Development and validation of a machine learning model using electronic health records to predict trauma- and stressor-related psychiatric disorders after hospitalization with sepsis.

A significant minority of individuals develop trauma- and stressor-related disorders (TSRD) after surviving sepsis, a life-threatening immune response to infections. Accurate prediction of risk for TSRD can facilitate targeted early intervention strategies, but many existing models rely on research measures that are impractical to incorporate to standard emergency department workflows. To increase the feasibility of implementation, we developed models that predict TSRD in the year after survival from sepsis using only electronic health records from the hospitalization (n = 217,122 hospitalizations from 2012-2015). The optimal model was evaluated in a temporally independent prospective test sample (n = 128,783 hospitalizations from 2016-2017), where patients in the highest-risk decile accounted for nearly one-third of TSRD cases. Our approach demonstrates that risk for TSRD after sepsis can be stratified without additional assessment burden on clinicians and patients, which increases the likelihood of model implementation in hospital settings.

The impact of timing of initiating invasive mechanical ventilation in COVID-19-related respiratory failure.

PURPOSE: Optimal timing of initiating invasive mechanical ventilation (IMV) in coronavirus disease 2019 (COVID-19)-related respiratory failure is unclear. We hypothesized that a strategy of IMV as opposed to continuing high flow oxygen or non-invasive mechanical ventilation each day after reaching a high FiO2 threshold would be associated with worse in-hospital mortality. METHODS: Using data from Kaiser Permanente Northern/Southern California's 36 medical centers, we identified patients with COVID-19-related acute respiratory failure who reached ≥80% FiO2 on high flow nasal cannula or non-invasive ventilation. Exposure was IMV initiation each day after reaching high FiO2 threshold (T0). We developed propensity scores with overlap weighting for receipt of IMV each day adjusting for confounders. We reported relative risk of inpatient death with 95% Confidence Interval. RESULTS: Of 28,035 hospitalizations representing 21,175 patient-days, 5758 patients were included (2793 received and 2965 did not receive IMV). Patients receiving IMV had higher unadjusted mortality (63.6% versus 18.2%, P < 0.0001). On each day after reaching T0 through day >10, the adjusted relative risk was higher for those receiving IMV compared to those not receiving IMV (Relative Risk>1). CONCLUSIONS: Initiation of IMV on each day after patients reach high FiO2 threshold was associated with higher inpatient mortality after adjusting for time-varying confounders. Remaining on high flow nasal cannula or non-invasive ventilation does not appear to be harmful compared to IMV. Prospective evaluation is needed.

Anakinra or high-dose corticosteroids in COVID-19 pneumonia patients who deteriorate on low-dose dexamethasone: an observational study of comparative effectiveness.

OBJECTIVES: To assess whether escalating to high-dose corticosteroids or anakinra compared with continuing low-dose corticosteroids reduced mortality in patients with severe COVID-19 whose respiratory function deteriorated while receiving dexamethasone 6 mg daily. METHODS: We conducted a retrospective cohort study between March 1 to December 31, 2020, of hospitalized patients with confirmed COVID-19 pneumonia. In-hospital death was analyzed using logistic regression with inverse probability of treatment weighting of receiving anakinra, high-dose corticosteroid (dexamethasone >10 mg daily), or remaining on low-dose corticosteroids on the day of first respiratory deterioration. RESULTS: We analyzed 6671 patients whose respiratory status deteriorated while receiving dexamethasone 6 mg daily for COVID-19 pneumonia, of whom 6265 stayed on low-dose corticosteroids, 232 were escalated to high-dose corticosteroids, and 174 to anakinra in addition to corticosteroids. The propensity score-adjusted odds of death were higher in the anakinra (odds ratio [OR] 1.76; 95% CI 1.13-2.72) and high-dose corticosteroid groups (OR 1.53; 95% CI 1.14-2.07) compared with those who continued low-dose corticosteroids on the day of respiratory deterioration. The odds of hospital-acquired infections were also higher in the anakinra (OR 2.00; 95% CI 1.28-3.11) and high-dose corticosteroid groups (OR 1.43; 95% CI 1.00-2.04) compared with low-dose corticosteroid group. CONCLUSION: Our findings do not support escalating patients with COVID-19 pneumonia who deteriorate on low-dose corticosteroids to high-dose corticosteroids or anakinra.

The intensity of anticoagulant dosing in hospitalized patients with COVID-19: An observational, comparative effectiveness study.

BACKGROUND: The question of anticoagulant dosing in hospitalized patients with coronavirus disease-2019 (COVID-19) is unresolved, with randomized trials showing mixed results and heterogeneity of treatment effects for in-hospital death. OBJECTIVE: To examine the association between the intensity of anticoagulation and clinical outcomes in hospitalized patients with COVID-19. DESIGN, SETTING AND PARTICIPANTS: Retrospective cohort study of patients with COVID-19 and respiratory impairment who were hospitalized between 3/1/2020-12/31/2020 in two Kaiser Permanente regions. EXPOSURE AND MAIN OUTCOME: We fit propensity score models using categorical regression to estimate the probability of receiving standard prophylactic, intermediate, or full-dose anticoagulation beginning on the day of admission or on the day of first respiratory deterioration. Exposure was defined by the highest dose on the day of admission or within 24 hours after deterioration. The primary outcome was in-hospital death. RESULTS: We included 17,130 patients in the day of admission analysis and 4,924 patients who experienced respiratory deterioration. There were no differences in propensity score-adjusted odds of in-hospital death for patients who received either intermediate (odds ratio [OR]: 1.00, 95% confidence intervals [CI] 0.89-1.12) or full anticoagulation (OR: 1.00, 95% CI: 0.85-1.17) compared with standard prophylaxis beginning on the day of admission. Similarly, there were no differences in in-hospital death for either intermediate (OR: 1.22, 95% CI: 0.82-1.82) or full anticoagulation (OR: 1.50, 95% CI: 0.90-2.51) compared with standard prophylaxis on the day of deterioration. CONCLUSION: Results of this real-world, comparative effectiveness study showed no differences in in-hospital death among newly admitted or deteriorating patients with COVID-19 who received intermediate-dose or full anticoagulation compared with standard prophylaxis.

High-dose corticosteroids in patients hospitalized for COVID-19 pneumonia: an observational study of comparative effectiveness.

OBJECTIVES: To assess whether high- compared with low-dose corticosteroids started upon hospitalization reduce mortality in patients with severe COVID-19 pneumonia or in subgroups stratified by severity of respiratory impairment on admission. METHODS: We conducted a retrospective cohort study of patients with confirmed SARS-CoV-2 infection who required oxygen supplementation upon hospitalization between March 1 and December 31, 2020. In-hospital death was analyzed using logistic regression with inverse probability of treatment weighting of receiving low- or high-dose corticosteroid (dexamethasone 6-10 mg daily or >10-20 mg daily or other corticosteroid equivalents). RESULTS: We analyzed 13,366 patients who received low-dose and 948 who received high-dose corticosteroids, of whom 31.3% and 40.4% had severe respiratory impairment (>15 l/min of oxygen or mechanical ventilation) upon admission, respectively. There were no differences in the propensity score-adjusted odds of death (odds ratio 1.17, 95% CI 0.72-1.90) or infections (odds ratio 0.70, 95% CI 0.44-1.11) for patients who received high-dose compared with low-dose corticosteroids, beginning on the day of admission. No significant differences in subgroups stratified by severity of respiratory impairment were found. CONCLUSION: Initiating high-dose compared with low-dose corticosteroids among newly hospitalized patients with COVID-19 pneumonia did not improve survival. However, benefit of high-dose corticosteroids in specific subgroups cannot be excluded.

Quantifying the breadth of antibiotic exposure in sepsis and suspected infection using spectrum scores.

A retrospective cohort study. Studies to quantify the breadth of antibiotic exposure across populations remain limited. Therefore, we applied a validated method to describe the breadth of antimicrobial coverage in a multicenter cohort of patients with suspected infection and sepsis. We conducted a retrospective cohort study across 21 hospitals within an integrated healthcare delivery system of patients admitted to the hospital through the ED with suspected infection or sepsis and receiving antibiotics during hospitalization from January 1, 2012, to December 31, 2017. We quantified the breadth of antimicrobial coverage using the Spectrum Score, a numerical score from 0 to 64, in patients with suspected infection and sepsis using electronic health record data. Of 364,506 hospital admissions through the emergency department, we identified 159,004 (43.6%) with suspected infection and 205,502 (56.4%) with sepsis. Inpatient mortality was higher among those with sepsis compared to those with suspected infection (8.4% vs 1.2%; P < .001). Patients with sepsis had higher median global Spectrum Scores (43.8 [interquartile range IQR 32.0-49.5] vs 43.5 [IQR 26.8-47.2]; P < .001) and additive Spectrum Scores (114.0 [IQR 57.0-204.5] vs 87.5 [IQR 45.0-144.8]; P < .001) compared to those with suspected infection. Increased Spectrum Scores were associated with inpatient mortality, even after covariate adjustments (adjusted odds ratio per 10-point increase in Spectrum Score 1.31; 95%CI 1.29-1.33). Spectrum Scores quantify the variability in antibiotic breadth among individual patients, between suspected infection and sepsis populations, over the course of hospitalization, and across infection sources. They may play a key role in quantifying the variation in antibiotic prescribing in patients with suspected infection and sepsis.

Unsupervised probabilistic models for sequential Electronic Health Records.

We develop an unsupervised probabilistic model for heterogeneous Electronic Health Record (EHR) data. Utilizing a mixture model formulation, our approach directly models sequences of arbitrary length, such as medications and laboratory results. This allows for subgrouping and incorporation of the dynamics underlying heterogeneous data types. The model consists of a layered set of latent variables that encode underlying structure in the data. These variables represent subject subgroups at the top layer, and unobserved states for sequences in the second layer. We train this model on episodic data from subjects receiving medical care in the Kaiser Permanente Northern California integrated healthcare delivery system. The resulting properties of the trained model generate novel insight from these complex and multifaceted data. In addition, we show how the model can be used to analyze sequences that contribute to assessment of mortality likelihood.

Evaluation of Incident 7-Day Infection and Sepsis Hospitalizations in an Integrated Health System.

Rationale: Prehospital opportunities to predict infection and sepsis hospitalization may exist, but little is known about their incidence following common healthcare encounters. Objectives: To evaluate the incidence and timing of infection and sepsis hospitalization within 7 days of living hospital discharge, emergency department discharge, and ambulatory visit settings. Methods: In each setting, we identified patients in clinical strata based on the presence of infection and severity of illness. We estimated number needed to evaluate values with hypothetical predictive model operating characteristics. Results: We identified 97,614,228 encounters, including 1,117,702 (1.1%) hospital discharges, 4,635,517 (4.7%) emergency department discharges, and 91,861,009 (94.1%) ambulatory visits between 2012 and 2017. The incidence of 7-day infection hospitalization varied from 37,140 (3.3%) following inpatient discharge to 50,315 (1.1%) following emergency department discharge and 277,034 (0.3%) following ambulatory visits. The incidence of 7-day infection hospitalization was increased for inpatient discharges with high readmission risk (10.0%), emergency department discharges with increased acute or chronic severity of illness (3.5% and 4.7%, respectively), and ambulatory visits with acute infection (0.7%). The timing of 7-day infection and sepsis hospitalizations differed across settings with an early rise following ambulatory visits, a later peak following emergency department discharges, and a delayed peak following inpatient discharge. Theoretical number needed to evaluate values varied by strata, but following hospital and emergency department discharge, were as low as 15-25. Conclusions: Incident 7-day infection and sepsis hospitalizations following encounters in routine healthcare settings were surprisingly common and may be amenable to clinical predictive models.

Association of Positive Fluid Balance at Discharge After Sepsis Management With 30-Day Readmission.

Importance: Although early fluid administration has been shown to lower sepsis mortality, positive fluid balance has been associated with adverse outcomes. Little is known about associations in non-intensive care unit settings, with growing concern about readmission from excess fluid accumulation in patients with sepsis. Objective: To evaluate whether positive fluid balance among non-critically ill patients with sepsis was associated with increased readmission risk, including readmission for heart failure. Design, Setting, and Participants: This multicenter retrospective cohort study was conducted between January 1, 2012, and December 31, 2017, among 57 032 non-critically ill adults hospitalized for sepsis at 21 hospitals across Northern California. Kaiser Permanente Northern California is an integrated health care system with a community-based population of more than 4.4 million members. Statistical analysis was performed from January 1 to December 31, 2019. Exposures: Intake and output net fluid balance (I/O) measured daily and cumulatively at discharge (positive vs negative). Main Outcomes and Measures: The primary outcome was 30-day readmission. The secondary outcomes were readmission stratified by category and mortality after living discharge. Results: The cohort included 57 032 patients who were hospitalized for sepsis (28 779 women [50.5%]; mean [SD] age, 73.7 [15.5] years). Compared with patients with positive I/O (40 940 [71.8%]), those with negative I/O (16 092 [28.2%]) were older, with increased comorbidity, acute illness severity, preexisting heart failure or chronic kidney disease, diuretic use, and decreased fluid administration volume. During 30-day follow-up, 8719 patients (15.3%) were readmitted and 3639 patients (6.4%) died. There was no difference in readmission between patients with positive vs negative I/O (HR, 1.00; 95% CI, 0.95-1.05). No association was detected between readmission and I/O using continuous, splined, and quadratic function transformations. Positive I/O was associated with decreased heart failure-related readmission (HR, 0.80 [95% CI, 0.71-0.91]) and increased 30-day mortality (HR, 1.23 [95% CI, 1.15-1.31]). Conclusions and Relevance: In this large observational study of non-critically ill patients hospitalized with sepsis, there was no association between positive fluid balance at the time of discharge and readmission. However, these findings may have been limited by variable recording and documentation of fluid intake and output; additional studies are needed to examine the association of fluid status with outcomes in patients with sepsis to reduce readmission risk.

The Presentation, Pace, and Profile of Infection and Sepsis Patients Hospitalized Through the Emergency Department: An Exploratory Analysis.

To characterize the signs and symptoms of sepsis, compare them with those from simple infection and other emergent conditions and evaluate their association with hospital outcomes. DESIGN SETTING PARTICIPANTS AND INTERVENTION: A multicenter, retrospective cohort study of 408,377 patients hospitalized through the emergency department from 2012 to 2017 with sepsis, suspected infection, heart failure, or stroke. Infected patients were identified based on Sepsis-3 criteria, whereas noninfected patients were identified through diagnosis codes. MEASUREMENTS AND MAIN RESULTS: Signs and symptoms were identified within physician clinical documentation in the first 24 hours of hospitalization using natural language processing. The time of sign and symptom onset prior to presentation was quantified, and sign and symptom prevalence was assessed. Using multivariable logistic regression, the association of each sign and symptom with four outcomes was evaluated: sepsis versus suspected infection diagnosis, hospital mortality, ICU admission, and time of first antibiotics (> 3 vs ≤ 3 hr from presentation). A total of 10,825 signs and symptoms were identified in 6,148,348 clinical documentation fragments. The most common symptoms overall were as follows: dyspnea (35.2%), weakness (27.2%), altered mental status (24.3%), pain (23.9%), cough (19.7%), edema (17.8%), nausea (16.9%), hypertension (15.6%), fever (13.9%), and chest pain (12.1%). Compared with predominant signs and symptoms in heart failure and stroke, those present in infection were heterogeneous. Signs and symptoms indicative of neurologic dysfunction, significant respiratory conditions, and hypotension were strongly associated with sepsis diagnosis, hospital mortality, and intensive care. Fever, present in only a minority of patients, was associated with improved mortality (odds ratio, 0.67, 95% CI, 0.64-0.70; p < 0.001). For common symptoms, the peak time of symptom onset before sepsis was 2 days, except for altered mental status, which peaked at 1 day prior to presentation. CONCLUSIONS: The clinical presentation of sepsis was heterogeneous and occurred with rapid onset prior to hospital presentation. These findings have important implications for improving public education, clinical treatment, and quality measures of sepsis care.

Nonstationary multivariate Gaussian processes for electronic health records.

Advances in the modeling and analysis of electronic health records (EHR) have the potential to improve patient risk stratification, leading to better patient outcomes. The modeling of complex temporal relations across the multiple clinical variables inherent in EHR data is largely unexplored. Existing approaches to modeling EHR data often lack the flexibility to handle time-varying correlations across multiple clinical variables, or they are too complex for clinical interpretation. Therefore, we propose a novel nonstationary multivariate Gaussian process model for EHR data to address the aforementioned drawbacks of existing methodologies. Our proposed model is able to capture time-varying scale, correlation and smoothness across multiple clinical variables. We also provide details on two inference approaches: Maximum a posteriori and Hamilton Monte Carlo. Our model is validated on synthetic data and then we demonstrate its effectiveness on EHR data from Kaiser Permanente Division of Research (KPDOR). Finally, we use the KPDOR EHR data to investigate the relationships between a clinical patient risk metric and the latent processes of our proposed model and demonstrate statistically significant correlations between these entities.

Automated Identification of Adults at Risk for In-Hospital Clinical Deterioration.

BACKGROUND: Hospitalized adults whose condition deteriorates while they are in wards (outside the intensive care unit [ICU]) have considerable morbidity and mortality. Early identification of patients at risk for clinical deterioration has relied on manually calculated scores. Outcomes after an automated detection of impending clinical deterioration have not been widely reported. METHODS: On the basis of a validated model that uses information from electronic medical records to identify hospitalized patients at high risk for clinical deterioration (which permits automated, real-time risk-score calculation), we developed an intervention program involving remote monitoring by nurses who reviewed records of patients who had been identified as being at high risk; results of this monitoring were then communicated to rapid-response teams at hospitals. We compared outcomes (including the primary outcome, mortality within 30 days after an alert) among hospitalized patients (excluding those in the ICU) whose condition reached the alert threshold at hospitals where the system was operational (intervention sites, where alerts led to a clinical response) with outcomes among patients at hospitals where the system had not yet been deployed (comparison sites, where a patient's condition would have triggered a clinical response after an alert had the system been operational). Multivariate analyses adjusted for demographic characteristics, severity of illness, and burden of coexisting conditions. RESULTS: The program was deployed in a staggered fashion at 19 hospitals between August 1, 2016, and February 28, 2019. We identified 548,838 non-ICU hospitalizations involving 326,816 patients. A total of 43,949 hospitalizations (involving 35,669 patients) involved a patient whose condition reached the alert threshold; 15,487 hospitalizations were included in the intervention cohort, and 28,462 hospitalizations in the comparison cohort. Mortality within 30 days after an alert was lower in the intervention cohort than in the comparison cohort (adjusted relative risk, 0.84, 95% confidence interval, 0.78 to 0.90; P<0.001). CONCLUSIONS: The use of an automated predictive model to identify high-risk patients for whom interventions by rapid-response teams could be implemented was associated with decreased mortality. (Funded by the Gordon and Betty Moore Foundation and others.).

Multiyear Rehospitalization Rates and Hospital Outcomes in an Integrated Health Care System.

Importance: Since the introduction of the rehospitalization rate as a quality measure, multiple changes have taken place in the US health care delivery system. Interpreting rehospitalization rates without taking a global view of these changes and new data elements from comprehensive electronic medical records yields a limited assessment of the quality of care. Objective: To examine hospitalization outcomes from a broad perspective, including the implications of numerator and denominator definitions, all adult patients with all diagnoses, and detailed clinical data. Design, Setting, and Participants: This cohort study obtained data from 21 hospitals in Kaiser Permanente Northern California (KPNC), an integrated health care delivery system that serves patients with Medicare Advantage plans, Medicaid, and/or Kaiser Foundation Health Plan. The KPNC electronic medical record system was used to capture hospitalization data for adult patients who were 18 years of age or older; discharged from June 1, 2010, through December 31, 2017; and hospitalized for reasons other than childbirth. Hospital stays for transferred patients were linked using public and internal sources. Exposures: Hospitalization type (inpatient, for observation only), comorbidity burden, acute physiology score, and care directives. Main Outcomes and Measures: Mortality (inpatient, 30-day, and 30-day postdischarge), nonelective rehospitalization, and discharge disposition (home, home with home health assistance, regular skilled nursing facility, or custodial skilled nursing facility). Results: In total, 1 384 025 hospitalizations were identified, of which 1 155 034 (83.5%) were inpatient and 228 991 (16.5%) were for observation only. These hospitalizations involved 679 831 patients (mean [SD] age, 61.4 [18.1] years; 362 582 female [53.3%]). The number of for-observation-only hospitalizations increased from 16 497 (9.4%) in the first year of the study to 120 215 (20.5%) in the last period of the study, whereas inpatient hospitalizations with length of stay less than 24 hours decreased by 33% (from 12 008 [6.9%] to 27 108 [4.6%]). Illness burden measured using administrative data or acute physiology score increased significantly. The proportion of patients with a Comorbidity Point Score of 65 or higher increased from 20.5% (range across hospitals, 18.4%-26.4%) to 28.8% (range, 22.3%-33.0%), as did the proportion with a Charlson Comorbidity Index score of 4 or higher, which increased from 28.8% (range, 24.6%-35.0%) to 38.4% (range, 31.9%-43.4%). The proportion of patients at or near critical illness (Laboratory-based Acute Physiology Score [LAPS2] ≥110) increased by 21.4% (10.3% [range across hospitals, 7.4%-14.7%] to 12.5% [range across hospitals, 8.3%-16.6%]; P < .001), reflecting a steady increase of 0.07 (95% CI, 0.04-0.10) LAPS2 points per month. Unadjusted inpatient mortality in the first year of the study was 2.78% and in the last year was 2.71%; the corresponding numbers for 30-day mortality were 5.88% and 6.15%, for 30-day postdischarge mortality were 3.94% and 4.22%, and for nonelective rehospitalization were 12.00% and 12.81%, respectively. All outcomes improved after risk adjustment. Compared with the first month, the final observed to expected ratio was 0.79 (95% CI, 0.73-0.84) for inpatient mortality, 0.86 (95% CI, 0.82-0.89) for 30-day mortality, 0.90 (95% CI, 0.85-0.95) for 30-day nonelective rehospitalization, and 0.87 (95% CI, 0.83-0.92) for 30-day postdischarge mortality. The proportion of nonelective rehospitalizations meeting public reporting criteria decreased substantially over the study period (from 58.0% in 2010-2011 to 45.2% in 2017); most of this decrease was associated with the exclusion of observation stays. Conclusions and Relevance: This study found that in this integrated system, the hospitalization rate decreased and risk-adjusted hospital outcomes improved steadily over the 7.5-year study period despite worsening case mix. The comprehensive results suggest that future assessments of care quality should consider the implications of numerator and denominator definitions, display multiple metrics concurrently, and include all hospitalization types and detailed data.

Assessing clinical heterogeneity in sepsis through treatment patterns and machine learning.

OBJECTIVE: To use unsupervised topic modeling to evaluate heterogeneity in sepsis treatment patterns contained within granular data of electronic health records. MATERIALS AND METHODS: A multicenter, retrospective cohort study of 29 253 hospitalized adult sepsis patients between 2010 and 2013 in Northern California. We applied an unsupervised machine learning method, Latent Dirichlet Allocation, to the orders, medications, and procedures recorded in the electronic health record within the first 24 hours of each patient's hospitalization to uncover empiric treatment topics across the cohort and to develop computable clinical signatures for each patient based on proportions of these topics. We evaluated how these topics correlated with common sepsis treatment and outcome metrics including inpatient mortality, time to first antibiotic, and fluids given within 24 hours. RESULTS: Mean age was 70 ± 17 years with hospital mortality of 9.6%. We empirically identified 42 clinically recognizable treatment topics (eg, pneumonia, cellulitis, wound care, shock). Only 43.1% of hospitalizations had a single dominant topic, and a small minority (7.3%) had a single topic comprising at least 80% of their overall clinical signature. Across the entire sepsis cohort, clinical signatures were highly variable. DISCUSSION: Heterogeneity in sepsis is a major barrier to improving targeted treatments, yet existing approaches to characterizing clinical heterogeneity are narrowly defined. A machine learning approach captured substantial patient- and population-level heterogeneity in treatment during early sepsis hospitalization. CONCLUSION: Using topic modeling based on treatment patterns may enable more precise clinical characterization in sepsis and better understanding of variability in sepsis presentation and outcomes.

Risk Adjustment for Hospital Characteristics Reduces Unexplained Hospital Variation in Pressure Injury Risk.

BACKGROUND: Research investigating risk factors for hospital-acquired pressure injury (HAPI) has primarily focused on the characteristics of patients and nursing staff. Limited data are available on the association of hospital characteristics with HAPI. OBJECTIVE: We aimed to quantify the association of hospital characteristics with HAPI and their effect on residual hospital variation in HAPI risk. METHODS: We employed a retrospective cohort study design with split validation using hierarchical survival analysis. This study extends the analysis "Hospital-Acquired Pressure Injury (HAPI): Risk Adjusted Comparisons in an Integrated Healthcare Delivery System" by Rondinelli et al. (2018) to include hospital-level factors. We analyzed 1,661 HAPI episodes among 728,266 adult hospitalization episodes across 35 California Kaiser Permanente hospitals, an integrated healthcare delivery system between January 1, 2013, and June 30, 2015. RESULTS: After adjusting for patient-level and hospital-level variables, 2 out of 12 candidate hospital variables were statistically significant predictors of HAPI. The hazard for HAPI decreased by 4.8% for every 0.1% increase in a hospital's mean mortality ([6.3%, 2.6%], p < .001), whereas every 1% increase in a hospital's proportion of patients with a history of diabetes increased HAPI hazard by 5% ([-0.04%, 10.0%], p = .072). Addition of these hierarchical variables decreased unexplained hospital variation of HAPI risk by 35%. DISCUSSION: We found hospitals with higher patient mortality had lower HAPI risk. Higher patient mortality may decrease the pool of patients who live to HAPI occurrence. Such hospitals may also provide more resources (specialty staff) to care for frail patient populations. Future research should aim to combine hospital data sets to overcome power limitations at the hospital level and should investigate additional measures of structure and process related to HAPI care.

Flexible, cluster-based analysis of the electronic medical record of sepsis with composite mixture models.

The widespread adoption of electronic medical records (EMRs) in healthcare has provided vast new amounts of data for statistical machine learning researchers in their efforts to model and predict patient health status, potentially enabling novel advances in treatment. In the case of sepsis, a debilitating, dysregulated host response to infection, extracting subtle, uncataloged clinical phenotypes from the EMR with statistical machine learning methods has the potential to impact patient diagnosis and treatment early in the course of their hospitalization. However, there are significant barriers that must be overcome to extract these insights from EMR data. First, EMR datasets consist of both static and dynamic observations of discrete and continuous-valued variables, many of which may be missing, precluding the application of standard multivariate analysis techniques. Second, clinical populations observed via EMRs and relevant to the study and management of conditions like sepsis are often heterogeneous; properly accounting for this heterogeneity is critical. Here, we describe an unsupervised, probabilistic framework called a composite mixture model that can simultaneously accommodate the wide variety of observations frequently observed in EMR datasets, characterize heterogeneous clinical populations, and handle missing observations. We demonstrate the efficacy of our approach on a large-scale sepsis cohort, developing novel techniques built on our model-based clusters to track patient mortality risk over time and identify physiological trends and distinct subgroups of the dataset associated with elevated risk of mortality during hospitalization.

Prediction of Recurrent Clostridium Difficile Infection Using Comprehensive Electronic Medical Records in an Integrated Healthcare Delivery System.

BACKGROUND Predicting recurrent Clostridium difficile infection (rCDI) remains difficult. METHODS: We employed a retrospective cohort design. Granular electronic medical record (EMR) data had been collected from patients hospitalized at 21 Kaiser Permanente Northern California hospitals. The derivation dataset (2007-2013) included data from 9,386 patients who experienced incident CDI (iCDI) and 1,311 who experienced their first CDI recurrences (rCDI). The validation dataset (2014) included data from 1,865 patients who experienced incident CDI and 144 who experienced rCDI. Using multiple techniques, including machine learning, we evaluated more than 150 potential predictors. Our final analyses evaluated 3 models with varying degrees of complexity and 1 previously published model. RESULTS Despite having a large multicenter cohort and access to granular EMR data (eg, vital signs, and laboratory test results), none of the models discriminated well (c statistics, 0.591-0.605), had good calibration, or had good explanatory power. CONCLUSIONS Our ability to predict rCDI remains limited. Given currently available EMR technology, improvements in prediction will require incorporating new variables because currently available data elements lack adequate explanatory power. Infect Control Hosp Epidemiol 2017;38:1196-1203.

The Timing of Early Antibiotics and Hospital Mortality in Sepsis.

RATIONALE: Prior sepsis studies evaluating antibiotic timing have shown mixed results. OBJECTIVES: To evaluate the association between antibiotic timing and mortality among patients with sepsis receiving antibiotics within 6 hours of emergency department registration. METHODS: Retrospective study of 35,000 randomly selected inpatients with sepsis treated at 21 emergency departments between 2010 and 2013 in Northern California. The primary exposure was antibiotics given within 6 hours of emergency department registration. The primary outcome was adjusted in-hospital mortality. We used detailed physiologic data to quantify severity of illness within 1 hour of registration and logistic regression to estimate the odds of hospital mortality based on antibiotic timing and patient factors. MEASUREMENTS AND MAIN RESULTS: The median time to antibiotic administration was 2.1 hours (interquartile range, 1.4-3.1 h). The adjusted odds ratio for hospital mortality based on each hour of delay in antibiotics after registration was 1.09 (95% confidence interval [CI], 1.05-1.13) for each elapsed hour between registration and antibiotic administration. The increase in absolute mortality associated with an hour's delay in antibiotic administration was 0.3% (95% CI, 0.01-0.6%; P = 0.04) for sepsis, 0.4% (95% CI, 0.1-0.8%; P = 0.02) for severe sepsis, and 1.8% (95% CI, 0.8-3.0%; P = 0.001) for shock. CONCLUSIONS: In a large, contemporary, and multicenter sample of patients with sepsis in the emergency department, hourly delays in antibiotic administration were associated with increased odds of hospital mortality even among patients who received antibiotics within 6 hours. The odds increased within each sepsis severity strata, and the increased odds of mortality were greatest in septic shock.

Data that drive: Closing the loop in the learning hospital system.

The learning healthcare system describes a vision of US healthcare that capitalizes on science, information technology, incentives, and care culture to drive improvements in the quality of health care. The inpatient setting, one of the most costly and impactful domains of healthcare, is an ideal setting in which to use data and information technology to foster continuous learning and quality improvement. The rapid digitization of inpatient medicine offers incredible new opportunities to use data from routine care to generate new discovery and thus close the virtuous cycle of learning. We use an object lesson-sepsis care within the 21 hospitals of the Kaiser Permanente Northern California integrated healthcare delivery system-to offer insight into the critical elements necessary for developing a learning hospital system. We then describe how a hospital-wide data-driven approach to inpatient care can facilitate improvements in the quality of hospital care. Journal of Hospital Medicine 2016;11:S11-S17. © 2016 Society of Hospital Medicine.

The Natural History of Changes in Preferences for Life-Sustaining Treatments and Implications for Inpatient Mortality in Younger and Older Hospitalized Adults.

OBJECTIVES: To compare changes in preferences for life-sustaining treatments (LSTs) and subsequent mortality of younger and older inpatients. DESIGN: Retrospective cohort study. SETTING: Kaiser Permanente Northern California (KPNC). PARTICIPANTS: Individuals hospitalized at 21 KPNC hospitals between 2008 and 2012 (N = 227,525). MEASUREMENTS: Participants were divided according to age (<65, 65-84, ≥85). The effect of age on adding new and reversing prior LST limitations was evaluated. Survival to inpatient discharge was compared according to age group after adding new LST limitations. RESULTS: At admission, 18,254 (54.2%) of those aged 85 and older, 18,349 (20.8%) of those aged 65 to 84, and 3,258 (3.1%) of those younger than 65 had requested that the use of LST be limited. Of the 187,664 participants who initially did not request limitations on the use of LST, 15,932 (8.5%) had new LST limitations added; of the 39,861 admitted with LST limitations, 3,017 (7.6%) had these reversed. New limitations were more likely to be seen in older participants (aged 65-84, odds ratio (OR) = 2.27, 95% confidence interval (CI) = 2.16-2.39; aged ≥85, OR = 6.43, 95% CI = 6.05-6.84), and reversals of prior limitations were less likely to be seen in older individuals (aged 65-84, OR = 0.73, 95% CI = 0.65-0.83; aged ≥85, OR = 0.46, 95% CI = 0.41-0.53) than in those younger than 65. Survival rates to inpatient discharge were 71.7% of subjects aged 85 and older who added new limitations, 57.2% of those aged 65 to 84, and 43.4% of those younger than 65 (P < .001). CONCLUSION: Changes in preferences for LSTs were common in hospitalized individuals. Age was an important determinant of likelihood of adding new or reversing prior LST limitations. Of subjects who added LST limitations, those who were older were more likely than those who were younger to survive to hospital discharge.