A comparison of a multistate inpatient EHR database to the HCUP Nationwide Inpatient Sample

Leveraging multi-site electronic health data for characterization of subtypes: a pilot study of dementia in the N3C Clinical Tenant

Objectives

To provide a foundational methodology for differentiating comorbidity patterns in subphenotypes through investigation of a multi-site dementia patient dataset.

Materials and Methods

Employing the National Clinical Cohort Collaborative Tenant Pilot (N3C Clinical) dataset, our approach integrates machine learning algorithms-logistic regression and eXtreme Gradient Boosting (XGBoost)-with a diagnostic hierarchical model for nuanced classification of dementia subtypes based on comorbidities and gender. The methodology is enhanced by multi-site EHR data, implementing a hybrid sampling strategy combining 65% Synthetic Minority Over-sampling Technique (SMOTE), 35% Random Under-Sampling (RUS), and Tomek Links for class imbalance. The hierarchical model further refines the analysis, allowing for layered understanding of disease patterns.

Results

The study identified significant comorbidity patterns associated with diagnosis of Alzheimer's, Vascular, and Lewy Body dementia subtypes. The classification models achieved accuracies up to 69% for Alzheimer's/Vascular dementia and highlighted challenges in distinguishing Dementia with Lewy Bodies. The hierarchical model elucidates the complexity of diagnosing Dementia with Lewy Bodies and reveals the potential impact of regional clinical practices on dementia classification.

Conclusion

Our methodology underscores the importance of leveraging multi-site datasets and tailored sampling techniques for dementia research. This framework holds promise for extending to other disease subtypes, offering a pathway to more nuanced and generalizable insights into dementia and its complex interplay with comorbid conditions.

Discussion

This study underscores the critical role of multi-site data analyzes in understanding the relationship between comorbidities and disease subtypes. By utilizing diverse healthcare data, we emphasize the need to consider site-specific differences in clinical practices and patient demographics. Despite challenges like class imbalance and variability in EHR data, our findings highlight the essential contribution of multi-site data to developing accurate and generalizable models for disease classification.

Longitudinal Analysis and Latent Growth Modeling of the Modified Hip Dysfunction and Osteoarthritis Outcome Score for Joint Replacement (HOOS-JR)

The Hip Dysfunction and Osteoarthritis Outcome Score for Joint Replacement (HOOS-JR) was developed as a short-form survey to measure progress after total hip arthroplasty (THA). However, the longitudinal validity of the scale structure pertaining to the modified five-item HOOS-JR has not been assessed. Therefore, the purpose of this study was to evaluate the structural validity, longitudinal invariance properties, and latent growth curve (LGC) modeling of the modified five-item HOOS-JR in a large multi-site sample of patients who underwent a THA. A longitudinal study was conducted using data from the Surgical Outcome System (SOS) database. Confirmatory factor analyses (CFAs) were conducted to assess the structural validity and longitudinal invariance across five time points. Additionally, LGC modeling was performed to assess the heterogeneity of the recovery patterns for different subgroups of patients. The resulting CFAs met most of the goodness-of-fit indices (CFI = 0.964-0.982; IFI = 0.965-0.986; SRMR = 0.021-0.035). Longitudinal analysis did not meet full invariance, exceeding the scalar invariance model (CFIDIFF = 0.012; χ2DIFF test = 702.67). Partial invariance requirements were met upon release of the intercept constraint associated with item five (CFIDIFF test = 0.010; χ2DIFF = 1073.83). The equal means model did not pass the recommended goodness-of-fit indices (CFIDIFF = 0.133; χ2DIFF = 3962.49). Scores significantly changed over time, with the highest scores identified preoperatively and the lowest scores identified at 2- and 3-years postoperatively. Upon conclusion, partial scalar invariance was identified within our model. We identified that patients self-report most improvements in their scores within 6 months postoperatively. Females reported more hip disability at preoperative time points and had faster improvement as measured by the scores of the modified five-item HOOS-JR.

Explaining Variability in Electronic Health Record Effort in Primary Care Ambulatory Encounters

BACKGROUND

 Electronic health record (EHR) user interface event logs are fast providing another perspective on the value and efficiency EHR technology brings to health care. Analysis of these detailed usage data has demonstrated their potential to identify EHR and clinical process design factors related to user efficiency, satisfaction, and burnout.

OBJECTIVE

 This study aimed to analyze the event log data across 26 different health systems to determine the variability of use of a single vendor's EHR based on four event log metrics, at the individual, practice group, and health system levels.

METHODS

 We obtained de-identified event log data recorded from June 1, 2018, to May 31, 2019, from 26 health systems' primary care physicians. We estimated the variability in total Active EHR Time, Documentation Time, Chart Review Time, and Ordering Time across health systems, practice groups, and individual physicians.

RESULTS

 In total, 5,444 physicians (Family Medicine: 3,042 and Internal Medicine: 2,422) provided care in a total of 2,285 different practices nested in 26 health systems. Health systems explain 1.29, 3.55, 3.45, and 3.30% of the total variability in Active Time, Documentation Time, Chart Review Time, and Ordering Time, respectively. Practice-level variability was estimated to be 7.96, 13.52, 8.39, and 5.57%, respectively, and individual physicians explained the largest proportion of the variability for those same outcomes 17.09, 27.49, 17.51, and 19.75%, respectively.

CONCLUSION

 The most variable physician EHR usage patterns occurs at the individual physician level and decreases as you move up to the practice and health system levels. This suggests that interventions to improve individual users' EHR usage efficiency may have the most potential impact compared with those directed at health system or practice levels.

Age-Stratified Sex Differences in Heart Failure With Preserved Ejection Fraction Among Adult Hospitalizations

BACKGROUND

There is evidence that heart failure with preserved ejection fraction (HFpEF)-related hospitalizations are increasing in the United States. However, there is a lack of knowledge about HFpEF-related hospitalizations among younger adults.

OBJECTIVE

The aims of this study were to perform a retrospective analysis using the Nationwide Inpatient Sample and to examine age-stratified sex differences in the prevalence, correlates, and outcomes of HFpEF-related hospitalization across the adult life span.

METHOD

Using the Nationwide Inpatient Sample (2002-2014), patient and hospital characteristics were determined. Joinpoint regression was used to describe age-stratified sex differences in the annual average percent change of hospitalizations with HFpEF. Survey logistic regression was used to estimate adjusted odds ratios representing the association of sex with HFpEF-related hospitalization and in-hospital mortality.

RESULTS

There were 8 599 717 HFpEF-related hospitalizations (2.43% of all hospitalizations). Women represented the majority (5 459 422 [63.48%]) of HFpEF-related adult hospitalizations, compared with men (3 140 295 [36.52%]). Compared with men younger than 50 years, women within the same age group were 6% to 28% less likely to experience HFpEF-related hospitalization. Comorbidities such as hypertensive heart disease, renal disease, hypertension, obstructive sleep apnea, atrial fibrillation, obesity, anemia, and pulmonary edema explained a greater proportion of the risk of HFpEF-related hospitalization in adults younger than 50 years than in adults 50 years or older.

CONCLUSION

Before the age of 50 years, women exhibit lower HFpEF-related hospitalization than men, a pattern that reverses with advancing age. Understanding and addressing the factors contributing to these sex-specific differences can have several potential implications for improving women's cardiovascular health.

Increased Risk of In-Hospital Complications and Costs After Total Hip Arthroplasty for Primary and Secondary Osteonecrosis

BACKGROUND

An increasing proportion of patients are undergoing total hip arthroplasty (THA) for osteonecrosis (ON). Comorbid conditions and surgical risk factors are known to be greater in ON patients compared with patients who have osteoarthritis (OA) alone. The purpose of our study was to quantify the specific in-hospital complications and resource utilization associated with patients undergoing THA for ON versus OA.

METHODS

A large national database was queried to identify patients undergoing primary THA from January 1, 2016 to December 31, 2019. A total of 1,383,880 OA, 21,080 primary ON, and 54,335 secondary ON patients were identified. Demographics, in-hospital complications, costs, lengths of stay, and discharge dispositions for primary and secondary ON cohorts were compared to OA only. Age, race, ethnicity, comorbidities, Medicaid, and income status were controlled with binary logistic regression analyses.

RESULTS

The ON patients were often younger, African American or Hispanic, and had more comorbidities. Those undergoing THA for primary and secondary ON had a significantly higher risk of perioperative complications, including myocardial infarction, postoperative blood transfusion, and intraoperative bleeding. Total hospital costs and lengths of stay were significantly higher for both primary ON and secondary ON and both cohorts were less likely to be discharged home.

CONCLUSION

While rates of most complications have decreased over recent decades in ON patients undergoing THA, the ON patients still have worse outcomes even when controlling for comorbidity differences. Bundled payment systems and perioperative management strategies for these different patient cohorts should be considered separately.

The Association of the Lactate-Albumin Ratio With Mortality and Multiple Organ Dysfunction in PICU Patients

OBJECTIVES

To compare the relative associations of lactate, albumin, and the lactate-albumin ratio (LAR) measured early in disease course against mortality and prevalence of multiple organ dysfunction syndrome (MODS) in a general sample of critically ill pediatric patients.

DESIGN

Retrospective analysis of the Health Facts (Cerner Corporation, Kansas City, MO) national database.

SETTING

U.S. hospitals with PICUs.

PATIENTS

Children admitted to the ICU ( n = 648) from 2009 to 2018 who had lactate and albumin measured within 6 hours of admission.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

A total of 648 admissions were included, with an overall mortality rate of 10.8% ( n = 70) and a MODS prevalence of 29.3% ( n = 190). Compared with survivors, deaths had higher initial lactates (7.3 mmol/L [2.6-11.7 mmol/L] vs 1.9 mmol/L [1.2-3.1 mmol/L]; p < 0.01), lower initial albumins (3.3 g/dL [2.7-3.8 g/dL] vs 4.2 g/dL [3.7-4.7 g/dL]; p < 0.01), and higher LARs (2.2 [1.0-4.2] vs 0.5 [0.3-0.8]; p < 0.01), with similar trends in patients with MODS versus those without MODS. LAR demonstrated a higher odds ratio (OR) for death than initial lactate alone (2.34 [1.93-2.85] vs 1.29 [1.22-1.38]) and a higher OR for MODS than initial lactate alone (2.10 [1.73-2.56] vs 1.22 [1.16-1.29]). Area under the receiver operating characteristic (AUROC) curve of LAR for mortality was greater than initial lactate (0.86 vs 0.82; p < 0.01). The LAR AUROC for MODS was greater than the lactate AUROC (0.71 vs 0.66; p < 0.01). Trends of lactate, albumin, and LAR for mortality were consistent across several diagnostic subgroups (trauma, primary respiratory failure, toxicology), but not all.

CONCLUSIONS

LAR measured early in the course of critical illness is significantly associated with mortality and development of MODS when compared with initial lactate or initial albumin alone in critically ill pediatric patients.

Percutaneous Left Atrial Appendage Closure Among Patients With Diabetes (Insights from a National Database)

Atrial fibrillation is a major risk factor for stroke. Left atrial appendage closure (LAAC) has emerged as an alternative to anticoagulation for patients with high risk of bleeding. Diabetes mellitus (DM) is associated with adverse events after cardiac procedures. We sought to compare procedural and hospital outcomes in patients who underwent LAAC with and without DM. The Nationwide Inpatient Database was queried for patients with atrial fibrillation who underwent LAAC between January 1, 2016, and December 31, 2019. The primary outcome was all adverse events that included in-hospital death, acute myocardial infarction, cardiac arrest, stroke, pericardial effusion, pericardial tamponade, pericardiocentesis, pericardial window, and postprocedural hemorrhage requiring blood transfusion. Analysis included 62,220 patients who underwent LAAC from 2016 to 2019; 34.9% of patients had DM. There was a slight increase in the percentage of patients who underwent LAAC who had DM during the study period, from 29.92% to 34.93%. In unadjusted and adjusted analysis, there was no significant difference in all adverse events between patients with and without DM who underwent LAAC (9.18% vs 8.77%, respectively, adjusted p = 0.63), and no difference in length of stay. Patients with DM have higher risk of acute kidney injury (3.75 vs 1.96%, p <0.001). This nationwide retrospective study demonstrates that DM is not associated with an increase in adverse event rates in patients who underwent LAAC.

Trimethoprim-sulfamethoxazole resistance patterns among Staphylococcus aureus in the United States, 2012-2018

We reviewed trimethoprim-sulfamethoxazole antibiotic susceptibility testing data among Staphylococcus aureus using 3 national inpatient databases. In all 3 databases, we observed an increases in the percentage of methicillin-resistant Staphylococcus aureus that were not susceptible to trimethoprim-sulfamethoxazole. Providers should select antibiotic regimens based on local resistance patterns and should report changes to the public health department.

Digital Health Data Quality Issues: Systematic Review

BACKGROUND

The promise of digital health is principally dependent on the ability to electronically capture data that can be analyzed to improve decision-making. However, the ability to effectively harness data has proven elusive, largely because of the quality of the data captured. Despite the importance of data quality (DQ), an agreed-upon DQ taxonomy evades literature. When consolidated frameworks are developed, the dimensions are often fragmented, without consideration of the interrelationships among the dimensions or their resultant impact.

OBJECTIVE

The aim of this study was to develop a consolidated digital health DQ dimension and outcome (DQ-DO) framework to provide insights into 3 research questions: What are the dimensions of digital health DQ? How are the dimensions of digital health DQ related? and What are the impacts of digital health DQ?

METHODS

Following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines, a developmental systematic literature review was conducted of peer-reviewed literature focusing on digital health DQ in predominately hospital settings. A total of 227 relevant articles were retrieved and inductively analyzed to identify digital health DQ dimensions and outcomes. The inductive analysis was performed through open coding, constant comparison, and card sorting with subject matter experts to identify digital health DQ dimensions and digital health DQ outcomes. Subsequently, a computer-assisted analysis was performed and verified by DQ experts to identify the interrelationships among the DQ dimensions and relationships between DQ dimensions and outcomes. The analysis resulted in the development of the DQ-DO framework.

RESULTS

The digital health DQ-DO framework consists of 6 dimensions of DQ, namely accessibility, accuracy, completeness, consistency, contextual validity, and currency; interrelationships among the dimensions of digital health DQ, with consistency being the most influential dimension impacting all other digital health DQ dimensions; 5 digital health DQ outcomes, namely clinical, clinician, research-related, business process, and organizational outcomes; and relationships between the digital health DQ dimensions and DQ outcomes, with the consistency and accessibility dimensions impacting all DQ outcomes.

CONCLUSIONS

The DQ-DO framework developed in this study demonstrates the complexity of digital health DQ and the necessity for reducing digital health DQ issues. The framework further provides health care executives with holistic insights into DQ issues and resultant outcomes, which can help them prioritize which DQ-related problems to tackle first.

Association of depression and anxiety disorders with outcomes after revascularization in chronic limb-threatening ischemia hospitalizations nationwide

OBJECTIVE

Patients with chronic limb-threatening ischemia (CLTI), the end stage of peripheral artery disease, often present with comorbid depression and anxiety disorders. The prevalence of these comorbidities in the inpatient context over time, and their association with outcomes after revascularization and resource usage is unknown.

METHODS

Using the 2011 to 2017 National Inpatient Sample, two cohorts were created-CLTI hospitalizations with endovascular revascularization and CLTI hospitalizations with surgical revascularization. Within each cohort, the annual prevalence of depression and anxiety disorder diagnoses was determined, and temporal trends were evaluated using the Cochran-Mantel-Haenszel test. Hierarchical multivariable logistic and linear regression analyses were used to examine the association of depression and anxiety disorder diagnoses with inpatient major amputation, mortality, length of stay (LOS), and cost, adjusting for illness severity, comorbidities, and potential bias in the documentation of depression and anxiety disorder diagnoses stratified by patient sociodemographic data.

RESULTS

Across the study period were a total of 245,507 CLTI-related hospitalizations with endovascular revascularization and 138,922 with surgical revascularization. Hospitalizations with a depression or anxiety disorder diagnosis increased from 10.8% in 2011 to 15.3% in 2017 in the endovascular revascularization cohort and from 11.7% in 2011 to 14.4% in 2017 in the surgical revascularization cohort (P_trend < .001). In the endovascular revascularization cohort, depression was associated with higher odds of major amputation (odds ratio, 1.15; 95% confidence interval, 1.03-1.30). In addition, depression (9 vs 8 days [P < .001]; $105,754 vs $102,481 [P = .018]) and anxiety disorder (9 vs 8 days [P < .001]; $109,496 vs $102,324 [P < .001]) diagnoses were associated with a longer median LOS and higher median costs. In the surgical revascularization cohort, depression was associated with a higher odds of major amputation (odds ratio, 1.33; 95% confidence interval, 1.13-1.58) and a longer LOS (median, 9 vs 9 days; P = .004).

CONCLUSIONS

Depression and anxiety disorder diagnoses have become increasingly prevalent among CLTI hospitalizations including revascularizations. When present, these psychiatric comorbidities are associated with an increased risk of amputation and greater resource usage.

Examining Geographic Variation of Opioid Use Disorder Encounters in the USA

OBJECTIVES

The objectives were (1) to characterize patient encounters of opioid use disorder (OUD) using Health Facts^® database; and (2) to identify geographic variation, patient characteristics, and facility characteristics impacting patients' reduced OUD encounters over time.

METHODS

Patient encounters were included if the patient (1) was 18 years old or greater; (2) had an index encounter; (3) survived at least 30 days after the discharge. The OUD encounter was based on ICD-10 codes. The date at which a patient first had an OUD encounter was the index date. For the first objective, OUD encounters were described according to patient characteristics, facility characteristics, and geographic variation. Patient characteristics were age, gender, marital status, race, health insurance coverage, discharge disposition, and patient type. Facility characteristics were care setting, medical specialty, census region, census division, urban vs. rural, acute vs. non-acute, and teaching hospital status. For the second objective, patients were examined 1 year prior to through 1 year after the index date. A logistic regression was used to determine the likelihood of reduced OUD encounters over time, conditional upon geographic variation, patient characteristics, and facility characteristics.

RESULTS

A total of 265,643 OUD encounters were identified. East South Central was associated with the highest population-based rate of OUD among nine census divisions. In the logistic regression (n = 10,762), discharged to home, outpatient, emergency room, psychiatry, East North Central, West North Central, and urban areas were significant positive predictors for reduced OUD encounters over time, whereas age and Mountain were significant negative predictors.

CONCLUSIONS

East South Central was associated with the highest population-based rate of OUD. Compared with East South Central, East North Central and West North Central had a significantly positive impact on fewer encounters of OUD over time.

Palliative Care Use in Patients With Acute Myocardial Infarction and Do-Not-Resuscitate Status From a Nationwide Inpatient Cohort

OBJECTIVE

To examine the predictors, treatments, and outcomes of the use of palliative care in patients hospitalized with acute myocardial infarction (AMI) who had a do-not-resuscitate (DNR) order.

PATIENTS AND METHODS

Using the National (Nationwide) Inpatient Sampling database for 2015-2018, we examined the predictors, in-hospital procedures, and outcomes of palliative care recipients among patients with AMI who had a DNR order.

RESULTS

We identified 339,270 admissions with AMI that had a DNR order, including patients who received palliative care (n=113,215 [33.4%]). Compared with patients who did not receive palliative care, these patients were more frequently younger (median age, 81 vs 83 years; P<.001), were less likely to be female (50.9% [57,626 of 113,215] vs 54.7% [123,652 of 226,055]; P<.001), and were more likely to present with cardiac arrest (11.6% [13,133 of 113,215] vs 6.9% [15,598 of 226,055]; P<.001). Patients were more likely to receive palliative care at a large (odds ratio [OR], 1.47; 95% CI, 1.44 to 1.50) or teaching (OR, 2.10; 95% CI, 2.04 to 2.16) hospitals compared with small or rural ones. Patients receiving palliative care were less likely to be treated invasively, with reduced rates of invasive coronary angiography (OR, 0.46; 95% CI, 0.45 to 0.47) and percutaneous coronary intervention (OR, 0.47; 95% CI, 0.45 to 0.48), and were more likely to die in the hospital (52.4% [59,325 of 113,215] vs 22.9% [51,766 of 226,055]).

CONCLUSION

In patients who had a DNR status and were hospitalized and received a diagnosis of AMI, only one-third received palliative care.

Machine Learning Approach to Predict In-Hospital Mortality in Patients Admitted for Peripheral Artery Disease in the United States

Background Peripheral artery disease (PAD) affects >10 million people in the United States. PAD is associated with poor outcomes, including premature death. Machine learning (ML) has been increasingly used on big data to predict clinical outcomes. This study aims to develop ML models to predict in-hospital mortality in patients hospitalized for PAD based on a national database. Methods and Results Inpatient hospitalization data were obtained from the 2016 to 2019 National Inpatient Sample. A total of 150 921 inpatients were identified with a primary diagnosis of PAD and PAD-related procedures using codes of the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) and International Classification of Diseases, Tenth Revision, Procedure Coding System (ICD-10-PCS). Four ML models, including logistic regression, random forest, light gradient boosting, and extreme gradient boosting models, were trained to predict the risk of in-hospital death based on a selection of variables, including patient characteristics, comorbidities, procedures, and hospital-related factors. In-hospital mortality occurred in 1.8% of patients. The performance of the 4 models was comparable, with the area under the receiver operating characteristic curve ranging from 0.83 to 0.85, sensitivity of 77% to 82%, and specificity of 72% to 75%. These results suggest adequate predictability for clinical decision-making. In all 4 models, the total number of diagnoses and procedures, age, endovascular revascularization procedure, congestive heart failure, diabetes, and diabetes with complications were critical predictors of in-hospital mortality. Conclusions This study demonstrates the feasibility of ML in predicting in-hospital mortality in patients with a primary PAD diagnosis. Findings highlight the potential of ML models in identifying high-risk patients for poor outcomes and guiding personalized intervention.

Mortality in ST-segment elevation myocardial infarction patients without standard modifiable risk factors: A race disaggregated analysis

Background

Individuals who present with STEMI without the standard cardiovascular risk factors (SMuRFs) of diabetes, hypercholesterolemia, hypertension, and smoking, coined SMuRF-less are not uncommon. Little is known about their outcomes as a cohort and how they differ by race.

Methods & Results

We identified 431,615 admissions with STEMI in the National Inpatient Sample (NIS) database 2015–2018, including patients with ≥ 1 SMuRF (n = 369,870) and those who were SMuRF-less (n = 234,745). SMuRF-less patients presented at a similar age (median age 63y vs 63y), were less likely to be female (33.6 % vs 34.6 %) and were almost twice as likely to present as a cardiac arrest (13.7 % vs 7.0 %), than those with ≥ 1 SMuRFs. SMuRF-less patients were less frequently in receipt of ICA (71.3 % vs 83.8 %) and PCI (58.0 % vs 72.2 %) compared to those with ≥ 1 SMuRF. Our race disaggregated analysis showed ethnic minority SMuRF-less patients were less likely than White patients to receive ICA and PCI, which was most apparent in Black patients with reduced odds of ICA (OR: 0.47, 95 % CI: 0.43–0.52) and PCI (OR: 0.46, 95 % CI: 0.52–0.50). Similarly, in ethnic minority subgroups within the SMuRF-less cohort, mortality and MACCE were significantly higher than in White patients. This was most profound in Black patients with in-hospital mortality (OR: 1.90, 95 % CI: 1.72–2.09) and MACCE (OR: 1.63, 95 % CI: 1.49–1.78) compared to White patients.

Conclusion

Ethnic Minority SMuRF-less patients were less likely than White SMuRF-less patients to receive ICA and PCI and had worse mortality outcomes.

Machine Learning in Infectious Disease for Risk Factor Identification and Hypothesis Generation: Proof of Concept Using Invasive Candidiasis

Background

Machine learning (ML) models can handle large datasets without assuming underlying relationships and can be useful for evaluating disease characteristics; yet, they are more commonly used for predicting individual disease risk rather than identifying factors at the population level. We offer a proof of concept applying random forest (RF) algorithms to Candida-positive hospital encounters in an electronic health record database of patients in the U.S.

Methods

Candida-positive encounters were extracted from the Cerner HealthFacts database; invasive infections were laboratory positive sterile site Candida infections. Features included demographics, admission source, care setting, physician specialty, diagnostic and procedure codes, and medications received prior to the first positive Candida culture. We used RF to assess risk factors for three outcomes: any invasive candidiasis (IC) vs non-IC, within-species IC vs non-IC (e.g. invasive C. glabrata vs non-invasive C. glabrata), and between-species IC (e.g. invasive C. glabrata vs all other IC).

Results

14 of 169 (8%) variables were consistently identified as important features in the ML models. When evaluating within-species IC, for example invasive C. glabrata vs non-invasive C. glabrata, we identified known features like central venous catheters, ICU stay, and gastrointestinal operations. In contrast, important variables for invasive C. glabrata vs all other IC included renal disease and medications like diabetes therapeutics, cholesterol medications, and antiarrhythmics.

Conclusions

Known and novel risk factors for IC were identified using ML, demonstrating the hypotheses generating utility of this approach for infectious disease conditions about which less is known, specifically at the species-level or for rarer diseases.

Dynamic Mortality Risk Predictions for Children in ICUs: Development and Validation of Machine Learning Models

OBJECTIVES

Assess a machine learning method of serially updated mortality risk.

DESIGN

Retrospective analysis of a national database (Health Facts; Cerner Corporation, Kansas City, MO).

SETTING

Hospitals caring for children in ICUs.

PATIENTS

A total of 27,354 admissions cared for in ICUs from 2009 to 2018.

INTERVENTIONS

None.

MAIN OUTCOME

Hospital mortality risk estimates determined at 6-hour time periods during care in the ICU. Models were truncated at 180 hours due to decreased sample size secondary to discharges and deaths.

MEASUREMENTS AND MAIN RESULTS

The Criticality Index, based on physiology, therapy, and care intensity, was computed for each admission for each time period and calibrated to hospital mortality risk (Criticality Index-Mortality [CI-M]) at each of 29 time periods (initial assessment: 6 hr; last assessment: 180 hr). Performance metrics and clinical validity were determined from the held-out test sample (n = 3,453, 13%). Discrimination assessed with the area under the receiver operating characteristic curve was 0.852 (95% CI, 0.843-0.861) overall and greater than or equal to 0.80 for all individual time periods. Calibration assessed by the Hosmer-Lemeshow goodness-of-fit test showed good fit overall (p = 0.196) and was statistically not significant for 28 of the 29 time periods. Calibration plots for all models revealed the intercept ranged from--0.002 to 0.009, the slope ranged from 0.867 to 1.415, and the R2 ranged from 0.862 to 0.989. Clinical validity assessed using population trajectories and changes in the risk status of admissions (clinical volatility) revealed clinical trajectories consistent with clinical expectations and greater clinical volatility in deaths than survivors (p < 0.001).

CONCLUSIONS

Machine learning models incorporating physiology, therapy, and care intensity can track changes in hospital mortality risk during intensive care. The CI-M's framework and modeling method are potentially applicable to monitoring clinical improvement and deterioration in real time.

Estimating the prevalence of neurosurgical interventions in adults with spina bifida using the Health Facts data set: implications for transition planning and the development of adult clinics

OBJECTIVE

As the care of patients with spina bifida continues to evolve, life expectancy is increasing, leading to a critical need for transition planning from pediatric-based to adult-based care. The burden of neurosurgical care for adults with spina bifida remains unknown. In this study, the authors sought to use a large national data set to estimate the prevalence of neurosurgical interventions in adults with spina bifida.

METHODS

This study utilized Health Facts, which is a de-identified proprietary data set abstracted from all Cerner electronic health records. It includes 69 million unique patients with > 500 million encounters in 580 centers. Validation, technical exclusions, and data filters were applied to obtain an appropriate cohort of patients. The ICD-9 and ICD-10 codes for all types of spinal dysraphism, as well as the Current Procedural Terminology (CPT) codes for hydrocephalus procedures, spinal cord untethering, and Chiari decompression, were queried and records were retrieved. Demographic variables along with differences in age groups and temporal trends were analyzed.

RESULTS

Overall, 24,764 unique patients with ≥ 1 encounter with a spinal dysraphism diagnosis between 2000 and 2017 were identified. The pediatric cohort included 11,123 patients with 60,027 separate encounters, and the adult cohort included 13,641 patients with 41,618 separate encounters. The proportion of females was higher in the adult (62.9%) than in the pediatric (51.4%) cohort. Annual encounters were stable from 2 to 18 years of age, but then decreased by approximately half with a precipitous drop after age 21 years. The sex distribution of adults and children who underwent procedures was similar (54.6% female adults vs 52.4% female children). Surgical interventions in adults were common. Between 2013 and 2017, there were 4913 procedures for hydrocephalus, with 2435 (49.6%) adult patients. Similarly, 273 (33.3%) of the 819 tethered cord procedures were performed in adults, as were 307 (32.9%) of 933 Chiari decompressions.

CONCLUSIONS

The Health Facts database offered another option for studying care delivery and utilization in patients aging with spina bifida. The median age of this population has now reached early adulthood, and a significant number of neurosurgical procedures were performed in adults. An abrupt drop in the rate of encounters occurred at 21 years of age, possibly reflecting transition issues such as access-to-care problems and lack of coordinated care.

High Rate of Emergency Department Care in Youth With Abdominal Pain-Associated Functional Gastrointestinal Disorders

OBJECTIVES

The primary objective was to describe patterns of care delivery locations in youth with abdominal pain-associated functional gastrointestinal disorders (AP-FGID) and assess for differences in patterns of care delivery by sex and race. A secondary objective was to describe cost variability within the emergency department (ED).

METHODS

Data were obtained using a large, single-vendor database that extracts and deidentifies data from the electronic health record across the outpatient, ED, and inpatient continuum of care. We evaluated patients 8 to 17 years of age seen over an 8-year period for a priority 1 diagnosis of an AP-FGID. Data collected included age, sex, race, encounter location, and total cost of ED encounters. We specifically assessed how often patients seen in the ED were also seen in outpatient or inpatient settings.

RESULTS

A total of 53,750 patients (64% female; mean age, 13.3 ± 2.8 years) were identified and assessed. The most common location of care was the ED (48.8%) followed by the outpatient setting (46.2%). Of patients seen for a priority 1 AP-FGID diagnosis in the ED, only 3.7% were seen for a priority 1 diagnosis in the outpatient setting, and only 1% were seen in an inpatient setting. Overall, females received 42.5% of their care and males received 44.8% of their care in the ED. The overall rate of ED care was 66.9% for Hispanic, 61.5% for African American, 55.1% for Asian, 46.6% for Native American, and 36.9% for Caucasian patients.

CONCLUSIONS

The ED is the most common location for care for youth with AP-FGIDs and, for the majority, seems to be the only location. This seems to be particularly true for Hispanic and African American patients. Given the often complex psychosocial needs of this patient group, processes need to be developed to transition these patients into the outpatient setting, ideally to programs specializing in chronic pain.

Epidemiologic evaluation of inhaled nitric oxide use among neonates with gestational age less than 35 weeks

BACKGROUND AND OBJECTIVES

The use of inhaled nitric oxide (iNO) in +late preterm and term infants with pulmonary hypertension is Food and Drug Administration (FDA) approved and has improved outcomes and survival. iNO use is not FDA approved for preterm infants and previous studies show no mortality benefit. The objectives were 1) to determine the usage of iNO among preterm neonates <35 weeks before and after the 2010 National Institutes of Health consensus statement and 2) to evaluate characteristics and outcomes among preterm neonates who received iNO.

METHODS

This is a population-based cross-sectional study. Billing and procedure codes were used to determine iNO usage. Data were queried from the National Inpatient Sample from 2004 to 2016. Neonates were included if gestational age was <35 weeks. The epochs were spilt into 2004-2010 (Epoch 1) and 2011-2016 (Epoch 2). Prevalence of iNO use, mortality, bronchopulmonary dysplasia (BPD), intraventricular hemorrhage, length of stay, mechanical ventilation, and cost of hospitalization.

RESULTS

There were 4865 preterm neonates <35 weeks who received iNO. There was a significant increase in iNO use during Epoch 2 (p < 0.001). There was significantly higher use in Epoch 2 among neonates small for gestational age (SGA) 2.3% versus 7.2%, congenital heart disease (CHD) 11.1% versus 18.6%, and BPD 35.2% versus 46.8%. Mortality was significantly lower in Epoch 2 19.8% versus 22.7%.

CONCLUSION

Usage of iNO was higher after the release of the consensus statement. The increased use of iNO among preterm neonates may be targeted at specific high-risk populations such as SGA and CHD neonates. There was lower mortality in Epoch 2; however, the cost was doubled.

Standardized Health data and Research Exchange (SHaRE): promoting a learning health system

Aggregate de-identified data from electronic health records (EHRs) provide a valuable resource for research. The Standardized Health data and Research Exchange (SHaRE) is a diverse group of US healthcare organizations contributing to the Cerner Health Facts (HF) and Cerner Real-World Data (CRWD) initiatives. The 51 facilities at the 7 founding organizations have provided data about more than 4.8 million patients with 63 million encounters to HF and 7.4 million patients and 119 million encounters to CRWD. SHaRE organizations unmask their organization IDs and provide 3-digit zip code (zip3) data to support epidemiology and disparity research. SHaRE enables communication between members, facilitating data validation and collaboration as we demonstrate by comparing imputed EHR module usage to actual usage. Unlike other data sharing initiatives, no additional technology installation is required. SHaRE establishes a foundation for members to engage in discussions that bridge data science research and patient care, promoting the learning health system.

Practice patterns and outcomes associated with intravenous albumin in patients with cirrhosis and acute kidney injury

BACKGROUND & AIMS

Guidelines recommend albumin as the plasma-expander of choice for acute kidney injury (AKI) in cirrhosis. However, the impact of these recommendations on patient outcomes remains unclear. We aimed to determine the practice-patterns and outcomes associated with albumin use in a large, nationwide-US cohort of hospitalized cirrhotics with AKI.

METHODS

A retrospective cohort study was performed in hospitalized cirrhotics with AKI using Cerner-Health-Facts database from January 2009 to March 2018. 6786 were included for analysis on albumin-practice-patterns, and 4126 had available outcomes data. Propensity-score-adjusted model was used to determine the association between albumin use, AKI-recovery and in-hospital survival.

RESULTS

Median age was 61-years (60% male, 70% white), median serum-creatinine was 1.8 mg/dL and median Model for End-stage Liver Disease Sodium (MELD-Na) score was 24. Albumin was given to 35% of patients, of which 50% received albumin within 48-hours of AKI-onset, and 17% received appropriate weight-based dosing. Albumin was used more frequently in patients with advanced complications of cirrhosis, higher MELD-Na scores and patients admitted to urban-teaching hospitals. After propensity-matching and multivariable adjustment, albumin use was not associated with AKI-recovery (odds ratio [OR] 0.70, 95% confidence-interval [CI]: 0.59-1.07, P = .130) or in-hospital survival (OR 0.76 [95% CI: 0.46-1.25], P = .280), compared with crystalloids. Findings were unchanged in subgroup analyses of patients with varying cirrhosis complications and disease severity.

CONCLUSIONS

USA hospitalized patients with cirrhosis and AKI frequently do not receive intravenous albumin, and albumin use was not associated with improved clinical outcomes. Prospective randomised trials are direly needed to evaluate the impact of albumin in cirrhotics with AKI.

Association of treatments for acute appendicitis with pregnancy outcomes in the United States from 2000 to 2016: Results from a multi-level analysis

Background

Open appendectomy, laparoscopic appendectomy, and non-surgical treatment are three options to treat acute appendicitis during pregnancy. Previous studies on the association of different treatment methods for acute appendicitis with pregnancy outcomes have been limited by small sample sizes and residual confounding, especially with respect to hospital-level factors. This study aimed to investigate the association of treatment method for acute appendicitis with pregnancy outcomes using a multi-level analysis.

Methods

A retrospective cohort study was conducted based on a large electronic health records database in the United States during the period 2000 to 2016. All pregnancies diagnosed with acute appendicitis and treated in participating hospitals during the study period were included. We conducted multi-level hierarchical logistic regression to analyze both individual- and hospital-level factors for abortion, preterm labor, and cesarean section.

Results

A total of 10,271 acute appendicitis during pregnancy were identified during the study period. Of them, 5,872 (57.2%) were treated by laparoscopic appendectomy, 1,403 (13.7%) by open appendectomy, and 2,996 (29.2%) by non-surgical treatment. Compared with open appendectomy, both laparoscopic appendectomy (adjusted OR, 0.6, 95% CI, 0.4, 0.9) and non-surgical treatment (adjusted OR, 0.4; 95% CI, 0.3–0.7) showed a decreased risk of preterm labor. Other important individual-level determinants of adverse pregnancy outcomes included maternal age, gestational hypertension, and anemia during pregnancy, the hospital-level determinant included the number of beds.

Conclusions

Compared with open appendectomy, both laparoscopic appendectomy and non-surgical treatment may be associated with a lower risk of preterm labor, without increased risks of abortion and cesarean section.

Rates and Predictors of Using Opioids in the Emergency Department to Treat Migraine in Adolescents and Young Adults

OBJECTIVES

This study aimed to determine the rate and context in which opioids are used to treat migraine in adolescents and young adults seen in emergency care settings.

METHODS

Data from 2010 to 2016 in the Cerner Health Facts electronic health record data warehouse were analyzed using multilevel logistic regression to estimate the population likelihood of an opioid being used in the emergency department (ED) to treat a primary diagnosis of migraine in adolescents and young adults and to evaluate the extent to which this likelihood varies as a function of characteristics of the patient (age, sex, race, and insurance), encounter (referral source, provider specialty, and encounter duration and year), and ED (region, setting, size, payer mix, and academic status).

RESULTS

The study identified 14,494 eligible ED encounters with unique patients, of which 23% involved an opioid. Likelihood of being treated with opioids was significantly higher for patients who were older, female, white, and seen by a surgeon and who had longer encounters and encounters earlier in the time period sampled. Sites varied widely in percentage of encounters involving opioids (mean, 26.4% ± 20.1%; range, 0-100%), with higher rates associated with smaller sites with relatively higher proportions of commercially insured patients.

CONCLUSIONS

Use of opioids in the ED to treat migraine in youth is fairly common, with rate variation reflecting broader trends in for whom opioids tend to be more likely to be prescribed. These findings may be helpful for benchmarking and informing quality improvement efforts aimed at reducing unwarranted opioid exposure in youth.

Trends of Sex Differences in Clinical Outcomes After Myocardial Infarction in the United States

BACKGROUND

Female patients have been shown to experience worse clinical outcomes after acute myocardial infarction (AMI) compared with male patients. However, it is unclear what trend these differences followed over time.

METHODS

Data from patients hospitalized with AMI between 2004 and 2015 in the National Inpatient Sample were retrospectively analyzed, stratified according to sex. Multivariable logistic regression analyses were performed to examine the adjusted odds ratios (aORs) of invasive management and in-hospital outcomes according to sex. The Mantel-Haenszel extension of the χ2 test was performed to examine the trend of management and in-hospital outcomes over the study period.

RESULTS

Of 7,026,432 AMI hospitalizations, 39.7% (n = 2,789,494) were women. Overall, women were older (median: 77 vs 70 years), with a higher prevalence of risk factors such as diabetes, hypertension, and depression. Women were less likely to receive coronary angiography (aOR, 0.92; 95% confidence interval [CI], 0.91-0.93) and percutaneous coronary intervention (aOR, 0.82; 95% CI, 0.81-0.83) compared with men. Odds of all-cause mortality were higher in women (aOR, 1.03; 95% CI, 1.02-1.04; P < 0.001) and these rates have not narrowed over time (2004 vs 2015: aOR, 1.07 [95% CI, 1.04-1.09] vs 1.11 [95% CI, 1.07-1.15), with similar observations recorded for major adverse cardiovascular and cerebrovascular events.

CONCLUSIONS

In this temporal analysis of AMI hospitalizations over 12 years, we showed lower receipt of invasive therapies and higher mortality rates in women, with no change in temporal trends. There needs to be a systematic and consistent effort toward exploring these disparities to identify strategies to mitigate them.

Association between obsessive-compulsive disorder and obstetrical and neonatal outcomes in the USA: a population-based cohort study

Obsessive-compulsive disorder (OCD) is a mental disorder linked to functional impairments and adverse health outcomes. We sought to examine the association between pregnant women with OCD and obstetrical and neonatal outcomes in the USA. A retrospective population-based cohort study was conducted using data provided by pregnant women from the Nationwide Inpatient Sample, a nationally representative database of hospitalizations in the USA, from 1999 to 2015. Using diagnostic and procedure codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM), we identified births and classified women by OCD status. Demographic and clinical characteristics were compared for women with and without OCD and multivariate logistic regressions were used to obtain odds ratios (OR) to compare obstetrical and neonatal outcomes between the two groups, adjusting for relevant demographic and clinical variables. Between 1999 and 2015, there were 3365 births to women with OCD, corresponding to an overall prevalence of 24.40 per 100,000 births. Women with OCD were more likely to be older than 25, Caucasian, of higher socioeconomic status, smokers or used drugs and alcohol, and have other comorbid psychiatric conditions. In adjusted models, OCD was associated with a higher risk of gestational hypertension, preeclampsia, premature rupture of membranes, caesarean and instrumental deliveries, venous thromboembolisms and preterm birth. Pregnancies in women with OCD are at high risk of adverse obstetrical and neonatal outcomes. A multidisciplinary approach should be used to identify high risk behaviours and ensure adequate prenatal follow-up and care be available for those with high risk pregnancies.

Predictors, Treatments, and Outcomes of Do-Not-Resuscitate Status in Acute Myocardial Infarction Patients (from a Nationwide Inpatient Cohort Study)

Little is known about how frequently do-not-resuscitate (DNR) orders are placed in patients with acute myocardial infarction (AMI), the types of patients in which they are placed, treatment strategies or clinical outcomes of such patients. Using the United States (US) National Inpatient Sample (NIS) database from 2015 to 2018, we identified 2,767,549 admissions that were admitted to US hospitals and during the hospitalization received a principle diagnosis of AMI, of which 339,270 (12.3%) patients had a DNR order (instigated both preadmission and during in-hospital stay). Patients with a DNR status were older (median age 83 vs 65, p < 0.001), more likely to be female (53.4% vs 39.3%, p < 0.001) and White (81.0% vs 73.3%, p < 0.001). Predictors of DNR status included comorbidities such as heart failure (OR: 1.47, 95% CI: 1.45 to 1.48), dementia (OR: 2.53, 95% CI: 2.50 to 2.55), and cancer. Patients with a DNR order were less likely to undergo invasive management or be discharged home (13.5% vs 52.8%), with only 1/3 receiving palliative consultation. In hospital mortality (32.7% vs 4.6%, p < 0.001) and MACCE (37.1% vs 8.8%, p < 0.001) were higher in the DNR group. Factors independently associated with in-hospital mortality among patients with a DNR order included a STEMI presentation (OR: 2.90, 95% CI: 2.84 to 2.96) and being of Black (OR: 1.29, 95% CI: 1.26 to 1.33), Hispanic (OR: 1.36, 95% CI: 1.32 to 1.41) or Asian/Pacific Islander (OR: 1.56, 95% CI:1.49-race. In conclusion, AMI patients with a DNR status were older, multimorbid, less likely to receive invasive management, with only one third of patients with DNR status referred for palliative care.

Healthcare System-to-System Cost Variability in the Care of Pediatric Abdominal Pain-Associated Functional Gastrointestinal Disorders

The purpose of this study was to assess cost variability in the care of abdominal pain-associated functional gastrointestinal disorders (AP-FGIDS) in youth across health systems, races, and specific AP-FGID diagnoses. Patients, aged 8-17 years, with a priority 1 diagnosis corresponding to a Rome IV defined AP-FGID were identified within the Health Facts® database. Total costs were obtained across the continuum of care including outpatient clinics, emergency department, and inpatient or observation units. Cost variability was described comparing different health systems, races, and diagnoses. Thirteen thousand two hundred and fourteen patients were identified accounting for 17,287 encounters. Total costs were available for 38.7% of the encounters. There was considerable variability in costs within and, especially, across health systems. Costs also varied across race, urban vs. rural site of care, and AP-FGID diagnoses. In conclusion, there was considerable variability in the costs for care of AP-FGIDs which is sufficient to support multi-site studies to understand the value of specific tests and treatments. Significant differences in costs by race merit further investigation to understand key drivers.

Distribution, management and outcomes of AMI according to principal diagnosis priority during inpatient admission

BACKGROUND

In recent years, there has been a growing interest in outcomes of patients with acute myocardial infarction (AMI) using large administrative datasets. The present study was designed to compare the characteristics, management strategies and acute outcomes between patients with primary and secondary AMI diagnoses in a national cohort of patients.

METHODS

All hospitalisations of adults (≥18 years) with a discharge diagnosis of AMI in the US National Inpatient Sample from January 2004 to September 2015 were included, stratified by primary or secondary AMI. The International Classification of Diseases, ninth revision and Clinical Classification Software codes were used to identify patient comorbidities, procedures and clinical outcomes.

RESULTS

A total of 10 864 598 weighted AMI hospitalisations were analysed, of which 7 186 261 (66.1%) were primary AMIs and 3 678 337 (33.9%) were secondary AMI. Patients with primary AMI diagnoses were younger (median 68 vs 74 years, P < .001) and less likely to be female (39.6% vs 48.5%, P < .001). Secondary AMI was associated with lower odds of receipt of coronary angiography (aOR 0.19; 95%CI 0.18-0.19) and percutaneous coronary intervention (0.24; 0.23-0.24). Secondary AMI was associated with increased odds of MACCE (1.73; 1.73-1.74), mortality (1.71; 1.70-1.72), major bleeding (1.64; 1.62-1.65), cardiac complications (1.69; 1.65-1.73) and stroke (1.68; 1.67-1.70) (P < .001 for all).

CONCLUSIONS

Secondary AMI diagnoses account for one-third of AMI admissions. Patients with secondary AMI are older, less likely to receive invasive care and have worse outcomes than patients with a primary diagnosis code of AMI. Future studies should consider both primary and secondary AMI diagnoses codes in order to accurately inform clinical decision-making and health planning.

Electronic Health Record Network Research in Infectious Diseases

With the marked increases in electronic health record (EHR) use for providing clinical care, there have been parallel efforts to leverage EHR data for research. EHR repositories offer the promise of vast amounts of clinical data not easily captured with traditional research methods and facilitate clinical epidemiology and comparative effectiveness research, including analyses to identify patients at higher risk for complications or who are better candidates for treatment. These types of studies have been relatively slow to penetrate the field of infectious diseases, but the need for rapid turnaround during the COVID-19 global pandemic has accelerated the uptake. This review discusses the rationale for her network projects, opportunities and challenges that such networks present, and some prior studies within the field of infectious diseases.

The Case against Antibiotics and for Anti-Virulence Therapeutics

Although antibiotics have been indispensable in the advancement of modern medicine, there are downsides to their use. Growing resistance to broad-spectrum antibiotics is leading to an epidemic of infections untreatable by first-line therapies. Resistance is exacerbated by antibiotics used as growth factors in livestock, over-prescribing by doctors, and poor treatment adherence by patients. This generates populations of resistant bacteria that can then spread resistance genes horizontally to other bacterial species, including commensals. Furthermore, even when antibiotics are used appropriately, they harm commensal bacteria leading to increased secondary infection risk. Effective antibiotic treatment can induce bacterial survival tactics, such as toxin release and increasing resistance gene transfer. These problems highlight the need for new approaches to treating bacterial infection. Current solutions include combination therapies, narrow-spectrum therapeutics, and antibiotic stewardship programs. These mediate the issues but do not address their root cause. One emerging solution to these problems is anti-virulence treatment: preventing bacterial pathogenesis instead of using bactericidal agents. In this review, we discuss select examples of potential anti-virulence targets and strategies that could be developed into bacterial infection treatments: the bacterial type III secretion system, quorum sensing, and liposomes.

Use of large scale EHR data to evaluate A1c utilization among sickle cell disease patients

BACKGROUND

The glycated hemoglobin (A1c) test is not recommended for sickle cell disease (SCD) patients. We examine ordering patterns of diabetes-related tests for SCD patients to explore misutilization of tests among this underserved population.

METHODS

We used de-identified electronic health record (EHR) data in the Cerner Health Facts™ (HF) data warehouse to evaluate the frequency of A1c and fructosamine tests during 2010 to 2016, for 37,151 SCD patients from 393 healthcare facilities across the United States. After excluding facilities with no A1c data, we defined three groups of facilities based on the prevalence of SCD patients with A1c test(s): adherent facilities (no SCD patients with A1c test(s)), minor non-adherent facilities, major non-adherent facilities.

RESULTS

We determined that 11% of SCD patients (3927 patients) treated at 393 facilities in the US received orders for at least one A1c test. Of the 3927 SCD patients with an A1c test, only 89 patients (2.3%) received an order for a fructosamine test. At the minor non-adherent facilities, 5% of the SCD patients received an A1c test while 58% of the SCD patients at the least adherent facilities had at least one A1c test. Overall, the percent of A1c tests ordered for SCD patients between 2010 and 2016 remained similar.

CONCLUSIONS

Inappropriate A1c test orders among a sickle cell population is a significant quality gap. Interventions to advance adoption of professional recommendations that advocate for alternate tests, such as fructosamine, can guide clinicians in test selection to reduce this quality gap are discussed. The informatics strategy used in this work can inform other largescale analyses of lab test utilization using de-identified EHR data.

Differential impact of type 1 and type 2 diabetes mellitus on outcomes among 1.4 million US patients undergoing percutaneous coronary intervention

BACKGROUND

The aim was to determine the impact of diabetes mellitus (DM) on outcomes after percutaneous coronary intervention (PCI). There is limited data on the impact of DM and its subtypes among patients who underwent PCI during hospitalization.

METHODS

All PCI hospitalizations from the National Inpatient Sample (October 2015-December 2018) were stratified by the presence and subtype of DM. Multivariable logistic regression was performed to determine the adjusted odds ratios (aOR) of in-hospital adverse outcomes in type 1 DM (T1DM) and type 2 DM (T2DM) compared to no-DM.

RESULTS

Out of 1,363,800 individuals undergoing PCI, 12,640 (0.9%) had T1DM and 539,690 (39.6%) had T2DM. T1DM patients had increased aOR of major adverse cardiovascular and cerebrovascular events (MACCE) (1.26, 95%CI 1.17-1.35), mortality (1.56, 95%CI 1.41-1.72), major bleeding (1.63, 95%CI 1.45-1.84), and stroke (1.75, 95%CI 1.51-2.02), while T2DM patients had only increased aOR of MACCE (1.02, 95%CI 1.01-1.04), mortality (1.10, 95%CI 1.08-1.13) and stroke (1.22, 95%CI 1.18-1.27), compared to no-DM patients. However, both T1DM and T2DM had lower aOR of cardiac complications (0.87, 95%CI 0.77-0.97 and 0.87, 95%CI 0.85-0.89, respectively), in comparison to no-DM patients. When accounting for the indication, both DM subgroups had higher aOR of MACCE, mortality, and stroke compared to no-DM patients in the acute coronary syndrome setting (p < 0.001, for all), while only increased aOR of stroke (1.59, 95%CI 1.17-2.15 for T1DM and 1.12, 95%CI 1.05-1.20 for T2DM) persisted in the elective setting.

CONCLUSIONS

Patients with DM who have undergone PCI during hospitalization are more likely to experience adverse in-hospital outcomes, and T1DM patients are a particularly high-risk cohort.

Predicting Future Care Requirements Using Machine Learning for Pediatric Intensive and Routine Care Inpatients

Supplemental Digital Content is available in the text.

OBJECTIVES:

Develop and compare separate prediction models for ICU and non-ICU care for hospitalized children in four future time periods (6–12, 12–18, 18–24, and 24–30 hr) and assess these models in an independent cohort and simulated children’s hospital.

DESIGN:

Predictive modeling used cohorts from the Health Facts database (Cerner Corporation, Kansas City, MO).

SETTING:

Children hospitalized in ICUs.

PATIENTS:

Children with greater than or equal to one ICU admission (n = 20,014) and randomly selected routine care children without ICU admission (n = 20,130) from 2009 to 2016 were used for model development and validation. An independent 2017–2018 cohort consisted of 80,089 children.

INTERVENTIONS:

None.

MEASUREMENT AND MAIN RESULTS:

Initially, we undersampled non-ICU patients for development and comparison of the models. We randomly assigned 64% of patients for training, 8% for validation, and 28% for testing in both clinical groups. Two additional validation cohorts were tested: a simulated children’s hospitals and the 2017–2018 cohort. The main outcome was ICU care or non-ICU care in four future time periods based on physiology, therapy, and care intensity. Four independent, sequential, and fully connected neural networks were calibrated to risk of ICU care at each time period. Performance for all models in the test sample were comparable including sensitivity greater than or equal to 0.727, specificity greater than or equal to 0.885, accuracy greater than 0.850, area under the receiver operating characteristic curves greater than or equal to 0.917, and all had excellent calibration (all R²s > 0.98). Model performance in the 2017–2018 cohort was sensitivity greater than or equal to 0.545, specificity greater than or equal to 0.972, accuracy greater than or equal to 0.921, area under the receiver operating characteristic curves greater than or equal to 0.946, and R²s greater than or equal to 0.979. Performance metrics were comparable for the simulated children’s hospital and for hospitals stratified by teaching status, bed numbers, and geographic location.

CONCLUSIONS:

Machine learning models using physiology, therapy, and care intensity predicting future care needs had promising performance metrics. Notably, performance metrics were similar as the prediction time periods increased from 6–12 hours to 24–30 hours.

A Large-Scale Observational Study on the Temporal Trends and Risk Factors of Opioid Overdose: Real-World Evidence for Better Opioids

Background

The USA is in the midst of an opioid overdose epidemic. To address the epidemic, we conducted a large-scale population study on opioid overdose.

Objectives

The primary objective of this study was to evaluate the temporal trends and risk factors of inpatient opioid overdose. Based on its patterns, the secondary objective was to examine the innate properties of opioid analgesics underlying reduced overdose effects.

Methods

A retrospective cross-sectional study was conducted based on a large-scale inpatient electronic health records database, Cerner Health Facts^®, with (1) inclusion criteria for participants as patients admitted between 1 January, 2009 and 31 December, 2017 and (2) measurements as opioid overdose prevalence by year, demographics, and prescription opioid exposures.

Results

A total of 4,720,041 patients with 7,339,480 inpatient encounters were retrieved from Cerner Health Facts^®. Among them, 30.2% patients were aged 65+ years, 57.0% female, 70.1% Caucasian, 42.3% single, 32.0% from the South, and 80.8% in an urban area. From 2009 to 2017, annual opioid overdose prevalence per 1000 patients significantly increased from 3.7 to 11.9 with an adjusted odds ratio (aOR): 1.16, 95% confidence interval (CI) 1.15–1.16. Compared to the major demographic counterparts, being in (1) age group: 41–50 years (overall aOR 1.36, 95% CI 1.31–1.40) or 51–64 years (overall aOR 1.35, 95% CI 1.32–1.39), (2) marital status: divorced (overall aOR 1.19, 95% CI 1.15–1.23), and (3) census region: West (overall aOR 1.32, 95% CI 1.28–1.36) were significantly associated with a higher odds of opioid overdose. Prescription opioid exposures were also associated with an increased odds of opioid overdose, such as meperidine (overall aOR 1.09, 95% CI 1.06–1.13) and tramadol (overall aOR 2.20, 95% CI 2.14–2.27). Examination on the relationships between opioid analgesic properties and their association strengths, aORs, and opioid overdose showed that lower aOR values were significantly associated with (1) high molecular weight, (2) non-interaction with multi-drug resistance protein 1 or interaction with cytochrome P450 3A4, and (3) non-interaction with the delta opioid receptor or kappa opioid receptor.

Conclusions

The significant increasing trends of opioid overdose at the inpatient care setting from 2009 to 2017 suggested an ongoing need for efforts to combat the opioid overdose epidemic in the USA. Risk factors associated with opioid overdose included patient demographics and prescription opioid exposures. Moreover, there are physicochemical, pharmacokinetic, and pharmacodynamic properties underlying reduced overdose effects, which can be utilized to develop better opioids.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40801-021-00253-8.

Ranking Methodology to Evaluate the Severity of a Quality Gap Using a National EHR Database

Selecting quality improvement projects can often be a reactive process. In order to demonstrate a data-driven strategy, we used multi-site, de-identified electronic health record (EHR) data to prioritize the severity of a quality concern: inappropriate A1c test orders for sickle cell disease patients in two randomly chosen facilities (Facility A & B). The best linear unbiased predictions (BLUP) generated from Generalized Linear Mixed Model (GLMM) was estimated for all 393 facilities with 37,151 SCD patients in the Cerner Health FactsTM (HF) data warehouse based on the ratio of inappropriate A1c orders. Ranking the BLUP after applying the GLMM indicates that the facility A being in the second quartile may not have a quality gap as significant as facility B in the top quartile for this quality concern. This study illustrates the utility of multisite EHR data for evaluating QI projects and the utility of GLMM to enable this analysis.

The Role of Labor Induction in Racial Disparities in Cesarean Delivery

We examined the interaction between race and labor induction in cesarean delivery in a cohort of 600,000 deliveries in the Cerner Health Facts database. Black women had higher likelihood cesarean (28.9 vs. 26.5%) and lower likelihood of induction of labor at delivery compared to white women (27.2 vs. 32.5%). Induction modified the association between race and cesarean-Black women (odds ratio=1.36, 95% confidence interval 1.30, 1.43) who were induced had significantly increased odds of cesarean delivery.

Incidence and predictors of postoperative ischemic stroke after coronary artery bypass grafting

BACKGROUND

Data on the incidence and outcomes of ischemic stroke in patients undergoing coronary artery bypass grafting (CABG) in the current era are limited. The goal of this study was to examine contemporary trends, predictors, and outcomes of ischemic stroke following CABG in a large nationally representative database over a 12-year-period.

METHODS

The National Inpatient Sample was used to identify all adult (≥18 years) patients who underwent CABG between 2004 and 2015. The incidence and predictors of post-CABG ischemic stroke were assessed and in-hospital outcomes of patients with and without post-CABG stroke were compared.

RESULTS

Out of 2 569 597 CABG operations, ischemic stroke occurred in 47 279 (1.8%) patients, with a rising incidence from 2004 (1.2%) to 2015 (2.3%) (P < .001). Patient risk profiles increased over time in both cohorts, with higher Charlson comorbidity scores observed amongst stroke patients. Stroke was independently associated with higher rates of in-hospital mortality (3-fold), longer lengths of hospital stay (~6 more days), and higher total hospitalisation cost (~$80 000 more). Age ≥60 years and female sex (OR 1.33, 95% CI 1.31-1.36) were the strongest predictors of stroke (both P < .001). Further, on-pump CABG was not an independent predictor of stroke (P = .784).

CONCLUSION

In this nationally representative study we have shown that the rates of postoperative stroke complications following CABG have increased over time to commensurate with a parallel increase in overall baseline patient risks. Given the adverse impact of stroke on in-hospital morbidity and mortality after CABG, further studies are warranted to systematically delineate factors contributing to this striking trend.

Identifying risk of opioid use disorder for patients taking opioid medications with deep learning

OBJECTIVE

The United States is experiencing an opioid epidemic. In recent years, there were more than 10 million opioid misusers aged 12 years or older annually. Identifying patients at high risk of opioid use disorder (OUD) can help to make early clinical interventions to reduce the risk of OUD. Our goal is to develop and evaluate models to predict OUD for patients on opioid medications using electronic health records and deep learning methods. The resulting models help us to better understand OUD, providing new insights on the opioid epidemic. Further, these models provide a foundation for clinical tools to predict OUD before it occurs, permitting early interventions.

METHODS

Electronic health records of patients who have been prescribed with medications containing active opioid ingredients were extracted from Cerner's Health Facts database for encounters between January 1, 2008, and December 31, 2017. Long short-term memory models were applied to predict OUD risk based on five recent prior encounters before the target encounter and compared with logistic regression, random forest, decision tree, and dense neural network. Prediction performance was assessed using F1 score, precision, recall, and area under the receiver-operating characteristic curve.

RESULTS

The long short-term memory (LSTM) model provided promising prediction results which outperformed other methods, with an F1 score of 0.8023 (about 0.016 higher than dense neural network (DNN)) and an area under the receiver-operating characteristic curve (AUROC) of 0.9369 (about 0.145 higher than DNN).

CONCLUSIONS

LSTM-based sequential deep learning models can accurately predict OUD using a patient's history of electronic health records, with minimal prior domain knowledge. This tool has the potential to improve clinical decision support for early intervention and prevention to combat the opioid epidemic.

Aligning EHR Data for Pediatric Leukemia With Standard Protocol Therapy

Children with acute lymphoblastic leukemia (ALL) are treated according to risk-based protocols defined by the Children's Oncology Group (COG). Alignment between real-world clinical practice and protocol milestones is not widely understood. Aggregate deidentified electronic health record (EHR) data offer a useful resource to evaluate real-world clinical practice.

Acute Myocardial Infarction in Autoimmune Rheumatologic Disease: A Nationwide Analysis of Clinical Outcomes and Predictors of Management Strategy

OBJECTIVES

To examine national-level differences in management strategies and outcomes in patients with autoimmune rheumatic disease (AIRD) with acute myocardial infarction (AMI) from 2004 through 2014.

METHODS

All AMI hospitalizations were analyzed from the National Inpatient Sample, stratified according to AIRD diagnosis into 4 groups: no AIRD, rheumatoid arthritis (RA), systemic lupus erythematosus (SLE), and systemic sclerosis (SSC). The associations between AIRD subtypes and (1) receipt of coronary angiography and percutaneous coronary intervention (PCI) and (2) clinical outcomes were examined compared with patients without AIRD.

RESULTS

Of 6,747,797 AMI hospitalizations, 109,983 patients (1.6%) had an AIRD diagnosis (RA: 1.3%, SLE: 0.3%, and SSC: 0.1%). The prevalence of RA rose from 1.0% (2004) to 1.5% (2014), and SLE and SSC remained stable. Patients with SLE were less likely to receive invasive management (odds ratio [OR] [95% CI]: coronary angiography-0.87; 0.84 to 0.91; PCI-0.93; 0.90 to 0.96), whereas no statistically significant differences were found in the RA and SSC groups. Subsequently, the ORs (95% CIs) of mortality (1.15; 1.07 to 1.23) and bleeding (1.24; 1.16 to 1.31) were increased in patients with SLE; SSC was associated with increased ORs (95% CIs) of major adverse cardiovascular and cerebrovascular events (1.52; 1.38 to 1.68) and mortality (1.81; 1.62 to 2.02) but not bleeding or stroke; the RA group was at no increased risk for any complication.

CONCLUSION

In a nationwide cohort of AMI hospitalizations we found lower use of invasive management in patients with SLE and worse outcomes after AMI in patients with SLE and SSC compared with those without AIRD.

Outpatient health care utilization for sleep disorders in the Cerner Health Facts database

STUDY OBJECTIVES

Sleep disorders are common in the general population. This study aimed to identify direct health care utilization for sleep disorders using big data through the Cerner Health Facts database.

METHODS

The Cerner Health Facts database has 68.7 million patients in the data warehouse, documenting approximately 506.9 million encounters from 100 nonaffiliated health care systems. To identify sleep-related health care utilization, we examined the frequency of outpatient encounters related to sleep disorders between the years 2000 and 2017. Sleep disorders were grouped-based on the International Classification of Sleep Disorders-Third Edition.

RESULTS

Approximately 20.5 million patients were identified with a total of 127.4 million outpatient encounters. In pediatric patients (ages 0-18 years), health care utilization for major sleep diagnoses was measured per 100,000 encounters. Sleep-related breathing disorders ranked first among common sleep disorders for pediatric patients followed by parasomnia, insomnia, sleep movement disorders, hypersomnolence, then circadian rhythm disorders (820.1, 258.1, 181.6, 68.3, 48.1, and 16.2 per 100,000 encounters, respectively). However, in adult patients, the ranking was slightly different, with sleep-related breathing disorders ranked first, followed by insomnia, sleep-related movement disorders, hypersomnolence, parasomnia, then circadian rhythm disorders (1352.6, 511.6, 166.3, 79.1, 25.7, and 4.2 per 100,000 encounters, respectively). In general, there was a bimodal pattern with a clear dip in sleep-related health care utilization in young adults age (age 19-29 years), with the exception of insomnia.

CONCLUSIONS

Patients with sleep disorders show relatively low health care utilization despite a known high prevalence of sleep disorders in the general population. This finding may highlight under-recognition of sleep problems or decreased access to health care for sleep disorders. In addition, this study highlights the effect of age-based variation on different sleep disorders, which may have an impact on allocating resources.

Inappropriate empirical antibiotic therapy for bloodstream infections based on discordant in-vitro susceptibilities: a retrospective cohort analysis of prevalence, predictors, and mortality risk in US hospitals

BACKGROUND

The prevalence and effects of inappropriate empirical antibiotic therapy for bloodstream infections are unclear. We aimed to establish the population-level burden, predictors, and mortality risk of in-vitro susceptibility-discordant empirical antibiotic therapy among patients with bloodstream infections.

METHODS

Our retrospective cohort analysis of electronic health record data from 131 hospitals in the USA included patients with suspected-and subsequently confirmed-bloodstream infections who were treated empirically with systemic antibiotics between Jan 1, 2005, and Dec 31, 2014. We included all patients with monomicrobial bacteraemia caused by common bloodstream pathogens who received at least one systemic antibiotic either on the day blood cultures were drawn or the day after, and for whom susceptibility data were available. We calculated the prevalence of discordant empirical antibiotic therapy-which was defined as receiving antibiotics on the day blood culture samples were drawn to which the cultured isolate was not susceptible in vitro-overall and by hospital type by using regression tree analysis. We used generalised estimating equations to identify predictors of receiving discordant empirical antibiotic therapy, and used logistic regression to calculate adjusted odds ratios for the relationship between in-hospital mortality and discordant empirical antibiotic therapy.

FINDINGS

21 608 patients with bloodstream infections received empirical antibiotic therapy on the day of first blood culture collection. Of these patients, 4165 (19%) received discordant empirical antibiotic therapy. Discordant empirical antibiotic therapy was independently associated with increased risk of mortality (adjusted odds ratio 1·46 [95% CI, 1·28-1·66]; p<0·0001), a relationship that was unaffected by the presence or absence of resistance or sepsis or septic shock. Infection with antibiotic-resistant species strongly predicted receiving discordant empirical therapy (adjusted odds ratio 9·09 [95% CI 7·68-10·76]; p<0·0001). Most incidences of discordant empirical antibiotic therapy and associated deaths occurred among patients with bloodstream infections caused by Staphylococcus aureus or Enterobacterales.

INTERPRETATION

Approximately one in five patients with bloodstream infections in US hospitals received discordant empirical antibiotic therapy, receipt of which was closely associated with infection with antibiotic-resistant pathogens. Receiving discordant empirical antibiotic therapy was associated with increased odds of mortality overall, even in patients without sepsis. Early identification of bloodstream pathogens and resistance will probably improve population-level outcomes.

FUNDING

US National Institutes of Health, US Centers for Disease Control and Prevention, and US Agency for Healthcare Research and Quality.

The Association of Laboratory Test Abnormalities With Mortality Risk in Pediatric Intensive Care

OBJECTIVES

To determine the bivariable associations between abnormalities of 28 common laboratory tests and hospital mortality and determine how mortality risks changes when the ranges are evaluated in the context of commonly used laboratory test panels.

DESIGN

A 2009-2016 cohort from the Health Facts (Cerner Corporation, Kansas City, MO) database.

SETTING

Hospitals caring for children in ICUs.

PATIENTS

Children cared for in ICUs with laboratory data.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

There were 2,987,515 laboratory measurements in 71,563 children. The distribution of laboratory test values in 10 groups defined by population percentiles demonstrated the midrange of tests was within the normal range except for those measured predominantly when significant abnormalities are suspected. Logistic regression analysis at the patient level combined the population-based groups into ranges with nonoverlapping mortality odds ratios. The most deviant test ranges associated with increased mortality risk (mortality odds ratios > 5.0) included variables associated with acidosis, coagulation abnormalities and blood loss, immune function, liver function, nutritional status, and the basic metabolic profile. The test ranges most associated with survival included normal values for chloride, pH, and bicarbonate/total Co2. When the significant test ranges from bivariable analyses were combined in commonly used test panels, they generally remained significant but were reduced as risk was distributed among the tests.

CONCLUSIONS

The relative importance of laboratory test ranges vary widely, with some ranges strongly associated with mortality and others strongly associated with survival. When evaluated in the context of test panels rather than isolated tests, the mortality odds ratios for the test ranges decreased but generally remained significant as risk was distributed among the components of the test panels. These data are useful to develop critical values for children in ICUs, to identify risk factors previously underappreciated, for education and training, and for future risk score development.

Clinical Characteristics, Management Strategies and Outcomes of Acute Myocardial Infarction Patients With Prior Coronary Artery Bypass Grafting

OBJECTIVE

To investigate the management strategies, temporal trends, and clinical outcomes of patients with a history of coronary artery bypass graft (CABG) surgery and presenting with acute myocardial infarction (MI).

PATIENTS AND METHODS

We undertook a retrospective cohort study using the National Inpatient Sample database from the United States (January 2004-September 2015), identified all inpatient MI admissions (7,250,768 records) and stratified according to history of CABG (group 1, CABG-naive [94%]; group 2, prior CABG [6%]).

RESULTS

Patients in group 2 were older, less likely to be female, had more comorbidities, and were more likely to present with non-ST-elevation myocardial infarction compared with group 1. More patients underwent coronary angiography (68% vs 48%) and percutaneous coronary intervention (PCI) (44% vs 26%) in group 1 compared with group 2. Following multivariable logistic regression analyses, the adjusted odd ratio (OR) of in-hospital major adverse cardiovascular and cerebrovascular events (OR, 0.98; 95% CI, 0.95 to 1.005; P=.11), all-cause mortality (OR, 1; 95% CI, 0.98 to 1.04; P=.6) and major bleeding (OR, 0.99; 95% CI, 0.94 to 1.03; P=.54) were similar to group 1. Lower adjusted odds of in-hospital major adverse cardiovascular and cerebrovascular events (OR, 0.64; 95% CI, 0.57 to 0.72; P<.001), all-cause mortality (OR, 0.45; 95% CI, 0.38 to 0.53; P<.001), and acute ischemic stroke (OR, 0.71; 95% CI, 0.59 to 0.86; P<.001) were observed in group 2 patients who underwent PCI compared with those managed medically without any increased risk of major bleeding (OR, 1.08; 95% CI, 0.94 to 1.23; P=.26).

CONCLUSIONS

In this national cohort, MI patients with prior-CABG had a higher risk profile, but similar in-hospital adverse outcomes compared with CABG-naive patients. Prior-CABG patients who received PCI had better in-hospital clinical outcomes compared to those who received medical management.

Severity Trajectories of Pediatric Inpatients Using the Criticality Index

OBJECTIVES

To assess severity of illness trajectories described by the Criticality Index for survivors and deaths in five patient groups defined by the sequence of patient care in ICU and routine patient care locations.

DESIGN

The Criticality Index developed using a calibrated, deep neural network, measures severity of illness using physiology, therapies, and therapeutic intensity. Criticality Index values in sequential 6-hour time periods described severity trajectories.

SETTING

Hospitals with pediatric inpatient and ICU care.

PATIENTS

Pediatric patients never cared for in an ICU (n = 20,091), patients only cared for in the ICU (n = 2,096) and patients cared for in both ICU and non-ICU care locations (n = 17,023) from 2009 to 2016 Health Facts database (Cerner Corporation, Kansas City, MO).

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

Criticality Index values were consistent with clinical experience. The median (25-75th percentile) ICU Criticality Index values (0.878 [0.696-0.966]) were more than 80-fold higher than the non-ICU values (0.010 [0.002-0.099]). Non-ICU Criticality Index values for patients transferred to the ICU were 40-fold higher than those never transferred to the ICU (0.164 vs 0.004). The median for ICU deaths was higher than ICU survivors (0.983 vs 0.875) (p < 0.001). The severity trajectories for the five groups met expectations based on clinical experience. Survivors had increasing Criticality Index values in non-ICU locations prior to ICU admission, decreasing Criticality Index values in the ICU, and decreasing Criticality Index values until hospital discharge. Deaths had higher Criticality Index values than survivors, steeper increases prior to the ICU, and worsening values in the ICU. Deaths had a variable course, especially those who died in non-ICU care locations, consistent with deaths associated with both active therapies and withdrawals/limitations of care.

CONCLUSIONS

Severity trajectories measured by the Criticality Index showed strong validity, reflecting the expected clinical course for five diverse patient groups.

Criticality: A New Concept of Severity of Illness for Hospitalized Children

OBJECTIVES

To validate the conceptual framework of "criticality," a new pediatric inpatient severity measure based on physiology, therapy, and therapeutic intensity calibrated to care intensity, operationalized as ICU care.

DESIGN

Deep neural network analysis of a pediatric cohort from the Health Facts (Cerner Corporation, Kansas City, MO) national database.

SETTING

Hospitals with pediatric routine inpatient and ICU care.

PATIENTS

Children cared for in the ICU (n = 20,014) and in routine care units without an ICU admission (n = 20,130) from 2009 to 2016. All patients had laboratory, vital sign, and medication data.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

A calibrated, deep neural network used physiology (laboratory tests and vital signs), therapy (medications), and therapeutic intensity (number of physiology tests and medications) to model care intensity, operationalized as ICU (versus routine) care every 6 hours of a patient's hospital course. The probability of ICU care is termed the Criticality Index. First, the model demonstrated excellent separation of criticality distributions from a severity hierarchy of five patient groups: routine care, routine care for those who also received ICU care, transition from routine to ICU care, ICU care, and high-intensity ICU care. Second, model performance assessed with statistical metrics was excellent with an area under the curve for the receiver operating characteristic of 0.95 for 327,189 6-hour time periods, excellent calibration, sensitivity of 0.817, specificity of 0.892, accuracy of 0.866, and precision of 0.799. Third, the performance in individual patients with greater than one care designation indicated as 88.03% (95% CI, 87.72-88.34) of the Criticality Indices in the more intensive locations was higher than the less intense locations.

CONCLUSIONS

The Criticality Index is a quantification of severity of illness for hospitalized children using physiology, therapy, and care intensity. This new conceptual model is applicable to clinical investigations and predicting future care needs.

Using multivariate long short-term memory neural network to detect aberrant signals in health data for quality assurance

BACKGROUND

The data quality of electronic health records (EHR) has been a topic of increasing interest to clinical and health services researchers. One indicator of possible errors in data is a large change in the frequency of observations in chronic illnesses. In this study, we built and demonstrated the utility of a stacked multivariate LSTM model to predict an acceptable range for the frequency of observations.

METHODS

We applied the LSTM approach to a large EHR dataset with over 400 million total encounters. We computed sensitivity and specificity for predicting if the frequency of an observation in a given week is an aberrant signal.

RESULTS

Compared with the simple frequency monitoring approach, our proposed multivariate LSTM approach increased the sensitivity of finding aberrant signals in 6 randomly selected diagnostic codes from 75 to 88% and the specificity from 68 to 91%. We also experimented with two different LSTM algorithms, namely, direct multi-step and recursive multi-step. Both models were able to detect the aberrant signals while the recursive multi-step algorithm performed better.

CONCLUSIONS

Simply monitoring the frequency trend, as is the common practice in systems that do monitor the data quality, would not be able to distinguish between the fluctuations caused by seasonal disease changes, seasonal patient visits, or a change in data sources. Our study demonstrated the ability of stacked multivariate LSTM models to recognize true data quality issues rather than fluctuations that are caused by different reasons, including seasonal changes and outbreaks.