The estimation of semantic similarity between words is an important task in many language related applications. In the past, several approaches to assess similarity by evaluating the knowledge modelled in an ontology have been proposed.... more
The estimation of semantic similarity between words is an important task in many language related applications. In the past, several approaches to assess similarity by evaluating the knowledge modelled in an ontology have been proposed. However, in many domains, knowledge is dispersed through several partial and/or overlapping ontologies. Because most previous works on semantic similarity only support a unique input ontology, we propose a method to enable similarity estimation across multiple ontologies. Our ...
Purpose To develop and validate a case-finding algorithm for the identification of Non-Small Cell Lung Cancer (NSCLC) cases in a region-wide Italian pathology registry (PR). Materials and methods Data collected between 2009 and 2017 in... more
Purpose To develop and validate a case-finding algorithm for the identification of Non-Small Cell Lung Cancer (NSCLC) cases in a region-wide Italian pathology registry (PR).
Materials and methods Data collected between 2009 and 2017 in the PR and the Pharmacy Database of the Univer- sity Hospital of Siena and the PR of Tuscany region were used. A NSCLC-identification algorithm based on free-text keywords and SNOMED morphology and topography codes was designed and tested on data from Siena: indication for drug use (i.e. NSCLC) was the reference standard for sensitivity (SE); positive predictive value (PPV) was estimated through manual review. Algorithm modifications were then tested to improve algorithm per- formance: PPV was calculated against validated dataset from PR of Siena; a range of SE [min-max] was estimated in PR of Tuscany using analytical formulae that assumed NSCLC incidence equal either to 80% or 90% of overall lung cancer incidence recorded in Tuscany. The algorithm modification with the best performance was chosen as the final version of the algorithm. A random sample of 200 cases was extracted from the PR of Tuscany for manual review.
Results The first version of the algorithm showed a PPV of 74.7% and SE of 79% in PR of Siena. The final version of the algorithm had a SE in PR of Tuscany that grew with calendar time (2009 = [24.7%-28%]; 2017 = [57.9%-65.1%]) and a PPV of 93%.
Conclusions The final NSCLC-finding algorithm showed with very high PPV. SE was in line with the expected contribution of PR to overall cases captured in the regional Cancer Registry, with a trend of increase over calendar time. Given the promising algorithm validity and the wide use of SNOMED terminology in electronic pathology records, the proposed algorithm is expected to be easily adapted to other electronic databases for (pharmaco)epidemiology purposes.