array(41) {
  ["project_status"]=>
  string(7) "ongoing"
  ["project_assoc_trials"]=>
  array(1) {
    [0]=>
    object(WP_Post)#4668 (24) {
      ["ID"]=>
      int(19134)
      ["post_author"]=>
      string(4) "1885"
      ["post_date"]=>
      string(19) "2026-04-17 11:05:15"
      ["post_date_gmt"]=>
      string(19) "2026-04-17 15:05:15"
      ["post_content"]=>
      string(0) ""
      ["post_title"]=>
      string(235) "NCT03201965 - A Randomized Phase 3 Study to Evaluate the Efficacy and Safety of Daratumumab in Combination With Cyclophosphamide, Bortezomib and Dexamethasone (CyBorD) Compared to CyBorD Alone in Newly Diagnosed Systemic AL Amyloidosis"
      ["post_excerpt"]=>
      string(0) ""
      ["post_status"]=>
      string(7) "publish"
      ["comment_status"]=>
      string(6) "closed"
      ["ping_status"]=>
      string(6) "closed"
      ["post_password"]=>
      string(0) ""
      ["post_name"]=>
      string(195) "nct03201965-a-randomized-phase-3-study-to-evaluate-the-efficacy-and-safety-of-daratumumab-in-combination-with-cyclophosphamide-bortezomib-and-dexamethasone-cybord-compared-to-cybord-alone-in-newl"
      ["to_ping"]=>
      string(0) ""
      ["pinged"]=>
      string(0) ""
      ["post_modified"]=>
      string(19) "2026-06-30 12:19:12"
      ["post_modified_gmt"]=>
      string(19) "2026-06-30 16:19:12"
      ["post_content_filtered"]=>
      string(0) ""
      ["post_parent"]=>
      int(0)
      ["guid"]=>
      string(60) "https://yoda.yale.edu/?post_type=clinical_trial&p=19134"
      ["menu_order"]=>
      int(0)
      ["post_type"]=>
      string(14) "clinical_trial"
      ["post_mime_type"]=>
      string(0) ""
      ["comment_count"]=>
      string(1) "0"
      ["filter"]=>
      string(3) "raw"
    }
  }
  ["project_title"]=>
  string(132) "Development and External Validation of Transfer-Learning Cardiac Recovery Models in AL Amyloidosis Using the Phase 3 ANDROMEDA Trial"
  ["project_narrative_summary"]=>
  string(630) "Our proposed project will develop new tools to predict recovery of heart function in patients with light chain (AL) amyloidosis, a rare but devastating disorder with disparate survival outcomes in underrepresented populations and significant unmet needs. By improving our capacity to identify patients at highest risk for poor outcomes, this work will guide the design of more efficient, patient-centered studies and support more equitable access to clinical trials of therapeutics which effect heart function recovery. Ultimately, this project aims to accelerate the development of new treatments for diverse patient populations."
  ["project_learn_source"]=>
  string(10) "web_search"
  ["principal_investigator"]=>
  array(7) {
    ["first_name"]=>
    string(7) "Michael"
    ["last_name"]=>
    string(6) "Hughes"
    ["degree"]=>
    string(2) "MD"
    ["primary_affiliation"]=>
    string(41) "Columbia University Irving Medical Center"
    ["email"]=>
    string(24) "mh4266@cumc.columbia.edu"
    ["state_or_province"]=>
    string(2) "NY"
    ["country"]=>
    string(13) "United States"
  }
  ["project_key_personnel"]=>
  array(6) {
    [0]=>
    array(6) {
      ["p_pers_f_name"]=>
      string(10) "Siddhardha"
      ["p_pers_l_name"]=>
      string(5) "Nanda"
      ["p_pers_degree"]=>
      string(2) "MS"
      ["p_pers_pr_affil"]=>
      string(41) "Columbia University School of Engineering"
      ["p_pers_scop_id"]=>
      string(0) ""
      ["requires_data_access"]=>
      string(3) "yes"
    }
    [1]=>
    array(6) {
      ["p_pers_f_name"]=>
      string(8) "Jiacheng"
      ["p_pers_l_name"]=>
      string(4) "Zhou"
      ["p_pers_degree"]=>
      string(2) "MS"
      ["p_pers_pr_affil"]=>
      string(41) "Columbia University School of Engineering"
      ["p_pers_scop_id"]=>
      string(0) ""
      ["requires_data_access"]=>
      string(3) "yes"
    }
    [2]=>
    array(6) {
      ["p_pers_f_name"]=>
      string(5) "Grace"
      ["p_pers_l_name"]=>
      string(7) "Cropper"
      ["p_pers_degree"]=>
      string(2) "MS"
      ["p_pers_pr_affil"]=>
      string(41) "Columbia University School of Engineering"
      ["p_pers_scop_id"]=>
      string(0) ""
      ["requires_data_access"]=>
      string(3) "yes"
    }
    [3]=>
    array(6) {
      ["p_pers_f_name"]=>
      string(5) "Yiran"
      ["p_pers_l_name"]=>
      string(2) "Xu"
      ["p_pers_degree"]=>
      string(2) "BS"
      ["p_pers_pr_affil"]=>
      string(41) "Columbia University School of Engineering"
      ["p_pers_scop_id"]=>
      string(0) ""
      ["requires_data_access"]=>
      string(3) "yes"
    }
    [4]=>
    array(6) {
      ["p_pers_f_name"]=>
      string(6) "Jiayin"
      ["p_pers_l_name"]=>
      string(3) "Luo"
      ["p_pers_degree"]=>
      string(2) "MS"
      ["p_pers_pr_affil"]=>
      string(41) "Columbia University School of Engineering"
      ["p_pers_scop_id"]=>
      string(0) ""
      ["requires_data_access"]=>
      string(3) "yes"
    }
    [5]=>
    array(6) {
      ["p_pers_f_name"]=>
      string(5) "Divya"
      ["p_pers_l_name"]=>
      string(11) "Empranthiri"
      ["p_pers_degree"]=>
      string(2) "MS"
      ["p_pers_pr_affil"]=>
      string(41) "Columbia University School of Engineering"
      ["p_pers_scop_id"]=>
      string(0) ""
      ["requires_data_access"]=>
      string(3) "yes"
    }
  }
  ["project_ext_grants"]=>
  array(2) {
    ["value"]=>
    string(2) "no"
    ["label"]=>
    string(68) "No external grants or funds are being used to support this research."
  }
  ["project_funding_source"]=>
  string(47) "NIH/NCI Cancer Center Support Grant P30CA013696"
  ["project_date_type"]=>
  string(18) "full_crs_supp_docs"
  ["property_scientific_abstract"]=>
  string(1639) "Background: Although anti-plasma cell therapy improves AL amyloidosis outcomes, patients with cardiac involvement still face poor survival. Prognosis depends on cardiac recovery, but no models predict its probability over time. While machine learning (ML) models typically require large datasets for stable prediction, transfer learning may enable robust prediction in this rare disease.

Objective: To develop two ML models (“Static”, “Dynamic”) to predict cardiac recovery probabilities over 18 months with external validation. Static will use early-course data; Dynamic will use landmark-updated data.

Design: We will pretrain the models on clinical features from MIMIC-IV and INSPECT. We will transfer to ANDROMEDA data for final encoder layer retraining and validate locked models on the CUIMC AL amyloidosis dataset.

Participants: Pretraining- all eligible MIMIC-IV and INSPECT patients. Fine-tuning, validation- ANDROMEDA and CUIMC patients with cardiac AL amyloidosis, respectively.

Outcome Measures: Primary outcomes are 12- and 18-month cardiac responses. Primary outputs are patient-level probabilities of 12- and 18-month cardiac response by Static and Dynamic models, respectively. Performance endpoint is model accuracy by IPCW-Brier score. Secondary outcomes are 7-18 month cardiac responses and times to response.

Statistical Analysis: We will report logistic regression results for association of candidate clinical inputs with cardiac response. Model performance will be assessed with IPCW-Brier score, calibration, and discrimination metrics. There is no pooled data comparison."
  ["project_brief_bg"]=>
  string(3027) "	Light chain (AL) amyloidosis is a devastating condition in which clonal CD38+ plasma cells (PC) generate misfolded immunoglobulin light chain (LC) proteins. These misfolded LCs aggregate into amyloid fibrils and deposit into tissues, causing organ failure and death. While anti-CD38 agents have produced remarkable success in hematologic and organ response as well as survival,[1,2] PC elimination does not translate to rapid endogenous fibril clearance. Median time to any organ response takes months and best recovery takes years, resulting in substantial interval morbidity and mortality.[3–7] Reversal of fibril-related organ damage is the main determinant of survival and quality of life, especially in patients with cardiac involvement.[8] Despite the prognostic importance of heart function improvement, current AL amyloidosis risk models are still only able to predict survival, and cardiac recovery prediction is currently limited to a singular future timepoint.[9] Reliable, individualized estimation of cardiac recovery probability over time remains a major unmet need for prognostication.

	For outcome prediction, machine and deep learning (ML/DL) models typically require training on large datasets to ensure generalizable models and accurate conclusions.[10–15]  Applications of AI models to rare disorders are prone to significant error.[16] ML/DL models have been previously explored in AL amyloidosis; however, current model predictive capacity remains modest at best due to disease rarity.[9, 17–19] Transfer learning, a proven ML/DL method, may overcome such dataset size limitations.[20–22] By leveraging shared statistical structures rather than disease-specific features, it enables the creation of a “foundational” model pretrained on large source datasets unrelated to the target domain to accurately predict outcomes after fine-tuning with a disease-specific dataset.[23] This has not yet been applied to AL amyloidosis.

	An ML/DL foundational model, adapted for AL amyloidosis outcomes prediction via transfer learning, would enable patient-level prognostication and represent a significant step forward in the application of ML/DL predictive algorithms to rare diseases. We have developed ML/DL model components which predict future physiologic parameters and are ready for transfer learning. We propose the following:



AIM 1: To develop a “static” prediction model for patient-level cardiac recovery using data from diagnosis to 6 months of treatment.

AIM 2: To develop a “dynamic” prediction model for patient-level cardiac recovery using serial data inputs.

AIM 3: To validate the finalized models in an independent real-world AL amyloidosis cohort.

If achieved, our aims will not only transform AL amyloidosis counseling, possibly leading to future individualized treatment strategies adapted to granular cardiac recovery probability estimates over time; but also provide a novel framework for rare disease outcome prediction."
  ["project_specific_aims"]=>
  string(1649) "Aim 1: Develop a static model for patient-level cardiac recovery in AL amyloidosis using early-course data.

2 original ML model subunits pretrained on MIMIC-IV and INSPECT [24,25] will have final encoder layers fine-tuned on ANDROMEDA data. Patient latent representations will feed a discrete-time survival model to estimate cardiac recovery probabilities, 7–18 months, and probabilities will be unified. Primary output: predicted probability of cardiac response by 12 months. Primary performance endpoint: accuracy of predicted 12-month cardiac recovery probability by IPCW-Brier score.



Aim 2: Develop a dynamic model for patient-level cardiac recovery using serial data.

2 model subunits pretrained on time-aware MIMIC-IV and INSPECT data will be fine-tuned on longitudinal ANDROMEDA data. Updated patient representations at monthly intervals will drive a discrete-time survival model that refreshes recovery probabilities over time; subunits will be ensembled. Primary output: landmark-updated predicted probability of cardiac response by 18 months. Primary performance endpoint: improvement in 18-month cardiac recovery estimate accuracy by change in ICPW Brier score.



Aim 3: Validate the finalized models in an independent real-world AL amyloidosis cohort.

Using locked outputs from Aims 1–2, we will apply the static, dynamic, and benchmark XGBoost[9] models to the CUIMC cohort and compare probabilistic accuracy, calibration, and discrimination over time. Primary performance endpoint: external validation accuracy of predicted 18-month cardiac recovery probability by IPCW-Brier score."
  ["project_study_design"]=>
  array(2) {
    ["value"]=>
    string(8) "meth_res"
    ["label"]=>
    string(23) "Methodological research"
  }
  ["project_purposes"]=>
  array(1) {
    [0]=>
    array(2) {
      ["value"]=>
      string(50) "research_on_clinical_prediction_or_risk_prediction"
      ["label"]=>
      string(50) "Research on clinical prediction or risk prediction"
    }
  }
  ["project_research_methods"]=>
  string(2039) "Requested YODA study: phase 3 ANDROMEDA (NCT03201965), a randomized trial comparing daratumumab plus cyclophosphamide, bortezomib, and dexamethasone with CyBorD alone in newly diagnosed systemic AL amyloidosis.



Primary development cohort: randomized ANDROMEDA participants with baseline cardiac involvement, evaluable baseline clinical features needed for prespecified model inputs, and sufficient follow-up to determine observed cardiac response by the relevant prediction horizon, death before response, or censoring status. Participants without baseline cardiac involvement will be excluded from the primary cardiac-response model because the endpoint is not meaningfully defined for them, but they may be retained for descriptive, auxiliary, or non-cardiac exploratory analyses where appropriate. For dynamic-model analyses, participants must have longitudinal data available at one or more prespecified postbaseline landmark timepoints; each landmark analysis will use only information available up to that landmark.



Planned exclusion criteria are limited to absence of baseline cardiac involvement for the primary cardiac-response analysis, inability to compute essential prespecified model inputs after harmonization, absence of the primary cardiac-response endpoint or required censoring information, or lack of follow-up sufficient for outcome assessment.



Data outside YODA: MIMIC-IV and INSPECT [24,25] will be used for model pretraining and will not be pooled with ANDROMEDA individual participant data. An independent CUIMC AL amyloidosis cohort assembled under local IRB approval will be used for external validation of locked model specifications. No pooling of YODA and non-YODA participant-level data is planned. ANDROMEDA analyses will be conducted within the YODA-approved secure environment. External validation will be conducted only through an approved workflow consistent with YODA and institutional requirements, with aggregate validation results compared across cohorts."
  ["project_main_outcome_measure"]=>
  string(2032) "Primary outcome measures: Observed cardiac organ response by 12 and 18 months from randomization/treatment initiation. Cardiac response will be defined using the trial-recorded cardiac response variable if available. If not directly available, cardiac response will be derived from serial cardiac biomarker data using prespecified, established AL amyloidosis cardiac response criteria. The primary model output will be each participant’s predicted probability of observed 12-/18-month cardiac responses. The primary model-performance endpoint will be probabilistic accuracy of this predicted 12-/18-month response probability, assessed by inverse probability of censoring weighted (IPCW) Brier score.



Secondary outcome measures/endpoints: Secondary clinical outcomes will include observed cardiac organ response by 12, 15, and 18 months; time to first observed cardiac response; death before cardiac response, treated as a competing event where feasible; and all-cause mortality as an exploratory supportive endpoint. Secondary model outputs will include predicted cardiac response probabilities at prespecified horizons and landmark-updated predicted 18-month cardiac response probabilities generated by the dynamic model using only data available up to each landmark. Secondary performance endpoints will include calibration slope/intercept, observed-versus-predicted calibration plots, time-dependent AUC, Uno’s C-index, IPCW-Brier score at additional horizons, integrated Brier score across the prediction interval, and decision-curve analysis across prespecified cardiac-response probability thresholds.



No outcome redefinition based on model performance is planned. If variable availability requires derivation rather than direct use of a recorded cardiac response endpoint, the derivation algorithm will be documented before analysis and applied uniformly across development and validation datasets. Any deviations from these definitions in the final publication will be explicitly reported."
  ["project_main_predictor_indep"]=>
  string(2014) "This is a multivariable prognostic-model development and validation study rather than a single-exposure causal-effect study: there is no single exposure variable being tested for an independent effect. The main independent variables will consist of a prespecified set of candidate clinical features used as inputs to the Static and Dynamic cardiac-response prediction models. The primary model-derived predictor for validation analyses will be each patient’s predicted probability of cardiac response at prespecified horizons as above.



Candidate predictors [Supplement] will be selected from prespecified clinically relevant domains available in ANDROMEDA, based on clinical expertise, published organ-response criteria, prior literature, variable availability, missingness, and harmonizability with the external CUIMC validation cohort. Predictor domains will include demographics; baseline laboratory features (blood counts, serum chemistries); baseline cardiac involvement and biomarkers (NT-proBNP, troponin, Mayo stage, NYHA class, global longitudinal strain); renal function measures (creatinine, eGFR, proteinuria); hematologic disease burden (involved FLC, dFLC, bone marrow plasma cell percentage, M-protein); treatment assignment and exposure; and longitudinal landmark data (serial biomarkers/laboratory values available up to each prediction landmark). The final harmonized predictor set, coding rules, transformations, normalization procedures, missingness handling, and time-window definitions will be prespecified in the analysis plan before final model evaluation.



For the Static model, predictors will be restricted to baseline or early-course variables and treatment assignment where appropriate. For the Dynamic model, longitudinal predictors will include only data observed up to the relevant landmark timepoint, avoiding use of future information. The same coding and time-window rules will be applied during external validation in the independent CUIMC cohort."
  ["project_other_variables_interest"]=>
  string(2006) "Other variables will be used to characterize the study population, define follow-up, support outcome ascertainment, and conduct subgroup/sensitivity analyses. These variables will not be treated as post hoc predictors unless prespecified in the analysis plan.



Baseline descriptive variables may include age, sex, race/ethnicity where available, body mass index, ECOG performance status where available, AL amyloidosis diagnosis date, treatment initiation/randomization date, involved organ pattern, cardiac stage, renal involvement status, and baseline hematologic characteristics. Treatment-related descriptive variables will include randomized treatment arm, regimen received, dates of therapy, duration of exposure, dose modifications/delays, discontinuation, and subsequent therapy where available.



Outcome-related variables will include observed cardiac response status and timing, renal response status and timing where available, hematologic response status and timing, death, date of death, last follow-up, censoring indicators, and major organ deterioration/progression variables if available. Death before cardiac response will be handled as a competing event where feasible. Time-to-event variables will be defined relative to randomization/treatment initiation.



For dynamic landmark analyses, additional variables will include landmark timepoint, availability of serial laboratory/biomarker measurements up to each landmark, and missingness indicators or data-density summaries. Sensitivity analyses will stratify model performance by treatment arm, baseline cardiac stage, renal involvement, hematologic response depth, degree of missingness, and follow-up duration. All categorizations, time windows, censoring rules, competing-event handling, and subgroup definitions will be prespecified before final model evaluation and applied consistently across ANDROMEDA and the external CUIMC validation cohort where corresponding variables are available."
  ["project_stat_analysis_plan"]=>
  string(5022) "We will analyze ANDROMEDA participant-level data to develop, internally evaluate, and prepare for external validation of static and dynamic prognostic models for cardiac response in AL amyloidosis. All analyses will be conducted according to a prespecified analysis plan. The primary clinical outcome will be observed cardiac organ response by 12 (Aim 1) and 18 months (Aim 2) from randomization/treatment initiation. Cardiac response will use the trial-recorded response variable if available; otherwise, it will be derived from serial cardiac biomarkers using established AL amyloidosis response criteria. Death before cardiac response will be handled as a competing event where feasible.



Descriptive analyses: We will summarize baseline demographics, disease characteristics, organ involvement, cardiac biomarkers, renal function, hematologic disease burden, treatment assignment/exposure, follow-up, missingness, cardiac response, death, and censoring. Continuous variables will be reported as mean/SD or median/IQR, as appropriate; categorical variables will be reported as counts and percentages. Missingness patterns and longitudinal data density will be summarized overall, by treatment arm, and across landmark timepoints.



Bivariate analyses: Associations between candidate predictors and observed cardiac response will be explored using t tests or Wilcoxon rank-sum tests for continuous variables, chi-square or Fisher exact tests for categorical variables, and cumulative incidence/time-to-event methods for response outcomes. Univariable time-to-event models may be used to describe predictor-outcome relationships. These analyses will be descriptive and will not determine final model inclusion, which will be prespecified.



Multivariable/advanced analyses: Candidate predictors have been selected from prespecified clinically relevant domains [Supplement], including demographics, baseline laboratory features, cardiac involvement/biomarkers, renal function, hematologic disease burden, treatment assignment/exposure, and longitudinal landmark data. Two original ML models, Hybrid and mTimelyGPT, will be pretrained using MIMIC-IV and INSPECT to encode general inpatient and outpatient physiologic variability. Final encoder layers will be fine-tuned on ANDROMEDA. The Static model will use baseline/early-course variables to estimate cardiac response probabilities at prespecified monthly horizons from 7–18 months. The Dynamic model will update patient representations at monthly landmarks using only data available up to each landmark, thereby avoiding future-information leakage. Latent representations from each model will feed a discrete-time survival hazard module to generate predicted cardiac response probabilities over time. Censoring will be handled using inverse probability of censoring weighting and interval masking, as appropriate. A probability-generating modified XGBoost benchmark model will also be trained and internally evaluated on ANDROMEDA.



Model evaluation: The primary model output will be predicted probability of observed 12- (Aim 1) and 18-month (Aim 2) cardiac response. The primary performance metric will be inverse probability of censoring weighted (IPCW) Brier score for predicted cardiac response probabilities at 12 and 18 months. Secondary performance metrics will include IPCW-Brier score at additional horizons, integrated Brier score across the prediction interval, calibration slope/intercept, calibration-in-the-large, observed-versus-predicted calibration plots, time-dependent AUC, Uno’s C-index, and decision-curve analysis across prespecified probability thresholds. Threshold-based sensitivity, specificity, PPV, NPV, and balanced accuracy may be reported as secondary clinical-use analyses after applying prespecified thresholds. Internal validation will use cross-validation and/or bootstrap optimism correction, with penalization, early stopping, and nested tuning to reduce overfitting.



External validation: After model architecture, predictors, coding rules, transformations, missingness handling, time windows, hyperparameters, and probability thresholds are finalized using ANDROMEDA only, locked Static, Dynamic, and XGBoost models will be applied without retraining, feature selection, or threshold optimization to the independent CUIMC real-world AL amyloidosis cohort. External validation will assess probabilistic accuracy, calibration, and discrimination for observed 18-month cardiac response and secondary horizons, with sensitivity analyses by cardiac stage, treatment exposure, hematologic response depth, missingness/data density, and follow-up duration where sample size permits. Any limited post-validation recalibration, if needed, will be clearly labeled exploratory.



Of note, we select 18 months as the end-time horizon because most updated median follow-up time is reported at 15.7 months for cardiac AL amyloidosis patients in ANDROMEDA.[26]"
  ["project_software_used"]=>
  array(3) {
    [0]=>
    array(2) {
      ["value"]=>
      string(6) "python"
      ["label"]=>
      string(6) "Python"
    }
    [1]=>
    array(2) {
      ["value"]=>
      string(1) "r"
      ["label"]=>
      string(1) "R"
    }
    [2]=>
    array(2) {
      ["value"]=>
      string(7) "rstudio"
      ["label"]=>
      string(7) "RStudio"
    }
  }
  ["project_timeline"]=>
  string(1467) "Month 0–2 after data access approval: finalize the Data Use Agreement, complete required onboarding, review the protocol/statistical analysis plan/case report forms, build the ANDROMEDA variable crosswalk, finalize the prespecified analysis plan, and establish the approved workflow for external validation using deidentified CUIMC data. Months 3–5: clean and harmonize ANDROMEDA data, derive analysis variables and cardiac-response endpoints, summarize missingness and data density, and complete Static model fine-tuning with internal cross-validation. Months 6–8: complete Dynamic model development, landmark-based prediction analyses, internal model-performance assessment, benchmark XGBoost model development, and prespecified sensitivity analyses. Months 9–10: apply locked Static, Dynamic, and XGBoost models to the independent CUIMC cohort, complete external validation analyses, and perform any clearly labeled exploratory recalibration analyses if substantial calibration drift is observed. Month 11: draft the main manuscript, tables, figures, and supplementary methods; circulate to coauthors for internal review. Month 12: submit the manuscript for peer-reviewed publication and return a results summary and project status update to the YODA Project. If additional time is required for journal-requested confirmatory analyses, external-validation workflow delays, or manuscript revision, an extension will be requested consistent with YODA policy."
  ["project_dissemination_plan"]=>
  string(1531) "The primary product will be a full-length peer-reviewed manuscript reporting ANDROMEDA-based development and internal validation of static and dynamic cardiac-response prediction models in AL amyloidosis, together with independent external validation in a real-world cohort. A second methods-focused manuscript may follow if analyses of external transportability, recalibration, dynamic updating, or missingness in irregular real-world data yield distinct lessons for rare-disease prognostic modeling.



Target audiences include hematologists/oncologists, amyloidosis specialists, cardiologists caring for patients with cardiac amyloidosis, clinical trialists, biostatisticians, and investigators developing machine-learning models for rare diseases. Potential journals for the primary manuscript include Blood, Journal of Clinical Oncology, Leukemia, Blood Advances, NPJ Digital Medicine, and JCO Clinical Cancer Informatics, depending on the final clinical versus methodological emphasis. Results will also be considered for presentation at ASH, ASCO, the International Society of Amyloidosis primary meetings, ISBI, and/or MICCAI.



Consistent with YODA policies and the data use agreement, we will report study results to the YODA Project and submit findings for publication regardless of whether model performance is favorable. We will share analytic code, model specifications, variable definitions, and evaluation scripts to the extent permitted, without releasing participant-level ANDROMEDA data."
  ["project_bibliography"]=>
  string(4642) "

Kastritis E, Palladini G, Minnema MC, et al: Daratumumab-Based Treatment for Immunoglobulin Light-Chain Amyloidosis. N Engl J Med 385:46–58, 2021
Chakraborty R, Bhutani D, Mapara M, et al: Reduced early mortality with Daratumumab-based frontline therapy in AL amyloidosis: A retrospective cohort study. Am J Hematol 99:477–479, 2024
Kumar S, Dispenzieri A, Lacy MQ, et al: Revised prognostic staging system for light chain amyloidosis incorporating cardiac biomarkers and serum free light chain measurements. J Clin Oncol 30:989–95, 2012
Staron A, Zheng L, Doros G, et al: Marked progress in AL amyloidosis survival: a 40-year longitudinal natural history study. Blood Cancer J 11:139, 2021
Kaufman GP, Dispenzieri A, Gertz MA, et al: Kinetics of organ response and survival following normalization of the serum free light chain ratio in AL amyloidosis. Am J Hematol 90:181–6, 2015
Muchtar E, Dispenzieri A, Leung N, et al: Depth of organ response in AL amyloidosis is associated with improved survival: grading the organ response criteria. Leukemia 32:2240–2249, 2018
Sanchorawala V: Systemic Light Chain Amyloidosis. N Engl J Med 390:2295–2307, 2024
Muchtar E, Geyer S, Merlini G, et al: Patients with a cardiac complete response in AL amyloidosis have survival rates similar to those of a matched general population. Blood 144:790–793, 2024
Hanna J, Zhao R, Anwer F, et al: Predicting Cardiac and Renal Response in Light Chain (AL) Amyloidosis Using an Ensemble Machine Learning Model. Hemasphere 9:977–978, 2025
LeCun Y, Bengio Y, Hinton G: Deep learning. Nature 521:436–444, 2015
Ramamoorthy D, Severson K, Ghosh S, et al: Identifying patterns in amyotrophic lateral sclerosis progression from sparse longitudinal data. Nat Comput Sci 2:605–616, 2022
Kwong GA, Ghosh S, Gamboa L, et al: Synthetic biomarkers: a twenty-first century path to early cancer detection. Nat Rev Cancer 21:655–668, 2021
Naik SN, Angelini ED, Barr RG, et al: Unsupervised Airway Tree Clustering with Deep Learning: The Multi-Ethnic Study of Atherosclerosis (Mesa) Lung Study. Proc IEEE Int Symp Biomed Imaging 2024, 2024
Singh M, Kumar A, Khanna NN, et al: Artificial intelligence for cardiovascular disease risk assessment in personalised framework: a scoping review. EClinicalMedicine 73:102660, 2024
McDermott M, Yan T, Naumann T, et al: Semi-supervised biomedical translation with cycle wasserstein regression GANs, in 2018
D’Amico S, Dall’Olio L, Rollo C, et al: MOSAIC: An Artificial Intelligence-Based Framework for Multimodal Analysis, Classification, and Personalized Prognostic Assessment in Rare Cancers. JCO Clin Cancer Inform 8:e2400008, 2024
Goto S, Mahara K, Beussink-Nelson L, et al: Artificial intelligence-enabled fully automated detection of cardiac amyloidosis using electrocardiograms and echocardiograms. Nat Commun 12:2726, 2021
Palstrom NB, Rojek AM, Moller HEH, et al: Classification of Amyloidosis by Model-Assisted Mass Spectrometry-Based Proteomics. Int J Mol Sci 23, 2021
Anand SK, Staron A, Mendelson LM, et al: Machine-learning based subgroups of AL amyloidosis and cumulative incidence of mortality and end stage kidney disease. Am J Hematol , 2024
Kowald A, Barrantes I, Moller S, et al: Transfer learning of clinical outcomes from preclinical molecular data, principles and perspectives. Brief Bioinform 23, 2022
Topkara VK, Elias P, Jain R, et al: Machine Learning-Based Prediction of Myocardial Recovery in Patients With Left Ventricular Assist Device Support. Circulation: Heart Failure 15:e008711, 2022
Yang PC, Jha A, Xu W, et al: Cloud-Based Machine Learning Platform to Predict Clinical Outcomes at Home for Patients With Cardiovascular Conditions Discharged From Hospital: Clinical Trial. JMIR Cardio 8:e45130, 2024
Pan SJ, Yang Q: A Survey on Transfer Learning. IEEE Transactions on Knowledge and Data Engineering 22:1345–1359, 2010
Johnson AEW, Bulgarelli L, Shen L, et al: MIMIC-IV, a freely accessible electronic health record dataset. Scientific Data 10:1, 2023
Huang S-C, Huo Z, Steinberg E, et al: INSPECT: A Multimodal Dataset for Pulmonary Embolism Diagnosis and Prognosis [Internet], 2023[cited 2025 Sept 15] Available from: http://arxiv.org/abs/2311.10798
Minnema MC, Dispenzieri A, Merlini G, et al: Outcomes by Cardiac Stage in Patients With Newly Diagnosed AL Amyloidosis: Phase 3 ANDROMEDA Trial. JACC CardioOncol 4:474–487, 2022

"
  ["project_suppl_material"]=>
  array(1) {
    [0]=>
    array(1) {
      ["suppl_file"]=>
      array(21) {
        ["ID"]=>
        int(19209)
        ["id"]=>
        int(19209)
        ["title"]=>
        string(44) "Candidate-clinical-feature-input-listPDF.pdf"
        ["filename"]=>
        string(44) "Candidate-clinical-feature-input-listPDF.pdf"
        ["filesize"]=>
        int(69973)
        ["url"]=>
        string(93) "https://yoda.yale.edu/wp-content/uploads/2026/05/Candidate-clinical-feature-input-listPDF.pdf"
        ["link"]=>
        string(90) "https://yoda.yale.edu/data-request/2026-0284/candidate-clinical-feature-input-listpdf-pdf/"
        ["alt"]=>
        string(0) ""
        ["author"]=>
        string(4) "2059"
        ["description"]=>
        string(0) ""
        ["caption"]=>
        string(0) ""
        ["name"]=>
        string(44) "candidate-clinical-feature-input-listpdf-pdf"
        ["status"]=>
        string(7) "inherit"
        ["uploaded_to"]=>
        int(19189)
        ["date"]=>
        string(19) "2026-05-02 19:24:51"
        ["modified"]=>
        string(19) "2026-05-02 19:24:53"
        ["menu_order"]=>
        int(0)
        ["mime_type"]=>
        string(15) "application/pdf"
        ["type"]=>
        string(11) "application"
        ["subtype"]=>
        string(3) "pdf"
        ["icon"]=>
        string(62) "https://yoda.yale.edu/wp/wp-includes/images/media/document.png"
      }
    }
  }
  ["project_coi"]=>
  array(7) {
    [0]=>
    array(1) {
      ["file_coi"]=>
      array(21) {
        ["ID"]=>
        int(19210)
        ["id"]=>
        int(19210)
        ["title"]=>
        string(40) "SV_57KskaKADT3U9Aq-R_1MssvBIB8ljDoCB.pdf"
        ["filename"]=>
        string(40) "SV_57KskaKADT3U9Aq-R_1MssvBIB8ljDoCB.pdf"
        ["filesize"]=>
        int(19419)
        ["url"]=>
        string(89) "https://yoda.yale.edu/wp-content/uploads/2026/05/SV_57KskaKADT3U9Aq-R_1MssvBIB8ljDoCB.pdf"
        ["link"]=>
        string(86) "https://yoda.yale.edu/data-request/2026-0284/sv_57kskakadt3u9aq-r_1mssvbib8ljdocb-pdf/"
        ["alt"]=>
        string(0) ""
        ["author"]=>
        string(4) "2059"
        ["description"]=>
        string(0) ""
        ["caption"]=>
        string(0) ""
        ["name"]=>
        string(40) "sv_57kskakadt3u9aq-r_1mssvbib8ljdocb-pdf"
        ["status"]=>
        string(7) "inherit"
        ["uploaded_to"]=>
        int(19189)
        ["date"]=>
        string(19) "2026-05-02 19:36:35"
        ["modified"]=>
        string(19) "2026-05-02 19:36:37"
        ["menu_order"]=>
        int(0)
        ["mime_type"]=>
        string(15) "application/pdf"
        ["type"]=>
        string(11) "application"
        ["subtype"]=>
        string(3) "pdf"
        ["icon"]=>
        string(62) "https://yoda.yale.edu/wp/wp-includes/images/media/document.png"
      }
    }
    [1]=>
    array(1) {
      ["file_coi"]=>
      array(21) {
        ["ID"]=>
        int(19211)
        ["id"]=>
        int(19211)
        ["title"]=>
        string(18) "COI_form_Yiran.pdf"
        ["filename"]=>
        string(18) "COI_form_Yiran.pdf"
        ["filesize"]=>
        int(18302)
        ["url"]=>
        string(67) "https://yoda.yale.edu/wp-content/uploads/2026/05/COI_form_Yiran.pdf"
        ["link"]=>
        string(64) "https://yoda.yale.edu/data-request/2026-0284/coi_form_yiran-pdf/"
        ["alt"]=>
        string(0) ""
        ["author"]=>
        string(4) "2059"
        ["description"]=>
        string(0) ""
        ["caption"]=>
        string(0) ""
        ["name"]=>
        string(18) "coi_form_yiran-pdf"
        ["status"]=>
        string(7) "inherit"
        ["uploaded_to"]=>
        int(19189)
        ["date"]=>
        string(19) "2026-05-02 20:01:39"
        ["modified"]=>
        string(19) "2026-05-02 20:01:41"
        ["menu_order"]=>
        int(0)
        ["mime_type"]=>
        string(15) "application/pdf"
        ["type"]=>
        string(11) "application"
        ["subtype"]=>
        string(3) "pdf"
        ["icon"]=>
        string(62) "https://yoda.yale.edu/wp/wp-includes/images/media/document.png"
      }
    }
    [2]=>
    array(1) {
      ["file_coi"]=>
      array(21) {
        ["ID"]=>
        int(19212)
        ["id"]=>
        int(19212)
        ["title"]=>
        string(7) "COI.pdf"
        ["filename"]=>
        string(7) "COI.pdf"
        ["filesize"]=>
        int(18579)
        ["url"]=>
        string(56) "https://yoda.yale.edu/wp-content/uploads/2026/05/COI.pdf"
        ["link"]=>
        string(56) "https://yoda.yale.edu/data-request/2026-0284/coi-pdf-16/"
        ["alt"]=>
        string(0) ""
        ["author"]=>
        string(4) "2059"
        ["description"]=>
        string(0) ""
        ["caption"]=>
        string(0) ""
        ["name"]=>
        string(10) "coi-pdf-16"
        ["status"]=>
        string(7) "inherit"
        ["uploaded_to"]=>
        int(19189)
        ["date"]=>
        string(19) "2026-05-02 20:02:20"
        ["modified"]=>
        string(19) "2026-05-02 20:02:21"
        ["menu_order"]=>
        int(0)
        ["mime_type"]=>
        string(15) "application/pdf"
        ["type"]=>
        string(11) "application"
        ["subtype"]=>
        string(3) "pdf"
        ["icon"]=>
        string(62) "https://yoda.yale.edu/wp/wp-includes/images/media/document.png"
      }
    }
    [3]=>
    array(1) {
      ["file_coi"]=>
      array(21) {
        ["ID"]=>
        int(19213)
        ["id"]=>
        int(19213)
        ["title"]=>
        string(25) "COI_form_gracecropper.pdf"
        ["filename"]=>
        string(25) "COI_form_gracecropper.pdf"
        ["filesize"]=>
        int(18440)
        ["url"]=>
        string(74) "https://yoda.yale.edu/wp-content/uploads/2026/05/COI_form_gracecropper.pdf"
        ["link"]=>
        string(71) "https://yoda.yale.edu/data-request/2026-0284/coi_form_gracecropper-pdf/"
        ["alt"]=>
        string(0) ""
        ["author"]=>
        string(4) "2059"
        ["description"]=>
        string(0) ""
        ["caption"]=>
        string(0) ""
        ["name"]=>
        string(25) "coi_form_gracecropper-pdf"
        ["status"]=>
        string(7) "inherit"
        ["uploaded_to"]=>
        int(19189)
        ["date"]=>
        string(19) "2026-05-02 20:44:46"
        ["modified"]=>
        string(19) "2026-05-02 20:44:48"
        ["menu_order"]=>
        int(0)
        ["mime_type"]=>
        string(15) "application/pdf"
        ["type"]=>
        string(11) "application"
        ["subtype"]=>
        string(3) "pdf"
        ["icon"]=>
        string(62) "https://yoda.yale.edu/wp/wp-includes/images/media/document.png"
      }
    }
    [4]=>
    array(1) {
      ["file_coi"]=>
      array(21) {
        ["ID"]=>
        int(19214)
        ["id"]=>
        int(19214)
        ["title"]=>
        string(11) "COI-pdf.pdf"
        ["filename"]=>
        string(11) "COI-pdf.pdf"
        ["filesize"]=>
        int(18349)
        ["url"]=>
        string(60) "https://yoda.yale.edu/wp-content/uploads/2026/05/COI-pdf.pdf"
        ["link"]=>
        string(57) "https://yoda.yale.edu/data-request/2026-0284/coi-pdf-pdf/"
        ["alt"]=>
        string(0) ""
        ["author"]=>
        string(4) "2059"
        ["description"]=>
        string(0) ""
        ["caption"]=>
        string(0) ""
        ["name"]=>
        string(11) "coi-pdf-pdf"
        ["status"]=>
        string(7) "inherit"
        ["uploaded_to"]=>
        int(19189)
        ["date"]=>
        string(19) "2026-05-03 16:26:37"
        ["modified"]=>
        string(19) "2026-05-03 16:26:39"
        ["menu_order"]=>
        int(0)
        ["mime_type"]=>
        string(15) "application/pdf"
        ["type"]=>
        string(11) "application"
        ["subtype"]=>
        string(3) "pdf"
        ["icon"]=>
        string(62) "https://yoda.yale.edu/wp/wp-includes/images/media/document.png"
      }
    }
    [5]=>
    array(1) {
      ["file_coi"]=>
      array(21) {
        ["ID"]=>
        int(19224)
        ["id"]=>
        int(19224)
        ["title"]=>
        string(27) "COI-Joyce-Jiacheng-Zhou.pdf"
        ["filename"]=>
        string(27) "COI-Joyce-Jiacheng-Zhou.pdf"
        ["filesize"]=>
        int(37808)
        ["url"]=>
        string(76) "https://yoda.yale.edu/wp-content/uploads/2026/05/COI-Joyce-Jiacheng-Zhou.pdf"
        ["link"]=>
        string(73) "https://yoda.yale.edu/data-request/2026-0284/coi-joyce-jiacheng-zhou-pdf/"
        ["alt"]=>
        string(0) ""
        ["author"]=>
        string(4) "2059"
        ["description"]=>
        string(0) ""
        ["caption"]=>
        string(0) ""
        ["name"]=>
        string(27) "coi-joyce-jiacheng-zhou-pdf"
        ["status"]=>
        string(7) "inherit"
        ["uploaded_to"]=>
        int(19189)
        ["date"]=>
        string(19) "2026-05-04 18:49:17"
        ["modified"]=>
        string(19) "2026-05-04 18:49:20"
        ["menu_order"]=>
        int(0)
        ["mime_type"]=>
        string(15) "application/pdf"
        ["type"]=>
        string(11) "application"
        ["subtype"]=>
        string(3) "pdf"
        ["icon"]=>
        string(62) "https://yoda.yale.edu/wp/wp-includes/images/media/document.png"
      }
    }
    [6]=>
    array(1) {
      ["file_coi"]=>
      array(21) {
        ["ID"]=>
        int(19226)
        ["id"]=>
        int(19226)
        ["title"]=>
        string(13) "divya-coi.pdf"
        ["filename"]=>
        string(13) "divya-coi.pdf"
        ["filesize"]=>
        int(37000)
        ["url"]=>
        string(62) "https://yoda.yale.edu/wp-content/uploads/2026/05/divya-coi.pdf"
        ["link"]=>
        string(59) "https://yoda.yale.edu/data-request/2026-0284/divya-coi-pdf/"
        ["alt"]=>
        string(0) ""
        ["author"]=>
        string(4) "2059"
        ["description"]=>
        string(0) ""
        ["caption"]=>
        string(0) ""
        ["name"]=>
        string(13) "divya-coi-pdf"
        ["status"]=>
        string(7) "inherit"
        ["uploaded_to"]=>
        int(19189)
        ["date"]=>
        string(19) "2026-05-04 19:22:16"
        ["modified"]=>
        string(19) "2026-05-04 19:22:17"
        ["menu_order"]=>
        int(0)
        ["mime_type"]=>
        string(15) "application/pdf"
        ["type"]=>
        string(11) "application"
        ["subtype"]=>
        string(3) "pdf"
        ["icon"]=>
        string(62) "https://yoda.yale.edu/wp/wp-includes/images/media/document.png"
      }
    }
  }
  ["data_use_agreement_training"]=>
  bool(true)
  ["human_research_protection_training"]=>
  bool(true)
  ["certification"]=>
  bool(true)
  ["search_order"]=>
  string(1) "0"
  ["project_send_email_updates"]=>
  bool(false)
  ["project_publ_available"]=>
  bool(true)
  ["project_year_access"]=>
  string(4) "2026"
  ["project_rep_publ"]=>
  bool(false)
  ["project_assoc_data"]=>
  array(0) {
  }
  ["project_due_dil_assessment"]=>
  array(21) {
    ["ID"]=>
    int(19541)
    ["id"]=>
    int(19541)
    ["title"]=>
    string(47) "YODA Project Due Diligence Assessment 2026-0284"
    ["filename"]=>
    string(51) "YODA-Project-Due-Diligence-Assessment-2026-0284.pdf"
    ["filesize"]=>
    int(107544)
    ["url"]=>
    string(100) "https://yoda.yale.edu/wp-content/uploads/2026/04/YODA-Project-Due-Diligence-Assessment-2026-0284.pdf"
    ["link"]=>
    string(93) "https://yoda.yale.edu/data-request/2026-0284/yoda-project-due-diligence-assessment-2026-0284/"
    ["alt"]=>
    string(0) ""
    ["author"]=>
    string(4) "1885"
    ["description"]=>
    string(0) ""
    ["caption"]=>
    string(0) ""
    ["name"]=>
    string(47) "yoda-project-due-diligence-assessment-2026-0284"
    ["status"]=>
    string(7) "inherit"
    ["uploaded_to"]=>
    int(19189)
    ["date"]=>
    string(19) "2026-06-30 15:31:13"
    ["modified"]=>
    string(19) "2026-06-30 15:31:13"
    ["menu_order"]=>
    int(0)
    ["mime_type"]=>
    string(15) "application/pdf"
    ["type"]=>
    string(11) "application"
    ["subtype"]=>
    string(3) "pdf"
    ["icon"]=>
    string(62) "https://yoda.yale.edu/wp/wp-includes/images/media/document.png"
  }
  ["project_title_link"]=>
  array(21) {
    ["ID"]=>
    int(19542)
    ["id"]=>
    int(19542)
    ["title"]=>
    string(46) "YODA Project Protocol - 2026-0284 - 2026-05-06"
    ["filename"]=>
    string(46) "YODA-Project-Protocol-2026-0284-2026-05-06.pdf"
    ["filesize"]=>
    int(141360)
    ["url"]=>
    string(95) "https://yoda.yale.edu/wp-content/uploads/2026/04/YODA-Project-Protocol-2026-0284-2026-05-06.pdf"
    ["link"]=>
    string(88) "https://yoda.yale.edu/data-request/2026-0284/yoda-project-protocol-2026-0284-2026-05-06/"
    ["alt"]=>
    string(0) ""
    ["author"]=>
    string(4) "1885"
    ["description"]=>
    string(0) ""
    ["caption"]=>
    string(0) ""
    ["name"]=>
    string(42) "yoda-project-protocol-2026-0284-2026-05-06"
    ["status"]=>
    string(7) "inherit"
    ["uploaded_to"]=>
    int(19189)
    ["date"]=>
    string(19) "2026-06-30 15:31:57"
    ["modified"]=>
    string(19) "2026-06-30 15:31:57"
    ["menu_order"]=>
    int(0)
    ["mime_type"]=>
    string(15) "application/pdf"
    ["type"]=>
    string(11) "application"
    ["subtype"]=>
    string(3) "pdf"
    ["icon"]=>
    string(62) "https://yoda.yale.edu/wp/wp-includes/images/media/document.png"
  }
  ["project_review_link"]=>
  array(21) {
    ["ID"]=>
    int(19543)
    ["id"]=>
    int(19543)
    ["title"]=>
    string(36) "YODA Project Review - 2026-0284_site"
    ["filename"]=>
    string(38) "YODA-Project-Review-2026-0284_site.pdf"
    ["filesize"]=>
    int(1074880)
    ["url"]=>
    string(87) "https://yoda.yale.edu/wp-content/uploads/2026/04/YODA-Project-Review-2026-0284_site.pdf"
    ["link"]=>
    string(80) "https://yoda.yale.edu/data-request/2026-0284/yoda-project-review-2026-0284_site/"
    ["alt"]=>
    string(0) ""
    ["author"]=>
    string(4) "1885"
    ["description"]=>
    string(0) ""
    ["caption"]=>
    string(0) ""
    ["name"]=>
    string(34) "yoda-project-review-2026-0284_site"
    ["status"]=>
    string(7) "inherit"
    ["uploaded_to"]=>
    int(19189)
    ["date"]=>
    string(19) "2026-06-30 15:32:20"
    ["modified"]=>
    string(19) "2026-06-30 15:32:20"
    ["menu_order"]=>
    int(0)
    ["mime_type"]=>
    string(15) "application/pdf"
    ["type"]=>
    string(11) "application"
    ["subtype"]=>
    string(3) "pdf"
    ["icon"]=>
    string(62) "https://yoda.yale.edu/wp/wp-includes/images/media/document.png"
  }
  ["project_highlight_button"]=>
  string(0) ""
  ["request_overridden_res"]=>
  string(1) "3"
  ["request_data_partner"]=>
  string(15) "johnson-johnson"
}
data partner
array(1) {
  [0]=>
  string(15) "johnson-johnson"
}


pi country
array(0) {
}


pi affil
array(0) {
}


products
array(1) {
  [0]=>
  string(8) "darzalex"
}


num of trials
array(1) {
  [0]=>
  string(1) "1"
}


res
array(1) {
  [0]=>
  string(1) "3"
}