Development and Validation of a Natural Language Processing Tool to Identify Patients Treated for Pneumonia across VA Emergency Departments.

BACKGROUND: Identifying pneumonia using diagnosis codes alone may be insufficient for research on clinical decision making. Natural language processing (NLP) may enable the inclusion of cases missed by diagnosis codes. OBJECTIVES: This article (1) develops a NLP tool that identifies the clinical assertion of pneumonia from physician emergency department (ED) notes, and (2) compares classification methods using diagnosis codes versus NLP against a gold standard of manual chart review to identify patients initially treated for pneumonia. METHODS: Among a national population of ED visits occurring between 2006 and 2012 across the Veterans Affairs health system, we extracted 811 physician documents containing search terms for pneumonia for training, and 100 random documents for validation. Two reviewers annotated span- and document-level classifications of the clinical assertion of pneumonia. An NLP tool using a support vector machine was trained on the enriched documents. We extracted diagnosis codes assigned in the ED and upon hospital discharge and calculated performance characteristics for diagnosis codes, NLP, and NLP plus diagnosis codes against manual review in training and validation sets. RESULTS: Among the training documents, 51% contained clinical assertions of pneumonia; in the validation set, 9% were classified with pneumonia, of which 100% contained pneumonia search terms. After enriching with search terms, the NLP system alone demonstrated a recall/sensitivity of 0.72 (training) and 0.55 (validation), and a precision/positive predictive value (PPV) of 0.89 (training) and 0.71 (validation). ED-assigned diagnostic codes demonstrated lower recall/sensitivity (0.48 and 0.44) but higher precision/PPV (0.95 in training, 1.0 in validation); the NLP system identified more "possible-treated" cases than diagnostic coding. An approach combining NLP and ED-assigned diagnostic coding classification achieved the best performance (sensitivity 0.89 and PPV 0.80). CONCLUSION: System-wide application of NLP to clinical text can increase capture of initial diagnostic hypotheses, an important inclusion when studying diagnosis and clinical decision-making under uncertainty.

Susceptibilities of clinical Clostridium difficile isolates to antimicrobials: a systematic review and meta-analysis of studies since 1970.

OBJECTIVES: Although exposure to antibiotics can cause Clostridium difficile infection, certain antibiotics are used to treat C. difficile. Measurements of antimicrobial C. difficile activity could help to identify antibiotic risk and emergent resistance. Here, we describe publication patterns relating to C. difficile susceptibilities and estimate minimum inhibitory concentrations (MIC) for antibiotic classes in the published literature between January 1970 and June 2014. METHODS: We queried PUBMED and EMBASE for studies reporting antibiotic C. difficile MIC in English or French. We used mixed-effects models to obtain pooled estimates of antibiotic class median MIC (MIC50), 90th percentile of MIC (MIC90), and MIC90:MIC50 ratio. RESULTS: Our search identified 182 articles that met our inclusion criteria, of which 27 were retained for meta-analysis. Aminoglycosides (MIC50 120 mg/L, 95% CI 62-250), 3rd (MIC50 75 mg/L, 95% CI 39-130) and 2nd generation cephalosporins (MIC50 64 mg/L, 95% CI 27-140) had the least C. difficile activity. Rifamycins (MIC50 0.034 mg/L, 95% CI 0.012-0.099) and tetracyclines (MIC50 0.29 mg/L, 95% CI 0.054-1.7) had the highest level of activity. The activity of 3rd generation cephalosporins was more than three times lower than that of 1st generation agents (MIC50 19 mg/L, 95% CI 7.0-54). Time-trends in MIC50 were increasing for carbapenems (70% increase per 10 years) while decreasing for tetracyclines (51% decrease per 10 years). CONCLUSIONS: We found a 3500-fold variation in antibiotic C. difficile MIC50, with aminoglycosides as the least active agents and rifamycins as the most active. Further research is needed to determine how in vitro measures can help assess patient C. difficile risk and guide antimicrobial stewardship.

Constructing Ebola transmission chains from West Africa and estimating model parameters using internet sources.

During the recent Ebola crisis in West Africa, individual person-level details of disease onset, transmissions, and outcomes such as survival or death were reported in online news media. We set out to document disease transmission chains for Ebola, with the goal of generating a timely account that could be used for surveillance, mathematical modeling, and public health decision-making. By accessing public web pages only, such as locally produced newspapers and blogs, we created a transmission chain involving two Ebola clusters in West Africa that compared favorably with other published transmission chains, and derived parameters for a mathematical model of Ebola disease transmission that were not statistically different from those derived from published sources. We present a protocol for responsibly gleaning epidemiological facts, transmission model parameters, and useful details from affected communities using mostly indigenously produced sources. After comparing our transmission parameters to published parameters, we discuss additional benefits of our method, such as gaining practical information about the affected community, its infrastructure, politics, and culture. We also briefly compare our method to similar efforts that used mostly non-indigenous online sources to generate epidemiological information.

Scaling-up NLP Pipelines to Process Large Corpora of Clinical Notes.

INTRODUCTION: This article is part of the Focus Theme of Methods of Information in Medicine on "Big Data and Analytics in Healthcare". OBJECTIVES: This paper describes the scale-up efforts at the VA Salt Lake City Health Care System to address processing large corpora of clinical notes through a natural language processing (NLP) pipeline. The use case described is a current project focused on detecting the presence of an indwelling urinary catheter in hospitalized patients and subsequent catheter-associated urinary tract infections. METHODS: An NLP algorithm using v3NLP was developed to detect the presence of an indwelling urinary catheter in hospitalized patients. The algorithm was tested on a small corpus of notes on patients for whom the presence or absence of a catheter was already known (reference standard). In planning for a scale-up, we estimated that the original algorithm would have taken 2.4 days to run on a larger corpus of notes for this project (550,000 notes), and 27 days for a corpus of 6 million records representative of a national sample of notes. We approached scaling-up NLP pipelines through three techniques: pipeline replication via multi-threading, intra-annotator threading for tasks that can be further decomposed, and remote annotator services which enable annotator scale-out. RESULTS: The scale-up resulted in reducing the average time to process a record from 206 milliseconds to 17 milliseconds or a 12- fold increase in performance when applied to a corpus of 550,000 notes. CONCLUSIONS: Purposely simplistic in nature, these scale-up efforts are the straight forward evolution from small scale NLP processing to larger scale extraction without incurring associated complexities that are inherited by the use of the underlying UIMA framework. These efforts represent generalizable and widely applicable techniques that will aid other computationally complex NLP pipelines that are of need to be scaled out for processing and analyzing big data.

Cost-effectiveness of adding decolonization to a surveillance strategy of screening and isolation for methicillin-resistant Staphylococcus aureus carriers.

We compared the cost-effectiveness of a methicillin-resistant Staphylococcus aureus (MRSA) programme of active surveillance plus decolonization with the current Veterans Health Administration (VHA) strategy of active surveillance alone, as well as a common strategy of no surveillance. A decision-analytical model was developed for an inpatient stay time horizon, using the VHA's perspective. Model inputs were taken from published literature where available, and supplemented with expert opinion when necessary. Effectiveness outcomes were hospital-acquired MRSA infections and deaths avoided. One-way and two-way sensitivity analyses and Monte Carlo simulations were performed. In the base-case analysis, the strategy of active surveillance plus decolonization dominated (i.e. lower cost and greater effectiveness) both the comparison strategies of active surveillance and no surveillance. In addition, the active surveillance strategy dominated the strategy of no surveillance. One-way and two-way sensitivity analyses demonstrated that at low levels of direct benefit of decolonization (1-4%), the strategy of active surveillance plus decolonization would no longer be dominant. In the probabilistic sensitivity analysis, active surveillance plus decolonization dominated both the other two strategies, and the active surveillance strategy dominated no surveillance in all of 1000 Monte Carlo simulations. These results provide a strong economic argument for adding an MRSA decolonization protocol to the current VHA active surveillance strategy.

Interventions to control MRSA: high time for time-series analysis?

Time-series methods are useful in quasi-experimental study designs in which rates of antibiotic-resistant infections are ascertained before and after an intervention. However, uncertainties remain regarding the use of time-series analysis as an appropriate research methodology for analysing the effect of infection control interventions and antibiotic policies on the epidemiology of methicillin-resistant Staphylococcus aureus (MRSA). In particular, there is still a substantial gap in our understanding of what actually happens to MRSA incidence when a planned intervention is made on use of one or more antibiotic drug classes.

Sources of disseminated Mycobacterium avium infection in AIDS.

OBJECTIVES: To identify the sources of disseminated Mycobacterium avium complex (MAC) infection in AIDS. METHODS: HIV positive subjects with CD4 counts <100/mm(3) in Atlanta, Boston, New Hampshire and Finland were entered in a prospective cohort study. Subjects were interviewed about potential MAC exposures, had phlebotomy performed for determination of antibody to mycobacterial lipoarabinomannin and for culture. Patient-directed water samples were collected from places of residence, work and recreation. Patients were followed for the development of disseminated MAC. Univariate and multivariate risk factors for MAC were analyzed. RESULTS: Disseminated MAC was identified in 31 (9%) subjects. Significant risks in univariate analysis included prior Pneumocystis carinii pneumonia (PCP) (hazard ratio 1.821), consumption of spring water (4.909), consumption of raw seafood (34.3), gastrointestinal endoscopy (2.894), and showering outside the home (0.388). PCP, showering and endoscopy remained significant in a Cox proportional hazards model. There was no association between M. avium colonization of home water and risk of MAC. In patients with CD4<25, median OD antibody levels to lipoarabinomannin at baseline were 0.054 among patients who did not develop MAC and 0.021 among patients who did develop MAC (P=0.077). CONCLUSIONS: MAC infection results from diverse and likely undetectable environmental and nosocomial exposures. Mycobacterial infection before HIV infection may confer protection against disseminated MAC in advanced AIDS.

Comparison of pulsed-field gel electrophoresis and amplified fragment-length polymorphism for epidemiological investigations of common nosocomial pathogens.

OBJECTIVE: To compare molecular typing by amplified fragment-length polymorphism (AFLP) analysis with pulsed-field gel electrophoresis (PFGE) with respect to the ability to differentiate between epidemiologically related and unrelated isolates of common nosocomial pathogens recovered during a period of endemicity. DESIGN: Retrospective laboratory analysis. SETTING: Tertiary-care institution. METHODS: 17 isolates of Acinetobacter baumannii, 22 isolates of Pseudomonas aeruginosa, and 22 vancomycin-resistant Enterococcusfaecium (VRE) were typed by both methods. RESULTS: AFLP generated comparable results to PFGE for A baumannii and P aeruginosa isolates; both methods identified epidemiologically related and unrelated isolates. However, strain typing of VRE isolates produced discordant results between the two methods. PFGE identified 10 different strain types and differentiated between all epidemiologically related and unrelated isolates. In contrast, AFLP generated only five different strain types, three of which contained both epidemiologically related and unrelated isolates. CONCLUSION: Molecular typing by AFLP is comparable to PFGE for A baumannii and P aeruginosa isolates. For VRE isolates, however, PFGE remains the method of choice.

The epidemiology of nephrotoxicity associated with conventional amphotericin B therapy.

PURPOSE: We sought to quantify the incidence of, define risk factors for, and examine the relation between renal functional impairment and treatment with conventional amphotericin B. SUBJECTS AND METHODS: We performed a 9-year retrospective analysis of amphotericin B-associated nephrotoxicity in 494 adult inpatients who received > or = 2 doses of amphotericin B. Nephrotoxicity was classified according to two nonmutually exclusive severity categories (50% increase or doubling in the baseline creatinine level). RESULTS: The median cumulative dosage of amphotericin B was 240 mg (interquartile range, 113 to 500 mg), with the majority of patients (n = 361) receiving it for empiric treatment. Overall, 139 (28%) patients experienced renal toxicity, including 58 (12%) with moderate-to-severe nephrotoxicity. The rate of nephrotoxicity was relatively constant during amphotericin B treatment. For each 10-mg increase in the mean daily amphotericin B dose, the adjusted rate of renal toxicity increased by a factor of 1.13 (95% confidence interval: 1.02 to 1.25). We defined 5 categorical risk factors: mean daily amphotericin B dose > or = 35 mg, male sex, weight > or = 90 kg, chronic renal disease, and use of amikacin or cyclosporine. The incidence of moderate-to-severe nephrotoxicity was 4% (6 of 137) in patients with none of these risk factors, 8% (14 of 181) in those with 1 risk factor, 18% (21 of 117) in those with 2 risk factors, and 29% (17 of 59) in patients with > or = 3 risk factors. Nephrotoxicity rarely led to hemodialysis (n = 3); however, at the time of discharge or death, 70% of patients with moderate-to-severe nephrotoxicity had a serum creatinine level that was > or = 0.5 mg/dL above baseline. CONCLUSION: Amphotericin B-related nephrotoxicity is an important dose-dependent and duration-dependent toxicity that is accentuated by certain nephrotoxic drugs and patient characteristics. Patients with more than two risk factors for nephrotoxicity are potential candidates for alternative antifungal therapy.

High rates of multiple antibiotic resistance in Streptococcus pneumoniae from healthy children living in isolated rural communities: association with cephalosporin use and intrafamilial transmission.

OBJECTIVE: Streptococcus pneumoniae is one of the most clinically significant pathogens with emerging antibiotic resistance. We performed a surveillance study in isolated rural populations of healthy children to estimate the prevalence of pneumococcal resistance and to contrast factors that predict pneumococcal carriage with those that specifically predict resistant pneumococcal carriage. METHODS: The study was conducted in 1998 in 2 rural communities in Utah. Families were recruited directly for participation through community canvassing. Surveillance nasopharyngeal cultures were obtained from children who were younger than 8 years. Antibiotic usage and information on other potential risk factors were obtained from questionnaires and local pharmacy records. Resistance was determined by testing isolates for susceptibility to penicillin, cefaclor, trimethoprim-sulfamethoxazole, erythromycin, ceftriaxone, and trovafloxacin. Selected resistant isolates were characterized further by serotyping, pulsed field gel electrophoresis, and Southern blot with DNA probes specific for the pneumococcal lytA gene and for antibiotic resistance genes. RESULTS: In April 1998, surveillance nasopharyngeal cultures were obtained from 368 children aged

Parallel analysis of individual and aggregated data on antibiotic exposure and resistance in gram-negative bacilli.

To evaluate the potential bias of analyzing aggregated data, we separately examined antibiotic exposure and resistance data for 35,423 patients admitted to a university hospital in Utah, from both an individual-patient perspective and group-level perspective. From 1994 through 1998, use of defined daily doses (per 1000 patient-days) of fluoroquinolones, third-generation cephalosporins, ampicillin-sulbactam, and imipenem increased by 82%, 38%, and 99%, and decreased by 38%, respectively, whereas group-level resistance rates of Enterobacteriaceae or Pseudomonas species changed only minimally. However, in individual-patient-level analyses performed by multivariable proportional hazards regression, exposure to a fluoroquinolone, third-generation cephalosporin, ampicillin-sulbactam, or imipenem was a strong risk factor for resistance to fluoroquinolones (adjusted hazard ratio [AHR], 4.0; P<.001), third-generation cephalosporins (AHR, 3.5; P<.001), ampicillin-sulbactam (AHR, 2.3; P=.008), or imipenem (AHR, 5.7; P<.001), respectively. Thus, group-level and individual-patient-level analyses of antibiotic-use-versus-susceptibility relations yielded divergent results. Multicenter studies should include individual-patient-level data to elucidate more fully the relation between antibiotic exposure and resistance.

Antibiotic prophylaxis and the risk of Clostridium difficile-associated diarrhoea.

To test the hypothesis that extended antibiotic prophylaxis increases the risk of Clostridium difficile -associated diarrhoea (CDAD), we conducted a retrospective cohort study of 2641 patients under-going cardiovascular surgery. Main outcome measures were the duration of prophylaxis (< 48 h vs. > 48 h) and the occurrence of CDAD. CDAD occurred in 31 patients (1.2%), who were significantly older (70 +/- 9 y vs. 66 +/- 10 y; P = 0.03), received more therapeutic antibiotics (2.2 +/- 1.9 vs. 0.4 +/- 0.9; P = 0.001) and had a longer postoperative hospital stay (26 +/- 19 d vs. 9 +/- 8 d; P < 0.001) than non-cases. After adjusting for confounding, we did not observe an association between prolonged prophylaxis and CDAD [adjusted odds ratio (AOR), 0.8; CI, 0.4-1.8]. In contrast, three independent predictors were identified: increasing length of hospital stay (AOR per one-day-increment, 1.03; CI, 1.01-1.05), and treatment with third generation cephalosporins (AOR, 5.9; CI, 2.2-16.0) or beta-lactam-beta-lactamase inhibitor combinations (AOR, 4.6; CI, 1.7-12.3). Our results did not confirm that extended prophylaxis after clean surgery increases the risk of CDAD, which remains an uncommon postoperative complication, associated even with short antibiotic exposure.

Methodological principles of case-control studies that analyzed risk factors for antibiotic resistance: a systematic review.

Case-control studies that analyze the risk factors for antibiotic-resistant organisms have varied epidemiological methodologies, which may lead to biased estimates of antibiotic risk factors. A systematic review of case-control studies that analyzed risk factors for antibiotic-resistant organisms addressed 3 methodological principles: method of control group selection, adjustment for time at risk, and adjustment for comorbid illness. A total of 406 abstracts were reviewed. Thirty-seven studies met the inclusion and exclusion criteria and were reviewed and evaluated for the 3 methodological principles. Thirteen (35%) of 37 studies chose the preferred control group. Eleven adjusted for time at risk. Twenty-seven adjusted for comorbid illness. Future studies need to consider more closely the optimization of control group selection, adjusting for confounding caused by time at risk, and adjusting for confounding caused by comorbid illness.

Benefits of negative penicillin skin test results persist during subsequent hospital admissions.

For an initial series of 38 patients with negative skin test results, we reviewed retrospectively all subsequent admissions over a 2-year period. For 38 patients with negative initial skin test results, there were 48 subsequent readmissions to our institution, of which 35 required antibiotics. beta-lactams were prescribed for 86% of admissions; a penicillin for 37%, and a cephalosporin for 51%. All infections were cured, and there were no allergic drug reactions during any of the admissions that were reviewed.

Clinical isolation and resistance patterns of and superinfection with 10 nosocomial pathogens after treatment with ceftriaxone versus ampicillin-sulbactam.

Isolation of pathogens from clinical cultures and their resistance patterns may be altered by antecedent antibiotic treatment. The objective of this study was to assess the influence of treatment with ceftriaxone versus that with ampicillin-sulbactam on recovery and superinfections with 10 nosocomial pathogens. The study was designed as a historical cohort study, using a propensity score to adjust for confounding by indication and multivariate survival analyses to adjust for other confounding. Two thousand four hundred forty-five patients were treated with ampicillin-sulbactam, and 1, 308 were treated with ceftriaxone. The study analyzed two outcomes: (i) recovery of pathogens from clinical cultures and (ii) microbiologically documented infections. Data were obtained from administrative, pharmacy, clinical, and laboratory databases and by chart extraction. Following treatment, new isolation of at least 1 of the 10 target pathogens occurred for 244 patients. After adjustment, more infections occurred in the ampicillin-sulbactam group (hazard ratio [HR], 1.55; P = 0.009). This was observed with all gram-negative rods combined (HR, 3.6; P < 0.001) and with each genus of the family Enterobacteriaceae. No differences in isolation of gram-positive bacteria were evident (P = 0.33). Microbiologically documented superinfections occurred in 172 patients and were less frequent in the ceftriaxone group (3.8% versus 5%; HR, 1.6; P = 0. 015). All the Escherichia coli and Klebsiella spp. isolates were susceptible to ceftriaxone, but half were resistant to ampicillin-sulbactam. The prevalence of oxacillin resistance among Staphylococcus aureus isolates was higher in the ceftriaxone group (63% versus 31%; odds ratio, 3.8; P = 0.08). Differences in the rates of superinfections and the likely causative organisms following treatment with ceftriaxone or ampicillin-sulbactam were evident. This may guide clinicians in empirical choices of antibiotics to treat superinfection.

Developing a taxonomy for research in adverse drug events: potholes and signposts.

Computerized decision support and order entry shows great promise for reducing adverse drug events (ADEs). The evaluation of these solutions depends on a framework of definitions and classifications that is clear and practical. Unfortunately the literature does not always provide a clear path to defining and classifying adverse drug events. While not a systematic review, this paper uses examples from the literature to illustrate problems that investigators will confront as they develop a conceptual framework for their research. It also proposes a targeted taxonomy that can facilitate a clear and consistent approach to the research of ADEs and aid in the comparison to results of past and future studies. The taxonomy addresses the definition of ADE, types, seriousness, error, and causality.

A survey on handwashing practices and opinions of healthcare workers.

Studies have consistently demonstrated rates of handwashing compliance are less than 50%. The objective of this study was to gain the following information about handwashing: self-reported compliance; attitudes towards handwashing in different patient settings; and attitudes towards interventions aimed at increasing compliance. A 74-question survey was administered to healthcare workers in two tertiary care hospitals. One hundred and ninety nine healthcare workers completed the survey and 89% reported that handwashing is an important means of preventing infection. Sixty-four percent believed that they washed their hands as often as their peers and 2% believed that they washed less often than their peers. Patients with diarrhoea, AIDS or patients on antibiotics led to increased handwashing. Relative to potential interventions, 76% reported that rewards for handwashing would have no effect, 73% reported that punishment would have no effect and 80% reported that easy access to sinks and availability of washing facilities would lead to increased compliance. This survey suggests that healthcare workers understand the importance of handwashing, but tend to overestimate their own compliance. Healthcare workers are not in favour of interventions involving rewards and punishments, but are more attracted to interventions that make handwashing easier.