Journal of Biomedical Informatics

Comparing clinical decision support systems for improving follow-up of abnormal cervical cancer screening test results

Many individuals with abnormal cervical cancer screening test results do not receive timely follow-up care. Clinical decision support systems (CDSS) to improve follow-up are challenged by difficulty identifying clinical elements and applying complex guideline recommendations. As part of a multisite trial, two CDSS models were implemented: one used natural language processes to evaluate extracted data outside of the electronic health record (EHR) (System A); the other used commercial EHR functionality using LOINC-defined result fields (System B). This secondary analysis compared the accuracy and trial outcomes among sites using these two CDSS models. Primary care clinics (32 in System A and 12 in System B) were randomly assigned to usual care, CDSS alone, or CDSS with patient outreach with or without navigation. CDSS identified individuals with overdue abnormal screening results and specified the recommended follow-up and time interval. CDSS accuracy was assessed by manual chart review. Patient outreach consisted of portal/mailed letters plus a single phone call. Navigation included one or more phone calls to address barriers to care. Completion of recommended follow-up at 120 days after enrollment was the primary outcome. Clinic was the unit of randomization, and the patient was the unit of analysis. Between October 2020 and December 2021, 2596 patients with abnormal results were identified by the CDSS. CDSS true positives were 61.3 % in System A and 70.4 % in System B. CDSS alone versus usual care did not improve outcomes in either system. CDSS with patient outreach with or without navigation versus usual care significantly increased follow-up rates in System A (38.2 % or 37.2 % vs 23.5 %, p < 0.001) and System B (25.4 % or 23 % vs. 19.7 %, p = 0.044). Two CDSS models developed to identify overdue abnormal cervical cancer screening test results had moderate accuracy. Both models with patient outreach with or without navigation - but not CDSS alone - increased recommended follow-up. Future CDSS for cervical cancer screening may be improved with open-source tools developed in public-private partnerships.

Explainable discovery of disease biomarkers: The case of ovarian cancer to illustrate the best practice in machine learning and Shapley analysis

Ovarian cancer is a significant health issue with lasting impacts on the community. Despite recent advances in surgical, chemotherapeutic and radiotherapeutic interventions, they have had only marginal impacts due to an inability to identify biomarkers at an early stage. Biomarker discovery is challenging, yet essential for improving drug discovery and clinical care. Machine learning (ML) techniques are invaluable for recognising complex patterns in biomarkers compared to conventional methods, yet they can lack physical insights into diagnosis. eXplainable Artificial Intelligence (XAI) is capable of providing deeper insights into the decision-making of complex ML algorithms increasing their applicability. We aim to introduce best practice for combining ML and XAI techniques for biomarker validation tasks. We focused on classification tasks and a game theoretic approach based on Shapley values to build and evaluate models and visualise results. We described the workflow and apply the pipeline in a case study using the CDAS PLCO Ovarian Biomarkers dataset to demonstrate the potential for accuracy and utility. The case study results demonstrate the efficacy of the ML pipeline, its consistency, and advantages compared to conventional statistical approaches. The resulting guidelines provide a general framework for practical application of XAI in medical research that can inform clinicians and validate and explain cancer biomarkers.

Federated causal discovery with missing data in a multicentric study on endometrial cancer

Establishing causal dependencies is crucial in applied domains, such as medicine and healthcare, where decision-making must be explainable. In these settings, small sample sizes and missing data call for federated approaches to maximise the amount of information we can use. We propose a novel federated causal discovery algorithm capable of pooling information from multiple sources with heterogeneous missing data to learn a graph representing cause-effect relationships. In particular, we learn a causal graph on a centralised server while taking into account both prior knowledge and missingness mechanism specific to each client. We applied the proposed algorithm to synthetic data and real-world data from a multicentric study on endometrial cancer, validating the obtained causal graph through quantitative analyses and a clinical literature review. Our approach learns an accurate model despite data missing not-at-random.