array(41) {
  ["request_overridden_res"]=>
  string(1) "3"
  ["project_status"]=>
  string(30) "approved_pending_dua_signature"
  ["project_assoc_trials"]=>
  array(1) {
    [0]=>
    object(WP_Post)#4549 (24) {
      ["ID"]=>
      int(1265)
      ["post_author"]=>
      string(4) "1363"
      ["post_date"]=>
      string(19) "2014-10-20 16:16:00"
      ["post_date_gmt"]=>
      string(19) "2014-10-20 16:16:00"
      ["post_content"]=>
      string(0) ""
      ["post_title"]=>
      string(284) "NCT01081834 - A Randomized, Double-Blind, Placebo-Controlled, Parallel-Group, Multicenter Study to Evaluate the Efficacy, Safety, and Tolerability of Canagliflozin as Monotherapy in the Treatment of Subjects With Type 2 Diabetes Mellitus Inadequately Controlled With Diet and Exercise"
      ["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(192) "nct01081834-a-randomized-double-blind-placebo-controlled-parallel-group-multicenter-study-to-evaluate-the-efficacy-safety-and-tolerability-of-canagliflozin-as-monotherapy-in-the-treatment-of-s"
      ["to_ping"]=>
      string(0) ""
      ["pinged"]=>
      string(0) ""
      ["post_modified"]=>
      string(19) "2025-10-24 15:48:34"
      ["post_modified_gmt"]=>
      string(19) "2025-10-24 19:48:34"
      ["post_content_filtered"]=>
      string(0) ""
      ["post_parent"]=>
      int(0)
      ["guid"]=>
      string(241) "https://dev-yoda.pantheonsite.io/clinical-trial/nct01081834-a-randomized-double-blind-placebo-controlled-parallel-group-multicenter-study-to-evaluate-the-efficacy-safety-and-tolerability-of-canagliflozin-as-monotherapy-in-the-treatment-of-s/"
      ["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(149) "Digital Twins Accelerate Progress in Randomized Controlled Trials of Type 2 Diabetes Mellitus With Inadequate Glycemic Control with diet and exercise"
  ["project_narrative_summary"]=>
  string(844) "Type 2 diabetes mellitus remains a leading public health challenge. New drug approval relies on randomized controlled trials. But trials are costly, slow, and face recruitment difficulties, with nearly 80% experiencing delays. Reducing sample size without compromising statistical power is a key priority. We propose using historical data from placebo-controlled trials of patients with inadequate control with diet and exercise. Using only placebo arm data, we will train an AI model to generate an individualized prognostic score predicting each patient's glycemic trajectory under placebo. In a new trial, this score will be incorporated as a covariate in the primary analysis. This adjustment reduces outcome variability, enhances precision, and enables substantial sample size reduction while maintaining power to detect treatment effects."
  ["project_learn_source"]=>
  string(4) "conf"
  ["principal_investigator"]=>
  array(7) {
    ["first_name"]=>
    string(5) "Qiong"
    ["last_name"]=>
    string(3) "Tan"
    ["degree"]=>
    string(3) "MPH"
    ["primary_affiliation"]=>
    string(31) "China Pharmaceutical University"
    ["email"]=>
    string(17) "nictee789@163.com"
    ["state_or_province"]=>
    string(7) "Jiangsu"
    ["country"]=>
    string(5) "China"
  }
  ["project_key_personnel"]=>
  array(1) {
    [0]=>
    array(6) {
      ["p_pers_f_name"]=>
      string(8) "Fangrong"
      ["p_pers_l_name"]=>
      string(3) "Yan"
      ["p_pers_degree"]=>
      string(3) "PhD"
      ["p_pers_pr_affil"]=>
      string(31) "China Pharmaceutical University"
      ["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_date_type"]=>
  string(18) "full_crs_supp_docs"
  ["property_scientific_abstract"]=>
  string(1568) "Background: T2DM trials require large samples, driving high costs and delays. This study uses machine learning to improve efficiency. By leveraging historical placebo data to generate prognostic scores, the method reduces outcome variability and enables sample size reduction. Objective: Evaluate whether a prognostic score—derived from a neural network trained on historical placebo data—as an MMRM covariate can reduce sample size and increase power in a placebo-controlled T2DM trial. Study Design: Methodological study using historical data. A neural network trained on prior placebo-arm data (T2DM patients inadequately controlled with diet and exercise) generates an individualized prognostic score—a digital twin—predicting HbA1c trajectory under placebo. This covariate in MMRM reduces residual variance, permitting placebo group size reduction without compromising precision. Participants: T2DM patients inadequately glycemic controlled with diet and exercise randomly assigned to placebo group. Primary Outcome Measure(s): Two co-primary metrics: 1) Variance reduction in estimated treatment effect for adjusted vs. unadjusted MMRM. 2) Percentage reduction in total sample size needed for 80% power to detect a prespecified HbA1c reduction. Secondary Outcome Measure(s): 1) Correlation between prognostic scores and observed HbA1c trajectories.  Statistical Analysis: MMRM with neural network-generated prognostic score as a fixed covariate. Efficiency gains quantified by comparing standard errors and sample size requirements vs. conventional MMRM."
  ["project_brief_bg"]=>
  string(2244) "T2DM affects more than 500 million individuals worldwide, creating an urgent and growing demand for novel drugs. RCTs remain the gold standard for regulatory approval and market authorization. However, RCTs in T2DM are increasingly inefficient: typically require large sample sizes—often exceeding 500 participants—to detect clinically meaningful reductions in HbA1c. This inefficiency translates into substantial financial costs, prolonged development timelines, and delayed patient access to effective treatments. Conventional analytic approaches rarely incorporate historical control data, representing a missed opportunity to reduce outcome variability and enhance statistical precision. 

Prognostic Covariate Adjustment (PROCOVA™), a statistical methodology qualified by the European Medicines Agency (EMA), addresses this gap by integrating deep learning with historical trial data. PROCOVA™ trains predictive models on past placebo-group outcomes to generate patient-specific prognostic scores, which are then incorporated as covariates in the primary RCT analysis. This approach reduces residual variance, enabling smaller sample sizes without inflating Type I error or compromising statistical power. While PROCOVA™ has been validated in clinical trials for Alzheimer’s disease, its application to T2DM remains unexplored—a critical gap given the abundance of historical diabetes trial data and the pressing need for more efficient trial designs in this therapeutic area.

This project aims to demonstrate the feasibility of PROCOVA™ in T2DM by quantifying its potential to reduce required sample sizes by 20–30% and improve the precision of treatment effect estimates. The insights generated will not only validate this methodology for diabetes but also provide a generalizable framework for applying PROCOVA™ to other chronic diseases—such as cardiovascular disease, obesity, and beyond—thereby fostering a paradigm shift toward more efficient, data-driven clinical research. By bridging advanced machine learning techniques with regulatory-grade trial design, this work will materially enhance generalizable knowledge in translational medicine and inform evidence-based policy for public health."
  ["project_specific_aims"]=>
  string(1120) "This project will develop and validate a prognostic covariate adjustment methodology tailored to T2DM clinical trials, leveraging historical placebo-group data to improve precision in estimating treatment effects on longitudinal glycemic outcomes. The specific aims are as follows:

Aim 1: Methodological Development and Validation. To develop a neural network-based framework that generates patient-specific prognostic scores—digital twins—predicting longitudinal HbA1c trajectories under placebo, using baseline covariates and historical trial data. We will validate the predictive accuracy of these scores against observed outcomes in held-out placebo datasets.

Aim 2: Quantifying Efficiency Gains. To evaluate the reduction in required sample size achieved by incorporating the prognostic score as a covariate in a MMRM, compared to conventional unadjusted MMRM. We hypothesize that the proposed method will achieve ≥20% reduction in sample size while maintaining 80% power to detect a clinically meaningful HbA1c reduction (e.g., 0.5 percentage points), with no inflation of Type I error.

"
  ["project_study_design"]=>
  array(2) {
    ["value"]=>
    string(8) "meth_res"
    ["label"]=>
    string(23) "Methodological research"
  }
  ["project_purposes"]=>
  array(3) {
    [0]=>
    array(2) {
      ["value"]=>
      string(37) "develop_or_refine_statistical_methods"
      ["label"]=>
      string(37) "Develop or refine statistical methods"
    }
    [1]=>
    array(2) {
      ["value"]=>
      string(34) "research_on_clinical_trial_methods"
      ["label"]=>
      string(34) "Research on clinical trial methods"
    }
    [2]=>
    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(1882) "This study will leverage patient-level data from participants randomized to the placebo arm in completed randomized controlled trials (NCT01081834), accessed through the Yale University Open Data Access (YODA) Project. To augment the training dataset and enhance model generalizability, additional patient-level data from historical placebo-controlled trials in type 2 diabetes will be incorporated: In addition to the previously requested trials, we will incorporate data from a Phase III clinical trial evaluating another SGLT2 inhibitor in Chinese adults with type 2 diabetes mellitus who have inadequate glycemic control with diet and exercise alone. This trial is registered under identifier CTR20240399. Data from the placebo arm of this study will be combined with placebo-arm data from NCT01081834 we have requested to form the training and validation sets for the prognostic model. The target population of CTR20240399 is aligned with that of NCT01081834, ensuring consistency in baseline characteristics and outcome measures. Permission to use these data has been obtained from the sponsoring company.



Inclusion Criteria:

Eligible participants for model training must meet the following criteria:

1. Randomized to the placebo group.

2. With inadequate glycemic control on diet and exercise, and patients on an AHA who had to undergo an 8-week AHA washout period with diet and exercise.

3. Complete baseline assessments .

4. At least one post-randomization HbA1c measurement available during follow-up.

Based on currently available information, we estimate that 192 participants from NCT01081834 will meet these criteria. The final sample size will be confirmed upon data access and verification of actual enrollment.



Exclusion Criteria:

Randomized to the active treatment group in NCT01081834."
  ["project_main_outcome_measure"]=>
  string(1758) "Primary Outcome Measures:

This methodological study evaluates two co-primary metrics to quantify the efficiency gains of the proposed prognostic covariate adjustment approach:

1. Precision of Treatment Effect Estimates (Variance Reduction). The primary metric is the relative reduction in the variance of the estimated treatment effect on longitudinal change in HbA1c (%) achieved by the novel methodology compared to an unadjusted MMRM. This will be calculated as the ratio of the treatment effect variance under the prognostic score-adjusted MMRM to that under the unadjusted MMRM. The final analysis will report the percentage reduction in variance relative to the conventional approach.

2. Sample Size Reduction. The second primary metric is the percentage reduction in required total sample size needed to maintain 80% statistical power to detect a prespecified clinically meaningful HbA1c reduction , comparing the novel methodology to traditional unadjusted methods. This will be estimated through simulation studies informed by variance components derived from historical data. Results will be reported as the percentage sample size reduction (e.g., "the novel method enabled a 25% reduction in sample size while preserving 80% power").



Secondary Outcome Measures:

Prognostic Score Predictive Performance. The correlation between time-matched prognostic scores (generated from baseline covariates) and observed longitudinal HbA1c trajectories will be evaluated using the Pearson correlation coefficient (r), calculated at each prespecified study time point (e.g., 12, 24, and 36 weeks). This assesses the extent to which the digital twin captures individual-level glycemic progression under placebo.

"
  ["project_main_predictor_indep"]=>
  string(371) "The main independent variable is treatment assignment, defined as the randomized allocation of participants to either the active intervention arm or the placebo control arm. This binary variable serves as the main predictor of interest, and its effect on longitudinal glycemic outcomes will be tested within the primary analytic framework.







"
  ["project_other_variables_interest"]=>
  string(1061) "While not independent variables in the causal sense, the following covariates are incorporated into the model to enhance precision and reduce residual variance:



Prognostic Score. A continuous variable generated for each participant using a deep learning model trained exclusively on historical placebo-arm data. The score represents the predicted longitudinal HbA1c trajectory for that individual under control conditions, conditional on baseline characteristics. Computed at each study timepoint using only baseline data, the prognostic score is included as a fixed covariate in MMRM to reduce outcome variability and improve the precision of treatment effect estimates.



Baseline Covariates. Selected baseline characteristics—including age, sex, baseline HbA1c, body mass index (BMI), and other prespecified variables—are used as inputs to generate the prognostic scores. In accordance with the study statistical analysis plan, these covariates may also be included directly in the MMRM if specified for additional adjustment."
  ["project_stat_analysis_plan"]=>
  string(3921) "Statistical Analysis:

This study comprises two integrated components. The first component involves artificial intelligence model development: a neural network will be trained using baseline covariates and longitudinal HbA1c data from approximately 192 placebo-arm participants in trials NCT01081834 to generate patient-specific prognostic scores. The second component applies this model in a methodological evaluation, comparing traditional MMRM with a prognostic score-adjusted MMRM. This section describes the statistical analysis plan for the second component.



General Principles:

For continuous variables, descriptive statistics will include sample size, arithmetic mean, least squares mean, standard deviation, standard error, quartiles (median, first and third quartiles), and minimum and maximum values. For categorical variables, frequencies and percentages will be reported to one decimal place; percentages will be suppressed when the numerator is zero. Missing categorical data will be handled by including "missing" as a distinct category. All hypothesis tests will be two-sided with a significance level of 0.05. Missing data related to intercurrent events will be addressed according to a prespecified intercurrent event handling strategy, aligned with the estimand framework.



Analysis Populations:

Randomized Set (RS): All randomized participants, regardless of treatment receipt. Used for disposition and protocol deviation summaries.

Safety Set (SS): All randomized participants who received at least one dose of investigational drug. Used for safety analyses, based on actual treatment received.

Full Analysis Set (FAS): All randomized participants who received at least one dose of investigational drug, following the intention-to-treat principle. Used for efficacy endpoints other than HbA1c.

Modified Full Analysis Set (mFAS): All FAS participants with a baseline and at least one post-baseline HbA1c measurement. Used for all HbA1c-related primary efficacy analyses, following the intention-to-treat principle.

Per-Protocol Set (PPS): A subset of the mFAS comprising participants without major protocol deviations affecting the primary endpoint, with good medication adherence, and without prohibited medication use. Used for sensitivity analyses of the primary efficacy endpoint.



Primary Efficacy Analysis:

The primary efficacy endpoint is the change from baseline in HbA1c at Week 26. The target population is patients with type 2 diabetes inadequately controlled with diet and exercise. The primary comparison is the difference in least squares mean change from baseline between the active intervention and placebo groups.



Two analytic approaches will be implemented and compared:

Conventional MMRM: A standard MMRM including treatment group, baseline HbA1c, visit, and treatment-by-visit interaction as fixed effects, with participant as a random effect.

Prognostic Score-Adjusted MMRM: The same MMRM structure, with the addition of the neural network-generated prognostic score as a continuous fixed covariate.

For both approaches, the MMRM will use an unstructured covariance matrix to model within-participant correlation. Kenward-Roger approximation will be used for degrees of freedom. Intercurrent events (e.g., rescue medication initiation, treatment discontinuation) will be addressed using a hypothetical estimand strategy, reflecting the treatment effect in the absence of these events.



Efficiency Gain Quantification:

The standard errors of the treatment effect estimates from the two MMRM approaches. The required sample sizes to achieve 80% power to detect a prespecified clinically meaningful HbA1c reduction (e.g., 0.5 percentage points), derived from the variance components estimated from each approach.



"
  ["project_software_used"]=>
  array(2) {
    [0]=>
    array(2) {
      ["value"]=>
      string(6) "python"
      ["label"]=>
      string(6) "Python"
    }
    [1]=>
    array(2) {
      ["value"]=>
      string(86) "not_analyzing_participant_level_data__plan_to_use_another_secure_data_sharing_platform"
      ["label"]=>
      string(92) "I am not analyzing participant-level data / plan to use another secure data sharing platform"
    }
  }
  ["project_timeline"]=>
  string(1558) "The proposed study will commence upon approval of the data request and execution of the Data Use Agreement with the YODA Project. All activities are planned within the 12-month data access period, with the following key milestones:



Months 1–5: Data Analysis and Results Generation



Data processing, model training and validation, and primary statistical analyses will be conducted. This phase includes all analyses described in the statistical analysis plan, culminating in the generation of final results and conclusions.



Months 4–6: Manuscript Preparation and Submission



Drafting of the manuscript will begin during the latter phase of analysis. The target is to submit the manuscript for peer-reviewed publication by the end of Month 6. Simultaneously with manuscript submission, a summary of results will be reported back to the YODA Project, fulfilling the data use agreement requirements.



Months 6–12: Revision and Resubmission



Following journal submission, this period will be dedicated to addressing reviewer comments, making necessary revisions, and resubmitting the manuscript as required. Any requests for additional analyses or manuscript modifications will be accommodated within this window.



Should additional time be required beyond the initial 12-month access period—for instance, to complete extensive revisions or conduct supplementary analyses requested by reviewers—we will request an extension  in accordance with policies."
  ["project_dissemination_plan"]=>
  string(413) "We plan to submit the manuscript to a high-impact peer-reviewed journal with a focus on clinical trial methodology, medical statistics, or diabetes research. Potential venues include NPJ Digital Medicine, Diabetologia, Journal of Biomedical Informatics, Statistics in Medicine, BMC Medical Research Methodology, Clinical Trials. The final selection will be guided by the scope and emphasis of the completed study."
  ["project_bibliography"]=>
  string(1679) "

Schuler A, Walsh D, Hall D, Walsh J, Fisher C. Increasing the efficiency of randomized trial estimates via linear adjustment for a prognostic score. The international journal of biostatistics. 2022;18(2):329-356.
EMA. Qualification opinion for Prognostic Covariate Adjustment (PROCOVA™) https://www.ema.europa.eu/en/documents/regulatory-procedural-guideline/qualification-opinion-prognostic-covariate-adjustment-procovatm_en.pdf. 2022.
Ross. JL, Sabbaghi. A, Zhuang. R, Bertolini. D. Enhancing Longitudinal Clinical Trial Efficiency with Digital Twins and Prognostic Covariate-Adjusted Mixed Models for Repeated Measures (PROCOVA-MMRM). ArXiv:200202779v2. 2024.
Daniele Bertolini ADL, Aaron Smith, David Li-Bland, Yannick Pouliot, Jonathan R. Walsh, Charles K. Fisher. Forecasting progression of mild cognitive impairment (MCI)and Alzheimer’s disease (AD) with digital twins. Alzheimer’s Dementia. 2021;17(Suppl. 9).
Jonathan R. Walsh AMS, Yannick Pouliot, David Li-Bland, Anton Loukianov, Charles K. Fisher. Generating Digital Twins with Multiple Sclerosis Using Probabilistic Neural Networks. arXiv:200202779v2. 2020.
Fisher CK, Smith AM, Walsh JR. Machine learning for comprehensive forecasting of Alzheimer’s Disease progression. Scientific reports. 2019;9(1):13622.
Bertolini D LA, Smith AM, et al. Modeling Disease Progression in Mild Cognitive Impairment and Alzheimer’s Disease with Digital Twins. ArXiv. 2020;ArXiv201213455 Cs Q-Bio. Published online December 24, 2020. Accessed March 6, 2021.(http://arxiv.org/abs/2012.13455).

"
  ["project_suppl_material"]=>
  bool(false)
  ["project_coi"]=>
  array(2) {
    [0]=>
    array(1) {
      ["file_coi"]=>
      array(21) {
        ["ID"]=>
        int(18926)
        ["id"]=>
        int(18926)
        ["title"]=>
        string(7) "COI.pdf"
        ["filename"]=>
        string(7) "COI.pdf"
        ["filesize"]=>
        int(189898)
        ["url"]=>
        string(56) "https://yoda.yale.edu/wp-content/uploads/2026/03/COI.pdf"
        ["link"]=>
        string(56) "https://yoda.yale.edu/data-request/2026-0188/coi-pdf-14/"
        ["alt"]=>
        string(0) ""
        ["author"]=>
        string(4) "2061"
        ["description"]=>
        string(0) ""
        ["caption"]=>
        string(0) ""
        ["name"]=>
        string(10) "coi-pdf-14"
        ["status"]=>
        string(7) "inherit"
        ["uploaded_to"]=>
        int(18908)
        ["date"]=>
        string(19) "2026-03-17 06:32:56"
        ["modified"]=>
        string(19) "2026-03-17 06:33:03"
        ["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(18927)
        ["id"]=>
        int(18927)
        ["title"]=>
        string(11) "COI_Yan.pdf"
        ["filename"]=>
        string(11) "COI_Yan.pdf"
        ["filesize"]=>
        int(180117)
        ["url"]=>
        string(60) "https://yoda.yale.edu/wp-content/uploads/2026/03/COI_Yan.pdf"
        ["link"]=>
        string(57) "https://yoda.yale.edu/data-request/2026-0188/coi_yan-pdf/"
        ["alt"]=>
        string(0) ""
        ["author"]=>
        string(4) "2061"
        ["description"]=>
        string(0) ""
        ["caption"]=>
        string(0) ""
        ["name"]=>
        string(11) "coi_yan-pdf"
        ["status"]=>
        string(7) "inherit"
        ["uploaded_to"]=>
        int(18908)
        ["date"]=>
        string(19) "2026-03-17 06:32:59"
        ["modified"]=>
        string(19) "2026-03-17 06:33:03"
        ["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(0) ""
  ["project_rep_publ"]=>
  bool(false)
  ["project_assoc_data"]=>
  array(0) {
  }
  ["project_due_dil_assessment"]=>
  bool(false)
  ["project_title_link"]=>
  array(21) {
    ["ID"]=>
    int(18897)
    ["id"]=>
    int(18897)
    ["title"]=>
    string(28) "Data Request Approved Notice"
    ["filename"]=>
    string(32) "Data-Request-Approved-Notice.pdf"
    ["filesize"]=>
    int(195663)
    ["url"]=>
    string(81) "https://yoda.yale.edu/wp-content/uploads/2026/02/Data-Request-Approved-Notice.pdf"
    ["link"]=>
    string(77) "https://yoda.yale.edu/data-request/2026-0088/data-request-approved-notice-69/"
    ["alt"]=>
    string(0) ""
    ["author"]=>
    string(4) "1885"
    ["description"]=>
    string(0) ""
    ["caption"]=>
    string(0) ""
    ["name"]=>
    string(31) "data-request-approved-notice-69"
    ["status"]=>
    string(7) "inherit"
    ["uploaded_to"]=>
    int(18715)
    ["date"]=>
    string(19) "2026-03-10 19:58:49"
    ["modified"]=>
    string(19) "2026-03-10 19:58:49"
    ["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"]=>
  bool(false)
  ["project_highlight_button"]=>
  string(0) ""
  ["request_data_partner"]=>
  string(0) ""
  ["project_software_used_exp"]=>
  string(121) " All analyses will be conducted using SAS (version 9.4 or higher), with the neural network models implemented in PyTorch."
}
data partner
array(1) {
  [0]=>
  string(0) ""
}


pi country
array(0) {
}


pi affil
array(0) {
}


products
array(0) {
}


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


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