With an increasing focus on measuring and withholding payment for complications of care not present on admission (POA), hospitals have great incentive to know the limitations of data that track potential patient safety problems. This study combined POA data collected from two statewide discharge databases and used them to analyze 13 of 20 AHRQ Patient Safety Indicators (PSIs) for which POA information was relevant. Investigators discovered that suspect coding of POA did exist in some circumstances, but more notably, significant limitations on the validity of 3 PSIs were found after incorporating POA information. Decubitus ulcers, postoperative hip fractures, and postoperative thromboembolic events all appear as invalid measures of in-hospital patient safety events, while the other PSIs studied remain potentially useful.