Background: We aimed at measuring the positive predictive value (PPV) of data in the French Hospital Medical Information Database (FHD).
Summary: This retrospective multicenter study included 31 hospitals from where 56 hospital stays were randomly selected among all hospitalizations for the years 2009 and 2010 with at least 1 principal diagnosis of stroke or transient ischemic attack (TIA). Three algorithms were evaluated. Algorithm 1 selected discharge abstracts with at least 1 principal diagnosis identified by one of the relevant International Classification of Diseases, 10th revision codes. Algorithm 2 selected stays with 1 principal diagnosis of the whole stay, but without the dates of the stay. Algorithm 3 took into account the kind of medical wards. The PPV of each algorithm was calculated using medical records as the reference. We found 1,669 discharge abstracts with a diagnosis of stroke among the 1,680 that were randomly selected. The neurologist's review revealed 196 false-positive cases providing a global PPV of 88.26% for algorithm 1, 89.96% for algorithm 2 and 92.74% for algorithm 3.
Key messages: It was possible to build an algorithm to optimize the FHD for stroke and TIA reporting, with a PPV at 90%. The FHD could be a good tool to measure the burden of stroke in France.
© 2015 S. Karger AG, Basel.