The use of count data models in biomedical informatics evaluation research

Evaluation of negative binomial and zero-inflated negative binomial models for the analysis of zero-inflated count data: application to the telemedicine for children with medical complexity trial

BACKGROUND

Two characteristics of commonly used outcomes in medical research are zero inflation and non-negative integers; examples include the number of hospital admissions or emergency department visits, where the majority of patients will have zero counts. Zero-inflated regression models were devised to analyze this type of data. However, the performance of zero-inflated regression models or the properties of data best suited for these analyses have not been thoroughly investigated.

METHODS

We conducted a simulation study to evaluate the performance of two generalized linear models, negative binomial and zero-inflated negative binomial, for analyzing zero-inflated count data. Simulation scenarios assumed a randomized controlled trial design and varied the true underlying distribution, sample size, and rate of zero inflation. We compared the models in terms of bias, mean squared error, and coverage. Additionally, we used logistic regression to determine which data properties are most important for predicting the best-fitting model.

RESULTS

We first found that, regardless of the rate of zero inflation, there was little difference between the conventional negative binomial and its zero-inflated counterpart in terms of bias of the marginal treatment group coefficient. Second, even when the outcome was simulated from a zero-inflated distribution, a negative binomial model was favored above its ZI counterpart in terms of the Akaike Information Criterion. Third, the mean and skewness of the non-zero part of the data were stronger predictors of model preference than the percentage of zero counts. These results were not affected by the sample size, which ranged from 60 to 800.

CONCLUSIONS

We recommend that the rate of zero inflation and overdispersion in the outcome should not be the sole and main justification for choosing zero-inflated regression models. Investigators should also consider other data characteristics when choosing a model for count data. In addition, if the performance of the NB and ZINB regression models is reasonably comparable even with ZI outcomes, we advocate the use of the NB regression model due to its clear and straightforward interpretation of the results.

Building on a novel bootstrapping modelling technique to predict region-wide critical care capacity requirements over the next decade

We have previously described an open-source data-driven modelling technique that has been used to model critical care resource provision as well as expanded to elective surgery and even whole-hospital modelling. Here, we describe the use of this technique to model patient flow and resource use across the West Yorkshire Critical Care Network, with the advantage that recommendations can be made at an individual unit level for future resource provision, taking into account changes in population numbers and demography over the coming decade. We will be using this approach in other regions around the UK to help predict future critical care capacity requirements.

One and done? Exploring linkages between households' intended adaptations to climate-induced floods

As climate change increases the probability and severity of natural hazards, the need for coordinated adaptation at all levels of society intensifies. Governmental-level adaptation measures are essential, but insufficient in the face of growing risks, necessitating complementary action from households. Apprehending the drivers of household adaptation is critical if governments are to stimulate protective behavior effectively. While past work has focused on the behavioral drivers of household adaptation, little attention has been paid to understanding the relationships between adaptation measures themselves-both previously undergone and additionally (planned) intended adaptation(s). Using survey data (N = 4,688) from four countries-the United States, China, Indonesia, and the Netherlands-we utilize protection motivation theory to account for the behavioral drivers of household adaptation to the most devastating climate-driven hazard: flooding. We analyze how past and additionally intended adaptations involving structural modification to one's home affect household behavior. We find that both prior adaptations and additionally intended adaptation have a positive effect on intending a specific adaptation. Further, we note that once links between adaptations are accounted for, the effect that worry has on motivating specific actions, substantially lessens. This suggests that while threat appraisal is important in initially determining if households intend to adapt, it is households' adaptive capacity that determines how. Our analysis reveals that household structural modifications may be nonmarginal. This could indicate that past action and intention to pursue one action trigger intentions for other adaptations, a finding with implications for estimating the speed and scope of household adaptation diffusion.

Count data models for outpatient health services utilisation

BACKGROUND

Count data from the national survey captures healthcare utilisation within a specific reference period, resulting in excess zeros and skewed positive tails. Often, it is modelled using count data models. This study aims to identify the best-fitting model for outpatient healthcare utilisation using data from the Malaysian National Health and Morbidity Survey 2019 (NHMS 2019) and utilisation factors among adults in Malaysia.

METHODS

The frequency of outpatient visits is the dependent variable, and instrumental variable selection is based on Andersen's model. Six different models were used: ordinary least squares (OLS), Poisson regression, negative binomial regression (NB), inflated models: zero-inflated Poisson, marginalized-zero-inflated negative binomial (MZINB), and hurdle model. Identification of the best-fitting model was based on model selection criteria, goodness-of-fit and statistical test of the factors associated with outpatient visits.

RESULTS

The frequency of zero was 90%. Of the sample, 8.35% of adults utilized healthcare services only once, and 1.04% utilized them twice. The mean-variance value varied between 0.14 and 0.39. Across six models, the zero-inflated model (ZIM) possesses the smallest log-likelihood, Akaike information criterion, Bayesian information criterion, and a positive Vuong corrected value. Fourteen instrumental variables, five predisposing factors, six enablers, and three need factors were identified. Data overdispersion is characterized by excess zeros, a large mean to variance value, and skewed positive tails. We assumed frequency and true zeros throughout the study reference period. ZIM is the best-fitting model based on the model selection criteria, smallest Root Mean Square Error (RMSE) and higher R2. Both Vuong corrected and uncorrected values with different Stata commands yielded positive values with small differences.

CONCLUSION

State as a place of residence, ethnicity, household income quintile, and health needs were significantly associated with healthcare utilisation. Our findings suggest using ZIM over traditional OLS. This study encourages the use of this count data model as it has a better fit, is easy to interpret, and has appropriate assumptions based on the survey methodology.

Consumer demand for healthy beverages in the hospitality industry: Examining willingness to pay a premium, and barriers to purchase

This study empirically examines consumer demand for healthy beverages within the hospitality industry. The research investigates sociodemographic and motivational factors that influence consumers’ ‘willingness to pay a premium’ (WTPP) price for healthy beverages using survey data from 1021 consumers in Australia and New Zealand (NZ). Water and juice are rated as representing ‘healthy’ beverages sold by hospitality businesses. Under 2% of respondents consider sugar free drinks as being healthy. Consumers rate a ‘healthy’ beverage as having low/no sugar, natural/no additives, or containing vitamins and minerals. Less than 1% of respondents identify ‘probiotics’ or ‘organic’ as a healthy beverage. Censored Poisson finds consumers who frequently eat out or are younger have higher WTPP. Healthy eating goals increase WTPP, whereas food economizing goals decreases WTPP. Food hedonism goals reduces consumers’ WTPP, and gender differences moderates this relationship. The findings present new insights on consumer behavior and healthy consumption in hospitality.

Application of a Poisson deep neural network model for the prediction of count data in genome-based prediction

Genomic selection (GS) is revolutionizing conventional ways of developing new plants and animals. However, because it is a predictive methodology, GS strongly depends on statistical and machine learning to perform these predictions. For continuous outcomes, more models are available for GS. Unfortunately, for count data outcomes, there are few efficient statistical machine learning models for large datasets or for datasets with fewer observations than independent variables. For this reason, in this paper, we applied the univariate version of the Poisson deep neural network (PDNN) proposed earlier for genomic predictions of count data. The model was implemented with (a) the negative log-likelihood of Poisson distribution as the loss function, (b) the rectified linear activation unit as the activation function in hidden layers, and (c) the exponential activation function in the output layer. The advantage of the PDNN model is that it captures complex patterns in the data by implementing many nonlinear transformations in the hidden layers. Moreover, since it was implemented in Tensorflow as the back-end, and in Keras as the front-end, the model can be applied to moderate and large datasets, which is a significant advantage over previous GS models for count data. The PDNN model was compared with deep learning models with continuous outcomes, conventional generalized Poisson regression models, and conventional Bayesian regression methods. We found that the PDNN model outperformed the Bayesian regression and generalized Poisson regression methods in terms of prediction accuracy, although it was not better than the conventional deep neural network with continuous outcomes.

Q-DEPICT: Qatar Determining Emergency Physician Incidence of COVID-Positive Testing

Despite protective measures such as personal protective equipment (PPE) and a COVID airway management program (CAMP), some emergency physicians will inevitably test positive for COVID. We aim to develop a model predicting weekly numbers of emergency physician COVID converters to aid operations planning.

The data were obtained from the electronic medical record (EMR) used throughout the national healthcare system. Hamad Medical Corporation's internal emergency medicine workforce data were used as a source of information on emergency physician COVID conversion and numbers of emergency physicians completing CAMP training. The study period included the spring and summer months of 2020 and started on March 7 and ran for 21 whole weeks through July 31. Data were extracted from the system's EMR database into a spreadsheet (Excel, Microsoft, Redmond, USA). The statistical software used for all analyses and plots was Stata (version 16.1 MP, StataCorp, College Station, USA). All data definitions were made a priori.

A total of 35 of 250 emergency physicians (14.0%, 95% CI 9.9%–19.9%) converted to a positive real-time reverse transcriptase-polymerase chain reaction (PCR) during the study's 21-week period. Of these. only two were hospitalized for having respiratory-only disease, and none required respiratory support. Both were discharged within a week of admission. The weekly number of newly COVID-positive emergency physicians was zero and was seen in eight of 21 (38.1%) weeks. The peak weekly counts of six emergency physicians with new COVID-positive were seen in week 14. The mean weekly number of newly COVID-positive emergency physicians was 1.7 ± 1.9, and the median was 1 (IQR, 0 to 3).

This study demonstrates that in the State of Qatar's Emergency Department (ED) system, knowing only four parameters allows the reliable prediction of the number of emergency physicians likely to convert COVID PCR tests within the next week. The results also suggest that attention to the details of minimizing endotracheal intubation (ETI) risk can eliminate the expected finding of the association between ETI numbers and emergency physician COVID numbers.

Link between redemption of a medical food pantry voucher and reduced hospital readmissions

This study investigated the relationship between redeeming a voucher at hospital-based Medical Food Pantry (MFP) and hospital readmissions in Greenville, NC. Admitted patients at Vidant Medical Center identified as food insecure were given a voucher to the MFP. A retrospective chart review identified demographic information, type of insurance, voucher provision, and redemption dates, food bag type and number of subsequent hospital readmissions for all patients issued a voucher (n = 542) between June 21, 2018 and July 1, 2019. Negative binomial regression analysis assessed the relationship between readmissions and voucher redemption. Sixty percent of patients receiving a voucher were minority (African American) with an average age of 55. Nearly half (48 percent) had Medicare. Thirty-eight percent of those vouchers that were issued were redeemed, usually within five days. Regression results indicate that the number of readmissions was higher among women and non-whites in the sample relative to men and whites. Those patients who redeemed a food voucher had a seven percent lower likelihood of being readmitted (CI, 0.05–0.27). Food insecure patients who redeemed MFP vouchers had a comparatively lower likelihood of subsequent readmissions. These findings suggest that programs targeting modifiable social determinants of health like food insecurity could improve health outcomes and reduce utilization of the healthcare system.

Risk of Mortality and Readmission among Patients with Pelvic Fracture and Urinary Tract Infection: A Population-Based Cohort Study

Patients with pelvic fractures could encounter various complications during or after treatments. This cohort study investigated the risk of mortality and readmissions in patients with pelvic fractures, with or without urinary tract infections (UTIs), within 30 days following the pelvic fractures. This retrospective cohort study examined claim records from the Longitudinal Health Insurance Database 2000 (LHID2000). We selected patients hospitalized with pelvic fractures between 1997 and 2013 for study. Patients who had index data before 2000 or after 2010 (n = 963), who died before the index date (n = 64), who were aged <18 years (n = 94), or who had a pelvic injury (n = 31) were excluded. In total, the study cohort comprised 1623 adult patients; 115 had UTIs, and 1508 patients without UTIs were used as a comparison cohort. Multivariate analysis with a multiple Cox regression model and Kaplan-Meier survival analysis were performed to analyze the data. Our results showed that the 1-year mortality rate (adjusted hazard ratio [HR]: 2.32; 95% CI: 1.25-4.29) and readmission rate (adjusted HR: 1.72; 95% CI: 1.26-3.34) of the UTI group were significantly higher than those of the non-UTI group. Moreover, the Kaplan-Meier curve for the 1-year follow-up indicated that the UTI group had a higher cumulative risk of both mortality and hospital readmission compared with the non-UTI group. In conclusion, among patients with pelvic fracture, patients with UTI were associated with increased risks of mortality and readmission. Physicians must pay more attention to such patients to prevent UTIs among patients with pelvic fractures during hospitalization and conduct a follow-up after discharge within at least 1 year.

Modelling count data with an excess of zero values applied to childhood bone tumour incidence in Iraq

Bone tumours are rarely found in children and adolescents (0- 19 years old), but there are reports from some provinces in Iraq indicating possible increases in the incidence of childhood bone cancer. Since counts are very low and often zero, or near zero, we fitted zero-inflated Poisson, zero-inflated negative binomial, Poisson hurdle, and negative binomial hurdle regression models to investigate these changes. We used data covering the 2000-2015 period taking age, gender and province into account with the aim of identifying potential health disparities. The results indicate that the zero-inflated Poisson is the most appropriate approach. We also found that, the incidence rate ratio of bone tumours for age groups of 5-9, 10-14 and 15-19 years were 134%, 490% and 723% higher, respectively, compared to the 0-4 year olds. The incidence rate was higher by 49% higher in males compared to females. Compared to 2000-2004, the rate was higher during 2005-2009 and 2010-2015 by 23% and 50%, respectively. In addition, the provinces Al-Muthana and Al-Diwaniyah in the South were found to have a higher incidence rate than other provinces. Join point analysis showed that the age-adjusted incidence rate had a significant, increasing trend, with an average percentage change of 3.1% during 2000-2015. The study suggests that further research into childhood tumours, bone tumours in particular, is needed. Reference to the effect of environmental factors in this group of medical disorders would be of special interest.

Phase III Randomized Controlled Trial of eRAPID: eHealth Intervention During Chemotherapy

PURPOSE

Electronic patient self-Reporting of Adverse-events: Patient Information and aDvice (eRAPID) is an online eHealth system for patients to self-report symptoms during cancer treatment. It provides automated severity-dependent patient advice guiding self-management or medical contact and displays the reports in electronic patient records. This trial evaluated the impact of eRAPID on symptom control, healthcare use, patient self-efficacy, and quality of life (QOL) in a patient population treated predominantly with curative intent.

METHODS

Patients with colorectal, breast, or gynecological cancers commencing chemotherapy were randomly assigned to usual care (UC) or the addition of eRAPID (weekly online symptom reporting for 18 weeks). Primary outcome was symptom control (Functional Assessment of Cancer Therapy-General, Physical Well-Being subscale [FACT-PWB]) assessed at 6, 12, and 18 weeks. Secondary outcomes were processes of care (admissions or chemotherapy delivery), patient self-efficacy, and global quality of life (Functional Assessment of Cancer Therapy–General, EQ5D-VAS, and EORTC QLQ-C30 summary score). Multivariable mixed-effects repeated-measures models were used for analyses. Trial registration: ISRCTN88520246.

RESULTS

Participants were 508 consenting patients (73.6% of 690 eligible) and 55 health professionals. eRAPID compared to UC showed improved physical well-being at 6 ( P = .028) and 12 ( P = .039) weeks and no difference at 18 weeks (primary end point) ( P = .69). Fewer eRAPID patients (47%) had clinically meaningful physical well-being deterioration than UC (56%) at 12 weeks. Subgroup analysis found benefit in the nonmetastatic group at 6 weeks ( P = .0426), but not in metastatic disease. There were no differences for admissions or chemotherapy delivery. At 18 weeks, patients using eRAPID reported better self-efficacy ( P = .007) and better health on EQ5D-VAS ( P = .009). Average patient compliance with weekly symptom reporting was 64.7%. Patient adherence was associated with clinician's data use and improved FACT-PWB at 12 weeks.

CONCLUSION

Real-time monitoring with electronic patient-reported outcomes improved physical well-being (6 and 12 weeks) and self-efficacy (18 weeks) in a patient population predominantly treated with curative intent, without increasing hospital workload.

A Multivariate Poisson Deep Learning Model for Genomic Prediction of Count Data

The paradigm called genomic selection (GS) is a revolutionary way of developing new plants and animals. This is a predictive methodology, since it uses learning methods to perform its task. Unfortunately, there is no universal model that can be used for all types of predictions; for this reason, specific methodologies are required for each type of output (response variables). Since there is a lack of efficient methodologies for multivariate count data outcomes, in this paper, a multivariate Poisson deep neural network (MPDN) model is proposed for the genomic prediction of various count outcomes simultaneously. The MPDN model uses the minus log-likelihood of a Poisson distribution as a loss function, in hidden layers for capturing nonlinear patterns using the rectified linear unit (RELU) activation function and, in the output layer, the exponential activation function was used for producing outputs on the same scale of counts. The proposed MPDN model was compared to conventional generalized Poisson regression models and univariate Poisson deep learning models in two experimental data sets of count data. We found that the proposed MPDL outperformed univariate Poisson deep neural network models, but did not outperform, in terms of prediction, the univariate generalized Poisson regression models. All deep learning models were implemented in Tensorflow as back-end and Keras as front-end, which allows implementing these models on moderate and large data sets, which is a significant advantage over previous GS models for multivariate count data.

Impact of care provider network characteristics on patient outcomes: Usage of social network analysis and a multi-scale community detection

Objective

We assess healthcare provider collaboration and the impact on patient outcomes using social network analysis, a multi-scale community detection algorithm, and generalized estimating equations.

Material and methods

A longitudinal analysis of health claims data of a large employer over a 3 year period was performed to measure how provider relationships impact patient outcomes. The study cohort included 4,230 patients with 167 providers. Social network analysis with a multi-scale community detection algorithm was used to identify groups of healthcare providers more closely working together. Resulting measures of provider collaboration were: 1) degree, 2) betweenness, and 3) closeness centrality. The three patient outcome measures were 1) emergency department visit, 2) inpatient hospitalization, and 3) unplanned hospitalization. Relationships between provider collaboration and patient outcomes were assessed using generalized estimating equations. General practitioner, family practice, and internal medicine were labeled as primary care. Cardiovascular, endocrinologists, etc. were labeled as specialists, and providers such as radiology and social workers were labeled as others.

Results

Higher connectedness (degree) and higher access (closeness) to other providers in the community were significant for reducing inpatient hospitalization and emergency department visits. Patients of specialists (e.g. cardiovascular) and providers specified as others (e.g. social worker) had higher rate of hospitalization and emergency department visits compared to patients of primary care providers.

Conclusion

Application of social network analysis for developing healthcare provider networks can be leveraged by community detection algorithms and predictive modeling to identify providers’ network characteristics and their impacts on patient outcomes. The proposed framework presents multi-scale measures to assess characteristics of healthcare providers and their impact on patient outcomes. This approach can be used by implementation experts for informed decision-making regarding the design of insurance coverage plans, and wellness promotion programs. Health services researchers can use the study approach for assessment of provider collaboration and impacts on patient outcomes.

Understanding the factors influencing health professionals' online voluntary behaviors: Evidence from YiXinLi, a Chinese online health community for mental health

BACKGROUND

Normal users' voluntary behaviors (e.g., knowledge sharing) in virtual communities (VCs) has been well investigated; however, research on health professionals' voluntary behaviors in online health communities (OHCs) is limited.

OBJECTIVE

This paper focuses on OHCs for mental health and aims to explore how intrinsic and extrinsic motivations influence mental health service providers' voluntary behaviors.

METHODS

Based on motivation theory and prior studies, we incorporated technical competence as intrinsic motivation and online reputation and economic rewards as extrinsic motivations, and proposed five hypotheses. We crawled objective data from YiXinLi, a Chinese OHC for mental health, and tested the hypotheses based on the Poisson regression model. All hypotheses are supported.

RESULTS

1) Technical competence, online reputation, and economic rewards positively influence mental health service providers' voluntary behaviors; 2) the interaction effect between technical competence and online reputation negatively influences mental health service providers' voluntary behaviors; 3) the interaction effect between technical competence and economic rewards negatively influences mental health service providers' voluntary behaviors.

CONCLUSIONS

Both intrinsic motivations and extrinsic motivations positively influence mental health service providers' voluntary behaviors, and their interaction effects negatively influence mental health service providers' voluntary behaviors. This study first contributes to the literature on health professionals' voluntary behaviors in OHCs by verifying the positive effect of economic rewards. It then contributes to motivation theory by incorporating a situation where intrinsic motivations and extrinsic motivations could negatively interact.

Migrants’ transnational political engagement in Spain and Italy

International migrants’ cross-border political activities challenge singular notions of national citizenship and political belonging. Yet most sociological studies of migrants’ transnational political engagement are based on single national groups in the USA, and limit themselves to examining how assimilation and contexts of reception determine migrants’ propensity to engage with homeland politics—thereby under theorizing the influence of origin countries. This study moves beyond this approach by recognizing the multi-directionality of migration, and testing the applicability of existing theoretical approaches across two different origins and receiving contexts. We compare a sample of Colombian and Dominican migrants in Spain and Italy, analyzing how contexts in countries of origin, as well as migrants’ social networks across borders, interact with assimilation and contexts of reception to determine migrants’ political transnational engagement. Findings reveal migrants’ transnational political engagement in Spain and Italy appears to be a highly selective process dominated by a small minority of well-educated males from high social status in origin. Findings also suggest immigrant incorporation and transnational political engagement form a dialectical relationship operating at different scales that is simultaneously complementary and contradictory. Contextual conditions in origin countries explain observed much of variation in Colombian and Dominican migrants’ transnational political engagement.

Community Health Center Use After Oregon's Randomized Medicaid Experiment

PURPOSE

There is debate about whether community health centers (CHCs) will experience increased demand from patients gaining coverage through Affordable Care Act Medicaid expansions. To better understand the effect of new Medicaid coverage on CHC use over time, we studied Oregon's 2008 randomized Medicaid expansion (the "Oregon Experiment").

METHODS

We probabilistically matched demographic data from adults (aged 19-64 years) participating in the Oregon Experiment to electronic health record data from 108 Oregon CHCs within the OCHIN community health information network (originally the Oregon Community Health Information Network) (N = 34,849). We performed intent-to-treat analyses using zero-inflated Poisson regression models to compare 36-month (2008-2011) usage rates among those selected to apply for Medicaid vs not selected, and instrumental variable analyses to estimate the effect of gaining Medicaid coverage on use. Use outcomes included primary care visits, behavioral/mental health visits, laboratory tests, referrals, immunizations, and imaging.

RESULTS

The intent-to-treat analyses revealed statistically significant differences in rates of behavioral/mental health visits, referrals, and imaging between patients randomly selected to apply for Medicaid vs those not selected. In instrumental variable analyses, gaining Medicaid coverage significantly increased the rate of primary care visits, laboratory tests, referrals, and imaging; rate ratios ranged from 1.27 (95% CI, 1.05-1.55) for laboratory tests to 1.58 (95% CI, 1.10-2.28) for referrals.

CONCLUSIONS

Our results suggest that use of many different types of CHC services will increase as patients gain Medicaid through Affordable Care Act expansions. To maximize access to critical health services, it will be important to ensure that the health care system can support increasing demands by providing more resources to CHCs and other primary care settings.

Do Pediatric Patients Who Receive Care Across Multiple Health Systems Have Higher Levels of Repeat Testing?

Repetition by clinicians of the same tests for a given patient is common. However, not all repeat tests are necessary for optimal care and can result in unnecessary hardship. Limited evidence suggests that an electronic health record may reduce redundant laboratory testing and imaging by making previous results accessible to physicians. The purpose of this study is to establish a baseline by characterizing repeat testing in a pediatric population and to identify significant risk factors associated with repeated tests, including the impact of using multiple health systems. A population-based retrospective cross-sectional design was used to examine initial and repeat test instances, defined as a second test following an initial test of the same type for the same patient. The study population consisted of 8760 children with 1-25 test claims over a 1-year period. The study setting included all health care service organizations in Minnesota that generated these claims. In all, 17.2% of tests met the definition of repeat test instances, with several risk factors associated with per patient repeat test levels. The incidence of repeat test instances per patient was significantly higher when patients received care from more than 1 health system (adjusted incidence rate ratio 1.4; 95% confidence interval: 1.3-1.5). Repeat test levels are significant in pediatric populations and potentially actionable. Interoperable health information technology may reduce the incidence of repeat test instances in pediatric populations by making prior test results readily accessible. (Population Health Management 2016;19:102-108).

The impact of electronic health records on people with diabetes in three different emergency departments

OBJECTIVE

To evaluate if electronic health records (EHR) with prior clinical information have observable effects for patients with diabetes presenting to emergency departments (ED), we examined measures of quality and resource utilization.

MATERIALS AND METHODS

Retrospective observational studies of patients in three ED (A=5510; B=4393; C=3324) were conducted comparing patients with prior information in the EHR to those without such information. Differences with respect to hospitalization, mortality, length of stay (LOS), and numbers of ED orders for tests, procedures and medications were examined after adjusting for age, gender, race, marital status, comorbidities and for acuity level within each ED.

RESULTS

There were 7% fewer laboratory test orders at one ED and 3% fewer at another; fewer diagnostic procedures were performed at two of the sites. At one site 36% fewer medications were ordered. The odds of being hospitalized were lower for EHR patients at one site and hospital LOS was shorter at two of the sites. EHR patient ED LOS was 18% longer at one site. There was no demonstrable impact of an EHR on mortality. Results varied in magnitude and direction by site.

DISCUSSION

The pattern of significant results varied by ED but tended to reveal reduced utilization and better outcomes for patients although EHR patients' ED LOS was longer at one site.

CONCLUSIONS

The presence of prior information in an EHR may be a valuable adjunct in the care of diabetes patients in ED settings but the pattern of impact may vary from ED to ED.

The impact of electronic health records on care of heart failure patients in the emergency room

OBJECTIVE

To evaluate if electronic health records (EHR) have observable effects on care outcomes, we examined quality and efficiency measures for patients presenting to emergency departments (ED).

MATERIALS AND METHODS

We conducted a retrospective study of 5166 adults with heart failure in three metropolitan EDs. Patients were termed internal if prior information was in the EHR upon ED presentation, otherwise external. Associations of internality with hospitalization, mortality, length of stay (LOS), and numbers of tests, procedures, and medications ordered in the ED were examined after adjusting for age, gender, race, marital status, comorbidities and hospitalization as a proxy for acuity level where appropriate.

RESULTS

At two EDs internals had lower odds of mortality if hospitalized (OR 0.55; 95% CI 0.38 to 0.81 and 0.45; 0.21 to 0.96), fewer laboratory tests during the ED visit (-4.6%; -8.9% to -0.1% and -14.0%; -19.5% to -8.1%) as well as fewer medications (-33.6%; -38.4% to -28.4% and -21.3%; -33.2% to -7.3%). At one of these two EDs, internals had lower odds of hospitalization (0.37; 0.22 to 0.60). At the third ED, internal patients only experienced a prolonged ED LOS (32.3%; 6.3% to 64.8%) but no other differences. There was no association with hospital LOS or number of procedures ordered.

DISCUSSION

EHR availability was associated with salutary outcomes in two of three ED settings and prolongation of ED LOS at a third, but evidence was mixed and causality remains to be determined.

CONCLUSIONS

An EHR may have the potential to be a valuable adjunct in the care of heart failure patients.