Can a Transparent Machine Learning Algorithm Predict Better than Its Black Box Counterparts? A Benchmarking Study Using 110 Data Sets.

We developed a novel machine learning (ML) algorithm with the goal of producing transparent models (i.e., understandable by humans) while also flexibly accounting for nonlinearity and interactions. Our method is based on ranked sparsity, and it allows for flexibility and user control in varying the shade of the opacity of black box machine learning methods. The main tenet of ranked sparsity is that an algorithm should be more skeptical of higher-order polynomials and interactions a priori compared to main effects, and hence, the inclusion of these more complex terms should require a higher level of evidence. In this work, we put our new ranked sparsity algorithm (as implemented in the open source R package, sparseR) to the test in a predictive model "bakeoff" (i.e., a benchmarking study of ML algorithms applied "out of the box", that is, with no special tuning). Algorithms were trained on a large set of simulated and real-world data sets from the Penn Machine Learning Benchmarks database, addressing both regression and binary classification problems. We evaluated the extent to which our human-centered algorithm can attain predictive accuracy that rivals popular black box approaches such as neural networks, random forests, and support vector machines, while also producing more interpretable models. Using out-of-bag error as a meta-outcome, we describe the properties of data sets in which human-centered approaches can perform as well as or better than black box approaches. We found that interpretable approaches predicted optimally or within 5% of the optimal method in most real-world data sets. We provide a more in-depth comparison of the performances of random forests to interpretable methods for several case studies, including exemplars in which algorithms performed similarly, and several cases when interpretable methods underperformed. This work provides a strong rationale for including human-centered transparent algorithms such as ours in predictive modeling applications.

Assessing Variable Importance for Best Subset Selection.

One of the primary issues that arises in statistical modeling pertains to the assessment of the relative importance of each variable in the model. A variety of techniques have been proposed to quantify variable importance for regression models. However, in the context of best subset selection, fewer satisfactory methods are available. With this motivation, we here develop a variable importance measure expressly for this setting. We investigate and illustrate the properties of this measure, introduce algorithms for the efficient computation of its values, and propose a procedure for calculating p-values based on its sampling distributions. We present multiple simulation studies to examine the properties of the proposed methods, along with an application to demonstrate their practical utility.

Bootstrap Approximation of Model Selection Probabilities for Multimodel Inference Frameworks.

Most statistical modeling applications involve the consideration of a candidate collection of models based on various sets of explanatory variables. The candidate models may also differ in terms of the structural formulations for the systematic component and the posited probability distributions for the random component. A common practice is to use an information criterion to select a model from the collection that provides an optimal balance between fidelity to the data and parsimony. The analyst then typically proceeds as if the chosen model was the only model ever considered. However, such a practice fails to account for the variability inherent in the model selection process, which can lead to inappropriate inferential results and conclusions. In recent years, inferential methods have been proposed for multimodel frameworks that attempt to provide an appropriate accounting of modeling uncertainty. In the frequentist paradigm, such methods should ideally involve model selection probabilities, i.e., the relative frequencies of selection for each candidate model based on repeated sampling. Model selection probabilities can be conveniently approximated through bootstrapping. When the Akaike information criterion is employed, Akaike weights are also commonly used as a surrogate for selection probabilities. In this work, we show that the conventional bootstrap approach for approximating model selection probabilities is impacted by bias. We propose a simple correction to adjust for this bias. We also argue that Akaike weights do not provide adequate approximations for selection probabilities, although they do provide a crude gauge of model plausibility.

The Increase in Household Transmission of Pertussis Associated With Diagnostic Delays.

BACKGROUND: Pertussis is a highly contagious respiratory illness that can be especially dangerous to young children. Transmission of pertussis often occurs in household settings and is impacted by the timing of treatment and postexposure chemoprophylaxis. This study analyzes the risk for secondary household transmission and if delays in diagnosing pertussis increased the risk for household transmission. METHODS: We conducted 2 population-based studies using a large nationally representative administrative claims database. The first study utilized a stratified monthly incidence model to compare the incidence of pertussis among enrollees exposed to a family member with pertussis versus those not exposed. The second study was conducted at a household level following the index case of pertussis in each household. We identified diagnostic delays in the initial household case and used a logistic regression model to evaluate if such delays were associated with a greater risk for transmission. RESULTS: The incidence rate ratio of pertussis was 938.99 [95% confidence interval (CI): 880.19-1001.73] among enrollees exposed to a family member with pertussis relative to those not exposed. The odds of secondary household transmission in households where the index case experienced a diagnostic delay was 5.10 (CI: 4.44-5.85) times the odds of transmission when the index case was not delayed. We found that longer delays were associated with a greater risk for secondary household transmission ( P < 0.0001). CONCLUSIONS: There is a high rate of secondary transmission of pertussis in household settings. Diagnostic delays increase the likelihood that pertussis will transmit in the household.

Cystic Fibrosis Carrier States Are Associated With More Severe Cases of Bronchiectasis.

Background: People with cystic fibrosis (CF) are at increased risk for bronchiectasis, and several reports suggest that CF carriers may also be at higher risk for developing bronchiectasis. The purpose of this study was to determine if CF carriers are at risk for more severe courses or complications of bronchiectasis. Methods: Using MarketScan data (2001-2021), we built a cohort consisting of 105 CF carriers with bronchiectasis and 300 083 controls with bronchiectasis but without a CF carrier diagnosis. We evaluated if CF carriers were more likely to be hospitalized for bronchiectasis. In addition, we examined if CF carriers were more likely to be infected with Pseudomonas aeruginosa or nontuberculous mycobacteria (NTM) or to have filled more antibiotic prescriptions. We considered regression models for incident and rate outcomes that controlled for age, sex, smoking status, and comorbidities. Results: The odds of hospitalization were almost 2.4 times higher (95% CI, 1.116-5.255) for CF carriers with bronchiectasis when compared with non-CF carriers with bronchiectasis. The estimated odds of being diagnosed with a Pseudomonas infection for CF carriers vs noncarriers was about 4.2 times higher (95% CI, 2.417-7.551) and 5.4 times higher (95% CI, 3.398-8.804) for being diagnosed with NTM. The rate of distinct antibiotic fill dates was estimated to be 2 times higher for carriers as compared with controls (95% CI, 1.735-2.333), and the rate ratio for the total number of days of antibiotics supplied was estimated as 2.8 (95% CI, 2.290-3.442). Conclusions: CF carriers with bronchiectasis required more hospitalizations and more frequent administration of antibiotics as compared with noncarriers. Given that CF carriers were also more likely to be diagnosed with Pseudomonas and NTM infections, CF carriers with bronchiectasis may have a phenotype more resembling CF-related bronchiectasis than non-CF bronchiectasis.

Diurnal temperature variation and the implications for diagnosis and infectious disease screening: a population-based study.

OBJECTIVES: Fevers have been used as a marker of disease for hundreds of years and are frequently used for disease screening. However, body temperature varies over the course of a day and across individual characteristics; such variation may limit the detection of febrile episodes complicating the diagnostic process. Our objective was to describe individual variation in diurnal temperature patterns during episodes of febrile activity using millions of recorded temperatures and evaluate the probability of recording a fever by sex and for different age groups. METHODS: We use timestamped deidentified temperature readings from thermometers across the US to construct illness episodes where continuous periods of activity in a single user included a febrile reading. We model the mean temperature recorded and probability of registering a fever across the course of a day using sinusoidal regression models while accounting for user age and sex. We then estimate the probability of recording a fever by time of day for children, working-age adults, and older adults. RESULTS: We find wide variation in body temperatures over the course of a day and across individual characteristics. The diurnal temperature pattern differed between men and women, and average temperatures declined for older age groups. The likelihood of detecting a fever varied widely by the time of day and by an individual's age or sex. CONCLUSIONS: Time of day and demographics should be considered when using body temperatures for diagnostic or screening purposes. Our results demonstrate the importance of follow-up thermometry readings if infectious diseases are suspected.

Recruitment of Row Crop Farmers into a Research Study to Assess Farm Hazards.

Farmers are at an elevated risk for injuries and are, therefore, highly sought after for research studies. However, their participation in research studies is low. We examine how characteristics of the farmer, farm location, and timing of recruitment contact impact the probability that farmers will engage and participate in a study of injuries and related farm hazards. Study data were obtained from the Farm Safety Study conducted at the University of Iowa between June 2019 and March 2020. We used recruitment data from participants enrolled using Farm Journal magazine subscription lists. Multinomial logistic regression was used for predictive modeling. Predictor variables included the time of day and the farm season in which phone contact for study recruitment was attempted, as well as the rurality of the farm. Two models were created to characterize screening and participation of farmers in the study. Farm season and time of day of the last recruitment call increased the likelihood of farmers being screened for study participation and completing the study. Specifically, contacting farmers during the growing season and during the daytime, regardless of farm rurality, resulted in higher probabilities of participation. Studies of agricultural injury may be more efficiently conducted, with higher participation responses, when circumstances of the recruitment call are considered. This work serves as a starting place for much-needed methodological research to identify factors that increase participation of farmers and farm workers in research studies.

Does Implementation Matter? Associations Between Implementation of Maine's Anti- Bullying Law and Bullying Victimization Among High School Youth.

PURPOSE: To characterize the relationship between implementation of an antibullying law and bullying rates among high school youth. METHODS: School staff (administrators, counselors, and teachers) from public high schools in Maine completed a survey assessing: (1) the frequency with which they implemented 17 components of their district's antibullying policy as mandated by state law; and (2) confidence in implementing the law. Their responses were linked to data on bullying victimization among high school respondents to the Maine Integrated Youth Health Survey, which created a population-based dataset of 84 high schools with 29,818 student responses. RESULTS: Students in schools where administrators (adjusted odds ratio = 0.93; 95% CI: 0.89, 0.97) and counselors (adjusted odds ratio = 0.86; 95% CI: 0.81, 0.92) reported implementing more mandated components of the law experienced notable reductions in the odds of bullying, controlling for student-level characteristics (sex, race, grade) and for school-level bullying rates assessed prior to the passage of the law. With respect to specific implementation components, bullying was most consistently reduced in schools where staff reported increased referrals for counseling and other supports for targets of bullying and in schools where counselors and teachers were interviewed as part of bullying investigations. Students in schools where teachers reported increased confidence in implementing the antibullying law also had reduced odds of bullying. DISCUSSION: These data provide some of the first evidence that the efficacy of a state's antibullying law depends in part on the extent to which school personnel implement the law.

Association between household opioid prescriptions and risk for overdose among family members not prescribed opioids.

BACKGROUND: Prescription opioids have contributed to the rise in opioid-related overdoses and deaths. The presence of opioids within households may increase the risk of overdose among family members who were not prescribed an opioid themselves. Larger quantities of opioids may further increase risk. OBJECTIVES: To determine the risk of opioid overdose among individuals who were not prescribed an opioid but were exposed to opioids prescribed to other family members in the household, and evaluate the risk in relation to the total morphine milligram equivalents (MMEs) present in the household. METHODS: We conducted a cohort study using a large database of commercial insurance claims from 2001 to 2021. For inclusion in the cohort, we identified individuals not prescribed an opioid in the prior 90 days from households with two or more family members, and determined the total MMEs prescribed to other family members. Individuals were stratified into monthly enrollment strata defined by household opioid exposure and other confounders. A generalized linear model was used to estimate incidence rate ratios (IRRs) for overdose. RESULTS: Overall, the incidence of overdose among enrollees in households where a family member was prescribed an opioid was 1.73 (95% confidence interval [CI]: 1.67-1.78) times greater than households without opioid prescriptions. The risk of overdose increased continuously with the level of potential MMEs in the household from an IRR of 1.23 (95% CI: 1.16-1.32) for 1-100 MMEs to 4.67 (95% CI: 4.18-5.22) for >12,000 MMEs. The risk of overdose associated with household opioid exposure was greatest for ages 1-2 years (IRR: 3.46 [95% CI: 2.98-4.01]) and 3-5 years (IRR: 3.31 [95% CI: 2.75-3.99]). CONCLUSIONS: The presence of opioids in a household significantly increases the risk of overdose among other family members who were not prescribed an opioid. Higher levels of MMEs, either in terms of opioid strength or quantity, were associated with increased levels of risk. Risk estimates may reflect accidental poisonings among younger family members.

Creation of a data commons for substance misuse related health research through privacy-preserving patient record linkage between hospitals and state agencies.

Objectives: Substance misuse is a complex and heterogeneous set of conditions associated with high mortality and regional/demographic variations. Existing data systems are siloed and have been ineffective in curtailing the substance misuse epidemic. Therefore, we aimed to build a novel informatics platform, the Substance Misuse Data Commons (SMDC), by integrating multiple data modalities to provide a unified record of information crucial to improving outcomes in substance misuse patients. Materials and Methods: The SMDC was created by linking electronic health record (EHR) data from adult cases of substance (alcohol, opioid, nonopioid drug) misuse at the University of Wisconsin hospitals to socioeconomic and state agency data. To ensure private and secure data exchange, Privacy-Preserving Record Linkage (PPRL) and Honest Broker services were utilized. The overlap in mortality reporting among the EHR, state Vital Statistics, and a commercial national data source was assessed. Results: The SMDC included data from 36 522 patients experiencing 62 594 healthcare encounters. Over half of patients were linked to the statewide ambulance database and prescription drug monitoring program. Chronic diseases accounted for most underlying causes of death, while drug-related overdoses constituted 8%. Our analysis of mortality revealed a 49.1% overlap across the 3 data sources. Nonoverlapping deaths were associated with poor socioeconomic indicators. Discussion: Through PPRL, the SMDC enabled the longitudinal integration of multimodal data. Combining death data from local, state, and national sources enhanced mortality tracking and exposed disparities. Conclusion: The SMDC provides a comprehensive resource for clinical providers and policymakers to inform interventions targeting substance misuse-related hospitalizations, overdoses, and death.

Predictors of rural driver self-reported passing behaviors when interacting with farm equipment on the roadway.

Background: Crashes involving farm equipment (FE) are a major safety concern for farmers as well as all other users of the public road system in both rural and urban areas. These crashes often involve passenger vehicle drivers striking the farm equipment from behind or attempting to pass, but little is known about drivers' perceived norms and self-reported passing behaviors. The objective of this study is to examine factors influencing drivers' farm equipment passing frequencies and their perceptions about the passing behaviors of other drivers. Methods: Data were collected via intercept surveys with adult drivers at local gas stations in two small rural towns in Iowa. The survey asked drivers about their demographic information, frequency of passing farm equipment, and perceptions of other drivers' passing behavior in their community and state when approaching farm equipment (proximal and distal descriptive norms). A multinomial logistic regression model was used to estimate the relationship between descriptive norms and self-reported passing behavior. Results: Survey data from 201 adult drivers showed that only 10% of respondents considered farm equipment crashes to be a top road safety concern. Respondents who perceived others passing farm equipment frequently in their community were more likely to report that they also frequently pass farm equipment. The results also showed interactions between gender and experience operating farm equipment in terms of self-reported passing behavior. Conclusions/Implications: Results from this study suggest local and state-level norms and perceptions of those norms may be important targets for intervention to improve individual driving behaviors around farm equipment.

Informing traffic enforcement leniency and discretion: Crash culpability and the effectiveness of written warnings versus citations.

OBJECTIVES: Deterrence of risky driving behavior is important for the prevention of crashes and injuries. Traffic law enforcement is a key strategy used to decrease risky driving, but there is little evidence on the deterrent effect of issuing warnings versus citations to drivers regarding the prevention of future crashes. The purpose of this study was to 1) investigate the difference between citations and written warnings in their association with future crash culpability and 2) investigate whether drivers who were issued written warnings or citations have different associations with future crash culpability likelihood than those without prior citations or written warnings. METHODS: Data for this study included Iowa Department of Transportation crash data for 2016 to 2019 linked to data from the Iowa Court Case Management System. A quasi-induced exposure method was used based on driver pairs involved in the same collision in which one driver was deemed culpable and one was non-culpable. Conditional logistic regression models were constructed to examine predictors of crash culpability. The main independent variable was traffic citation and warnings history categorized into moving warning, non-moving warning, moving citation, non-moving citation, or no citation or warning in the 30 days prior to the crash. RESULTS: The study sample included a total of 152,986 drivers. Among drivers with moving violations, previously cited drivers were more likely to be crash culpable than previously warned drivers (OR = 1.64, 95% CI = 1.29-2.08). Drivers with prior non-moving citations were less likely to be the culpable party in a crash than a driver who had no recent warnings or citations (OR = 0.72, 95% CI = 0.58-0.89). Drivers with prior warnings (moving or non-moving) did not appreciably differ in crash culpability relative to drivers who had not received any citations or warnings in the previous 30 days. CONCLUSIONS: Drivers with prior moving citations were more likely to be culpable in a future crash than drivers with prior moving warnings, which may relate to overall driving riskiness as opposed to effectiveness of citations in deterring risky driving behaviors. Results from this study also suggest that officer discretion was being appropriately applied by citing the riskiest drivers, while giving lower risk drivers warnings. Results from this study may be useful to support strengthening of state driver improvement programming.

A comprehensive framework to estimate the frequency, duration, and risk factors for diagnostic delays using bootstrapping-based simulation methods.

BACKGROUND: The incidence of diagnostic delays is unknown for many diseases and specific healthcare settings. Many existing methods to identify diagnostic delays are resource intensive or difficult to apply to different diseases or settings. Administrative and other real-world data sources may offer the ability to better identify and study diagnostic delays for a range of diseases. METHODS: We propose a comprehensive framework to estimate the frequency of missed diagnostic opportunities for a given disease using real-world longitudinal data sources. We provide a conceptual model of the disease-diagnostic, data-generating process. We then propose a bootstrapping method to estimate measures of the frequency of missed diagnostic opportunities and duration of delays. This approach identifies diagnostic opportunities based on signs and symptoms occurring prior to an initial diagnosis, while accounting for expected patterns of healthcare that may appear as coincidental symptoms. Three different bootstrapping algorithms are described along with estimation procedures to implement the resampling. Finally, we apply our approach to the diseases of tuberculosis, acute myocardial infarction, and stroke to estimate the frequency and duration of diagnostic delays for these diseases. RESULTS: Using the IBM MarketScan Research databases from 2001 to 2017, we identified 2,073 cases of tuberculosis, 359,625 cases of AMI, and 367,768 cases of stroke. Depending on the simulation approach that was used, we estimated that 6.9-8.3% of patients with stroke, 16.0-21.3% of patients with AMI and 63.9-82.3% of patients with tuberculosis experienced a missed diagnostic opportunity. Similarly, we estimated that, on average, diagnostic delays lasted 6.7-7.6 days for stroke, 6.7-8.2 days for AMI, and 34.3-44.5 days for tuberculosis. Estimates for each of these measures was consistent with prior literature; however, specific estimates varied across the different simulation algorithms considered. CONCLUSIONS: Our approach can be easily applied to study diagnostic delays using longitudinal administrative data sources. Moreover, this general approach can be customized to fit a range of diseases to account for specific clinical characteristics of a given disease. We summarize how the choice of simulation algorithm may impact the resulting estimates and provide guidance on the statistical considerations for applying our approach to future studies.

The incidence, duration, risk factors, and age-based variation of missed opportunities to diagnose pertussis: A population-based cohort study.

OBJECTIVE: To estimate the incidence, duration and risk factors for diagnostic delays associated with pertussis. DESIGN: We used longitudinal retrospective insurance claims from the Marketscan Commercial Claims and Encounters, Medicare Supplemental (2001-2020), and Multi-State Medicaid (2014-2018) databases. SETTING: Inpatient, emergency department, and outpatient visits. PATIENTS: The study included patients diagnosed with pertussis (International Classification of Diseases [ICD] codes) and receipt of macrolide antibiotic treatment. METHODS: We estimated the number of visits with pertussis-related symptoms before diagnosis beyond that expected in the absence of diagnostic delays. Using a bootstrapping approach, we estimated the number of visits representing a delay, the number of missed diagnostic opportunities per patient, and the duration of delays. Results were stratified by age groups. We also used a logistic regression model to evaluate potential factors associated with delay. RESULTS: We identified 20,828 patients meeting inclusion criteria. On average, patients had almost 2 missed opportunities prior to diagnosis, and delay duration was 12 days. Across age groups, the percentage of patients experiencing a delay ranged from 29.7% to 37.6%. The duration of delays increased considerably with age from an average of 5.6 days for patients aged <2 years to 13.8 days for patients aged ≥18 years. Factors associated with increased risk of delays included emergency department visits, telehealth visits, and recent prescriptions for antibiotics not effective against pertussis. CONCLUSIONS: Diagnostic delays for pertussis are frequent. More work is needed to decrease diagnostic delays, especially among adults. Earlier case identification may play an important role in the response to outbreaks by facilitating treatment, isolation, and improved contact tracing.

Epidemiology of Tuberculosis Among People Living With HIV in the African Cohort Study From 2013 to 2021.

BACKGROUND: The prevalence and incidence of tuberculosis (TB) is high among people living with HIV (PLWH) but is often underdiagnosed in HIV programmatic settings. SETTING: President's Emergency Plan for AIDS Relief (PEPFAR)-supported research sites in Uganda, Kenya, Tanzania, and Nigeria. METHODS: All patients underwent molecular testing at entry into a longitudinal cohort of PLWH and annually thereafter. We assessed the prevalence and incidence of TB and identified clinical and demographic factors associated with prevalent and incident TB using logistic regression and Cox proportional hazard models. RESULTS: From 21 January, 2013, to 1 December 2021, 3171 PLWH were enrolled with a TB prevalence of 3% (n = 93). Of the cases with prevalent TB, 66% (n = 61) were bacteriologically confirmed. The adjusted odds of prevalent TB were significantly higher among those with higher educational attainment, PLWH for 1-5 years since their HIV diagnosis, those who were underweight, and those with CD4 counts <200 cells/mm 3 . The overall TB incidence rate was 600 per 100,000 person-years (95% CI: 481-748). We found that shorter time since HIV diagnosis, being underweight, taking antiretroviral therapy <6 months, and having a CD4 count <200 cells/mm 3 were significantly associated with incident TB. PLWH on dolutegravir/lamivudine/tenofovir had a 78% lower risk of incident TB compared with those on tenofovir/lamivudine/efavirenz (hazard ratio: 0.22; 95% CI: 0.08-0.63). CONCLUSION: The prevalence and incidence of TB was notably high in this cohort sourced from PEPFAR clinics. Aggressive efforts to enhance HIV diagnosis and optimize treatment in programmatic settings are warranted to reduce the risk of HIV-TB co-occurrence in this cohort.

A clinically-guided unsupervised clustering approach to recommend symptoms of disease associated with diagnostic opportunities.

OBJECTIVES: A first step in studying diagnostic delays is to select the signs, symptoms and alternative diseases that represent missed diagnostic opportunities. Because this step is labor intensive requiring exhaustive literature reviews, we developed machine learning approaches to mine administrative data sources and recommend conditions for consideration. We propose a methodological approach to find diagnostic codes that exhibit known patterns of diagnostic delays and apply this to the diseases of tuberculosis and appendicitis. METHODS: We used the IBM MarketScan Research Databases, and consider the initial symptoms of cough before tuberculosis and abdominal pain before appendicitis. We analyze diagnosis codes during healthcare visits before the index diagnosis, and use k-means clustering to recommend conditions that exhibit similar trends to the initial symptoms provided. We evaluate the clinical plausibility of the recommended conditions and the corresponding number of possible diagnostic delays based on these diseases. RESULTS: For both diseases of interest, the clustering approach suggested a large number of clinically-plausible conditions to consider (e.g., fever, hemoptysis, and pneumonia before tuberculosis). The recommended conditions had a high degree of precision in terms of clinical plausibility: >70% for tuberculosis and >90% for appendicitis. Including these additional clinically-plausible conditions resulted in more than twice the number of possible diagnostic delays identified. CONCLUSIONS: Our approach can mine administrative datasets to detect patterns of diagnostic delay and help investigators avoid under-identifying potential missed diagnostic opportunities. In addition, the methods we describe can be used to discover less-common presentations of diseases that are frequently misdiagnosed.

Analysis of Mortality in Patients Treated With Phenylephrine in Septic Shock.

BACKGROUND: Phenylephrine is a selective α1-receptor agonist used to manage shock. Current guidelines for septic shock recommend limited utilization of phenylephrine due to the lack of evidence available. This deviates from previous guidelines, which had recommendations of when utilization may be appropriate. OBJECTIVE: The primary objective of this study was to evaluate mortality in patients receiving phenylephrine for the management of septic shock. METHODS: This was a retrospective chart review from September 2015 to September 2017 evaluating all adult patients admitted to an intensive care unit (ICU) on vasopressors for management of septic shock. Patients were divided into 2 groups, those treated with phenylephrine and those treated without phenylephrine. The primary outcome was mortality. Secondary objectives included days on vasopressors and ICU length of stay. Two subgroup analyses were performed: 1 for phenylephrine as first-line therapy and 1 for patients with tachycardia at initiation of vasopressors. Patients started on phenylephrine for salvage therapy were excluded from this study. RESULTS: 499 patients enrolled in the study. 148 (32%) were enrolled in the phenylephrine group. Phenylephrine was associated with an increase in mortality (56% vs 41%; p = 0.003). There was no difference in the days on vasopressors or ICU length of stay. Patients who had ongoing tachycardia were associated with increased mortality with phenylephrine (54% vs 36%, p = 0.02). There was no difference in mortality when phenylephrine was started as the initial vasopressor. CONCLUSION: Utilization of phenylephrine in septic shock patients, especially those with ongoing tachycardia, was associated with an increased rate of mortality.

Development of a Checklist to Identify Injury Hazards on Row Crop Farms in the Midwestern United States.

Agriculture is among the most dangerous industries in the U.S., yet routine surveillance of injury hazards is not currently being conducted on a national level. The objectives of this study were to describe a new tool, called the Hazard Assessment Checklist (HAC), to identify and characterize farm hazards that increase injury risk to farmers and farm workers, and (2) report the inter-rater reliability of the new tool when administered on row-crop farms in Iowa. Based on a literature review and a consensus of expert opinion, the HAC included hazards related to self-propelled vehicles, powered portable implements, fixed machinery and equipment, farm buildings and structures, fall risks, and portable equipment associated with fall risk. A scoring metric indicating the extent of compliance with recommended safety guidelines and standards was developed for each item of the HAC, which included compliant, minimal improvement needed, substantial improvement needed, and not compliant. Inter-rater reliability was assessed from data collected by research staff on 52 row crop farms in Iowa. Cohen's weighted Kappa values demonstrated high inter-rater reliability, ranging between 0.86 and 0.94, for all HAC sections. The HAC can be completed in 1.5-2 hours on each farm and requires about three hours of training, two hours of which are spent in field training. The ability to monitor injury-related hazards over time using an empirically driven tool will contribute significantly to injury prevention efforts in an industry with consistently high rates of fatal and nonfatal injury.

Should we throw the book at 'em? Charge combinations and conviction rates among alcohol-influenced drivers involved in motorcycle crashes.

INTRODUCTION: Motorcycle fatality rates are increasing, and impaired driving is a major contributing factor. Impaired driving laws are a main component of state efforts to reduce drunk driving, but motorcycle crash charge and conviction outcomes have yet to be studied. The purpose of this study was to evaluate driver charge and conviction outcomes following alcohol-related motorcycle crashes. METHODS: Data for this study were drawn from Iowa crash, charge, and conviction data from 2011 to 2018. The study sample included 480 alcohol-influenced drivers (428 motorcyclists and 52 other vehicle drivers) involved in motorcycle crashes. Driver crash-related charges were categorized by type: Alcohol, Moving Violations, and Administrative/Miscellaneous. Factors associated with convictions were determined and estimated with multivariable logistic regression models. The main factor of interest was charge combination. RESULTS: Over three-quarters (78.5%) of the 480 alcohol-influenced drivers in crashes received any charge type and 68.1% received an alcohol-related charge. Among drivers with any charge, 88.6% were convicted, and among drivers with alcohol charges, 87.2% were convicted on an alcohol charge. After adjusting for BAC, drivers with a combination of Alcohol, Administrative, and Moving Violation charges had more than three times the odds of conviction of any charge compared to drivers with alcohol only charges (OR = 3.21, 95% CI = 1.00-10.26). However, charge combinations had little impact on alcohol-related convictions. CONCLUSIONS: Convictions were more likely when the impaired driver was charged with multiple types of offenses than with a single offense. An increased variety of charges was not associated with greater rates of conviction on alcohol-specific charges, which had high conviction rates overall. PRACTICAL APPLICATIONS: Law enforcement officers should be informed that lesser infractions impact driver conviction outcomes in alcohol-related crashes and procedures for issuing charges should be evaluated to assure equitable enforcement and to hold drivers accountable for unsafe driving behaviors.