Accelerating Surgical Site Infection Abstraction With a Semi-automated Machine-learning Approach.

OBJECTIVE: To demonstrate that a semi-automated approach to health data abstraction provides significant efficiencies and high accuracy. BACKGROUND: Surgical outcome abstraction remains laborious and a barrier to the sustainment of quality improvement registries like ACS-NSQIP. A supervised machine learning algorithm developed for detecting SSi using structured and unstructured electronic health record data was tested to perform semi-automated SSI abstraction. METHODS: A Lasso-penalized logistic regression model with 2011-3 data was trained (baseline performance measured with 10-fold cross-validation). A cutoff probability score from the training data was established, dividing the subsequent evaluation dataset into "negative" and "possible" SSI groups, with manual data abstraction only performed on the "possible" group. We evaluated performance on data from 2014, 2015, and both years. RESULTS: Overall, 6188 patients were in the 2011-3 training dataset and 5132 patients in the 2014-5 evaluation dataset. With use of the semi-automated approach, applying the cut-off score decreased the amount of manual abstraction by >90%, resulting in < 1% false negatives in the "negative" group and a sensitivity of 82%. A blinded review of 10% of the "possible" group, considering only the features selected by the algorithm, resulted in high agreement with the gold standard based on full chart abstraction, pointing towards additional efficiency in the abstraction process by making it possible for abstractors to review limited, salient portions of the chart. CONCLUSION: Semi-automated machine learning-aided SSI abstraction greatly accelerates the abstraction process and achieves very good performance. This could be translated to other post-operative outcomes and reduce cost barriers for wider ACS-NSQIP adoption.

Strategies for building robust prediction models using data unavailable at prediction time.

OBJECTIVE: Hospital-acquired infections (HAIs) are associated with significant morbidity, mortality, and prolonged hospital length of stay. Risk prediction models based on pre- and intraoperative data have been proposed to assess the risk of HAIs at the end of the surgery, but the performance of these models lag behind HAI detection models based on postoperative data. Postoperative data are more predictive than pre- or interoperative data since it is closer to the outcomes in time, but it is unavailable when the risk models are applied (end of surgery). The objective is to study whether such data, which is temporally unavailable at prediction time (TUP) (and thus cannot directly enter the model), can be used to improve the performance of the risk model. MATERIALS AND METHODS: An extensive array of 12 methods based on logistic/linear regression and deep learning were used to incorporate the TUP data using a variety of intermediate representations of the data. Due to the hierarchical structure of different HAI outcomes, a comparison of single and multi-task learning frameworks is also presented. RESULTS AND DISCUSSION: The use of TUP data was always advantageous as baseline methods, which cannot utilize TUP data, never achieved the top performance. The relative performances of the different models vary across the different outcomes. Regarding the intermediate representation, we found that its complexity was key and that incorporating label information was helpful. CONCLUSIONS: Using TUP data significantly helped predictive performance irrespective of the model complexity.

Bariatric Surgery is Associated With Decreased Progression of Nonalcoholic Fatty Liver Disease to Cirrhosis: A Retrospective Cohort Analysis.

OBJECTIVE: This study sought to compare trends in the development of cirrhosis between patients with NAFLD who underwent bariatric surgery and a well-matched group of nonsurgical controls. SUMMARY OF BACKGROUND DATA: Patients with NAFLD who undergo bariatric surgery generally have improvements in liver histology. However, the long-term effect of bariatric surgery on clinically relevant liver outcomes has not been investigated. METHODS: From a large insurance database, patients with a new NAFLD diagnosis and at least 2 years of continuous enrollment before and after diagnosis were identified. Patients with traditional contraindications to bariatric surgery were excluded. Patients who underwent bariatric surgery were identified and matched 1:2 with patients who did not undergo bariatric surgery based on age, sex, and comorbid conditions. Kaplan-Meier analysis and Cox proportional hazards modeling were used to evaluate differences in progression from NAFLD to cirrhosis. RESULTS: A total of 2942 NAFLD patients who underwent bariatric surgery were identified and matched with 5884 NAFLD patients who did not undergo surgery. Cox proportional hazards modeling found that bariatric surgery was independently associated with a decreased risk of developing cirrhosis (hazard ratio 0.31, 95% confidence interval 0.19-0.52). Male gender was associated with an increased risk of cirrhosis (hazard ratio 2.07, 95% confidence interval 1.31-3.27). CONCLUSIONS: Patients with NAFLD who undergo bariatric surgery are at a decreased risk for progression to cirrhosis compared to well-matched controls. Bariatric surgery should be considered as a treatment strategy for otherwise eligible patients with NAFLD. Future bariatric surgery guidelines should include NAFLD as a comorbid indication when determining eligibility.

Quantifying the Effect of Data Quality on the Validity of an eMeasure.

Objective The objective of this study was to demonstrate the utility of a healthcare data quality framework by using it to measure the impact of synthetic data quality issues on the validity of an eMeasure (CMS178­urinary catheter removal after surgery). Methods Data quality issues were artificially created by systematically degrading the underlying quality of EHR data using two methods: independent and correlated degradation. A linear model that describes the change in the events included in the eMeasure quantifies the impact of each data quality issue. Results Catheter duration had the most impact on the CMS178 eMeasure with every 1% reduction in data quality causing a 1.21% increase in the number of missing events. For birth date and admission type, every 1% reduction in data quality resulted in a 1% increase in missing events. Conclusion This research demonstrated that the impact of data quality issues can be quantified using a generalized process and that the CMS178 eMeasure, as currently defined, may not measure how well an organization is meeting the intended best practice goal. Secondary use of EHR data is warranted only if the data are of sufficient quality. The assessment approach described in this study demonstrates how the impact of data quality issues on an eMeasure can be quantified and the approach can be generalized for other data analysis tasks. Healthcare organizations can prioritize data quality improvement efforts to focus on the areas that will have the most impact on validity and assess whether the values that are reported should be trusted.

Causal Phenotyping for Susceptibility to Cardiotoxicity from Antineoplastic Breast Cancer Medications.

Cardiotoxicity is a relatively common and particularly important adverse event caused by chemotherapy for breast cancer patients. Typical associative phenotypes, such as risk factors associated with diabetes, can often be detected solely based on the data elements existing in electronic health records; however, causal phenotypes, such as risk factors causing cardiotoxicity, require establishing causation between chemotherapy and determining new heart disease, and cannot be directly observedfrom EHR. We propose three phenotyping algorithms to assess breast cancer patients' susceptibility to cardiotoxicity caused by five first-line antineoplastic drugs: (1) causal phenotype model to predict the patients' risk of cardiotoxicity as the difference between the heart disease risks with exposure and nonexposure to the drugs; (2) regular predictive model; (3) combined predictive model of the above two models. Concordances for three methods were 0.60, 0.62, and 0.68. When considering all exposed patients, concordances were 0.66, 0.58 and 0.65 at 280 days after treatment. The study demonstrates the potential utility of causal phenotyping.

Statin Use, Diabetes Incidence and Overall Mortality in Normoglycemic and Impaired Fasting Glucose Patients.

BACKGROUND: The association between the use of statins and the risk of diabetes and increased mortality within the same population has been a source of controversy, and may underestimate the value of statins for patients at risk. OBJECTIVE: We aimed to assess whether statin use increases the risk of developing diabetes or affects overall mortality among normoglycemic patients and patients with impaired fasting glucose (IFG). DESIGN AND PARTICIPANTS: Observational cohort study of 13,508 normoglycemic patients (n = 4460; 33% taking statins) and 4563 IFG patients (n = 1865; 41% taking statin) among residents of Olmsted County, Minnesota, with clinical data in the Mayo Clinic electronic medical record and at least one outpatient fasting glucose test between 1999 and 2004. Demographics, vital signs, tobacco use, laboratory results, medications and comorbidities were obtained by electronic search for the period 1999-2004. Results were analyzed by Cox proportional hazards models, and the risk of incident diabetes and mortality were analyzed by survival curves using the Kaplan-Meier method. MAIN MEASURES: The main endpoints were new clinical diagnosis of diabetes mellitus and total mortality. KEY RESULTS: After a mean of 6 years of follow-up, statin use was found to be associated with an increased risk of incident diabetes in the normoglycemic (HR 1.19; 95% CI, 1.05 to 1.35; p = 0.007) and IFG groups (HR 1.24; 95%CI, 1.11 to 1.38; p = 0.0001). At the same time, overall mortality decreased in both normoglycemic (HR 0.70; 95% CI, 0.66 to 0.80; p < 0.0001) and IFG patients (HR 0.77, 95% CI, 0.64 to 0.91; p = 0.0029) with statin use. CONCLUSION: In general, recommendations for statin use should not be affected by concerns over an increased risk of developing diabetes, since the benefit of reduced mortality clearly outweighs this small (19-24%) risk.

Leveraging an Electronic Health Record-Linked Biorepository to Generate a Metformin Pharmacogenomics Hypothesis.

Metformin is a first-line antihyperglycemic agent commonly prescribed in type 2 diabetes mellitus (T2DM), but whose pharmacogenomics are not clearly understood. Further, due to accumulating evidence highlighting the potential for metformin in cancer prevention and treatment efforts it is imperative to understand molecular mechanisms of metformin. In this electronic health record(EHR)-based study we explore the potential association of the flavin-containing monooxygenase(FMO)-5 gene, a biologically plausible biotransformer of metformin, and modifying glycemic response to metformin treatment. Using a cohort of 258 T2DM patients who had new metformin exposure, existing genetic data, and longitudinal electronic health records, we compared genetic variation within FMO5 to change in glycemic response. Gene-level and SNP-level analysis identified marginally significant associations for FMO5 variation, representing an EHR-driven pharmacogenetics hypothesis for a potential novel mechanism for metformin biotransformation. However, functional validation of this EHR-based hypothesis is necessary to ascertain its clinical and biological significance.

Studying the Confounding Effects of Socio-Ecological Conditions in Retrospective Clinical Research: A Use Case of Social Stress.

Socio-ecological Conditions (SECs) are important to include in clinical research models as they have been known to impact the health of patients. However, current clinical research models account for these factors only in an unsatisfyingly rudimentary way. In this study, we developed an SEC Index that captured the latent and direct effects of social stress, one of the many kinds of SEC, on patients' general health as measured by the Charlson Comorbidity Index. We demonstrated that the above SEC Index had a significant effect in a clinical model, a patient-level model with the specific clinical outcome of breast cancer prevalence. Further, we demonstrated that including the SEC Index of social stress into the clinical models significantly increased their performance. Our study demonstrated a viable approach that is interchangeable to include any SEC of interest, to more appropriately account for SECs in clinical research models.

Mining Biomedical Literature to Explore Interactions between Cancer Drugs and Dietary Supplements.

Interactions between cancer drugs and dietary supplements are clinically important and have not been extensively investigated through mining of the biomedical literature. We report on a previously introduced method now enhanced by machine learning-based filtering. Potential interactions are extracted by using relationships in the form of semantic predications. Semantic predications stored in SemMedDB, a database of structured knowledge generated from MEDLINE, were filtered and connected by two interaction pathways to explore potential drug-supplement interactions (DSIs). The lasso regression filter was trained by using SemRep output features in an expert annotated corpus and used to rank retrieved predications by predicted precision. We found not only known interactions but also inferred several unknown potential DSIs by appropriate filtering and linking of semantic predications.

Automated Detection of Postoperative Surgical Site Infections Using Supervised Methods with Electronic Health Record Data.

The National Surgical Quality Improvement Project (NSQIP) is widely recognized as "the best in the nation" surgical quality improvement resource in the United States. In particular, it rigorously defines postoperative morbidity outcomes, including surgical adverse events occurring within 30 days of surgery. Due to its manual yet expensive construction process, the NSQIP registry is of exceptionally high quality, but its high cost remains a significant bottleneck to NSQIP's wider dissemination. In this work, we propose an automated surgical adverse events detection tool, aimed at accelerating the process of extracting postoperative outcomes from medical charts. As a prototype system, we combined local EHR data with the NSQIP gold standard outcomes and developed machine learned models to retrospectively detect Surgical Site Infections (SSI), a particular family of adverse events that NSQIP extracts. The built models have high specificity (from 0.788 to 0.988) as well as very high negative predictive values (>0.98), reliably eliminating the vast majority of patients without SSI, thereby significantly reducing the NSQIP extractors' burden.

Using EHR-Linked Biobank Data to Study Metformin Pharmacogenomics.

Metformin is a commonly prescribed diabetes medication whose mechanism of action is poorly understood. In this study we utilized EHR-linked biobank data to elucidate the impact of genomic variation on glycemic response to metformin. Our study found significant gene- and SNP-level associations within the beta-2 subunit of the heterotrimeric adenosine monophosphate-activated protein kinase complex. Using EHR phenotypes where were able to add additional clarity to ongoing metformin pharmacogenomic dialogue.