Studies are cited which correspond to the way in which we conceive studies described with these labels.
Randomized controlled trial (RCT)	Individual participants, or clusters of participants, are randomly allocated to intervention or comparator. This design is the same as the RCT design described in Box 1.
Quasi-randomized controlled trial (Q-RCT)	Individual participants, or clusters of participants, are allocated to intervention or comparator in a quasi-random manner. In health care evaluation studies, the allocation rule is often by alternation, day of the week, odd/even hospital, or social security number [39]. The allocation rule may be as good as random but, typically, gives rise to a less credible study (compared to health system studies, where the allocation rule is applied by a higher level decision maker); if allocation is not concealed, research personnel who know the rule can recruit selectively or allocate participants in a biased way. This design is essentially the same as the Q-RCT design described in Box 1 but with different mechanisms for allocation.
Controlled before-and-after study (CBA)	Study in which outcomes are assessed at two time periods for several clusters (usually geographic). Clusters are classified into intervention and comparator groups. All clusters are studied without the intervention during period 1. Between periods 1 and 2, clusters in the intervention group implement the intervention of interest whereas clusters in the comparator group do not. The outcome for clusters receiving the intervention is compared to the outcome for comparator clusters during period 2, adjusted for the outcomes observed during period 1 (when no clusters had had the intervention). Observations usually represent episodes of care, so may or may not correspond to the same individuals during the two time periods. Data at either an aggregate [40] or individual level [41] can be analyzed. This design has similarities to the DID design described in Box 1.
Nonrandomized controlled trial (NRCT)	This is usually a prospective cohort study in which allocation to intervention and comparator is not random or quasi-random and is applied by research personnel [42]. The involvement of research personnel in the allocation rule may be difficult to discern; such studies may be labeled observational if the personnel responsible for the allocation rule are not clearly described or some personnel have both health care decision making and researcher roles. Individual-level data are usually analyzed. Note that nonrandom allocation of a health care intervention is often defined in relation to organizational factors (ward, clinic, doctor, provider organization) [43], and the analysis should take account of the data hierarchy if one exists.
Interrupted time series (ITS)	When used to study health care interventions, observations usually represent episodes of care or events, the cohorts studied may or may not correspond to the same individuals at different time points and are often clustered in organizational units (e.g., a health facility or district). (Such studies may be considered to consist of multiple cross-sectional “snapshots.”) The analysis may be aggregated at the level of the clusters [44] or at the level of individual episodes of care [45]. If ITS do not have the benefit of analyzing multiple measurements from the same cohort over time (Box 1), confounding by secular trends needs to be assessed, for example, with reference to a contemporaneous comparison group (controlled interrupted time series, CITS, below). NB. Entries in Table 2 are for ITS as defined in Box 1; for ITS as defined here, entries for some cells would change. This design is similar to the ITS design described in Box 1.
Controlled interrupted time series (CITS)	As above for an ITS but with data for a contemporaneous comparison group in which the intervention was not implemented [46]. Measurements for the comparison group should be collected using the same methods. This design is similar to the CITS design described in Box 1.
Concurrently controlled prospective cohort study (PCS)	A cohort study in which subjects are identified prospectively and classified as having received the intervention or comparator of interest on the basis of the prospectively collected information [47]. Data for individuals are usually analyzed. However, it is important to note that nonrandom receipt of a health care intervention is almost always defined in relation to organizational factors (ward, clinic, doctor, provider organization), and the analysis should take into account the data hierarchy. This is equivalent to a “pipeline design” used in health systems program evaluation. It is very similar to a NRCT, except with respect to the method of allocation.
Concurrently controlled retrospective cohort study (RCS)	A cohort study in which subjects are identified from historic records and classified as having received the intervention or comparator of interest on the basis of the historic information [48]. As for a PCS, data for individuals are usually analyzed, but the analysis should take account of the data hierarchy.
Historically controlled cohort study (HCS)	This type of cohort study is a combination of an RCS (for one group, usually receiving the comparator) and a PCS (for the second group, usually receiving the intervention) [49]. Thus, the comparison between groups is not contemporaneous. The analysis should take into account the data hierarchy.
Case–control study (CC)	Consecutive individuals experiencing an outcome of interest are identified, preferably prospectively, from within a defined population (but for whom relevant data have not been collected) and form a group of “cases” [50]. Individuals, sometimes matched to the cases, who did not experience the outcome of interest are also identified from within the defined population and form the group of “controls.” Data characterizing the intervention or comparator received in the past are collected retrospectively from existing records or by interviewing participants. The receipt of the intervention or comparator of interest is compared among cases and controls. If applicable, the analysis should take into account the data hierarchy.
Nested case–control study (NCC)	Individuals experiencing an outcome of interest are identified from within a defined cohort (for which some data have already been collected) and form a group of “cases.” Individuals, often matched to the cases, who did not experience the outcome of interest are also identified from within the defined cohort and form the group of “controls” [51]. Additional data required for the study, characterizing the intervention or comparator received in the past, are collected retrospectively from existing records or by interviewing participants. The receipt of the intervention or comparator of interest is compared among cases and controls. If applicable, the analysis should take into account the data hierarchy.
Before after study (BA)	As for CBA but without data for a control group of clusters [52]. An uncontrolled comparison is made between frequencies of outcomes for the two time points. This term may also be applied to a study in which a cohort of individuals have the outcome (e.g., function, symptoms, or quality of life) measured before an intervention and after the intervention [53]. This type of study comprises a single “exposed” cohort [54] (often called a “case series”), with the outcome measured before and after exposure. If applicable, the analysis should take into account the data hierarchy.
Cross-sectional study (XS)	The feature of this study design is that information required to classify individuals according to receipt of the intervention or comparator of interest and according to outcome are collected at the same time, sometimes preventing researchers from knowing whether the intervention preceded the outcome [55]. In cross-sectional studies of health interventions, despite collecting data about the intervention/comparator and outcome at one point in time, the nature of the intervention and outcome may allow one to be confident about whether the intervention preceded the outcome. This design is similar to the XS design described in Box 1.