Exploration of different statistical approaches in the comparison of dopamine and norepinephrine in the treatment of shock: SOAP II

Abstract

Background: Exploring clinical trial data using alternative methods may enhance original study's findings and provide new insights. The SOAP II trial has been published more than 10 years ago; but there is still some speculation that some patients may benefit from dopamine administration for shock management. We aimed to reanalyse the trial under different approaches and evaluate for heterogeneity in treatment effect (HTE).

Methods: All patients enrolled in SOAP II were eligible for reanalysis. We used a variety of methods including the win-ratio (WR), a Bayesian reanalysis stratified according to shock type, and both a risk-based and effect-based explorations for HTE. The methods were applied to different endpoints, including a hierarchy of death, new use of renal-replacement therapy (RRT), and new-onset arrhythmia; 28-day mortality; a composite endpoint (mortality, new use of RRT, and new-onset arrhythmia), and days alive and free of ICU at 28-days (DAFICU28).

Results: A total of 1679 patients were included (average age was 64.9 years, 57% male, 62% with septic and 17% with cardiogenic shock). All analysis favoured norepinephrine over dopamine. Under the WR approach, dopamine had fewer wins compared to norepinephrine (WR 0.79; 95% confidence intervals [CI] 0.68-0.92; p = 0.003), evident in both cardiogenic and septic shock subgroups. The Bayesian reanalysis for type of shock showed, for dopamine, a probability of harm of 0.95 for mortality, > 0.99 probability of harm for composite endpoint, and 0.91 probability of harm for DAFICU28. The fewer DAFICU28 with dopamine was more apparent in those with cardiogenic shock (0.92). Under the risk-based HTE, there was a high probability that dopamine resulted fewer DAFICU28 in the highest quartile of predicted mortality risk. The effect-based HTE assessment model did not recommended dopamine over norepinephrine for any combination of possible modifiers including age, type of shock, presence of cardiomyopathy, and SOFA score. Receiving dopamine when the effect-based model recommended norepinephrine was associated with an absolute increase in composite endpoint of 6%.

Conclusion: The harm associated with the use of dopamine for the management of shock appears to be present in both septic and cardiogenic shock patients. There was no suggestion of any subgroup in which dopamine was found to be favourable over norepinephrine.

Keywords: Bayesian; Critical care; Dopamine; Norepinephrine; Randomized; Shock.

Fig. 1

Results for the win ration (WR) analysis for the first hierarchical approach (A and B). A Stratified win ratio results for the first hierarchical approach (mortality, use of renal replacement therapy—RRT, and occurrence of arrhythmia). B Stratified win ratio results for the second hierarchical approach (mortality and ICU length-of-stay—LOS)

Fig. 2

Bayesian reanalysis for 28-day mortality and composite endpoint. Blue areas represent benefit for dopamine, red areas represent harm for dopamine (benefit for norepinephrine), and dark red areas represent the region of practical equivalence. A Posterior distribution of adjusted risk difference for mortality. B Posterior distribution of adjusted risk difference for composite endpoint of mortality, use of renal replacement therapy, and mortality

Fig. 3

Bayesian reanalysis for day-alive and free of ICU at 28 days (DAFICU28). A Overall difference in DAFICU28 in a model adjusted for shock type. B, C, and D show the posterior distribution according to shock type. Areas in blue mark benefit for dopamine; areas in red mark harm for dopamine; region of practical equivalence—ROPE—is highlighted in dark red

Fig. 4

Risk-based HTE results for days alive and free of ICU at 28 days—DAFICU28—according to customized APACHE II predicted mortality and randomization arm. Note that the differences in DAFICU28 are more pronounced for more severe patients (last quartile). Numeric values are shown in Table 3. Parenthesis are used to denote open intervals and brackets are used to denote the interval is closed at that end

Fig. 5

A “Individualized” (conditional) treatment effects (ITE) for patients in the test set according to the model adjusted for adjusted for age, type of shock, presence of cardiomyopathy at baseline, and SOFA score, all interacting with intervention arm. The differences between counterfactuals for each patient are displayed in (A), with patient identification sorted according to increasing ITE and y-axis showing the median difference and 95% HDI for the counterfactuals; a high value represents an increase in the counterfactuals between dopamine and norepinephrine, hence the higher the conceivable benefit for norepinephrine versus dopamine the higher the ITE. Note that the model could not make a recommendation in 27% of all patients. B Differences in outcome according to enrollment arm and suggestion obtained from the S-Learner. Receiving norepinephrine when the model recommended norepinephrine was associated with a decrease in composite endpoint; for patients for whom no recommendation was made, no clear differences between norepinephrine and dopamine were found for patients for whom no recommendation was made