Supervised learning for infection risk inference using pathology data

Abstract

BackgroundAntimicrobial Resistance is threatening our ability to treat common infectious diseases and overuse of antimicrobials to treat human infections in hospitals is accelerating this process. Clinical Decision Support Systems (CDSSs) have been proven to enhance quality of care by promoting change in prescription practices through antimicrobial selection advice. However, bypassing an initial assessment to determine the existence of an underlying disease that justifies the need of antimicrobial therapy might lead to indiscriminate and often unnecessary prescriptions.MethodsFrom pathology laboratory tests, six biochemical markers were selected and combined with microbiology outcomes from susceptibility tests to create a unique dataset with over one and a half million daily profiles to perform infection risk inference. Outliers were discarded using the inter-quartile range rule and several sampling techniques were studied to tackle the class imbalance problem. The first phase selects the most effective and robust model during training using ten-fold stratified cross-validation. The second phase evaluates the final model after isotonic calibration in scenarios with missing inputs and imbalanced class distributions.ResultsMore than 50% of infected profiles have daily requested laboratory tests for the six biochemical markers with very promising infection inference results: area under the receiver operating characteristic curve (0.80-0.83), sensitivity (0.64-0.75) and specificity (0.92-0.97). Standardization consistently outperforms normalization and sensitivity is enhanced by using the SMOTE sampling technique. Furthermore, models operated without noticeable loss in performance if at least four biomarkers were available.ConclusionThe selected biomarkers comprise enough information to perform infection risk inference with a high degree of confidence even in the presence of incomplete and imbalanced data. Since they are commonly available in hospitals, Clinical Decision Support Systems could benefit from these findings to assist clinicians in deciding whether or not to initiate antimicrobial therapy to improve prescription practices.

Predicting Quality of Overnight Glycaemic Control in Type 1 Diabetes Using Binary Classifiers

“…The work-flow proposed by Hernandez et al was used to build and evaluate the models (see Fig. 3) [34]. In this process, the data was initially divided into cross-validation (CVS) and hold-out datasets (HOS), with the latter dataset comprising the 25% of the observations.…”

Machine learning in infection management using routine electronic health records: tools, techniques, and reporting of future technologies

“…transformation to binary variables indicating missingness [22], carryforward of last observation [12,14,20,23e26], including complete cases only [18,27e30], or applying multiple imputation [11,17,31e35]. Two studies assessed the effect of missing data on model performance through a stepwise introduction of missing variables [36,37]. Class imbalance of the labelled outcome variable was explicitly mentioned if applicable in 39% (n ¼ 18) of the studies.…”

“…Thirty studies (57.7%) compared multiple ML techniques. Among these studies, the best performing techniques per research area were long short-term memory networks (LSTM) in the sepsis group [12,46], ANN in the HAI group [28], L1-regularized logistic regression (L1LR) in the SSI and other postoperative infections group [26], SVM in the infections (general) group [37], classification and regression tree (CART) in the microbiological test results group [47], and stochastic gradient boosting (SGB) in the musculoskeletal infections group [35]. However, as outlined below, the definition of the predicted outcome can be very heterogenous.…”

Corrigendum

“…Connectivity with supporting laboratories would also facilitate audits and quality assurance. 5 Expanding connectivity would allow clinical decision support to use the 6 This capacity would be important in settings where specialist input is unavailable and would support optimization of antimicrobial use.…”