A machine learning model for predicting serum neutralizing activity against Omicron SARS-CoV-2 BA.2 and BA.4/5 sublineages in the general population

Supervised machine learning (ML) methods have been used to predict antibody responses elicited by COVID-19 vaccines in a variety of clinical settings. Here, we explored the reliability of a ML approach to predict the presence of detectable neutralizing antibody responses (NtAb) against Omicron BA.2 and BA.4/5 sublineages in the general population. Anti-SARS-CoV-2 receptor-binding domain (RBD) total antibodies were measured by the Elecsys® Anti-SARS-CoV-2 S assay (Roche Diagnostics) in all participants. NtAbs against Omicron BA.2 and BA4/5 were measured using a SARS-CoV-2 S pseudotyped neutralization assay in 100 randomly selected sera. A ML model was built using the variables of age, vaccination (number of doses) and SARS-CoV-2 infection status. The model was trained in a cohort (TC) comprising 931 participants and validated in an external cohort (VC) including 787 individuals. Receiver operating characteristics analysis indicated that an anti-SARS-CoV-2 RBD total antibody threshold of 2300 BAU/mL best discriminated between participants either exhibiting or not detectable Omicron BA.2 and Omicron BA.4/5-Spike targeted NtAb responses (87% and 84% precision, respectively). The ML model correctly classified 88% (793/901) of participants in the TC: 717/749 (95.7%) of those displaying ≥2300 BAU/mL and 76/152 (50%) of those exhibiting antibody levels <2300 BAU/mL. The model performed better in vaccinated participants, either with or without prior SARS-CoV-2 infection. The overall accuracy of the ML model in the VC was comparable. Our ML model, based upon a few easily collected parameters for predicting neutralizing activity against Omicron BA.2 and BA.4/5 (sub)variants circumvents the need to perform not only neutralization assays, but also anti-S serological tests, thus potentially saving costs in the setting of large seroprevalence studies.

Diagnosis of pain in the right iliac fossa. A new diagnostic score based on Decision-Tree and Artificial Neural Network Methods

INTRODUCTION

Pain in the right iliac fossa (RIF) continues to pose diagnostic challenges. The objective of this study is the development of a RIF pain diagnosis model based on classification trees of type CHAID (Chi-Square Automatic Interaction Detection) and on an artificial neural network (ANN).

METHODS

Prospective study of 252 patients who visited the hospital due to RIF pain. Demographic, clinical, physical examination and analytical data were registered. Patients were classified into 4 groups: NsP (nonspecific RIFP group), AA (acute appendicitis), NIRIF (RIF pain with no inflammation) and IRIF (RIF pain with inflammation). A CHAID-type classification tree model and an ANN were constructed. The classic models (Alvarado [ALS], Appendicitis Inflammatory Response [AIR] and Fenyö-Linberg [FLS]) were also evaluated. Discrimination was assessed using ROC curves (AUC [95% CI]) and the correct classification rate (CCR).

RESULTS

53% were men. Mean age 33.3±16 years. The largest group was the NsP (45%), AA (37%), NRIF (12%) and IRIF (6%). The analytical model results were: ALS (0.82 [0.76-0.87]), AIR (0.83 [0.77-0.88]) and FLS (0.88 [0.84-0.92]). CHAID determined 10 decision groups: 3 with high probability for NsP, 3 high for AA and 4 special groups with no predominant diagnosis. CCR of ANN and CHAID were 75% and 74.2%, respectively.

CONCLUSIONS

The methodology based on CHAID-type classification trees establishes a diagnostic model based on four pain groups in RIF and generates decision rules that can help us in the diagnosis of processes with RIF pain.

Aid decision algorithms to estimate the risk in congenital heart surgery

BACKGROUND AND OBJECTIVE

In this paper, we have tested the suitability of using different artificial intelligence-based algorithms for decision support when classifying the risk of congenital heart surgery. In this sense, classification of those surgical risks provides enormous benefits as the a priori estimation of surgical outcomes depending on either the type of disease or the type of repair, and other elements that influence the final result. This preventive estimation may help to avoid future complications, or even death.

METHODS

We have evaluated four machine learning algorithms to achieve our objective: multilayer perceptron, self-organizing map, radial basis function networks and decision trees. The architectures implemented have the aim of classifying among three types of surgical risk: low complexity, medium complexity and high complexity.

RESULTS

Accuracy outcomes achieved range between 80% and 99%, being the multilayer perceptron method the one that offered a higher hit ratio.

CONCLUSIONS

According to the results, it is feasible to develop a clinical decision support system using the evaluated algorithms. Such system would help cardiology specialists, paediatricians and surgeons to forecast the level of risk related to a congenital heart disease surgery.

Identifying Thoracic Malignancies Through Pleural Fluid Biomarkers: A Predictive Multivariate Model

The diagnosis of malignant pleural effusions may be challenging when cytological examination of aspirated pleural fluid is equivocal or noncontributory. The purpose of this study was to identify protein candidate biomarkers differentially expressed in the pleural fluid of patients with mesothelioma, lung adenocarcinoma, lymphoma, and tuberculosis (TB).A multiplex protein biochip comprising 120 biomarkers was used to determine the pleural fluid protein profile of 29 mesotheliomas, 29 lung adenocarcinomas, 12 lymphomas, and 35 tuberculosis. The relative abundance of these predetermined biomarkers among groups served to establish the differential diagnosis of: malignant versus benign (TB) effusions, lung adenocarcinoma versus mesothelioma, and lymphoma versus TB. The selected putative markers were validated using widely available commercial techniques in an independent sample of 102 patients.Significant differences were found in the protein expressions of metalloproteinase-9 (MMP-9), cathepsin-B, C-reactive protein, and chondroitin sulfate between malignant and TB effusions. When integrated into a scoring model, these proteins yielded 85% sensitivity, 100% specificity, and an area under the curve (AUC) of 0.98 for labeling malignancy in the verification sample. For lung adenocarcinoma-mesothelioma discrimination, combining CA19-9, CA15-3, and kallikrein-12 had maximal discriminatory capacity (65% sensitivity, 100% specificity, AUC 0.94); figures which also refer to the validation set. Last, cathepsin-B in isolation was only moderately useful (sensitivity 89%, specificity 62%, AUC 0.75) in separating lymphomatous and TB effusions. However, this last differentiation improved significantly when cathepsin-B was used with respect to the patient's age (sensitivity 72%, specificity 100%, AUC 0.94).In conclusion, panels of 4 (i.e., MMP-9, cathepsin-B, C-reactive protein, chondroitin sulfate), or 3 (i.e., CA19-9, CA15-3, kallikrein-12) different protein biomarkers on pleural fluid samples are highly discriminative for signaling a malignant versus tuberculous effusion, or lung adenocarcinoma versus mesothelioma, respectively. Cathepsin-B could also be helpful in establishing the presence of a lymphomatous effusion versus that of TB, if the patient's age is simultaneously taken into consideration.

[Incidence and predictive factors of iron deficiency anemia after acute non-variceal upper gastrointestinal bleeding without portal hypertension]

INTRODUCTION

There are few studies on iron deficiency anemia (IDA) after non-variceal acute upper gastrointestinal bleeding (UGIB) in patients without portal hypertension.

OBJECTIVES

To define the incidence of IDA after UGIB, to characterize the predictive factors for IDA and to design algorithms that could help physicians identify those patients who could benefit from iron therapy.

MATERIAL AND METHOD

We registered 391 patients with UGIB between April 2007 and May 2009. Patients with portal hypertension and those with clinical or/and biological conditions that could affect the ferrokinetic pattern were excluded. Blood analyses were performed, including ferric parameters upon admission, on the 5th day, and on the 30th day after the hemorrhage episode. We used a multiple logistic regression model and a classification and regression tree model.

RESULTS

A total of 124 patients were included, of which 76 (61.3%) developed IDA 30 days after UGIB. The predictive variables were age >75 years (P=.037; OR 3.9; 95% CI: 1.3-11.6), initial urea level >80mg/dL (P=.027; OR 2.9; 95% CI: 1.1-7.6), initial ferritin level ≤65ng/dL (P=.002; OR 7.6; 95% CI: 2.9-18.5), initial hemoglobin level ≤100g/L (P=.003; OR 3.2; 95% CI: 1.3-8.0), hemoglobin level on the 5th day ≤100g/L (P<.001; OR 14.9; 95% CI: 3.6-61.1) and the value of the transferrin saturation index on the 5th day <10% (p<0.001; OR 7.2; 95% CI: 2.6-20.3).

CONCLUSIONS

Most patients with UGIB developed IDA 30 days after the episode. Identification of the predictive factors for IDA may help to establish guidelines for the administration of iron therapy.

New symptom-based predictive tool for survival at seven and thirty days developed by palliative home care teams

AIM

This study sought to develop models to predict survival at 7 and 30 days based on symptoms detected by palliative home care teams (PHCTs).

MATERIALS AND METHODS

This prospective analytic study included a 6-month recruitment period with patient monitoring until death or 180 days after recruitment. The inclusion criteria consisted of age greater than 18 years, advanced cancer, and treatment provided by participating PHCTs between April and July 2009. The study variables included death at 7 or 30 days, survival time, age, gender, place of residence, type of tumor and extension, presence of 11 signs and symptoms measured with a 0-3 Likert scale, functional and cognitive status, and use of a subcutaneous butterfly needle. The statistics applied included a descriptive analysis according to the percentage or mean±standard deviation. For symptom comparison between surviving and nonsurviving patients, the χ(2) test was used. Classification and regression tree (CART) methodology was used for model development. An internal validation system (cross-validation with 10 partitions) was used to ensure generalization of the models. The area under the receiver operating characteristics (ROC) curve was calculated (with a 95% confidence interval) to assess the validation of the models.

RESULTS

A total of 698 patients were included. The mean age of the patients was 73.7±12 years, and 60.3% were male. The most frequent type of neoplasm was digestive (37.6%). The mean Karnofsky score was 51.8±14, the patients' cognitive status according to the Pfeiffer test was 2.6±4 errors, and 8.3% of patients required a subcutaneous butterfly needle. Each model provided 8 decision rules with a probability assignment range between 2.2% and 99.1%. The model used to predict the probability of death at 7 days included the presence of anorexia and dysphagia and the level of consciousness, and this model produced areas under the curve (AUCs) of 0.88 (0.86-0.90) and 0.81 (0.79-0.83). The model used to predict the probability of death at 30 days included the presence of asthenia and anorexia and the level of consciousness, and this model produced AUCs of 0.78 (0.77-0.80) and 0.77 (0.75-0.79).

CONCLUSION

For patients with advanced cancer treated by PHCTs, the use of classification schemes and decision trees based on specific symptoms can help clinicians predict survival at 7 and 30 days.

Prognostic factors in head and neck squamous cell carcinoma: comparison of CHAID decision trees technology and Cox analysis

BACKGROUND

The purpose of this study was to compare the risk factors obtained from a classical statistical method (Cox proportional hazards model) and the results obtained with classification trees (Chi-square Automatic Interaction Detection [CHAID] model) in head and neck squamous cell carcinoma (HNSCC).

METHODS

We conducted a retrospective study of 3373 patients with HNSCC and a follow-up longer than 2 years.

RESULTS

The most decisive variable classified by CHAID was T category. N classification was an important prognostic factor in the Cox analysis, but this was not considered in CHAID except in T2 and supraglottic T3. CHAID also indicated that women with oral cavity T3/T4 to N0 tumors had poorer prognosis (28%) than men (58.5%; p value < .001). In oropharynx location, men had lower survival than women (41% vs 72%; p value < .001).

CONCLUSIONS

The main benefit of CHAID analysis is that it identifies a relatively small number of patients with a singular behavior, which is more discriminatory for an optimal diagnosis and treatment.

Stratification of the severity of critically ill patients with classification trees

Background

Development of three classification trees (CT) based on the CART (Classification and Regression Trees), CHAID (Chi-Square Automatic Interaction Detection) and C4.5 methodologies for the calculation of probability of hospital mortality; the comparison of the results with the APACHE II, SAPS II and MPM II-24 scores, and with a model based on multiple logistic regression (LR).

Methods

Retrospective study of 2864 patients. Random partition (70:30) into a Development Set (DS) n = 1808 and Validation Set (VS) n = 808. Their properties of discrimination are compared with the ROC curve (AUC CI 95%), Percent of correct classification (PCC CI 95%); and the calibration with the Calibration Curve and the Standardized Mortality Ratio (SMR CI 95%).

Results

CTs are produced with a different selection of variables and decision rules: CART (5 variables and 8 decision rules), CHAID (7 variables and 15 rules) and C4.5 (6 variables and 10 rules). The common variables were: inotropic therapy, Glasgow, age, (A-a)O2 gradient and antecedent of chronic illness. In VS: all the models achieved acceptable discrimination with AUC above 0.7. CT: CART (0.75(0.71-0.81)), CHAID (0.76(0.72-0.79)) and C4.5 (0.76(0.73-0.80)). PCC: CART (72(69-75)), CHAID (72(69-75)) and C4.5 (76(73-79)). Calibration (SMR) better in the CT: CART (1.04(0.95-1.31)), CHAID (1.06(0.97-1.15) and C4.5 (1.08(0.98-1.16)).

Conclusion

With different methodologies of CTs, trees are generated with different selection of variables and decision rules. The CTs are easy to interpret, and they stratify the risk of hospital mortality. The CTs should be taken into account for the classification of the prognosis of critically ill patients.