Past Data Studio Presentations

This study will build on the World Health Organization Infection Prevention and Controls (WHO IPC) guidance to address infection and prevention control research gaps about the safety and efficacy of the source- and infection-control equipment used in the field during pandemic conditions. It will provide evidence for design improvement of existing infection-control strategies in order to use hospital equipment effectively in less-than-optimal conditions and will provide evidence concerning personal protective equipment (PPE) and the role of powered air-purifying respirators (PAPR) and their implementation during pandemic conditions. This study consists of three aims. The first aim addresses Human Factors Simulations. We will compare and test auditory and visual acuity plus cognitive load and error tracing across equipment combinations N95/face shield, Traditional PAPR, and Light PAPR. The second aim addresses Field Testing Applications. We will support the field studies in Bologna and Sao Paola and offer evidence for optimal PPE choices in specific pandemic environments and offer ways to compensate for unavoidable loss of dexterity, auditory or visual acuity on the field. The third aim addresses Aerosol Dispersion Evaluation. We will evaluate the dispersion of aerosol particles under various medical environments and aerosol generating procedures (AGP) to characterize and potentially risk stratify such environments and potential mitigating PPE.

California is one of the epicenters of America’s mass incarceration crisis: 25% of its 115,000 state prisoners serve a life sentence. These individuals will die in correctional facilities unless released by the Board of Parole Hearings (BPH) with gubernatorial approval. Each year, BPH schedules thousands of parole hearings for eligible candidates, after which the candidates are either granted or denied parole for some number of years. In 2019 alone, BPH held 6,061 hearings and granted parole in 1,181 cases, a grant rate of 19.5%. The parole system has remained largely opaque, since the only public information it produces is a 150-page transcript of the 2+ hour free-form hearing dialogue. We have obtained a complete dataset of 35,000 parole hearing transcripts from the State of California dating from 2007 to 2019. In 2020, we sued the State of California to obtain the race and attorney representation data for all of the 15,766 prisoners covered by these transcripts. We are now attempting to construct the most complete picture to date of the California parole system by stitching together information from a number of sources: (1) covariates automatically extracted from the transcripts using Natural Language Processing (NLP) techniques and automated parsing of the PDF documents, (2) manually annotated information covering a subset of approximately 1,000 transcripts, (3) information scraped from California Department of Corrections & Rehabilitation websites, and (4) the results of our lawsuit. We are conducting a retrospective observational analysis seeking to describe the independent effect of approximately 50 covariates on the parole outcome (grant or deny). Most of these covariates are binary (e.g., Did the district attorney oppose release?), some are continuous (e.g., How many years ago was the candidates last disciplinary infraction in prison?), and a few are ordinal (e.g., Was the psychologists assessment one of high, medium, or low risk?)

We are trying to demonstrate that COVID-19 diagnosis data merged with rideshare trip data may enable detection of novel, more infectious viral variants. We have developed Rideshare Infection Detection (RIDE), a mathematical model of rideshare vehicle transmission and analytical method for analysis of aggregated rideshare plus COVID-19 data. RIDE allows comparison of expected and observed infection numbers in the rideshare network, and thus can detect the infectivity of the current strain. Because aggregated rideshare plus COVID-19 data is not currently available, we create synthetic data via a simulation of viral propagation through Los Angeles with a hypothetical baseline viral variant A. We then analyze the results with RIDE, assuming either variant A or an alternative, incorrectly hypothesized, more infectious variant B. When analyzed with the true variant (variant A), the difference in the number of observed infections and the number of expected infections should be about 0. Our initial simulations suggest that, despite noise, analysis assuming variant A leads to results that are closer to 0 than analysis with variant B.

The field of hip arthroscopy has been growing exponentially over the past several years. Both the UK and Denmark have hip arthroscopy registries. However, there is no such registry in North America. Our vision is to develop a hip arthroscopy registry to track pre-operative baseline measurements, intra-operative findings, and post-operative patient outcomes for up to 10 years. This will be a multi-site QA/QI initiative. Data will be collected from both patients and surgeons. We intend to use the data for future research projects.

We aim to demonstrate that, compared to controls, high school youth from disadvantaged, underrepresented communities will increase problem solving skills, health knowledge, self-efficacy, patient activation, healthy behaviors, and talking about health at home after participating in the 8-week Stanford Youth Diabetes Coaching Program (SYDCP). Over the past 10 years, we have developed, piloted, and evaluated SYDCP using quasi-experimental design with small convenience samples which, when aggregated, have shown statistically significant improvement in these measures. We are exploring the feasibility of using a cluster randomized trial with delayed intervention. We estimate that we could recruit up to 10 high schools (5 per year over 2 years or 10 in one year), each with 2 participating classes of approximately 20 tenth-grade students, which would yield 200 participants and 200 controls. Participants would complete the program in the Fall, while controls will complete the program in the Spring. We aim to have all participants complete validated surveys at 3 points in time: baseline (beginning of fall semester), after the intervention group completes training (end of fall semester), and after the control group completes training (end of spring semester).

Stroke survivors often have involuntary muscle contraction in their hand and arm (spastic hypertonia). Promising preliminary evidence suggests that vibrotactile stimulation may be associated with reductions in this condition. This project will measure the impact of four stimulation conditions. The study has a within-subjects design and each participant will experience four visits. A different stimulation condition will be applied during each visit. Symptoms of spastic hypertonia will be measured at three joints using two ordinal scales, two angle measures recorded using motion capture, and a custom tool. Measures will be taken at four timepoints, including before and after stimulation. Aim 1 is to evaluate which condition showed the greatest change in measures. Aim 2 is to evaluate change between each two time points within each condition, providing us data on the duration of effects.

Time-to-event endpoints are often used to determine treatment efficacy in clinical trials. It is common practice to collect alongside the time-to-event endpoint additional longitudinal information on important biomarkers. Examples include the collection of prostate-specific antigen alongside time-to-death in prostate cancer trials, or fasting plasma glucose levels alongside time-to-diabetes in preventive clinical trials for type 1 diabetes. In many instances, the longitudinal biomarker could be seen as potential surrogate marker for the event of interest, and may provide additional insights in the treatment effect. In this project we use a joint modeling strategy to explicitly model the intrinsic relationship between the event and the biomarker by incorporating the longitudinal biomarker process into the hazard function of the event. In this framework, treatment could either affect the time-to-event endpoint through the biomarker, or treatment could act through a separate pathway that is not accounted for by the biomarker. Characterizing the different pathways and the total treatment effect are the parameters of interest. We will illustrate the approach using observed clinical trial data in amyotrophic lateral sclerosis, a rapidly progressive neurological disease with an average survival time of 18 months after diagnosis.

Each year pain costs the U.S. economy more than diabetes, cancer, and heart disease combined. Pain is ubiquitous across most health conditions, disproportionately affects women and minorities, and is the leading cause of disability. Pain is cited as the main reason people seek primary care, and for first use of illicit substances among people with addiction. Strikingly, behavioral pain treatment is largely inaccessible in primary care settings and addiction treatment programs. There is in urgent need for low-cost integrated technological infrastructure that can characterize the needs of each individual early on and deploy low-risk, scalable, tailored, and on-demand behavioral treatment.

The study design we are planning will be prospective, randomized-controlled, multi-site, multisetting, pragmatic, implementation research study that includes a subsequent responsive adaptive clinical trial design (N=1,200). We will provide free national access to an existing cloud-based learning health system (LHS) and treatment platform that will deliver on-demand and personalized digital treatments for acute and chronic pain. The LHS will be accessed via the internet. Any report of pain triggers an automatic invitation to enroll and all participants complete a common streamlined survey battery.

The Data Studio Workshop brings together a biomedical investigator with a group of experts for an in-depth session to solicit advice about statistical and study design issues that arise while planning or conducting a research project. This week, the investigators will discuss the following project with the group.

Post-operative pain management in pediatric patients requires intermittent administration of opioid medications that are available in concentrated form. Small weight pediatric patients (less than 20kg) are at a particular risk for receiving potentially inaccurate volumes of opioid medications given low milligram per kilogram dosing, lack of commercially available diluted formulations, and variability in administration techniques among providers. We hypothesize that specific administration techniques and the use of pharmacy compounded diluted opioid products is associated with greater accuracy of administered small volume opioid doses in the post-operative setting.

The Data Studio Workshop brings together a biomedical investigator with a group of experts for an in-depth session to solicit advice about statistical and study design issues that arise while planning or conducting a research project. This week, the investigator(s) will discuss the following project with the group.

A substantial proportion of adults with B-cell acute lymphoblastic leukemia (ALL) will develop relapsed or refractory (r/r) leukemia. Outcomes for this population are poor. The development and recent FDA approval of chimeric antigen receptor T-cell (CART) therapy directed at CD19 for ALL has revolutionized the approach to this disease. However, no commercial CART product exists for r/r ALL in adults. Among the published single center clinical trials of CART in adult ALL, response rates are universally high; 50-70% will achieve complete response. However, at least two major barriers to progress stand out. One is CAR-mediated toxicities (cytokine release syndrome [CRS] and neurotoxicity) induced by the early and rapid expansion and proliferation of these cells. The other is subsequent T-cell exhaustion with loss of clinical response.

We screened several small molecule kinase inhibitors on CAR T-cell functionality and identified dasatinib, an FDA-approved drug for the treatment of Ph+ ALL and chronic myeloid leukemia (CML). It is a potent and reversible inhibitor of CAR T cell cytotoxicity, cytokine secretion, and proliferation. Dasatinib inhibits CAR T-cell functionality in vitro and in vivo at nanomolar concentrations similar to those detected in the serum of Ph+ ALL and CML patients treated with dasatinib. These data suggest that dasatinib could be safely used as an OFF switch to prevent or mitigate CAR-mediated toxicities.

In a separate study, we discovered that transient disruption of CAR signaling (i.e. ”rest”) via downregulation of surface CAR could reprogram exhausted CAR T cells to resemble memory-like cells, thereby leading to functional reinvigoration. We hypothesized that transient application of dasatinib could recapitulate these results by inducing ”rest”. Indeed, using an in vitro model of CAR T-cell exhaustion, dasatinib treatment of exhausted CAR T-cells for 4-7 days enhanced the expansion of CD8+ cells, decreased the expression of canonical exhaustion markers, and enhanced antigen-dependent cytotoxicity and cytokine secretion. Notably, repeated 3-day dasatinib pulses prolonged the anti-tumor response and enhanced survival in 2 separate dasatinib-insensitive tumor xenograft murine models. Tumor-infiltrating CAR T-cells in mice treated with dasatinib exhibited an increased percentage of memory-like cells and enhanced degranulation when re-stimulated ex vivo. Collectively, these results indicate that CAR T-cell and dasatinib combination therapy may result in enhanced safety and efficacy.

The Data Studio Workshop brings together a biomedical investigator with a group of experts for an in-depth session to solicit advice about statistical and study design issues that arise while planning or conducting a research project. This week, the investigator(s) will discuss the following project with the group.

The Center for Medicare and Medicaid Services (CMS) has recently focused on unplanned care events in cancer patients as a measure of both the quality of care delivered and a source of potential cost savings. These events have appeared in a pay-for-reporting program that contains a metric, OP-35, which measures the rate of hospital admissions and emergency department visits in patients receiving outpatient chemotherapy. These chemotherapy-related events are felt to be potentially avoidable with appropriate access, outreach, and preventative measures.

For practices with very large patient populations such as academic medical centers, generic interventions to address unplanned care can quickly overwhelm available resources. Focused interventions on higher risk patients present an opportunity to decrease the rate of unplanned care events in patients receiving outpatient chemotherapy. Leveraging the data available in the electronic medical record to identify patients may allow for alerts to be seamlessly pushed to frontline caregivers and give them the opportunity to reach out and provide high-touch care to those in greatest need.

We are developing a predictive model for identifying patients at risk for chemotherapy-related unplanned care events during outpatient treatment as defined by the CMS OP-35 metric. This effort encompasses many aspects of data science, some of which are the following: how to clean multiple datasets each with a different structure, how to combine multiple datasets, how should the data be structured for a predictive model, how to handle missing values among the covariates, how to define the response variable, how to deal with dependencies among the covariates, how should the covariates be represented in the model, how to choose a predictive model, should the data be split into construction-validation-test sets, and how should the predictive value of the model be measured.

The Data Studio Workshop brings together a biomedical investigator with a group of experts for an in-depth session to solicit advice about statistical and study design issues that arise while planning or conducting a research project. This week, the investigator(s) will discuss the following project with the group.

The use of cognitive aids during crises has been shown to increase appropriate care and decrease omission of vital actions across a variety of fields, including in medicine. While cognitive aids are widely used and mandated in crises in the military and aviation, adoption of these tools has been slower in medicine, and research on actual use once introduced is scarce. An emergency manual (a cognitive aid for anesthesia providers, detailing management of rare crises) was introduced to a large US hospital system in 2014. Data were collected during the first six months (period A) on whether and for what purpose an emergency manual was used in every surgical or procedural case in the hospital. Similar data were collected during a follow-up period one year later (period B), and are currently being collected at the five-year follow-up mark (period C). It is our aim to evaluate the long-term sustainability of the use of emergency manuals. The current study is specifically focusing on changes from period B to period C, as it is well known that use decreases from period A to B when even an intervention for routine use is not actively sustained. Specifically, we intend to compare overall use for each period, as well as use for rare crises.

The Data Studio Workshop brings together a biomedical investigator with a group of experts for an in-depth session to solicit advice about statistical and study design issues that arise while planning or conducting a research project. This week, the investigators will discuss the following project with the group. Patients are asymptomatically infected with Human Herpesvirsuses in their youth, which remain latent in the body but can later reactivate and cause potentially serious or fatal disease in immunocompromised patients. The viruses typically have intrinsic reactivation times relative to the day of allogeneic hematopoietic cell transplant (HCT) infusion but these can be influenced by various factors. We have a retrospective, single center study of 77 umbilical cord blood (UCB) transplant patients transplanted at Stanford University Medical Center (Jan 2013–Dec 2019).

Patients were grouped by initial antiviral prophylaxis (PPX), treated as Intention-to-treat. Clinically significant infection (CSI) due to Human Herpesvirus 6 (HHV-6) or Cytomegalovirus (CMV) was defined as receipt of directed antiviral therapy. With our unique population of UCB recipients who have received either valacyclovir or letermovir (LET), we would like to evaluate the impact of high-dose valacyclovir versus standard-dose acyclovir prophylaxis on the incidence of HHV6-CSI and CMV-CSI, report the incidence of HHV6-CSI and CMV-CSI with or without concomitant LET PPX, and expand our understanding of interactions between CMV and HHV-6 reactivation.

The Data Studio Workshop brings together a biomedical investigator with a group of experts for an in-depth session to solicit advice about statistical and study design issues that arise while planning or conducting a research project. This week, the Data Studio Workshop will be a joint effort with our speakers, Elizabeth Lee and Andrew Brooks. Elizabeth is a Senior Privacy Officer with the Office of the Chief Risk Officer and Andrew is a Postdoctoral Fellow in Mike Snyders lab. They will discuss the topics of data privacy, re-identification risk, and shared computing for genomics, imaging, sensors, and biomedical datasets.

The HIPAA Safe Harbor Method for Deidentification does not provide the risk mitigation that research data scientists need for the long-run, especially as machine learning and natural language processing become more prevalent. During the workshop, we will consider a use case where the HIPAA Safe Harbor Method for Deidentification might not mitigate risk sufficiently. Elizabeth and Andrew will present a COVID-19 monitoring data set as an example to discuss the potential risks for its release and mitigation strategies to reduce the risk.

Given that this will be the first time we have dealt with data privacy in the workshop, we need to clarify terminology, define the problem, and set expectations. The specific strategies we will focus on during the workshop include differential computing, federated algorithms, adversarial attacks, re-identification risk rating, de-identification, and differential privacy. We expect to have multiple workshops to develop consensus around goals and ways to collaborate. The ultimate goal is to hand over to an initiative led by the University Privacy Office with participation of faculty and students who are interested in this project.

The Data Studio Workshop brings together a biomedical investigator with a group of experts for an in-depth session to solicit advice about statistical and study design issues that arise while planning or conducting a research project. This week, the investigator(s) will discuss the following project with the group.

Myalgic Encephalomyelitis/Chronic Fatigue Syndrome (ME/CFS) is a chronic, debilitating illness of unknown etiology. An ME/CFS diagnosis is based solely on symptoms with case definitions made by expert consensus. According to the most recent Institute of Medicine (2015) case definition, the core symptoms of ME/CFS include debilitating fatigue, unrefreshing sleep, post-exertional malaise, and cognitive dysfunction or orthostatic intolerance. Up to 85% of patients with ME/CFS report cognitive impairment symptoms which include difficulty with memory, attention, and information processing.

Although the cause of the illness is unknown, a growing body of evidence suggests that ME/CFS involves inflammation of the brain which includes changes in inflammatory cytokines in both plasma and cerebrospinal fluid correlated with the severity of symptoms. Studies using positron emission tomography (PET) show evidence of activated microglia or astrocytes in various regions of the brain in ME/CFS patients.

Dopamine D2 receptor agonists have been shown to mediate neuroinflammation, microglial activation, and cell death in animal models and humans (4-6). This suggests that dopaminemodulating drugs may lead to clinical improvement in fatigue and cognitive symptoms in ME/CFS. Given the lack of approved drugs for treating this condition, we are interested in exploring the potential benefit of low doses of Aripiprazole.

In a retrospective study, we reviewed the medical records of 101 patients who met the criteria for an ME/CFS diagnosis according to three separate case definitions (Fukuda, CCC, and IOM) and who received off-label Aripiprazole. Medical records were included for individuals evaluated in the clinic at least twice, representing periods before and after the use of the medication. The age range was from 18 to 84 years old (mean 51 y), with a gender distribution of 67% female and 33% male, and the duration of illness was from 1 to 54 years (median 10 years). The daily oral dose of Aripiprazole ranged from 0.2-2.0 mg/day (mean 0.94 mg/day). During each clinic visit, patients were asked to rate their symptoms on a scale of 0-10.

Of the 101 patients taking Aripiprazole, 75/101 (74%) experienced an improvement in one or more categories: fatigue, brain fog, unrefreshing sleep, and frequency of post-exertional malaise (PEM) episodes. Twelve individuals (12%) had no observable difference in symptoms at the maximum dose of 2 mg, and 14 individuals (14%) reported worsening of symptoms or onset of side effects that led to discontinuation of the drug. These results suggest that Aripiprazole may effectively reduce symptoms of ME/CFS and warrants further investigation in a randomized clinical trial.

Currently in the United States, over 100,000 patients undergoing dialysis for renal failure are on a wait list for transplantation because it offers the best chance for a near-normal lifestyle and improved life expectancy. Recipients of donor kidneys require lifelong adherence to combinations of immune suppression (IS) medications to prevent immune-mediated rejection of the graft. The unmet medical needs in solid organ transplantation are to eliminate this lifelong adherence to combinations of IS medications with their serious side effects and to prevent immune-mediated rejection. We are planning a Phase 1 clinical trial to study whether conditioning the recipient using total lymphoid irradiation (TLI), anti-thymocyte globulin (ATG), and a single very low dose of total body irradiation (svldTBI) followed by an infusion of vertebral body (VB) hematopoietic cells (HCs) will result in persistent mixed chimerism after a standard-of-care deceased donor kidney transplant. We expect that persistent mixed chimerism will result in IS drug minimization/cessation and lead to an acquired immune tolerance of the graft. One of the primary objectives is to identify the safe and effective dose of svldTBI (range 40–120 centiGray) that will support persistent mixed chimerism. We are developing a novel dose finding method that is an extension of the modified Toxicity Probability Interval design version 2 (mTPI2). This method allows escalation and de-escalation according to a pre-tabulated decision table. Our method is based on the Bayesian probability interval of toxicity and efficacy that employs a simple beta binomial hierarchical model to find the safe and effective svldTBI dose in the interval of efficacy for mixed chimerism.

Previous studies have shown strong relation between exposure to air pollution and several health conditions, including cardiovascular and respiratory ailments. Air pollution due to wildfires have several additional complexities due to its intense and sudden nature. Therefore, as the severity and frequency of wildfires increase, understanding their effect on healthcare utilization is becoming a pressing concern. For public health officials, it is important to know how exposure to wildfire changes healthcare utilization at a population level in order to improve resource planning, preparation and decision making. We are building a predictive model that takes historical healthcare utilization and smoke exposure for individual geographical regions as inputs to predict the shortterm change in healthcare utilization.

Intracranial atherosclerotic disease (ICAD) is the cause of ischemic stroke in 40% of Asians, 29% of African-Americans, and 10% of Caucasians, making it the most common cause of stroke in the world. A primary goal of stroke treatment is to prevent recurrence. In cohorts with real-world standard-ofcare secondary prevention, the rate of recurrent ICAD stroke is high: over 20% of patients at 30 days from the initial stroke. The average annual rate of recurrence is 5% for all causes of stroke over the period 2000–2010. For patients with treated carotid atherosclerosis who have excellent medical and surgical options, the rate is less than 5%. This is the result of numerous major clinical trials aimed at improving secondary prevention in patients with cervical carotid atherosclerosis. In contrast, there have been only two major clinical trials for ICAD patients: WASID and SAMMPRIS. The WASID trial reported that ICAD patients with 50–99% stenosis had a 15% annual stroke recurrence with medical management. The SAMMPRIS trial included ICAD patients with 70–99% stenosis. This trial reported that treatment with aggressive medical management (including a wellness coach) was associated with a 15% annual recurrence rate compared to a 20% rate among patients who underwent an endovascular stenting procedure plus aggressive medical management. The high rates of stroke recurrence in these two trials with maximum medical management indicate that there is an urgent need for better stroke prevention regimens in this high-risk population. The same treatments that dramatically reduce the risk of stroke for patients with cervical carotid atherosclerosis do not work well enough for ICAD. We therefore need to look at new approaches for stroke prevention in ICAD patients. The most promising sources of residual risk to mitigate include thrombosis, inflammation, and lipids. We want to develop innovative clinical trials that will test promising and safe drug regimens that mitigate these sources of residual risk for secondary stroke prevention in patients with ICAD. Through these trials, we aim to make necessary progress in the fight against ICAD—the most treatment-resistant and common form of ischemic stroke in the world.

In December 2019, a series of pneumonia cases emerged in Wuhan, Hubei, China. Sequencing subsequently identified a novel coronavirus, which was named COVID-19. On March 11th 2020, the World Health Organization declared a pandemic and as of April 18th 2020, approximately 2.5 million people have been infected and at least 158,000 have died. The primary driver of COVID-19 morbidity and mortality is respiratory compromise. This is thought to be linked with increases of certain elements of the extracellular matrix, one of which is a matrix polymer known to drive inflammation in multiple other tissues and disease models. We are planning a proof-of-concept study to evaluate the clinical utility of a repurposed agent that inhibits the synthesis of an extracellular matrix polymer which may be a means of managing the respiratory morbidity of COVID-19 infection. The repurposed agent was developed in the 1970s and approved in Europe. However, it has been neither reviewed nor approved by the FDA. If successful, this may prove to be a low cost and easily adhered to medication with significant benefit in both the USA and the developing world. For the study design, we are considering an international, multi-center, 3-week, 2:1 randomized, phase 1/2 study of this repurposed agent as treatment versus standard of care with at most 40–45 subjects. One of the primary objectives is to identify the safe and effective dose. Other endpoints, that will be looked on in a largely exploratory manner, will include the ability of the repurposed agent to improve time to worsening in oxygenation requirement, time to mechanical ventilation, duration of hospital stay, and mortality.

In December 2019, a series of pneumonia cases emerged in Wuhan, Hubei, China. Sequencing subsequently identified a novel coronavirus, which was named COVID-19. On March 11th 2020, the World Health Organization declared a pandemic and as of April 18th 2020, approximately 2.5 million people have been infected and at least 158,000 have died. The primary driver of COVID-19 morbidity and mortality is respiratory compromise. This is thought to be linked with increases of certain elements of the extracellular matrix, one of which is a matrix polymer known to drive inflammation in multiple other tissues and disease models. We are planning a proof-of-concept study to evaluate the clinical utility of a repurposed agent that inhibits the synthesis of an extracellular matrix polymer which may be a means of managing the respiratory morbidity of COVID-19 infection. The repurposed agent was developed in the 1970s and approved in Europe. However, it has been neither reviewed nor approved by the FDA. If successful, this may prove to be a low cost and easily adhered to medication with significant benefit in both the USA and the developing world. For the study design, we are considering an international, multi-center, 3-week, 2:1 randomized, phase 1/2 study of this repurposed agent as treatment versus standard of care with at most 40–45 subjects. One of the primary objectives is to identify the safe and effective dose. Other endpoints, that will be looked on in a largely exploratory manner, will include the ability of the repurposed agent to improve time to worsening in oxygenation requirement, time to mechanical ventilation, duration of hospital stay, and mortality.

More than 260,000 US emergency department visits for pediatric mild traumatic brain injury (mild TBI) each year are complicated by intracranial bleeding that is often undiagnosed or sparsely monitored after hospital discharge. Mild TBI is defined as a head injury resulting in a Glasgow Coma Scale score of 13 or higher in a child with amnesia, transient neurologic symptoms, and/or brief loss of consciousness. National guidelines and preliminary research suggest that children with intracranial bleeding in the setting of mild TBI, commonly referred to as complicated mild TBI, may be at risk for prolonged recovery and future disability. One-third of all children with mild TBI have prolonged symptom-recovery beyond one month that impacts their families, reintegration to school and learning, and quality of life. However, the proportion of children with prolonged recovery secondary specifically to complicated mild TBI is unknown. Children with and without complicated mild TBI are currently managed almost identically and therefore, there is a critical need to understand differences in their outcomes. Without such information, informing the need for diagnostic neuroimaging and quantifying the benefits of management with early therapeutic intervention in this population remains unknown. This study’s overall objectives are to: 1) test if children with complicated mild TBI have longer recovery times when compared to those without complicated mild TBI and 2) test if children with complicated mild TBI at one year post-injury have a higher proportion of academic difficulty and lower quality of life when compared with those without complicated mild TBI. My central hypothesis, supported by preliminary data, is that children with complicated mild TBI have longer recovery times and higher risk for future academic difficulties at one year post-injury as compared to children without complicated mild TBI.

The COVID-19 pandemic has placed extraordinary demands on health systems. These unprecedented demands create the urgent need for medical resource allocation. Our goal is to build a transparent data-driven system that will assist decision making for front-line workers. The first step of our project is to build a patient-level risk assessment tool based on survival analysis. We integrated over 15 publicly available data sources and performed extensive data cleaning, de-duplication, and filtering to build a database for more than 10 countries and regions in the world. For each subgroup, we performed elementary survival analyses and presented our results on a website (https://zren.shinyapps.io/covid19/). The main challenge we are facing is to handle the missing values (e.g. age and gender) in those publicly available datasets that are unlikely to be missing at random. We therefore use the population statistics and apply inverse probability weighting techniques to adjust for the non-random missing values under some assumptions about the missingness mechanism. We further propose a method without those assumptions based on partial identification, although it is conservative and computationally intensive.

The VA SOLID study will test the utility of a Contextual Multi-Arm Bandit (CMAB) design for improving patient outcomes in the surgical treatment of spinal stenosis across multiple VA systems. Surgical treatment of spinal stenosis is broadly described as decompression of trapped neural components) alone or in conjunction with spinal fusion (to stabilize the spine). While decompression alone is associated with lower morbidity and cost, both failure to relieve symptoms and the possibility of post-operative spinal instability encourage the addition of spinal fusion. The Stanford statistical advisory group is responsible for proposing a specific method for outcomedependent, contextual adaptation of randomized assignment of treatment in order to (1) improve within-sample average patient outcome and (2) generate reliable decision rules (based on patient context) for future treatment decisions. As a first step, Michael Sklar has completed a review of technical publications that might inform design of a CMAB for medical/surgical decision-making, and begun work on simulations of the performance of the leading contenders.

Acute leukemia is the most common cancer in children and teens, accounting for almost 1 out of 3 cancers. Despite recent advances in the treatment of childhood leukemia, a substantial proportion of patients are resistant to conventional treatment. Allogeneic hematopoietic stem cell transplantation (HSCT) is effective for high risk or refractory disease but is limited by a lack of HLA-matched donors. Haploidentical HSCT after αβT-cell/CD19 B-cell depletion (αβhaplo-HSCT) can address this issue. It is especially valuable for patients of African-American or mixed heritage because this demographic has a significantly lower likelihood of finding a matched donor. αβhaplo-HSCT has extremely low rates of transplant-related mortality (TRM) and severe graft-versus-host disease (GvHD). However, the rate of leukemic relapse (25–30%) after αβhaplo-HSCT remains high. Thus, there is a clinical need to improve the anti-leukemic efficacy of αβhaplo-HSCT without increasing the risk of GvHD. We are planning a Phase I/II clinical trial combining αβhaplo-HSCT with T-allo10 cell infusion. The donor cells in the T-allo10 infusion are processed to strengthen immunosuppression thereby decreasing the risk of GvHD and to promote graft-versus-leukemia (GvL) activity. The primary objective of the Phase I part is to identify the safe and effective dose of T-allo10 cells (range 1 ×105 to 1×106 cells/kg recipient body weight) which can be infused in patients without increasing risk of GvHD. The primary objective of the Phase II part is to evaluate the improvement of the adaptive immune reconstitution defined as an increase of absolute CD4 cell count at Day 100 compared with historical controls.

Congenital birth defects affect 3% of births in the United States. Several parental factors, such as smoking and increasing age, been shown to modify the incidence. Given the longer and more obvious influence a mother has on gestation, most risk factors for birth defects have focused on women. However, as the age of paternity at conception increases around the developed world, paternal comorbidities increase. With that, more prospective fathers are exposed to prescription medications during spermatogenesis cycles, which lead to live births. Our ability to adequately counsel patients regarding paternal exposures to medicines is limited by sparse data. Investigators have performed studies on limited numbers of patients exposed to certain medications strongly suspected of impacting sperm DNA (e.g. anti-neoplastics). Some common medications, such as antidepressants, may affect the DNA integrity of sperm. To date, large studies of diverse medications are rare despite the crucial role the sperm contributes to fertilization and the first few days of development. This project explores the association between paternal medication use and adverse birth outcomes, including congenital birth defects. Using unique national identification numbers, we will link Danish demographic, birth, health, and medication registries for the approximately 1.3 million births from 1995 to 2015. Next, we will determine if individual medications or classes of medications with known spermatogenic effects taken by fathers just before conception are associated with a higher risk of congenital birth defects. Such findings are likely to have wide implications for couples attempting to conceive and provide guidance for the safety of medications during reproductive efforts.

Clinical trials use strict eligibility criteria to exclude patients who may have poor outcomes. Those criteria are designed based on physician’s expectation and are usually duplicated across protocols without solid clinical validation. Unnecessarily stringent criteria have been a large barrier in trial accrual—the enrollment is significantly slowed down and the generalizability of results is limited. Relaxing the eligibility criteria has been a popular advocate but how to really do it remains underexplored. In this work, we emulate oncology cancer trials from observational data and systematically analyze how different eligibility criteria can affect a trial’s final performance. Our results validate the necessity to broaden overly restrictive enrollment because patients excluded by the original criteria often derive similar clinical benefits as the enrolled ones. We propose a principled framework to encode different eligibility criteria and emulate corresponding trials. Given standard inclusion and exclusion criteria, we are able to evaluate the influence of each criterion and recommend an optimal strategy to relax them. It is a powerful tool to identify potential label expansions for existing trials and facilitate the design of eligibility criteria for future trials. It can also be used to generate synthetic external control arms with its unified pipeline to conduct cohort selection on observational data.

Preeclampsia (PE), a pregnancy-related placental vascular disorder affecting 5–8% of all pregnancies, is thought to be a multisystem disorder of pregnancy driven by alterations in placental function and resolved by the delivery of the placenta and fetus. Untreated PE can cause serious health problems for both mothers and babies. Early detection of risk of developing PE remains a challenge in clinical settings. The ratio of soluble fms-like tyrosine kinase (sFlt-1) and placental growth factor (PIGF) levels has been proposed as a useful index to diagnose and manage PE. However, the ratio only works at mid or late gestation. Previous transcriptomic and proteomic profiling of normal and complicated pregnancies have identified disease-specific expression patterns and signaling networks, which suggests candidate biomarkers for possible early clinical assessment. We hypothesized that there are maternal serological biomarkers differentiating impending PE from normal pregnancy at early stage of gestation. We quantified proteins and lipids in maternal serum by ELISA and LC MS/MS methods, and evaluated the performance of these candidate markers in predicting PE on an age-matched discovery cohort (32 PE subjects and 32 controls) and a longitudinal testing cohort (20 PE subjects and 20 controls). Results showed that a ratio of an elevated marker and a decreased marker differentiated women with impending PE from normal pregnant women at 1st and early 2nd trimesters, which is earlier than the well-established biomarkers (sFlt-1 and PlGF). Our finding may offer a new investigational approach towards the understanding of biology during pregnancy as well as guiding innovative methods for PE assessment.

Each year in the United States, 1.8 million new patients are diagnosed with cancers. Beyond this coverage by the NCI SEER registry, 3.3 million additional new patients are diagnosed with nonmelanoma skin cancers in US annually. Even a moderate improvement in earlier detection (e.g., 6 months) could have meaningful benefits for survival and costs. The cancer development has been intensively investigated using samples collected after diagnosis. However, cancer development prior to clinical presentation remains mostly unobserved due to scarcity of samples. At Stanford, we have created a prospective longitudinal cohort of individuals followed for up to 10 years. Immune profiling of these individuals is systematically characterized through annual visits. During the period of observation, 48 incidences of malignancies were recorded (67% of them were non-melanoma skin cancers, consistent with the overall US population). Thus, the immune profiling of blood samples collected prior to the cancer diagnosis offers a rare opportunity to understand changes in the immune system during this crucial period.

In these samples, we found that the average concentrations of Cancer-Preceding (CanDx) cytokines were already elevated up to 3 years before the clinical diagnosis. Even at this time point, the impact of cancers on the immune system is large. It is comparable or more significant than the impact associated with over 50 years of aging. We also observed that average concentration of CanDx cytokines decreased immediately in the years following the cancer diagnosis. This is consistent with the fact that most of the incidences of cancers in our cohort, as well as the general US population, were basal cell carcinoma (BCC) and squamous cell carcinoma (SCC). These are nearly always surgically removed soon after diagnosis. Lastly, we also found evidence suggesting that a tilted inflammation balance in aged population may contribute to accelerated cancer development.

Our current dataset is not powered for detailed analyses on cancer subtypes other than nonmelanoma skin cancers (BCC/SCC). We are expanding our current study cohort by both active recruitments targeting cancer-prone populations and large-scale electronic health record analyses. Our aim is to examine the role of immunity in early cancer development beyond skin cancers.

We hypothesize that there are different sex-specific microbial signatures linked to regional adiposity in men and women. Such differences could be largely attributable to the biological functions of sex hormones and fatty acids. Sex hormones have been shown to be associated with both regional adiposity and gut microbiota. Fatty acids from the lipid pathways have shown pronounced effect in the mechanism of determining the adiposity site, whereas amino acids were shown to play catalyst roles in this process. To examine the above hypothesis, our current study has two aims. First, identify the sex-specific microbial compositions that associate with fat distribution in men and women. Second, identify the metabolites that could be potential mediators to the association between the microbial signature and fat distribution. To achieve these two aims, we conducted the following analyses separately in women and men. First, we univariately screened the metabolites that are associated with fat distribution by sex. Second, we explored the sex-specific microbial signatures using the Wald test. Third, we implemented sparse partial least square discriminant analysis to synthesize the joint distribution across microbiome, metabolites, and fat distribution. Fourth, we conducted mediation analysis to identify the metabolic mediators for the association between microbiome and fat distribution.

Hepatocellular carcinoma (HCC) is a common and aggressive form of liver cancer that results in 800,000 deaths globally each year. In the US, HCC accounts for the fastest-rising cause of cancer mortality. While various treatment options are available for HCC ranging from different types of surgeries (e.g. liver transplant, chemoembolization, resection, ablation) to different types of chemotherapies (e.g. systemic chemotherapy, chemotherapy infusion), selection of patient-specific treatment options for HCC remains challenging.

As an initial attempt to investigate this question, we study a cross-sectional dataset of 2,625 HCC patients derived from the Stanford University Medical Center. We estimate the conditional average treatment effect (CATE) using causal random forests on a subset of 206 patients that received systemic chemotherapy as first-line treatment (controlled group) versus another subset of 953 patients that received chemoembolization as first-line treatment (treatment group). We adjusted for covariates including demographics, TNM staging covariates, lab test values, and liver disease history. We observe a CATE of 13.0% ± 4.0% absolute difference in the three-year survival probability between the treatment group and the controlled group.

Our team has recently discovered a general, safe, and remarkably effective concept for RNA delivery based on a new class of synthetic cationic materials, Charge-Altering Releasable Transporters (CARTs). These new materials operate by an unprecedented mechanism for the delivery and transcription of proteins in both cell culture and live animals. The function and selectivity of these materials has proven to be remarkably sensitive to aspects of the chemical structure, leading to (for instance) in vivo selective accumulation changing from spleen (> 99%) to lung (75%) upon changes in chemical structure. Due to the modular design of these materials and our ability to rapidly synthesize new candidates, there is a vast chemical space to explore. Our hope is to apply statistical design of experiment (DoE) methods to explore this space efficiently and establish meaningful structure-function correlations to guide our understanding of the remarkable efficacy of these materials. Further convoluting the possible space to explore, we have found that mixing chemically different CARTs in a single transfection often leads to superior function over either component individually, sometimes dramatically superior.

Opioid-sparing analgesic strategies are desperately needed for pregnant women at risk for prolonged pain and opioid use after surgical delivery. This is the largest iatrogenic exposure of young women to opioids in the United States. Each year, approximately 1.2 million American women undergo the most common surgical procedure. Up to 11% will persist with analgesic use, reporting prolonged pain and reduced function.

I propose a double-blind, randomized clinical trial of gabapentin which is titrated over 12 weeks compared to placebo. The primary outcome is time to opioid cessation. Secondary outcomes are pain resolution, return of physical function, and psychological wellbeing. I have enrolled over 30 subjects with 2 lost to follow-up before achieving the primary outcome variable of time to opioid cessation.

My need for statistical advice concerns the statistical analysis plan and the sample size calculation. The study is designed with a time to event outcome. In my current plan, event times are measured every 2 weeks. However, most of the events occur in the 26 week range. I also have count data measured at each of the 2-week measurements but need help to incorporate these into the study design. The patient population is relatively limited. Therefore, I need help finding an appropriate design that will be sufficiently powered.

Organic forms of mercury (Hg) are ubiquitous contaminants in fish, but the toxic health effects of mercury are mitigated by binding with the essential trace metal selenium. Selenium (Se) is also a required cofactor for enzymes that manage oxidative stress and for thyroid hormone metabolism. However, functional deficiency in these enzymes may not be apparent by measuring selenium concentrations or using comprehensive thyroid hormone assessment panels. Chronic Fatigue Syndrome, also known as Myalgic Encephalomyelitis (ME/CFS), is a complex multi-system disorder that is characterized by increased levels of oxidative stress, neuroinflammation, disturbances in energy metabolism, and other nonspecific markers of illness. We have identified a subset of ME/CFS patients with elevated levels of mercury in hair and a molar ratio of Se:Hg less than 1, possibly indicating lowered bioavailability of selenium in vivo. Our findings may explain the recent report of a Low Free T3 Syndrome in a subset of ME/CFS patients, as free T3 levels are dependent on selenium for thyroxine (T4) deiodinase activity.

We need help with the study design and statistical analyses needed to address the following scientific questions.

1. ME/CFS patients

(a) Does the Se:Hg ratio correlate with selenoenzyme activity?

(b) Does selenoenzyme activity correlate with phenotypes defined by two different diagnostic standards?

2. Non-ME/CFS controls

(a) Does the Se:Hg ratio correlate with selenoenzyme activity?

3. Is there a way to tease apart the causal direction between diet and a subgroup of ME/CFS patients?

In renal cell carcinoma (RCC), the 7th most commonly diagnosed cancer in the U.S., time of diagnosis has a profound impact on survival. For stage I patients (localized non-invading tumor < 7 cm), the 5-year overall survival rate is 96%, compared to 23% for stage IV (metastatic) patients. There are currently no screening assays for RCC, hence it is often found incidentally and ˜30–40% of patients are still diagnosed with poor-prognosis, advanced (stage III or IV) RCC, which highlights the need for early detection and practical screening assays.

Once a small renal mass has been found by imaging, the treatment decision for surgery versus surveillance is made based on tumor features (histologic subtype, growth rate, tumor grade, which is a measure of tumor aggressiveness) and condition of the patient (age, comorbidities). However, information on tumor grade cannot be readily obtained through imaging or tumor biopsies, hence there is a need for a surrogate marker for tumor grade to aid in diagnosis and clinical management of small renal masses.

Platelets splice pre-mRNA deposited by megakaryocytes, which is the bone marrow cell type from which platelets are derived, splice it into mature mRNA, and translate it into protein. While the platelet transcriptome is remarkably stable over time in healthy individuals, platelets have been shown to respond to various disease states, including cardiovascular and autoimmune disease, and more recently, cancer, with likely disease-specific changes in the platelet transcriptome. However, platelet transcriptomic changes have never been specifically investigated in RCC or in early stage cancer or in high-grade tumors.

Maternal hemorrhage is globally the leading cause of maternal mortality. Furthermore, 80% of postpartum hemorrhage is caused by uterine atony which is a failure of the uterus to contract after separation of the placenta. Current drugs used to address uterine atony refractory to standard management with oxytocin are limited by cost, adverse side effect profiles, and common contraindications such as hypertension and asthma.

Calcium chloride is a widely-available, shelf-stable, cheap, and well-tolerated drug. Calcium is a key element in uterine muscle contractility. The mechanism of action for all drugs currently used for uterine atony depends upon myometrial calcium. Multiple studies utilizing uterine muscle strips from pregnant animals and humans have shown that increased extracellular calcium can improve strength and frequency of uterine contractions. Finally, many experienced obstetric anesthesiologists anecdotally use calcium to address cases of refractory uterine atony. However, there are no clinical studies that have examined the role of calcium chloride in this setting.

This pilot study is aimed to elucidate the potential role of calcium chloride plus standard of care for prevention of uterine atony in women at high risk undergoing Cesarean section. The design of this interventional pilot clinical trial is safety and efficacy with double-blind masking and parallel assignment via randomized allocation to either treatment or placebo for the purpose of prevention of uterine atony. A total of 40 parturients undergoing Cesarean with 2 or more accepted risk factors for uterine atony were block-randomized in a one-to-one ratio to receive an intravenous infusion of either calcium chloride or placebo at the time of fetal delivery. The primary aim is to determine if a single dose of calcium chloride compared to placebo administered immediately after fetal delivery can reduce the incidence of uterine atony in women at risk. The secondary aims are to determine whether intravenous calcium chloride reduces blood loss, improves maternal hemodynamics, and is well-tolerated with a favorable side-effect profile. The primary outcome is dichotomous: presence or absence of clinical uterine atony. The secondary outcomes include quantitative blood loss, subjective assessment of uterine tone from 0% (no tone) to 100% (excellent tone) by the obstetrician immediately after completion of study drug infusion, and change in hematocrit defined as preoperative baseline minus postpartum day 1.

FIRST SESSION: 

Platelets are the blood component responsible for clot formation. The greatest demand for this blood product comes from trauma, oncology, and major surgery patients. Hospitals must always maintain a sufficient inventory of platelets. Platelets have a shelf life of only five days but spend two in testing for bacterial and viral contamination. This means that hospitals have only three days to use platelets upon receiving them. This constraint, along with highly variable demand, a smaller donor pool, longer donation process, and less collection capacity due to the cost of specialized equipment, makes platelets the most difficult blood product for hospitals to manage. It is estimated that nationwide, 11% of all collected platelet units, or 250,000 units, expire before being transfused, amounting to $80 million lost and 500,000 donor-hours wasted. In 2017, the Stanford Blood Center sought to address this wastage problem by engaging a team of researchers to forecast platelet demand using daily aggregated hospital data. They formulated a three-day ordering strategy by solving an optimization problem using features aggregated from the available data. However, the ordering strategy was not put into production due to lack of trust in model predictions by the inventory manager. In order to address this problem, we decided to separately predict platelet demand, and develop ordering strategy as a secondary step. The goal of our current study is to supply the user with a decision support tool which will build confidence in the accuracy of computer predictions. Our four-day daily prediction is based on both aggregate data and patient-level information. Our questions concern how to improve the prediction models.

SECOND SESSION:

Improved understanding of the relationship between altered hemodynamics and cerebral oxygenation may inform future strategies to prevent brain injury in preterm infants. The objective of this study is to determine whether decreased cerebral oxygenation or altered cerebral autoregulation in the first 96 hours of life is associated with mortality or severe neuroradiographic abnormalities in preterm infants. The study design is a prospective cohort over multiple centers. We enrolled 111 preterm infants from 6 neonatal intensive care units. These preterm infants had birth weight less than 1250 g during the first 24 hours of life and received simultaneous monitoring of mean arterial blood pressure and cerebral oxygenation using near-infrared spectroscopy until at least 96 hours of age. The primary outcome measure is mortality before hospital discharge or severe neuroradiographic abnormalities. Among the enrolled preterm infants, 103 had sufficient spectroscopic data for analysis. We computed positive or negative moving window correlations between cerebral oxygen saturation (Csat) and mean arterial blood pressure (MAP) to determine time periods with altered cerebral autoregulation. We compared percentage of time with altered autoregulation using Spearman rank correlation between infants with and without adverse outcome. We have specific questions about how to analyze the longitudinal data collected from these preterm infants.

Concussion affects millions of children in the United States each year with recovery times averaging 4 weeks, subsequently impacting quality of life. Current acute concussion care is solely supportive care with a gradual return to activity and is not based on newer concussion subtype management. While a proportion of children with head injury present to the pediatric emergency department (ED), these often represent the most severe presentations representing the most ideal therapeutic target.

We hypothesize that children diagnosed in the pediatric ED can be feasibly classified into concussion subtypes and directed to acute subtype-based targeted therapies with shorter recovery times as compared to those who receive standard of care. We propose an observational study on epidemiology followed by a cross-sectional trial for interventional feasibility. Our questions concern study design and data analysis methods.

Individuals with narcolepsy suffer from abnormal sleep patterns due to loss of neurons that uniquely supply hypocretin (HCRT), a wakefulness-promoting hormone in human brains. Previous studies found the associations of narcolepsy with human leukocyte antigen (HLA)-DQ6 allele and T-cell-receptor-alpha (TRA) J24 gene segment. These led to an autoimmune hypothesis for narcolepsy that the presentation of HCRT-derived peptides by DQ6-expressing antigen presenting cells to TRAJ24-expressing T-cells drives T-cell clonal expansion and downstream self-attacking immune responses against HCRT-producing neurons. Although T-cells from narcolepsy patients were recently reported to recognize HCRT peptides, autoimmunity-related evidence that HCRT-reactive T-cells have experienced in vivo clonal expansion is still absent.

We directly sequence T-cell-receptor (TCR) and selected phenotypic marker genes in over 5000 single ex vivo T-cells that are isolated from DQ6+ donors using DQ6-HCRT peptide tetramers. Statistical comparisons suggest that DQ6-HCRT(87–100) tetramers, a peptide composed of amino acids 87–97 from HCRT, identify significantly more expanded CD4+ T cells from 8 patients than from 8 DQ6 allele-matched healthy controls. We also identify related TRAJ24+ TCR clonotypes encoded by identical genes from 2 patients and 2 controls. TRAJ24-G-allele+ clonotypes only expand in the two patients, whereas a TRAJ24-C-allele+ clonotype expands in a control (the G/C SNP in TRAJ24 gene is narcolepsy-associated). In addition, clonally expanded TRAJ24-G-allele+ T-cells from the two patients exhibit an unconventional effector phenotype. Our identification of features distinguishing in vivo expansion of HCRT-specific TRAJ24+ T-cells in pat

This project is studying the natural trajectory of various biomarkers during pregnancy. The ultimate goal is to choose a panel of selected biomarkers that can be monitored in order to identify women with higher risk of miscarriage or premature delivery.

This research is focused on studying systemic juvenile idiopathic arthritis (sJIA), a pediatric autoinflammatory disease of unknown etiology. Diagnosis can often be challenging, and not all patients with sJIA respond optimally to biologic cytokine inhibitors, highlighting two of the many unmet needs in understanding and treating this disease. I am working with plasma samples and implementing high-resolution flow cytometry to characterize the differences in extracellular vesicle (EV) populations (based on surface marker expression) present in active relative to quiescent disease states as well as compared to healthy controls. The overall objective of this study aims to identify EV signatures that distinguish active and quiescent states of sJIA, which could reveal novel biomarkers with exciting diagnostic, prognostic, and therapeutic potential.

The Nulliparous, Term, Singleton, Vertex (NTSV) Cesarean Birth Rate measures the proportion of cesarean births among a population of women deemed to be at very low risk of needing a cesarean section (CS). Distributions of many of the risk factors (such as obesity, age and gestational diabetes) may still differ significantly between providers and the adjustment may result in different comparison results. The outstanding question is whether those factors should be taken into account when comparing CS rates between hospitals. If so, what is the most appropriate statistical method for adjustment?

The Williams PanLab is designing the first prospective trial of depression using brain activity from fMRI in particular regions to guide subjects to treatment. We are inspired by innovative clinical trial designs used in cancer medicine but have questions about the most appropriate design and statistical approaches given the nature of our data and the questions we are trying to answer.

We are interested in the question of whether adverse pregnancy outcomes (APOs) in women, like gestational diabetes, are risk factors for developing heart disease in later life. We are working with the Women's Health Initiative (WHI) data for the study. This is a large cohort of women (about 161,000) enrolled about 20 years ago, with extensive baseline data collection and scrupulous follow-up for detecting the development of heart disease. So the cohort is ideal for our study, BUT the investigators didn’t collect baseline data on APOs, the exposure of interest. So we did a survey in 2017 to collect data on APOs from about 51,000 women, and are trying to devise the best analysis strategy to relate APOs to heart disease. Our "missing data problems" range from small (some of the 50,000 respondents did not answer all the survey questions, but did answer some of them) to very large (we are missing baseline APO data on about 110,000 women, many of whom have died from heart disease or other causes).

Unobserved technical and biological factors can affect RNA-seq data, resulting in both false positives and loss of power when testing for differential expression between groups.  This problem is widespread and commonly handled with surrogate variable analysis (SVA), using the sva or SmartSVA package in R to estimate the number of latent factors as well as the factors themselves.  Unfortunately, these algorithms may fail under some conditions such as strong correlations between the factors and the variable being tested for differential expression.  Newer packages such as cate address these issues.  I will discuss both theory and practical implications and give examples from real world data. I will make the slides and sample code publicly available, although the data cannot yet be released. For slides and R code, click here.

This will be an interactive session between the members of Cancer Clinical Trials Office (CCTO) at Stanford and Department of Biomedical Data Science (BDS) to explore processes that might expedite NIH reporting for clinical trials.  Neal Birkett from CCTO will speak about reporting workflow in use and present some examples. Balasubramanian Narasimhan from BDS will talk about some R based tools that may be useful in this context using a recent report he created as example. Time permitting, we will discuss the possibility of piloting some reporting templates and associated issues.

The integrative, precision health-based science and inter-departmental collaborations fostered at the Stanford School of Medicine have created transformative change in preventing and predicting allergic and asthma disorders.The Sean N. Parker Center for Allergy and Asthma Research focuses on finding the cause and cure of allergies and asthma. The center has been very successful in basic and clinical research, including 20 clinical trials in 2018, most in phase 1 and 2 but some of phase 3 studies. We have worked with biostatisticians from QSU.

Recent medical discoveries in allergic and asthma disorders are similar to oncology that we have more precise disease types and multiple drugs are being developed. We are interested in using precision approaches, such as basket trials, platform trials, and master protocols to accelerate our clinical trial research. We plan to discuss a phase 2 trial on managing gastrointestinal adverse events in patients undergoing milk protein oral immunotherapy. We would seek inputs on design features to maximize the utility of this planned phase 2 trial. Background information on this particular trial and related clinical trial challenges in allergic and asthma disorders in general can be found from the following literature.

After sampling a subset of cells from a population, the cells may be classified into different types based on next generation sequencing. We are interested in finding the number of different cell types in the original population based on the sequencing results. There are two complications: (1) we don't observe the cell type of the sample cells directly and we only observe sequenced DNA segments after PCR amplification; and (2) the sequencing process is prone to errors.

The advent of anti-IL1/IL6 biologics transformed the management of systemic juvenile idiopathic arthritis (sJIA), a debilitating early-onset auto-inflammatory condition.  However, in recent years, the rheumatology community started to spot cases of interstitial lung disease (ILD) within the sJIA population, a complication rarely seen in previous decades.  To define its clinical presentations and identify potential risk factors, we organized a multi-center retrospective study to initiate a comprehensive investigation on the sJIA-ILD. We found that the complication presented several distinctive clinical features, such as acute clubbing (61 % prevalence), unique radiological findings and pulmonary alveolar proteinosis in biopsy (PAP, 62 %).  

However, whole-exome sequencing on 20 patients ruled out the conventional congenital PAP in most cases. Thus, the etiology of the disease is not clear, while its mortality cost is high (75% within 6 years). To elucidate on the relationships between the clinical features, we performed a correlation analysis within the sJIA-ILD cohort. We found a history of exposure to anti-IL1/6 is associated with various distinctive clinical features, including PAP pathology.  

Meanwhile, the early onset age of sJIA (< 5 years) was strongly associated with pathological findings of PAP in ILD later (OR = 20). When compared to sJIA not developing ILD, sJIA-ILD tended to be significantly younger only in the anti-IL1/6 exposed group but not in the non-exposed group. They also suffer from unusually high frequency of serious adverse reaction to Tocilizumab (40%, >100 times higher than the background). In addition, patients with both sJIA and Down’s Syndrome also have elevated risk for ILD.

The augmented risk of young age by pre-exposure status to anti-IL1/IL6 biologics indicates a potential developmental cause for the disease. In clinical practice, we recommend to routinely monitor potential ILD symptoms (such as digital clubbing and non-resolving respiratory symptoms) among early-onset sJIA (< 5 years old) patients treated with anti-IL1/IL6 biologics, as well as Down Syndrome patients with sJIA.

To devise predictive/preemptive big data solutions for population health, we have employed a comprehensive unbiased multi-’omics’ approach, integrating mass spectrometric data production of metabolomics, lipidomics, peptidomics, and proteomics to define the disease associated molecular “portrait” and relative health risk against the population baseline.  With the statistical learning of these mass spec data sets, we aim to provide a more technically and bioinformatically tractable, physiologically relevant, chemically comprehensive, and cost effective assessment of multi-factorial non-communicable diseases. However, analytical challenges are obvious when the needs go beyond the qualitative understanding of the underlying biology but to report the quantitative relative/absolute disease risks.  In this Data Studio, case studies will be presented and suggestions will be solicited.

Accurate and early detection  of breast cancer and relapse is a critical part of the strategy to reduce the morbidity and mortality associated with it.  While some guidelines recommend mammography for screening, the sensitivity, specificity and patient acceptance of mammograms remains less than optimal. Thermography has the potential to become a reliable diagnostic adjunct for breast cancer screening due to recent advances, including the use of machine learning to enhance its diagnostic accuracy.  Achanrya's Support Vector Machine (SVM) model classified normal and malignant breast tissues from a thermogram with 88% accuracy. In this Data Studio, the investigators will identify and seek advice on aspects that need further refinement. Recent advances and suggestions for future work in the field include adding advanced simulation methods, inverse modelling, imaging protocols, and the introduction of artificial neural networks to predict the location and activity of tumours.

“It is important to understand what you CAN DO before you learn to measure how WELL you seem to have DONE it” (Exploratory Data Analysis: John Tukey, 1977). Data analysts and statistics practitioners nowadays are facing difficulties in understanding higher and higher dimensional data with more and more complexity while conventional graphics/visualization tools do not answer the needs. It is a more difficult challenge to understand overall structure in big data sets, so good and appropriate Exploratory Data Analysis (EDA) practices are going to play more important roles in understanding what one can do in the big data era.

Matrix Visualization (MV) has been shown to be more efficient than conventional EDA tools such as box plots, scatter plots (with dimension reduction techniques), and parallel coordinates plots for extracting information embedded in moderate to large data sets of binary, continuous, ordinal, and nominal nature. In this study we plan to investigate feasibility of and potential difficulties in applying MV techniques to visualizing and exploring structure in big data: 1) memory/computation (permutation with clustering) of proximity matrices for variables and subject; 2) display of data and proximity matrices for variables and subject. We shall integrate techniques from Hadoop computing environment, image scaling, and symbolic data analysis into the framework of GAP (Generalized Association Plots) to develop an appropriate package for conducting Big Data EDA with visualization. Several international data sets (HCAHPS 2013 Data, MS-DRG 2011 Data) and local data sets (National Health Insurance Data) will be used to demonstrate the bigGAP environment.