Use of electronic health record data to identify skin and soft tissue infections in primary care settings: a validation study

Automating detection of diagnostic error of infectious diseases using machine learning

Diagnostic error, a cause of substantial morbidity and mortality, is largely discovered and evaluated through self-report and manual review, which is costly and not suitable to real-time intervention. Opportunities exist to leverage electronic health record data for automated detection of potential misdiagnosis, executed at scale and generalized across diseases. We propose a novel automated approach to identifying diagnostic divergence considering both diagnosis and risk of mortality. Our objective was to identify cases of emergency department infectious disease misdiagnoses by measuring the deviation between predicted diagnosis and documented diagnosis, weighted by mortality. Two machine learning models were trained for prediction of infectious disease and mortality using the first 24h of data. Charts were manually reviewed by clinicians to determine whether there could have been a more correct or timely diagnosis. The proposed approach was validated against manual reviews and compared using the Spearman rank correlation. We analyzed 6.5 million ED visits and over 700 million associated clinical features from over one hundred emergency departments. The testing set performances of the infectious disease (Macro F1 = 86.7, AUROC 90.6 to 94.7) and mortality model (Macro F1 = 97.6, AUROC 89.1 to 89.1) were in expected ranges. Human reviews and the proposed automated metric demonstrated positive correlations ranging from 0.231 to 0.358. The proposed approach for diagnostic deviation shows promise as a potential tool for clinicians to find diagnostic errors. Given the vast number of clinical features used in this analysis, further improvements likely need to either take greater account of data structure (what occurs before when) or involve natural language processing. Further work is needed to explain the potential reasons for divergence and to refine and validate the approach for implementation in real-world settings.

Space-Time Trends of Community Onset Staphylococcus aureus Infections in Children: A Group Based Trajectory Modeling Approach

PURPOSE

Staphylococcus aureus (S. aureus) remains a serious cause of infections in the U.S. and worldwide. In the U.S., methicillin resistant S. aureus (MRSA) is the leading cause of skin and soft tissue infections. This study identifies 'best' to 'worst' infection trends from 2002 to 2016, using group-based trajectory modelling approach.

METHODS

Electronic health records of children living in the southeastern U.S. with S. aureus infections from 2002-2016 were retrospectively studied, by applying a group-based trajectory model to estimate infection trends (low, high, very high), and then assess spatial significance of these trends at the census tract level; we focused on community onset (CO) infections and not those considered healthcare acquired.

RESULTS

Three methicillin sensitive (MSSA) infection trends (low, high, very high) and three MRSA trends (low, high, very high) were identified from 2002-2016. Among census tracts with community onset (CO) S. aureus cases, 29% of tracts belonged to the best trend (low infection) for both MRSA and MSSA; higher proportions occurring in the less densely populated areas. Race disparities were seen with the worst MRSA infection trends and were more often in urban areas.

CONCLUSIONS

Group based trajectory modeling identified unique trends of S. aureus infection rates over time and space, giving insight into the associated population characteristics which reflect these trends of community onset infection.

CT utilization in evaluation of skin and soft tissue extremity infections in the ED: Retrospective cohort study

OBJECTIVE

Skin and soft tissue infections (SSTI) are commonly diagnosed in the emergency department (ED). While most SSTI are diagnosed with patient history and physical exam alone, ED clinicians may order CT imaging when they suspect more serious or complicated infections. Patients who inject drugs are thought to be at higher risk for complications from SSTI and may undergo CT imaging more frequently. The objective of this study is to characterize CT utilization when evaluating for SSTI in ED patients particularly in patients with intravenous drug use (IVDU), the frequency of significant and actionable findings from CT imaging, and its impact on subsequent management and ED operations.

METHODS

We performed a retrospective analysis of encounters involving a diagnosis of SSTI in seven EDs across an integrated health system between October 2019 and October 2021. Descriptive statistics were used to assess overall trends, compare CT utilization frequencies, actionable imaging findings, and surgical intervention between patients who inject drugs and those who do not. Multivariable logistic regression was used to analyze patient factors associated with higher likelihood of CT imaging.

RESULTS

There were 4833 ED encounters with an ICD-10 diagnosis of SSTI during the study period, of which 6% involved a documented history of IVDU and 30% resulted in admission. 7% (315/4833) of patients received CT imaging, and 22% (70/315) of CTs demonstrated evidence of possible deep space or necrotizing infections. Patients with history of IVDU were more likely than patients without IVDU to receive a CT scan (18% vs 6%), have a CT scan with findings suspicious for deep-space or necrotizing infection (4% vs 1%), and undergo surgical drainage in the operating room within 48 h of arrival (5% vs 2%). Male sex, abnormal vital signs, and history of IVDU were each associated with higher likelihood of CT utilization. Encounters involving CT scans had longer median times to ED disposition than those without CT scans, regardless of whether these encounters resulted in admission (9.0 vs 5.5 h), ED observation (5.5 vs 4.1 h), or discharge (6.8 vs 2.9 h).

DISCUSSION

ED clinicians ordered CT scans in 7% of encounters when evaluating for SSTI, most frequently in patients with abnormal vital signs or a history of IV drug use. Patients with a history of IVDU had higher rates of CT findings suspicious for deep space infections or necrotizing infections and higher rates of incision and drainage procedures in the OR. While CT scans significantly extended time spent in the ED for patients, this appeared justified by the high rate of actionable findings found on imaging, particularly for patients with a history of IVDU.

Consensus Current Procedural Terminology Code Definition of Source Control for Sepsis

INTRODUCTION

Unlike antibiotic and perfusion support, guidelines for sepsis source control lack high-quality evidence and are ungraded. Internally valid administrative data methods are needed to identify cases representing source control procedures to evaluate outcomes.

METHODS

Over five modified Delphi rounds, two independent reviewers identified Current Procedural Terminology (CPT) codes pertinent to source control. In each round, codes with perfect agreement were retained or excluded, whereas disagreements were reviewed by the panelists. Manual review of 400 patient records meeting Sepsis-3 criteria (2010-2017) clinically adjudicated which encounters included source control procedures (gold standard). The performance of consensus codes was compared with the gold standard to assess sensitivity, specificity, predictive values, and likelihood ratios.

RESULTS

Of 5752 CPT codes, 609 consensus codes represented source control procedures. Of 400 hospitalizations for sepsis, 39 (9.8%; 95% confidence interval [CI] 7.0%-13.1%) underwent gold standard source control procedures and 29 (7.3%; 95% CI 4.9-10.3%) consensus code-defined source control procedures. Thirty consensus codes were identified (20.0% gastrointestinal/intraabdominal, 10.0% genitourinary, 13.3% hepatopancreatobiliary, 23.3% orthopedic/cranial, 23.3% soft tissue, and 10.0% intrathoracic), which had 61.5% (95% CI 44.6%-76.6%) sensitivity, 98.6% (95% CI 96.8%-99.6%) specificity, 83.2% (95% CI 66.6%-92.4%) positive, and 95.9% (95% CI 93.9%-97.2%) negative predictive values. With pretest probability at sample prevalence, an identified consensus code had a posttest probability of 83.0% (95% CI 66.0%-92.0%), whereas consensus code absence had a probability of 4.0% (95% CI 3.0-6.0) for undergoing a source control procedure.

CONCLUSIONS

Using modified Delphi methodology, we created and validated CPT codes identifying source control procedures, providing a framework for evaluation of the surgical care of patients with sepsis.

Validation of algorithms for identifying outpatient infections in MS patients using electronic medical records

Background Our multiple sclerosis (MS) stakeholder groups expressed concerns about whether MS disease-modifying therapies (DMTs) increase the risk of specific outpatient infections. Validated methods for identifying the risk of these selected outpatient infections in the general population either do not exist, exclude the clinically important possibility of recurrent infections, or are inaccurate, largely because existing studies relied primarily on International Classification of Diseases (ICD) codes to identify infectious outcomes. Additionally, no studies have validated methods among the MS population, where some MS symptoms can be mistaken for infections (e.g., urinary tract infections (UTIs)). Objective To utilize multiple data elements in the electronic health record (EHR) to improve accurate identification of selected outpatient infections in an MS cohort and general population controls. Methods We searched Kaiser Permanente Southern California's EHR based on ICD-9/10 codes for specified outpatient infections from 1/1/2008-12/31/2018 among our MS cohort (n=6000) and 5:1 general population controls matched on age, sex, and race/ethnicity (n=30,010). Random sample chart abstractions from each group were used to identify common coding errors for outpatient pneumonia, upper and lower respiratory tract infection, UTIs, herpetic infections (herpes zoster (HZ), herpes simplex virus (HSV)), fungal infections, otitis media, cellulitis, and influenza. This information was used to define discrete infectious episodes and to identify the algorithm with the highest positive predictive value (PPV) after supplementing the ICD-coded episodes with radiology, laboratory and/or pharmacy data. Results PPVs relying on ICD codes alone were inaccurate, particularly for identifying recurrent herpetic infections (HZ (42%) and HSV (60%)), UTIs (42%) and outpatient pneumonia (20%) in MS patients. Defining and validating episodes improved the PPVs for all the selected infections. The final algorithms' PPVs were 80-100% in MS and 75-100% in the general population, after including dispensed treatments (UTI, herpetic infections and yeast vaginitis), timing of dispensed treatments (UTI, herpetic infections and yeast vaginitis), removal of prophylactic antiviral use (herpetic infections), and inclusion of selected laboratory (UTIs) and imaging results (pneumonia). The only exception was outpatient pneumonia, where PPVs improved but remained ≤70%. There were no significant differences in the PPVs for the final algorithms between the MS and general population. Conclusions Provided herein are accurate and validated algorithms that can be used to improve our understanding of how the risk of recurrent outpatient infections are influenced by MS treatments, MS-related disability, and co-morbidities. Findings from such studies will be important in helping patients and clinicians engage in shared decision-making and in developing strategies to mitigate risks of recurrent infections.

Emergency department utilization for impetigo among the pediatric population: A retrospective study of the national emergency department sample 2013-2015

BACKGROUND

Despite the large burden of impetigo in childhood and high frequency of pediatric emergency department (ED) visits for skin conditions, limited information exists on the use of EDs for impetigo among US children.

OBJECTIVE

Our study aimed to generate national estimates of ED utilization and to identify sociodemographic predictors of impetigo-related ED visits.

METHODS

This was a retrospective, cross-sectional study of children ages 1-17 presenting to EDs with a primary diagnosis of impetigo using years 2013-2015 of the Nationwide Emergency Department Sample.

RESULTS

Impetigo accounted for 163 909 of the 71 488, 511 pediatric ED visits and was the fourth most common presenting skin diagnosis. Controlling for sociodemographic factors, patients presenting to the ED with impetigo were most likely to be 6-11 years old, male, and from lower-income quartiles. Patients were most likely to be uninsured and most likely to present on weekends in the summer.

CONCLUSION

This study provided national-level estimates of ED use for impetigo among US children. Ultimately, the identification of factors associated with increased ED utilization may help in developing targeted interventions to reduce the use of emergency care for impetigo.

Applying Machine Learning Across Sites: External Validation of a Surgical Site Infection Detection Algorithm

BACKGROUND

Surgical complications have tremendous consequences and costs. Complication detection is important for quality improvement, but traditional manual chart review is burdensome. Automated mechanisms are needed to make this more efficient. To understand the generalizability of a machine learning algorithm between sites, automated surgical site infection (SSI) detection algorithms developed at one center were tested at another distinct center.

STUDY DESIGN

NSQIP patients had electronic health record (EHR) data extracted at one center (University of Minnesota Medical Center, Site A) over a 4-year period for model development and internal validation, and at a second center (University of California San Francisco, Site B) over a subsequent 2-year period for external validation. Models for automated NSQIP SSI detection of superficial, organ space, and total SSI within 30 days postoperatively were validated using area under the curve (AUC) scores and corresponding 95% confidence intervals.

RESULTS

For the 8,883 patients (Site A) and 1,473 patients (Site B), AUC scores were not statistically different for any outcome including superficial (external 0.804, internal [0.784, 0.874] AUC); organ/space (external 0.905, internal [0.867, 0.941] AUC); and total (external 0.855, internal [0.854, 0.908] AUC) SSI. False negative rates decreased with increasing case review volume and would be amenable to a strategy in which cases with low predicted probabilities of SSI could be excluded from chart review.

CONCLUSIONS

Our findings demonstrated that SSI detection machine learning algorithms developed at 1 site were generalizable to another institution. SSI detection models are practically applicable to accelerate and focus chart review.

Uncontrolled diabetes mellitus increases risk of infection in patients with advanced cirrhosis

BACKGROUND

Diabetes mellitus (DM) is common in patients with cirrhosis and is associated with increased risk of infection.

AIM

To analyze the impact of uncontrolled DM on infection and mortality among inpatients with advanced cirrhosis.

METHODS

This study utilized the Nationwide Inpatient Sample from 1998 to 2014. We defined advanced cirrhosis using a validated ICD-9-CM algorithm requiring a diagnosis of cirrhosis and clinically significant portal hypertension or decompensation. The primary outcome was bacterial infection. Secondary outcomes included inpatient mortality stratified by elderly age (age≥70). Multivariable logistic regression analyzed outcomes.

RESULTS

906,559 (29.2%) patients had DM and 109,694 (12.1%) were uncontrolled. Patients who had uncontrolled DM were younger, had less ascites, but more encephalopathy. Bacterial infection prevalence was more common in uncontrolled DM (34.2% vs. 28.4%, OR 1.33, 95% CI 1.29-1.37, p<0.001). Although uncontrolled DM was not associated with mortality, when stratified by age, elderly patients with uncontrolled DM had a significantly higher risk of inpatient mortality (OR 1.62, 95% CI 1.46-1.81).

CONCLUSIONS

Uncontrolled DM is associated with increased risk of infection, and when combined with elderly age is associated with increased risk of inpatient mortality. Glycemic control is a modifiable target to improve morbidity and mortality in patients with advanced cirrhosis.

National Public Health Burden Estimates of Endocarditis and Skin and Soft-Tissue Infections Related to Injection Drug Use: A Review

BACKGROUND

Despite concerns about the burden of the bacterial and fungal infection syndromes related to injection drug use (IDU), robust estimates of the public health burden of these conditions are lacking. The current article reviews and compares data sources and national burden estimates for infective endocarditis (IE) and skin and soft-tissue infections related to IDU in the United States.

METHODS

A literature review was conducted for estimates of skin and soft-tissue infection and endocarditis disease burden with related IDU or substance use disorder terms since 2011. A range of the burden is presented, based on different methods of obtaining national projections from available data sources or published data.

RESULTS

Estimates using available data suggest the number of hospital admissions for IE related to IDU ranged from 2900 admissions in 2013 to more than 20 000 in 2017. The only source of data available to estimate the annual number of hospitalizations and emergency department visits for skin and soft-tissue infections related to IDU yielded a crude estimate of 98 000 such visits. Including people who are not hospitalized, a crude calculation suggests that 155 000-540 000 skin infections related to IDU occur annually.

DISCUSSION

These estimates carry significant limitations. However, regardless of the source or method, the burden of disease appears substantial, with estimates of thousands of episodes of IE among persons with IDU and at least 100 000 persons who inject drugs (PWID) with skin and soft-tissue infections annually in the United States. Given the importance of these types of infections, more robust and reliable estimates are needed to better quantitate the occurrence and understand the impact of interventions.

Skin and Soft Tissue Infection in People Living With Human Immunodeficiency Virus in a Large, Urban, Public Healthcare System in Houston, Texas, 2009-2014

BACKGROUND

Skin and soft tissue infections (SSTIs) disproportionately impact patients with human immunodeficiency virus (HIV). Recent declines in the incidence of SSTIs have been noted in the non-HIV population. We sought to study the epidemiology and microbiology of SSTIs in a population of 8597 patients followed for HIV primary care in a large, urban county system from January 2009 to December 2014.

METHODS

SSTIs were identified from the electronic medical record by use of International Classification of Diseases-9 billing codes. Charts were reviewed to confirm each patient's diagnosis of acute SSTI and abstract culture and susceptibility data. We calculated the yearly SSTI incidences using Poisson regression with clustering by patient.

RESULTS

There were 2202 SSTIs identified. Of 503 (22.8%) cultured SSTIs, 332 (66.0%) recovered Staphylococcus aureus as a pathogen, of which 287/332 (86.4%) featured S. aureus as the sole isolated organism. Among the S. aureus isolates that exhibited antibiotic susceptibilities, 231/331 (69.8%) were methicillin resistant, and the proportion did not change by year. The observed incidence of SSTI was 78.0 per 1000 person-years (95% confidence interval 72.9-83.4) and declined from 96.0 infections per 1000 person-years in 2009 to 56.5 infections per 1000 person-years in 2014 (P < .001). Other significant predictors of SSTI incidences in both univariate as well as multivariate analyses included a low CD4 count, high viral load, and not being a Spanish-speaking Hispanic.

CONCLUSIONS

SSTIs remain a significant problem in the outpatients living with HIV, although rates of SSTIs appear to have declined by approximately 40% between 2009 and 2014.

Association of Infections and Use of Fluoroquinolones With the Risk of Aortic Aneurysm or Aortic Dissection

IMPORTANCE

Prior observational studies have suggested that fluoroquinolone use may be associated with more than 2-fold increased risk of aortic aneurysm or aortic dissection (AA/AD). These studies, however, did not fully consider the role of coexisting infections and the risk of fluoroquinolones relative to other antibiotics.

OBJECTIVE

To estimate the risk of AA/AD associated with infections and to assess the comparative risk of AA/AD associated with fluoroquinolones vs other antibiotics with similar indication profiles among patients with the same types of infections.

DESIGNS, SETTINGS, AND PARTICIPANTS

This nested case-control study identified 21 651 176 adult patients from a nationwide population-based health insurance claims database from January 1, 2009, to November 30, 2015. Each incident case of AA/AD was matched with 10 control individuals by age, sex, and follow-up duration in the database using risk-set sampling. Analysis of the data was conducted from April 2019 to March 2020.

EXPOSURES

Infections and antibiotic use within a 60-day risk window before the occurrence of AA/AD.

MAIN OUTCOMES AND MEASURES

Conditional logistic regression was used to estimate the odds ratios (ORs) and 95% CIs comparing infections for which fluoroquinolones are commonly used with no infection within a 60-day risk window before outcome occurrence, adjusting for baseline confounders and concomitant antibiotic use. The adjusted ORs comparing fluoroquinolones with antibiotics with similar indication profiles within patients with indicated infections were also estimated.

RESULTS

A total of 28 948 cases and 289 480 matched controls were included (71.37% male; mean [SD] age, 67.41 [15.03] years). Among these, the adjusted OR of AA/AD for any indicated infections was 1.73 (95% CI, 1.66-1.81). Septicemia (OR, 3.16; 95% CI, 2.63-3.78) and intra-abdominal infection (OR, 2.99; 95% CI, 2.45-3.65) had the highest increased risk. Fluoroquinolones were not associated with an increased AA/AD risk when compared with combined amoxicillin-clavulanate or combined ampicillin-sulbactam (OR, 1.01; 95% CI, 0.82-1.24) or with extended-spectrum cephalosporins (OR, 0.88; 95% CI, 0.70-1.11) among patients with indicated infections. The null findings for fluoroquinolone use remained robust in different subgroup and sensitivity analyses.

CONCLUSIONS AND RELEVANCE

These results highlight the importance of accounting for coexisting infections while examining the safety of antibiotics using real-world data; the findings suggest that concerns about AA/AD risk should not deter fluoroquinolone use for patients with indicated infections.

Infection as a Cardiovascular Trigger: Associations Between Different Organ System Infections and Cardiovascular Events

BACKGROUND

Infection may be an acute precipitant of cardiovascular events. However, the relationships between different types of infection and cardiovascular events are less known. Our objective was to determine if exposure to infections of different organ systems in different time periods increases risk of myocardial infarction or venous thromboembolism.

METHOD

We used case-crossover analysis with conditional logistic regression to estimate odds ratios (OR) for the association for each infection type during 3 case periods (30, 60, and 90 days prior to index event) compared with control periods (exactly 1 year before).

RESULTS

This study had a total number of index admissions of 338,021 individuals, of which 82,986 were female; the mean age for individuals with myocardial infarction and venous thromboembolism was 68.48 years and 62.33 years, respectively. With every infection type, there was an increased likelihood of venous thromboembolism. The greatest association was for skin infections, with an OR of 5.39 (95% confidence interval, 4.08- 7.12) within the 7-day window. The association between myocardial infarction and skin infections was of lesser magnitude, with an OR of 2.89 (confidence interval, 1.97-4.24) in the 7-day exposure period.

CONCLUSION

We found a gradient of decreasing magnitudes of association with longer time periods, across the majority of infection types and cardiovascular events. This warrants potential interventions to prevent infection or cardiovascular disease prophylaxis in individuals with infection.

Gender Differences Among Patients Hospitalized With Cirrhosis in the United States

BACKGROUND AND AIMS

Gender disparities exist in outcomes among patients with cirrhosis. We sought to evaluate the role of gender on hospital course and in-hospital outcomes in patients with cirrhosis to help better understand these disparities.

STUDY

We analyzed data from the National Inpatient Sample (NIS), years 2009 to 2013, to identify patients with any diagnosis of cirrhosis. We calculated demographic and clinical characteristics by gender, as well as cirrhosis complications. Our primary outcome was inpatient mortality. We used logistic regression to associate baseline characteristics and cirrhosis complications with inpatient mortality.

RESULTS

Our cohort included 553,017 patients with cirrhosis admitted from 2009 to 2013. Women made up 39% of the cohort; median age was 57 with 66% non-Hispanic white. Women were more likely than men to have noncirrhosis comorbidities, including diabetes and hypertension but were less likely to have most cirrhosis complications, including ascites and variceal bleeding. Women were more likely than men to have acute bacterial infections (34.9% vs. 28.2%; P<0.001), and were less likely than men to die in the hospital on univariable (odds ratio, 0.88; 95% confidence interval, 0.86-0.90; P<0.001) and multivariable (odds ratio, 0.86; 95% confidence interval, 0.83-0.88; P<0.001) analysis.

CONCLUSIONS

In patients hospitalized with cirrhosis, women have lower rates of hepatic decompensating events and higher rates of nonhepatic comorbidities and infections, resulting in lower in-hospital mortality. Understanding differences in indications for and disposition following hospitalization may help with the development of gender-specific cirrhosis management programs to improve long-term outcomes in women and men living with cirrhosis.

Identification of validated case definitions for medical conditions used in primary care electronic medical record databases: a systematic review

Objectives

Data derived from primary care electronic medical records (EMRs) are being used for research and surveillance. Case definitions are required to identify patients with specific conditions in EMR data with a degree of accuracy. The purpose of this study is to identify and provide a summary of case definitions that have been validated in primary care EMR data.

Materials and Methods

We searched MEDLINE and Embase (from inception to June 2016) to identify studies that describe case definitions for clinical conditions in EMR data and report on the performance metrics of these definitions.

Results

We identified 40 studies reporting on case definitions for 47 unique clinical conditions. The studies used combinations of International Classification of Disease version 9 (ICD-9) codes, Read codes, laboratory values, and medications in their algorithms. The most common validation metric reported was positive predictive value, with inconsistent reporting of sensitivity and specificity.

Discussion

This review describes validated case definitions derived in primary care EMR data, which can be used to understand disease patterns and prevalence among primary care populations. Limitations include incomplete reporting of performance metrics and uncertainty regarding performance of case definitions across different EMR databases and countries.

Conclusion

Our review found a significant number of validated case definitions with good performance for use in primary care EMR data. These could be applied to other EMR databases in similar contexts and may enable better disease surveillance when using clinical EMR data. Consistent reporting across validation studies using EMR data would facilitate comparison across studies.

Systematic review registration

PROSPERO CRD42016040020 (submitted June 8, 2016, and last revised June 14, 2016).

Infection as a Stroke Trigger

Background and Purpose- The relationships between different infection types and stroke subtype are not well-characterized. We examined exposure to infections in different organ systems in different time periods before the acute ischemic stroke, intracerebral hemorrhage, and subarachnoid hemorrhage. Methods- We used the New York State Inpatient Databases and Emergency Department Databases (2006-2013). Validated International Classification of Diseases, Ninth Edition definitions identified index hospitalizations for acute ischemic stroke, intracerebral hemorrhage, and subarachnoid hemorrhage, and emergency department visits and hospitalizations for infection (skin, urinary tract infection, septicemia, abdominal, and respiratory). We used case cross-over analysis with conditional logistic regression to estimate odds ratios (OR) for the association between each infection type during case periods compared with control periods 1 year before. Results- Every infection type was associated with an increased likelihood of acute ischemic stroke. The greatest association was for urinary tract infection, with OR of 5.32 (95% CI, 3.69-7.68) within the 7-day window. The magnitude of association between urinary tract infection and intracerebral hemorrhage was of lesser magnitude, with OR of 1.80 (1.04-3.11) in the 14-day exposure period and OR of 1.54 (1.23-1.94) in the 120-day exposure period. Only respiratory infection was associated with subarachnoid hemorrhage, with OR of 3.67 (1.49-9.04) in the 14-day window and 1.95 (1.44-2.64) in the 120-day window. Conclusions- All infection types were associated with subsequent acute ischemic stroke, with the greatest association for urinary tract infection.

Conditional validation sampling for consistent risk estimation with binary outcome data subject to misclassification

PURPOSE

Misclassification of a binary outcome can introduce bias in estimation of the odds-ratio associated with an exposure of interest in pharmacoepidemiology research. It has been previously demonstrated that utilizing information from an internal randomly selected validation sample can help mitigate this bias.

METHODS

Using a Monte Carlo simulation-based approach, we study the properties of misclassification bias-adjusted odds-ratio estimators in a contingency table setting. We consider two methods of internal validation sampling; namely, simple random sampling and sampling conditional on the original (possibly incorrect) outcome status. Additional simulation studies are conducted to investigate these sampling approaches in a multi-table setting.

RESULTS

We demonstrate that conditional validation sampling, across a range of subsampling fractions, can produce better estimates than those based on an unconditional simple random sample. This approach allows for greater flexibility in the chosen categorical composition of the validation data, as well as the potential for obtaining a more efficient estimator of the odds-ratio. We further demonstrate that this relationship holds for the Mantel-Haenszel misclassification bias-adjusted odds-ratio in stratified samples. Recommendations for the choice of validation subsampling fraction are also provided.

CONCLUSIONS

Careful consideration when choosing the sampling scheme used to draw internal validation samples can improve the properties of the outcome misclassification bias-adjusted odds-ratio estimator in a (multiple) contingency table.

A validation sampling approach for consistent estimation of adverse drug reaction risk with misclassified right-censored survival data

Patient electronic health records, viewed as continuous-time right-censored survival data, can be used to estimate adverse drug reaction risk. Temporal outcome misclassification may occur as a result of errors in follow-up. These errors can be due to a failure to observe the incidence time of the adverse event of interest (due to misdiagnosis or nonreporting, etc) or an actual misdiagnosis of a competing adverse event. As the misclassifying event is often unobservable in the original data, we apply an internal validation sampling approach to produce consistent estimation in the presence of such errors. We introduce a univariate survival model and a cause-specific hazards model in which misclassification may also manifest as a diagnosis of an alternate adverse health outcome other than that of interest. We develop a method of maximum likelihood estimation of the model parameters and establish consistency and asymptotic normality of the estimators using standard results. We also conduct simulation studies to numerically investigate the finite sample properties of these estimators and the impact of ignoring the misclassification error.

Development and validation of a heart failure with preserved ejection fraction cohort using electronic medical records

Background

Heart failure (HF) with preserved ejection fraction (HFpEF) comprises nearly half of prevalent HF, yet is challenging to curate in a large database of electronic medical records (EMR) since it requires both accurate HF diagnosis and left ventricular ejection fraction (EF) values to be consistently ≥50%.

Methods

We used the national Veterans Affairs EMR to curate a cohort of HFpEF patients from 2002 to 2014. EF values were extracted from clinical documents utilizing natural language processing and an iterative approach was used to refine the algorithm for verification of clinical HFpEF. The final algorithm utilized the following inclusion criteria: any International Classification of Diseases-9 (ICD-9) code of HF (428.xx); all recorded EF ≥50%; and either B-type natriuretic peptide (BNP) or aminoterminal pro-BNP (NT-proBNP) values recorded OR diuretic use within one month of diagnosis of HF. Validation of the algorithm was performed by 3 independent reviewers doing manual chart review of 100 HFpEF cases and 100 controls.

Results

We established a HFpEF cohort of 80,248 patients (out of a total 1,155,376 patients with the ICD-9 diagnosis of HF). Mean age was 72 years; 96% were males and 12% were African-Americans. Validation analysis of the HFpEF algorithm had a sensitivity of 88%, specificity of 96%, positive predictive value of 96%, and a negative predictive value of 87% to identify HFpEF cases.

Conclusion

We developed a sensitive, highly specific algorithm for detecting HFpEF in a large national database. This approach may be applicable to other large EMR databases to identify HFpEF patients.

Strategies for handling missing clinical data for automated surgical site infection detection from the electronic health record

Proper handling of missing data is important for many secondary uses of electronic health record (EHR) data. Data imputation methods can be used to handle missing data, but their use for analyzing EHR data is limited and specific efficacy for postoperative complication detection is unclear. Several data imputation methods were used to develop data models for automated detection of three types (i.e., superficial, deep, and organ space) of surgical site infection (SSI) and overall SSI using American College of Surgeons National Surgical Quality Improvement Project (NSQIP) Registry 30-day SSI occurrence data as a reference standard. Overall, models with missing data imputation almost always outperformed reference models without imputation that included only cases with complete data for detection of SSI overall achieving very good average area under the curve values. Missing data imputation appears to be an effective means for improving postoperative SSI detection using EHR clinical data.

Decreasing Incidence of Skin and Soft Tissue Infections With a Seasonal Pattern at an Academic Medical Center, 2006-2014

The incidence of SSTIs at University of Chicago Medicine decreased significantly in children and adults with seasonal variation, peaking during the summer months. This suggests a reversal of the massive increase in SSTI incidence after 2000 in the U.S.

Background. The incidence of skin and soft tissue infections (SSTIs) in the United States increased sharply after 2000 with the emergence of USA300 methicillin-resistant Staphylococcus aureus. We examined trends in SSTI incidence in 2006–2014 at the University of Chicago Medicine (UCM).

Methods. Data were obtained for patient encounters at UCM with an International Classification of Diseases, Ninth Revision-coded SSTI diagnosis between January 1, 2006 and March 31, 2014. Incidence density was calculated per 1000 encounters by quarter and year. Encounters were stratified by inpatient, outpatient clinic and emergency department (ED) encounters and by age group, gender, and race. Poisson regression was used to assess change over time.

Results. In 2006–2014, data were collected for 38 201 SSTI-associated encounters among 31 869 subjects. Among all patients treated at UCM, there was a decrease of 1% per year in the incidence of SSTIs during 2006–2013, with an overall decrease of 16%. There was a significant decrease in SSTI-related encounters among inpatients (rate ratio [RR] = 0.97; 95% confidence interval [CI], .96–.98), ED patients (RR = 0.98; 95% CI, .97–.98), adults (RR = 0.98; 95% CI, .97–.98), children (RR = 0.96; 95% CI, .95–.97), and African Americans (RR = 0.99; 95% CI, .98–.99). There was an annual seasonal trend, with the peak incidence occurring during the late summer.

Conclusions. The incidence of SSTIs at UCM decreased in children and adults with seasonal variation, peaking during the summer months. This suggests a reversal of the massive increase in SSTI incidence in the United States after 2000.

Obesity is independently associated with infection in hospitalised patients with end-stage liver disease

BACKGROUND

Infection is the most common cause of mortality in end-stage liver disease (ESLD). The impact of obesity on infection risk in ESLD is not established.

AIM

To characterise the impact of obesity on infection risk in ESLD.

METHODS

We evaluated the association between infection and obesity in patients with ESLD. Patients grouped as non-obese, obesity class I-II and obesity class III were studied using the Nationwide Inpatient Sample. Validated diagnostic code based algorithms were utilised to determine weight category and infections, including bacteraemia, skin/soft tissue infection, urinary tract infection (UTI), pneumonia/respiratory infection, Clostridium difficile infection (CDI) and spontaneous bacterial peritonitis (SBP). Risk factors for infection and mortality were assessed using multivariable logistic regression analysis.

RESULTS

Of 115 465 patients identified, 100 957 (87.5%) were non-obese and 14 508 (12.5%) were obese, with 9489 (8.2%) as obesity class I-II and 5019 (4.3%) as obesity class III. 37 117 patients (32.1%) had an infection diagnosis. Infection was most prevalent among obesity class III (44.0%), followed by obesity class I-II (38.9%) and then non-obese (31.9%). In multivariable modelling, class III obesity (OR = 1.41; 95% CI 1.32-1.51; P < 0.001), and class I-II obesity (OR = 1.08; 95% CI 1.01-1.15; P = 0.026) were associated with infection. Compared to non-obese patients, obese individuals had greater prevalence of bacteraemia, UTI, and skin/soft tissue infection as compared to non-obese patients.

CONCLUSIONS

Obesity is newly identified to be independently associated with infection in end-stage liver disease. The distribution of infection sites varies based on weight category.

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 (&gt;0.98), reliably eliminating the vast majority of patients without SSI, thereby significantly reducing the NSQIP extractors' burden.

Data-driven discovery of seasonally linked diseases from an Electronic Health Records system

BACKGROUND

Patterns of disease incidence can identify new risk factors for the disease or provide insight into the etiology. For example, allergies and infectious diseases have been shown to follow periodic temporal patterns due to seasonal changes in environmental or infectious agents. Previous work searching for seasonal or other temporal patterns in disease diagnosis rates has been limited both in the scope of the diseases examined and in the ability to distinguish unexpected seasonal patterns. Electronic Health Records (EHR) compile extensive longitudinal clinical information, constituting a unique source for discovery of trends in occurrence of disease. However, the data suffer from inherent biases that preclude an identification of temporal trends.

METHODS

Motivated by observation of the biases in this data source, we developed a method (Lomb-Scargle periodograms in detrended data, LSP-detrend) to find periodic patterns by adjusting the temporal information for broad trends in incidence, as well as seasonal changes in total hospitalizations. LSP-detrend can sensitively uncover periodic temporal patterns in the corrected data and identify the significance of the trend. We apply LSP-detrend to a compilation of records from 1.5 million patients encoded by ICD-9-CM (International Classification of Diseases, Ninth Revision, Clinical Modification), including 2,805 disorders with more than 500 occurrences across a 12 year period, recorded from 1.5 million patients.

RESULTS AND CONCLUSIONS

Although EHR data, and ICD-9 coded records in particular, were not created with the intention of aggregated use for research, these data can in fact be mined for periodic patterns in incidence of disease, if confounders are properly removed. Of all diagnoses, around 10% are identified as seasonal by LSP-detrend, including many known phenomena. We robustly reproduce previous findings, even for relatively rare diseases. For instance, Kawasaki disease, a rare childhood disease that has been associated with weather patterns, is detected as strongly linked with winter months. Among the novel results, we find a bi-annual increase in exacerbations of myasthenia gravis, a potentially life threatening complication of an autoimmune disease. We dissect the causes of this seasonal incidence and propose that factors predisposing patients to this event vary through the year.