Statistical-learning strategies generate only modestly performing predictive models for urinary symptoms following external beam radiotherapy of the prostate: A comparison of conventional and machine-learning methods

Frequently asked questions on erectile dysfunction: evaluating artificial intelligence answers with expert mentorship

The present study assessed the accuracy of artificiaI intelligence-generated responses to frequently asked questions on erectile dysfunction. A cross-sectional analysis involved 56 erectile dysfunction-related questions searched on Google, categorized into nine sections: causes, diagnosis, treatment options, treatment complications, protective measures, relationship with other illnesses, treatment costs, treatment with herbal agents, and appointments. Responses from ChatGPT 3.5, ChatGPT 4, and BARD were evaluated by two experienced urology experts using the F1 and global quality scores (GQS) for accuracy, relevance, and comprehensibility. ChatGPT 3.5 and ChatGPT 4 achieved higher GQS than BARD in categories such as causes (4.5 ± 0.54, 4.5 ± 0.51, 3.15 ± 1.01, respectively, p < 0.001), treatment options (4.35 ± 0.6, 4.5 ± 0.43, 2.71 ± 1.38, respectively, p < 0.001), protective measures (5.0 ± 0, 5.0 ± 0, 4 ± 0.5, respectively, p = 0.013), relationships with other illnesses (4.58 ± 0.58, 4.83 ± 0.25, 3.58 ± 0.8, respectively, p = 0.006), and treatment with herbal agents (3 ± 0.61, 3.33 ± 0.83, 1.8 ± 1.09, respectively, p = 0.043). F1 scores in categories: causes (1), diagnosis (0.857), treatment options (0.726), and protective measures (1), indicated their alignment with the guidelines. There was no significant difference between ChatGPT 3.5 and ChatGPT 4 regarding answer quality, but both outperformed BARD in the GQS. These results emphasize the need to continually enhance and validate AI-generated medical information, underscoring the importance of artificiaI intelligence systems in delivering reliable information on erectile dysfunction.

Comparing Performances of Predictive Models of Toxicity after Radiotherapy for Breast Cancer Using Different Machine Learning Approaches

PURPOSE

Different ML models were compared to predict toxicity in RT on a large cohort (n = 1314).

METHODS

The endpoint was RTOG G2/G3 acute toxicity, resulting in 204/1314 patients with the event. The dataset, including 25 clinical, anatomical, and dosimetric features, was split into 984 for training and 330 for internal tests. The dataset was standardized; features with a high p-value at univariate LR and with Spearman ρ>0.8 were excluded; synthesized data of the minority were generated to compensate for class imbalance. Twelve ML methods were considered. Model optimization and sequential backward selection were run to choose the best models with a parsimonious feature number. Finally, feature importance was derived for every model.

RESULTS

The model's performance was compared on a training-test dataset over different metrics: the best performance model was LightGBM. Logistic regression with three variables (LR3) selected via bootstrapping showed performances similar to the best-performing models. The AUC of test data is slightly above 0.65 for the best models (highest value: 0.662 with LightGBM).

CONCLUSIONS

No model performed the best for all metrics: more complex ML models had better performances; however, models with just three features showed performances comparable to the best models using many (n = 13-19) features.

Assessment of gene expressions from squamous cell carcinoma of the head and neck to predict radiochemotherapy-related xerostomia and dysphagia

PURPOSE

We tested the hypothesis that gene expressions from biopsies of locally advanced head and neck squamous cell carcinoma (HNSCC) patients can supplement dose-volume parameters to predict dysphagia and xerostomia following primary radiochemotherapy (RCTx).

MATERIAL AND METHODS

A panel of 178 genes previously related to radiochemosensitivity of HNSCC was considered for nanoString analysis based on tumour biopsies of 90 patients with locally advanced HNSCC treated by primary RCTx. Dose-volume parameters were extracted from the parotid, submandibular glands, oral cavity, larynx, buccal mucosa, and lips. Normal tissue complication probability (NTCP) models were developed for acute, late, and for the improvement of xerostomia grade ≥2 and dysphagia grade ≥3 using a cross-validation-based least absolute shrinkage and selection operator (LASSO) approach combined with stepwise logistic regression for feature selection. The final signatures were included in a logistic regression model with optimism correction. Performance was assessed by the area under the receiver operating characteristic curve (AUC).

RESULTS

NTCP models for acute and late xerostomia and the improvement of dysphagia resulted in optimism-corrected AUC values of 0.84, 0.76, and 0.70, respectively. The minimum dose to the contralateral parotid was selected for both acute and late xerostomia and the minimum dose to the larynx was selected for dysphagia improvement. For the xerostomia endpoints, the following gene expressions were selected: RPA2 (cellular response to DNA damage), TCF3 (salivary gland cells development), GBE1 (glycogen storage and regulation), and MAPK3 (regulation of cellular processes). No gene expression features were selected for the prediction of dysphagia.

CONCLUSION

This hypothesis-generating study showed the potential of improving NTCP models using gene expression data for HNSCC patients. The presented models require independent validation before potential application in clinical practice.

Machine Learning for Predicting Discharge Disposition After Traumatic Brain Injury

BACKGROUND

Current traumatic brain injury (TBI) prognostic calculators are commonly used to predict the mortality and Glasgow Outcome Scale, but these outcomes are most relevant for severe TBI. Because mild and moderate TBI rarely reaches severe outcomes, there is a need for novel prognostic endpoints.

OBJECTIVE

To generate machine learning (ML) models with a strong predictive capacity for trichotomized discharge disposition, an outcome not previously used in TBI prognostic models. The outcome can serve as a proxy for patients' functional status, even in mild and moderate patients with TBI.

METHODS

Using a large data set (n = 5292) of patients with TBI from a quaternary care center and 84 predictors, including vitals, demographics, mechanism of injury, initial Glasgow Coma Scale, and comorbidities, we trained 6 different ML algorithms using a nested-stratified-cross-validation protocol. After optimizing hyperparameters and performing model selection, isotonic regression was applied to calibrate models.

RESULTS

When maximizing the microaveraged area under the receiver operating characteristic curve during hyperparameter optimization, a random forest model exhibited top performance. A random forest model was also selected when maximizing the microaveraged area under the precision-recall curve. For both models, the weighted average area under the receiver operating characteristic curves was 0.84 (95% CI 0.81-0.87) and the weighted average area under the precision-recall curves was 0.85 (95% CI 0.82-0.88).

CONCLUSION

Our group presents high-performing ML models to predict trichotomized discharge disposition. These models can assist in optimization of patient triage and treatment, especially in cases of mild and moderate TBI.

Machine Learning for Predicting In-Hospital Mortality After Traumatic Brain Injury in Both High-Income and Low- and Middle-Income Countries

BACKGROUND

Machine learning (ML) holds promise as a tool to guide clinical decision making by predicting in-hospital mortality for patients with traumatic brain injury (TBI). Previous models such as the international mission for prognosis and clinical trials in TBI (IMPACT) and the corticosteroid randomization after significant head injury (CRASH) prognosis calculators can potentially be improved with expanded clinical features and newer ML approaches.

OBJECTIVE

To develop ML models to predict in-hospital mortality for both the high-income country (HIC) and the low- and middle-income country (LMIC) settings.

METHODS

We used the Duke University Medical Center National Trauma Data Bank and Mulago National Referral Hospital (MNRH) registry to predict in-hospital mortality for the HIC and LMIC settings, respectively. Six ML models were built on each data set, and the best model was chosen through nested cross-validation. The CRASH and IMPACT models were externally validated on the MNRH database.

RESULTS

ML models built on National Trauma Data Bank (n = 5393, 84 predictors) demonstrated an area under the receiver operating curve (AUROC) of 0.91 (95% CI: 0.85-0.97) while models constructed on MNRH (n = 877, 31 predictors) demonstrated an AUROC of 0.89 (95% CI: 0.81-0.97). Direct comparison with CRASH and IMPACT models showed significant improvement of the proposed LMIC models regarding AUROC (P = .038).

CONCLUSION

We developed high-performing well-calibrated ML models for predicting in-hospital mortality for both the HIC and LMIC settings that have the potential to influence clinical management and traumatic brain injury patient trajectories.

Relative Performance of Machine Learning and Linear Regression in Predicting Quality of Life and Academic Performance of School Children in Norway: Data Analysis of a Quasi-Experimental Study

BACKGROUND

Machine learning techniques are increasingly being applied in health research. It is not clear how useful these approaches are for modeling continuous outcomes. Child quality of life is associated with parental socioeconomic status and physical activity and may be associated with aerobic fitness and strength. It is unclear whether diet or academic performance is associated with quality of life.

OBJECTIVE

The purpose of this study was to compare the predictive performance of machine learning techniques with that of linear regression in examining the extent to which continuous outcomes (physical activity, aerobic fitness, muscular strength, diet, and parental education) are predictive of academic performance and quality of life and whether academic performance and quality of life are associated.

METHODS

We modeled data from children attending 9 schools in a quasi-experimental study. We split data randomly into training and validation sets. Curvilinear, nonlinear, and heteroscedastic variables were simulated to examine the performance of machine learning techniques compared to that of linear models, with and without imputation.

RESULTS

We included data for 1711 children. Regression models explained 24% of academic performance variance in the real complete-case validation set, and up to 15% in quality of life. While machine learning techniques explained high proportions of variance in training sets, in validation, machine learning techniques explained approximately 0% of academic performance and 3% to 8% of quality of life. With imputation, machine learning techniques improved to 15% for academic performance. Machine learning outperformed regression for simulated nonlinear and heteroscedastic variables. The best predictors of academic performance in adjusted models were the child's mother having a master-level education (P<.001; β=1.98, 95% CI 0.25 to 3.71), increased television and computer use (P=.03; β=1.19, 95% CI 0.25 to 3.71), and dichotomized self-reported exercise (P=.001; β=2.47, 95% CI 1.08 to 3.87). For quality of life, self-reported exercise (P<.001; β=1.09, 95% CI 0.53 to 1.66) and increased television and computer use (P=.002; β=-0.95, 95% CI -1.55 to -0.36) were the best predictors. Adjusted academic performance was associated with quality of life (P=.02; β=0.12, 95% CI 0.02 to 0.22).

CONCLUSIONS

Linear regression was less prone to overfitting and outperformed commonly used machine learning techniques. Imputation improved the performance of machine learning, but not sufficiently to outperform regression. Machine learning techniques outperformed linear regression for modeling nonlinear and heteroscedastic relationships and may be of use in such cases. Regression with splines performed almost as well in nonlinear modeling. Lifestyle variables, including physical exercise, television and computer use, and parental education are predictive of academic performance or quality of life. Academic performance is associated with quality of life after adjusting for lifestyle variables and may offer another promising intervention target to improve quality of life in children.

Spatial descriptions of radiotherapy dose: normal tissue complication models and statistical associations

For decades, dose-volume information for segmented anatomy has provided the essential data for correlating radiotherapy dosimetry with treatment-induced complications. Dose-volume information has formed the basis for modelling those associations via normal tissue complication probability (NTCP) models and for driving treatment planning. Limitations to this approach have been identified. Many studies have emerged demonstrating that the incorporation of information describing the spatial nature of the dose distribution, and potentially its correlation with anatomy, can provide more robust associations with toxicity and seed more general NTCP models. Such approaches are culminating in the application of computationally intensive processes such as machine learning and the application of neural networks. The opportunities these approaches have for individualising treatment, predicting toxicity and expanding the solution space for radiation therapy are substantial and have clearly widespread and disruptive potential. Impediments to reaching that potential include issues associated with data collection, model generalisation and validation. This review examines the role of spatial models of complication and summarises relevant published studies. Sources of data for these studies, appropriate statistical methodology frameworks for processing spatial dose information and extracting relevant features are described. Spatial complication modelling is consolidated as a pathway to guiding future developments towards effective, complication-free radiotherapy treatment.

Machine learning models for streamflow regionalization in a tropical watershed

This study aims to assess different machine learning approaches for streamflow regionalization in a tropical watershed, analyzing their advantages and limitations, and to point the benefits of using them for water resources management. The algorithms applied were: Random Forest, Earth and linear model. The response variables were the three types of minimum streamflow (Q_7.10, Q₉₅ and Q₉₀), besides the long-term average streamflow (Q_mld). The database involved 76 environmental covariates related to morphometry, topography, climate, land use and cover, and surface conditions. The elimination of covariates was performed using two processes: Pearson's correlation analysis and importance analysis by Recursive Feature Elimination (RFE). To validate the models, the following statistical metrics were used: Nash-Sutcliffe coefficient (NSE), percent bias (PBIAS), Willmott's index of agreement (d), coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE) and relative error (RE). The linear model was unsatisfactory for all response variables. The results show that nonlinear models performed well, and their covariate of greatest predictive importance was flow equivalent to the precipitated volume, considering the subtraction of an abstraction factor of 750 mm (Peq750). Generally, the Random Forest and Earth models showed similar performances and great ability to predict the minimum streamflow and long-term average streamflow assessed, constituting powerful and promising alternatives for the streamflow regionalization in support to the management and integrated planning of water resources at the level of river basins.

Machine Learning-Based Models for Prediction of Toxicity Outcomes in Radiotherapy

In order to limit radiotherapy (RT)-related side effects, effective toxicity prediction and assessment schemes are essential. In recent years, the growing interest toward artificial intelligence and machine learning (ML) within the science community has led to the implementation of innovative tools in RT. Several researchers have demonstrated the high performance of ML-based models in predicting toxicity, but the application of these approaches in clinics is still lagging, partly due to their low interpretability. Therefore, an overview of contemporary research is needed in order to familiarize practitioners with common methods and strategies. Here, we present a review of ML-based models for predicting and classifying RT-induced complications from both a methodological and a clinical standpoint, focusing on the type of features considered, the ML methods used, and the main results achieved. Our work overviews published research in multiple cancer sites, including brain, breast, esophagus, gynecological, head and neck, liver, lung, and prostate cancers. The aim is to define the current state of the art and main achievements within the field for both researchers and clinicians.

Harnessing repeated measurements of predictor variables for clinical risk prediction: a review of existing methods

BACKGROUND

Clinical prediction models (CPMs) predict the risk of health outcomes for individual patients. The majority of existing CPMs only harness cross-sectional patient information. Incorporating repeated measurements, such as those stored in electronic health records, into CPMs may provide an opportunity to enhance their performance. However, the number and complexity of methodological approaches available could make it difficult for researchers to explore this opportunity. Our objective was to review the literature and summarise existing approaches for harnessing repeated measurements of predictor variables in CPMs, primarily to make this field more accessible for applied researchers.

METHODS

MEDLINE, Embase and Web of Science were searched for articles reporting the development of a multivariable CPM for individual-level prediction of future binary or time-to-event outcomes and modelling repeated measurements of at least one predictor. Information was extracted on the following: the methodology used, its specific aim, reported advantages and limitations, and software available to apply the method.

RESULTS

The search revealed 217 relevant articles. Seven methodological frameworks were identified: time-dependent covariate modelling, generalised estimating equations, landmark analysis, two-stage modelling, joint-modelling, trajectory classification and machine learning. Each of these frameworks satisfies at least one of three aims: to better represent the predictor-outcome relationship over time, to infer a covariate value at a pre-specified time and to account for the effect of covariate change.

CONCLUSIONS

The applicability of identified methods depends on the motivation for including longitudinal information and the method's compatibility with the clinical context and available patient data, for both model development and risk estimation in practice.

Predicting Emergency Visits and Hospital Admissions During Radiation and Chemoradiation: An Internally Validated Pretreatment Machine Learning Algorithm

PURPOSE

Patients undergoing radiotherapy (RT) or chemoradiotherapy (CRT) may require emergency department evaluation or hospitalization. Early identification may direct preventative supportive care, improving outcomes and reducing health care costs. We developed and evaluated a machine learning (ML) approach to predict these events.

METHODS

A total of 8,134 outpatient courses of RT and CRT from a single institution from 2013 to 2016 were identified. Extensive pretreatment data were programmatically extracted and processed from the electronic health record (EHR). Training and internal validation cohorts were randomly generated (3:1 ratio). Gradient tree boosting (GTB), random forest, support vector machine, and least absolute shrinkage and selection operator logistic regression approaches were trained and internally validated based on area under receiver operating characteristic (AUROC) curve. The most predictive ML approach was also evaluated using only disease- and treatment-related factors to assess predictive gain of extensive EHR data.

RESULTS

All methods had high predictive accuracy, particularly GTB (validation AUROC, 0.798). Extensive EHR data beyond disease and treatment information improved accuracy (delta AUROC, 0.056). A Youden-based cutoff corresponded to validation sensitivity of 81.0% (175 of 216 courses with events) and specificity of 67.3% (1,218 of 1811 courses without events). Interpretability is an important advantage of GTB. Variable importance identified top predictive factors, including treatment (planned RT and systemic therapy), pretreatment encounters (emergency department visits and admissions in the year before treatment), vital signs (weight loss and pain score in the year before treatment), and laboratory values (albumin level at weeks before treatment).

CONCLUSION

ML predicts emergency visits and hospitalization during cancer therapy. Incorporating predictions into clinical care algorithms may help direct personalized supportive care, improve quality of care, and reduce costs. A prospective trial investigating ML-assisted direction of increased clinical assessments during RT is planned.

Machine Learning Models of Survival Prediction in Trauma Patients

BACKGROUND

We aimed to build a model using machine learning for the prediction of survival in trauma patients and compared these model predictions to those predicted by the most commonly used algorithm, the Trauma and Injury Severity Score (TRISS).

METHODS

Enrolled hospitalized trauma patients from 2009 to 2016 were divided into a training dataset (70% of the original data set) for generation of a plausible model under supervised classification, and a test dataset (30% of the original data set) to test the performance of the model. The training and test datasets comprised 13,208 (12,871 survival and 337 mortality) and 5603 (5473 survival and 130 mortality) patients, respectively. With the provision of additional information such as pre-existing comorbidity status or laboratory data, logistic regression (LR), support vector machine (SVM), and neural network (NN) (with the Stuttgart Neural Network Simulator (RSNNS)) were used to build models of survival prediction and compared to the predictive performance of TRISS. Predictive performance was evaluated by accuracy, sensitivity, and specificity, as well as by area under the curve (AUC) measures of receiver operating characteristic curves.

RESULTS

In the validation dataset, NN and the TRISS presented the highest score (82.0%) for balanced accuracy, followed by SVM (75.2%) and LR (71.8%) models. In the test dataset, NN had the highest balanced accuracy (75.1%), followed by the TRISS (70.2%), SVM (70.6%), and LR (68.9%) models. All four models (LR, SVM, NN, and TRISS) exhibited a high accuracy of more than 97.5% and a sensitivity of more than 98.6%. However, NN exhibited the highest specificity (51.5%), followed by the TRISS (41.5%), SVM (40.8%), and LR (38.5%) models.

CONCLUSIONS

These four models (LR, SVM, NN, and TRISS) exhibited a similar high accuracy and sensitivity in predicting the survival of the trauma patients. In the test dataset, the NN model had the highest balanced accuracy and predictive specificity.

A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models

OBJECTIVES

The objective of this study was to compare performance of logistic regression (LR) with machine learning (ML) for clinical prediction modeling in the literature.

STUDY DESIGN AND SETTING

We conducted a Medline literature search (1/2016 to 8/2017) and extracted comparisons between LR and ML models for binary outcomes.

RESULTS

We included 71 of 927 studies. The median sample size was 1,250 (range 72-3,994,872), with 19 predictors considered (range 5-563) and eight events per predictor (range 0.3-6,697). The most common ML methods were classification trees, random forests, artificial neural networks, and support vector machines. In 48 (68%) studies, we observed potential bias in the validation procedures. Sixty-four (90%) studies used the area under the receiver operating characteristic curve (AUC) to assess discrimination. Calibration was not addressed in 56 (79%) studies. We identified 282 comparisons between an LR and ML model (AUC range, 0.52-0.99). For 145 comparisons at low risk of bias, the difference in logit(AUC) between LR and ML was 0.00 (95% confidence interval, -0.18 to 0.18). For 137 comparisons at high risk of bias, logit(AUC) was 0.34 (0.20-0.47) higher for ML.

CONCLUSION

We found no evidence of superior performance of ML over LR. Improvements in methodology and reporting are needed for studies that compare modeling algorithms.

Mortality prediction in patients with isolated moderate and severe traumatic brain injury using machine learning models

BACKGROUND

The purpose of this study was to build a model of machine learning (ML) for the prediction of mortality in patients with isolated moderate and severe traumatic brain injury (TBI).

METHODS

Hospitalized adult patients registered in the Trauma Registry System between January 2009 and December 2015 were enrolled in this study. Only patients with an Abbreviated Injury Scale (AIS) score ≥ 3 points related to head injuries were included in this study. A total of 1734 (1564 survival and 170 non-survival) and 325 (293 survival and 32 non-survival) patients were included in the training and test sets, respectively.

RESULTS

Using demographics and injury characteristics, as well as patient laboratory data, predictive tools (e.g., logistic regression [LR], support vector machine [SVM], decision tree [DT], naive Bayes [NB], and artificial neural networks [ANN]) were used to determine the mortality of individual patients. The predictive performance was evaluated by accuracy, sensitivity, and specificity, as well as by area under the curve (AUC) measures of receiver operator characteristic curves. In the training set, all five ML models had a specificity of more than 90% and all ML models (except the NB) achieved an accuracy of more than 90%. Among them, the ANN had the highest sensitivity (80.59%) in mortality prediction. Regarding performance, the ANN had the highest AUC (0.968), followed by the LR (0.942), SVM (0.935), NB (0.908), and DT (0.872). In the test set, the ANN had the highest sensitivity (84.38%) in mortality prediction, followed by the SVM (65.63%), LR (59.38%), NB (59.38%), and DT (43.75%).

CONCLUSIONS

The ANN model provided the best prediction of mortality for patients with isolated moderate and severe TBI.

Predicting hospital and emergency department utilization among community-dwelling older adults: Statistical and machine learning approaches

OBJECTIVE

The objective of this study was to compare the performance of several commonly used machine learning methods to traditional statistical methods for predicting emergency department and hospital utilization among patients receiving publicly-funded home care services.

STUDY DESIGN AND SETTING

We conducted a population-based retrospective cohort study of publicly-funded home care recipients in the Hamilton-Niagara-Haldimand-Brant region of southern Ontario, Canada between 2014 and 2016. Gradient boosted trees, neural networks, and random forests were tested against two variations of logistic regression for predicting three outcomes related to emergency department and hospital utilization within six months of a comprehensive home care clinical assessment. Models were trained on data from years 2014 and 2015 and tested on data from 2016. Performance was compared using logarithmic score, Brier score, AUC, and diagnostic accuracy measures.

RESULTS

Gradient boosted trees achieved the best performance on all three outcomes. Gradient boosted trees provided small but statistically significant performance gains over both traditional methods on all three outcomes, while neural networks significantly outperformed logistic regression on two of three outcomes. However, sensitivity and specificity gains from using gradient boosted trees over logistic regression were only in the range of 1%-2% at several classification thresholds.

CONCLUSION

Gradient boosted trees and simple neural networks yielded small performance benefits over logistic regression for predicting emergency department and hospital utilization among patients receiving publicly-funded home care. However, the performance benefits were of negligible clinical importance.

Artificial neural network approach to predict surgical site infection after free-flap reconstruction in patients receiving surgery for head and neck cancer

Background

The aim of this study was to develop an effective surgical site infection (SSI) prediction model in patients receiving free-flap reconstruction after surgery for head and neck cancer using artificial neural network (ANN), and to compare its predictive power with that of conventional logistic regression (LR).

Materials and methods

There were 1,836 patients with 1,854 free-flap reconstructions and 438 postoperative SSIs in the dataset for analysis. They were randomly assigned tin ratio of 7:3 into a training set and a test set. Based on comprehensive characteristics of patients and diseases in the absence or presence of operative data, prediction of SSI was performed at two time points (pre-operatively and post-operatively) with a feed-forward ANN and the LR models. In addition to the calculated accuracy, sensitivity, and specificity, the predictive performance of ANN and LR were assessed based on area under the curve (AUC) measures of receiver operator characteristic curves and Brier score.

Results

ANN had a significantly higher AUC (0.892) of post-operative prediction and AUC (0.808) of pre-operative prediction than LR (both P<0.0001). In addition, there was significant higher AUC of post-operative prediction than pre-operative prediction by ANN (p<0.0001). With the highest AUC and the lowest Brier score (0.090), the post-operative prediction by ANN had the highest overall predictive performance.

Conclusion

The post-operative prediction by ANN had the highest overall performance in predicting SSI after free-flap reconstruction in patients receiving surgery for head and neck cancer.

Comparison of Models for the Prediction of Medical Costs of Spinal Fusion in Taiwan Diagnosis-Related Groups by Machine Learning Algorithms

Objectives

The aims of this study were to compare the performance of machine learning methods for the prediction of the medical costs associated with spinal fusion in terms of profit or loss in Taiwan Diagnosis-Related Groups (Tw-DRGs) and to apply these methods to explore the important factors associated with the medical costs of spinal fusion.

Methods

A data set was obtained from a regional hospital in Taoyuan city in Taiwan, which contained data from 2010 to 2013 on patients of Tw-DRG49702 (posterior and other spinal fusion without complications or comorbidities). Naïve-Bayesian, support vector machines, logistic regression, C4.5 decision tree, and random forest methods were employed for prediction using WEKA 3.8.1.

Results

Five hundred thirty-two cases were categorized as belonging to the Tw-DRG49702 group. The mean medical cost was US $4,549.7, and the mean age of the patients was 62.4 years. The mean length of stay was 9.3 days. The length of stay was an important variable in terms of determining medical costs for patients undergoing spinal fusion. The random forest method had the best predictive performance in comparison to the other methods, achieving an accuracy of 84.30%, a sensitivity of 71.4%, a specificity of 92.2%, and an AUC of 0.904.

Conclusions

Our study demonstrated that the random forest model can be employed to predict the medical costs of Tw-DRG49702, and could inform hospital strategy in terms of increasing the financial management efficiency of this operation.

Derivation and validation of different machine-learning models in mortality prediction of trauma in motorcycle riders: a cross-sectional retrospective study in southern Taiwan

OBJECTIVES

This study aimed to build and test the models of machine learning (ML) to predict the mortality of hospitalised motorcycle riders.

SETTING

The study was conducted in a level-1 trauma centre in southern Taiwan.

PARTICIPANTS

Motorcycle riders who were hospitalised between January 2009 and December 2015 were classified into a training set (n=6306) and test set (n=946). Using the demographic information, injury characteristics and laboratory data of patients, logistic regression (LR), support vector machine (SVM) and decision tree (DT) analyses were performed to determine the mortality of individual motorcycle riders, under different conditions, using all samples or reduced samples, as well as all variables or selected features in the algorithm.

PRIMARY AND SECONDARY OUTCOME MEASURES

The predictive performance of the model was evaluated based on accuracy, sensitivity, specificity and geometric mean, and an analysis of the area under the receiver operating characteristic curves of the two different models was carried out.

RESULTS

In the training set, both LR and SVM had a significantly higher area under the receiver operating characteristic curve (AUC) than DT. No significant difference was observed in the AUC of LR and SVM, regardless of whether all samples or reduced samples and whether all variables or selected features were used. In the test set, the performance of the SVM model for all samples with selected features was better than that of all other models, with an accuracy of 98.73%, sensitivity of 86.96%, specificity of 99.02%, geometric mean of 92.79% and AUC of 0.9517, in mortality prediction.

CONCLUSION

ML can provide a feasible level of accuracy in predicting the mortality of motorcycle riders. Integration of the ML model, particularly the SVM algorithm in the trauma system, may help identify high-risk patients and, therefore, guide appropriate interventions by the clinical staff.

Multi-atlas-based segmentation of prostatic urethra from planning CT imaging to quantify dose distribution in prostate cancer radiotherapy

BACKGROUND AND PURPOSE

Segmentation of intra-prostatic urethra for dose assessment from planning CT may help explaining urinary toxicity in prostate cancer radiotherapy. This work sought to: i) propose an automatic method for urethra segmentation in CT, ii) compare it with previously proposed surrogate models and iii) quantify the dose received by the urethra in patients treated with IMRT.

MATERIALS AND METHODS

A weighted multi-atlas-based urethra segmentation method was devised from a training data set of 55 CT scans of patients receiving brachytherapy with visible urinary catheters. Leave-one-out cross validation was performed to quantify the error between the urethra segmentation and the catheter ground truth with two scores: the centerlines distance (CLD) and the percentage of centerline within a certain distance from the catheter (PWR). The segmentation method was then applied to a second test data set of 95 prostate cancer patients having received 78Gy IMRT to quantify dose to the urethra.

RESULTS

Mean CLD was 3.25±1.2mm for the whole urethra and 3.7±1.7mm, 2.52±1.5mm, and 3.01±1.7mm for the top, middle, and bottom thirds, respectively. In average, 53% of the segmented centerlines were within a radius<3.5mm from the centerline ground truth and 83% in a radius<5mm. The proposed method outperformed existing surrogate models. In IMRT, urethra DVH was significantly higher than prostate DVH from V74Gy to V79Gy.

CONCLUSION

A multi-atlas-based segmentation method was proposed enabling assessment of the dose within the prostatic urethra.

Internet-based computer technology on radiotherapy

Recent rapid development of Internet-based computer technologies has made possible many novel applications in radiation dose delivery. However, translational speed of applying these new technologies in radiotherapy could hardly catch up due to the complex commissioning process and quality assurance protocol. Implementing novel Internet-based technology in radiotherapy requires corresponding design of algorithm and infrastructure of the application, set up of related clinical policies, purchase and development of software and hardware, computer programming and debugging, and national to international collaboration. Although such implementation processes are time consuming, some recent computer advancements in the radiation dose delivery are still noticeable. In this review, we will present the background and concept of some recent Internet-based computer technologies such as cloud computing, big data processing and machine learning, followed by their potential applications in radiotherapy, such as treatment planning and dose delivery. We will also discuss the current progress of these applications and their impacts on radiotherapy. We will explore and evaluate the expected benefits and challenges in implementation as well.

Modeling Urinary Dysfunction After External Beam Radiation Therapy of the Prostate Using Bladder Dose-Surface Maps: Evidence of Spatially Variable Response of the Bladder Surface

PURPOSE

We assessed the association of the spatial distribution of dose to the bladder surface, described using dose-surface maps, with the risk of urinary dysfunction.

METHODS AND MATERIALS

The bladder dose-surface maps of 754 participants from the TROG 03.04-RADAR trial were generated from the volumetric data by virtually cutting the bladder at the sagittal slice, intersecting the bladder center-of-mass through to the bladder posterior and projecting the dose information on a 2-dimensional plane. Pixelwise dose comparisons were performed between patients with and without symptoms (dysuria, hematuria, incontinence, and an International Prostate Symptom Score increase of ≥10 [ΔIPSS10]). The results with and without permutation-based multiple-comparison adjustments are reported. The pixelwise multivariate analysis findings (peak-event model for dysuria, hematuria, and ΔIPSS10; event-count model for incontinence), with adjustments for clinical factors, are also reported.

RESULTS

The associations of the spatially specific dose measures to urinary dysfunction were dependent on the presence of specific symptoms. The doses received by the anteroinferior and, to lesser extent, posterosuperior surface of the bladder had the strongest relationship with the incidence of dysuria, hematuria, and ΔIPSS10, both with and without adjustment for clinical factors. For the doses to the posteroinferior region corresponding to the area of the trigone, the only symptom with significance was incontinence.

CONCLUSIONS

A spatially variable response of the bladder surface to the dose was found for symptoms of urinary dysfunction. Limiting the dose extending anteriorly might help reduce the risk of urinary dysfunction.