Defining sepsis in 2016: clarity or confusion

Transcription

1 Defining sepsis in 2016: clarity or confusion Christopher W. Seymour, MD MSc The CRISMA Center Assistant Professor of Critical Care Medicine & Emergency Medicine University of Pittsburgh School of Medicine Can an otherwise healthy 58-year-old man die from a bad cold? He can, and he did. Through an unfortunate cascade of events, starting with a missed diagnosis of viral pneumonia, Tom Wilson, a systems analyst for Westinghouse, went from bad to worse until every major organ system -- kidneys, liver, lungs and finally his heart -- stopped working. After 10 days in intensive care during which doctors struggled in vain to get ahead of the rampaging disorder, Mr. Wilson died. Cause of death: septic shock. New York Times, March 5,

2 Minor illness Healthy guy Delay in the diagnosis of septic shock Absence of specific tests Rampaging disorder with organ dysfunction Why is defining sepsis important? Why is defining sepsis difficult? Conceptual approach by the 2016 Sepsis Definitions Task Force Review of Sepsis-3 Controversies post-release Objectives 2

3 Sepsis is everywhere. 2 5 million US cases each year percent of US healthcare spending Gaieski et al. Crit Care Med, 2014 Singer et al., JAMA, 2016 Sepsis is everywhere. 1 out of every 2 to 3 hospital deaths Liu et al., JAMA,

4 We don t talk about it. Stroke Heart attack Sepsis 12% 10% 20% Seymour et al., Am J Resp Crit Care Med, 2014 Why is defining sepsis difficult? Sepsis is common We don t agree on the terms Surface phenomena lead to small zone of rarity with lots of patients Time-sensitive diagnosis Knowledge is rapidly evolving Angus et al., Crit Care Med,

5 2016 Sepsis Definitions Task Force To re-examine existing criteria for sepsis and septic shock Does current pathophysiology, epidemiology mandate an update? Use expert consensus to develop a definition Use data to develop clinical criteria Focus is on the bedside clinician 5

6 Current state prior to Sepsis-3 Variety of terms Septicaemia, septic, severe sepsis, septic shock, sepsis 2 or more SIRS criteria to identify sepsis among those with suspected infection Organ dysfunction is key, but uncertain how to measure Multiple criteria for septic shock Shankar-Hari et al., JAMA, 2016 Infection Organ dysfunction Life threatening Dysregulated host response Singer et al., JAMA,

7 We have a definition for sepsis. Criteria for the bedside Criteria for Infection? Clinical diagnosis Not the prevue of the Task Force Criteria for organ dysfunction? Seymour et al., JAMA,

8 Developing sepsis criteria Use large electronic health record databases Identify those with suspected infection Study various existing OD criteria SOFA score LODS score SIRS criteria Levy et al., Crit Care Med, 2003 Use of predictive validity No gold standard for sepsis We can t rely on tests like sensitivity, specificity etc Use outcome more common in septic patients than healthy patients Outcome Septic Not septic 8

9 Candidate clinical criteria Primary cohort 9

10 Patient characteristics CRISMA conducted primary analyses on UPMC data Suspected infection patients 12 hospitals 4% mortality rate Patient characteristics 10

11 Variable Threshold Units All patients (N=74,453) ICU patients (N=7,836) Non-ICU patients (N=66,617) Heart rate >90 BPM Respiratory rate >20 BPM SIRS variables Temperature <36 C White blood cell count >12 k/ul Temperature >38 C White blood cell count <4 k/ul Bands >10 % Systolic blood pressure <=100 mmhg Serum creatinine >=1.2 mg/dl Pa0 2 / Fi0 2 ratio <=300 SOFA variables Platelets <=150 k/ul Glasgow coma scale <15 Bilirubin >=1.2 mg/dl Mechanical ventilation Present/absent Vasopressors Present/absent Vasopressors More than one Bicarbonate <=26 mmol/l Saturation <=94 % Glucose <=109 mg/dl AST >=36 IU/L Additional candidate variables ALT >=37 IU/L INR >=1.4 Albumin <=2.5 g/dl Troponin >=0.1 ng/ml ph <=7.36 Lactate >=2.5 mmol/l Fibrinogen <=300 mg/dl ScvO 2 <=69 % Abnormal Normal Missing Proportion (%) Proportion (%) Proportion (%) What do we already know at the bedside We built a baseline risk model using only age, demographics, race, co-morbidity Divide patients into deciles No. of patients Risk of in-hospital mortality 7,449 7,456 7,515 7,372 7,572 7,301 7,523 7,390 7,515 7,346 Deciles of baseline risk of in-hospital mortality Compare validity within and across deciles 11

12 Predictive validity of criteria Fold change, in-hospital mortality Baseline risk (%) ICU encounters N = 7,932 SIRS 2 vs. SIRS <2 SOFA 2 vs. SOFA <2 LODS 2 vs. LODS <2 qsofa 2 vs. qsofa < Decile of baseline risk of in-hospital mortality Predictive validity of criteria Baseline risk (%) Fold change, in-hospital mortality Outside the ICU encounters N = 66,522 SIRS 2 vs. SIRS <2 SOFA 2 vs. SOFA <2 LODS 2 vs. LODS <2 qsofa 2 vs. qsofa < Decile of baseline risk of in-hospital mortality 12

13 But SOFA is complex Sepsis criteria should be easy SOFA is complex, requires 12 variables, costly, range from 0 to 24 points Laboratory tests take time to result We need more simple parsimonious criteria for the bedside quick Sepsis - Related Organ Failure Assessment 13

14 qsofa as a clinical prompt 3 variables Measured near onset of infection No laboratory tests Studied in 72 6 hr windows around infection qsofa as a clinical prompt SIRS 0.64 (0.62, 0.66) ICU encounters N = 7,932 AUROC in-hospital mortality SIRS 0.76 (0.75, 0.77) Outside the ICU encounters N = 66,522 AUROC in-hospital mortality SOFA < (0.73, 0.76) SOFA < (0.78, 0.80) LODS < (0.73, 0.76) LODS <0.01 < (0.80, 0.82) qsofa 0.01 <0.01 < (0.64, 0.68) qsofa <0.01 < (0.80, 0.82) SOFA and LODS superior in the ICU qsofa similar to complex scores outside the ICU 14

15 qsofa in external datasets 15

16 Serum lactate as an adjunct Proportion in hospital mortality (%) Missing qsofa = 0 qsofa = 1 qsofa = 2 qsofa = 3 < 2.0 mmol/l 2.0 to 4.0 mmol/l Serum lactate 4.0 mmol/l Fold change, in-hospital mortality Baseline risk (%) Median Minimum Maximum qsofa 2 vs. qsofa <2 (qsofa + serum lactate) 2 vs. (qsofa + lactate) < Decile of baseline risk for in-hospital mortality Addressed missing data Post hoc analyses Measurement windows for qsofa and SOFA 24 hrs after infection 6 hr window around infection Agreement of SOFA and qsofa exceeded 70% Delta of 2 SOFA points same predictive validity 16

17 Conclusions In the ICU, the SOFA and LODS have greater predictive validity than qsofa or SIRS Outside the ICU, the qsofa has similar predictive validity to more complex scores Clinical criteria for sepsis Infection plus 2 or more SOFA points above baseline Prompt to consider sepsis outside the ICU Infection plus 2 or more qsofa points 17

18 Controversies Where did severe sepsis go? Billing implications Was SIRS just left for dead? Delays in treatment if infection not suspected Why isn t lactate in the criteria? Lacks face validity to not include Prospective evaluation? No association of new definitions with better outcomes Controversies Change in SOFA from what baseline value? Practical implementation is challenging Measure altered mentation? Multiple different scales available What to do if intermediate risk? qsofa = 1 Isn t the AUROC a bad measure of clinical usefulness? 18

19 Why Sepsis-3 provides clarity Speak the same language Redundant terms like severe sepsis are removed Objective criteria for organ dysfunction recommended Data driven Why Sepsis-3 may lead to confusion Other criteria are available CMS, CDC surveillance criteria, RCT inclusion criteria Unclear how to chose time windows to measure criteria Blessing and curse of EHR data Suspected infection is a clinical decision No check boxes proposed by Task Force 19

20 My approach outside the ICU My approach outside the ICU Pt arrives to ED This looks like pneumonia! Frequent reassessment Did qsofa already occur? Send lactate and SOFA labs 3 6 hr look back Sepsis 20

21 My approach inside the ICU Resources

22 Questions 22