Clinical Translation of a Deep Learning Model of Radiation-Induced Lymphopenia for Esophageal Cancer.

Purpose: Radiation-induced lymphopenia is a common immune toxicity that adversely impacts treatment outcomes. We report here our approach to translate a deep-learning (DL) model developed to predict severe lymphopenia risk among esophageal cancer into a strategy for incorporating the immune system as an organ-at-risk (iOAR) to mitigate the risk. Materials and Methods: We conducted "virtual clinical trials" utilizing retrospective data for 10 intensity-modulated radiation therapy (IMRT) and 10 passively-scattered proton therapy (PSPT) esophageal cancer patients. For each patient, additional treatment plans of the modality other than the original were created employing standard-of-care (SOC) dose constraints. Predicted values of absolute lymphocyte count (ALC) nadir for all plans were estimated using a previously-developed DL model. The model also yielded the relative magnitudes of contributions of iOARs dosimetric factors to ALC nadir, which were used to compute iOARs dose-volume constraints, which were incorporated into optimization criteria to produce "IMRT-enhanced" and "intensity-modulated proton therapy (IMPT)-enhanced" plans. Results: Model-predicted ALC nadir for the original IMRT (IMRT-SOC) and PSPT plans agreed well with actual values. IMPT-SOC showed greater immune sparing vs IMRT and PSPT. The average mean body doses were 13.10 Gy vs 7.62 Gy for IMRT-SOC vs IMPT-SOC for patients treated with IMRT-SOC; and 8.08 Gy vs 6.68 Gy for PSPT vs IMPT-SOC for patients treated with PSPT. For IMRT patients, the average predicted ALC nadir of IMRT-SOC, IMRT-enhanced, IMPT-SOC, and IMPT-enhanced was 281, 327, 351, and 392 cells/µL, respectively. For PSPT patients, the average predicted ALC nadir of PSPT, IMPT-SOC, and IMPT-enhanced was 258, 316, and 350 cells/µL, respectively. Enhanced plans achieved higher predicted ALC nadir, with an average improvement of 40.8 cells/µL (20.6%). Conclusion: The proposed DL model-guided strategy to incorporate the immune system as iOAR in IMRT and IMPT optimization has the potential for radiation-induced lymphopenia mitigation. A prospective clinical trial is planned.

Radiation-Induced Lymphopenia is a Causal Mediator of Survival After Chemoradiation Therapy for Esophagus Cancer.

Purpose: Radiation-induced lymphopenia (RIL) is common during chemoradiation therapy. Severe lymphopenia is associated with reduced survival. Proton beam therapy (PBT), with its substantially more compact dose distributions, spares circulating lymphocytes and immune organs at risk to a greater extent than photon therapy. Recent studies comparing PBT to photon radiation therapy, specifically intensity-modulated radiation therapy (IMRT) for esophageal cancer (EC), showed that the incidence of grade 4 RIL (G4RIL) is significantly reduced among patients receiving PBT for EC. However, whether the extent of this reduction has a direct causative link with improved survival is unknown. This study applies causal mediation analysis to answer this question. Methods and Materials: We retrospectively assessed 734 patients treated with concurrent chemoradiation therapy for biopsy-proven EC from 2004 to 2017. To address the potential for bias in the choice of radiation modality, propensity score analysis was used to evaluate and reduce imbalances between the PBT and IMRT cohorts. Causal mediation analysis was applied to decompose the total effect of radiation modality on overall survival (OS) into indirect (mediated through G4RIL) and direct effects. Results: We found that PBT was associated with a significantly lower incidence of G4RIL and prolonged OS compared with IMRT (odds ratio, 0.41; 95% CI, 0.28-0.60; P < .001). In the propensity-matched cohort of 506 patients (253 PBT, 253 IMRT), G4RIL risk reduction with PBT versus IMRT translated into a 5% reduction in the relative rate of death (P = .032). Mediation of G4RIL explained ∼14.5% of the difference in OS. Conclusions: G4RIL was found to mediate survival; however, a statistically significant direct effect of PBT on survival was not observed. In other words, the statistical significance of survival benefit from protons over photons in this EC cohort was lost in the absence of G4RIL risk reduction.

Machine learning identifies prognostic subtypes of the tumor microenvironment of NSCLC.

The tumor microenvironment (TME) plays a fundamental role in tumorigenesis, tumor progression, and anti-cancer immunity potential of emerging cancer therapeutics. Understanding inter-patient TME heterogeneity, however, remains a challenge to efficient drug development. This article applies recent advances in machine learning (ML) for survival analysis to a retrospective study of NSCLC patients who received definitive surgical resection and immune pathology following surgery. ML methods are compared for their effectiveness in identifying prognostic subtypes. Six survival models, including Cox regression and five survival machine learning methods, were calibrated and applied to predict survival for NSCLC patients based on PD-L1 expression, CD3 expression, and ten baseline patient characteristics. Prognostic subregions of the biomarker space are delineated for each method using synthetic patient data augmentation and compared between models for overall survival concordance. A total of 423 NSCLC patients (46% female; median age [inter quantile range]: 67 [60-73]) treated with definite surgical resection were included in the study. And 219 (52%) patients experienced events during the observation period consisting of a maximum follow-up of 10 years and median follow up 78 months. The random survival forest (RSF) achieved the highest predictive accuracy, with a C-index of 0.84. The resultant biomarker subtypes demonstrate that patients with high PD-L1 expression combined with low CD3 counts experience higher risk of death within five-years of surgical resection.

Personalized Composite Dosimetric Score-Based Machine Learning Model of Severe Radiation-Induced Lymphopenia Among Patients With Esophageal Cancer.

PURPOSE: Radiation-induced lymphopenia (RIL) is common among patients undergoing radiation therapy (RT)' Severe RIL has been linked to adverse outcomes. The severity and risk of RIL can be predicted from baseline clinical characteristics and dosimetric parameters. However, dosimetric parameters, e.g. dose-volume (DV) indices, are highly correlated with one another and are only weakly associated with RIL. Here we introduce the novel concept of "composite dosimetric score" (CDS) as the index that condenses the dose distribution in immune tissues of interest to study the dosimetric dependence of RIL. We derived an improved multivariate classification scheme for risk of grade 4 RIL (G4RIL), based on this novel RT dosimetric feature, for patients receiving chemo RT for esophageal cancer. METHODS AND MATERIALS: DV indices were extracted for 734 patients who received chemo RT for biopsy-proven esophageal cancer. Nonnegative matrix factorization was used to project the DV indices of lung, heart, and spleen into a single CDS; XGBoost was employed to explore significant interactions among predictors; and logistic regression was applied to combine the resultant CDS with baseline clinical factors and interaction terms to facilitate individualized prediction of immunotoxicity. Five-fold cross-validation was applied to evaluate the model performance. RESULTS: The CDS for selected immune organs at risk (ie, heart, lung, and spleen) (OR 1.791; 95 CI [1.350, 2.377]) was a statistically significant risk determinant for G4RIL. Pearson correlation coefficients for CDS versus G4RIL risk for individual immune organs at risk were greater than any single DV indicx. Personalized prediction of G4RIL based on CDS and 4 clinical risk factors yielded an area under the curve value of 0.78. Interaction between age and CDS revealed that G4RIL risk increased more sharply with increasing CDS for patients aged ≥65 years. CONCLUSIONS: Risk of immunotoxicity for patients undergoing chemo RT for esophageal cancer can be predicted by CDS. The CDS concept can be extended to immunotoxicity in other cancer types and in dose-response models currently based on DV indices. Personalized treatment planning should leverage composite dosimetric scoring methods rather than using individual or subsets of DV indices.

Comparing 90-Day Postoperative Mortality After Neoadjuvant Proton-Based Versus Photon-Based Chemoradiotherapy for Esophageal Cancer.

Purpose: Evidence suggests that proton-beam therapy (PBT) results in less toxicity and postoperative complications compared to photon-based radiotherapy in patients who receive chemoradiotherapy followed by esophagectomy for cancer. Ninety-day mortality (90DM) is an important measure of the postoperative (nononcologic) outcome as proxy of quality-of-care. We hypothesize that PBT could reduce 90DM compared to photon-based radiotherapy. Materials and Methods: From a single-center retrospective database patients treated with chemoradiotherapy before esophagectomy for cancer were selected (1998-2022). Univariable logistic regression was used to study the association of radiotherapy modality with 90DM. Three separate methods were applied to adjust for confounding bias, including multivariable logistic regression, propensity score matching, and inverse probability of treatment weighting. Stratified analysis for the age threshold that maximized the difference in 90DM (ie, ≥67 vs <67 years) was performed. Results: A total of 894 eligible patients were included and 90DM was 5/202 (2.5%) in the PBT versus 29/692 (4.2%) in the photon-based radiotherapy group (P = .262). After adjustment for age and tumor location, PBT versus photon-based radiotherapy was not significantly associated with 90DM (P = .491). The 90DM was not significantly different for PBT versus photon-based radiotherapy in the propensity score matching (P = .379) and inverse probability of treatment weighting cohort (P = .426). The stratified analysis revealed that in patients aged ≥67 years, PBT was associated with decreased 90DM (1.3% vs 8.8%; P = .026). Higher age significantly increased 90DM risk within the photon-based radiotherapy (8.8% vs 2.7%; P = .001), but not within the PBT group (1.3% vs 3.2%; P = .651). Conclusion: No statistically significant difference was observed in postoperative 90DM after esophagectomy for cancer between PBT and photon-based neoadjuvant chemoradiotherapy. However, among older patients a signal was observed that PBT may reduce 90DM risk.

Severe Lymphopenia During Chemoradiation Therapy for Esophageal Cancer: Comprehensive Analysis of Randomized Phase 2B Trial of Proton Beam Therapy Versus Intensity Modulated Radiation Therapy.

PURPOSE: Lymphocytes play an important role in antitumor immunity; however, they are also especially vulnerable to depletion during chemoradiation therapy (CRT). The purpose of this study was to compare the incidence of grade 4 lymphopenia (G4L) between proton beam therapy (PBT) and intensity modulated photon radiation therapy (IMRT) in patients with esophageal cancer treated with CRT in a completed randomized trial and to ascertain patient heterogeneity to G4L risk based on treatment and established prognostic factors. METHODS AND MATERIALS: Between April 2012 and March 2019, a single-institution, open-label, nonblinded, phase 2 randomized trial (NCT01512589) was conducted at the University of Texas MD Anderson Cancer Center. Patients were randomly assigned to IMRT or PBT, either definitively or preoperatively. This secondary analysis of the randomized trial was G4L during concurrent CRT according to Common Terminology Criteria for Adverse Events version 5.0. RESULTS: Among 105 patients evaluable for analysis, 44 patients (42%) experienced G4L at a median of 28 days after the start date of concurrent CRT. Induction chemotherapy (P = .003), baseline absolute lymphocyte count (P < .001), radiation therapy modality (P = .002), and planning treatment volume (P = .033) were found to be significantly associated with G4L. Multivariate classification analysis partitioned patients into 5 subgroups for whom the incidence of G4L was observed in 0%, 14%, 35%, 70%, and 100% of patients. The benefit of PBT over IMRT was most pronounced in patients with an intermediate baseline absolute lymphocyte count and large planning treatment volume (P = .011). CONCLUSIONS: This is the first prospective evidence that limiting dose scatter by PBT significantly reduced the incidence of G4L, especially in the intermediate-risk patients. The implication of this immune-sparing effect of PBT, especially in the context of standard adjuvant immunotherapy, needs further examination in the current phase 3 randomized trials.

An ERK5-NRF2 Axis Mediates Senescence-Associated Stemness and Atherosclerosis.

BACKGROUND: ERK5 (extracellular signal-regulated kinase 5) is a dual kinase transcription factor containing an N-terminal kinase domain and a C-terminal transcriptional activation domain. Many ERK5 kinase inhibitors have been developed and tested to treat cancer and inflammatory diseases. However, recent data have raised questions about the role of the catalytic activity of ERK5 in proliferation and inflammation. We aimed to investigate how ERK5 reprograms myeloid cells to the proinflammatory senescent phenotype, subsequently leading to atherosclerosis. METHODS: A ERK5 S496A (dephosphorylation mimic) knock in (KI) mouse model was generated using CRISPR/Cas9 (clustered regularly interspaced short palindromic repeats/clustered regularly interspaced short palindromic repeat-associated 9), and atherosclerosis was characterized by hypercholesterolemia induction. The plaque phenotyping in homozygous ERK5 S496A KI and wild type (WT) mice was studied using imaging mass cytometry. Bone marrow-derived macrophages were isolated from hypercholesterolemic mice and characterized using RNA sequencing and functional in vitro approaches, including senescence, mitochondria reactive oxygen species, and inflammation assays, as well as by metabolic extracellular flux analysis. RESULTS: We show that atherosclerosis was inhibited in ERK5 S496A KI mice. Furthermore, ERK5 S496 phosphorylation mediates both senescence-associated secretory phenotype and senescence-associated stemness by upregulating AHR (aryl hydrocarbon receptor) in plaque and bone marrow-derived macrophages isolated from hypercholesterolemic mice. We also discovered that ERK5 S496 phosphorylation could induce NRF2 (NFE2-related factor 2) SUMOylation at a novel K518 site to inhibit NRF2 transcriptional activity without altering ERK5 catalytic activity and mediates oxidized LDL (low-density lipoprotein)-induced senescence-associated secretory phenotype. Specific ERK5 kinase inhibitors (AX15836 and XMD8-92) also inhibited ERK5 S496 phosphorylation, suggesting the involvement of ERK5 S496 phosphorylation in the anti-inflammatory effects of these ERK5 kinase inhibitors. CONCLUSIONS: We discovered a novel mechanism by which the macrophage ERK5-NRF2 axis develops a unique senescence-associated secretory phenotype/stemness phenotype by upregulating AHR to engender atherogenesis. The finding of senescence-associated stemness phenotype provides a molecular explanation to resolve the paradox of senescence in proliferative plaque by permitting myeloid cells to escape the senescence-induced cell cycle arrest during atherosclerosis formation.

Effect of scan duration on CT perfusion values in metastases from renal cell carcinoma.

Objective: CT perfusion (CTp) values are affected by CT scan acquisition duration (tacq); their reproducibility is adversely affected by uncertainty in their measurement. The objectives were to assess the effects of tacq on CTp parameter values in metastases from renal cell carcinoma (mRCC) in thoracic and abdominal locations. Materials and Methods: 131 CTp evaluations in 53 patients with mRCC were retrospectively analyzed by distributed parameter modeling to yield tissue blood flow (BF), blood volume (BV), mean transit time (MTT), permeability (PS), and also hepatic arterial perfusion (HAP) and hepatic arterial fraction (HAF) for liver metastases and normal liver, with tacq from 25 to 590 s. Penalized piecewise polynomial regression (SPLINE) characterized functional relationships between CTp parameters and acquisition duration, tacq. Evidence for time-invariance was evaluated for each parameter at multiple time points by conducting inference on the fitted derivative to assess its proximity to zero as a function of acquisition time. Equivalence testing was implemented with three levels of confidence (low (20%), moderate (70%), high (95%)). Results: Systematic and non-systematic variability was observed for CTp parameter values with limited tacq. All parameters in all locations approached increasing stability with increasing tacq. PS, HAP and HAF required longer acquisition times than BF, BV and MTT to attain comparable levels of stability. Stabilization tended to require longer acquisition in liver than other tissues. tacq=380 s was required to obtain at least moderate level of confidence for all parameters and organs. Conclusion: Increasing tacq yields increasingly more stable CT perfusion parameters, and thereby better reproducibility.

Optimal predictive probability designs for randomized biomarker-guided oncology trials.

Introduction: Efforts to develop biomarker-targeted anti-cancer therapies have progressed rapidly in recent years. With efforts to expedite regulatory reviews of promising therapies, several targeted cancer therapies have been granted accelerated approval on the basis of evidence acquired in single-arm phase II clinical trials. And yet, in the absence of randomization, patient prognosis for progression-free survival and overall survival may not have been studied under standard of care chemotherapies for emerging biomarker subpopulations prior to the submission of an accelerated approval application. Historical control rates used to design and evaluate emerging targeted therapies often arise as population averages, lacking specificity to the targeted genetic or immunophenotypic profile. Thus, historical trial results are inherently limited for inferring the potential "comparative efficacy" of novel targeted therapies. Consequently, randomization may be unavoidable in this setting. Innovations in design methodology are needed, however, to enable efficient implementation of randomized trials for agents that target biomarker subpopulations. Methods: This article proposes three randomized designs for early phase biomarker-guided oncology clinical trials. Each design utilizes the optimal efficiency predictive probability method to monitor multiple biomarker subpopulations for futility. Only designs with type I error between 0.05 and 0.1 and power of at least 0.8 were considered when selecting an optimal efficiency design from among the candidate designs formed by different combinations of posterior and predictive threshold. A simulation study motivated by the results reported in a recent clinical trial studying atezolizumab treatment in patients with locally advanced or metastatic urothelial carcinoma is used to evaluate the operating characteristics of the various designs. Results: Out of a maximum of 300 total patients, we find that the enrichment design has an average total sample size under the null of 101.0 and a total average sample size under the alternative of 218.0, as compared to 144.8 and 213.8 under the null and alternative, respectively, for the stratified control arm design. The pooled control arm design enrolled a total of 113.2 patients under the null and 159.6 under the alternative, out of a maximum of 200. These average sample sizes that are 23-48% smaller under the alternative and 47-64% smaller under the null, as compared to the realized sample size of 310 patients in the phase II study of atezolizumab. Discussion: Our findings suggest that potentially smaller phase II trials to those used in practice can be designed using randomization and futility stopping to efficiently obtain more information about both the treatment and control groups prior to phase III study.

Augmenting control arms with real-world data for cancer trials: Hybrid control arm methods and considerations.

Background: Hybrid controlled trials with real-world data (RWD), where the control arm is composed of both trial and real-world patients, could facilitate research when the feasibility of randomized controlled trials (RCTs) is challenging and single-arm trials would provide insufficient information. Methods: We propose a frequentist two-step borrowing method to construct hybrid control arms. We use parameters informed by a completed randomized trial in metastatic triple-negative breast cancer to simulate the operating characteristics of dynamic and static borrowing methods, highlighting key trade-offs and analytic decisions in the design of hybrid studies. Results: Simulated data were generated under varying residual-bias assumptions (no bias: HRRWD = 1) and experimental treatment effects (target trial scenario: HRExp = 0.78). Under the target scenario with no residual bias, all borrowing methods achieved the desired 88% power, an improvement over the reference model (74% power) that does not borrow information externally. The effective number of external events tended to decrease with higher bias between RWD and RCT (i.e. HRRWD away from 1), and with weaker experimental treatment effects (i.e. HRExp closer to 1). All dynamic borrowing methods illustrated (but not the static power prior) cap the maximum Type 1 error over the residual-bias range considered. Our two-step model achieved comparable results for power, type 1 error, and effective number of external events borrowed compared to other borrowing methodologies. Conclusion: By pairing high-quality external data with rigorous simulations, researchers have the potential to design hybrid controlled trials that better meet the needs of patients and drug development.

Severe Radiation-Induced Lymphopenia Attenuates the Benefit of Durvalumab After Concurrent Chemoradiotherapy for NSCLC.

Introduction: Durvalumab after concurrent chemoradiation (CCRT) for NSCLC improves survival, but only in a subset of patients. We investigated the effect of severe radiation-induced lymphopenia (sRIL) on survival in these patients. Methods: Outcomes after CCRT (2010-2019) or CCRT followed by durvalumab (2018-2019) were reviewed. RIL was defined by absolute lymphocyte count (ALC) nadir in samples collected at end of CCRT; sRIL was defined as nadir ALC less than 0.23 × 109/L (the lowest tertile). Progression-free survival (PFS) and overall survival (OS) were calculated by the Kaplan-Meier method. Cox proportional hazard modeling evaluated associations between clinical variables and survival. Results: Of 309 patients, 192 (62%) received CCRT only and 117 (38%) CCRT plus durvalumab. Multivariable logistic regression analysis indicated that sRIL was associated with planning target volume (OR = 1.002, p = 0.001), stage IIIB disease (OR = 2.77, p = 0.04), and baseline ALC (OR = 0.36, p < 0.01). Durvalumab extended median PFS (23.3 versus 14.1 mo, p = 0.003) and OS (not reached versus 30.8 mo, p < 0.01). sRIL predicted poorer PFS and OS in both treatment groups. Among patients with sRIL, durvalumab did not improve survival (median = 24.6 mo versus 18.1 mo CCRT only, p = 0.079). On multivariable analyses, sRIL (OR = 1.81, p < 0.01) independently predicted poor survival. Conclusions: Severe RIL compromises survival benefits from durvalumab after CCRT for NSCLC. Measures to mitigate RIL after CCRT may be warranted to enhance the benefit of consolidation durvalumab.

Basket Trials: Review of Current Practice and Innovations for Future Trials.

Advances in biology and immunology have elucidated genetic and immunologic origins of cancer. Innovations in sequencing technologies revealed that distinct cancer histologies shared common genetic and immune phenotypic traits. Pharmacologic developments made it possible to target these alterations, yielding novel classes of targeted agents whose therapeutic potential span multiple tumor types. Basket trials, one type of master protocol, emerged as a tool for evaluating biomarker-targeted therapies among multiple tumor histologies. Conventionally conducted within the phase II setting and designed to estimate high and durable objective responses, basket trials pose challenges to statistical design and interpretation of results. This article reviews basket trials implemented in oncology studies and discusses issues related to their statistical design and analysis.

Optimal Sequential Predictive Probability Designs for Early-Phase Oncology Expansion Cohorts.

PURPOSE: The customary approach to early-phase clinical trial design, where the focus is on identification of the maximum tolerated dose, is not always suitable for noncytotoxic or other targeted therapies. Many trials have continued to follow the 3 + 3 dose-escalation design, but with the addition of phase I dose-expansion cohorts to further characterize safety and assess efficacy. Dose-expansion cohorts are not always planned in advance nor rigorously designed. We introduce an approach to the design of phase I expansion cohorts on the basis of sequential predictive probability monitoring. METHODS: Two optimization criteria are proposed that allow investigators to stop for futility to preserve limited resources while maintaining traditional control of type I and type II errors. We demonstrate the use of these designs through simulation, and we elucidate their implementation with a redesign of the phase I expansion cohort for atezolizumab in metastatic urothelial carcinoma. RESULTS: A sequential predictive probability design outperforms Simon's two-stage designs and posterior probability monitoring with respect to both proposed optimization criteria. The Bayesian sequential predictive probability design yields increased power while significantly reducing the average sample size under the null hypothesis in the context of the case study, whereas the original study design yields too low type I error and power. The optimal efficiency design tended to have more desirable properties, subject to constraints on type I error and power, compared with the optimal accuracy design. CONCLUSION: The optimal efficiency design allows investigators to preserve limited financial resources and to maintain ethical standards by halting potentially large dose-expansion cohorts early in the absence of promising efficacy results, while maintaining traditional control of type I and II error rates.

The Impact of Radiation Dose to Heart Substructures on Major Coronary Events and Patient Survival after Chemoradiation Therapy for Esophageal Cancer.

Background: There is a paucity of data regarding the association between radiation exposure of heart substructures and the incidence of major coronary events (MCEs) in patients with esophageal cancer (ESOC) undergoing chemoradiation therapy. We studied radiation dosimetric determinants of MCE risk and measured their impact on patient prognosis using a cohort of ESOC patients treated at a single institution. Methods: Between March 2005 and October 2015, 355 ESOC patients treated with concurrent chemoradiotherapy were identified from a prospectively maintained and institutional-regulatory-board-approved clinical database. Dose-distribution parameters of the whole heart, the atria, the ventricles, the left main coronary artery, and three main coronary arteries were extracted for analysis. Results: Within a median follow-up time of 67 months, 14 patients experienced MCEs at a median of 16 months. The incidence of MCEs was significantly associated with the left anterior descending coronary artery (LAD) receiving ≥30 Gy (V30Gy) (p = 0.048). Patients receiving LAD V30Gy ≥ 10% of volume experienced a higher incidence of MCEs versus the LAD V30Gy < 10% group (p = 0.044). The relative rate of death increased with the left main coronary artery (LMA) mean dose (Gy) (p = 0.002). Furthermore, a mutual promotion effect of hyperlipidemia and RT on MCEs was observed. Conclusion: Radiation dose to coronary substructures is associated with MCEs and overall survival in patients with ESOC. In this study, the doses to these substructures appeared to be better predictors of toxicity outcomes than mean heart dose (MHD) or whole-heart V30Gy. These findings have implications for reducing coronary events through radiation therapy planning.

A hybrid deep learning model for forecasting lymphocyte depletion during radiation therapy.

PURPOSE: Recent studies have shown that severe depletion of the absolute lymphocyte count (ALC) induced by radiation therapy (RT) has been associated with poor overall survival of patients with many solid tumors. In this paper, we aimed to predict radiation-induced lymphocyte depletion in esophageal cancer patients during the course of RT based on patient characteristics and dosimetric features. METHODS: We proposed a hybrid deep learning model in a stacked structure to predict a trend toward ALC depletion based on the clinical information before or at the early stages of RT treatment. The proposed model consisted of four channels, one channel based on long short-term memory (LSTM) network and three channels based on neural networks, to process four categories of features followed by a dense layer to integrate the outputs of four channels and predict the weekly ALC values. Moreover, a discriminative kernel was developed to extract temporal features and assign different weights to each part of the input sequence that enabled the model to focus on the most relevant parts. The proposed model was trained and tested on a dataset of 860 esophageal cancer patients who received concurrent chemoradiotherapy. RESULTS: The performance of the proposed model was evaluated based on several important prediction metrics and compared to other commonly used prediction models. The results showed that the proposed model outperformed off-the-shelf prediction methods with at least a 30% reduction in the mean squared error (MSE) of weekly ALC predictions based on pretreatment data. Moreover, using an extended model based on augmented first-week treatment, data reduced the MSE of predictions by 70% compared to the model based on the pretreatment data. CONCLUSIONS: In conclusion, our model performed well in predicting radiation-induced lymphocyte depletion for RT treatment planning. The ability to predict ALC will enable physicians to evaluate individual RT treatment plans for lymphopenia risk and to identify patients at high risk who would benefit from modified treatment approaches.

Bayesian basket trial design with false-discovery rate control.

BACKGROUND: Recent advances in developing "tumor agnostic" oncology therapies have identified molecular targets that define patient subpopulations in a manner that supersedes conventional criteria for cancer classification. These successes have produced effective targeted therapies that are administered to patients regardless of their tumor histology. Trials have evolved as well with master protocol designs. By blending translational and clinical science, basket trials in particular are well-suited to investigate and develop targeted therapies among multiple cancer histologies. However, basket trials intrinsically involve more complex design decisions, including issues of multiple testing across baskets, and guidance for investigators is needed. METHODS: The sensitivity of the multisource exchangeability model to prior specification under differing degrees of response heterogeneity is explored through simulation. Then, a multisource exchangeability model design that incorporates control of the false-discovery rate is presented and a simulation study compares the operating characteristics to a design where the family-wise error rate is controlled and to the frequentist approach of treating the baskets as independent. Simulations are based on the original design of a real-world clinical trial, the SUMMIT trial, which investigated Neratinib treatment for a variety of solid tumors. The methods studied here are specific to single-arm phase II trials with binary outcomes. RESULTS: Values of prior probability of exchangeability in the multisource exchangeability model between 0.1 and 0.3 provide the best trade-offs between gain in precision and bias, especially when per-basket sample size is below 30. Application of these calibration results to a re-analysis of the SUMMIT trial showed that the breast basket exceeded the null response rate with posterior probability of 0.999 while having low posterior probability of exchangeability with all other baskets. Simulations based on the design of the SUMMIT trial revealed that there is meaningful improvement in power even in baskets with small sample size when the false-discovery rate is controlled as opposed to the family-wise error rate. For example, when only the breast basket was active, with a sample size of 25, the power was 0.76 when the false-discovery rate was controlled at 0.05 but only 0.56 when the family-wise error rate was controlled at 0.05, indicating that impractical sample sizes for the phase II setting would be needed to achieve acceptable power while controlling the family-wise error rate in this setting of a trial with 10 baskets. CONCLUSION: Selection of the prior exchangeability probability based on calibration and incorporation of false-discovery rate control result in multisource exchangeability model designs with high power to detect promising treatments in the context of phase II basket trials.

Predicting outcomes of phase III oncology trials with Bayesian mediation modeling of tumor response.

Pivotal cancer trials often fail to yield evidence in support of new therapies thought to offer promising alternatives to standards-of-care. Conducting randomized controlled trials in oncology tends to be considerably more expensive than studies of other diseases with comparable sample size. Moreover, phase III trial design often takes place with a paucity of survival data for experimental therapies. Experts have explained the failures on the basis of design flaws which produce studies with unrealistic expectations. This article presents a framework for predicting outcomes of phase III oncology trials using Bayesian mediation models. Predictions, which arise from interim analyses, derive from multivariate modeling of the relationships among treatment, tumor response, and their conjoint effects on survival. Acting as a safeguard against inaccurate pre-trial design assumptions, the methodology may better facilitate rapid closure of negative studies. Additionally the models can be used to inform re-estimations of sample size for under-powered trials that demonstrate survival benefit via tumor response mediation. The methods are applied to predict the outcomes of two colorectal cancer studies. Simulation is used to evaluate and compare models in the absence versus presence of reliable surrogate markers of survival.

Landmark mediation survival analysis using longitudinal surrogate.

Clinical cancer trials are designed to collect radiographic measurements of each patient's baseline and residual tumor burden at regular intervals over the course of study. For solid tumors, the extent of reduction in tumor size following treatment is used as a measure of a drug's antitumor activity. Statistical estimation of treatment efficacy routinely reduce the longitudinal assessment of tumor burden to a binary outcome describing the presence versus absence of an objective tumor response as defined by RECIST criteria. The objective response rate (ORR) is the predominate method for evaluating an experimental therapy in a single-arm trial. Additionally, ORR is routinely compared against a control therapy in phase III randomized controlled trials. The longitudinal assessments of tumor burden are seldom integrated into a formal statistical model, nor integrated into mediation analysis to characterize the relationships among treatment, residual tumor burden, and survival. This article presents a frameworkfor landmark mediation survival analyses devised to incorporate longitudinal assessment of tumor burden. R 2 effect-size measures are developed to quantify the survival treatment mediation effects using longitudinal predictors. Analyses are demonstrated with applications to two colorectal cancer trials. Survival prediction is compared in the presence versus absence of longitudinal analysis. Simulation studies elucidate settings wherein patterns of tumor burden dynamics require longitudinal analysis.

A group-sequential randomized trial design utilizing supplemental trial data.

Definitive clinical trials are resource intensive, often requiring a large number of participants over several years. One approach to improve the efficiency of clinical trials is to incorporate historical information into the primary trial analysis. This approach has tremendous potential in the areas of pediatric or rare disease trials, where achieving reasonable power is difficult. In this article, we introduce a novel Bayesian group-sequential trial design based on Multisource Exchangeability Models, which allows for dynamic borrowing of historical information at the interim analyses. Our approach achieves synergy between group sequential and adaptive borrowing methodology to attain improved power and reduced sample size. We explore the frequentist operating characteristics of our design through simulation and compare our method to a traditional group-sequential design. Our method achieves earlier stopping of the primary study while increasing power under the alternative hypothesis but has a potential for type I error inflation under some null scenarios. We discuss the issues of decision boundary determination, power and sample size calculations, and the issue of information accrual. We present our method for a continuous and binary outcome, as well as in a linear regression setting.

Identifying Individualized Risk Profiles for Radiotherapy-Induced Lymphopenia Among Patients With Esophageal Cancer Using Machine Learning.

PURPOSE: Radiotherapy (RT)-induced lymphopenia (RIL) is commonly associated with adverse clinical outcomes in patients with cancer. Using machine learning techniques, a retrospective study was conducted for patients with esophageal cancer treated with proton and photon therapies to characterize the principal pretreatment clinical and radiation dosimetric risk factors of grade 4 RIL (G4RIL) as well as to establish G4RIL risk profiles. METHODS: A single-institution retrospective data of 746 patients with esophageal cancer treated with photons (n = 500) and protons (n = 246) was reviewed. The primary end point of our study was G4RIL. Clustering techniques were applied to identify patient subpopulations with similar pretreatment clinical and radiation dosimetric characteristics. XGBoost was built on a training set (n = 499) to predict G4RIL risks. Predictive performance was assessed on the remaining n = 247 patients. SHapley Additive exPlanations were used to rank the importance of individual predictors. Counterfactual analyses compared patients' risk profiles assuming that they had switched modalities. RESULTS: Baseline absolute lymphocyte count and volumes of lung and spleen receiving ≥ 15 and ≥ 5 Gy, respectively, were the most important G4RIL risk determinants. The model achieved sensitivitytesting-set 0.798 and specificitytesting-set 0.667 with an area under the receiver operating characteristics curve (AUCtesting-set) of 0.783. The G4RIL risk for an average patient receiving protons increased by 19% had the patient switched to photons. Reductions in G4RIL risk were maximized with proton therapy for patients with older age, lower baseline absolute lymphocyte count, and higher lung and heart dose. CONCLUSION: G4RIL risk varies for individual patients with esophageal cancer and is modulated by radiotherapy dosimetric parameters. The framework for machine learning presented can be applied broadly to study risk determinants of other adverse events, providing the basis for adapting treatment strategies for mitigation.