Big Data - Epidemiology of Critical Illness and Sepsis

危重疾病和败血症（包括COVID-19引起的危重疾病）与显著的发病率和死亡率相关，特别是在现有治疗策略仍不理想的情况下。我们的主要目标是利用大量的临床数据来更好地了解危重疾病、败血症和严重感染的临床流行病学，包括定义疾病负担、风险因素和临床影响。

Critical illness and sepsis (including critical disease from COVID-19) are associated with significant morbidity and mortality, especially in conditions where existing therapeutic strategies remain suboptimal. Our primary aim is to leverage large repositories of granular clinical data to better understand the clinical epidemiology of critical illness, sepsis and serious infections, including defining illness burden, risk factors and clinical impact. The pathophysiology of organ dysfunction in streptococcal infection is attributed to inflammation mediated by exotoxin-mediated cytokine cascade. Clindamycin neutralizes exotoxin released by Group A Streptococcus. In a large propensity-matched cohort of patients we found that clindamycin added as an adjunct to beta lactam therapy was associated with improved survival in Group A streptococcal infection but also identified a trend toward harm when used in non-Group A, Group B streptococcal infections. These findings warrant confirmation in randomized trials. Antibiotic overuse remains a significant problem in the critically ill and is associated with toxicity and development of antimicrobial resistance. Among critically ill patients with suspected or confirmed sepsis in a larger cohort of US hospitals, we found that use of procalcitonin significantly reduced duration of antibiotic use without worsening outcomes. We confirmed these findings in a meta-analysis of randomized controlled trials. We also described real world use patterns of this biomarker using large an enhanced administrative dataset and studied its performance characteristics in patients with suspected bloodstream infection. Most estimates of sepsis incidence and mortality in existing literature are estimated using claims data. Unfortunately claims based data are subject to a variety of biases. We estimate 10-year trends in the incidence and outcome of septic shock using clinical indicators in a cohort of academic medical centers in the United States and compared it to estimates obtained using claims based data. We found that the prevalence of septic shock was rising and mortality declining over time, albeit, both less vigorously than suggested by claims based methods. In addition, we identified obesity as being associated with better outcomes in more than 50,000 patients with clinical indicators of sepsis at US hospitals. In collaboration with investigators at Harvard Medical School, we were able to study the differences in characteristics and outcomes of sepsis that originates in the community versus the hospital as well as studied variation in identifying sepsis and organ dysfunction using claims versus electronic health records. Along with the same group, we were able to assess how a severity score performs to identify patients with undifferentiated sepsis. In collaboration with investigators at Harvard Department of Population Medicine, we found that models incorporating electronic health record data accurately predict hospital mortality for patients who meet an operational definition of sepsis based on clinical indicators and outperforms models using administrative data alone. This operational definition may enable more meaningful comparisons of hospital sepsis outcomes and provide an important window into quality of care. We also found that incorporating clinical data into risk adjustment substantially changes rankings of hospitals' sepsis mortality rates compared with using administrative data alone. Comprehensive risk adjustment using both administrative and clinical data is necessary before comparing hospitals by sepsis mortality rates. Our study on short stays for sepsis underscore the incomplete uptake of Sepsis-3 definitions, the breadth of illness severities encompassed by both traditional and new sepsis definitions, and the possibility that some patients with sepsis recover very rapidly. Our study on association between opioid use and sepsis highlight the disproportionately high mortality burden associated with younger opioid users developing sepsis. With this group, we also identified potential biases in time-to-antibiotic studies in sepsis. As part of the ICU-CAR initiative, a research group of critical care and oncology providers, we participated in a study that surveyed clinicians from multiple ICUs in the US caring for patients with hematologic malignancies receiving T-cell directed therapy to understand prevailing practices around the management of cytokine release syndrome T Cell toxicities. Immune suppressed patients often develop invasive candidiasis. We used this entity as a proof of concept to show how machine learning can capture risk factors for infectious diseases of interest. In collaboration with the CDC, we found that hospital surges are detrimental to survival for patients admitted with COVID-19. With the CDC group, we also found that risk of hospital-onset COVID-19 increases in the setting of high community transmission. In collaboration with a Canadian research group, we found that surges were particularly detrimental for non-COVID patients with conditions that tend to require the same healthcare resources as patients with COVID-19. With investigator from the same group, we found that vaccine booster doses were associated with improved outcomes in vaccinated patients with COVID-19 who were admitted to the ICU. We also found that the risk of COVID-19 reinfection remained low up to 9 months since index infection but our findings do not include the period where variants of concern were prevalent in the U.S. We found that hydroxychloroquine use fell sharply even before the emergency use authorization for this drug was revoked. We also found that corticosteroids were being used frequently in ventilated COVID-19 patients well before its benefit was publicly known from clinical trials. We found that inter hospital transport of patients with lower respiratory tract infections during the pandemic was just as safe as compared to during pre-pandemic times. We also found that the SOFA score, an acute severity score was a poor predictor of mortality in intubated patients with COVID-19, and yet most US States use it in their Crisis Standards of Care ventilator triage algorithms were heavily weighted by the SOFA score. We completed an investigation to test whether the modified 4 C score, a COVID-19 specific severity score performs adequately in a surge setting and over time, to test its candidacy for triage algorithms.