There is a huge increase in the use of medico-administrative databases for research purposes. For this reason, some selection algorithms have been validated with cancer registry data to detect incident cancer cases. However, there is no validated algorithm to identify Non-Hodgkin's Lymphoma (NHL) and Hodgkin's Lymphoma (HL) incident cases. Aims of this study were to assess the validity of diagnosis codes in the PMSI database to identify incident cases of Lymphomas according to results of the Tarn cancer registry and to compare baseline characteristics of lymphoma cases using both sources.

Between 2010 and 2013, inhabitants of the tarn department with lymphomas were identified in the PMSI through two algorithms of selection (combination of diagnosis codes of lymphomas and chemotherapy for Algorithm 1, only diagnosis code of lymphomas for algorithm 2). Then, only incident lymphoma cases were selected using the past history of patients. An incident case must have no previous record of lymphoma diagnosis during an observation period of 24months. The results of the selection and characteristics of incident lymphomas cases were compared with data from the tarn cancer registry considered as the “gold standard” in this area. Each algorithm performance was assessed by estimating sensitivity (Se), predictive positive value (PPV), specificity (Spe) and negative predictive value (NPV).

Between 2010 and 2013, the registry identified 485 validated incident cases of lymphomas, of which 52 HL cases and 433 NHL cases. For corresponding area and period, algorithm 1 provides a number of incident cases close to the registry whereas Algorithm 2 overestimated the number of incident cases by approximately 30%. For algorithm 1, sensitivity and PPV were closed, respectively 66.2% and 67.7%. For algorithm 2, sensitivity was increased by up 10% whereas PPV was decreased by up 9%. Both algorithms presented high specificity and NPV (≈98%). The comparative analysis illustrates that similar distribution and characteristics are observed when using both sources.

Given these results, identifying incident Lymphoma patients through data collected from the French national health insurance database may be useful to conduct national pharmaco-epidemiological studies on Lymphomas. These results suggest that using both diagnosis codes and chemotherapy codes appears to be a more efficient method to detect incident cases.