Predicting pediatric cardiac surgery-associated acute kidney injury using machine learning.

BACKGROUND: Prediction of cardiac surgery-associated acute kidney injury (CS-AKI) in pediatric patients is crucial to improve outcomes and guide clinical decision-making. This study aimed to develop a supervised machine learning (ML) model for predicting moderate to severe CS-AKI at postoperative day 2 (POD2). METHODS: This retrospective cohort study analyzed data from 402 pediatric patients who underwent cardiac surgery at a university-affiliated children's hospital, who were separated into an 80%-20% train-test split. The ML model utilized demographic, preoperative, intraoperative, and POD0 clinical and laboratory data to predict moderate to severe AKI categorized by Kidney Disease: Improving Global Outcomes (KDIGO) stage 2 or 3 at POD2. Input feature importance was assessed by SHapley Additive exPlanations (SHAP) values. Model performance was evaluated using accuracy, area under the receiver operating curve (AUROC), precision, recall, area under the precision-recall curve (AUPRC), F1-score, and Brier score. RESULTS: Overall, 13.7% of children in the test set experienced moderate to severe AKI. The ML model achieved promising performance, with accuracy of 0.91 (95% CI: 0.82-1.00), AUROC of 0.88 (95% CI: 0.72-1.00), precision of 0.92 (95% CI: 0.70-1.00), recall of 0.63 (95% CI: 0.32-0.96), AUPRC of 0.81 (95% CI: 0.61-1.00), F1-score of 0.73 (95% CI: 0.46-0.99), and Brier score loss of 0.09 (95% CI: 0.00-0.17). The top ten most important features assessed by SHAP analyses in this model were preoperative serum creatinine, surgery duration, POD0 serum pH, POD0 lactate, cardiopulmonary bypass duration, POD0 vasoactive inotropic score, sex, POD0 hematocrit, preoperative weight, and POD0 serum creatinine. CONCLUSIONS: A supervised ML model utilizing demographic, preoperative, intraoperative, and immediate postoperative clinical and laboratory data showed promising performance in predicting moderate to severe CS-AKI at POD2 in pediatric patients.

Acceptance of Acute Kidney Injury Alert by Providers in Cardiac Surgery Intensive Care Unit.

BACKGROUND: Acute kidney injury (AKI) is a common complication after cardiac surgery and is associated with worse outcomes. Its management relies on early diagnosis, and therefore, electronic alerts have been used to alert clinicians for development of AKI. Electronic alerts are, however, associated with high rates of alert fatigue. OBJECTIVES: We designed this study to assess the acceptance of user-centered electronic AKI alert by clinicians. METHODS: We developed a user-centered electronic AKI alert that alerted clinicians of development of AKI in a persistent yet noninterruptive fashion. As the goal of the alert was to alert toward new or worsening AKI, it disappeared 48 hours after being activated. We assessed the acceptance of the alert using surveys at 6 and 12 months after the alert went live. RESULTS: At 6 months after their implementation, 38.9% providers reported that they would not have recognized AKI as early as they did without this alert. This number increased to 66.7% by 12 months of survey. Most providers also shared that they re-dosed or discontinued medications earlier, provided earlier management of volume status, avoided intravenous contrast use, and evaluated patients by using point-of-care ultrasounds more due to the alert. Overall, 83.3% respondents reported satisfaction with the electronic AKI alerts at 6 months and 94.4% at 12 months. CONCLUSION: This study showed high rates of acceptance of a user-centered electronic AKI alert over time by clinicians taking care of patients with AKI.

Utility of telemedicine in pediatric rheumatology during the COVID-19 pandemic.

BACKGROUND: During the COVID-19 pandemic, telemedicine has provided an alternative to in-person visits for patients practicing social distancing and undergoing quarantine. During this time, there has been a rapid expansion of telemedicine and its implementation in various clinical specialties and settings. In this observational study we aim to examine the utility of telemedicine in a pediatric rheumatology clinic, for 3 months during the COVID-19 pandemic. METHODS: A review of outpatient pediatric rheumatology telemedicine encounters were conducted from April-June 2020. Telemedicine visits (n = 75) were compared to patients seen in practice over the prior year in office-based visits (March 2019-March 2020) (n = 415). Patient characteristics, information on no-show visits, completed visits, new patient or follow-up status, and if new patients had received a visit within 2 weeks of calling to schedule an appointment were analyzed by chart review. An independent sample t-test and Chi Square statistic was used to determine statical significance between the two groups. A two-proportion z-test was used to compare visit metrics. RESULTS: The percentage of new patients utilizing telemedicine (60%) was lower and statistically significant compared to the percentage of new patient office visits (84%) the previous year (p < 0.0001). There was no change in no-show rate between groups and patient characteristics were similar. CONCLUSIONS: This study demonstrates a statistically significant decrease in new patient visits during the pandemic with telemedicine-only appointments compared to in-office visits over the previous year. This suggests a possible hesitation to seek care during this time. However, there was no significant difference among patient characteristics between telemedicine visits during the pandemic and during in-office visits in the previous year. In our experience, patient visits were able to be conducted via telemedicine with a limited physical exam using caregiver's help during the pandemic. However, further studies will need to ascertain patient satisfaction and preference for telemedicine in the future.

Coronary Artery Disease Phenotype Detection in an Academic Hospital System Setting.

BACKGROUND: The United States, and especially West Virginia, have a tremendous burden of coronary artery disease (CAD). Undiagnosed familial hypercholesterolemia (FH) is an important factor for CAD in the U.S. Identification of a CAD phenotype is an initial step to find families with FH. OBJECTIVE: We hypothesized that a CAD phenotype detection algorithm that uses discrete data elements from electronic health records (EHRs) can be validated from EHR information housed in a data repository. METHODS: We developed an algorithm to detect a CAD phenotype which searched through discrete data elements, such as diagnosis, problem lists, medical history, billing, and procedure (International Classification of Diseases [ICD]-9/10 and Current Procedural Terminology [CPT]) codes. The algorithm was applied to two cohorts of 500 patients, each with varying characteristics. The second (younger) cohort consisted of parents from a school child screening program. We then determined which patients had CAD by systematic, blinded review of EHRs. Following this, we revised the algorithm by refining the acceptable diagnoses and procedures. We ran the second algorithm on the same cohorts and determined the accuracy of the modification. RESULTS: CAD phenotype Algorithm I was 89.6% accurate, 94.6% sensitive, and 85.6% specific for group 1. After revising the algorithm (denoted CAD Algorithm II) and applying it to the same groups 1 and 2, sensitivity 98.2%, specificity 87.8%, and accuracy 92.4; accuracy 93% for group 2. Group 1 F1 score was 92.4%. Specific ICD-10 and CPT codes such as "coronary angiography through a vein graft" were more useful than generic terms. CONCLUSION: We have created an algorithm, CAD Algorithm II, that detects CAD on a large scale with high accuracy and sensitivity (recall). It has proven useful among varied patient populations. Use of this algorithm can extend to monitor a registry of patients in an EHR and/or to identify a group such as those with likely FH.

Deep learning for biological age estimation.

Modern machine learning techniques (such as deep learning) offer immense opportunities in the field of human biological aging research. Aging is a complex process, experienced by all living organisms. While traditional machine learning and data mining approaches are still popular in aging research, they typically need feature engineering or feature extraction for robust performance. Explicit feature engineering represents a major challenge, as it requires significant domain knowledge. The latest advances in deep learning provide a paradigm shift in eliciting meaningful knowledge from complex data without performing explicit feature engineering. In this article, we review the recent literature on applying deep learning in biological age estimation. We consider the current data modalities that have been used to study aging and the deep learning architectures that have been applied. We identify four broad classes of measures to quantify the performance of algorithms for biological age estimation and based on these evaluate the current approaches. The paper concludes with a brief discussion on possible future directions in biological aging research using deep learning. This study has significant potentials for improving our understanding of the health status of individuals, for instance, based on their physical activities, blood samples and body shapes. Thus, the results of the study could have implications in different health care settings, from palliative care to public health.

Cardiometabolic risk factors in siblings from a statewide screening program.

BACKGROUND: The Coronary Artery Risk Detection in Appalachian Communities (CARDIAC) Project is a state-wide risk factor screening program that operated in West Virginia for 19 years and screened more than 100,000 5th graders for obesity, hypertension, and dyslipidemia. OBJECTIVES: We investigated siblings in the CARDIAC Project to assess whether cardiometabolic risk factors (CMRFs) correlate in siblings. METHODS: We identified 12,053 children from 5752 families with lipid panel, blood pressure, and anthropometric data. A linkage application (LinkPlus from the U.S. Centers for Disease Control and Prevention) matched siblings based on parent names, addresses, telephone numbers, and school to generate a linkage probability curve. Graphical and statistical analyses demonstrate the relationships between CMRFs in siblings. RESULTS: Siblings showed moderate intraclass correlation coefficient of 0.375 for low-density lipoprotein cholesterol (LDL-C), 0.34 for high-density lipoprotein cholesterol (HDL-C), and 0.22 for triglyceride levels. The body mass index (BMI) intraclass correlation coefficient (0.383) is slightly better (2%) than LDL-C or HDL-C, but the standardized beta values from linear regression suggest a 3-fold impact of sibling LDL-C over the child's own BMI. The odds ratio of a second sibling having LDL-C < 110 mg/dL with a first sibling at that level is 3.444:1 (Confidence Limit 3.031-3.915, P < .05). The odds ratio of a sibling showing an LDL-C ≥ 160 mg/dL, given a first sibling with that degree of elevated LDL-C is 29.6:1 (95% Confidence Limit: 15.54-56.36). The individual LDL-C level correlated more strongly with sibling LDL-C than with the individual's own BMI. Seventy-eight children with LDL-C > 160 mg/dL and negative family history would have been missed, which represents more than half of those with LDL-C > 160 mg/dL (78 vs 67 or 54%). CONCLUSIONS: Sibling HDL-C levels, LDL-C levels, and BMIs correlate within a family. Triglyceride and blood pressure levels are less well correlated. The identified CMRF relationships strengthen the main findings of the overall CARDIAC Project: an elevated BMI is not predictive of elevated LDL-C and family history of coronary artery disease poorly predicts cholesterol abnormality at screening. Family history does not adequately identify children who should be screened for cholesterol abnormality. Elevated LDL-C (>160 mg/dL) in a child strongly suggests that additional siblings and parents be screened if universal screening is not practiced.

Foreign Body Aspiration Presenting as Pneumothorax in a Child.

A typical presentation of a foreign body aspiration (FBA) in a child includes witnessed choking, respiratory distress, cyanosis, coughing, wheezing, diminished breath sounds, and/or altered mental status. Following an extensive literature review, we found pneumothorax occurring secondary to FBA is a rare occurrence and should elicit prompt treatment. This 17-month-old female was admitted for respiratory syncytial virus (RSV) bronchiolitis and developed a subsequent pneumothorax during her hospital stay, consequent to aspiration of a cashew fragment two weeks before presentation. In light of the National Institute of Allergy and Infectious Diseases (NIAID)-sponsored expert panel's addended guidelines, published and endorsed by the American Academy of Pediatrics (AAP) in 2017, we highlight a potential complication of increasing encouragement of peanut consumption in children as young as four months.

Lessons From Starting a New Subspecialty in a Rural State.

In starting a new pediatric rheumatology service in a rural state, we designed the practice to focus on patient access, patient quality, and patient experience. We created a clinical experience that starts with an intake call to optimize the face-to-face visit. A team-based care approach is used. Weekend appointments are offered to avoid school and work absence. The social determinants of health are addressed. In our first year, our patients have reported their appreciation for a high-touch, patient-centered experience.

Epidemiology of abusive head trauma in West Virginia children <24 months: 2000-2010.

BACKGROUND: Few states have published statewide epidemiology of abusive head trauma (AHT). OBJECTIVE: To examine the statewide epidemiology of AHT in West Virginia (WV), with the primary objective of establishing AHT incidence for comparison to national data, and to use as a baseline for comparison to incidence post-implementation of a statewide AHT prevention program. PARTICIPANTS AND SETTING: AHT cases in children less than 2 years old were identified from the 3 tertiary pediatric centers in WV. METHODS: Cases were identified by using ICD-9 codes for initially identifying those with injuries which might be consistent with AHT, followed by medical record review to determine which of these met the criteria for inclusion as a case. Medical examiner data was used to find additional cases of AHT. Using the number of cases identified along with relevant census data, incidence of AHT was calculated. RESULTS: There were 120 cases of AHT treated in WV hospitals from 2000 to 2010, 100 of which were WV residents. The incidence was 36.1/100,000 children <1 year of age and was 21.9 cases per 100,000 children <2 years of age. Incidence in infants increased during the latter years (2006-2010) of the study to 51.8/100,000 compared to the incidence during 2000-2005, which was 24.0/100,000 (p < .01). CONCLUSIONS: Compared to US national, state and regional figures, the WV incidence of AHT was among the highest. In addition, the incidence of AHT increased significantly over the study period. Possible factors contributing to the rise in incidence are discussed.

An Electronic Health Record Data-driven Model for Identifying Older Adults at Risk of Unintentional Falls.

Screening for risk of unintentional falls remains low in the primary care setting because of the time constraints of brief office visits. National studies suggest that physicians caring for older adults provide recommended fall risk screening only 30 to 37 percent of the time. Given prior success in developing methods for repurposing electronic health record data for the identification of fall risk, this study involves building a model in which electronic health record data could be applied for use in clinical decision support to bolster screening by proactively identifying patients for whom screening would be beneficial and targeting efforts specifically to those patients. The final model, consisting of priority and extended measures, demonstrates moderate discriminatory power, indicating that it could prove useful in a clinical setting for identifying patients at risk of falls. Focus group discussions reveal important contextual issues involving the use of fall-related data and provide direction for the development of health systems-level innovations for the use of electronic health record data for fall risk identification.

LDL cholesterol level in fifth-grade schoolchildren associates with stature.

Short stature is associated with increased LDL-cholesterol levels and coronary artery disease in adults. We investigated the relationship of stature to LDL levels in children in the West Virginia Coronary Artery Risk Detection in Appalachian Communities (CARDIAC) Project to determine whether the genetically determined inverse relationship observed in adults would be evident in fifth graders. A cross-sectional survey of schoolchildren was assessed for cardiovascular risk factors. Data collected at school screenings over 18 years in WV schools were analyzed for 63,152 fifth-graders to determine relationship of LDL to stature with consideration of age, gender, and BMI. The first (shortest) quartile showed an LDL level of 93.6 mg/dl compared with an LDL level of 89.7 mg/dl for the fourth (tallest) quartile. Each incremental increase of 1 SD of height lowered LDL by 0.049 mg/dl (P < 0.0001). Multivariate analysis showed LDL to vary inversely as a function of the first (lowest) quartile of height after controlling for gender, median age, BMI percentile for age and gender, and year of screening. The odds ratio for LDL ≥ 130 mg/dl for shortest versus tallest quartile is 1.266 (95% CL 1.162-1.380). The odds ratio for LDL ≥ 160 mg/dl is 1.456 (95% CL 1.163-1.822). The relationship between short stature and LDL, noted in adults, is confirmed in childhood.

Comparison of Intraoperative Aminophylline Versus Furosemide in Treatment of Oliguria During Pediatric Cardiac Surgery.

OBJECTIVES: To determine if intraoperative aminophylline was superior to furosemide to prevent or attenuate postoperative cardiac surgery-associated acute kidney injury. DESIGN: Single-center, historical control, retrospective cohort study. SETTING: PICU, university-affiliated children's hospital. PATIENTS: Children with congenital heart disease in PICU who received furosemide or aminophylline to treat intraoperative oliguria. INTERVENTIONS: Intraoperative oliguria was treated either with furosemide (September 2007 to February 2012) or with aminophylline (February 2012 to June 2013). The postoperative 48 hours renal outcomes of the aminophylline group were compared with the furosemide group. The primary outcomes were acute kidney injury and renal replacement therapy use at 48 hours postoperatively. Surgical complexity was accounted for by the use of Risk Adjustment for Congenital Heart Surgery-1 score. MEASUREMENTS AND MAIN RESULTS: The study involves 69 months of observation. There were 200 cases younger than 21 years old reviewed for this study. Eighty-five cases (42.5%) developed acute kidney injury. The aminophylline group patients produced significantly more urine (mL/kg/hr) during the first 8 hours postoperatively than furosemide patients (5.1 vs 3.4 mL/kg/hr; p = 0.01). The urine output at 48 hours postoperatively was similar between the two groups. There was no difference in acute kidney injury incidence at 48 hours between the aminophylline and furosemide groups (38% vs 47%, respectively; p = 0.29). Fewer aminophylline group subjects required renal replacement therapy compared to the furosemide group subjects (n = 1 vs 7, respectively; p = 0.03). In the multi-variant predictive model, intraoperative aminophylline infusion was noted as a negative predictive factor for renal replacement therapy, but not for cardiac surgery-associated acute kidney injury. CONCLUSION: The intraoperative use of aminophylline was more effective than furosemide in reversal of oliguria in the early postoperative period. There were less renal replacement therapy-requiring acute kidney injury in children in the aminophylline group. Future prospective studies of intraoperative aminophylline to prevent cardiac surgery-associated acute kidney injury may be warranted.

Developing Methods of Repurposing Electronic Health Record Data for Identification of Older Adults at Risk of Unintentional Falls.

Nationally, nearly 40 percent of community-dwelling adults age 65 and older fall at least once a year, making unintentional falls the leading cause of both fatal and nonfatal injuries among this age group. Addressing this public health problem in primary care offers promise. However, challenges in incorporating fall risk screening into primary care result in a problem of missed opportunities for screening, counseling, intervention, and ultimately prevention. Given these barriers, this study examines the potential for the innovative use of routinely collected electronic health record data to provide enhanced clinical decision support in busy, often resource-thin primary care environments. Using de-identified data from a sample of West Virginia primary care centers, we find that it is both feasible and worthwhile to repurpose routinely collected data for the purpose of identification of older adults at risk of falls. Searching of both free-text and semistructured data was particularly valuable.

Machine learning-based coreference resolution of concepts in clinical documents.

OBJECTIVE: Coreference resolution of concepts, although a very active area in the natural language processing community, has not yet been widely applied to clinical documents. Accordingly, the 2011 i2b2 competition focusing on this area is a timely and useful challenge. The objective of this research was to collate coreferent chains of concepts from a corpus of clinical documents. These concepts are in the categories of person, problems, treatments, and tests. DESIGN: A machine learning approach based on graphical models was employed to cluster coreferent concepts. Features selected were divided into domain independent and domain specific sets. Training was done with the i2b2 provided training set of 489 documents with 6949 chains. Testing was done on 322 documents. RESULTS: The learning engine, using the un-weighted average of three different measurement schemes, resulted in an F measure of 0.8423 where no domain specific features were included and 0.8483 where the feature set included both domain independent and domain specific features. CONCLUSION: Our machine learning approach is a promising solution for recognizing coreferent concepts, which in turn is useful for practical applications such as the assembly of problem and medication lists from clinical documents.

Ketogenic diet: rapid onset of selenium deficiency-induced cardiac decompensation.

Selenium-deficiency cardiomyopathy is a known secondary complication from long-term treatment with a ketogenic diet for medical refractory epilepsy. Our patient, a 5-year-old boy on a ketogenic diet for intractable seizures, had a normal selenium level before starting the diet, but he shortly thereafter developed acute reversible cardiomyopathy and ventricular tachycardia, which was unmasked during a hospitalization for an elective operative procedure. His cardiomyopathy was suspected to be secondary to a selenium-deficient state and was confirmed by way of a markedly low serum selenium level and supported by rapid improvement with the initiation of selenium supplementation and cessation of the ketogenic diet. For patients being initiated on a ketogenic diet, current screening guidelines call for baseline and follow-up selenium levels every 3 months during the first year along with RDA selenium supplementation, which is 30 mcg/day. Most of the new ketogenic diet formulas meet this requirement. Our patient underwent elective surgery before his planned 3-month selenium level check and had potentially preventable complications. Secondary to this experience, we suggest that all patients initiated on a ketogenic diet should have a preoperative electrocardiogram (EKG), an echocardiogram, and selenium level determined before any elective surgery. These steps would prevent unnecessary perioperative morbidity and mortality.

Impact of a Multimodal Antimicrobial Stewardship Program on Pseudomonas aeruginosa Susceptibility and Antimicrobial Use in the Intensive Care Unit Setting.

Objective. To study the impact of our multimodal antibiotic stewardship program on Pseudomonas aeruginosa susceptibility and antibiotic use in the intensive care unit (ICU) setting. Methods. Our stewardship program employed the key tenants of published antimicrobial stewardship guidelines. These included prospective audits with intervention and feedback, formulary restriction with preauthorization, educational conferences, guidelines for use, antimicrobial cycling, and de-escalation of therapy. ICU antibiotic use was measured and expressed as defined daily doses (DDD) per 1,000 patient-days. Results. Certain temporal relationships between antibiotic use and ICU resistance patterns appeared to be affected by our antibiotic stewardship program. In particular, the ICU use of intravenous ciprofloxacin and ceftazidime declined from 148 and 62.5 DDD/1,000 patient-days to 40.0 and 24.5, respectively, during 2004 to 2007. An increase in the use of these agents and resistance to these agents was witnessed during 2008-2010. Despite variability in antibiotic usage from the stewardship efforts, we were overall unable to show statistical relationships with P. aeruginosa resistance rate. Conclusion. Antibiotic resistance in the ICU setting is complex. Multimodal stewardship efforts attempt to prevent resistance, but such programs clearly have their limits.

Acute subdural hematoma: potential soccer injury in an otherwise healthy child.

A 16-year-old adolescent boy presented with headache, dizziness, loss of consciousness, and a tonic-clonic seizure after heading a soccer ball in a competitive match. A computed tomographic scan of the head revealed an acute subdural hematoma with a mass effect. The patient was emergently referred to a tertiary care facility where he eventually recovered completely with conservative care. No predisposing medical conditions were found. To the best of our knowledge, this is the first report of an intracranial hemorrhage secondary to the heading of a soccer ball alone in an otherwise healthy child without any underlying predisposing central nervous system abnormalities.

Natural language processing framework to assess clinical conditions.

OBJECTIVE The authors developed a natural language processing (NLP) framework that could be used to extract clinical findings and diagnoses from dictated physician documentation. DESIGN De-identified documentation was made available by i2b2 Bio-informatics research group as a part of their NLP challenge focusing on obesity and its co-morbidities. The authors describe their approach, which used a combination of concept detection, context validation, and the application of a variety of rules to conclude patient diagnoses. RESULTS The framework was successful at correctly identifying diagnoses as judged by NLP challenge organizers when compared with a gold standard of physician annotations. The authors overall kappa values for agreement with the gold standard were 0.92 for explicit textual results and 0.91 for intuited results. The NLP framework compared favorably with those of the other entrants, placing third in textual results and fourth in intuited results in the i2b2 competition. CONCLUSIONS The framework and approach used to detect clinical conditions was reasonably successful at extracting 16 diagnoses related to obesity. The system and methodology merits further development, targeting clinically useful applications.

Assessment of commercial NLP engines for medication information extraction from dictated clinical notes.

PURPOSE: We assessed the current state of commercial natural language processing (NLP) engines for their ability to extract medication information from textual clinical documents. METHODS: Two thousand de-identified discharge summaries and family practice notes were submitted to four commercial NLP engines with the request to extract all medication information. The four sets of returned results were combined to create a comparison standard which was validated against a manual, physician-derived gold standard created from a subset of 100 reports. Once validated, the individual vendor results for medication names, strengths, route, and frequency were compared against this automated standard with precision, recall, and F measures calculated. RESULTS: Compared with the manual, physician-derived gold standard, the automated standard was successful at accurately capturing medication names (F measure=93.2%), but performed less well with strength (85.3%) and route (80.3%), and relatively poorly with dosing frequency (48.3%). Moderate variability was seen in the strengths of the four vendors. The vendors performed better with the structured discharge summaries than with the clinic notes in an analysis comparing the two document types. CONCLUSION: Although automated extraction may serve as the foundation for a manual review process, it is not ready to automate medication lists without human intervention.