ABSTRACT
In resource-limited settings, timely treatment of acute stroke is dependent upon accurate diagnosis that draws on non-contrast computed tomography (NCCT) scans of the head. Artificial Intelligence (AI) based devices may be able to assist non-specialist physicians in NCCT interpretation, thereby enabling faster interventions for acute stroke patients in these settings. We evaluated the impact of an AI device by comparing the time to intervention (TTI) from NCCT imaging to significant intervention before (baseline) and after the deployment of AI, in patients diagnosed with stroke (ischemic or hemorrhagic) through a retrospective interrupted time series analysis at a rural hospital managed by non-specialists in India. Significant intervention was defined as thrombolysis or antiplatelet therapy in ischemic strokes, and mannitol for hemorrhagic strokes or mass effect. We also evaluated the diagnostic accuracy of the software using the teleradiologists' reports as ground truth. Impact analysis in a total of 174 stroke patients (72 in baseline and 102 after deployment) demonstrated a significant reduction of median TTI from 80 minutes (IQR: 56·8-139·5) during baseline to 58·50 (IQR: 30·3-118.2) minutes after AI deployment (Wilcoxon rank sum test-location shift: -21·0, 95% CI: -38·0, -7·0). Diagnostic accuracy evaluation in a total of 864 NCCT scans demonstrated the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) in detecting intracranial hemorrhage to be 0·89 (95% CI: 0·83-0·93), 0·99 (0·98-1·00), 0·96 (0·91-0·98) and 0·97 (0·96-0·98) respectively, and for infarct these were 0·84 (0·79-0·89), 0·81 (0·77-0·84), 0·58 (0·52-0·63), and 0·94 (0·92-0·96), respectively. AI-based NCCT interpretation supported faster abnormality detection with high accuracy, resulting in persons with acute stroke receiving significantly earlier treatment. Our results suggest that AI-based NCCT interpretation can potentially improve uptake of lifesaving interventions for acute stroke in resource-limited settings.
ABSTRACT
Deep learning is a subfield of artificial intelligence and machine learning based mostly on neural networks and often combined with attention algorithms that has been used to detect and identify objects in text, audio, images, and video. Serghiou and Rough (Am J Epidemiol. 0000;000(00):0000-0000) present a primer for epidemiologists on deep learning models. These models provide substantial opportunities for epidemiologists to expand and amplify their research in both data collection and analyses by increasing the geographic reach of studies, including more research subjects, and working with large or high dimensional data. The tools for implementing deep learning methods are not quite yet as straightforward or ubiquitous for epidemiologists as traditional regression methods found in standard statistical software, but there are exciting opportunities for interdisciplinary collaboration with deep learning experts, just as epidemiologists have with statisticians, healthcare providers, urban planners, and other professionals. Despite the novelty of these methods, epidemiological principles of assessing bias, study design, interpretation and others still apply when implementing deep learning methods or assessing the findings of studies that have used them.
ABSTRACT
Development of new therapeutics for a rare disease such as cystic fibrosis (CF) is hindered by challenges in accruing enough patients for clinical trials. Using external controls from well-matched historical trials can reduce prospective trial sizes, and this approach has supported regulatory approval of new interventions for other rare diseases. We consider three statistical methods that incorporate external controls into a hypothetical clinical trial of a new treatment to reduce pulmonary exacerbations in CF patients: 1) inverse probability weighting, 2) Bayesian modeling with propensity score-based power priors, and 3) hierarchical Bayesian modeling with commensurate priors. We compare the methods via simulation study and in a real clinical trial data setting. Simulations showed that bias in the treatment effect was <4% using any of the methods, with type 1 error (or in the Bayesian cases, posterior probability of the null hypothesis) usually <5%. Inverse probability weighting was sensitive to similarity in prevalence of the covariates between historical and prospective trial populations. The commensurate prior method performed best with real clinical trial data. Using external controls to reduce trial size in future clinical trials holds promise and can advance the therapeutic pipeline for rare diseases.
ABSTRACT
BACKGROUND: Although road traffic injuries and deaths have decreased globally, there is substantial national and sub-national heterogeneity, particularly in low- and middle-income countries (LMICs). Ghana is one of few countries in Africa collecting comprehensive, spatially detailed data on motor vehicle collisions (MVCs). This data is a critical step towards improving roadway safety, as accurate and reliable information is essential for devising targeted countermeasures. METHODS: Here, we analyze 16 years of police-report data using emerging hot spot analysis in ArcGIS to identify hot spots with trends of increasing injury severity (a weighted composite measure of MVCs, minor injuries, severe injuries, and deaths), and counts of injuries, severe injuries, and deaths along major roads in urban and rural areas of Ghana. RESULTS: We find injury severity index sums and minor injury counts are significantly decreasing over time in Ghana while severe injury and death counts are not, indicating the latter should be the focus for road safety efforts. We identify new, consecutive, intensifying, and persistent hot spots on 2.65% of urban roads and 4.37% of rural roads. Hot spots are intensifying in terms of severity and frequency on major roads in rural areas. CONCLUSIONS: A few key road sections, particularly in rural areas, show elevated levels of road traffic injury severity, warranting targeted interventions. Our method for evaluating spatiotemporal trends in MVC, road traffic injuries, and deaths in a LMIC includes sufficient detail for replication and adaptation in other countries, which is useful for targeting countermeasures and tracking progress.
Subject(s)
Accidents, Traffic , Spatio-Temporal Analysis , Wounds and Injuries , Ghana/epidemiology , Accidents, Traffic/statistics & numerical data , Accidents, Traffic/mortality , Humans , Wounds and Injuries/epidemiology , Longitudinal Studies , Trauma Severity IndicesABSTRACT
Background: Antipsychotics carry a higher-risk profile than other psychotropic medications and may be prescribed for youth with conditions in which other first-line treatments are more appropriate. This study aimed to evaluate the population-level effect of the Safer Use of Antipsychotics in Youth (SUAY) trial, which aimed to reduce person-days of antipsychotic use among participants. Methods: We conducted an interrupted time series analysis using segmented regression to measure changes in prescribing trends of antipsychotic initiation rates pre-SUAY and post-SUAY trial at four U.S. health systems between 2013 and 2020. Results: In our overall model, adjusted for age and insurance type, antipsychotic initiation rates decreased by 0.73 (95% confidence interval [CI]: 0.30, 1.16, p = 0.002) prescriptions per 10,000 person-months before the SUAY trial. In the first quarter following the start of the trial, there was an immediate decrease in the rate of antipsychotic initiations of 6.57 (95% CI: 0.99, 12.15) prescriptions per 10,000 person-months. When comparing the posttrial period to the pretrial period, there was an increase of 1.09 (95% CI: 0.32, 1.85) prescriptions per 10,000 person-months, but the increasing rate in the posttrial period alone was not statistically significant (0.36 prescriptions per 10,000 person-months, 95% CI: -0.27, 0.99). Conclusion: The declining trend of antipsychotic initiation seen between 2013 and 2018 (pre-SUAY trial) may have naturally reached a level at which prescribing was clinically warranted and appropriate, resulting in a floor effect. The COVID-19 pandemic, which began in the final three quarters of the posttrial period, may also be related to increased antipsychotic medication initiation.
ABSTRACT
BACKGROUND: Early detection of lung cancer reduces cancer mortality; yet uptake for lung cancer screening (LCS) has been limited in Washington State. Geographic disparities contribute to low uptake, but do not wholly explain gaps in access for underserved populations. Other factors, such as an adequate workforce to meet population demand and the capacity of accredited screening facility sites, must also be considered. RESEARCH QUESTION: What proportion of the eligible population for LCS has access to LCS facilities in Washington State? STUDY DESIGN AND METHODS: We used the enhanced two-step floating catchment area (E2SFCA) model to evaluate how geographic accessibility in addition to availability of LCS imaging centers contribute to disparities. We used available data on radiologic technologist volume at each American College of Radiology (ACR)-accredited screening facility site to estimate the capacity of each site to meet potential population demand. Spearman rank correlation coefficients of the spatial access ratios were compared with the 2010 Rural-Urban Commuting Area (RUCA) codes and area deprivation index quintiles to identify characteristics of populations at risk for lung cancer with greater and lesser levels of access. RESULTS: A total of 549 radiologic technologists were identified across the 95 ACR-accredited screening facilities. We observed that 95% of the eligible population had proximate geographic access to any ACR facility. However, when we incorporated the E2SFCA method, we found significant variation of access for eligible populations. The inclusion of the availability measure attenuated access for most of the eligible population. Furthermore, we observed that rural areas were substantially correlated, and areas with greater socioeconomic disadvantage were modestly correlated, with lower access. INTERPRETATION: Rural and socioeconomically disadvantaged areas face significant disparities. The E2SFCA models demonstrated that capacity is an important component and how geographic access and availability jointly contribute to disparities in access to LCS.
ABSTRACT
Importance: Sweetened beverage taxes have been associated with reduced purchasing of taxed beverages. However, few studies have assessed the association between sweetened beverage taxes and health outcomes. Objective: To evaluate the association between the Seattle sweetened beverage tax and change in body mass index (BMI) among children. Design, Setting, and Participants: In this longitudinal cohort study, anthropometric data were obtained from electronic medical records of 2 health care systems (Kaiser Permanente Washington [KP] and Seattle Children's Hospital Odessa Brown Children's Clinic [OBCC]). Children were included in the study if they were aged 2 to 18 years (between January 1, 2014, and December 31, 2019); had at least 1 weight measurement every year between 2015 and 2019; lived in Seattle or in urban areas of 3 surrounding counties (King, Pierce, and Snohomish); had not moved between taxed (Seattle) and nontaxed areas; received primary health care from KP or OBCC; did not have a recent history of cancer, bariatric surgery, or pregnancy; and had biologically plausible height and BMI (calculated as weight in kilograms divided by height in meters squared). Data analysis was conducted between August 5, 2022, and March 4, 2024. Exposure: Seattle sweetened beverage tax (1.75 cents per ounce on sweetened beverages), implemented on January 1, 2018. Main Outcomes and Measures: The primary outcome was BMIp95 (BMI expressed as a percentage of the 95th percentile; a newly recommended metric for assessing BMI change) of the reference population for age and sex, using the Centers for Disease Control and Prevention growth charts. In the primary (synthetic difference-in-differences [SDID]) model used, a comparison sample was created by reweighting the comparison sample to optimize on matching to pretax trends in outcome among 6313 children in Seattle. Secondary models were within-person change models using 1 pretax measurement and 1 posttax measurement in 22â¯779 children and fine stratification weights to balance baseline individual and neighborhood-level confounders. Results: The primary SDID analysis included 6313 children (3041 female [48%] and 3272 male [52%]). More than a third of children (2383 [38%]) were aged 2 to 5 years); their mean (SE) age was 7.7 (0.6) years. With regard to race and ethnicity, 789 children (13%) were Asian, 631 (10%) were Black, 649 (10%) were Hispanic, and 3158 (50%) were White. The primary model results suggested that the Seattle tax was associated with a larger decrease in BMIp95 for children living in Seattle compared with those living in the comparison area (SDID: -0.90 percentage points [95% CI, -1.20 to -0.60]; P < .001). Results from secondary models were similar. Conclusions and Relevance: The findings of this cohort study suggest that the Seattle sweetened beverage tax was associated with a modest decrease in BMIp95 among children living in Seattle compared with children living in nearby nontaxed areas who were receiving care within the same health care systems. Taken together with existing studies in the US, these results suggest that sweetened beverage taxes may be an effective policy for improving children's BMI. Future research should test this association using longitudinal data in other US cities with sweetened beverage taxes.
Subject(s)
Body Mass Index , Pediatric Obesity , Sugar-Sweetened Beverages , Taxes , Humans , Female , Male , Child , Child, Preschool , Taxes/statistics & numerical data , Sugar-Sweetened Beverages/economics , Sugar-Sweetened Beverages/statistics & numerical data , Adolescent , Washington , Longitudinal Studies , Pediatric Obesity/prevention & controlABSTRACT
Introduction: This study investigates the associations between built environment features and 3-year BMI trajectories in children and adolescents. Methods: This retrospective cohort study utilized electronic health records of individuals aged 5-18 years living in King County, Washington, from 2005 to 2017. Built environment features such as residential density; counts of supermarkets, fast-food restaurants, and parks; and park area were measured using SmartMaps at 1,600-meter buffers. Linear mixed-effects models performed in 2022 tested whether built environment variables at baseline were associated with BMI change within age cohorts (5, 9, and 13 years), adjusting for sex, age, race/ethnicity, Medicaid, BMI, and residential property values (SES measure). Results: At 3-year follow-up, higher residential density was associated with lower BMI increase for girls across all age cohorts and for boys in age cohorts of 5 and 13 years but not for the age cohort of 9 years. Presence of fast food was associated with higher BMI increase for boys in the age cohort of 5 years and for girls in the age cohort of 9 years. There were no significant associations between BMI change and counts of parks, and park area was only significantly associated with BMI change among boys in the age cohort of 5 years. Conclusions: Higher residential density was associated with lower BMI increase in children and adolescents. The effect was small but may accumulate over the life course. Built environment factors have limited independent impact on 3-year BMI trajectories in children and adolescents.
ABSTRACT
Nine in 10 road traffic deaths occur in low- and middle-income countries (LMICs). Despite this disproportionate burden, few studies have examined built environment correlates of road traffic injury in these settings, including in Latin America. We examined road traffic collisions in Bogotá, Colombia, occurring between 2015 and 2019, and assessed the association between neighborhood-level built environment features and pedestrian injury and death. We used descriptive statistics to characterize all police-reported road traffic collisions that occurred in Bogotá between 2015 and 2019. Cluster detection was used to identify spatial clustering of pedestrian collisions. Adjusted multivariate Poisson regression models were fit to examine associations between several neighborhood-built environment features and rate of pedestrian road traffic injury and death. A total of 173,443 police-reported traffic collisions occurred in Bogotá between 2015 and 2019. Pedestrians made up about 25% of road traffic injuries and 50% of road traffic deaths in Bogotá between 2015 and 2019. Pedestrian collisions were spatially clustered in the southwestern region of Bogotá. Neighborhoods with more street trees (RR, 0.90; 95% CI, 0.82-0.98), traffic signals (0.89, 0.81-0.99), and bus stops (0.89, 0.82-0.97) were associated with lower pedestrian road traffic deaths. Neighborhoods with greater density of large roads were associated with higher pedestrian injury. Our findings highlight the potential for pedestrian-friendly infrastructure to promote safer interactions between pedestrians and motorists in Bogotá and in similar urban contexts globally.
Subject(s)
Accidents, Traffic , Built Environment , Pedestrians , Residence Characteristics , Wounds and Injuries , Humans , Colombia/epidemiology , Accidents, Traffic/statistics & numerical data , Accidents, Traffic/mortality , Cross-Sectional Studies , Adult , Male , Female , Pedestrians/statistics & numerical data , Young Adult , Middle Aged , Wounds and Injuries/epidemiology , Adolescent , Residence Characteristics/statistics & numerical data , Child , Child, Preschool , Aged , Environment DesignABSTRACT
BACKGROUND: Opioid overdose mortality in the US has exceeded one million deaths over the last two decades. A regulated opioid supply may help prevent future overdose deaths by reducing exposure to the unregulated opioid supply. We examined the acceptability, delivery model preference, and anticipated effectiveness of different regulated opioid models among people in the Seattle area who inject opioids. METHODS: We enrolled people who inject drugs in the 2022 Seattle-area National HIV Behavior Surveillance (NHBS) survey. Participants were recruited between July and December 2022 using respondent-driven sampling. Participants who reported injecting opioids (N = 453) were asked whether regulated opioids would be acceptable, their preferred model of receiving regulated opioids, and the anticipated change in individual overdose risk from accessing a regulated opioid supply. RESULTS: In total, 369 (81 %) participants who injected opioids reported that a regulated opioid supply would be acceptable to them. Of the 369 who found a regulated opioid supply to be acceptable, the plurality preferred a take-home model where drugs are prescribed (35 %), followed closely by a dispensary model that required no prescription (28 %), and a prescribed model where drugs need to be consumed on site (13 %), a model where no prescription is required and drugs can be accessed in a community setting with a one-time upfront payment was the least preferred model (5 %). Most participants (69 %) indicated that receiving a regulated opioid supply would be "a lot less risky" than their current supply, 20 % said, "a little less risky", 10 % said no difference, and 1 % said a little or a lot more risky. CONCLUSION: A regulated opioid supply would be acceptable to most participants, and participants reported it would greatly reduce their risk of overdose. As overdose deaths continue to increase in Washington state pragmatic and effective solutions that reduce exposure to unregulated drugs are needed.
Subject(s)
Analgesics, Opioid , Opioid-Related Disorders , Substance Abuse, Intravenous , Humans , Male , Adult , Female , Analgesics, Opioid/supply & distribution , Analgesics, Opioid/administration & dosage , Analgesics, Opioid/poisoning , Substance Abuse, Intravenous/epidemiology , Washington , Middle Aged , Opioid-Related Disorders/epidemiology , Opioid-Related Disorders/prevention & control , Opiate Overdose/prevention & control , Opiate Overdose/epidemiology , Young Adult , Drug Overdose/prevention & control , Drug Overdose/mortality , Drug and Narcotic Control/legislation & jurisprudenceABSTRACT
OBJECTIVE: To examine whether built environment and food metrics are associated with glycemic control in people with type 2 diabetes. RESEARCH DESIGN AND METHODS: We included 14,985 patients with type 2 diabetes using electronic health records from Kaiser Permanente Washington. Patient addresses were geocoded with ArcGIS using King County and Esri reference data. Built environment exposures estimated from geocoded locations included residential unit density, transit threshold residential unit density, park access, and having supermarkets and fast food restaurants within 1600-m Euclidean buffers. Linear mixed effects models compared mean changes of HbA1c from baseline at 1, 3 (primary) and 5 years by each built environment variable. RESULTS: Patients (mean age = 59.4 SD = 13.2, 49.5% female, 16.6% Asian, 9.8% Black, 5.5% Latino/Hispanic, 57.1% White, 20% insulin dependent, mean BMI = 32.7±7.7) had an average of 6 HbA1c measures available. Participants in the 1st tertile of residential density (lowest) had a greater decline in HbA1c (-0.42, -0.43, and -0.44 in years 1, 3, and 5 respectively) than those in the 3rd tertile (HbA1c = -0.37 at 1- and 3-years and -0.36 at 5-years; all p-values <0.05). Having any supermarkets within 1600 m of home was associated with a greater decrease in HbA1c at 1-year and 3-years compared to having none (all p-values <0.05). CONCLUSIONS: Lower residential density and better proximity to supermarkets may benefit HbA1c control in people with people with type 2 diabetes. However, effects were small and indicate limited clinical significance.
Subject(s)
Diabetes Mellitus, Type 2 , Humans , Female , Middle Aged , Male , Glycated Hemoglobin , Glycemic Control , Residence Characteristics , FoodABSTRACT
INTRODUCTION: Neighborhoods are complex and multi-faceted. Analytic strategies used to model neighborhoods should reflect this complexity, with the potential to better understand how neighborhood characteristics together impact health. We used latent profile analysis (LPA) to derive a residential neighborhood typology applicable for census tracts across the US. METHODS: From tract-level 2015-2019 American Community Survey (ACS) five-year estimates, we selected five indicators that represent four neighborhood domains: demographic composition, commuting, socioeconomic composition, and built environment. We compared model fit statistics for up to eight profiles to identify the optimal number of latent profiles of the selected neighborhood indicators for the entire US. We then examined differences in national tract-level 2019 prevalence estimates of physical and mental health derived from CDC's PLACES dataset between derived profiles using one-way analysis of variance (ANOVA). RESULTS: The 6-profile LPA model was the optimal categorization of neighborhood profiles based on model fit statistics and interpretability. Neighborhood types were distinguished most by demographic composition, followed by commuting and built environment domains. Neighborhood profiles were associated with meaningful differences in the prevalence of health outcomes. Specifically, tracts characterized as "Less educated non-immigrant racial and ethnic minority active transiters" (n = 3,132, 4%) had the highest poor health prevalence (Mean poor physical health: 18.6 %, SD: 4.30; Mean poor mental health: 19.6 %, SD: 3.85), whereas tracts characterized as "More educated metro/micropolitans" (n = 15, 250, 21%) had the lowest prevalence of poor mental and physical health (Mean poor physical health: 10.6 %, SD: 2.41; Mean poor mental health: 12.4 %, SD: 2.67; p < 0.001). CONCLUSION: LPA can be used to derive meaningful and standardized profiles of tracts sensitive to the spatial patterning of social and built conditions, with observed differences in mental and physical health by neighborhood type in the US.
Subject(s)
Ethnicity , Minority Groups , Humans , Residence Characteristics , Racial GroupsABSTRACT
PRCIS: Residence in a middle-class neighborhood correlated with lower follow-up compared with residence in more affluent neighborhoods. The most common explanations for not following up were the process of making an appointment and lack of symptoms. PURPOSE: To explore which individual-level and neighborhood-level factors influence follow-up as recommended after positive ophthalmic and primary care screening in a vulnerable population using novel methodologies. PARTICIPANTS AND METHODS: From 2017 to 2018, 957 participants were screened for ophthalmic disease and cardiovascular risk factors as part of the Real-Time Mobile Teleophthalmology study. Individuals who screened positive for either ophthalmic or cardiovascular risk factors were contacted to determine whether or not they followed up with a health care provider. Data from the Social Vulnerability Index, a novel virtual auditing system, and personal demographics were collected for each participant. A multivariate logistic regression was performed to determine which factors significantly differed between participants who followed up and those who did not. RESULTS: As a whole, the study population was more socioeconomically vulnerable than the national average (mean summary Social Vulnerability Index score=0.81). Participants whose neighborhoods fell in the middle of the national per capita income distribution had a lower likelihood of follow-up compared with those who resided in the most affluent neighborhoods (relative risk ratio=0.21, P -value<0.01). Participants cited the complicated process of making an eye care appointment and lack of symptoms as the most common reasons for not following up as instructed within 4 months. CONCLUSIONS: Residence in a middle-class neighborhood, difficulty accessing eye care appointments, and low health literacy may influence follow-up among vulnerable populations.
Subject(s)
Ophthalmology , Telemedicine , Humans , Follow-Up Studies , Intraocular Pressure , Risk FactorsABSTRACT
Few studies have used longitudinal imagery of Google Street View (GSV) despite its potential for measuring changes in urban streetscapes characteristics relevant to health, such as neighborhood disorder. Neighborhood disorder has been previously associated with health outcomes. We conducted a feasibility study exploring image availability over time in the Philadelphia metropolitan region and describing changes in neighborhood disorder in this region between 2009, 2014, and 2019. Our team audited Street View images from 192 street segments in the Philadelphia Metropolitan Region. On each segment, we measured the number of images available through time, and for locations where imagery from more than one time point was available, we collected 8 neighborhood disorder indicators at 3 different times (up to 2009, up to 2014, and up to 2019). More than 70% of streets segments had at least one image. Neighborhood disorder increased between 2009 and 2019. Future studies should study the determinants of change of neighborhood disorder using longitudinal GSV imagery.
ABSTRACT
BACKGROUND: Firearm suicide has been more prevalent among males, but age-adjusted female firearm suicide rates increased by 20% from 2010 to 2020, outpacing the rate increase among males by about 8 percentage points, and female firearm suicide may have different contributing circumstances. In the United States, the National Violent Death Reporting System (NVDRS) is a comprehensive source of data on violent deaths and includes unstructured incident narrative reports from coroners or medical examiners and law enforcement. Conventional natural language processing approaches have been used to identify common circumstances preceding female firearm suicide deaths but failed to identify rarer circumstances due to insufficient training data. OBJECTIVE: This study aimed to leverage a large language model approach to identify infrequent circumstances preceding female firearm suicide in the unstructured coroners or medical examiners and law enforcement narrative reports available in the NVDRS. METHODS: We used the narrative reports of 1462 female firearm suicide decedents in the NVDRS from 2014 to 2018. The reports were written in English. We coded 9 infrequent circumstances preceding female firearm suicides. We experimented with predicting those circumstances by leveraging a large language model approach in a yes/no question-answer format. We measured the prediction accuracy with F1-score (ranging from 0 to 1). F1-score is the harmonic mean of precision (positive predictive value) and recall (true positive rate or sensitivity). RESULTS: Our large language model outperformed a conventional support vector machine-supervised machine learning approach by a wide margin. Compared to the support vector machine model, which had F1-scores less than 0.2 for most infrequent circumstances, our large language model approach achieved an F1-score of over 0.6 for 4 circumstances and 0.8 for 2 circumstances. CONCLUSIONS: The use of a large language model approach shows promise. Researchers interested in using natural language processing to identify infrequent circumstances in narrative report data may benefit from large language models.
ABSTRACT
Introduction: This study aimed to evaluate the rate of pediatric emergency department (ED) visits for pedestrian injuries in relation to the enactment of the Complete Streets policy. Methods: The National Complete Streets policies were codified by county and associated with each hospital's catchment area and date of enactment. Pedestrian injury-related ED visits were identified across 40 children's hospitals within the Pediatric Health Information System (PHIS) from 2004 to 2014. We calculated the proportion of the PHIS hospitals' catchment areas covered by any county policy. We used a generalized linear model to assess the impact of the proportion of the policy coverage on the rate of pedestrian injury-related ED visits. Results: The proportion of the population covered by Complete Streets policies increased by 23.9%, and pedestrian injury rates at PHIS hospitals decreased by 29.8% during the study period. After controlling for years, pediatric ED visits for pedestrian injuries did not change with increases in the PHIS catchment population with enacted Complete Streets policies. Conclusion: After accounting for time trends, Complete Streets policy enactment was not related to observed changes in ED visits for pedestrian injuries at PHIS hospitals.
Subject(s)
Pedestrians , Humans , Child , Emergency Service, Hospital , Hospitals, Pediatric , Linear Models , PolicyABSTRACT
PURPOSE: Individual matching in case-control studies improves statistical efficiency over random selection of controls but can lead to selection bias if cases are excluded due to the lack of appropriate controls or residual confounding with less strict matching criteria. We introduce flex matching, an algorithm using multiple rounds of control selection with successively relaxed matching criteria to select controls for cases. METHODS: We simulated exposure-disease relationships in multiple cohort data sets with a range of confounding scenarios and conducted 16,800,000 nested case-control studies, comparing random selection of controls, strict matching, and flex matching. We computed average bias and statistical efficiency in estimates of exposure-disease relationships under each matching strategy. RESULTS: On average, flex matching produced the least biased estimates of exposure-disease associations with the smallest standard errors. Strict matching algorithms that excluded cases for whom matched controls could not be identified produced biased estimates with larger standard errors. Estimates from studies with random assignment of controls were relatively unbiased, but the standard errors were larger than from studies using flex matching. CONCLUSIONS: Flex matching should be considered for case-control designs, especially for biomarker studies where matching on technical artifacts is necessary and maximizing efficiency is a priority.
ABSTRACT
Introduction: : People of lower socioeconomic position (SEP) and people of color (POC) experience higher risks of severe COVID-19, but understanding of these associations beyond the effect of underlying health conditions (UHCs) is limited. Moreover, few studies have focused on young adults, who have had the highest incidence of COVID-19 during much of the pandemic. Methods: : We conducted a retrospective cohort study using electronic health record data from the University of Washington Medicine healthcare system. Our study population included individuals aged 18-39 years who tested positive for SARS-CoV-2 from February 2020 to March 2021. Using regression modeling, we estimated adjusted risk ratios (aRRs) and differences (aRDs) of COVID-19 hospitalization by SEP (using health insurance as a proxy) and race and ethnicity. We adjusted for any UHC to examine these associations beyond the effect of UHCs. Results: Among 3,101 individuals, the uninsured/publicly insured had a 1.9-fold higher risk of hospitalization (aRR [95% CI]=1.9 [1.0, 3.6]) and 9 additional hospitalizations per 1,000 SARS-CoV-2 positive persons (aRD [95% CI]=9 [-1, 20]) compared to the privately insured. Hispanic or Latine, non-Hispanic (NH) Asian, NH Black, and NH Native Hawaiian or Pacific Islander patients had a 1.5-, 2.7-, 1.4-, and 2.1-fold-higher risk of hospitalization (aRR [95% CI]=1.5 [0.7, 3.1]; 2.7 [1.1, 6.5]; 1.4 [0.6, 3.3]; 2.1 [0.5, 9.1]), respectively, compared to NH White patients. Conclusions: Though they should be interpreted with caution given low precision, our findings suggest the increased risk of COVID-19 hospitalization among young adults of lower SEP and young adults of color may be driven by forces other than UHCs, including social determinants of health.
ABSTRACT
Real-world data (RWD) like electronic health records (EHR) has great potential for secondary use by health systems and researchers. However, collected primarily for efficient health care, EHR data may not equitably represent local regions and populations, impacting the generalizability of insights learned from it. We assessed the geospatial representativeness of regions in a large health system EHR data using a spatial analysis workflow, which provides a data-driven way to quantify geospatial representation and identify adequately represented regions. We applied the workflow to investigate geospatial patterns of overweight/obesity and depression patients to find regional "hotspots" for potential targeted interventions. Our findings show the presence of geospatial bias in EHR and demonstrate the workflow to identify spatial clusters after adjusting for bias due to the geospatial representativeness. This work highlights the importance of evaluating geospatial representativeness in RWD to guide targeted deployment of limited healthcare resources and generate equitable real-world evidence.
ABSTRACT
Importance: International Classification of Diseases-coded hospital discharge data do not accurately reflect whether firearm injuries were caused by assault, unintentional injury, self-harm, legal intervention, or were of undetermined intent. Applying natural language processing (NLP) and machine learning (ML) techniques to electronic health record (EHR) narrative text could be associated with improved accuracy of firearm injury intent data. Objective: To assess the accuracy with which an ML model identified firearm injury intent. Design, Setting, and Participants: A cross-sectional retrospective EHR review was conducted at 3 level I trauma centers, 2 from health care institutions in Boston, Massachusetts, and 1 from Seattle, Washington, between January 1, 2000, and December 31, 2019; data analysis was performed from January 18, 2021, to August 22, 2022. A total of 1915 incident cases of firearm injury in patients presenting to emergency departments at the model development institution and 769 from the external validation institution with a firearm injury code assigned according to International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) or International Statistical Classification of Diseases and Related Health Problems, 10th Revision, Clinical Modification (ICD-10-CM), in discharge data were included. Exposures: Classification of firearm injury intent. Main Outcomes and Measures: Intent classification accuracy by the NLP model was compared with ICD codes assigned by medical record coders in discharge data. The NLP model extracted intent-relevant features from narrative text that were then used by a gradient-boosting classifier to determine the intent of each firearm injury. Classification accuracy was evaluated against intent assigned by the research team. The model was further validated using an external data set. Results: The NLP model was evaluated in 381 patients presenting with firearm injury at the model development site (mean [SD] age, 39.2 [13.0] years; 348 [91.3%] men) and 304 patients at the external development site (mean [SD] age, 31.8 [14.8] years; 263 [86.5%] men). The model proved more accurate than medical record coders in assigning intent to firearm injuries at the model development site (accident F-score, 0.78 vs 0.40; assault F-score, 0.90 vs 0.78). The model maintained this improvement on an external validation set from a second institution (accident F-score, 0.64 vs 0.58; assault F-score, 0.88 vs 0.81). While the model showed some degradation between institutions, retraining the model using data from the second institution further improved performance on that site's records (accident F-score, 0.75; assault F-score, 0.92). Conclusions and Relevance: The findings of this study suggest that NLP ML can be used to improve the accuracy of firearm injury intent classification compared with ICD-coded discharge data, particularly for cases of accident and assault intents (the most prevalent and commonly misclassified intent types). Future research could refine this model using larger and more diverse data sets.