Table 2. Assessing the research goal and types of questions answered using existing data.

Assessing the Research Goal	Types of Questions Answered	Need for Clinical Detail	Notes
Determine the causal relationship between risk factor and outcome	What is the relationship between to recruitment maneuvers and complications in patients with ARDS?(123) What is the association between refractory hypotension and long-term cognitive function among survivors of sepsis?	High	In select situations, one can use advanced methods for causal inference to attempt to overcome confounding by severity of illness such as propensity scores, instrumental variables, or quasi-experimental design.
Compare outcomes between two treatments, tests, or care delivery models (comparative effectiveness)	Does the routine use of pulmonary-artery catheters improve outcomes for critically ill patients?(8) Do non-physician providers compared to physicians achieve better or worse mortality rates for their ICU patients?(124)	High	Many therapies in critical care are confounded by indication - that is, the choice of treatment depends upon how sick patients are. Because most data sources do not collect the rationale behind medical decision making, overcoming this bias is difficult even with detailed data.
Determine the association between a risk factor or group of risk factors and outcome	Are septic patients with a trial fibrillation at greater risk of stroke?(100) Can we predict ARDS after trauma?(125)	Moderate	Predictive models are usually not evaluated on variables in model, but rather on their effectiveness at determining outcome
Describe epidemiology of disease, clinical practice, health service use, or health care spending	What are the longitudinal trends in hospitalizations for sepsis?(94) What proportion of patients who receive CPR in the ICU are alive at six months?(126) To what extent does ICU admission for patients with DKA vary across hospitals?(99)	Low	Descriptive epidemiology questions rarely require risk adjustment; comparisons between groups are not made. Such questions will also require richer data to approach well.
Evaluate policy or quality improvement intervention	Did participation in a clinical trial of low tidal volume ventilation in ARDS change clinical practice?(127) Did resident work hour reform impact outcomes of ICU mortality?(128) Did health care reform in Massachusetts result in changes in ICU utilization rates?(129)	Low	Although extensive clinical information is not often necessary for such analyses, when present it can greatly enhance the analysis.