The ability of triggers to retrospectively predict potentially preventable adverse events in a sample of deceased patients

Several trigger systems have been developed to screen medical records of hospitalized patients for adverse events (AEs). Because it's too labor-intensive to screen the records of all patients, usually a sample is screened. Our sample consists of patients who died during their stay because chances of finding preventable AEs in this subset are highest. Records were reviewed for fifteen triggers (n = 2182). When a trigger was present, the records were scrutinized by specialized medical doctors who searched for AEs. The positive predictive value (PPV) of the total trigger system and of the individual triggers was calculated. Additional analyses were performed to identify a possible optimization of the trigger system. In our sample, the trigger system had an overall PPV for AEs of 47%, 17% for potentially preventable AEs. More triggers present in a record increased the probability of detecting an AE. Adjustments to the trigger system slightly increased the positive predictive value but missed about 10% of the AEs detected with the original system. In our sample of deceased patients the trigger system has a PPV comparable to other samples. However still, an enormous amount of time and resources are spent on cases without AEs or with non-preventable AEs. Possibly, the performance could be further improved by combining triggers with clinical scores and laboratory results. This could be promising in reducing the costly and labor-intensive work of screening medical records.