Over 50 RCTs have examined resuscitation strategies in sepsis, but the optimal approach remains unclear. Early goal-directed therapy (EGDT), a 6-hour protocol involving fluids, vasopressors, inotropes, and blood products, showed a significant reduction in mortality in a landmark trial. However, subsequent larger trials did not replicate these benefits, and EGDT is no longer routinely followed in practice. Some of its components, like fluid administration and vasopressor use, remain crucial to resuscitation, but the best delivery method is still debated.


Increasing awareness of varying treatment effects among individuals in RCTs suggests that the average treatment effect (ATE) may not reflect the diverse responses of patients. While an individual patient data meta-analysis of EGDT trials did not identify specific subgroups benefiting, treatment heterogeneity (HTE) could explain inconsistent results. Additionally, there is variability in the dose of interventions like fluid or vasopressor use, complicating the assessment of HTE. 


A recent study explored HTE in large EGDT trials and investigated patient characteristics or intervention factors that could explain these variations.The study divided sites from the Australian Resuscitation of Sepsis Evaluation (ARISE) and Protocolized Care for Early Septic Shock (ProCESS) trials into derivation and validation cohorts. Machine learning models were trained to predict individual absolute risk differences (iARDs) in 90-day mortality in the derivation cohorts and tested for HTE in the validation cohorts. Sensitivity analyses involved swapping the cohorts. The best-performing model was applied to a combined dataset to explore the impact of patient characteristics and specific components of EGDT on treatment responses.


The study took place at 81 sites across Australia, New Zealand, Hong Kong, Finland, Ireland, and the United States. It involved adult patients presenting to the emergency department with severe sepsis or septic shock. The interventions tested were EGDT versus usual care.


The local-linear random forest model was the most accurate in predicting iARDs. In the validation cohort, HTE was confirmed, with a significant interaction between iARD prediction and treatment. Treatment response varied across quintiles based on predicted iARDs, with lower quintiles showing potential harm from EGDT and higher quintiles showing benefit. Sensitivity analyses supported these findings. Pre-intervention albumin levels were the most significant factor in explaining HTE, but the analyses of individual EGDT components yielded inconclusive results.


The study found significant variation in individual treatment effects (ITEs) of EGDT compared to usual care. While neither trial showed a statistically significant average treatment effect (ATE), the range of iARDs varied greatly, with some patients benefiting similarly to the original EGDT trial and others experiencing harm. Circulating albumin levels were the strongest predictor of treatment response, with younger adults and those with lower albumin levels predicted to experience the most harm from EGDT. Exploratory analyses suggested that EGDT might harm patients with cirrhosis and benefit those with heart failure, though these findings are preliminary and highlight the importance of considering comorbidities in future research.


The discrepancy between results from the ARISE, ProCESS, and PROMISE trials and the original Rivers trial suggests that EGDT may be effective only for certain subsets of patients. These variations could be due to differences in patient populations and their responses to EGDT. The study emphasises that similar heterogeneity in treatment effects likely exists in other sepsis resuscitation trials and suggests that future trials should be larger, randomise patients to multiple intervention arms, and account for potential treatment heterogeneity for each component of resuscitation.


This research underscores the need for more personalised treatment approaches, with the potential to tailor resuscitation strategies based on individual patient characteristics. Using machine learning tools like SHAP to better understand predictors of treatment response could offer valuable insights. The findings point to the need for future trials to incorporate strategies that learn and adapt to heterogeneity in treatment effects, allowing for more personalised resuscitation plans.


Source: Critical Care Medicine
Image Credit: iStock 

 


References:




Latest Articles

Sepsis, septic shock, heterogeneity of treatment effect, early goal-directed therapy, EGDT Heterogeneity in the Effect of Early Goal-Directed Therapy for Septic Shock