Incorporation of quantitative imaging data using artificial intelligence improves risk prediction in veterans with liver disease

BACKGROUND AND AIMS

Utilization of electronic health records data to derive predictive indexes such as the electronic Child-Turcotte-Pugh (eCTP) Score can have significant utility in health care delivery. Within the records, CT scans contain phenotypic data which have significant prognostic value. However, data extractions have not traditionally been applied to imaging data. In this study, we used artificial intelligence to automate biomarker extraction from CT scans and examined the value of these features in improving risk prediction in patients with liver disease.

APPROACH AND RESULTS

Using a regional liver disease cohort from the Veterans Health System, we retrieved administrative, laboratory, and clinical data for Veterans who had CT scans performed for any clinical indication between 2008 and 2014. Imaging biomarkers were automatically derived using the analytic morphomics platform. In all, 4614 patients were included. We found that the eCTP Score had a Concordance index of 0.64 for the prediction of overall mortality while the imaging-based model alone or with eCTP Score performed significantly better [Concordance index of 0.72 and 0.73 ( p <0.001)]. For the subset of patients without hepatic decompensation at baseline (n=4452), the Concordance index for predicting future decompensation was 0.67, 0.79, and 0.80 for eCTP Score, imaging alone, or combined, respectively.

CONCLUSIONS

This proof of concept demonstrates that the potential of utilizing automated extraction of imaging features within CT scans either alone or in conjunction with classic health data can improve risk prediction in patients with chronic liver disease.

Different gut microbiota in U.S. formula-fed infants consuming a meat vs. dairy-based complementary foods: A randomized controlled trial

Objective

This project aimed to evaluate the impact of meat- vs. dairy-based complementary foods on gut microbiota and whether it relates to growth.

Design

Full-term, formula-fed infants were recruited from the metro Denver area (Colorado, US) and randomized to a meat- or dairy-based complementary diet from 5 to 12 months of age. Infant's length and weight were measured, and stool samples were collected at 5, 10, and 12 months for 16S rRNA gene sequencing and short-chain fatty acids (SCFAs) quantification.

Results

Sixty-four infants completed the dietary intervention (n = 32/group). Weight-for-age Z (WAZ) scores increased in both groups and length-for-age Z scores (LAZ) increased in the meat group only, which led to a significant group-by-time interaction (P = 0.02) of weight-for-length Z (WLZ) score. Microbiota composition (Beta-diversity) differed between groups at 12 months (weighted PERMANOVA P = 0.01) and had a group-by-time interaction of P = 0.09. Microbial community richness (Chao1) increased in the meat group only. Genus Akkermansia had a significant group-by-time interaction and increased in the dairy group and decreased in the meat group. A significant fold change of butyric acid from 5 to 12 months was found in the meat group (+1.75, P = 0.011) but not in the dairy group. Regression analysis showed that Chao1 had a negative association with WLZ and WAZ. Several genera also had significant associations with all growth Z scores.

Conclusion

Complementary feeding not only impacts infant growth but also affects gut microbiota maturation. Complementary food choices can affect both the gut microbiota diversity and structures and these changes in gut microbiota are associated with infant growth.

Variation in Provider Connectedness Associates With Outcomes of Inflammatory Bowel Diseases in an Analysis of Data From a National Health System

BACKGROUND & AIMS

Inflammatory bowel diseases (IBD) often require multidisciplinary care with tight coordination among providers. Provider connectedness, a measure of the relationship among providers, is an important aspect of care coordination that has been linked to higher quality care. We aimed to assess variation in provider connectedness among medical centers, and to understand the association between this established measure of care coordination and outcomes of patients with IBD.

METHODS

We conducted a national cohort study of 32,949 IBD patients with IBD from 2005 to 2014. We used network analysis to examine provider connectedness, defined using network properties that measure the strength of the collaborative relationship, team cohesiveness, and between-facility collaborations. We used multilevel modeling to examine variations in provider connectedness and association with patient outcomes.

RESULTS

There was wide variation in provider connectedness among facilities in complexity, rural designation, and volume of patients with IBD. In a multivariable model, patients followed in a facility with team cohesiveness (odds ratio, 0.38; 95% CI, 0.16-0.88) and where providers often collaborated with providers outside their facility (odds ratio, 0.48; 95% CI, 0.31-0.75) were less likely to have clinically active disease, defined by a composite of outpatient flare, inpatient flare, and IBD-related surgery.

CONCLUSIONS

A national study found evidence for heterogeneity in patient-sharing among IBD care teams. Patients with IBD seen at health centers with higher provider connectedness appear to have better outcomes. Understanding provider connectedness is a step toward designing network-based interventions to improve coordination and quality of care.

Development and Validation of an Administrative Codes Algorithm to Identify Abdominal Surgery and Bowel Obstruction in Patients With Inflammatory Bowel Disease

Background

Validated administrative codes (CPT and ICD) can permit the use of large databases to study diseases and outcomes. The aim of this study was to validate administrative codes for surgery and obstructive complications in patients with inflammatory bowel disease (IBD).

Methods

We performed a retrospective study of IBD patients within the Veterans Affairs Health Administration (VA) from 2000 to 2015 with administrative codes for bowel surgery and complications validated by chart review. Positive predictive values (PPVs) and negative predictive value (NPV) were calculated.

Results

The PPV for bowel surgery was 96.4%; PPV of obstruction codes for bowel obstruction was 80.5% (95% confidence interval: 69.1%, 89.2%).

Conclusions

CPT and ICD codes for abdominal surgery and obstructive complications can be accurately utilized in IBD patients in VA.

Predicting 30-day hospital readmissions using artificial neural networks with medical code embedding

Reducing unplanned readmissions is a major focus of current hospital quality efforts. In order to avoid unfair penalization, administrators and policymakers use prediction models to adjust for the performance of hospitals from healthcare claims data. Regression-based models are a commonly utilized method for such risk-standardization across hospitals; however, these models often suffer in accuracy. In this study we, compare four prediction models for unplanned patient readmission for patients hospitalized with acute myocardial infarction (AMI), congestive health failure (HF), and pneumonia (PNA) within the Nationwide Readmissions Database in 2014. We evaluated hierarchical logistic regression and compared its performance with gradient boosting and two models that utilize artificial neural networks. We show that unsupervised Global Vector for Word Representations embedding representations of administrative claims data combined with artificial neural network classification models improves prediction of 30-day readmission. Our best models increased the AUC for prediction of 30-day readmissions from 0.68 to 0.72 for AMI, 0.60 to 0.64 for HF, and 0.63 to 0.68 for PNA compared to hierarchical logistic regression. Furthermore, risk-standardized hospital readmission rates calculated from our artificial neural network model that employed embeddings led to reclassification of approximately 10% of hospitals across categories of hospital performance. This finding suggests that prediction models that incorporate new methods classify hospitals differently than traditional regression-based approaches and that their role in assessing hospital performance warrants further investigation.

Opioid and benzodiazepine prescription among patients with cirrhosis compared to other forms of chronic disease

Objective

Data on patterns and correlates of opioid and benzodiazepines prescriptions among patients with chronic conditions are limited. Given a diminished capacity for hepatic clearance, patients with cirrhosis represent a high risk group for use. The aim of this study was to characterise the patterns and correlates of prescription opioid, benzodiazepine and dual drug prescriptions among individuals with common chronic diseases.

Design

Analysis of Truven Marketscan database to evaluate individuals with drug coverage with cirrhosis (n=169,181), chronic hepatitis C without cirrhosis (n=210 191), congestive heart failure (n=766 840) or chronic obstructive pulmonary disease (n=1 438 798). Pharmacy files were examined for outpatient prescriptions.

Results

Patients with cirrhosis had a significantly higher prevalence of opioid prescriptions (37.1 per 100 person-years vs 24.3–26.0, p≤0.001) and benzodiazepine prescriptions (21.3 per 100 person-years vs 12.1–12.9, p<0.001). High dose opioid prescription (>90 daily oral morphine equivalents) (29.1% vs 14.4%, p<0.001) and dual opioid and benzodiazepine prescription (17.5% vs 9.6%–10.5 %, p<0.001) were also significantly more prevalent in cirrhosis. High dose opioid prescription was greater in men, individuals ages 40–59, in the Western USA, and among those with a mental health or substance abuse condition. Dual opioid and benzodiazepine prescription were highest among those with alcoholic cirrhosis and middle aged-adults.

Conclusion

Persons with cirrhosis have significantly higher rates of prescription opioid and benzodiazepine prescription compared to others with chronic diseases despite their high risk for adverse drug reactions. Demographics and mental health or substance abuse history can help identify high risk groups to target interventions.

Abstract 225: Determinants of Compensation for US Academic Physicians: Does Gender Matter?

Introduction: Academic physician compensation remains opaque, despite its critical importance for negotiating equitable salaries. In particular, prior studies have suggested inequity between male and female physicians. We set out to broadly define determinants of compensation in US academic physicians focusing on gender.

Methods: We performed an online pilot survey using a convenience sample of US academic physicians in clinical practice recruited through social media. Survey questions focused on demographic information, practice environment, job requirements, compensation, and overall satisfaction. Respondents failing to provide information on gender (n=2), salary (n=43) or key compensation questions (n=7) were excluded. Predictors of total salary, defined as the sum of base salary plus bonuses, and satisfaction with position (range 0-100) were explored in a generalized regression model.

Results: 252 respondents met inclusion criteria for analysis of total salary. Of the respondents, 32.9% were female, 59.9% Caucasian, 35.7% were associate professors or professors, 54.8% were employed in an urban setting, 44.9% performed major procedures, and 72.8% practiced in internal medicine or an internal medicine subspecialty. Median total salary was $270,000 (Interquartile range 208,000-348,500). After multivariable adjustment, female physicians made $44,512.55 less in total salary. Performing major procedures (p<0.001), living in a major urban area (p=0.034), and being a professor or associate professor (p<0.001) also predicted greater total salary (Table). Age and years since clinical training ended were significant predictors of total salary in a univariable but not a multivariable model. There was no significant interaction between gender and rank or between gender and performing major procedures. Median job satisfaction in 241 respondents was 77.0 (Interquartile range 64.0-86.0). In a multivariable model, only years since clinical training ended but not total salary or other covariates predicted physician satisfaction.

Conclusions: This pilot online study suggests gender disparities exist with respect to US academic physician compensation. Further studies are needed to better define these disparities and to enhance transparency with respect to physician compensation.

Variation in Laboratory Test Naming Conventions in EHRs Within and Between Hospitals: A Nationwide Longitudinal Study

BACKGROUND

Electronic health records provide clinically rich data for research and quality improvement work. However, the data are often unstructured text, may be inconsistently recorded and extracted into centralized databases, making them difficult to use for research.

OBJECTIVES

We sought to quantify the variation in how key laboratory measures are recorded in the Department of Veterans Affairs (VA) Corporate Data Warehouse (CDW) across hospitals and over time. We included 6 laboratory tests commonly drawn within the first 24 hours of hospital admission (albumin, bilirubin, creatinine, hemoglobin, sodium, white blood cell count) from fiscal years 2005-2015.

RESULTS

We assessed laboratory test capture for 5,454,411 acute hospital admissions at 121 sites across the VA. The mapping of standardized laboratory nomenclature (Logical Observation Identifiers Names and Codes, LOINCs) to test results in CDW varied within hospital by laboratory test. The relationship between LOINCs and laboratory test names improved over time; by FY2015, 109 (95.6%) hospitals had >90% of the 6 laboratory tests mapped to an appropriate LOINC. All fields used to classify test results are provided in an Appendix (Supplemental Digital Content 1, http://links.lww.com/MLR/B635).

CONCLUSIONS

The use of electronic health record data for research requires assessing data consistency and quality. Using laboratory test results requires the use of both unstructured text fields and the identification of appropriate LOINCs. When using data from multiple facilities, the results should be carefully examined by facility and over time to maximize the capture of data fields.

Development and Internal Validation of a Model for Early Detection of Hepatocellular Carcinoma in Patients With Cirrhosis

BACKGROUND

Surveillance for hepatocellular carcinoma (HCC) is recommended in patients with cirrhosis; however, early detection efforts are limited by suboptimal effectiveness.

AIM

To derive and validate a model to accurately distinguish cirrhotic patients with and without HCC and compare the accuracy of the model to that of α-fetoprotein (AFP) alone.

METHODS

We conducted a case-control study of cirrhotic patients with and without HCC seen at a large urban hospital system between January 2005 and June 2012. We derived multivariate logistic regression models for the presence of HCC and early-stage HCC. Discriminatory power was evaluated using receiver operating characteristic curve analysis in derivation and validation cohorts using a 10-fold cross-validation approach.

RESULTS

We identified 1356 patients with cirrhosis, with (n=455, 147 early stage) and without (n=901) HCC. We found that AFP>20 ng/mL and FIB-4, a noninvasive marker of fibrosis, were significantly associated with the presence of HCC (OR=10.5; 95% CI, 7.9-13.9 and OR=1.05; 95% CI, 1.03-1.07, respectively) and early-stage HCC (OR=4.4; 95% CI, 2.9-6.5 and OR=1.06; 95% CI, 1.03-1.09, respectively). Models incorporating AFP and FIB-4 had good discriminatory power, with c-statistics of approximately 0.80, in both derivation and validation cohorts. The model for early-stage HCC had higher discriminatory power than AFP alone (c-statistic 0.73; 95% CI, 0.69-0.78) in derivation and validation cohorts (P=0.02 and 0.15, respectively).

CONCLUSIONS

Models including AFP and FIB-4 can accurately discriminate cirrhotic patients with early-stage HCC from those without HCC.

Threefold increased risk of infertility: a meta-analysis of infertility after ileal pouch anal anastomosis in ulcerative colitis

BACKGROUND

Increased infertility in women has been reported after ileal pouch-anal anastomosis (IPAA) for ulcerative colitis but reported infertility rates vary substantially.

AIMS

(1) To perform a systematic review and meta-analysis of the relative risk of infertility post-IPAA compared with medical management; (2) to estimate the rate of infertility post-IPAA; and (3) to identify modifiable risk factors which contribute to infertility.

METHODS

Medline, EMBASE, Current Contents, meeting abstracts, and bibliographies were searched independently by two investigators. The titles and abstracts of 189 potentially relevant studies were reviewed; eight met the criteria and all data were extracted independently. Consensus was achieved on each data point, and fixed effects meta-analyses, a funnel plot, and sensitivity analyses were performed.

RESULTS

The initial meta-analysis of eight studies had significant heterogeneity (p = 0.004) due to one study with very high preoperative infertility (38%). When this study was omitted, the relative risk of infertility after IPAA was 3.17 (2.41-4.18), with non-significant heterogeneity. The weighted average infertility rate in medically treated ulcerative colitis was 15% for all seven studies, and the weighted average infertility rate was 48% after IPAA (50% if all eight studies are included). We were unable to identify any procedural factors that consistently affected the risk of infertility.

CONCLUSIONS

IPAA increases the risk of infertility in women with ulcerative colitis by approximately threefold. Infertility, defined as achieving pregnancy in 12 months of attempting conception, increased from 15% to 48% in women post-IPAA for ulcerative colitis. This provides a basis for counselling patients considering colectomy with IPAA. Further studies of modifiable risk factors are needed.