ABSTRACT
OBJECTIVES: To assess the relationship between prior exposure to immune checkpoint inhibitors (ICIs) and the risk of postoperative complications in cancer patients. DESIGN: Single-center retrospective cohort study. INTERVENTIONS: The main exposure was treatment with an FDA-approved ICI within 6 months before surgery. MEASUREMENTS AND MAIN RESULTS: Exposure to ICIs and covariates was determined from the electronic health record. The primary outcome was a composite of postoperative complications, including prolonged pressor or oxygen dependence, kidney injury, or myocardial injury. Secondary outcomes included each subcomponent of the primary outcome. Of 7674 subjects with cancer admitted to the ICU after surgery, 247 were exposed to one or more ICIs in the 6 months before surgery. After propensity score matching, 197 ICI-exposed subjects were matched to 777 nonexposed. The composite outcome occurred in 70 of 197 (35.5%) ICI-exposed subjects and 251 of 777 (32.3%) nonexposed. There was no difference between exposed and nonexposed groups in the primary composite outcome (odds ratio [OR], 1.12; 95% CI, 0.80-1.58) by conditional logistic regression. Risk of the secondary outcome of prolonged pressor dependence was significantly higher in ICI-exposed subjects (OR, 1.64; 95% CI, 1.01-2.67). Risks of oxygen dependence (OR, 1.13; 95% CI, 0.75-1.73), kidney injury (OR, 1.15; 95% CI, 0.77-1.71), and myocardial injury (OR, 1.76; 95% CI, 1.00-3.10) were not significantly different. There was no difference between groups in the time to hospital discharge alive (p = 0.62). CONCLUSIONS: Exposure to ICIs within 6 months before high-risk surgery was not associated with the composite outcome of cardiopulmonary instability or organ injury in patients with cancer. The potential for an association with the secondary outcomes of cardiac instability and injury is worthy of future study.
ABSTRACT
BACKGROUND: With recent prioritization of equity in pediatric health outcomes, a shift to examine neighborhood-level health care disparities within pediatric populations has occurred, specifically in the context of critical illness. RESEARCH QUESTION: Does an association exist between individual indicators of neighborhood-level disadvantage and incidence of PICU admission? STUDY DESIGN AND METHODS: Pediatric patients younger than 18 years admitted to a PICU in a large urban tertiary pediatric hospital from January 1, 2016, through December 31, 2019, with a residential address in the city of Baltimore or Baltimore County on the day of admission were included in this ecological study. Demographic and clinical characteristics of children admitted to the PICU were summarized, with the primary outcome being PICU admission. Unadjusted negative binomial regression was used to examine the association between census tract-level PICU admissions and the previously described census tract-level indicators of neighborhood socioeconomic position. Regression models included an offset term for the population younger than 18 years for each census tract; results of models are reported as incidence rate ratios (IRRs) with corresponding 95% CIs. RESULTS: We identified 2,476 PICU admissions: 1,351 patients from the city of Baltimore (10.25 per 1,000 children) and 1,125 patients from Baltimore County (6.31 per 1,000 children). Most PICU admissions (n = 906 [68%]) for the city of Baltimore represented an area deprivation index (ADI) of > 60, whereas most Baltimore County PICU admissions (n = 919 [82.3%]) represented an ADI of < 60. At the neighborhood level, the percentage of families living below the poverty line was associated with greater incidence of PICU admission in the city of Baltimore (IRR, 1.09; 95% CI, 1.00-1.18) and Baltimore County (IRR, 1.19; 95% CI, 1.05-1.36). For every $10,000 increase in median household income, PICU admission rates dropped by 9% for the city of Baltimore (IRR, 0.91; 95% CI, 0.86-0.95) and Baltimore County (IRR, 0.91; 95% CI, 0.88-0.94). Neighborhoods with vacant housing units also were associated with a higher incidence of PICU admission in the city of Baltimore (IRR, 1.10; 95% CI, 1.01-1.21) and Baltimore County (IRR, 1.46; 95% CI, 1.21-1.77), as was a 10% increase in occupied homes without vehicles (city of Baltimore: IRR, 1.14; 95% CI, 1.07-1.21; Baltimore County: IRR, 1.23; 95% CI, 1.11-1.37). INTERPRETATION: Health outcomes of pediatric critical illness should be examined in the context of structural determinants of health, including neighborhood-level and environmental characteristics.
Subject(s)
Critical Illness , Residence Characteristics , Child , Humans , Critical Illness/epidemiology , Critical Illness/therapy , Poverty , Hospitalization , IncomeABSTRACT
Introduction: Mechanical ventilation is a life-saving treatment in the Intensive Care Unit (ICU), but often causes patients to be at risk of further respiratory complication. We created a statistical model utilizing electronic health record and physiologic vitals data to predict the Center for Disease Control and Prevention (CDC) defined Ventilator Associated Complications (VACs). Further, we evaluated the effect of data temporal resolution and feature generation method choice on the accuracy of such a constructed model. Methods: We constructed a random forest model to predict occurrence of VACs using health records and chart events from adult patients in the Medical Information Mart for Intensive Care III (MIMIC-III) database. We trained the machine learning models on two patient populations of 1921 and 464 based on low and high frequency data availability. Model features were generated using both basic statistical summaries and tsfresh, a python library that generates a large number of derived time-series features. Classification to determine whether a patient will experience VAC one hour after 35 h of ventilation was performed using a random forest classifier. Two different sample spaces conditioned on five varying feature extraction techniques were evaluated to identify the most optimal selection of features resulting in the best VAC discrimination. Each dataset was assessed using K-folds cross-validation (k = 10), giving average area under the receiver operating characteristic curves (AUROCs) and accuracies. Results: After feature selection, hyperparameter tuning, and feature extraction, the best performing model used automatically generated features on high frequency data and achieved an average AUROC of 0.83 ± 0.11 and an average accuracy of 0.69 ± 0.10. Discussion: Results show the potential viability of predicting VACs using machine learning, and indicate that higher-resolution data and the larger feature set generated by tsfresh yield better AUROCs compared to lower-resolution data and manual statistical features.
ABSTRACT
OBJECTIVES: There is variation in microbiology testing among PICU patients with fever offering opportunities to reduce avoidable testing and treatment. Our objective is to describe the development and assess the impact of a novel comprehensive testing algorithm to support judicious testing practices and expanded diagnostic differentials for PICU patients with new fever or instability. DESIGN: A mixed-methods quality improvement study. SETTING: Single-center academic PICU and pediatric cardiac ICU. SUBJECTS: Admitted PICU patients and physicians. INTERVENTIONS: A multidisciplinary team developed a clinical decision-support algorithm. MEASUREMENTS AND MAIN RESULTS: We evaluated blood, endotracheal, and urine cultures, urinalyses, and broad-spectrum antibiotic use per 1,000 ICU patient-days using statistical process control charts and incident rate ratios (IRRs) and assessed clinical outcomes 24 months pre- and 18 months postimplementation. We surveyed physicians weekly for 12 months postimplementation. Blood cultures declined by 17% (IRR, 0.83; 95% CI, 0.77-0.89), endotracheal cultures by 26% (IRR, 0.74; 95% CI, 0.63-0.86), and urine cultures by 36% (IRR, 0.64; 95% CI, 0.56-0.73). There was an anticipated rise in urinalysis testing by 23% (IRR, 1.23; 95% CI, 1.14-1.33). Despite higher acuity and fewer brief hospitalizations, mortality, hospital, and PICU readmissions were stable, and PICU length of stay declined. Of the 108 physician surveys, 46 replied (43%), and 39 (85%) recently used the algorithm; 0 reported patient safety concerns, two (4%) provided constructive feedback, and 28 (61%) reported the algorithm improved patient care. CONCLUSIONS: A comprehensive fever algorithm was associated with reductions in blood, endotracheal, and urine cultures and anticipated increase in urinalyses. We detected no patient harm, and physicians reported improved patient care.
Subject(s)
Physicians , Trachea , Child , Humans , Infant , Surveys and Questionnaires , Hospitalization , Length of Stay , Intensive Care Units, PediatricABSTRACT
BACKGROUND: Cardiac arrest is a leading cause of mortality prior to discharge for children admitted to the pediatric intensive care unit. To address this problem, we used machine learning to predict cardiac arrest up to three hours in advance. METHODS: Our data consists of 240 Hz ECG waveform data, 0.5 Hz physiological time series data, medications, and demographics from 1,145 patients in the pediatric intensive care unit at the Johns Hopkins Hospital, 15 of whom experienced a cardiac arrest. The data were divided into training, validating, and testing sets, and features were generated every five minutes. 23 heart rate variability (HRV) metrics were determined from ECG waveforms. 96 summary statistics were calculated for 12 vital signs, such as respiratory rate and blood pressure. Medications were classified into 42 therapeutic drug classes. Binary features were generated to indicate the administration of these different drugs. Next, six machine learning models were evaluated: logistic regression, support vector machine, random forest, XGBoost, LightGBM, and a soft voting ensemble. RESULTS: XGBoost performed the best, with 0.971 auROC, 0.797 auPRC, 99.5% sensitivity, and 69.6% specificity on an independent test set. CONCLUSION: We have created high-performing models that identify signatures of in-hospital cardiac arrest (IHCA) that may not be evident to clinicians. These signatures include a combination of heart rate variability metrics, vital signs data, and therapeutic drug classes. These machine learning models can predict IHCA up to three hours prior to onset with high performance, allowing clinicians to intervene earlier, improving patient outcomes.
Subject(s)
Heart Arrest , Child , Humans , Pilot Projects , Intensive Care Units, Pediatric , Vital Signs , Machine Learning , Intensive Care UnitsABSTRACT
BACKGROUND: Acute kidney injury (AKI) after major noncardiac surgery is commonly attributed to cardiovascular dysfunction. Identifying novel associations between preoperative cardiovascular markers and kidney injury may guide risk stratification and perioperative intervention. Increased left ventricular relative wall thickness (RWT), routinely measured on echocardiography, is associated with myocardial dysfunction and long-term risk of heart failure in patients with preserved left ventricular ejection fraction (LVEF); however, its relationship to postoperative complications has not been studied. We evaluated the association between preoperative RWT and AKI in high-risk noncardiac surgical patients with preserved LVEF. METHODS: Patients ≥18 years of age having major noncardiac surgery (high-risk elective intra-abdominal or noncardiac intrathoracic surgery) between July 1, 2016, and June 30, 2018, who had transthoracic echocardiography in the previous 12 months were eligible. Patients with preoperative creatinine ≥2 mg/dL or reduced LVEF (<50%) were excluded. The association between RWT and AKI, defined as an increase in serum creatinine by 0.3 mg/dL from baseline within 48 hours or by 50% within 7 days after surgery, was assessed using multivariable logistic regression adjusted for preoperative covariates. An additional model adjusted for intraoperative covariates, which are strongly associated with AKI, especially hypotension. RWT was modeled continuously, associating the change in odds of AKI for each 0.1 increase in RWT. RESULTS: The study included 1041 patients (mean ± standard deviation [SD] age 62 ± 15 years; 59% female). A total of 145 subjects (13.9%) developed AKI within 7 days. For RWT quartiles 1 through 4, respectively, 20 of 262 (7.6%), 40 of 259 (15.4%), 39 of 263 (14.8%), and 46 of 257 (17.9%) developed AKI. Log-odds and proportion with AKI increased across the observed RWT values. After adjusting for confounders (demographics, American Society of Anesthesiologists [ASA] physical status, comorbidities, baseline creatinine, antihypertensive medications, and left ventricular mass index), each RWT increase of 0.1 was associated with an estimated 26% increased odds of developing AKI (odds ratio [OR]; 95% confidence interval [CI]) of 1.26 (1.09-1.46; P = .002). After adjusting for intraoperative covariates (length of surgery, presence of an arterial line, intraoperative hypotension, crystalloid administration, transfusion, and urine output), RWT remained independently associated with the odds of AKI (OR; 95% CI) of 1.28 (1.13-1.47; P = .001). Increased RWT was also independently associated with hospital length of stay and adjusted hazard ratio (HR [95% CI]) of 0.94 (0.89-0.99; P = .018). CONCLUSIONS: Left ventricular RWT is a novel cardiovascular factor associated with AKI within 7 days after high-risk noncardiac surgery among patients with preserved LVEF. Application of this commonly available measurement of risk stratification or perioperative intervention warrants further investigation.
Subject(s)
Acute Kidney Injury , Hypotension , Acute Kidney Injury/diagnosis , Acute Kidney Injury/epidemiology , Acute Kidney Injury/etiology , Creatinine , Female , Humans , Hypotension/complications , Male , Postoperative Complications/epidemiology , Postoperative Complications/etiology , Retrospective Studies , Risk Factors , Stroke Volume , Ventricular Function, LeftABSTRACT
Background: High flow nasal cannula (HFNC) is commonly used as non-invasive respiratory support in critically ill children. There are limited data to inform consensus on optimal device parameters, determinants of successful patient response, and indications for escalation of support. Clinical scores, such as the respiratory rate-oxygenation (ROX) index, have been described as a means to predict HFNC non-response, but are limited to evaluating for escalations to invasive mechanical ventilation (MV). In the presence of apparent HFNC non-response, a clinician may choose to increase the HFNC flow rate to hypothetically prevent further respiratory deterioration, transition to an alternative non-invasive interface, or intubation for MV. To date, no models have been assessed to predict subsequent escalations of HFNC flow rates after HFNC initiation. Objective: To evaluate the abilities of tree-based machine learning algorithms to predict HFNC flow rate escalations. Methods: We performed a retrospective, cohort study assessing children admitted for acute respiratory failure under 24 months of age placed on HFNC in the Johns Hopkins Children's Center pediatric intensive care unit from January 2019 through January 2020. We excluded encounters with gaps in recorded clinical data, encounters in which MV treatment occurred prior to HFNC, and cases electively intubated in the operating room. The primary study outcome was discriminatory capacity of generated machine learning algorithms to predict HFNC flow rate escalations as compared to each other and ROX indices using area under the receiver operating characteristic (AUROC) analyses. In an exploratory fashion, model feature importance rankings were assessed by comparing Shapley values. Results: Our gradient boosting model with a time window of 8 h and lead time of 1 h before HFNC flow rate escalation achieved an AUROC with a 95% confidence interval of 0.810 ± 0.003. In comparison, the ROX index achieved an AUROC of 0.525 ± 0.000. Conclusion: In this single-center, retrospective cohort study assessing children under 24 months of age receiving HFNC for acute respiratory failure, tree-based machine learning models outperformed the ROX index in predicting subsequent flow rate escalations. Further validation studies are needed to ensure generalizability for bedside application.
Subject(s)
Respiratory Distress Syndrome , Child , Humans , Lung , Respiration, Artificial , Respiratory Distress Syndrome/therapyABSTRACT
OBJECTIVES: Sepsis and septic shock are leading causes of in-hospital mortality. Timely treatment is crucial in improving patient outcome, yet treatment delays remain common. Early prediction of those patients with sepsis who will progress to its most severe form, septic shock, can increase the actionable window for interventions. We aim to extend a time-evolving risk score, previously developed in adult patients, to predict pediatric sepsis patients who are likely to develop septic shock before its onset, and to determine whether or not these risk scores stratify into groups with distinct temporal evolution once this prediction is made. DESIGN: Retrospective cohort study. SETTING: Academic medical center from July 1, 2016, to December 11, 2020. PATIENTS: Six-thousand one-hundred sixty-one patients under 18 admitted to the Johns Hopkins Hospital PICU. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We trained risk models to predict impending transition into septic shock and compute time-evolving risk scores representative of a patient's probability of developing septic shock. We obtain early prediction performance of 0.90 area under the receiver operating curve, 43% overall positive predictive value, patient-specific positive predictive value as high as 62%, and an 8.9-hour median early warning time using Sepsis-3 labels based on age-adjusted Sequential Organ Failure Assessment score. Using spectral clustering, we stratified pediatric sepsis patients into two clusters differing in septic shock prevalence, mortality, and proportion of patients adequately fluid resuscitated. CONCLUSIONS: We demonstrate the applicability of our methodology for early prediction and stratification for risk of septic shock in pediatric sepsis patients. Through analyses of risk score evolution over time, we corroborate our past finding of an abrupt transition preceding onset of septic shock in children and are able to stratify pediatric sepsis patients using their risk score trajectories into low and high-risk categories.
ABSTRACT
BACKGROUND: Clinicians commonly obtain endotracheal aspirate cultures (EACs) in the evaluation of suspected ventilator-associated infections. However, bacterial growth in EACs does not distinguish bacterial colonization from infection and may lead to overtreatment with antibiotics. We describe the development and impact of a clinical decision support algorithm to standardize the use of EACs from ventilated PICU patients. METHODS: We monitored EAC use using a statistical process control chart. We compared the rate of EACs using Poisson regression and a quasi-experimental interrupted time series model and assessed clinical outcomes 1 year before and after introduction of the algorithm. RESULTS: In the preintervention year, there were 557 EACs over 5092 ventilator days; after introduction of the algorithm, there were 234 EACs over 3654 ventilator days (an incident rate of 10.9 vs 6.5 per 100 ventilator days). There was a 41% decrease in the monthly rate of EACs (incidence rate ratio [IRR]: 0.59; 95% confidence interval [CI] 0.51-0.67; P < .001). The interrupted time series model revealed a preexisting 2% decline in the monthly culture rate (IRR: 0.98; 95% CI 0.97-1.0; P = .01), immediate 44% drop (IRR: 0.56; 95% CI 0.45-0.70; P = .02), and stable rate in the postintervention year (IRR: 1.03; 95% CI 0.99-1.07; P = .09). In-hospital mortality, hospital length of stay, 7-day readmissions, and All Patients Refined Diagnosis Related Group severity and mortality scores were stable. The estimated direct cost savings was $26 000 per year. CONCLUSIONS: A clinical decision support algorithm standardizing EAC obtainment from ventilated PICU patients was associated with a sustained decline in the rate of EACs, without changes in mortality, readmissions, or length of stay.