Indicators of inappropriate tumour marker use through the mining of electronic health records.

RATIONALE, AIMS, AND OBJECTIVES: Although the issue of monitoring appropriateness of tumour markers (TMs) request in outpatients remains crucial, proper indicators are still demanding. The present study developed and explored indicators of inappropriate TM ordering in outpatients through the data mining of electronic health records (EHRs). METHODS: Carcinoembryonic antigen (CEA), alfa-fetoprotein (AFP), carbohydrate antigen (CA)125, CA15.3, CA19.9, and prostate-specific antigen (PSA) ordered in outpatients during a year were examined by mining EHRs of a Local Health Authority in Italy. Evidence-based criteria were used to develop performance indicators. Demographic and clinical information associated with TM orders were examined. RESULTS: A total of 80 813 TMs were ordered in 52 536 outpatients (1.54 markers/patient). Indicators related to disease codes, gender, age, and TM repetitions were developed, and their application showed that (1) CA15.3 and CEA are prevalently requested in patients with cancer (79.2% and 65.6%) whereas the other TMs are largely requested also in patients without cancer; (2) requests of PSA in women and of CA125 or CA15.3 in men are negligible; (3) although requests in people older than 80 years are relevant (16.4% of total), the highest rate of request of all markers occurs in patients aged 40 to 79 years; (4) CA15.3 and CEA are mainly requested in cancer cases between 50 and 79 years and AFP, CA19.9, and CA125 in those between 60 and 69 years; (5) <50% of PSA orders are associated with cancer code for all age intervals; and (6) multiple repetitions of AFP, CA125, CA15.3, CA19.9, and CEA are prevalent in cancer patients or benign diseases to which TMs are appropriate, whereas PSA repetitions occur mainly in patients without cancer. CONCLUSIONS: The developed indicators resulted suitable to monitor TM overordering in outpatients through the mining of EHRs. The present study is a first approach towards the use of big-data mining for TM appropriateness evaluation purposes.