Background
Sepsis and acute respiratory distress syndrome are two heterogeneous acute illnesses with high risk of death and for which there are many ‘statistically negative’ randomised controlled trials. We hypothesised that negative randomised controlled trials occur because of between-participant differences in response to treatment, illness manifestation (phenotype) and risk of outcomes (heterogeneity).
Objectives
To assess (1) heterogeneity of treatment effect, which tests whether or not treatment effect varies with a patient’s pre-randomisation risk of outcome; and (2) whether or not subphenotypes explain the treatment response differences in sepsis and acute respiratory distress syndrome demonstrated in randomised controlled trials.
Study population
We performed secondary analysis of two randomised controlled trials in patients with sepsis [i.e. the Vasopressin vs Noradrenaline as Initial Therapy in Septic Shock (VANISH) trial and the Levosimendan for the Prevention of Acute oRgan Dysfunction in Sepsis (LeoPARDS) trial] and one acute respiratory distress syndrome multicentre randomised controlled trial [i.e. the Hydroxymethylglutaryl-CoA reductase inhibition with simvastatin in Acute lung injury to Reduce Pulmonary dysfunction (HARP-2) trial], conducted in the UK. The VANISH trial is a 2 × 2 factorial randomised controlled trial of vasopressin (Pressyn AR®; Ferring Pharmaceuticals, Saint-Prex, Switzerland) and hydrocortisone sodium phosphate (hereafter referred to as hydrocortisone) (EfcortesolTM; Amdipharm plc, St Helier, Jersey) compared with placebo. The LeoPARDS trial is a two-arm-parallel-group randomised controlled trial of levosimendan (Simdax®; Orion Pharma, Espoo, Finland) compared with placebo. The HARP-2 trial is a parallel-group randomised controlled trial of simvastatin compared with placebo.
Methods
To test for heterogeneity of the effect on 28-day mortality of vasopressin, hydrocortisone and levosimendan in patients with sepsis and of simvastatin in patients with acute respiratory distress syndrome. We used the total Acute Physiology And Chronic Health Evaluation II (APACHE II) score as the baseline risk measurement, comparing treatment effects in patients with baseline APACHE II scores above (high) and below (low) the median using regression models with an interaction between treatment and baseline risk. To identify subphenotypes, we performed latent class analysis using only baseline clinical and biomarker data, and compared clinical outcomes across subphenotypes and treatment groups.
Results
The odds of death in the highest APACHE II quartile compared with the lowest quartile ranged from 4.9 to 7.4, across the three trials. We did not observe heterogeneity of treatment effect for vasopressin, hydrocortisone and levosimendan. In the HARP-2 trial, simvastatin reduced mortality in the low-APACHE II group and increased mortality in the high-APACHE II group. In the VANISH trial, a two-subphenotype model provided the best fit for the data. Subphenotype 2 individuals had more inflammation and shorter survival. There were no treatment effect differences between the two subphenotypes. In the LeoPARDS trial, a three-subphenotype model provided the best fit for the data. Subphenotype 3 individuals had the greatest inflammation and lowest survival. There were no treatment effect differences between the three subphenotypes, although survival was lowest in the levosimendan group for all subphenotypes. In the HARP-2 trial, a two-subphenotype model provided the best fit for the data. The inflammatory subphenotype was associated with fewer ventilator-free days and higher 28-day mortality.
Limitations
The lack of heterogeneity of treatment effect and any treatment effect differences between sepsis subphenotypes may be secondary to the lack of statistical power to detect such effects, if they truly exist.
Conclusions
We highlight lack of heterogeneity of treatment effect in all three trial populations. We report three subphenotypes in sepsis and two subphenotypes in acute respiratory distress syndrome, with an inflammatory phenotype with greater risk of death as a consistent finding in both sepsis and acute respiratory distress syndrome.
Future work
Our analysis highlights the need to identify key discriminant markers to characterise subphenotypes in sepsis and acute respiratory distress syndrome with an observational cohort study.
Funding
This project was funded by the Efficacy and Mechanism Evaluation (EME) programme, a MRC and National Institute for Health Research (NIHR) partnership. This will be published in full in Efficacy and Mechanism Evaluation; Vol. 8, No. 10. See the NIHR Journals Library website for further project information.
Comparison of machine learning clustering algorithms for detecting heterogeneity of treatment effect in acute respiratory distress syndrome: A secondary analysis of three randomised controlled trials
