Machine Learning Algorithms to Predict Colistin-Induced Nephrotoxicity from Electronic Health Records in Patients with Multidrug-Resistant Gram-Negative Infection

OBJECTIVES

Colistin-induced nephrotoxicity prolongs hospitalization and increases mortality. The study aimed to construct machine learning models to predict colistin-induced nephrotoxicity in patients with multidrug-resistant gram-negative infection.

METHODS

Patients receiving colistin from three hospitals in the Clinical Research Database were included. Data were divided into a derivation cohort (2011∼2017) and a temporal validation cohort (2018∼2020). Fifteen machine learning models were established by categorical boosting, light gradient boosting machine, and random forest. Classifier performances were compared by the sensitivity, F1 score, Matthews correlation coefficient (MCC), area under the receiver operating characteristic (AUROC) curve, and area under the precision-recall curve (AUPRC). SHapley Additive exPlanations plots were drawn to understand feature importance and interactions.

RESULTS

The study included 1392 patients, with 360 (36.4%) and 165 (40.9%) experiencing nephrotoxicity in the derivation and temporal validation cohorts, respectively. The categorical boosting with oversampling achieved the highest performance with a sensitivity of 0.860, an F1 score of 0.740, an MCC of 0.533, an AUROC curve of 0.823, and an AUPRC of 0.737. The feature importance demonstrated that the days of colistin use, cumulative dose, daily dose, latest C-reactive protein, and baseline hemoglobin were the most important risk factors, especially for vulnerable patients. A cutoff colistin dose of 4.0 mg/kg body weight/day was identified for patients at higher risk of nephrotoxicity.

CONCLUSIONS

Machine learning techniques can be an early identification tool to predict colistin-induced nephrotoxicity. The observed interactions suggest a modification in dose adjustment guidelines. Future geographic and prospective validation studies are warranted to strengthen the real-world applicability.

Estimating the Heavy Metal Contents in Entisols from a Mining Area Based on Improved Spectral Indices and Catboost

In the study of the inversion of soil multi-species heavy metal element concentrations using hyperspectral techniques, the selection of feature bands is very important. However, interactions among soil elements can lead to redundancy and instability of spectral features. In this study, heavy metal elements (Pb, Zn, Mn, and As) in entisols around a mining area in Harbin, Heilongjiang Province, China, were studied. To optimise the combination of spectral indices and their weights, radar plots of characteristic-band Pearson coefficients (RCBP) were used to screen three-band spectral index combinations of Pb, Zn, Mn, and As elements, while the Catboost algorithm was used to invert the concentrations of each element. The correlations of Fe with the four heavy metals were analysed from both concentration and characteristic band perspectives, while the effect of spectral inversion was further evaluated via spatial analysis. It was found that the regression model for the inversion of the Zn elemental concentration based on the optimised spectral index combinations had the best fit, with R2 = 0.8786 for the test set, followed by Mn (R2 = 0.8576), As (R2 = 0.7916), and Pb (R2 = 0.6022). As far as the characteristic bands are concerned, the best correlations of Fe with the Pb, Zn, Mn and As elements were 0.837, 0.711, 0.542 and 0.303, respectively. The spatial distribution and correlation of the spectral inversion concentrations of the As and Mn elements with the measured concentrations were consistent, and there were some differences in the results for Zn and Pb. Therefore, hyperspectral techniques and analysis of Fe elements have potential applications in the inversion of entisols heavy metal concentrations and can improve the quality monitoring efficiency of these soils.

Factors affecting injury severity in motorcycle crashes: Different age groups analysis using Catboost and SHAP techniques

OBJECTIVE

Motorcycle crashes often result in severe injuries on roads that affect people's lives physically, financially, and psychologically. These injuries could be notably harmful to drivers of all age groups. The main objective of this study is to investigate the risk factors contributing to the severity of crash injuries in different age groups.

METHODS

This Objective is achieved by developing accurate machine learning (ML) based prediction models. This research examines the relationship between potential risk factors of motorcycle-associated crashes using (ML) and Shapley Additive explanations (SHAP) technique. The SHAP technique further helped interpreting ML methods for traffic injury severity prediction. It indicates the significant non-linear interactions between dependent and independent variables. The data for this study was collected from the Provincial Emergency Response Service RESCUE 1122 for the Rawalpindi region (Pakistan) over three years (from 2017 to 2020). The Synthetic Minority Oversampling Technique (SMOTE) is employed to balance injury severity classes in the pre-processing phase.

RESULTS

The results demonstrate that age, gender, posted speed limit, the number of lanes, and month of the year are positively associated with severe and fatal injuries. This research also assesses how the modeling framework varies between the ML and classical statistical methods. The predictive performance of proposed ML models was assessed using several evaluation metrics, and it is found that Catboost outperformed the XGBoost, Random Forest (RF) and Multinomial Logit (MNL) model.

CONCLUSION

The findings of this study will assist road users, road safety authorities, stakeholders, policymakers, and decision-makers in obtaining substantial and essential guidance for reducing the severity of crash injuries in Pakistan and other countries with prevailing conditions.