Development and internal validation of predictive models to assess risk of post-acute care facility discharge in adults undergoing multi-level instrumented fusions for lumbar degenerative pathology and spinal deformity

PURPOSE

To develop a model for factors predictive of Post-Acute Care Facility (PACF) discharge in adult patients undergoing elective multi-level (≥ 3 segments) lumbar/thoracolumbar spinal instrumented fusions.

METHODS

The State Inpatient Databases acquired from the Healthcare Cost and Utilization Project from 2005 to 2013 were queried for adult patients who underwent elective multi-level thoracolumbar fusions for spinal deformity. Outcome variables were classified as discharge to home or PACF. Predictive variables included demographic, pre-operative, and operative factors. Univariate and multivariate logistic regression analyses informed development of a logistic regression-based predictive model using seven selected variables. Performance metrics included area under the curve (AUC), sensitivity, and specificity.

RESULTS

Included for analysis were 8866 patients. The logistic model including significant variables from multivariate analysis yielded an AUC of 0.75. Stepwise logistic regression was used to simplify the model and assess number of variables needed to reach peak AUC, which included seven selected predictors (insurance, interspaces fused, gender, age, surgical region, CCI, and revision surgery) and had an AUC of 0.74. Model cut-off for predictive PACF discharge was 0.41, yielding a sensitivity of 75% and specificity of 59%.

CONCLUSIONS

The seven variables associated significantly with PACF discharge (age > 60, female gender, non-private insurance, primary operations, instrumented fusion involving 8+ interspaces, thoracolumbar region, and higher CCI scores) may aid in identification of adults at risk for discharge to a PACF following elective multi-level lumbar/thoracolumbar spinal fusions for spinal deformity. This may in turn inform discharge planning and expectation management.

Multivariable regression: understanding one of medicine's most fundamental statistical tools

Multivariable regression is a fundamental tool that drives observational research in orthopaedic surgery. However, regression analyses are not always implemented correctly. This study presents a basic overview of regression analyses and reviews frequent points of confusion. Topics include linear, logistic, and time-to-event regressions, causal inference, confounders, overfitting, missing data, multicollinearity, interactions, and key differences between multivariable versus multivariate regression. The goal is to provide clarity regarding the use and interpretation of multivariable analyses for those attempting to increase their statistical literacy in orthopaedic research.

Diagnostic nomogram for closed-loop small bowel obstruction requiring emergency surgery

PURPOSE

This study aimed to build a diagnostic model of closed-loop small bowel obstruction (CL-SBO) using clinical information, blood test results, and computed tomography (CT) findings.

METHODS

All patients who were diagnosed with small bowel obstruction (SBO) and underwent surgery between January 1, 2018, and October 31, 2021, in the affiliated hospital of Qingdao university were reviewed, and their relevant preoperative information was collected. All variables were selected using univariate analysis and backward stepwise regression to build a diagnostic nomogram model. K-fold cross-validation and bootstrap resampling techniques were used for internal validation, and data from Qingdao Central Hospital were used for external validation. We also evaluated the diagnostic performance of each CT finding and performed subgroup analysis according to bowel ischemia in the closed-loop small bowel obstruction (CL-SBO) group.

RESULTS

A total of 219 patients (95 in the CL-SBO group and 124 in the open-loop small bowel obstruction [OL-SBO] group) were included in our research. D-dimers (median 1085 vs. 690, P = 0.019), tenderness (77.9% vs. 59.7%, P = 0.004), more than one beak sign (65.3% vs. 30.6%, P < 0.001), radial distribution (18.9% vs. 6.5%, P = 0.005), whirl sign (35.8% vs. 8.9%, P < 0.001), and ascites (71.6% vs. 53.2%, P = 0.006) were selected as the predictive variables of the nomogram. This model's Harrell's C statistic was 0.786 (95% confidence interval (CI), 0.724-0.848), and the Brier score was 0.182. The Harrell's C statistic of external validation was 0.784 (95%CI, 0.664-0.905); the Brier score was 0.190. Regarding the CT findings, radial distribution, U/C-shaped loop, and whirl sign had high specificity (93.5%, 96.0%, and 91.1%, respectively), but low sensitivity (18.9%, 8.4%, and 35.8%, respectively). D-dimer levels and tenderness were also associated with bowel ischemia.

CONCLUSION

The nomogram accurately predicted CL-SBO in patients with SBO, and surgery should be considered when patients have a high risk for developing CL-SBO.

Predicting treatment outcome in depression: an introduction into current concepts and challenges

Improving response and remission rates in major depressive disorder (MDD) remains an important challenge. Matching patients to the treatment they will most likely respond to should be the ultimate goal. Even though numerous studies have investigated patient-specific indicators of treatment efficacy, no (bio)markers or empirical tests for use in clinical practice have resulted as of now. Therefore, clinical decisions regarding the treatment of MDD still have to be made on the basis of questionnaire- or interview-based assessments and general guidelines without the support of a (laboratory) test. We conducted a narrative review of current approaches to characterize and predict outcome to pharmacological treatments in MDD. We particularly focused on findings from newer computational studies using machine learning and on the resulting implementation into clinical decision support systems. The main issues seem to rest upon the unavailability of robust predictive variables and the lacking application of empirical findings and predictive models in clinical practice. We outline several challenges that need to be tackled on different stages of the translational process, from current concepts and definitions to generalizable prediction models and their successful implementation into digital support systems. By bridging the addressed gaps in translational psychiatric research, advances in data quantity and new technologies may enable the next steps toward precision psychiatry.

Predicting water quality from geospatial lake, catchment, and buffer zone characteristics in temperate lowland lakes

Lakes provide essential ecosystem services and strongly influence landscape nutrient and carbon cycling. Therefore, monitoring water quality is essential for the management of element transport, biodiversity, and public goods in lakes. We investigated the ability of machine learning models to predict eight important water quality variables (alkalinity, pH, total phosphorus, total nitrogen, chlorophyll a, Secchi depth, color, and pCO₂) using monitoring data from 924 to 1054 lakes. The geospatial predictor variables comprise a wide range of potential drivers at the lake, buffer zone, and catchment level. We compared the performance of nine predictive models of varying complexity for each of the eight water quality variables. The best models (Random Forest and Support Vector Machine in six and two cases, respectively) generally performed well on the test set (R² = 0.28-0.60). Models were then used to predict water quality for all 180,377 mapped Danish lakes. Additionally, we trained models to predict each water quality variable by using the predictions we had generated for the remaining seven variables. This improved model performance (R² = 0.45-0.78). Overall, the uncovered relationships were in line with the findings of previous studies, e.g., total nitrogen was positively related to catchment agriculture and chlorophyll a, Secchi depth, and alkalinity were influenced by soil type and landscape history. Remarkably, buffer zone geomorphology (curvature, ruggedness, and elevation) had a strong influence on nutrients, chlorophyll a, and Secchi depth, e.g., curvature was positively related to nutrients and chlorophyll a and negatively to Secchi depth. Lake area was a strong predictor of multiple variables, especially its relationship with pH (positive), pCO₂ (negative), and color (negative). Our analysis shows that the combination of machine learning methods and geospatial data can be used to predict lake water quality and improve national upscaling of predictions related to nutrient and carbon cycling.

Task-based co-activation patterns reliably predict resting state canonical network engagement during development

Neurodevelopmental research has traditionally focused on development of individual structures, yet multiple lines of evidence indicate parallel development of large-scale systems, including canonical neural networks (i.e., default mode, frontoparietal). However, the relationship between region- vs. network-level development remains poorly understood. The current study tests the ability of a recently developed multi-task coactivation matrix approach to predict canonical resting state network engagement at baseline and at two-year follow-up in a large and cohort of young adolescents. Pre-processed tabulated neuroimaging data were obtained from the Adolescent Brain and Cognitive Development (ABCD) study, assessing youth at baseline (N = 6073, age = 10.0 ± 0.6 years, 3056 female) and at two-year follow-up (N = 3539, age = 11.9 ± 0.6 years, 1726 female). Individual multi-task co-activation matrices were constructed from the beta weights of task contrasts from the stop signal task, the monetary incentive delay task, and emotional N-back task. Activation-based predictive modeling, a cross-validated machine learning approach, was adopted to predict resting-state canonical network engagement from multi-task co-activation matrices at baseline. Note that the tabulated data used different parcellations of the task fMRI data ("ASEG" and Desikan) and the resting-state fMRI data (Gordon). Despite this, the model successfully predicted connectivity within the default mode network (DMN, rho = 0.179 ± 0.002, p < 0.001) across participants and identified a subset of co-activations within parietal and occipital macroscale brain regions as key contributors to model performance, suggesting an underlying common brain functional architecture across cognitive domains. Notably, predictive features for resting-state connectivity within the DMN identified at baseline also predicted DMN connectivity at two-year follow-up (rho = 0.258). These results indicate that multi-task co-activation matrices are functionally meaningful and can be used to predict resting-state connectivity. Interestingly, given that predictive features within the co-activation matrices identified at baseline can be extended to predictions at a future time point, our results suggest that task-based neural features and models are valid predictors of resting state network level connectivity across the course of development. Future work is encouraged to verify these findings with more consistent parcellations between task-based and resting-state fMRI, and with longer developmental trajectories.

Socioeconomic resources are associated with distributed alterations of the brain's intrinsic functional architecture in youth

Little is known about how exposure to limited socioeconomic resources (SER) in childhood gets "under the skin" to shape brain development, especially using rigorous whole-brain multivariate methods in large, adequately powered samples. The present study examined resting state functional connectivity patterns from 5821 youth in the Adolescent Brain Cognitive Development (ABCD) study, employing multivariate methods across three levels: whole-brain, network-wise, and connection-wise. Across all three levels, SER was associated with widespread alterations across the connectome. However, critically, we found that parental education was the primary driver of neural associations with SER. These parental education associations with the developing connectome exhibited notable concentrations in somatosensory and subcortical regions, and they were partially accounted for by home enrichment activities, child's cognitive abilities, and child's grades, indicating interwoven links between parental education, child stimulation, and child cognitive performance. These results add a new data-driven, multivariate perspective on links between household SER and the child's developing functional connectome.

Myosteatosis predicting risk of transition to severe COVID-19 infection

BACKGROUND

About 10-20% of patients with Coronavirus disease 2019 (COVID-19) infection progressed to severe illness within a week or so after initially diagnosed as mild infection. Identification of this subgroup of patients was crucial for early aggressive intervention to improve survival. The purpose of this study was to evaluate whether computer tomography (CT) - derived measurements of body composition such as myosteatosis indicating fat deposition inside the muscles could be used to predict the risk of transition to severe illness in patients with initial diagnosis of mild COVID-19 infection.

METHODS

Patients with laboratory-confirmed COVID-19 infection presenting initially as having the mild common-subtype illness were retrospectively recruited between January 21, 2020 and February 19, 2020. CT-derived body composition measurements were obtained from the initial chest CT images at the level of the twelfth thoracic vertebra (T12) and were used to build models to predict the risk of transition. A myosteatosis nomogram was constructed using multivariate logistic regression incorporating both clinical variables and myosteatosis measurements. The performance of the prediction models was assessed by receiver operating characteristic (ROC) curve including the area under the curve (AUC). The performance of the nomogram was evaluated by discrimination, calibration curve, and decision curve.

RESULTS

A total of 234 patients were included in this study. Thirty-one of the enrolled patients transitioned to severe illness. Myosteatosis measurements including SM-RA (skeletal muscle radiation attenuation) and SMFI (skeletal muscle fat index) score fitted with SMFI, age and gender, were significantly associated with risk of transition for both the training and validation cohorts (P < 0.01). The nomogram combining the SM-RA, SMFI score and clinical model improved prediction for the transition risk with an AUC of 0.85 [95% CI, 0.75 to 0.95] for the training cohort and 0.84 [95% CI, 0.71 to 0.97] for the validation cohort, as compared to the nomogram of the clinical model with AUC of 0.75 and 0.74 for the training and validation cohorts respectively. Favorable clinical utility was observed using decision curve analysis.

CONCLUSION

We found CT-derived measurements of thoracic myosteatosis to be associated with higher risk of transition to severe illness in patients affected by COVID-19 who presented initially as having the mild common-subtype infection. Our study showed the relevance of skeletal muscle examination in the overall assessment of disease progression and prognosis of patients with COVID-19 infection.

Calibrated Regression Models Based on the Risk of Clinical Nodal Metastasis Should be Used as Decision Aids for Prostate Cancer Staging to Reduce Unnecessary Imaging

INTRODUCTION

Radionuclide imaging will change the role of computed tomography and magnetic resonance imaging (CT/MRI) for prostate cancer (CaP) staging. Current guidelines recommend abdominopelvic imaging for new cases of CaP categorized as unfavorable intermediate risk (UIR) or higher. We assessed the performance characteristics of CT/MRI based on the National Comprehensive Cancer Network (NCCN) guidelines and developed a model that predicts cN1 disease using conventional imaging.

PATIENTS AND METHODS

We selected patients in the National Cancer Database diagnosed with CaP from 2010 to 2016 with available age, prostate specific antigen, clinical locoregional staging, biopsy Gleason grading, and core information. Multivariate logistic regression (MLR) was used on a undersampled training dataset using cN1 as the outcome. Performance characteristics were compared to those of the three most recent versions of the NCCN guidelines.

RESULTS

A total of 443,640 men were included, and 2.5% had cN1 disease. Using CT/MRI only, the current NCCN guidelines have a sensitivity of 99%, and the number needed to image (NNI) is 24. At the same sensitivity, the cN1 risk was 1.6% using the MLR. The NNI for UIR alone is 341. Using the MLR model and a threshold of 10%, the PPV is 10.3% and 64% of CTs/MRIs could be saved at a cost of missing 6% of cN1 patients (or 0.15% of all patients).

CONCLUSION

The NCCN guidelines are sensitive for detecting cN1 with CT/MRI, however, the number needed to image is 24. Obtaining CT/MRI for nodal staging when patients have a cN1 risk of 10% would reduce total imaging while still remaining sensitive. As novel PET tracers becomes increasingly used for initial CaP staging, well calibrated prediction models trained on the outcome of interest should be developed as decision aids for obtaining imaging.

Continuous direct compression: Development of an empirical predictive model and challenges regarding PAT implementation

In this study, an empirical predictive model was developed based on the quantitative relationships between blend properties, critical quality attributes (CQA) and critical process parameters (CPP) related to blending and tableting. The blend uniformity and API concentration in the tablets were used to elucidate challenges related to the processability as well as the implementation of PAT tools. Thirty divergent ternary blends were evaluated on a continuous direct compression line (ConsiGma™ CDC-50). The trials showed a significant impact of the impeller configuration and impeller speed on the blending performance, whereas a limited impact of blend properties was observed. In contrast, blend properties played a significant role during compression, where changes in blend composition significantly altered the tablet quality. The observed correlations allowed to develop an empirical predictive model for the selection of process configurations based on the blend properties, reducing the number of trial runs needed to optimize a process and thus reducing development time and costs of new drug products. Furthermore, the trials elucidated several challenges related to blend properties that had a significant impact on PAT implementation and performance of the CDC-platform, highlighting the importance of further process development and optimization in order to solve the remaining challenges.

Prognostic value of exercise capacity in incident diabetes: a country with high prevalence of diabetes

BACKGROUND

Diabetes Mellitus (DM) is a fast-growing health problem that imposes an enormous economic burden. Several studies demonstrated the association between physical inactivity and predicting the incidence of diabetes. However, these prediction models have limited validation locally. Therefore, we aim to explore the predictive value of exercise capacity in the incidence of diabetes within a high diabetes prevalence population.

METHODOLOGY

A retrospective cohort study including consecutive patients free of diabetes who underwent clinically indicated treadmill stress testing. Diabetic patients at baseline or patients younger than 18 years of age were excluded. Incident diabetes was defined as an established clinical diagnosis post-exercise testing date. The predictive value of exercise capacity was examined using Harrell's c-index, net reclassification index (NRI), and integrated discrimination index (IDI).

RESULTS

A total of 8,722 participants (mean age 46 ± 12 years, 66.3% were men) were free of diabetes at baseline. Over a median follow-up period of 5.24 (2.17-8.78) years, there were 2,280 (≈ 26%) new cases of diabetes. In a multivariate model adjusted for conventional risk factors, we found a 12% reduction in the risk of incident diabetes for each METs achieved (HR, 0.9; 95% CI, 0.88-0.92; P < 0.001). Using Cox regression, exercise capacity improved the prediction ability beyond the conventional risk factors (AUC = 0.62 to 0.66 and c-index = 0.62 to 0.68).

CONCLUSION

Exercise capacity improved the overall predictability of diabetes. Patients with reduced exercise capacity are at high risk for developing incidence diabetes. Improvement of both physical activity and functional capacity represents a preventive measure for the general population.

Assessing the long-term evolution of mine water quality in abandoned underground mine workings using first-flush based models

Coal mining activities can leave an extensive network of abandoned underground workings that gradually flood after operations cease. This rising mine water can eventually lead to uncontrolled releases of harmful acid mine drainage (AMD) to the environment. Treatment plants are used to extract and treat the mine water to maintain its elevation below suspected discharge points. Accurate predictions of long-term water quality, and treatment plant operations, are highly challenging due to the complexity and volume of the underground workings. As numerical models require considerable effort to effectively implement, empirical models that are based on the 'first-flush' phenomenon, where mine water concentrations peak shortly after flooding and then exponentially decline, may provide suitable long-term predictions. The objective of this study was to assess the robustness of first-flush based models for describing mine water behavior at large, complex mine pools in the Sydney Coalfield (Nova Scotia, Canada). Numerous mine pools across the coalfield flooded at various times over 100+ years, with extensive mine water quality data available in various pools of different ages. The historical evolution of mine water quality demonstrated first-flush behavior across key AMD indicator parameters (acidity, sulfate, iron), concentration ranges, and mine pool depths. Two 'newer' mine pools, which only flooded in the past 10-15 years, rely on an active treatment plant to manage mine water levels below environmental discharge points. Influent water quality from each mine pool was sampled bi-weekly between 2011 and 2022, and first-flush models were then applied to predict the future quality of mine water entering the treatment plant over the long-term. Knowledge on long-term influent quality can help to optimize treatment plant requirements and related expenses.

Development and validation of machine learning-based predictive model for clinical outcome of decompression surgery for lumbar spinal canal stenosis

BACKGROUND CONTEXT

Although the results of decompression surgery for lumbar spinal canal stenosis (LSS) are favorable, it is still difficult to predict the postoperative health-related quality of life of patients before surgery.

PURPOSE

The purpose of this study was to develop and validate a machine learning model to predict the postoperative outcome of decompression surgery for patients with LSS.

STUDY DESIGN/SETTING

A multicentered retrospective study.

PATIENT SAMPLE

A total of 848 patients who underwent decompression surgery for LSS at an academic hospital, tertiary center, and private hospital were included (age 71±9 years, 68% male, 91% LSS, level treated 1.8±0.8, operation time 69±37 minutes, blood loss 48±113 mL, and length of hospital stay 12±5 days).

OUTCOME MEASURES

Baseline and 2 years postoperative health-related quality of life.

METHODS

The subjects were randomly assigned in a 7:3 ratio to a model building cohort and a testing cohort to test the models' accuracy. Twelve predictive algorithms using 68 preoperative factors were used to predict each domain of the Japanese Orthopedic Association Back Pain Evaluation Questionnaire and visual analog scale scores at 2 years postoperatively. The final predictive values were generated using an ensemble of the top five algorithms in prediction accuracy.

RESULTS

The correlation coefficients of the top algorithms for each domain established using the preoperative factors were excellent (correlation coefficient: 0.95-0.97 [relative error: 0.06-0.14]). The performance evaluation of each Japanese Orthopedic Association Back Pain Evaluation Questionnaire domain and visual analog scale score by the ensemble of the top five algorithms in the testing cohort was favorable (mean absolute error [MAE] 8.9-17.4, median difference [MD] 8.1-15.6/100 points), with the highest accuracy for mental status (MAE 8.9, MD 8.1) and the lowest for buttock and leg numbness (MAE 1.7, MD 1.6/10 points). A strong linear correlation was observed between the predicted and measured values (linear correlation 0.82-0.89), while 4% to 6% of the subjects had predicted values of greater than±3 standard deviations of the MAE.

CONCLUSIONS

We successfully developed a machine learning model to predict the postoperative outcomes of decompression surgery for patients with LSS using patient data from three different institutions in three different settings. Thorough analyses for the subjects with deviations from the actual measured values may further improve the predictive probability of this model.

Minimal important change was on the lower spectrum of previous estimates and responsiveness was sufficient for core outcomes in chronic low back pain

OBJECTIVES

The objective of this study was to estimate the minimal important change (MIC) and responsiveness of core patient reported outcome measures for chronic low back pain (LBP) and Modic changes.

STUDY DESIGN AND SETTING

In the Antibiotics in Modic changes (AIM) trial we measured disability (RMDQ, ODI), LBP intensity (NRS) and health-related quality of life (EQ5D) electronically at baseline, three- and 12-month follow-up. MICs were estimated using Receiver Operating Curve (ROC) curve and Predictive modeling analyses against the global perceived effect. Credibility of the estimates was assessed by a standardized set of criteria. Responsiveness was assessed by a construct and criterion approach according to COSMIN guidelines.

RESULTS

The MIC estimates of RMDQ, ODI and NRS scores varied between a 15-40% reduction, depending on including "slightly improved" in the definition of MIC or not. The MIC estimates for EQ5D were lower. The credibility of the estimates was moderate. For responsiveness, five out of six hypotheses were confirmed and AUC was >0.7 for all PROMs.

CONCLUSION

When evaluated in a clinical trial of patients with chronic LBP and Modic changes, MIC thresholds for all PROMs were on the lower spectrum of previous estimates, varying depending on the definition of MIC. Responsiveness was sufficient.

Brachytherapy outcome modeling in cervical cancer patients: A predictive machine learning study on patient-specific clinical, physical and dosimetric parameters

PURPOSE

To predict clinical response in locally advanced cervical cancer (LACC) patients by a combination of measures, including clinical and brachytherapy parameters and several machine learning (ML) approaches.

METHODS

Brachytherapy features such as insertion approaches, source metrics, dosimetric, and clinical measures were used for modeling. Four different ML approaches, including LASSO, Ridge, support vector machine (SVM), and Random Forest (RF), were applied to extracted measures for model development alone or in combination. Model performance was evaluated using the area under the curve (AUC) of receiver operating characteristics curve, sensitivity, specificity, and accuracy. Our results were compared with a reference model developed by simple logistic regression applied to three distinct clinical features identified by previous papers.

RESULTS

One hundred eleven LACC patients were included. Nine data sets were obtained based on the features, and 36 predictive models were built. In terms of AUC, the model developed using RF applied to dosimetric, physical, and total BT sessions features were found as the most predictive [AUC; 0.82 (0.95 confidence interval (CI); 0.79 -0.93), sensitivity; 0.79, specificity; 0.76, and accuracy; 0.77]. The AUC (0.95 CI), sensitivity, specificity, and accuracy for the reference model were found as 0.56 (0.52 ...0.68), 0.51, 0.51, and 0.48, respectively. Most RF models had significantly better performance than the reference model (Bonferroni corrected p-value < 0.0014).

CONCLUSION

Brachytherapy response can be predicted using dosimetric and physical parameters extracted from treatment parameters. Machine learning algorithms, including Random Forest, could play a critical role in such predictive modeling.

Exploiting Genomic Features to Improve the Prediction of Transcription Factor-Binding Sites in Plants

The identification of transcription factor (TF) target genes is central in biology. A popular approach is based on the location by pattern matching of potential cis-regulatory elements (CREs). During the last few years, tools integrating next-generation sequencing data have been developed to improve the performance of pattern matching. However, such tools have not yet been comprehensively evaluated in plants. Hence, we developed a new streamlined method aiming at predicting CREs and target genes of plant TFs in specific organs or conditions. Our approach implements a supervised machine learning strategy, which allows decision rule models to be learnt using TF ChIP-chip/seq experimental data. Different layers of genomic features were integrated in predictive models: the position on the gene, the DNA sequence conservation, the chromatin state and various CRE footprints. Among the tested features, the chromatin features were crucial for improving the accuracy of the method. Furthermore, we evaluated the transferability of predictive models across TFs, organs and species. Finally, we validated our method by correctly inferring the target genes of key TFs controlling metabolite biosynthesis at the organ level in Arabidopsis. We developed a tool-Wimtrap-to reproduce our approach in plant species and conditions/organs for which ChIP-chip/seq data are available. Wimtrap is a user-friendly R package that supports an R Shiny web interface and is provided with pre-built models that can be used to quickly get predictions of CREs and TF gene targets in different organs or conditions in Arabidopsis thaliana, Solanum lycopersicum, Oryza sativa and Zea mays.

A novel image-based machine learning model with superior accuracy and predictability for knee arthroplasty loosening detection and clinical decision making

Background

Loosening is the leading cause of total knee arthroplasty (TKA) revision. This is a heavy burden toward the healthcare system owing to the difficulty in diagnosis and complications occurring from the delay management. Based on automatic analytical model building, machine learning, may potentially help to automatically recognize the risk of loosening based on radiographs alone. The aim of this study was to build an image-based machine-learning model for detecting TKA loosening.

Methods

Image-based machine-learning model was developed based on ImageNet, Xception model and a TKA patient X-ray image dataset. Based on a dataset with TKA patient clinical parameters, another system was then created for developing the clinical-information-based machine learning model with random forest classifier. In addition, the Xception Model was pre-trained on the ImageNet database with python and TensorFlow deep learning library for the prediction of loosening. Class activation maps were also used to interpret the prediction decision made by model. Two senior orthopaedic specialists were invited to assess loosening from X-ray images for 3 attempts in setting up comparison benchmark.

Result

In the image-based machine learning loosening model, the precision rate and recall rate were 0.92 and 0.96, respectively. While for the accuracy rate, 96.3% for visualization classification was observed. However, the addition of clinical-information-based model, with precision rate of 0.71 and recall rate of 0.20, did not further showed improvement on the accuracy. Moreover, as class activation maps showed corresponding signals over bone-implant interface that is loosened radiographically, this confirms that the current model utilized a similar image recognition pattern as that of inspection by clinical specialists.

Conclusion

The image-based machine learning model developed demonstrated high accuracy and predictability of knee arthroplasty loosening. And the class activation heatmap matched well with the radiographic features used clinically to detect loosening, which highlighting its potential role in assisting clinicians in their daily practice. However, addition of clinical-information-based machine-learning model did not offer further improvement in detection. As far as we know, this is the first report of pure image-based machine learning model with high detection accuracy. Importantly, this is also the first model to show relevant class activation heatmap corresponding to loosening location.

Translational potential

The finding in this study indicated image-based machine learning model can detect knee arthroplasty loosening with high accuracy and predictability, which the class activation heatmap can potentially assist surgeons to identify the sites of loosening.

Predicting future cognitive decline from non-brain and multimodal brain imaging data in healthy and pathological aging

Previous literature has focused on predicting a diagnostic label from structural brain imaging. Since subtle changes in the brain precede a cognitive decline in healthy and pathological aging, our study predicts future decline as a continuous trajectory instead. Here, we tested whether baseline multimodal neuroimaging data improve the prediction of future cognitive decline in healthy and pathological aging. Nonbrain data (demographics, clinical, and neuropsychological scores), structural MRI, and functional connectivity data from OASIS-3 (N = 662; age = 46-96 years) were entered into cross-validated multitarget random forest models to predict future cognitive decline (measured by CDR and MMSE), on average 5.8 years into the future. The analysis was preregistered, and all analysis code is publicly available. Combining non-brain with structural data improved the continuous prediction of future cognitive decline (best test-set performance: R2 = 0.42). Cognitive performance, daily functioning, and subcortical volume drove the performance of our model. Including functional connectivity did not improve predictive accuracy. In the future, the prognosis of age-related cognitive decline may enable earlier and more effective individualized cognitive, pharmacological, and behavioral interventions.

A Survey on Methods for Predicting Polyadenylation Sites from DNA Sequences, Bulk RNA-seq, and Single-cell RNA-seq

Alternative polyadenylation (APA) plays important roles in modulating mRNA stability, translation, and subcellular localization, and contributes extensively to shaping eukaryotic transcriptome complexity and proteome diversity. Identification of poly(A) sites (pAs) on a genome-wide scale is a critical step toward understanding the underlying mechanism of APA-mediated gene regulation. A number of established computational tools have been proposed to predict pAs from diverse genomic data. Here we provided an exhaustive overview of computational approaches for predicting pAs from DNA sequences, bulk RNA sequencing (RNA-seq) data, and single-cell RNA sequencing (scRNA-seq) data. Particularly, we examined several representative tools using bulk RNA-seq and scRNA-seq data from peripheral blood mononuclear cells and put forward operable suggestions on how to assess the reliability of pAs predicted by different tools. We also proposed practical guidelines on choosing appropriate methods applicable to diverse scenarios. Moreover, we discussed in depth the challenges in improving the performance of pA prediction and benchmarking different methods. Additionally, we highlighted outstanding challenges and opportunities using new machine learning and integrative multi-omics techniques, and provided our perspective on how computational methodologies might evolve in the future for non-3' untranslated region, tissue-specific, cross-species, and single-cell pA prediction.

Machine learning for prediction of bronchopulmonary dysplasia-free survival among very preterm infants

Background

Bronchopulmonary dysplasia (BPD) is one of the most common and serious sequelae of prematurity. Prompt diagnosis using prediction tools is crucial for early intervention and prevention of further adverse effects. This study aims to develop a BPD-free survival prediction tool based on the concept of the developmental origin of BPD with machine learning.

Methods

Datasets comprising perinatal factors and early postnatal respiratory support were used for initial model development, followed by combining the two models into a final ensemble model using logistic regression. Simulation of clinical scenarios was performed.

Results

Data from 689 infants were included in the study. We randomly selected data from 80% of infants for model development and used the remaining 20% for validation. The performance of the final model was assessed by receiver operating characteristics which showed 0.921 (95% CI: 0.899–0.943) and 0.899 (95% CI: 0.848–0.949) for the training and the validation datasets, respectively. Simulation data suggests that extubating to CPAP is superior to NIPPV in BPD-free survival. Additionally, successful extubation may be defined as no reintubation for 9 days following initial extubation.

Conclusions

Machine learning-based BPD prediction based on perinatal features and respiratory data may have clinical applicability to promote early targeted intervention in high-risk infants.

Prognostic Models in Severe Traumatic Brain Injury: A Systematic Review and Meta-analysis

This review aimed to analyze the results of investigations that performed external validation or that compared prognostic models to identify the models and their variations that showed the best performance in predicting mortality, survival, and unfavorable outcome after severe traumatic brain injury. Pubmed, Embase, Scopus, Web of Science, Cumulative Index to Nursing and Allied Health Literature, Google Scholar, TROVE, and Open Grey databases were searched. A total of 1616 studies were identified and screened, and 15 studies were subsequently included for analysis after applying the selection criteria. The Corticosteroid Randomization After Significant Head Injury (CRASH) and International Mission for Prognosis and Analysis of Clinical Trials in Traumatic Brain Injury (IMPACT) models were the most externally validated among studies of severe traumatic brain injury. The results of the review showed that most publications encountered an area under the curve ≥ 0.70. The area under the curve meta-analysis showed similarity between the CRASH and IMPACT models and their variations for predicting mortality and unfavorable outcomes. Calibration results showed that the variations of CRASH and IMPACT models demonstrated adequate calibration in most studies for both outcomes, but without a clear indication of uncertainties in the evaluations of these models. Based on the results of this meta-analysis, the choice of prognostic models for clinical application may depend on the availability of predictors, characteristics of the population, and trauma care services.

High-Risk Drinkers Engage Distinct Stress-Predictive Brain Networks

BACKGROUND

Excessive alcohol intake is a major public health problem and can be triggered by stress. Heavy drinking in patients with alcohol use disorder also alters neural, physiological, and emotional stress responses. However, it is unclear whether adaptations in stress-predictive brain networks can be an early marker of risky drinking behavior.

METHODS

Risky social drinkers (regular bingers; n = 53) and light drinker control subjects (n = 51) aged 18 to 53 years completed a functional magnetic resonance imaging-based sustained stress protocol with repeated measures of subjective stress state, during which whole-brain functional connectivity was computed. This was followed by prospective daily ecological momentary assessment for 30 days. We used brain computational predictive modeling with cross-validation to identify unique brain connectivity predictors of stress in risky drinkers and determine the prospective utility of stress-brain networks for subsequent loss of control over drinking.

RESULTS

Risky drinkers had anatomically and functionally distinct stress-predictive brain networks (showing stronger predictions from visual and motor networks) compared with light drinkers (default mode and frontoparietal networks). Stress-predictive brain networks defined for risky drinkers selectively predicted future real-world stress levels for risky drinkers and successfully predicted prospective future real-world loss of control over drinking across all participants.

CONCLUSIONS

These results indicate adaptations in computationally derived stress-related brain circuitry among high-risk drinkers, suggesting potential targets for early preventive intervention and revealing the malleability of the neural processes that govern stress responses.

Artificial intelligence in predicting early-onset adjacent segment degeneration following anterior cervical discectomy and fusion

PURPOSE

Anterior cervical discectomy and fusion (ACDF) is a common surgical treatment for degenerative disease in the cervical spine. However, resultant biomechanical alterations may predispose to early-onset adjacent segment degeneration (EO-ASD), which may become symptomatic and require reoperation. This study aimed to develop and validate a machine learning (ML) model to predict EO-ASD following ACDF.

METHODS

Retrospective review of prospectively collected data of patients undergoing ACDF at a quaternary referral medical center was performed. Patients > 18 years of age with > 6 months of follow-up and complete pre- and postoperative X-ray and MRI imaging were included. An ML-based algorithm was developed to predict EO-ASD based on preoperative demographic, clinical, and radiographic parameters, and model performance was evaluated according to discrimination and overall performance.

RESULTS

In total, 366 ACDF patients were included (50.8% male, mean age 51.4 ± 11.1 years). Over 18.7 ± 20.9 months of follow-up, 97 (26.5%) patients developed EO-ASD. The model demonstrated good discrimination and overall performance according to precision (EO-ASD: 0.70, non-ASD: 0.88), recall (EO-ASD: 0.73, non-ASD: 0.87), accuracy (0.82), F1-score (0.79), Brier score (0.203), and AUC (0.794), with C4/C5 posterior disc bulge, C4/C5 anterior disc bulge, C6 posterior superior osteophyte, presence of osteophytes, and C6/C7 anterior disc bulge identified as the most important predictive features.

CONCLUSIONS

Through an ML approach, the model identified risk factors and predicted development of EO-ASD following ACDF with good discrimination and overall performance. By addressing the shortcomings of traditional statistics, ML techniques can support discovery, clinical decision-making, and precision-based spine care.

Aiding the prescriber: developing a machine learning approach to personalized risk modeling for chronic opioid therapy amongst US Army soldiers

The opioid epidemic is a major policy concern. The widespread availability of opioids, which is fueled by physician prescribing patterns, medication diversion, and the interaction with potential illicit opioid use, has been implicated as proximal cause for subsequent opioid dependence and mortality. Risk indicators related to chronic opioid therapy (COT) at the point of care may influence physicians' prescribing decisions, potentially reducing rates of dependency and abuse. In this paper, we investigate the performance of machine learning algorithms for predicting the risk of COT. Using data on over 12 million observations of active duty US Army soldiers, we apply machine learning models to predict the risk of COT in the initial months of prescription. We use the area under the curve (AUC) as an overall measure of model performance, and we focus on the positive predictive value (PPV), which reflects the models' ability to accurately target military members for intervention. Of the many models tested, AUC ranges between 0.83 and 0.87. When we focus on the top 1% of members at highest risk, we observe a PPV value of 8.4% and 20.3% for months 1 and 3, respectively. We further investigate the performance of sparse models that can be implemented in sparse data environments. We find that when the goal is to identify patients at the highest risk of chronic use, these sparse linear models achieve a performance similar to models trained on hundreds of variables. Our predictive models exhibit high accuracy and can alert prescribers to the risk of COT for the highest risk patients. Optimized sparse models identify a parsimonious set of factors to predict COT: initial supply of opioids, the supply of opioids in the month being studied, and the number of prescriptions for psychotropic medications. Future research should investigate the possible effects of these tools on prescriber behavior (e.g., the benefit of clinician nudging at the point of care in outpatient settings).

Proteomic analysis of cardiometabolic biomarkers and predictive modeling of severe outcomes in patients hospitalized with COVID-19

BACKGROUND

The high heterogeneity in the symptoms and severity of COVID-19 makes it challenging to identify high-risk patients early in the disease. Cardiometabolic comorbidities have shown strong associations with COVID-19 severity in epidemiologic studies. Cardiometabolic protein biomarkers, therefore, may provide predictive insight regarding which patients are most susceptible to severe illness from COVID-19.

METHODS

In plasma samples collected from 343 patients hospitalized with COVID-19 during the first wave of the pandemic, we measured 92 circulating protein biomarkers previously implicated in cardiometabolic disease. We performed proteomic analysis and developed predictive models for severe outcomes. We then used these models to predict the outcomes of out-of-sample patients hospitalized with COVID-19 later in the surge (N = 194).

RESULTS

We identified a set of seven protein biomarkers predictive of admission to the intensive care unit and/or death (ICU/death) within 28 days of presentation to care. Two of the biomarkers, ADAMTS13 and VEGFD, were associated with a lower risk of ICU/death. The remaining biomarkers, ACE2, IL-1RA, IL6, KIM1, and CTSL1, were associated with higher risk. When used to predict the outcomes of the future, out-of-sample patients, the predictive models built with these protein biomarkers outperformed all models built from standard clinical data, including known COVID-19 risk factors.

CONCLUSIONS

These findings suggest that proteomic profiling can inform the early clinical impression of a patient's likelihood of developing severe COVID-19 outcomes and, ultimately, accelerate the recognition and treatment of high-risk patients.