Search

Background: Cancer stage at diagnosis is essential for understanding cancer outcomes, guiding cancer control activities and healthcare services, and enabling benchmarking nationally and internationally. Yet, most cancer registries in Australia do not routinely collect this data. This study explored key stakeholders' perceptions of implementing cancer staging utilising Natural Language Processing and Machine Learning algorithms within the Western Australian Cancer Registry. Methods: Perceptions of key breast and colorectal cancer stakeholders, including registry staff, clinicians, consumers, data scientists, biostatisticians, data management, healthcare staff, and health researchers, were collected. Prospective and retrospective qualitative proformas at two-time points of the Western Australian Cancer Staging Project were employed. The Consolidated Framework for Implementation Research was used to guide data collection, analysis and interpretation embedded in a Participatory Action Research approach. Data analysis also incorporated Framework Analysis and an adapted version of grading qualitative data using a visual traffic light labelling system to highlight the levels of positivity, negativity, and implementation concern. Results: Twenty-nine pre-proformas and 18 post-proformas were completed online via REDCap. The grading and visual presentation of barriers and enablers aided interpretation and reviewing predicted intervention outcomes. Of the selected constructs, complexity (the perceived difficulty of the intervention) was the strongest barrier and tension for change (the situation needing change) was the strongest enabler. Implementing cancer staging into the Western Australian Cancer Registry was considered vital. Benefits included improved knowledge and understanding of various outcomes (e.g., treatment received as per Optimum Care Pathways) and benchmarking. Barriers included compatibility issues with current systems/workflows, departmental/higher managerial support, and future sustainment. Conclusions: The findings aid further review of data gaps, additional cancer streams, standardising cancer staging and future improvements. The study offers an adapted version of a rapid qualitative data collection and analytic approach for establishing barriers and enablers. The findings may also assist other population-based cancer registries considering collecting cancer stage at diagnosis. [ABSTRACT FROM AUTHOR]

Objectives: We aimed to automate routine extraction of clinically relevant unstructured information from uro-oncological histopathology reports by applying rule-based and machine learning (ML)/deep learning (DL) methods to develop an oncology focused natural language processing (NLP) algorithm. Methods: Our algorithm employs a combination of a rule-based approach and support vector machines/neural networks (BioBert/Clinical BERT), and is optimised for accuracy. We randomly extracted 5772 uro-oncological histology reports from 2008 to 2018 from electronic health records (EHRs) and split the data into training and validation datasets in an 80:20 ratio. The training dataset was annotated by medical professionals and reviewed by cancer registrars. The validation dataset was annotated by cancer registrars and defined as the gold standard with which the algorithm outcomes were compared. The accuracy of NLP-parsed data was matched against these human annotation results. We defined an accuracy rate of >95% as “acceptable” by professional human extraction, as per our cancer registry definition. Results: There were 11 extraction variables in 268 free-text reports. We achieved an accuracy rate of between 61.2% and 99.0% using our algorithm. Of the 11 data fields, a total of 8 data fields met the acceptable accuracy standard, while another 3 data fields had an accuracy rate between 61.2% and 89.7%. Noticeably, the rule-based approach was shown to be more effective and robust in extracting variables of interest. On the other hand, ML/DL models had poorer predictive performances due to highly imbalanced data distribution and variable writing styles between different reports and data used for domain-specific pre-trained models. Conclusion: We designed an NLP algorithm that can automate clinical information extraction accurately from histopathology reports with an overall average micro accuracy of 93.3%.

Histopathology reports are a primary data source for the case definition phase of a Cancer Registry. By reading the histopathology report, the operator that evaluates an oncology case can define the morphology and topography of cancer, and validate the case with the highest diagnosis base. The key problem of the Catania-Messina-Enna Integrated Cancer Registry (RTI) is that these reports are written in natural language and relevant information for cancer evaluation is only a little part of the total annual histopathological reports. In this population-based retrospective cohort study, we try to optimize the working time spent by the RTI operators in seeking and selecting the right information among the histopathology reports in the east Sicily population, by developing a binary classifier on a training set of labeled historical data and validating its outcome by a test set of labeled data created by the operators during the years. Using a machine learning algorithm we built a classification model that evaluates each free text report and returns a score that indicates the probability that it contains oncologic relevant information. The best performing algorithm, among the eight analyzed in this study, was the LightGBM that reached an F1-Score of 98.9%. Using the chosen classifier we shortened the time for case evaluation, improving the timeliness of cancer statistics. [ABSTRACT FROM AUTHOR]