A clustering approach for detecting implausible observation values in electronic health records data