Learning statistical models of phenotypes using noisy labeled training data.

OBJECTIVE: Traditionally, patient groups with a phenotype are selected through rule-based definitions whose creation and validation are time-consuming. Machine learning approaches to electronic phenotyping are limited by the paucity of labeled training datasets. We demonstrate the feasibility of utilizing semi-automatically labeled training sets to create phenotype models via machine learning, using a comprehensive representation of the patient medical record. METHODS: We use a list of keywords specific to the phenotype of interest to generate noisy labeled training data. We train L1 penalized logistic regression models for a chronic and an acute disease and evaluate the performance of the models against a gold standard. RESULTS: Our models for Type 2 diabetes mellitus and myocardial infarction achieve precision and accuracy of 0.90, 0.89, and 0.86, 0.89, respectively. Local implementations of the previously validated rule-based definitions for Type 2 diabetes mellitus and myocardial infarction achieve precision and accuracy of 0.96, 0.92 and 0.84, 0.87, respectively.We have demonstrated feasibility of learning phenotype models using imperfectly labeled data for a chronic and acute phenotype. Further research in feature engineering and in specification of the keyword list can improve the performance of the models and the scalability of the approach. CONCLUSIONS: Our method provides an alternative to manual labeling for creating training sets for statistical models of phenotypes. Such an approach can accelerate research with large observational healthcare datasets and may also be used to create local phenotype models.

A nurse-driven screening tool for the early identification of sepsis in an intermediate care unit setting.

BACKGROUND: Use of a screening tool as a decision support mechanism for early detection of sepsis has been widely advocated, yet studies validating tool performance are scarce, especially in non-intensive care unit settings. METHODS: For this pilot study we prospectively screened consecutive patients admitted to a medical/surgical intermediate care unit at an academic medical center over a 1-month period and retrospectively analyzed their clinical data. Patients were screened with a 3-tiered, paper-based, nurse-driven sepsis assessment tool every 8 hours. For patients screening positive for sepsis or severe sepsis, the primary treatment team was notified and the team's clinical actions were recorded. Results of the screening test were then compared to patient International Classification of Diseases, Ninth Revision (ICD-9) codes for sepsis, severe sepsis, and septic shock identified during the study time period, and performance of the screening test was assessed. RESULTS: A total of 2143 screening tests were completed in 245 patients (169 surgical, 76 medical). ICD-9 codes confirmed sepsis incidence was 9%. Of the 39 patients who screened positive, 51% were positive for sepsis, and 49% screened positive for severe sepsis. Screening tool sensitivity and specificity were 95% and 92%, respectively. Negative predictive value was 99% and positive predictive value was 54%. Overall test accuracy was 92%. There was no statistically significant difference in tool performance between medical and surgical patients. CONCLUSIONS: A simple screening tool for sepsis utilized as part of nursing assessment may be a useful way of identifying early sepsis in both medical and surgical patients in an intermediate care unit setting.

Factors impacting follow-up care after placement of temporary inferior vena cava filters.

OBJECTIVE: Rates of inferior vena cava (IVC) filter retrieval have remained suboptimal, in part because of poor follow-up. The goal of our study was to determine demographic and clinical factors predictive of IVC filter follow-up care in a university hospital setting. METHODS: We reviewed 250 consecutive patients who received an IVC filter placement with the intention of subsequent retrieval between March 2009 and October 2010. Patient demographics, clinical factors, and physician specialty were evaluated. Multivariate logistic regression analysis was performed to identify variables predicting follow-up care. RESULTS: In our cohort, 60.7% of patients received follow-up care; of those, 93% had IVC filter retrieval. Major indications for IVC filter placement were prophylaxis for high risk surgery (53%) and venous thromboembolic event with contraindication and/or failure of anticoagulation (39%). Follow-up care was less likely for patients discharged to acute rehabilitation or skilled nursing facilities (P < .0001), those with central nervous system pathology (eg, cerebral hemorrhage or spinal fracture; P < .0001), and for those who did not receive an IVC filter placement by a vascular surgeon (P < .0001). In a multivariate analysis, discharge home (odds ratio [OR], 4.0; 95% confidence interval [CI], 1.99-8.2; P < .0001), central nervous system pathology (OR, 0.46; 95% CI, 0.22-0.95; P = .04), and IVC filter placement by the vascular surgery service (OR, 4.7; 95% CI, 2.3-9.6; P < .0001) remained independent predictors of follow-up care. Trauma status and distance of residence did not significantly impact likelihood of patient follow-up. CONCLUSIONS: Service-dependent practice paradigms play a critical role in patient follow-up and IVC filter retrieval rates. Nevertheless, specific patient populations are more prone to having poorer rates of follow-up. Such trends should be factored into institutional quality control goals and patient-centered care.

Effect of chronic red cell transfusion therapy on vasculopathies and silent infarcts in patients with sickle cell disease.

Regular, chronic red cell transfusions (CTX) have been shown to be effective prophylaxis against stroke in sickle cell disease (SCD) in those at risk. Because serial brain imaging is not routinely performed, little is known about the impact of CTX on silent infarcts (SI) and cerebral vascular pathology. Thus, we retrospectively evaluated the magnetic resonance imaging reports of a cohort of SCD patients who were prescribed CTX for either primary or secondary stroke prophylaxis. Seventeen patients with Hb SS were included (mean age 15 years, mean follow-up 4.3 years). Eight patients were on CTX for primary prophylaxis. New SI occurred in 17.6% of patients corresponding to an SI rate of 5.42 per 100 patient-years. Vasculopathy of the cerebral arteries was present in 65% of patients and progressed in 63% of these patients. Those who developed progressive vasculopathy were on CTX for an average of 8 years before lesions progressed. Patients on CTX for secondary prophylaxis had more SIs and evidence of progressive vascular disease than patients on CTX for primary prophylaxis. We conclude that adherence to CTX does not necessarily prevent SI or halt cerebral vasculopathy progression, especially in those with a history of overt stroke.