Relational machine learning for electronic health record-driven phenotyping.

OBJECTIVE: Electronic health records (EHR) offer medical and pharmacogenomics research unprecedented opportunities to identify and classify patients at risk. EHRs are collections of highly inter-dependent records that include biological, anatomical, physiological, and behavioral observations. They comprise a patient's clinical phenome, where each patient has thousands of date-stamped records distributed across many relational tables. Development of EHR computer-based phenotyping algorithms require time and medical insight from clinical experts, who most often can only review a small patient subset representative of the total EHR records, to identify phenotype features. In this research we evaluate whether relational machine learning (ML) using inductive logic programming (ILP) can contribute to addressing these issues as a viable approach for EHR-based phenotyping. METHODS: Two relational learning ILP approaches and three well-known WEKA (Waikato Environment for Knowledge Analysis) implementations of non-relational approaches (PART, J48, and JRIP) were used to develop models for nine phenotypes. International Classification of Diseases, Ninth Revision (ICD-9) coded EHR data were used to select training cohorts for the development of each phenotypic model. Accuracy, precision, recall, F-Measure, and Area Under the Receiver Operating Characteristic (AUROC) curve statistics were measured for each phenotypic model based on independent manually verified test cohorts. A two-sided binomial distribution test (sign test) compared the five ML approaches across phenotypes for statistical significance. RESULTS: We developed an approach to automatically label training examples using ICD-9 diagnosis codes for the ML approaches being evaluated. Nine phenotypic models for each ML approach were evaluated, resulting in better overall model performance in AUROC using ILP when compared to PART (p=0.039), J48 (p=0.003) and JRIP (p=0.003). DISCUSSION: ILP has the potential to improve phenotyping by independently delivering clinically expert interpretable rules for phenotype definitions, or intuitive phenotypes to assist experts. CONCLUSION: Relational learning using ILP offers a viable approach to EHR-driven phenotyping.

The use of dietary supplements and their association with blood pressure in a large Midwestern cohort.

BACKGROUND: There have been numerous studies assessing the association of diet and blood pressure but little is known about the association between less commonly used nutritional supplements and blood pressured. The purpose of this study was to quantify the use of dietary supplements and their potential association with blood pressure in a large population-based cohort of adults in the Midwest. METHODS: The Personalized Medicine Research Project cohort was the population source for the current study. The current study includes subjects with Dietary History Questionnaire (DHQ) data available as well as at least one clinical blood pressure measurement recorded in their electronic medical record. After excluding extreme outlying measurements, median systolic and diastolic blood pressure measurements were calculated for each individual and were compared for subjects who did and did not report taking one of a list of 37 different supplements listed on the DHQ more than once per week over the previous 12 months. RESULTS: 9,732 subjects had both blood pressure and DHQ data available. They ranged in age from 18 to 98 years (mean 56 years) and 3,625 (37%) were male. Nine of 37 supplements showed evidence for association with blood pressure: coenzyme Q10, fish oil, iron, bilberry, echinacea, evening primrose oil, garlic, goldenseal and milk thistle. With the exception of the mineral iron, mean systolic and diastolic blood pressures were higher for users of the specific supplements than non-users. CONCLUSIONS: These results should not be interpreted as causal, nor can the direction of the association be assumed to be correct because the temporality of the association is unknown. We hope the observed significant associations will foster future research to evaluate blood pressure effects of dietary supplements.

A randomized controlled trial of genotype-based Coumadin initiation.

PURPOSE: A randomized controlled trial was conducted in patients initiating warfarin to determine whether algorithms that incorporate genotypes affecting warfarin metabolism and function, and Vitamin K metabolism improve prediction of therapeutic warfarin dose and anticoagulation management. METHODS: A total of 230 patients were randomized to either a clinical arm where dosing algorithms considered only clinical information or an interventional arm where dosing algorithms used clinical and genotypic variables (CYP2C9, CYP4F2, and VKORC1). Subjects in the interventional arm were genotyped within 5 hours, and the initial dose was informed by genotype. Primary endpoints were absolute prediction error relative to therapeutic dose, and time in therapeutic target range during the first 14 days. Secondary endpoints included time to stable dose in therapeutic range, time to first international normalization ratio >4, and warfarin-related adverse events. RESULTS: The model including genetics more accurately identified therapeutic dose twice as often as the clinical model (65.3% vs. 34.7%) (P < 0.0001). Patients in the interventional arm did not achieve greater time in therapeutic range. Study arms were similar regarding time to international normalization ratio >4 and adverse events. CONCLUSION: Genotype-informed dosing clearly improved prediction of therapeutic dose beyond that available with clinical parameters. Genetic information did not affect time in therapeutic target range during the first 14 days of therapy. Current management practices with the vagaries in dose adjustment after warfarin initiation exert a strong influence on traditional clinical outcomes.

Medical providers' dental information needs: a baseline survey.

Articulation of medical and dental practices has been strongly called for based on the many oral-systemic connections. With the rapid development and adoption of electronic health records, the feasibility of integrating medical and dental patient data should be strongly considered. The objective of this study was to develop an initial understanding of the medical providers' core dental information needs and opinion of integrated medical-dental electronic health record (iEHR) environment in their workflow. This was achieved by administering a 13 question survey to a group of 1,197 medical care providers employed by Marshfield Clinic in Wisconsin, United States. The survey received a response rate of 35%. The responses were analyzed based on provider 'Role' and 'Specialty'. The majority of the respondents felt the need for patient's dental information to coordinate or provide effective medical care. An integrated electronic health record environment could facilitate this holistic patient care approach.

Validation of PhenX measures in the personalized medicine research project for use in gene/environment studies.

BACKGROUND: The purpose of this paper is to describe the data collection efforts and validation of PhenX measures in the Personalized Medicine Research Project (PMRP) cohort. METHODS: Thirty-six measures were chosen from the PhenX Toolkit within the following domains: demographics; anthropometrics; alcohol, tobacco and other substances; cardiovascular; environmental exposures; cancer; psychiatric; neurology; and physical activity and physical fitness. Eligibility criteria for the current study included: living PMRP subjects with known addresses who consented to future contact and were not currently living in a nursing home, available GWAS data from eMERGE I for subjects where age-related cataract, HDL, dementia and resistant hypertension were the primary phenotypes, thus biasing the sample to the older PMRP participants. The questionnaires were mailed twice. Data from the PhenX measures were compared with information from PMRP questionnaires and data from Marshfield Clinic electronic medical records. RESULTS: Completed PhenX questionnaires were returned by 2271 subjects for a final response rate of 70%. The mean age reported on the PhenX questionnaire (73.1 years) was greater than the PMRP questionnaire (64.8 years) because the data were collected at different time points. The mean self-reported weight, and subsequently calculated BMI, were less on the PhenX survey than the measured values at the time of enrollment into PMRP (PhenX means 173.5 pounds and BMI 28.2 kg/m2 versus PMRP 182.9 pounds and BMI 29.6 kg/m2). There was 95.3% agreement between the two questionnaires about having ever smoked at least 100 cigarettes. 139 (6.2%) of subjects indicated on the PhenX questionnaire that they had been told they had a stroke. Of them, only 15 (10.8%) had no electronic indication of a prior stroke or TIA. All of the age-and gender-specific 95% confidence limits around point estimates for major depressive episodes overlap and show that 31% of women aged 50-64 reported symptoms associated with a major depressive episode. CONCLUSIONS: The approach employed resulted in a high response rate and valuable data for future gene/environment analyses. These results and high response rate highlight the utility of the PhenX Toolkit to collect valid phenotypic data that can be shared across groups to facilitate gene/environment studies.

Enhancing the power of genetic association studies through the use of silver standard cases derived from electronic medical records.

The feasibility of using imperfectly phenotyped "silver standard" samples identified from electronic medical record diagnoses is considered in genetic association studies when these samples might be combined with an existing set of samples phenotyped with a gold standard technique. An analytic expression is derived for the power of a chi-square test of independence using either research-quality case/control samples alone, or augmented with silver standard data. The subset of the parameter space where inclusion of silver standard samples increases statistical power is identified. A case study of dementia subjects identified from electronic medical records from the Electronic Medical Records and Genomics (eMERGE) network, combined with subjects from two studies specifically targeting dementia, verifies these results.