Consolidated Reporting Guidelines for Prognostic and Diagnostic Machine Learning Models (CREMLS).

The number of papers presenting machine learning (ML) models that are being submitted to and published in the Journal of Medical Internet Research and other JMIR Publications journals has steadily increased. Editors and peer reviewers involved in the review process for such manuscripts often go through multiple review cycles to enhance the quality and completeness of reporting. The use of reporting guidelines or checklists can help ensure consistency in the quality of submitted (and published) scientific manuscripts and, for example, avoid instances of missing information. In this Editorial, the editors of JMIR Publications journals discuss the general JMIR Publications policy regarding authors' application of reporting guidelines and specifically focus on the reporting of ML studies in JMIR Publications journals, using the Consolidated Reporting of Machine Learning Studies (CREMLS) guidelines, with an example of how authors and other journals could use the CREMLS checklist to ensure transparency and rigor in reporting.

Determining optimal air leak resolution criteria when using digital pleural drainage device after lung resection.

Objective: There is limited clinical evidence to support any specific parenchymal air leak resolution criteria when using digital pleural drainage devices following lung resection. The aim of this study is to determine an optimal air leak resolution criteria, where duration of chest tube drainage is minimized while avoiding complications from premature chest tube removal. Methods: Airflow data averaged at 10-minute intervals was collected prospectively using a digital pleural drainage device (Thopaz; Medela) in 400 patients from 2015 to 2019. All permutations of air leak resolution criteria from <10 to 100 mL/minute for 4 to 12 hours were applied retrospectively to the pleural drainage data to determine air leak duration, and air leak recurrence frequency and volume. Air leak recurrence indicates potential for rather than occurrence of adverse events. Descriptive statistics were used to identify the optimal criteria based on patient safety (low frequency and volume of air leak recurrences), and efficiency (shortest initial air leak duration). Results: The majority of the 400 patients underwent lobectomy (57% [227 out of 400]), wedge resections (29% [115 out of 400]), or segmentectomies (8% [32 out of 400]) for lung cancer (90% [360 out of 400]). An airflow threshold <50 mL/minute resulted in longer air leak duration before meeting the criteria for air leak resolution (P < .0001). Air leak recurrence frequency and volume were greater in patients with a monitoring period <8 consecutive hours (P < .0001). Conclusions: When using a digital pleural drainage device, a postoperative air leak resolution criteria <50 mL/minute for 8 consecutive hours was associated with the best safety and efficiency profile.

Consolidated Reporting Guidelines for Prognostic and Diagnostic Machine Learning Modeling Studies: Development and Validation.

BACKGROUND: The reporting of machine learning (ML) prognostic and diagnostic modeling studies is often inadequate, making it difficult to understand and replicate such studies. To address this issue, multiple consensus and expert reporting guidelines for ML studies have been published. However, these guidelines cover different parts of the analytics lifecycle, and individually, none of them provide a complete set of reporting requirements. OBJECTIVE: We aimed to consolidate the ML reporting guidelines and checklists in the literature to provide reporting items for prognostic and diagnostic ML in in-silico and shadow mode studies. METHODS: We conducted a literature search that identified 192 unique peer-reviewed English articles that provide guidance and checklists for reporting ML studies. The articles were screened by their title and abstract against a set of 9 inclusion and exclusion criteria. Articles that were filtered through had their quality evaluated by 2 raters using a 9-point checklist constructed from guideline development good practices. The average κ was 0.71 across all quality criteria. The resulting 17 high-quality source papers were defined as having a quality score equal to or higher than the median. The reporting items in these 17 articles were consolidated and screened against a set of 6 inclusion and exclusion criteria. The resulting reporting items were sent to an external group of 11 ML experts for review and updated accordingly. The updated checklist was used to assess the reporting in 6 recent modeling papers in JMIR AI. Feedback from the external review and initial validation efforts was used to improve the reporting items. RESULTS: In total, 37 reporting items were identified and grouped into 5 categories based on the stage of the ML project: defining the study details, defining and collecting the data, modeling methodology, model evaluation, and explainability. None of the 17 source articles covered all the reporting items. The study details and data description reporting items were the most common in the source literature, with explainability and methodology guidance (ie, data preparation and model training) having the least coverage. For instance, a median of 75% of the data description reporting items appeared in each of the 17 high-quality source guidelines, but only a median of 33% of the data explainability reporting items appeared. The highest-quality source articles tended to have more items on reporting study details. Other categories of reporting items were not related to the source article quality. We converted the reporting items into a checklist to support more complete reporting. CONCLUSIONS: Our findings supported the need for a set of consolidated reporting items, given that existing high-quality guidelines and checklists do not individually provide complete coverage. The consolidated set of reporting items is expected to improve the quality and reproducibility of ML modeling studies.

Dissecting the impact of molecular T-cell HLA mismatches in kidney transplant failure: A retrospective cohort study.

Introduction: Kidney transplantation is the optimal treatment in end-stage kidney disease, but de-novo donor specific antibody development continues to negatively impact patients undergoing kidney transplantation. One of the recent advances in solid organ transplantation has been the definition of molecular mismatching between donors and recipients' Human Leukocyte Antigens (HLA). While not fully integrated in standard clinical care, cumulative molecular mismatch at the level of eplets (EMM) as well as the PIRCHE-II score have shown promise in predicting transplant outcomes. In this manuscript, we sought to study whether certain T-cell molecular mismatches (TcEMM) were highly predictive of death-censored graft failure (DCGF). Methods: We studied a retrospective cohort of kidney donor:recipient pairs from the Scientific Registry of Transplant Recipients (2000-2015). Allele level HLA-A, B, C, DRB1 and DQB1 types were imputed from serologic types using the NMDP algorithm. TcEMMs were then estimated using the PIRCHE-II algorithm. Multivariable Accelerated Failure Time (AFT) models assessed the association between each TcEMM and DCGF. To discriminate between TcEMMs most predictive of DCGF, we fit multivariable Lasso penalized regression models. We identified co-expressed TcEMMs using weighted correlation network analysis (WGCNA). Finally, we conducted sensitivity analyses to address PIRCHE and IMGT/HLA version updates. Results: A total of 118,309 donor:recipient pairs meeting the eligibility criteria were studied. When applying the PIRCHE-II algorithm, we identified 1,935 distinct TcEMMs at the population level. A total of 218 of the observed TcEMM were independently associated with DCGF by AFT models. The Lasso penalized regression model with post selection inference identified a smaller subset of 86 TcEMMs (56 and 30 TcEMM derived from HLA Class I and II, respectively) to be highly predictive of DCGF. Of the observed TcEMM, 38.14% appeared as profiles of highly co-expressed TcEMMs. In addition, sensitivity analyses identified that the selected TcEMM were congruent across IMGT/HLA versions. Conclusion: In this study, we identified subsets of TcEMMs highly predictive of DCGF and profiles of co-expressed mismatches. Experimental verification of these TcEMMs determining immune responses and how they may interact with EMM as predictors of transplant outcomes would justify their consideration in organ allocation schemes and for modifying immunosuppression regimens.

Antifungals influence the immune-related transcriptomic landscape of human monocytes after Aspergillus fumigatus infection.

The heterogeneity of clinical responses to antifungals in aspergillosis is partially understood. We hypothesized that besides direct antifungal effects, these discrepancies may be related to different immunomodulatory profiles. Human THP-1 monocytes were coincubated in vitro with Aspergillus fumigatus and variable concentrations of voriconazole (0.5, 1 and 2 mg/l), caspofungin (1 and 2 mg/l), amphotericin B deoxycholate (0.25, 0.5 and 1 mg/l) and liposomal amphotericin B (1, 2 and 3 mg/l). After 6 h of coincubation, total cellular RNA was extracted, converted into cDNA, and transcription of 84 genes involved in antifungal immunity was measured through RT-qPCR. The presence of A. fumigatus was the main driver of the global immune-related transcriptomic response. After Aspergillus infection, thirty genes were upregulated, while 19 genes were downregulated. Discrepancies across antifungals were also evident; voriconazole-containing conditions showed similar reaction to natural infection, while the use of liposomal Amphotericin B significantly decreased the inflammatory response. Chemokines (notably CCL20 and CXCL2) and pro-inflammatory cytokines (IL1A, IL1B, IL23, G-CSF) exhibited the most pronounced differences across antifungals. Pattern recognition receptors and adaptor protein transcription were minimally affected. Protein-protein-interaction network analysis showed that IL23A played a dominant role in upregulated genes. Pathway enrichment analysis indicated that cytokine-cytokine receptor integration, TNF signaling pathways and Toll-like receptor pathways were highly involved. This exploratory study confirms the heterogeneous immunomodulatory role of antifungals. Overall, voriconazole appears to maintain an early pro-inflammatory response seen in natural infection. Assessment of immunomodulatory response with clinical response may provide a better rationale for differences observed across antifungals.

Bronchoalveolar lavage cytokine-based risk stratification of minimal acute rejection in clinically stable lung transplant recipients.

BACKGROUND: Acute cellular rejection (ACR) remains the most significant risk factor for chronic lung allograft dysfunction (CLAD). While clinically significant or higher-grade (≥A2) ACR is generally treated with augmented immunosuppression (IS), the management of clinically stable grade A1 ACR remains controversial. At our center, patients with clinically stable grade A1 ACR are routinely not treated with augmented IS. While the overall outcomes in this group of patients at our center are equivalent to patients with stable A0 pathology, CLAD and death rates remain overall high. We hypothesized that a distinct cytokine signature at the time of early minimal rejection state would be associated with worse outcomes. Specifically, we aimed to determine whether bronchoalveolar lavage (BAL) biomarkers at the time of first clinically stable grade A1 ACR (CSA1R) are predictive of subsequent CLAD or death. METHODS: Among all adult, bilateral, first lung transplants, performed 2010-2016, transbronchial biopsies obtained within the first-year post-transplant were categorized as clinically stable or unstable based on the presence or absence of ≥10% concurrent drop in forced expiratory volume in 1 second (FEV1). We assessed BAL samples obtained at the time of CSA1R episodes, which were not preceded by another ACR (i.e., first episodes). Twenty-one proteins previously associated with ACR or CLAD were measured in the BAL using a multiplex bead assay. Association between protein levels and subsequent CLAD or death was assessed using Cox Proportional Hazards models, adjusted for relevant peri-transplant clinical covariates. RESULTS: We identified 75 patients with first CSA1R occurring at a median time of 98 days (range 48.5-197) post-transplant. Median time from transplant to CLAD or death was 1247 (756.5-1921.5) and 1641 days (1024.5-2326.5), respectively. In multivariable models, levels of MCP1/CCL2, S100A8, IL10, TNF-receptor 1, and pentraxin 3 (PTX3) were associated with both CLAD development and death (p < 0.05 for all). PTX3 remained significantly associated with both CLAD and death after adjusting for multiple comparisons. CONCLUSION: Our data indicate that a focused BAL protein signature, with PTX3 having the strongest association, may be useful in determining a subset of CSA1R patients at increased risk and may benefit from a more aggressive management strategy.

On Path to Informing Hierarchy of Eplet Mismatches as Determinants of Kidney Transplant Loss.

INTRODUCTION: To mitigate risks related to human leukocyte antigen (HLA) incompatibility, we assessed whether certain structurally defined HLA targets present in donors but absent from recipients, known as eplet mismatches (EMM), are associated with death-censored graft failure (DCGF). METHODS: We studied a cohort of 118,313 American 0% panel reactive antibodies (PRA) first kidney transplant recipients (2000 to 2015) from the Scientific Registry of Transplant Recipients. Imputed allele-level donor and recipient HLA-A, -B, -C, -DRB1, and -DQB1 genotypes were converted to the repertoire of EMM. We fit survival models for each EMM with significance thresholds corrected for false discovery rate and validated those in an independent PRA > 0% cohort. We conducted network-based analyses to model relationships among EMM and developed models to select the subset of EMM most predictive of DCGF. RESULTS: Of 412 EMM observed, 119 class I and 118 class II EMM were associated with DCGF. Network analysis showed that although 210 eplets formed profiles of 2 to 12 simultaneously occurring EMMs, 202 were singleton EMMs that were not involved in any profile. A variable selection procedure identified 55 single HLA class I and II EMMs in 70% of the dataset; of those, 15 EMMs (9 singleton and 6 involved in profiles) were predictive of DCGF in the remaining dataset. CONCLUSION: Our analysis distinguished increasingly smaller subsets of EMMs associated with increased risk of DCGF. Validation of these EMMs as important predictors of transplant outcomes (in contrast to acceptable EMMs) in datasets with measured allele-level genotypes will support their role as immunodominant EMMs worthy of consideration in organ allocation schemes.

Matchmaker, matchmaker make me a match: Opportunities and challenges in optimizing compatibility of HLA eplets in transplantation.

The development of donor-specific antibodies (DSAs) is a major complication in transplantation, which is associated with inferior graft survival, impaired quality of life, and increased healthcare costs. DSA develop upon recognition of nonself HLA by the recipient's immune system. HLA molecules contain epitopes, which are the surface regions of HLA molecules recognized by antibodies. HLAMatchmaker is an algorithm for assessing donor:recipient HLA compatibility at the level of structurally defined HLA targets called eplets. The consideration of eplets, rather than the whole HLA molecule, could offer some advantages when classifying the immune risk associated with particular donor:recipient pairs. Assessing compatibility at the level of HLA eplets could decrease misclassification of post-transplant immune risk by improving specificity, when antibodies are confirmed to be directed against donor eplets missing from the recipient's repertoire of eplets. Consideration of eplets may also increase the sensitivity of immune risk assessment, when identifying mismatched eplets that could give rise to new, not previously detected, donor-specific antibodies post-transplant. Eplet matching can serve as a rational strategy for immune risk mitigation. Herein, we review the evolution of HLA (in) compatibility assessment for organ allocation. We outline challenges in the implementation of eplet-based donor:recipient matching, including unavailability of allele-level donor genotypes for 11 HLA loci at the time of organ allocation and difficulty in assessing the hierarchy of immune risk associated with particular HLA eplet mismatches. Opportunities to address some of the current shortcomings of donor genotyping and HLAMatchmaker are also discussed. While there is a demonstrated benefit in the application of HLAMatchmaker for donor: recipient HLA (in)compatibility assessment, evolving long-read genotyping methods, compilation of large data sets with allele-level genotypes, and standardization of methods to verify eplets as determinants of immune-mediated injuries are required before HLA eplet matching is implemented in organ allocation to improve upon transplant outcomes.

Eosinophils in transbronchial biopsies: a predictor of chronic lung allograft dysfunction and reduced survival after lung transplantation - a retrospective single-center cohort study.

Long-term outcomes after lung transplantation remain inferior to those of other solid organ groups. The significance of eosinophils detected on transbronchial biopsies (TBBx) after lung transplantation and their relationship to long-term outcomes remain unknown. A retrospective single-center cohort study was performed of patients transplanted between January 01, 2001, and July 31, 2018, who had at least 1 TBBx with evaluable parenchymal tissue. Multivariable Cox proportional hazard models were used to assess the associations between eosinophil detection and: all-cause mortality and Chronic Lung Allograft Dysfunction (CLAD). 8887 TBBx reports from 1440 patients were reviewed for the mention of eosinophils in the pathology report. 112 (7.8%) patients were identified with eosinophils on at least one TBBx. The median (95% CI) survival time for all patients was 8.28 (7.32-9.31) years. Multivariable analysis, adjusted for clinical variables known to affect post-transplant outcomes, showed that the detection of eosinophils was independently associated with an increased risk of death (HR 1.51, 95% CI 1.24-1.85, p < 0.01) and CLAD (HR 1.35, 95% CI 1.07-1.70, P = 0.01). Eosinophils detected in TBBx are associated with an increased risk of CLAD and death. There may be benefit in specifically reporting the presence of eosinophils in TBBx reports and incorporating their presence in clinical decision-making.

Identifying host microRNAs in bronchoalveolar lavage samples from lung transplant recipients infected with Aspergillus.

BACKGROUND: MicroRNAs (miRNAs) are small non-coding RNAs of ∼22 nucleotides that play a crucial role in post-transcriptional regulation of gene expression. Dysregulation of miRNA expression has been shown during microbial infections. We sought to identify miRNAs that distinguish invasive aspergillosis (IA) from non-IA in lung transplant recipients (LTRs). METHODS: We used NanoString nCounter Human miRNA, version 3, panel to measure miRNAs in bronchoalveolar lavage (BAL) samples from LTRs with Aspergillus colonization (ASP group) (n = 10), those with Aspergillus colonization and chronic lung allograft dysfunction (CLAD) (ASPCLAD group) (n = 7), those with IA without CLAD (IA group) (n = 10), those who developed IA with CLAD (IACLAD group) (n = 9), and control patients (controls) (n = 9). The miRNA profile was compared using the permutation test of 100,000 trials for each of the comparisons. We used mirDIP to obtain their gene targets and pathDIP to determine the pathway enrichment. RESULTS: We performed pairwise comparisons between patient groups to identify differentially expressed miRNAs. A total of 5 miRNAs were found to be specific to IA, including 4 (miR-145-5p, miR-424-5p, miR-99b-5p, and miR-4488) that were upregulated and the pair (miR-4454 + miR-7975) that was downregulated in IA group vs controls. The expression change for these miRNAs was specific to patients with IA; they were not significantly differentiated between IACLAD and IA groups. Signaling pathways associated with an immunologic response to IA were found to be significantly enriched. CONCLUSIONS: We report a set of 5 differentially expressed miRNAs in the BAL of LTRs with IA that might help in the development of diagnostic and prognostic tools for IA in LTRs. However, further investigation is needed in a larger cohort to validate the findings.

A machine learning approach predicts future risk to suicidal ideation from social media data.

Machine learning analysis of social media data represents a promising way to capture longitudinal environmental influences contributing to individual risk for suicidal thoughts and behaviors. Our objective was to generate an algorithm termed "Suicide Artificial Intelligence Prediction Heuristic (SAIPH)" capable of predicting future risk to suicidal thought by analyzing publicly available Twitter data. We trained a series of neural networks on Twitter data queried against suicide associated psychological constructs including burden, stress, loneliness, hopelessness, insomnia, depression, and anxiety. Using 512,526 tweets from N = 283 suicidal ideation (SI) cases and 3,518,494 tweets from 2655 controls, we then trained a random forest model using neural network outputs to predict binary SI status. The model predicted N = 830 SI events derived from an independent set of 277 suicidal ideators relative to N = 3159 control events in all non-SI individuals with an AUC of 0.88 (95% CI 0.86-0.90). Using an alternative approach, our model generates temporal prediction of risk such that peak occurrences above an individual specific threshold denote a ~7 fold increased risk for SI within the following 10 days (OR = 6.7 ± 1.1, P = 9 × 10-71). We validated our model using regionally obtained Twitter data and observed significant associations of algorithm SI scores with county-wide suicide death rates across 16 days in August and in October, 2019, most significantly in younger individuals. Algorithmic approaches like SAIPH have the potential to identify individual future SI risk and could be easily adapted as clinical decision tools aiding suicide screening and risk monitoring using available technologies.

The impact of first untreated subclinical minimal acute rejection on risk for chronic lung allograft dysfunction or death after lung transplantation.

Acute cellular rejection (ACR) is a significant risk factor for chronic lung allograft dysfunction (CLAD). Although clinically manifest and higher grade (≥A2) ACR is generally treated with augmented immunosuppression, management of minimal (grade A1) ACR remains controversial. In our program, patients with subclinical and spirometrically stable A1 rejection (StA1R) are routinely not treated with augmented immunosuppression. We hypothesized that an untreated first StA1R does not increase the risk of CLAD or death compared to episodes of spirometrically stable no ACR (StNAR). The cohort was drawn from all consecutive adult, first, bilateral lung transplantations performed between 1999 and 2017. Biopsies obtained in the first-year posttransplant were paired with (forced expiratory volume in 1 second FEV1 ). The first occurrence of StA1R was compared to a time-matched StNAR. The risk of CLAD or death was assessed using univariable and multivariable Cox proportional hazards models. The analyses demonstrated no significant difference in risk of CLAD or death in patients with a first StA1R compared to StNAR. This largest study to date shows that, in clinically stable patients, an untreated first A1 ACR in the first-year posttransplant is not significantly associated with an increased risk for CLAD or death. Watchful-waiting approach may be an acceptable tactic for stable A1 episodes in lung transplant recipients.

Donor human leukocyte antigen-G single nucleotide polymorphisms are associated with post-lung transplant mortality.

Human leukocyte antigen (HLA)-G is a non-classical HLA that inhibits immune responses. Its expression is modified by single nucleotide polymorphisms (SNPs), which are associated with transplant outcomes. Our aim was to investigate the association of donor and recipient HLA-G SNPs with chronic lung allograft dysfunction (CLAD) and mortality after lung transplantation.In this single-centre study, we examined 11 HLA-G SNPs in 345 consecutive recipients and 297 donors of a first bilateral lung transplant. A multivariable Cox proportional hazards model assessed associations of SNPs with death and CLAD. Transbronchial biopsies (TBBx) and bronchoalveolar lavage (BAL) samples were examined using quantitative PCR, ELISA and immunofluorescence.Over a median of 4.75 years, 142 patients (41%) developed CLAD; 170 (49%) died. Multivariable analysis revealed donor SNP +3142 (GG+CG versus CC) was associated with increased mortality (hazard ratio 1.78, 95% CI 1.12-2.84; p=0.015). In contrast, five donor SNPs, -201(CC), -716(TT), -56(CC), G*01:03(AA) and 14 bp INDEL, conferred reduced mortality risk. Specific donor-recipient SNP pairings reduced CLAD risk. Predominantly epithelial HLA-G expression was observed on TBBx without rejection. Soluble HLA-G was present in higher concentrations in the BAL samples of patients who later developed CLAD.Specific donor SNPs were associated with mortality risk after lung transplantation, while certain donor-recipient SNP pairings modulated CLAD risk. TBBx demonstrated predominantly epithelial, and therefore presumably donor-derived, HLA-G expression in keeping with these observations. This study is the first to demonstrate an effect of donor HLA-G SNPs on lung transplantation outcome.

Towards donor lung recovery-gene expression changes during ex vivo lung perfusion of human lungs.

We and others have demonstrated that acellular normothermic ex vivo lung perfusion of high-risk donor lungs can result in posttransplant outcomes equivalent to that of contemporaneous lung transplantation using standard donor lungs. However, the mechanism of this effect remains unclear. Given the restoration of cellular metabolic activity during normothermic perfusion, one possibility is that of lung healing via natural innate recovery mechanisms. We explored this by examining the gene expression changes occurring in human lungs during ex vivo lung perfusion. Human lungs clinically rejected for transplantation were perfused for 12 hours of EVLP with biopsies taken at the start, at 1 hour, at 3 hours, and then every 3 hours thereafter to 12 hours. Temporal changes were identified in 2585 genes using the Short Time-series Expression Miner and used for pathway analysis. Despite increases in endothelial markers of inflammation, circulating leukocyte cell-specific gene expression fell over 12 hours of ex vivo lung perfusion (EVLP), suggesting an interrupted inflammation response secondary to washout of circulating leukocytes. Analysis of these gene changes suggests lung recovery follows specific stages: cellular death, cellular preservation, cellular reorganization, and cellular invasion. EVLP may improve posttransplant lung function by washout of leukocytes and facilitating innate mechanisms of repair.

An Integrated Approach Identifies Mediators of Local Recurrence in Head and Neck Squamous Carcinoma.

Purpose: Head and neck squamous cell carcinomas (HNSCCs) cause more than 300,000 deaths worldwide each year. Locoregional and distant recurrences represent worse prognostic events and accepted surrogate markers of patients' overall survival. No valid biomarker and salvage therapy exist to identify and treat patients at high-risk of recurrence. We aimed to verify if selected miRNAs could be used as biomarkers of recurrence in HNSCC.Experimental Design: A NanoString array was used to identify miRNAs associated with locoregional recurrence in 44 patients with HNSCC. Bioinformatic approaches validated the signature and identified potential miRNA targets. Validation experiments were performed using an independent cohort of primary HNSCC samples and a panel of HNSCC cell lines. In vivo experiments validated the in vitro results.Results: Our data identified a four-miRNA signature that classified HNSCC patients at high- or low-risk of recurrence. These miRNAs collectively impinge on the epithelial-mesenchymal transition process. In silico and wet lab approaches showed that miR-9, expressed at high levels in recurrent HNSCC, targets SASH1 and KRT13, whereas miR-1, miR-133, and miR-150, expressed at low levels in recurrent HNSCC, collectively target SP1 and TGFß pathways. A six-gene signature comprising these targets identified patients at high risk of recurrences, as well. Combined pharmacological inhibition of SP1 and TGFß pathways induced HNSCC cell death and, when timely administered, prevented recurrence formation in a preclinical model of HNSCC recurrence.Conclusions: By integrating different experimental approaches and competences, we identified critical mediators of recurrence formation in HNSCC that may merit to be considered for future clinical development. Clin Cancer Res; 23(14); 3769-80. ©2017 AACR.

Potential role of CC chemokine receptor 6 in prediction of late-onset cytomegalovirus infection following solid organ transplant.

CCR6 is a chemokine receptor involved in homing memory T cells, particularly Th17 cells, to sites of mucosal inflammation. Despite the critical role of memory T cells in long-term protective immunity against cytomegalovirus (CMV), a virus that reactivates at multiple mucosal sites, the ability of CCR6 or other Th17 marker expression to predict CMV reactivation following transplantation is not clear. Using 11-color flow cytometry, in this prospective single-center pilot study, we measured the expression of CCR6 and other markers of T-cell function in peripheral blood samples obtained from 21 SOT recipients at the time of discontinuation of anti-CMV prophylaxis. CMV viremia was monitored on a monthly basis after discontinuation of prophylaxis. Eleven patients (52%) developed CMV viremia during the six-month follow-up period. Late-onset CMV infection was preceded by an immune phenotype characterized by increased CCR6 expression on bulk CD4(+) T cells and a reduced number of circulating CMV IE-1-specific Th1 (CD4(+) IFN-γ(+)) cells. Among the markers evaluated, CCR6 was the best single predictor of late-onset CMV infection. Our results suggest that CCR6 expression at the time of discontinuation of antiviral prophylaxis might be a useful predictor of late-onset CMV reactivation and provide the basis for future larger prospective studies.

Predicting the need for CT imaging in children with minor head injury using an ensemble of Naive Bayes classifiers.

OBJECTIVE: Using an automatic data-driven approach, this paper develops a prediction model that achieves more balanced performance (in terms of sensitivity and specificity) than the Canadian Assessment of Tomography for Childhood Head Injury (CATCH) rule, when predicting the need for computed tomography (CT) imaging of children after a minor head injury. METHODS AND MATERIALS: CT is widely considered an effective tool for evaluating patients with minor head trauma who have potentially suffered serious intracranial injury. However, its use poses possible harmful effects, particularly for children, due to exposure to radiation. Safety concerns, along with issues of cost and practice variability, have led to calls for the development of effective methods to decide when CT imaging is needed. Clinical decision rules represent such methods and are normally derived from the analysis of large prospectively collected patient data sets. The CATCH rule was created by a group of Canadian pediatric emergency physicians to support the decision of referring children with minor head injury to CT imaging. The goal of the CATCH rule was to maximize the sensitivity of predictions of potential intracranial lesion while keeping specificity at a reasonable level. After extensive analysis of the CATCH data set, characterized by severe class imbalance, and after a thorough evaluation of several data mining methods, we derived an ensemble of multiple Naive Bayes classifiers as the prediction model for CT imaging decisions. RESULTS: In the first phase of the experiment we compared the proposed ensemble model to other ensemble models employing rule-, tree- and instance-based member classifiers. Our prediction model demonstrated the best performance in terms of AUC, G-mean and sensitivity measures. In the second phase, using a bootstrapping experiment similar to that reported by the CATCH investigators, we showed that the proposed ensemble model achieved a more balanced predictive performance than the CATCH rule with an average sensitivity of 82.8% and an average specificity of 74.4% (vs. 98.1% and 50.0% for the CATCH rule respectively). CONCLUSION: Automatically derived prediction models cannot replace a physician's acumen. However, they help establish reference performance indicators for the purpose of developing clinical decision rules so the trade-off between prediction sensitivity and specificity is better understood.

Exploiting the systematic review protocol for classification of medical abstracts.

OBJECTIVE: To determine whether the automatic classification of documents can be useful in systematic reviews on medical topics, and specifically if the performance of the automatic classification can be enhanced by using the particular protocol of questions employed by the human reviewers to create multiple classifiers. METHODS AND MATERIALS: The test collection is the data used in large-scale systematic review on the topic of the dissemination strategy of health care services for elderly people. From a group of 47,274 abstracts marked by human reviewers to be included in or excluded from further screening, we randomly selected 20,000 as a training set, with the remaining 27,274 becoming a separate test set. As a machine learning algorithm we used complement naïve Bayes. We tested both a global classification method, where a single classifier is trained on instances of abstracts and their classification (i.e., included or excluded), and a novel per-question classification method that trains multiple classifiers for each abstract, exploiting the specific protocol (questions) of the systematic review. For the per-question method we tested four ways of combining the results of the classifiers trained for the individual questions. As evaluation measures, we calculated precision and recall for several settings of the two methods. It is most important not to exclude any relevant documents (i.e., to attain high recall for the class of interest) but also desirable to exclude most of the non-relevant documents (i.e., to attain high precision on the class of interest) in order to reduce human workload. RESULTS: For the global method, the highest recall was 67.8% and the highest precision was 37.9%. For the per-question method, the highest recall was 99.2%, and the highest precision was 63%. The human-machine workflow proposed in this paper achieved a recall value of 99.6%, and a precision value of 17.8%. CONCLUSION: The per-question method that combines classifiers following the specific protocol of the review leads to better results than the global method in terms of recall. Because neither method is efficient enough to classify abstracts reliably by itself, the technology should be applied in a semi-automatic way, with a human expert still involved. When the workflow includes one human expert and the trained automatic classifier, recall improves to an acceptable level, showing that automatic classification techniques can reduce the human workload in the process of building a systematic review.