Table 4.

Comparative strengths and weaknesses of observational studies and randomized trials in AMI-CS.

Randomized controlled trials	Observational analyses
Strengths
•Able to prove causality •Randomization ensures that unmeasured confounding variables are balanced •Isolates treatment effect under ideal circumstances •Stringent diagnostic criteria ensure a homogenous population •Ideally includes patients most likely to benefit •Detailed case report form prospectively collects all baseline features and outcomes •Stratified randomization can further balance groups on important covariates •Low loss to follow-up with end point adjudication ensures accurate outcomes •Can provide insights into pathophysiology, utility of biomarkers, imaging, etc •Detailed assessment of the severity of shock and organ failure is possible •Temporary relationships between variables and outcomes can be ascertained	•Effectiveness under real-world conditions •Enhanced external generalizability owing to representative population sample •Nationally representative cohorts may be queried •Enrolls a broader population such as underrepresented groups •Lower cost •Large sample size improves statistical power, especially for subgroup analyses •Can explore low frequency safety outcomes
Weaknesses
•Only a few patients in contemporary practice may be trial-eligible •Enrolled patients may not be representative of the general disease population •Highly selected population with strict inclusion/exclusion criteria •Many eligible patients cannot be enrolled leading to limited sample size, especially for subgroups •Slow enrollment may bias results due to changes in care over time and uncertainty about clinical equipoise •Data not recorded on case report form may not be available in retrospect •Enrollment occurs at tertiary-care centers with experience and established treatment protocols which may reduce external generalizability •Risk of selection bias •High cost •Randomization may not ensure balance in measured and unmeasured covariates when sample size is small	•Unmeasured confounding variables may mediate observed effects •Confounding by indication often occurs, with substantial differences between groups based on treatments received •Typically includes a mix of patients who may and may not benefit •Can only demonstrate associations •May not differentiate cause vs consequence due to uncertainties about timing •Poor granularity of data, especially retrospective administrative or claims databases •Differences in care practices between centers may affect outcomes •Risk of selection bias •Limited information regarding disease severity and indications for device use •Limited mechanistic insights available •Differential loss to follow-up and inconsistent end point definitions can bias results •Variable/changing diagnostic criteria results in a mix of disease states in cohort •Changes in care during study period can affect results •Data not recorded in the health record cannot be obtained