Electronic Medical Records Dataset Research Articles

Automated Survival Prediction in Metastatic Cancer Patients Using High-Dimensional Electronic Medical Record Data.

Oncologists use patients' life expectancy to guide decisions and may benefit from a tool that accurately predicts prognosis. Existing prognostic models generally use only a few predictor variables. We used an electronic medical record dataset to train a prognostic model for patients with metastatic cancer. The model was trained and tested using 12 588 patients treated for metastatic cancer in the Stanford Health Care system from 2008 to 2017. Data sources included provider note text, labs, vital signs, procedures, medication orders, and diagnosis codes. Patients were divided randomly into a training set used to fit the model coefficients and a test set used to evaluate model performance (80%/20% split). A regularized Cox model with 4126 predictor variables was used. A landmarking approach was used due to the multiple observations per patient, with t0 set to the time of metastatic cancer diagnosis. Performance was also evaluated using 399 palliative radiation courses in test set patients. The C-index for overall survival was 0.786 in the test set (averaged across landmark times). For palliative radiation courses, the C-index was 0.745 (95% confidence interval [CI] = 0.715 to 0.775) compared with 0.635 (95% CI = 0.601 to 0.669) for a published model using performance status, primary tumor site, and treated site (two-sided P < .001). Our model's predictions were well-calibrated. The model showed high predictive performance, which will need to be validated using external data. Because it is fully automated, the model can be used to examine providers' practice patterns and could be deployed in a decision support tool to help improve quality of care.

Achievement of Target A1C

Objective.Many patients with type 2 diabetes do not reach glycemic goals despite basal insulin treatment. This study assessed the achievement of a target A1C <7.0% (<53 mmol/mol) after initiation of basal insulin in two settings.Methods.This was a retrospective analysis of pooled randomized controlled trial (RCT) data, from 11 24-week studies of patients initiating basal insulin performed between 2000 and 2005 and of outpatient electronic medical record (EMR) data from the General Electric Centricity database for insulin-naive patients initiating basal insulin between 2005 and 2012. Baseline characteristics stratified by target A1C and fasting plasma glucose (FPG) attainment were compared descriptively.Results.In the RCT dataset, 49.0% of patients failed to achieve the target A1C at 6 months versus 72.4% and 72.9% at 6 and 12 months in the EMR dataset, respectively. Despite this, in the RCT dataset, 79.4% of patients achieved the target A1C and/or an FPG <130 mg/dL. In the EMR dataset, only 47.6% and 47.3% of patients achieved an A1C <7.0% and/or FPG <130 mg/dL at 6 and 12 months, respectively. Overall, patients with an A1C >7.0% had a longer diabetes duration and were more likely to be female, nonwhite, and self-funding or covered by Medicaid. Among patients with an A1C >7.0%, more RCT patients (58.0%) had an FPG <130 mg/dL than EMR patients at 6 months (27.8%) and 12 months (27.7%).Conclusion.Unmet needs remain after basal insulin initiation, particularly in real-world settings, where many patients require further insulin titration. In both populations, patients failing to achieve the target A1C despite attaining an FPG <130 mg/dL require interventions to improve postprandial control.

1512O - Automated survival prediction in metastatic cancer patients using high-dimensional electronic medical record data

Background: Oncologists use patients’ life expectancy to guide medical decisions and may benefit from a tool that provides accurate, unbiased assessments of prognosis. Existing prognostic models generally use only a few predictor variables. We used a large electronic medical record dataset to train a prognostic model for patients with metastatic cancer. Methods: The model was trained and tested using data from 12,588 patients treated for metastatic cancer in the Stanford Health Care system from 2008-2017. Data sources included provider note text, labs, vital signs, procedures, medication orders, and diagnosis codes. Patients were divided randomly into training and test sets (80%/20% split). A regularized Cox proportional hazards model with 4,126 predictor variables was fit to the training set and evaluated on the test set. A landmarking approach was used due to the multiple observations per patient, with t0 set to the time of metastatic cancer diagnosis. Performance was also evaluated using 399 palliative radiation courses in test set patients. An existing published model that uses performance status, primary tumor site, and treated site was used as a baseline [Chow, JCO 2008:20;26(36)]. Results: From the first visit after metastatic cancer diagnosis, median follow-up was 14.5 months and median overall survival was 20.9 months. Patients were seen for 384,402 daily visits. The prognostic model’s C-index for overall survival was 0.79 in the test set (averaged across landmark times). For palliative radiation courses, the C-index was 0.75 (95% CI 0.72-0.78), compared to 0.64 (95% CI 0.60-0.67) for an existing published model (p < 0.001).Table: 1512OPredicted vs actual survival for 2,518 test set patients at landmark time t0 (first visit after diagnosis of metastatic cancer)Predicted median survival in monthsActual median survival in months (95% CI)0-3 (n = 106)1.3 (0.9-2.0)3.1-6 (n = 172)3.7 (2.5-5.2)6.1-12 (n = 382)6.7 (5.6-7.9)>12 (n = 1858)35.7 (31.8-39.1) Open table in a new tab Conclusions: The model showed high predictive performance, which was significantly better than that of an existing model. Because it is fully automated, the model can be used to examine providers’ practice patterns and could deployed in a decision support tool to help improve quality of care. Legal entity responsible for the study: Michael Gensheimer. Funding: National Institutes of Health. Disclosure: M.F. Gensheimer: Research funding: Varian Medical Systems, Philips Healthcare. E. Cho: Employee of Genentech. D. Rubin: Research funding: Philips Healthcare. D.T. Chang: Research funding, honoraria: Varian Medical Systems; Stock ownership: ViewRay. All other authors have declared no conflicts of interest.

Assessing Statewide All-Cause Future One-Year Mortality: Prospective Study With Implications for Quality of Life, Resource Utilization, and Medical Futility.

BackgroundFor many elderly patients, a disproportionate amount of health care resources and expenditures is spent during the last year of life, despite the discomfort and reduced quality of life associated with many aggressive medical approaches. However, few prognostic tools have focused on predicting all-cause 1-year mortality among elderly patients at a statewide level, an issue that has implications for improving quality of life while distributing scarce resources fairly.ObjectiveUsing data from a statewide elderly population (aged ≥65 years), we sought to prospectively validate an algorithm to identify patients at risk for dying in the next year for the purpose of minimizing decision uncertainty, improving quality of life, and reducing futile treatment.MethodsAnalysis was performed using electronic medical records from the Health Information Exchange in the state of Maine, which covered records of nearly 95% of the statewide population. The model was developed from 125,896 patients aged at least 65 years who were discharged from any care facility in the Health Information Exchange network from September 5, 2013, to September 4, 2015. Validation was conducted using 153,199 patients with same inclusion and exclusion criteria from September 5, 2014, to September 4, 2016. Patients were stratified into risk groups. The association between all-cause 1-year mortality and risk factors was screened by chi-squared test and manually reviewed by 2 clinicians. We calculated risk scores for individual patients using a gradient tree-based boost algorithm, which measured the probability of mortality within the next year based on the preceding 1-year clinical profile.ResultsThe development sample included 125,896 patients (72,572 women, 57.64%; mean 74.2 [SD 7.7] years). The final validation cohort included 153,199 patients (88,177 women, 57.56%; mean 74.3 [SD 7.8] years). The c-statistic for discrimination was 0.96 (95% CI 0.93-0.98) in the development group and 0.91 (95% CI 0.90-0.94) in the validation cohort. The mortality was 0.99% in the low-risk group, 16.75% in the intermediate-risk group, and 72.12% in the high-risk group. A total of 99 independent risk factors (n=99) for mortality were identified (reported as odds ratios; 95% CI). Age was on the top of list (1.41; 1.06-1.48); congestive heart failure (20.90; 15.41-28.08) and different tumor sites were also recognized as driving risk factors, such as cancer of the ovaries (14.42; 2.24-53.04), colon (14.07; 10.08-19.08), and stomach (13.64; 3.26-86.57). Disparities were also found in patients’ social determinants like respiratory hazard index (1.24; 0.92-1.40) and unemployment rate (1.18; 0.98-1.24). Among high-risk patients who expired in our dataset, cerebrovascular accident, amputation, and type 1 diabetes were the top 3 diseases in terms of average cost in the last year of life.ConclusionsOur study prospectively validated an accurate 1-year risk prediction model and stratification for the elderly population (≥65 years) at risk of mortality with statewide electronic medical record datasets. It should be a valuable adjunct for helping patients to make better quality-of-life choices and alerting care givers to target high-risk elderly for appropriate care and discussions, thus cutting back on futile treatment.

EMR-based medical knowledge representation and inference via Markov random fields and distributed representation learning.

Electronic medical records (EMRs) contain medical knowledge that can be used for clinical decision support (CDS). Our objective is to develop a general system that can extract and represent knowledge contained in EMRs to support three CDS tasks-test recommendation, initial diagnosis, and treatment plan recommendation-given the condition of a patient. We extracted four kinds of medical entities from records and constructed an EMR-based medical knowledge network (EMKN), in which nodes are entities and edges reflect their co-occurrence in a record. Three bipartite subgraphs (bigraphs) were extracted from the EMKN, one to support each task. One part of the bigraph was the given condition (e.g., symptoms), and the other was the condition to be inferred (e.g., diseases). Each bigraph was regarded as a Markov random field (MRF) to support the inference. We proposed three graph-based energy functions and three likelihood-based energy functions. Two of these functions are based on knowledge representation learning and can provide distributed representations of medical entities. Two EMR datasets and three metrics were utilized to evaluate the performance. As a whole, the evaluation results indicate that the proposed system outperformed the baseline methods. The distributed representation of medical entities does reflect similarity relationships with respect to knowledge level. Combining EMKN and MRF is an effective approach for general medical knowledge representation and inference. Different tasks, however, require individually designed energy functions.

Association of baseline absolute neutrophil counts and survival in patients with metastatic colorectal cancer treated with second-line antiangiogenic therapies: exploratory analyses of the RAISE trial and validation in an electronic medical record data set

BackgroundIn the RAISE trial, ramucirumab+leucovorin/fluorouracil/irinotecan (FOLFIRI) improved the median overall survival (mOS) of patients with previously treated metastatic colorectal cancer versus patients treated with placebo+FOLFIRI but had a higher incidence of neutropaenia, leading to more chemotherapy dose modifications and discontinuations. Thus, we conducted an exploratory post-hoc analysis of RAISE and a retrospective, observational analysis of electronic medical record (EMR) data to determine and verify the association of neutropaenia, baseline absolute neutrophil count (ANC) and survival.MethodsThe RAISE analysis used the study safety population (n=1057). IMS Health Oncology Database (IMS EMR) was the source for the real-world data set (n=617).ResultsRAISE patients with treatment-emergent neutropaenia had improved mOS compared with those without (ramucirumab arm: 16.1 vs 10.7 months, HR=0.57, p<0.0001; placebo arm: 12.7 vs 10.7 months, HR=0.76, p=0.0065). RAISE patients with low ANC versus high baseline ANC also had longer mOS (ramucirumab arm: 15.2 vs 8.9 months, HR=0.49, p<0.0001; placebo arm: 13.2 vs 7.3 months, HR=0.50, p<0.0001). The results were similar for IMS EMR low versus high baseline ANC (bevacizumab+FOLFIRI patients: 14.9 vs 7.7 months, HR=0.59, p<0.0001; FOLFIRI alone: 14.6 vs 5.4 months, HR=0.37, p<0.0001). Patients in the RAISE trial with low baseline ANC were more likely to develop neutropaenia (OR: ramucirumab arm=2.62, p<0.0001; placebo arm=2.16, p=0.0003).ConclusionNeutropaenia during treatment, and subsequent dose modifications or discontinuations, do not compromise treatment efficacy. Baseline ANC is a strong prognostic factor for survival and is associated with treatment-emergent neutropaenia in the analysed population.Trial registration numberNCT01183780, Results.

Automatic infection detection based on electronic medical records

BackgroundMaking accurate patient care decision, as early as possible, is a constant challenge, especially for physicians in the emergency department. The increasing volumes of electronic medical records (EMRs) open new horizons for automatic diagnosis. In this paper, we propose to use machine learning approaches for automatic infection detection based on EMRs. Five categories of information are utilized for prediction, including personal information, admission note, vital signs, diagnose test results and medical image diagnose.ResultsExperimental results on a newly constructed EMRs dataset from emergency department show that machine learning models can achieve a decent performance for infection detection with area under the receiver operator characteristic curve (AUC) of 0.88. Out of all the five types of information, admission note in text form makes the most contribution with the AUC of 0.87.ConclusionsThis study provides a state-of-the-art EMRs processing system to automatically make medical decisions. It extracts five types of features associated with infection and achieves a decent performance on automatic infection detection based on machine learning models.

Flexible, cluster-based analysis of the electronic medical record of sepsis with composite mixture models

The widespread adoption of electronic medical records (EMRs) in healthcare has provided vast new amounts of data for statistical machine learning researchers in their efforts to model and predict patient health status, potentially enabling novel advances in treatment. In the case of sepsis, a debilitating, dysregulated host response to infection, extracting subtle, uncataloged clinical phenotypes from the EMR with statistical machine learning methods has the potential to impact patient diagnosis and treatment early in the course of their hospitalization. However, there are significant barriers that must be overcome to extract these insights from EMR data. First, EMR datasets consist of both static and dynamic observations of discrete and continuous-valued variables, many of which may be missing, precluding the application of standard multivariate analysis techniques. Second, clinical populations observed via EMRs and relevant to the study and management of conditions like sepsis are often heterogeneous; properly accounting for this heterogeneity is critical. Here, we describe an unsupervised, probabilistic framework called a composite mixture model that can simultaneously accommodate the wide variety of observations frequently observed in EMR datasets, characterize heterogeneous clinical populations, and handle missing observations. We demonstrate the efficacy of our approach on a large-scale sepsis cohort, developing novel techniques built on our model-based clusters to track patient mortality risk over time and identify physiological trends and distinct subgroups of the dataset associated with elevated risk of mortality during hospitalization.

Privacy-preserving data cube for electronic medical records: An experimental evaluation

Introduction: The aim of this study is to evaluate the effectiveness and efficiency of privacy-preserving data cubes of electronic medical records (EMRs). An EMR data cube is a complex of EMR statistics that are summarized or aggregated by all possible combinations of attributes. Data cubes are widely utilized for efficient big data analysis and also have great potential for EMR analysis. For safe data analysis without privacy breaches, we must consider the privacy preservation characteristics of the EMR data cube. In this paper, we introduce a design for a privacy-preserving EMR data cube and the anonymization methods needed to achieve data privacy. We further focus on changes in efficiency and effectiveness that are caused by the anonymization process for privacy preservation. Thus, we experimentally evaluate various types of privacy-preserving EMR data cubes using several practical metrics and discuss the applicability of each anonymization method with consideration for the EMR analysis environment. Methods: We construct privacy-preserving EMR data cubes from anonymized EMR datasets. A real EMR dataset and demographic dataset are used for the evaluation. There are a large number of anonymization methods to preserve EMR privacy, and the methods are classified into three categories (i.e., global generalization, local generalization, and bucketization) by anonymization rules. According to this classification, three types of privacy-preserving EMR data cubes were constructed for the evaluation. We perform a comparative analysis by measuring the data size, cell overlap, and information loss of the EMR data cubes. Results: Global generalization considerably reduced the size of the EMR data cube and did not cause the data cube cells to overlap, but incurred a large amount of information loss. Local generalization maintained the data size and generated only moderate information loss, but there were cell overlaps that could decrease the search performance. Bucketization did not cause cells to overlap and generated little information loss; however, the method considerably inflated the size of the EMR data cubes. Conclusions: The utility of anonymized EMR data cubes varies widely according to the anonymization method, and the applicability of the anonymization method depends on the features of the EMR analysis environment. The findings help to adopt the optimal anonymization method considering the EMR analysis environment and goal of the EMR analysis.

Utilizing uncoded consultation notes from electronic medical records for predictive modeling of colorectal cancer.

Machine learning techniques can be used to extract predictive models for diseases from electronic medical records (EMRs). However, the nature of EMRs makes it difficult to apply off-the-shelf machine learning techniques while still exploiting the rich content of the EMRs. In this paper, we explore the usage of a range of natural language processing (NLP) techniques to extract valuable predictors from uncoded consultation notes and study whether they can help to improve predictive performance. We study a number of existing techniques for the extraction of predictors from the consultation notes, namely a bag of words based approach and topic modeling. In addition, we develop a dedicated technique to match the uncoded consultation notes with a medical ontology. We apply these techniques as an extension to an existing pipeline to extract predictors from EMRs. We evaluate them in the context of predictive modeling for colorectal cancer (CRC), a disease known to be difficult to diagnose before performing an endoscopy. Our results show that we are able to extract useful information from the consultation notes. The predictive performance of the ontology-based extraction method moves significantly beyond the benchmark of age and gender alone (area under the receiver operating characteristic curve (AUC) of 0.870 versus 0.831). We also observe more accurate predictive models by adding features derived from processing the consultation notes compared to solely using coded data (AUC of 0.896 versus 0.882) although the difference is not significant. The extracted features from the notes are shown be equally predictive (i.e. there is no significant difference in performance) compared to the coded data of the consultations. It is possible to extract useful predictors from uncoded consultation notes that improve predictive performance. Techniques linking text to concepts in medical ontologies to derive these predictors are shown to perform best for predicting CRC in our EMR dataset.

Hierarchical Bayesian nonparametric models for knowledge discovery from electronic medical records

Electronic Medical Record (EMR) has established itself as a valuable resource for large scale analysis of health data. A hospital EMR dataset typically consists of medical records of hospitalized patients. A medical record contains diagnostic information (diagnosis codes), procedures performed (procedure codes) and admission details. Traditional topic models, such as latent Dirichlet allocation (LDA) and hierarchical Dirichlet process (HDP), can be employed to discover disease topics from EMR data by treating patients as documents and diagnosis codes as words. This topic modeling helps to understand the constitution of patient diseases and offers a tool for better planning of treatment. In this paper, we propose a novel and flexible hierarchical Bayesian nonparametric model, the word distance dependent Chinese restaurant franchise (wddCRF), which incorporates word-to-word distances to discover semantically-coherent disease topics. We are motivated by the fact that diagnosis codes are connected in the form of ICD-10 tree structure which presents semantic relationships between codes. We exploit a decay function to incorporate distances between words at the bottom level of wddCRF. Efficient inference is derived for the wddCRF by using MCMC technique. Furthermore, since procedure codes are often correlated with diagnosis codes, we develop the correspondence wddCRF (Corr-wddCRF) to explore conditional relationships of procedure codes for a given disease pattern. Efficient collapsed Gibbs sampling is derived for the Corr-wddCRF. We evaluate the proposed models on two real-world medical datasets – PolyVascular disease and Acute Myocardial Infarction disease. We demonstrate that the Corr-wddCRF model discovers more coherent topics than the Corr-HDP. We also use disease topic proportions as new features and show that using features from the Corr-wddCRF outperforms the baselines on 14-days readmission prediction. Beside these, the prediction for procedure codes based on the Corr-wddCRF also shows considerable accuracy.

Hospitals' Patterns of Use of Noninvasive Ventilation in Patients With Asthma Exacerbation.

Limited data are available on the use of noninvasive ventilation in patients with asthma exacerbations. The objective of this study was to characterize hospital patterns of noninvasive ventilation use in patients with asthma and to evaluate the association with the use of invasive mechanical ventilation and case fatality rate. This cross-sectional study used an electronic medical record dataset, which includes comprehensive pharmacy and laboratory results from 58 hospitals. Data on 13,558 patients admitted from 2009 to 2012 were analyzed. Initial noninvasive ventilation (NIV) or invasive mechanical ventilation (IMV) was defined as the first ventilation method during hospitalization. Hospital-level risk-standardized rates of NIV among all admissions with asthma were calculated by using a hierarchical regression model. Hospitals were grouped into quartiles of NIV to compare the outcomes. Overall, 90.3% of patients with asthma were not ventilated, 4.0% were ventilated with NIV, and 5.7% were ventilated with IMV. Twenty-two (38%) hospitals did not use NIV for any included admissions. Hospital-level adjusted NIV rates varied considerably (range, 0.4-33.1; median, 5.2%). Hospitals in the highest quartile of NIV did not have lower IMV use (5.4% vs 5.7%), but they did have a small but significantly shorter length of stay. Higher NIV rates were not associated with lower risk-adjusted case fatality rates. Large variation exists in hospital use of NIV for patients with an acute exacerbation of asthma. Higher hospital rates of NIV use does not seem to be associated with lower IMV rates. These results indicate a need to understand contextual and organizational factors contributing to this variability.

Identification of subjects with polycystic ovary syndrome using electronic health records.

BackgroundPolycystic ovary syndrome (PCOS) is a heterogeneous disorder because of the variable criteria used for diagnosis. Therefore, International Classification of Diseases 9 (ICD-9) codes may not accurately capture the diagnostic criteria necessary for large scale PCOS identification. We hypothesized that use of electronic medical records text and data would more specifically capture PCOS subjects.MethodsSubjects with PCOS were identified in the Partners Healthcare Research Patients Data Registry by searching for the term “polycystic ovary syndrome” using natural language processing (n = 24,930). A training subset of 199 identified charts was reviewed and categorized based on likelihood of a true Rotterdam PCOS diagnosis, i.e. two out of three of the following: irregular menstrual cycles, hyperandrogenism and/or polycystic ovary morphology. Data from the history, physical exam, laboratory and radiology results were codified and extracted from notes of definite PCOS subjects. Thirty-two terms were used to build an algorithm for identifying definite PCOS cases and applied to the rest of the dataset. The positive predictive value cutoff was set at 76.8 % to maximize the number of subjects available for study. A true positive predictive value for the algorithm was calculated after review of 100 charts from subjects identified as definite PCOS cases with at least two documented Rotterdam criteria. The positive predictive value was compared to that calculated using 200 charts identified using the ICD-9 code for PCOS (256.4; n = 13,670). In addition, a cohort of previously recruited PCOS subjects was submitted for algorithm validation.ResultsChart review demonstrated that 64 % were confirmed as definitely PCOS using the algorithm, with a 9 % false positive rate. 66 % of subjects identified by ICD-9 code for PCOS could be confirmed as definitely PCOS, with an 8.5 % false positive rate. There was no significant difference in the positive predictive values using the two methods (p = 0.2). However, the number of charts that had insufficient confirmatory data was lower using the algorithm (5 % vs 11 %; p < 0.04). Of 477 subjects with PCOS recruited and examined individually and present in the database as patients, 451 were found within the algorithm dataset.ConclusionsExtraction of text parameters along with codified data improves the confidence in PCOS patient cohorts identified using the electronic medical record. However, the positive predictive value was not significantly different when using ICD-9 codes or the specific algorithm. Further studies are needed to determine the positive predictive value of the two methods in additional electronic medical record datasets.Electronic supplementary materialThe online version of this article (doi:10.1186/s12958-015-0115-z) contains supplementary material, which is available to authorized users.

An empirical evaluation of supervised learning approaches in assigning diagnosis codes to electronic medical records.

Diagnosis codes are assigned to medical records in healthcare facilities by trained coders by reviewing all physician authored documents associated with a patient's visit. This is a necessary and complex task involving coders adhering to coding guidelines and coding all assignable codes. With the popularity of electronic medical records (EMRs), computational approaches to code assignment have been proposed in the recent years. However, most efforts have focused on single and often short clinical narratives, while realistic scenarios warrant full EMR level analysis for code assignment. We evaluate supervised learning approaches to automatically assign international classification of diseases (ninth revision) - clinical modification (ICD-9-CM) codes to EMRs by experimenting with a large realistic EMR dataset. The overall goal is to identify methods that offer superior performance in this task when considering such datasets. We use a dataset of 71,463 EMRs corresponding to in-patient visits with discharge date falling in a two year period (2011-2012) from the University of Kentucky (UKY) Medical Center. We curate a smaller subset of this dataset and also use a third gold standard dataset of radiology reports. We conduct experiments using different problem transformation approaches with feature and data selection components and employing suitable label calibration and ranking methods with novel features involving code co-occurrence frequencies and latent code associations. Over all codes with at least 50 training examples we obtain a micro F-score of 0.48. On the set of codes that occur at least in 1% of the two year dataset, we achieve a micro F-score of 0.54. For the smaller radiology report dataset, the classifier chaining approach yields best results. For the smaller subset of the UKY dataset, feature selection, data selection, and label calibration offer best performance. We show that datasets at different scale (size of the EMRs, number of distinct codes) and with different characteristics warrant different learning approaches. For shorter narratives pertaining to a particular medical subdomain (e.g., radiology, pathology), classifier chaining is ideal given the codes are highly related with each other. For realistic in-patient full EMRs, feature and data selection methods offer high performance for smaller datasets. However, for large EMR datasets, we observe that the binary relevance approach with learning-to-rank based code reranking offers the best performance. Regardless of the training dataset size, for general EMRs, label calibration to select the optimal number of labels is an indispensable final step.

Use of augmentation agents for treating depression: analysis of a psychiatric electronic medical record data set.

This study evaluated the relationship between patient characteristics and augmentation strategies for the treatment of major depressive disorder. This retrospective, cross-sectional study used data from a psychiatric electronic medical record database for patients with depression without psychosis or psychotic features who initiated augmentation therapy between January 2001 and June 2011. Medical records were evaluated to identify factors predicting use of specific augmentation agents, and a multivariate logistic regression model was used to assess clinical and demographic predictors of augmentation strategy. Of 3,209 patients initiating augmentation therapy for depression, 75% received augmentation with an antidepressant combination and 11% received augmentation with second-generation antipsychotics. Baseline clinical severity (Clinical Global Impressions-Severity score) most strongly and consistently predicted augmentation with second-generation antipsychotics. Treatment of patients in specialty settings with depression was often augmented with an antidepressant combination, whereas those with severe depression had an increased likelihood of augmentation with second-generation antipsychotics.

Supervised Extraction of Diagnosis Codes from EMRs: Role of Feature Selection, Data Selection, and Probabilistic Thresholding.

Extracting diagnosis codes from medical records is a complex task carried out by trained coders by reading all the documents associated with a patient's visit. With the popularity of electronic medical records (EMRs), computational approaches to code extraction have been proposed in the recent years. Machine learning approaches to multi-label text classification provide an important methodology in this task given each EMR can be associated with multiple codes. In this paper, we study the the role of feature selection, training data selection, and probabilistic threshold optimization in improving different multi-label classification approaches. We conduct experiments based on two different datasets: a recent gold standard dataset used for this task and a second larger and more complex EMR dataset we curated from the University of Kentucky Medical Center. While conventional approaches achieve results comparable to the state-of-the-art on the gold standard dataset, on our complex in-house dataset, we show that feature selection, training data selection, and probabilistic thresholding provide significant gains in performance.

Assessment of time to follow-up visits in newly-treated hypertensive patients using an electronic medical record database

Objective:Use of electronic medical record (EMR) data for evaluating healthcare processes and outcomes is relatively new. Using EMR data, this study evaluated the time from antihypertensive initiation to the first follow-up office visit controlling for adverse events (AEs) and other factors that could influence follow-up timing. Findings were compared to treatment guidelines which recommend monthly follow-up in treatment naive patients until blood pressure (BP) levels are controlled.Research design and methods:Treatment-naïve hypertensive adult patients in the General Electric Centricity EMR database (1996–2006) with a new antihypertensive prescription were evaluated. Time from treatment initiation to first office visit was identified and stratified by occurrence of AEs and therapy change. BP was assessed at 120 ± 30 days.Results:The mean ± SD time from first antihypertensive prescription (index date) to the first office visit was 96.2 ± 160.6 days; 38% returned within a month of treatment initiation. Controlling for baseline demographic and clinical characteristics, the adjusted time until first office visit was shorter for those with an AE and therapy change than for those with neither event (61 vs. 158 days). Of the patients with follow-up BP data for analysis (n = 27 875), more of those seen within a month of treatment initiation achieved BP goal at 120 days (<140/90 mmHg) than those who were not seen within a month (64.3 vs. 61.7% respectively; p < 0.001).Conclusions:This study demonstrates that EMR data can be used to assess quality measures which in turn can inform efforts to improve treatment outcomes. Specifically, this study evaluated mean time to first office visit after antihypertensive therapy initiation controlling for clinical factors that could influence office visit intervals based on data available in a national EMR dataset. A key limitation of this study is that the EMR may not represent patient care delivered by other providers, thus, use of antihypertensives, changes in therapy, and office visits may be underreported.

Semi‐automated risk estimation using large databases: quinolones and clostridium difficile associated diarrhea

The availability of large databases with person time information and appropriate statistical methods allow for relatively rapid pharmacovigilance analyses. A semi-automated method was used to investigate the effect of fluoroquinolones on the incidence of C. difficile associated diarrhea (CDAD). Two US databases, an electronic medical record (EMR) and a large medical claims database for the period 2006-2007 were evaluated using a semi-automated methodology. The raw EMR and claims datasets were subject to a normalization procedure that aligns the drug exposures and conditions using ontologies; Snowmed for medications and MedDRA for conditions. A retrospective cohort design was used together with matching by means of the propensity score. The association between exposure and outcome was evaluated using a Poisson regression model after taking into account potential confounders. A comparison between quinolones as the target cohort and macrolides as the comparison cohort produced a total of 564,797 subjects exposed to a quinolone in the claims data and 233,090 subjects in the EMR. They were matched with replacement within six strata of the propensity score. Among the matched cohorts there were a total of 488 and 158 outcomes in the claims and the EMR respectively. Quinolones were found to be twice more likely to be significantly associated with CDAD than macrolides adjusting for risk factors (IRR 2.75, 95%CI 2.18-3.48). Use of a semi-automated method was successfully applied to two observational databases and was able to rapidly identify a potential for increased risk of developing CDAD with quinolones.

Registries for Robust Evidence

ALTHOUGH RANDOMIZED CLINICAL TRIALS (RCTS) ARE the bedrock for establishing which interventions are efficacious, there is increasing recognition that they cannot address all needs, especially the need to determine, in a timely manner, the safety and effectiveness of different interventions used in the diverse array of patients and settings that make up a health care system. Interest has increased in the role of observational studies, and more specifically in registries and other electronic data sets, as a way to fill these critical gaps in evidence and as useful guides for helping to determine formulary placement. The recently released “highest priority challenge topics” from the National Institutes of Health (NIH) pointedly reference registries more than 20 times. Yet little guidance is available to help patients, physicians, payers, researchers, and policy makers evaluate the quality of information derived from these nonexperimental sources. The Agency for Healthcare Research and Quality defines a patient registry for evaluating outcomes as “an organized system that uses observational study methods to collect uniform data (clinical and other) to evaluate specified outcomes for a population defined by a particular disease, condition, or exposure, and that serves a predetermined scientific, clinical, or policy purpose(s).” Registries and other noninterventional studies are often referred to as real-world data to distinguish them from most clinical trials. These realworld studies are a heterogeneous mix ranging from prospective observational studies like registries to studies of large administrative and electronic medical record data sets collected for other purposes. As more attention is focused on using real-world data to build an evidence base, it is important to know how to evaluate the quality and usability of studies derived from these types of observations. Registries could be organized around conditions or exposures, such as a particular disease; a health care service (eg, procedure); or a product (drug or device) and can address questions ranging from treatment effectiveness and safety to the quality of care delivered. Registries vary in complexity from simply recording product use as a requirement for reimbursement to more systematic efforts to collect prospective data on many types of treatment, risk factors, and clinical events in a defined population. Follow-up could be retrospective, prospective, or a combination of both. The mode and duration of follow-up could range from days (eg, hospital admission registry) to decades (eg, orthopedic implant registry). How the denominator population is defined and enumerated depends on the research question. For example, to estimate the proportion of a population that has been vaccinated, the underlying population would need to be systematically sampled along with those who receive vaccines. In contrast, the safety and effectiveness of a new treatment could be studied by following typical patients who receive the new treatment to evaluate how their condition resolves and whether any untoward events occur that appear to be related to the treatment. Registries are being used to fill important gaps in evidence and contribute to understanding how trial results can be applied in practice. For example, observational data were used to compare coronary artery bypass graft (CABG) surgery with percutaneous coronary intervention (PCI). The available trial evidence was derived mainly from patients with singleor 2-vessel coronary disease—and did not reflect other therapies in use at that time, such as minimally invasive surgery, off-pump surgery, and drug-eluting stents. Physicians and health insurers sought evidence to help guide therapeutic decisions, but there were no trial data for the clinically diverse range of patients commonly treated. Although it appeared that myocardial infarction and mortality were comparable for PCI and CABG among patients with similar levels of coronary disease, registry data showed a strong gradient of benefit of CABG by severity of disease. Data from registries are also used to support timely decisions by regulatory agencies about safety and about coverage (payment). Safety data from the acyclovir pregnancy registry were used to change the Food and Drug Administration (FDA) pregnancy labeling category from category C (risk cannot be ruled out) to category B (no evidence of risk in humans). The FDA has also used observational data to expand labeled indications, such as broadening the age groups for intra-ocular lenses through review of registry data,

Compliance (COMP) with colon cancer (CoC) national guidelines (NG) for chemotherapy (CT) by implementing electronic medical records (EMR)

17527 Background: Quality of care (QC) improves when NG are followed in patient (pt) management. Demonstration of COMP with NG is costly and slow (QOPI, and ASCO 2007 abstract 17050). We extended our prior experience to apply an EMR to monitor our pts for COMP with NCCN CoC CT guidelines. Methods: An EMR was implemented in 2003 in Wilshire Oncology Medical Group (6 CT sites and 8 medical oncologists (MO)). All pts seen in the offices were entered into the EMR by MO (ICD9 diagnosis (DX), stage (STG), history, orders) and staff (CT given). All pts were entered into a separate nonintegrated billing system (BIS) by staff. All pts with CoC seen between 6–1-06 and 5–31- 07 were evaluated electronically for DX, CT, and STG using EMR and checked in BIS. To study changes in COMP with NCCN NG for CT over time, 2 cohorts of pts were compared: pts initially seen prior to 6–1-07 (PRIOR); versus pts new after 6–1-07 (NEW). Results: Of 345 CoC pts seen by MO, 78 (25%) were seen in hospital only and not followed in office due to insurance, transfer of care, hospice, or no show. 33 (9.6%) had an incorrect ICD9 DX by MO or staff. Of 234 CoC pt in office, no STG was entered in 15 (6.4%, range per MO 0% to 13%). This was more frequent in PRIOR pt (8.9%) than NEW pt (2.3%, p=0.05). Pts were STG 0 2%, STG 1 12%, STG 2 22%, STG 3 21%, and STG 4 13%. CT was given to 0% STG 0, 3% STG 1, 24% STG 2, 29% STG 3 and 78% STG 4. CT improved by NG from PRIOR to NEW pt: STG 1 4% PRIOR to 0% NEW; STG 2 12% PRIOR to 44% NEW (p=0.02); STG 3 20% PRIOR to 52% NEW (p=0.008); but no change in STG 4 (81% to 75%, p=0.6). CT for STGs 2,3, or 4 increased from 28% to 53.6% (p=0.008), and varied by MO for NEW pt from 33% to 71%. Evaluation of reason for non-COMP with NG could not be evaluated electronically using this EMR due to lack of explanatory codes. Conclusions: EMRs can be used to monitor COMP with NG, track improvement over time, and evaluate variances in COMP between MO. MO completion of EMR data elements improves with time, but EMR data sets must be frequently monitored for completeness and accuracy by MO. EMRs should be integrated with BIS to avoid incorrect DX. Monitoring lack of 100% COMP with NG requires an additional explanatory code in EMR. QC can be improved by appropriate EMR use. This EMR has been updated after 6–07 to implement these conclusions. No significant financial relationships to disclose.