Funded Projects


AI Approaches for Diagnostic Error: Improving Patients Lives

  Internal Funding, Details Coming

 

Individual-level predictive models for management of pain 

Digital Health Cooporative Research Centre and Healthcare Management Solutions



Improving Quality of Postoperative Pain Care Through use of Electronic Health Records

Agency for Healthcare Research and Quality (AHRQ) (09/01/2015 - 6/30/2020)

Millions of Americans undergo surgery every year and postoperative pain is common and often poorly managed. Poorly managed postoperative pain may cause severe functional impairment, impaired care of the underlying diseases, transition to chronic pain, and decreased quality of life. Many controlled studies have demonstrated a variety of interventions that benefit postoperative pain, yet their application in a large and more diverse population is unknown and a nationally endorsed, concise quality process metric for postoperative pain management does not exist. One roadblock is that postoperative pain and its related outcomes are complex. The gathering of evidence from electronic health data, which draw from and inform real-world practice, could bypass this roadblock and inform decisions that lead to effective and efficient postoperative pain management. This project seeks to measure quality of care for postoperative pain, assess proposed evidence-based interventions from randomized controlled trails, lay the ground work for systematic pain-related research using EMRs, and produce population-based evidence for a nationally-endorsed postoperative pain management quality metric. To achieve these objectives, this project has three specific aims: (1) to develop standardized electronic definitions of pain-related care processes and outcomes (e.g. prolonged opioid use, readmission for pain, etc.); (2) to extract clinically meaningful data from both structured data and free text in electronic medical records (EMR) and examine the relationship between recommended care processes and outcomes for postoperative pain using EMRs; (3) to validate pain-related process-outcome relationships at a national level and to develop a National Quality Forum submission and evaluation form for a postoperative pain quality metric(s). This project will achieve these aims by developing data capture algorithms on Palo Alto Veterans Administration (VA) Healthcare data, refining algorithms at a tertiary academic hospital, and validating algorithms on the National VA healthcare system. Data will be identified and extracted from the EMR using an extended version of our validated data-mining workflow. Established experience with quality metric development and NQF endorsement will facilitate the dissemination of this work. These approaches are the basis of a learning healthcare system and the proposed research directly aligns with AHRQ’s mission and goals to improve healthcare quality through health information technology and data resources.

Selected Results

Predicting inadequate postoperative pain management in depressed patients: A machine learning approach. Parthipan A, Banerjee I, Humphreys K, Asch SM, Curtin C, Carroll I, Hernandez-Boussard T. PLOS One 2019. Feb 6 https://doi.org/10.1371/journal.pone.0210575

Utilization and effectiveness of multimodal discharge analgesia for postoperative pain management. Desai K, Carroll I, Asch SM, Seto T, McDonald K, Curtin C, Hernandez-Boussard T. August 2018, 228:160-169

Drug-Free Interventions to Reduce Pain or Opioid Consumption After Total Knee Arthroplasty: A Systematic Review and Meta-analysis. Tedesco D, Gori D, Desai KR, Asch S, Carroll IR, Curtin C, McDonald KM, Fantini MP, Hernandez-Boussard T. JAMA Surg. 2017 Oct 18;152(10):e172872.

Defining Postoperative Opioid Needs Among Preoperative Opioid Users.  Hah J, Hernandez-Boussard T.  JAMA Surg. 2018 Jul 1;153(7):689-690. doi: 10.1001/jamasurg.2018.0217. 

 

The Fifth Vital Sign: Postoperative Pain Predicts 30-day Readmissions and Subsequent Emergency Department Visits. Hernandez-Boussard T, Graham LA, Desai K, Wahl TS, Aucoin E, Richman JS, Morris MS, Itani KM, Telford GL, Hawn MT. Ann Surg. 2017 Sep;266(3):516-524. 

Opioid Abuse And Poisoning: Trends In Inpatient And Emergency Department Discharges. Tedesco D, Asch SM, Curtin C, Hah J, McDonald KM, Fantini MP, Hernandez-Boussard T. Health Aff (Millwood). 2017 Oct 1;36(10):1748-1753.

Collaborators

Professor of Medicine (Primary Care and Population Health)
Associate Professor of Anesthesiology, Perioperative and Pain Medicine (Adult Pain) at the Stanford University Medical Center
Professor of Surgery (Plastic & Reconstructive Surgery) and, by courtesy, of Orthopaedic Surgery at the Stanford University Medical Center

Utilizing Electronic Health Records to Measure and Improve Prostate Cancer Care

National Cancer Institute (NCI) (7/1/2015 - 6/30/2020)

Prostate cancer is the most common malignancy in men. Newly diagnosed men face complex treatment choices, each with different risks of acquired morbidities, including patient-centered outcomes (PCOs). The widespread implementation of electronic health records (EHRs) provides opportunities to incorporate PCOs into healthcare quality metric evaluations. However, efforts to assess quality metrics in EHRs have been limited because most relevant data are not reliably captured in structured formats. Our proposal innovates in three ways. First, we will develop an EHR prostate cancer database that will allow for clinical care data to be analyzed alongside diagnostic details. Second, we will create novel ontological representations of quality metrics that will be public and reliably calculable across EHR-systems. Third, we will assemble a robust data-mining workflow that expands on existing quality assessment methods by focusing on ontology-based dictionaries to annotate free text. Combining this set of innovative components will uniquely allow us to use existing EHRs to efficiently study the association between treatment processes and outcomes. Our methods are applicable not only to prostate cancer, but any disease with associated quality metrics. Our primary hypothesis is important prostate cancer PCOs will differ significantly across treatments (i.e. robotic surgery, open prostatectomy, and radiation therapy). To gather data to test our hypothesis, we assemble a data-mining workflow to extract quality metrics from both structured and free-text components of EHRs. 

Selected Results

Comparison of Orthogonal NLP Methods for Clinical Phenotyping and Assessment of Bone Scan Utilization among Prostate Cancer Patients. Coquet J, Bozkurt S, Kan KM, Ferrari MK, Blayney DW, Brooks JD, Hernandez-Boussard T.J Biomed Inform. 2019 Apr 20:103184. 

Weakly supervised natural language processing for assessing patient-centered outcome following prostate cancer treatment.Banerjee I, Li K, Seneviratne M, Ferrari M, Seto T,  Brooks JD, Rubin DL, Hernandez-Boussard T. JAMIA Open, ooy057, https://doi.org/10.1093/jamiaopen/ooy057

Utilization of Prostate Cancer Quality Metrics for Research and Quality Improvement: A Structured Review.  Gori D, Dulal R, Blayney DW, Brooks JD, Fantini MP, McDonald KM, Hernandez-Boussard T. Jt Comm J Qual Patient Saf. 2018 Sep 18. pii: S1553-7250(18)30094-1. 

Distribution of global health measures from routinely collected PROMIS surveys in patients with breast cancer or prostate cancer.  Seneviratne MG, Bozkurt S, Patel MI, Seto T, Brooks JD, Blayney DW, Kurian AW, Hernandez-Boussard T.  Cancer. 2018 Dec 4. doi: 10.1002/cncr.31895.

Architecture and Implementation of a Clinical Research Data Warehouse for Prostate Cancer. Seneviratne M, Seto T, Blayney D, Brooks J, Hernandez-Boussard T. eGEMs (Generating Evidence & Methods to improve patient outcomes) . 2018; 6 (1) :13 .

Mining Electronic Health Records to Extract Patient-Centered Outcomes Following Prostate Cancer Treatment. T Hernandez-Boussard, P Kourdis, T Seto, M Ferrari, D. Blayney, J.D. Brooks. AMIA Annu Sypo Proc. 2018 Apr 16; 2017:876-882. eCollection 2017.

A natural language processing algorithm to measure quality prostate cancer care. T Hernandez-Boussard, P Kourdis, R Dulal, M Ferrari, S Henry, T Seto, K. McDonald, D. Blayney, J.D. Brooks. 2017. Journal of Clinical Oncology 35 (8_suppl), 232-232

New Paradigms for Patient-Centered Outcomes Research in Electronic Medical Records: An Example of Detecting Urinary Incontinence Following Prostatectomy. Hernandez-Boussard T, Tamang S, Blayney D, Brooks J, Shah N. EGEMS (Wash DC). 2016 May 12;4(3):1231.

Collaborators

Professor of Medicine (Oncology) at the Stanford University Medical Center
Associate Professor of Medicine (Biomedical Informatics) and of Biomedical Data Science
Keith and Jan Hurlbut Professor in the Stanford University School of Medicine
Professor of Biomedical Data Science and of Statistics

Machine Learning Classifiers for Automated Staging of Prostate Cancer Patients

Stanford-AstraZeneca grant (09/01/2017-10/31/2018)

Prostate cancer is the most common malignancy in men, with an estimated 21% prevalence among new cancer cases in males for 2016. The staging of newly diagnosed prostate cancer is one of the most important factors in determining treatment options and predicting patient survival. Prostate staging is divided into clinical and pathological stage. Clinical stage is the physician’s estimate of the disease level, based on results from digital rectal exam (DRE), biopsy, lab tests, and imaging. The pathological stage includes information from the clinical stage plus surgical results (e.g. unknown metastasis). Manual documentation of clinical and pathological stage of cancer is extremely labor-intensive. When staging is not recorded in the health record, care coordination is jeopardized.  Therefore, we propose a tool to automate the extraction and classification of data for prostate cancer staging that would benefit patient care (improve care coordination), physician workload (reduce physician documentation), hospital certification (improve monitored quality metric reporting), and national cancer surveillance efforts (increase reported staging). Here we propose to build machine learning classifier models for prostate cancer staging. We propose to use tree-based models, and specifically random forests. The successful completion of our aims will allow the development of a tool that will enable the automated staging evaluation for patients diagnosed with prostate cancer by leveraging information documented in their Electronic Health Record (EHR). We envision that this tool, once validated in prostate cancer patients, could be used across multiple cancer types.

Collaborators

Keith and Jan Hurlbut Professor in the Stanford University School of Medicine

Demographic biases in machine learning algorithms using electronic health record data: a systematic review

AI in Medicine, Inclusion & Equity (AIMIE) Seed Grant

 The rise of electronic health records (EHRs) has enabled the development of machine learning algorithms to assist with clinical decision-making for diagnosis, treatment and prognosis. As these algorithms infiltrate into clinical practice, it is important to understand hidden biases in how they were developed and whether machine learning algorithms in healthcare will perpetuate discrimination if they are trained on historical data. To date, little evidence exists on the equitable benefit of machine learning algorithms in healthcare, which begins with transparency of the demographic distribution of the population(s) studied. Such information is critical to understand how such advances in medical artificial intelligence may benefit everyone equally. We propose a systematic review of machine learning algorithms using EHR data, evaluating 1) whether studies disclose the breakdown of their training/test datasets in terms of ethnicity, gender, age groups, and socioeconomic/insurance status; 2) how representative the training data is of the broader population; 3) whether these demographic parameters are included in the model; and 4) whether any comparative assessment of model performance on vulnerable populations has been performed. This work will fill a critical literature gap regarding the potential benefit of emerging medical informatic technologies across all populations.

Collaborators

C. F. Rehnborg Professor in Disease Prevention in the School of Medicine, Professor of Medicine, of Health Research and Policy (Epidemiology) and by courtesy, of Statistics and of Biomedical Data Science

Individual-level predictive models for management of pain

Stanford-DHCRC/HMS grant (09/01/2019-10/31/2022)

Opioids are a first-line treatment of pain following surgery, but this may be a gateway to opioid misuse and addiction. Over the past decade, opioid misuse and abuse has become a major epidemic crisis in the USA. Most surgical patients receive opioids regardless of co-morbidities, prior opioid-related problems, or possible drug-drug interactions. In addition, the success of treatment using opioids is likely to be a complex function of a variety of individual and societal level factors, which are currently not well understood. This project will use linked claims data – sourced from a de-identified Medicaid (USA) dataset made available by HMS in conjunction with the Digital Health CRC – to gain a deep understanding of patterns of opioid use and prescription, and develop new insights into what constitutes successful treatment. It will accomplish this utilising novel machine learning methods to extract interpretable patterns and associations in opioid prescribing and use.

Collaborators