Towards similarity-based differential diagnostics for common diseases

Abstract

Ontology-based phenotype profiles have been utilised for the purpose of differential diagnosis of rare genetic diseases, and for decision support in specific disease domains. Particularly, semantic similarity facilitates diagnostic hypothesis generation through comparison with disease phenotype profiles. However, the approach has not been applied for differential diagnosis of common diseases, or generalised clinical diagnostics from uncurated text-derived phenotypes. In this work, we describe the development of an approach for deriving patient phenotype profiles from clinical narrative text, and apply this to text associated with MIMIC-III patient visits. We then explore the use of semantic similarity with those text-derived phenotypes to classify primary patient diagnosis, comparing the use of patient-patient similarity and patient-disease similarity using phenotype-disease profiles previously mined from literature. We also consider a combined approach, in which literature-derived phenotypes are extended with the content of text-derived phenotypes we mined from 500 patients. The results reveal a powerful approach, showing that in one setting, uncurated text phenotypes can be used for differential diagnosis of common diseases, making use of information both inside and outside the setting. While the methods themselves should be explored for further optimisation, they could be applied to a variety of clinical tasks, such as differential diagnosis, cohort discovery, document and text classification, and outcome prediction.

Evaluating Semantic Similarity Methods for Comparison of Text-derived Phenotype Profiles

“…We also tested the setting that was used in our previous work, that is the combination of Resnik pairwise, BMA, and Resnik IC [23]. In this experiment, this yielded an AUC of 0.73, and an MRR-0 of 0.23, lower than the previously reported results of 0.77 and 0.42, respectively.…”

“…One such area that the experimental platform could be turned towards, is the prediction of patient diagnosis through comparison with disease profiles, rather than with other patients. In our previous work, we identified that prediction of patient diagnosis from text-derived phenotypes was best when using SS to compare patients to disease profiles mined from literature, and extended with in-context training from patient profiles [23]. We plan to follow this study up with another, comparing patients with different sizes and derivations of disease profile.…”

“…We evaluated performance using Area Under the receiver operating characteristic Curve (AUC), Mean Reciprocal Rank (MRR), and Top Ten Accuracy (the percentage of patients for whom the correct diagnosis was in the top ten most similar entities). One previously identified limitation of analyses that compared patient phenotype profiles for prediction of shared diagnosis was that no correct diagnosis can be found for an admission if there is no other admission in the set with a matching primary diagnosis [23]. To 4/12 identify whether, and to what extent, a lack of matching diagnoses for any admissions in our sample negatively affected scores, we also evaluated a modified MRR metric that removed values of MRR that were 0, since values of 0 indicate there were no admissions with a matching primary diagnosis (and therefore the correct label did not appear at all in the ranking).…”

Multi-faceted Semantic Clustering With Text-derived Phenotypes

“…With respect to clustering, what remains to be done, is to determine a method that enables the re-integration of these scores, clusters, and groupings, into a single representation that minimises loss of information. Such an approach could lead to powerful insights into multi-and co-morbidity, as well as to improve semantic similarity based classification using text-derived phenotypes, such as that described by our previous work [11]. One method of approaching this specific to clustering problems could be the consideration of Multi-View Clustering, which could consider each measure of facet-wise similarity as a different view of the same patient admission [54], which could then be further analysed to determine an optimal set of clusters.…”

Klarigi: Characteristic Explanations for Semantic Data

“…Ontology-based analysis has been leveraged across many tasks including prediction of protein interaction and rare disease variants [1]. In the clinical space, similar analysis methods have been applied across a wide range of applications including diagnosis of rare and common diseases [2,3], as well as the identification of subtypes of diseases, such as autism [4]. In addition, the synthesis of ontology-based methods and machine learning is increasingly common [5].…”