Table 1.
Characteristics of Included Reviews Comparing Observational Studies Using Propensity Score Methods with Randomized Trials
| Systematic comparison | Topic area | Number of comparisons (number of observational studies/randomized trials) | Median number of patients in observational studies/randomized trials | Sources of observational studies | Sources of randomized trials | Pairing of observational studies and randomized trials | Methods for handling topics with multiple studies per design |
|---|---|---|---|---|---|---|---|
| Dahabreh (2012)14 | Acute coronary syndromes | 17 (21/63) | 5194/177 | MEDLINE (top 8 journals in “Cardiac and Cardiovascular systems” and top 4 journals in “Medicine, general and internal” by impact factor) | CDSR, MEDLINE, evidence-based guidelines of the American Heart Association/American College of Cardiologists, a compendium of medical therapeutics (Washington Manual of Medical Therapeutics), and reference lists | Two reviewers matched each observational study to at least one randomized trial, using a PICO scheme and a structured search protocol | Meta-analysis using random effects models (DerSimonian-Laird); common effect models (sensitivity analysis) |
| Lonjon (2014)18 | Surgery | 48 (70/94) | 2049/179 | MEDLINE via PubMed | MEDLINE via PubMed | For each eligible observational study, a single reviewer searched for matching randomized trials using a PICO scheme | Meta-analysis using random effects models (DerSimonian-Laird); common effect models |
| Zhang (2014a)19 | Sepsis | 8 (14/40) | 4641/359* | PubMed, Scopus, EBSCO | PubMed | Two investigators independently matched observational studies to randomized trials or most updated systematic review of randomized trials using a PICO scheme | Meta-analysis using random effects models (DerSimonian-Laird) |
| Zhang (2014b)20 | Critical care medicine | 20 (20/130†) | 433/150 | PubMed | PubMed | Each observational study was matched to one or more randomized trials using PICO | Meta-analysis using random effects models (DerSimonian-Laird) |
| Kitsios (2015)15 | Critical care medicine | 18 (21/58) | 1327/118 | MEDLINE (top 5 journals that publish primary clinical research studies and top 4 journals in “Medicine, general and internal” by total citations), reference lists | CDSR, PubMed, reference lists, and reference files of intensivists on review team | Two reviewers independently matched each observational study to at least one randomized trial, using a PICO scheme and a structured search protocol | Meta-analysis using random effects models (REML) |
| Hemkens (2016)21 | Mixed clinical areas | 16 (16/36) | 2086/985 | PubMed | PubMed, Cochrane Library | Searched for observational studies published before an randomized trial was conducted, up to 2010; then, searched for randomized trials, systematic reviews, or meta-analyses of trials conducted after 2010 on the same clinical topics | Meta-analysis using random effects models; common effect models (sensitivity analysis); Peto’s method for event rates < 1% |
CDSR, Cochrane Database of Systematic Reviews; EBSCO, Elton B. Stephens Co. database; NRS, nonrandomized studies; PICO, population, interventions/comparators, outcomes
*Average study size, by design
†Exact number of randomized trials not specified so estimate is used