Table 2.

Summary of analyses, results, and interpretations discussed in the manuscript

Analysis	Endpoint	Motivation	Result	Interpretation
Win ratio	Hierarchy of death → RRT → arrhythmia	Combine mortality with two outcomes that were previously suggested with dopamine use (namely possible renal protective effects and possible higher arrhythmia occurrence	Less wins for dopamine (WR = 0.79 (95% CI 0.68–0.92; p = 0.003) for untied comparisons. Results consistent for cardiogenic and septic shock groups	Dopamine was associated with less wins than norepinephrine suggesting, under this approach, that it was inferior to norepinephrine. Results of WR, however, are not easily translatable to clinicians
	Hierarchy of death → ICU LOS)	Combine mortality and ICU LOS in a single endpoint	Neutral results (WR = 0.96; 95% CI 0.86–1.08; p = 0.51)	Using this hierarchy under a WR approach, results were neutral hence corroborating with the idea that arrhythmia was the larger driver factor in the first hierarchy approach
Bayesian reanalysis for shock type	28-day mortality	Reassess the main trial while adjusting for type of shock an interaction	No clear interaction for shock type and type of vasopressor; however, there was a high probability of harm with dopamine (0.95), with a median increase in mortality of 0.04 (95% HDI from − 0.01 to 0.09)	A simple Bayesian reanalysis under a neutral prior highlight that there was a high probability of harm with dopamine
	Composite endpoint	Reassess the main trial using a composite endpoint of death, RRT, and arrhythmia while adjusting for type of shock an interaction	No clear interaction for shock type and type of vasopressor; however, there was a very high probability of harm with dopamine (0.99), with a median increase in endpoint of 0.09 (95% HDI from 0.04 to 0.12)	Very high probability that dopamine was harmful was harmful, which should suffice by itself as evidence against the drug
	DAFICU28	Reassess the main trial using a more granular endpoint that has become common over last decade while adjusting for type of shock an interaction	There was a suggestion of association between type of shock and DAFICU, with the highest probability on cardiogenic shock (0.92). Overall, dopamine was associated with − 0.7 DAFICU28 (95% HDI − 1.58 to 0.92 days)	There was a suggestion for the interaction between vasopressor type and DAFICU28, but overall signal for harm is sustained
Risk-based HTE assessment	28-day mortality	Assess risk-based HTE using customized APACHE II predictions for 28-day mortality	No clear interaction between predicted APACHE II quartiles and outcomes. Results suggesting harm of dopamine similar to analysis stratified by type of shock	No clear risk-based HTE for primary endpoint for dopamine versus dopamine
	Composite endpoint	Assess risk-based HTE using customized APACHE II predictions for composite endpoint	No clear interaction between predicted APACHE II quartiles and outcomes. Results suggesting harm of dopamine similar to analysis stratified by type of shock	No clear risk-based HTE for composite endpoint for dopamine versus norepinephrine
	DAFICU28	Assess risk-based HTE using customized APACHE II predictions for DAFICU	Risk-based HTE was probable, with a suggestion of greatest harm for dopamine in the highest quartile of APACHE II predictions	Using a more granular endpoint, some suggestion of risk-based HTE can be found in SOAP II, with increasing harm for dopamine as illness severity increased
Effect-based HTE assessment	Composite endpoint	Use a causal HTE model adjusted for key covariates to try to obtain more information regarding possible THE	A model recommendation recommended norepinephrine for most patients and did not recommend dopamine	A simple S-learner recommended norepinephrine for most patients in the test set; no specific combination of possible mediators resulted in a recommendation favouring dopamine

RRT, Renal replacement therapy; DAFICU28, Days alive and free of ICU at 28 days; WR, Win Ratio; THE, Heterogeneity in treatment effects