Learning From Past Respiratory Infections to Predict COVID-19 Outcomes: Retrospective Study.

BACKGROUND: For the clinical care of patients with well-established diseases, randomized trials, literature, and research are supplemented with clinical judgment to understand disease prognosis and inform treatment choices. In the void created by a lack of clinical experience with COVID-19, artificial intelligence (AI) may be an important tool to bolster clinical judgment and decision making. However, a lack of clinical data restricts the design and development of such AI tools, particularly in preparation for an impending crisis or pandemic. OBJECTIVE: This study aimed to develop and test the feasibility of a "patients-like-me" framework to predict the deterioration of patients with COVID-19 using a retrospective cohort of patients with similar respiratory diseases. METHODS: Our framework used COVID-19-like cohorts to design and train AI models that were then validated on the COVID-19 population. The COVID-19-like cohorts included patients diagnosed with bacterial pneumonia, viral pneumonia, unspecified pneumonia, influenza, and acute respiratory distress syndrome (ARDS) at an academic medical center from 2008 to 2019. In total, 15 training cohorts were created using different combinations of the COVID-19-like cohorts with the ARDS cohort for exploratory purposes. In this study, two machine learning models were developed: one to predict invasive mechanical ventilation (IMV) within 48 hours for each hospitalized day, and one to predict all-cause mortality at the time of admission. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, positive predictive value, and negative predictive value. We established model interpretability by calculating SHapley Additive exPlanations (SHAP) scores to identify important features. RESULTS: Compared to the COVID-19-like cohorts (n=16,509), the patients hospitalized with COVID-19 (n=159) were significantly younger, with a higher proportion of patients of Hispanic ethnicity, a lower proportion of patients with smoking history, and fewer patients with comorbidities (P<.001). Patients with COVID-19 had a lower IMV rate (15.1 versus 23.2, P=.02) and shorter time to IMV (2.9 versus 4.1 days, P<.001) compared to the COVID-19-like patients. In the COVID-19-like training data, the top models achieved excellent performance (AUROC>0.90). Validating in the COVID-19 cohort, the top-performing model for predicting IMV was the XGBoost model (AUROC=0.826) trained on the viral pneumonia cohort. Similarly, the XGBoost model trained on all 4 COVID-19-like cohorts without ARDS achieved the best performance (AUROC=0.928) in predicting mortality. Important predictors included demographic information (age), vital signs (oxygen saturation), and laboratory values (white blood cell count, cardiac troponin, albumin, etc). Our models had class imbalance, which resulted in high negative predictive values and low positive predictive values. CONCLUSIONS: We provided a feasible framework for modeling patient deterioration using existing data and AI technology to address data limitations during the onset of a novel, rapidly changing pandemic.

Integrating Adjuvant Analgesics into Perioperative Pain Practice: Results from an Academic Medical Center.

BACKGROUND: Opioid-sparing postoperative pain management therapies are important considering the opioid epidemic. Total knee arthroplasty (TKA) is a common and painful procedure accounting for a large number of opioid prescriptions. Adjuvant analgesics, nonopioid drugs with primary indications other than pain, have shown beneficial pain management and opioid-sparing effects following TKA in clinical trials. We evaluated the adjuvant analgesic gabapentin for its usage patterns and its effects on opioid use, pain, and readmissions. METHODS: This retrospective, observational study included 4,046 patients who received primary TKA between 2009 and 2017 using electronic health records from an academic tertiary care medical institute. Descriptive statistics and multivariate modeling were used to estimate associations between inpatient gabapentin use and adverse pain outcomes as well as inpatient oral morphine equivalents per day (OME). RESULTS: Overall, there was an 8.72% annual increase in gabapentin use (P < 0.001). Modeled estimates suggest that gabapentin is associated with a significant decrease in opioid consumption (estimate = 0.63, 95% confidence interval = 0.49-0.82, P < 0.001) when controlling for patient characteristics. Patients receiving gabapentin had similar discharge pain scores, follow-up pain scores, and 30-day unplanned readmission rates compared with patients receiving no adjuvant analgesics (P > 0.05). CONCLUSIONS: When assessed in a real-world setting over a large cohort of TKA patients, gabapentin is an effective pain management therapy that is associated with reduced opioid consumption-a national priority in this time of opioid crisis-while maintaining the same quality of pain management.

Distribution of global health measures from routinely collected PROMIS surveys in patients with breast cancer or prostate cancer.

BACKGROUND: The collection of patient-reported outcomes (PROs) is an emerging priority internationally, guiding clinical care, quality improvement projects and research studies. After the deployment of Patient-Reported Outcomes Measurement Information System (PROMIS) surveys in routine outpatient workflows at an academic cancer center, electronic health record data were used to evaluate survey completion rates and self-reported global health measures across 2 tumor types: breast and prostate cancer. METHODS: This study retrospectively analyzed 11,657 PROMIS surveys from patients with breast cancer and 4411 surveys from patients with prostate cancer, and it calculated survey completion rates and global physical health (GPH) and global mental health (GMH) scores between 2013 and 2018. RESULTS: A total of 36.6% of eligible patients with breast cancer and 23.7% of patients with prostate cancer completed at least 1 survey, with completion rates lower among black patients for both tumor types (P < .05). The mean T scores (calibrated to a general population mean of 50) for GPH were 48.4 ± 9 for breast cancer and 50.6 ± 9 for prostate cancer, and the GMH scores were 52.7 ± 8 and 52.1 ± 9, respectively. GPH and GMH were frequently lower among ethnic minorities, patients without private health insurance, and those with advanced disease. CONCLUSIONS: This analysis provides important baseline data on patient-reported global health in breast and prostate cancer. Demonstrating that PROs can be integrated into clinical workflows, this study shows that supportive efforts may be needed to improve PRO collection and global health endpoints in vulnerable populations.

Genomic landscape of ductal carcinoma in situ and association with progression.

PURPOSE: The detection rate of breast ductal carcinoma in situ (DCIS) has increased significantly, raising the concern that DCIS is overdiagnosed and overtreated. Therefore, there is an unmet clinical need to better predict the risk of progression among DCIS patients. Our hypothesis is that by combining molecular signatures with clinicopathologic features, we can elucidate the biology of breast cancer progression, and risk-stratify patients with DCIS. METHODS: Targeted exon sequencing with a custom panel of 223 genes/regions was performed for 125 DCIS cases. Among them, 60 were from cases having concurrent or subsequent invasive breast cancer (IBC) (DCIS + IBC group), and 65 from cases with no IBC development over a median follow-up of 13 years (DCIS-only group). Copy number alterations in chromosome 1q32, 8q24, and 11q13 were analyzed using fluorescence in situ hybridization (FISH). Multivariable logistic regression models were fit to the outcome of DCIS progression to IBC as functions of demographic and clinical features. RESULTS: We observed recurrent variants of known IBC-related mutations, and the most commonly mutated genes in DCIS were PIK3CA (34.4%) and TP53 (18.4%). There was an inverse association between PIK3CA kinase domain mutations and progression (Odds Ratio [OR] 10.2, p < 0.05). Copy number variations in 1q32 and 8q24 were associated with progression (OR 9.3 and 46, respectively; both p < 0.05). CONCLUSIONS: PIK3CA kinase domain mutations and the absence of copy number gains in DCIS are protective against progression to IBC. These results may guide efforts to distinguish low-risk from high-risk DCIS.

PSA Testing Use and Prostate Cancer Diagnostic Stage After the 2012 U.S. Preventive Services Task Force Guideline Changes.

BACKGROUND: Most patients with prostate cancer are diagnosed with low-grade, localized disease and may not require definitive treatment. In 2012, the U.S. Preventive Services Task Force (USPSTF) recommended against prostate cancer screening to address overdetection and overtreatment. This study sought to determine the effect of guideline changes on prostate-specific antigen (PSA) screening and initial diagnostic stage for prostate cancer. PATIENTS AND METHODS: A difference-in-differences analysis was conducted to compare changes in PSA screening (exposure) relative to cholesterol testing (control) after the 2012 USPSTF guideline changes, and chi-square test was used to determine whether there was a subsequent decrease in early-stage, low-risk prostate cancer diagnoses. Data were derived from a tertiary academic medical center's electronic health records, a national commercial insurance database (OptumLabs), and the SEER database for men aged ≥35 years before (2008-2011) and after (2013-2016) the guideline changes. RESULTS: In both the academic center and insurance databases, PSA testing significantly decreased for all men compared with the control. The greatest decrease was among men aged 55 to 74 years at the academic center and among those aged ≥75 years in the commercial database. The proportion of early-stage prostate cancer diagnoses (

Utilization and effectiveness of multimodal discharge analgesia for postoperative pain management.

BACKGROUND: Although evidence-based guidelines recommend a multimodal approach to pain management, limited information exists on adherence to these guidelines and its association with outcomes in a generalized population. We sought to assess the association between discharge multimodal analgesia and postoperative pain outcomes in two diverse health care settings. METHODS: We evaluated patients undergoing four common surgeries associated with high pain in electronic health records from an academic hospital (AH) and Veterans Health Administration (VHA). Multimodal analgesia at discharge was characterized as opioids in combination with acetaminophen (O + A) and nonsteroidal antiinflammatory (O + A + N) drugs. Hierarchical models estimated associations of analgesia with 45-d follow-up pain scores and 30-d readmissions. RESULTS: We identified 7893 patients at AH and 34,581 at VHA. In both settings, most patients were discharged with O + A (60.6% and 54.8%, respectively), yet a significant proportion received opioids alone (AH: 24.3% and VHA: 18.8%). Combining acetaminophen with opioids was associated with decreased follow-up pain in VHA (Odds ratio [OR]: 0.86, 95% confidence interval [CI]: 0.79, 0.93) and readmissions (AH OR: 0.74, CI: 0.60, 0.90; VHA OR: 0.89, CI: 0.82, 0.96). Further addition of nonsteroidal antiinflammatory drugs was associated with further decreased follow-up pain (AH OR: 0.71, CI: 0.53, 0.96; VHA OR: 0.77, CI: 0.69, 0.86) and readmissions (AH OR: 0.46, CI: 0.31, 0.69; VHA OR: 0.84, CI: 0.76, 0.93). In both systems, patients receiving multimodal analgesia received 10%-40% less opioids per day compared to opioids only. CONCLUSIONS: A majority of surgical patients receive a multimodal pain approach at discharge yet many receive only opioids. Multimodal regimen at discharge was associated with better follow-up pain and all-cause readmissions compared to the opioid-only regimen.

Chromosomal copy number alterations for associations of ductal carcinoma in situ with invasive breast cancer.

INTRODUCTION: Screening mammography has contributed to a significant increase in the diagnosis of ductal carcinoma in situ (DCIS), raising concerns about overdiagnosis and overtreatment. Building on prior observations from lineage evolution analysis, we examined whether measuring genomic features of DCIS would predict association with invasive breast carcinoma (IBC). The long-term goal is to enhance standard clinicopathologic measures of low- versus high-risk DCIS and to enable risk-appropriate treatment. METHODS: We studied three common chromosomal copy number alterations (CNA) in IBC and designed fluorescence in situ hybridization-based assay to measure copy number at these loci in DCIS samples. Clinicopathologic data were extracted from the electronic medical records of Stanford Cancer Institute and linked to demographic data from the population-based California Cancer Registry; results were integrated with data from tissue microarrays of specimens containing DCIS that did not develop IBC versus DCIS with concurrent IBC. Multivariable logistic regression analysis was performed to describe associations of CNAs with these two groups of DCIS. RESULTS: We examined 271 patients with DCIS (120 that did not develop IBC and 151 with concurrent IBC) for the presence of 1q, 8q24 and 11q13 copy number gains. Compared to DCIS-only patients, patients with concurrent IBC had higher frequencies of CNAs in their DCIS samples. On multivariable analysis with conventional clinicopathologic features, the copy number gains were significantly associated with concurrent IBC. The state of two of the three copy number gains in DCIS was associated with a risk of IBC that was 9.07 times that of no copy number gains, and the presence of gains at all three genomic loci in DCIS was associated with a more than 17-fold risk (P = 0.0013). CONCLUSIONS: CNAs have the potential to improve the identification of high-risk DCIS, defined by presence of concurrent IBC. Expanding and validating this approach in both additional cross-sectional and longitudinal cohorts may enable improved risk stratification and risk-appropriate treatment in DCIS.

Breast cancer treatment across health care systems: linking electronic medical records and state registry data to enable outcomes research.

BACKGROUND: Understanding of cancer outcomes is limited by data fragmentation. In the current study, the authors analyzed the information yielded by integrating breast cancer data from 3 sources: electronic medical records (EMRs) from 2 health care systems and the state registry. METHODS: Diagnostic test and treatment data were extracted from the EMRs of all patients with breast cancer treated between 2000 and 2010 in 2 independent California institutions: a community-based practice (Palo Alto Medical Foundation; "Community") and an academic medical center (Stanford University; "University"). The authors incorporated records from the population-based California Cancer Registry and then linked EMR-California Cancer Registry data sets of Community and University patients. RESULTS: The authors initially identified 8210 University patients and 5770 Community patients; linked data sets revealed a 16% patient overlap, yielding 12,109 unique patients. The percentage of all Community patients, but not University patients, treated at both institutions increased with worsening cancer prognostic factors. Before linking the data sets, Community patients appeared to receive less intervention than University patients (mastectomy: 37.6% vs 43.2%; chemotherapy: 35% vs 41.7%; magnetic resonance imaging: 10% vs 29.3%; and genetic testing: 2.5% vs 9.2%). Linked Community and University data sets revealed that patients treated at both institutions received substantially more interventions (mastectomy: 55.8%; chemotherapy: 47.2%; magnetic resonance imaging: 38.9%; and genetic testing: 10.9% [P < .001 for each 3-way institutional comparison]). CONCLUSIONS: Data linkage identified 16% of patients who were treated in 2 health care systems and who, despite comparable prognostic factors, received far more intensive treatment than others. By integrating complementary data from EMRs and population-based registries, a more comprehensive understanding of breast cancer care and factors that drive treatment use was obtained.

Characterizing patient flow after an academic hospital merger and acquisition.

OBJECTIVES: Hospital mergers and acquisitions are increasingly used as a strategy to facilitate value-based care. However, no studies have assessed health care utilization (HCU) and patient flow across merged institutions. We aim to evaluate patient population distribution, HCU, and patient flow across a recent hospital merger of an academic medical center (AMC), a primary and specialty care alliance (PSC), and a community-based medical center (CMC). STUDY DESIGN: This was a retrospective observational study. METHODS: The study used 2018 adult demographic and encounter data from electronic health records. Patients' parent health care institution was determined by the most frequently visited site of face-to-face visits. Differences in patient demographics and HCU (ie, emergency department [ED] visits, hospitalizations, primary care visits) were compared. Independent factors associated with utilization were identified using adjusted logistic regression models. RESULTS: A total of 406,303 adult patients were identified in the cohort. The PSC setting, compared with the AMC and the CMC, had significantly more female (62.7% vs 54.4% and 58.5%, respectively), older (mean [SD] age, 52.0 [18.1] vs 51.1 [17.8] and 49.2 [17.8] years), and privately insured (63.6% vs 51.3% and 56.0%) patients. A higher proportion of patients at the CMC (27.5%) visited the ED compared with patients at the AMC (10.8%). Approximately 1645 primary care patients (7%) at the CMC setting went to the AMC for specialized care such as oncology, surgery, and neurology. CONCLUSIONS: Hospital mergers are increasing across the United States, allowing AMCs to expand their reach. These findings suggest that patients mainly sought care at their parent health care institution, yet appropriately received specialized care at the AMC. These results provide insights for future mergers and guide resource allocation and opportunities for improving care delivery.

Benchmark Method for Cost Computations Across Health Care Systems: Cost of Care per Patient per Day in Breast Cancer Care.

PURPOSE: To estimate the value of cancer care and to compare value among episodes of care, a transparent, reproducible, and standardized cost computation methodology is needed. Charges, claims, and reimbursements are related to cost but are nontransparent and proprietary. We developed a method to measure the cost of the following phases of care: (1) initial treatment with curative intent, (2) surveillance and survivorship care, and (3) relapse and end-of-life care. METHODS: We combined clinical data from our electronic health record, the state cancer registry, and the Social Security Death Index. We analyzed the care of patients with breast cancer and mapped Common Procedural Terminology (CPT) codes to the corresponding cost conversion factor and date in the CMS Medicare fee schedule. To account for varying duration of episodes of care, we computed a cost of care per day (CCPD) for each patient. RESULTS: Median CCPD for initial treatment was $29.45 in US dollars (USD), the CCPD for surveillance and survivorship care was $2.45 USD, and the CCPD for relapse care was $13.80 USD. Among the three breast cancer types (hormone receptor-positive or human epidermal growth factor receptor 2 [HER2]-negative, HER2-positive, and triple-negative), there was no difference in CCPD. Relapsed patients in the most expensive surveillance CCPD group had significantly shorter survival. CONCLUSION: We developed a method to identify high-value oncology care-cost of care per patient per day (CCPD)-in episodes of initial, survivorship, and relapse care. The methodology can help identify positive deviants (who have developed best practices) delivering high-value care. Merging our data with claims data from third-party payers can increase the accuracy and validity of the CCPD.

A framework for making predictive models useful in practice.

OBJECTIVE: To analyze the impact of factors in healthcare delivery on the net benefit of triggering an Advanced Care Planning (ACP) workflow based on predictions of 12-month mortality. MATERIALS AND METHODS: We built a predictive model of 12-month mortality using electronic health record data and evaluated the impact of healthcare delivery factors on the net benefit of triggering an ACP workflow based on the models' predictions. Factors included nonclinical reasons that make ACP inappropriate: limited capacity for ACP, inability to follow up due to patient discharge, and availability of an outpatient workflow to follow up on missed cases. We also quantified the relative benefits of increasing capacity for inpatient ACP versus outpatient ACP. RESULTS: Work capacity constraints and discharge timing can significantly reduce the net benefit of triggering the ACP workflow based on a model's predictions. However, the reduction can be mitigated by creating an outpatient ACP workflow. Given limited resources to either add capacity for inpatient ACP versus developing outpatient ACP capability, the latter is likely to provide more benefit to patient care. DISCUSSION: The benefit of using a predictive model for identifying patients for interventions is highly dependent on the capacity to execute the workflow triggered by the model. We provide a framework for quantifying the impact of healthcare delivery factors and work capacity constraints on achieved benefit. CONCLUSION: An analysis of the sensitivity of the net benefit realized by a predictive model triggered clinical workflow to various healthcare delivery factors is necessary for making predictive models useful in practice.

Weakly supervised natural language processing for assessing patient-centered outcome following prostate cancer treatment.

Background: The population-based assessment of patient-centered outcomes (PCOs) has been limited by the efficient and accurate collection of these data. Natural language processing (NLP) pipelines can determine whether a clinical note within an electronic medical record contains evidence on these data. We present and demonstrate the accuracy of an NLP pipeline that targets to assess the presence, absence, or risk discussion of two important PCOs following prostate cancer treatment: urinary incontinence (UI) and bowel dysfunction (BD). Methods: We propose a weakly supervised NLP approach which annotates electronic medical record clinical notes without requiring manual chart review. A weighted function of neural word embedding was used to create a sentence-level vector representation of relevant expressions extracted from the clinical notes. Sentence vectors were used as input for a multinomial logistic model, with output being either presence, absence or risk discussion of UI/BD. The classifier was trained based on automated sentence annotation depending only on domain-specific dictionaries (weak supervision). Results: The model achieved an average F1 score of 0.86 for the sentence-level, three-tier classification task (presence/absence/risk) in both UI and BD. The model also outperformed a pre-existing rule-based model for note-level annotation of UI with significant margin. Conclusions: We demonstrate a machine learning method to categorize clinical notes based on important PCOs that trains a classifier on sentence vector representations labeled with a domain-specific dictionary, which eliminates the need for manual engineering of linguistic rules or manual chart review for extracting the PCOs. The weakly supervised NLP pipeline showed promising sensitivity and specificity for identifying important PCOs in unstructured clinical text notes compared to rule-based algorithms.

Architecture and Implementation of a Clinical Research Data Warehouse for Prostate Cancer.

BACKGROUND: Electronic health record (EHR) based research in oncology can be limited by missing data and a lack of structured data elements. Clinical research data warehouses for specific cancer types can enable the creation of more robust research cohorts. METHODS: We linked data from the Stanford University EHR with the Stanford Cancer Institute Research Database (SCIRDB) and the California Cancer Registry (CCR) to create a research data warehouse for prostate cancer. The database was supplemented with information from clinical trials, natural language processing of clinical notes and surveys on patient-reported outcomes. RESULTS: 11,898 unique prostate cancer patients were identified in the Stanford EHR, of which 3,936 were matched to the Stanford cancer registry and 6153 in the CCR. 7158 patients with EHR data and at least one of SCIRDB and CCR data were initially included in the warehouse. CONCLUSIONS: A disease-specific clinical research data warehouse combining multiple data sources can facilitate secondary data use and enhance observational research in oncology.

Higher Absolute Lymphocyte Counts Predict Lower Mortality from Early-Stage Triple-Negative Breast Cancer.

Purpose: Tumor-infiltrating lymphocytes (TIL) in pretreatment biopsies are associated with improved survival in triple-negative breast cancer (TNBC). We investigated whether higher peripheral lymphocyte counts are associated with lower breast cancer-specific mortality (BCM) and overall mortality (OM) in TNBC.Experimental Design: Data on treatments and diagnostic tests from electronic medical records of two health care systems were linked with demographic, clinical, pathologic, and mortality data from the California Cancer Registry. Multivariable regression models adjusted for age, race/ethnicity, socioeconomic status, cancer stage, grade, neoadjuvant/adjuvant chemotherapy use, radiotherapy use, and germline BRCA1/2 mutations were used to evaluate associations between absolute lymphocyte count (ALC), BCM, and OM. For a subgroup with TIL data available, we explored the relationship between TILs and peripheral lymphocyte counts.Results: A total of 1,463 stage I-III TNBC patients were diagnosed from 2000 to 2014; 1,113 (76%) received neoadjuvant/adjuvant chemotherapy within 1 year of diagnosis. Of 759 patients with available ALC data, 481 (63.4%) were ever lymphopenic (minimum ALC <1.0 K/µL). On multivariable analysis, higher minimum ALC, but not absolute neutrophil count, predicted lower OM [HR = 0.23; 95% confidence interval (CI), 0.16-0.35] and BCM (HR = 0.19; CI, 0.11-0.34). Five-year probability of BCM was 15% for patients who were ever lymphopenic versus 4% for those who were not. An exploratory analysis (n = 70) showed a significant association between TILs and higher peripheral lymphocyte counts during neoadjuvant chemotherapy.Conclusions: Higher peripheral lymphocyte counts predicted lower mortality from early-stage, potentially curable TNBC, suggesting that immune function may enhance the effectiveness of early TNBC treatment. Clin Cancer Res; 24(12); 2851-8. ©2018 AACR.

Mining Electronic Health Records to Extract Patient-Centered Outcomes Following Prostate Cancer Treatment.

The clinical, granular data in electronic health record (EHR) systems provide opportunities to improve patient care using informatics retrieval methods. However, it is well known that many methodological obstacles exist in accessing data within EHRs. In particular, clinical notes routinely stored in EHR are composed from narrative, highly unstructured and heterogeneous biomedical text. This inherent complexity hinders the ability to perform automated large-scale medical knowledge extraction tasks without the use of computational linguistics methods. The aim of this work was to develop and validate a Natural Language Processing (NLP) pipeline to detect important patient-centered outcomes (PCOs) as interpreted and documented by clinicians in their dictated notes for male patients receiving treatment for localized prostate cancer at an academic medical center.

Synergistic drug combinations from electronic health records and gene expression.

OBJECTIVE: Using electronic health records (EHRs) and biomolecular data, we sought to discover drug pairs with synergistic repurposing potential. EHRs provide real-world treatment and outcome patterns, while complementary biomolecular data, including disease-specific gene expression and drug-protein interactions, provide mechanistic understanding. METHOD: We applied Group Lasso INTERaction NETwork (glinternet), an overlap group lasso penalty on a logistic regression model, with pairwise interactions to identify variables and interacting drug pairs associated with reduced 5-year mortality using EHRs of 9945 breast cancer patients. We identified differentially expressed genes from 14 case-control human breast cancer gene expression datasets and integrated them with drug-protein networks. Drugs in the network were scored according to their association with breast cancer individually or in pairs. Lastly, we determined whether synergistic drug pairs found in the EHRs were enriched among synergistic drug pairs from gene-expression data using a method similar to gene set enrichment analysis. RESULTS: From EHRs, we discovered 3 drug-class pairs associated with lower mortality: anti-inflammatories and hormone antagonists, anti-inflammatories and lipid modifiers, and lipid modifiers and obstructive airway drugs. The first 2 pairs were also enriched among pairs discovered using gene expression data and are supported by molecular interactions in drug-protein networks and preclinical and epidemiologic evidence. CONCLUSIONS: This is a proof-of-concept study demonstrating that a combination of complementary data sources, such as EHRs and gene expression, can corroborate discoveries and provide mechanistic insight into drug synergism for repurposing.

Use of Gene Expression Profiling and Chemotherapy in Early-Stage Breast Cancer: A Study of Linked Electronic Medical Records, Cancer Registry Data, and Genomic Data Across Two Health Care Systems.

PURPOSE: The 21-gene recurrence score (RS) identifies patients with breast cancer who derive little benefit from chemotherapy; it may reduce unwarranted variability in the use of chemotherapy. We tested whether the use of RS seems to guide chemotherapy receipt across different cancer care settings. METHODS: We developed a retrospective cohort of patients with breast cancer by using electronic medical record data from Stanford University (hereafter University) and Palo Alto Medical Foundation (hereafter Community) linked with demographic and staging data from the California Cancer Registry and RS results from the testing laboratory (Genomic Health Inc., Redwood City, CA). Multivariable analysis was performed to identify predictors of RS and chemotherapy use. RESULTS: In all, 10,125 patients with breast cancer were diagnosed in the University or Community systems from 2005 to 2011; 2,418 (23.9%) met RS guidelines criteria, of whom 15.6% received RS. RS was less often used for patients with involved lymph nodes, higher tumor grade, and age < 40 or ≥ 65 years. Among RS recipients, chemotherapy receipt was associated with a higher score (intermediate v low: odds ratio, 3.66; 95% CI, 1.94 to 6.91). A total of 293 patients (10.6%) received care in both health care systems (hereafter dual use); although receipt of RS was associated with dual use (v University: odds ratio, 1.73; 95% CI, 1.18 to 2.55), there was no difference in use of chemotherapy after RS by health care setting. CONCLUSION: Although there was greater use of RS for patients who sought care in more than one health care setting, use of chemotherapy followed RS guidance in University and Community health care systems. These results suggest that precision medicine may help optimize cancer treatment across health care settings.

Oncoshare: lessons learned from building an integrated multi-institutional database for comparative effectiveness research.

Comparative effectiveness research (CER) using observational data requires informatics methods for the extraction, standardization, sharing, and integration of data derived from a variety of electronic sources. In the Oncoshare project, we have developed such methods as part of a collaborative multi-institutional CER study of patterns, predictors, and outcome of breast cancer care. In this paper, we present an evaluation of the approaches we undertook and the lessons we learned in building and validating the Oncoshare data resource. Specifically, we determined that 1) the state or regional cancer registry makes the most efficient starting point for determining inclusion of subjects; 2) the data dictionary should be based on existing registry standards, such as Surveillance, Epidemiology and End Results (SEER), when applicable; 3) the Social Security Administration Death Master File (SSA DMF), rather than clinical resources, provides standardized ascertainment of mortality outcomes; and 4) CER database development efforts, despite the immediate availability of electronic data, may take as long as two years to produce validated, reliable data for research. Through our efforts using these methods, Oncoshare integrates complex, longitudinal data from multiple electronic medical records and registries and provides a rich, validated resource for research on oncology care.