were selected as best performing algorithms. We used the contributive regimen Capitation Payment Unit administrative database of Colombia for 2011-2014 as source of information. Results: Breast cancer case-identification algorithms have a higher VPP reported in literature (83-100%) compared to colorectal (41.7-94%) and stomach cancer (35-59.7%) algorithms. The closest breast cancer incidence estimates to the official registries Infocancer and National Cancer Institute were yielded by the specific algorithm with ICD-10 codes persisting for four months (n= 672 vs 649 and 397 vs 212, respectively). The closest colorectal cancer cancer incidence estimates to official registries were yielded by the specific algorithm with ICD-10 codes persisting for three months (n= 219 vs 230 and 168 vs 139, respectively). The closest stomach cancer incidence estimates to official registries were yielded by the specific algorithm with ICD-10 codes persisting for one month (n= 122 vs 146 and 99 vs 153, respectively). Sensitive algorithms were less stable than specific algorithms across all three cancer types. Conclusions: Breast, stomach and colorectal incident cancer cases can be identified through administrative databases. VPP vary among types of algorithms and cancers. Specific algorithms provide better breast and colorectal incident cancer case-identification in Colombian administrative databases, compared to stomach cancer algorithms. This is a potential approach for estimating nationwide cancer incidence in Colombia.

e18808 Background: Fragmentation in healthcare leads to adverse outcomes in cancer patients. Currently there is no fragmentation measurement that has been acknowledged to reliably assess healthcare fragmentation across different health systems. We aimed to measure cancer healthcare fragmentation through administrative databases in Colombia and to calculate its effect on breast, stomach and colorectal cancer mortality. Methods: We conducted a cohort study based on health administrative databases from 2013 to 2017. We combined data from two Colombian national health databases (Capitation Payment Unit database and Vital Statistics from DANE). We developed an algorithm based on ICD-10 codes and oncological procedures to select incident cases of breast, stomach and colorectal cancer. To measure healthcare fragmentation, we identified the number of providers between the dates of the first and last registered services. For patients who died during observation we adjusted the number of providers for survival time in days, otherwise survival time was set to 31 december, 2017. We categorized fragmentation in quartiles and evaluated its effect on mortality rate by Kaplan Meier estimates. Results: We identified three cohorts of patients based on primary tumor site. Age distribution was similar in stomach and colorectal cancer. Fragmentation measured as a continuous variable has a non-parametric distribution in all cohorts. The median of follow-up time ranged between 2.4 to 4.4 years. All-cause mortality rates were highest in stomach cancer, lowest in breast cancer. When measured as quartiles, fragmentation has a consistent dose-response effect increasing all-cause mortality rates. Conclusions: Healthcare fragmentation can be measured through algorithms applied to administrative databases in Colombia. Fragmentation is a predictor for all-cause mortality across different oncologic populations. This measurement based on real-world national administrative data could be used as an indicator of high-quality oncological healthcare for the Colombian healthcare system.[Table: see text]