Table 2.
Strategies and methods for test comparisons
| Characteristic | Comparative reviews |
Multiple test reviews | Total | |
|---|---|---|---|---|
| Statistical analyses to compare test accuracy | ||||
| Yes | No or unclear | |||
| Number of reviewsa | 53 (42) | 29 (23) | 45 (35) | 127 (100) |
| Study type | ||||
| Comparative only | 8 (15) | 8 (28) | 0 | 16 (13) |
| Any study type | 45 (85) | 21 (72) | 45 (100) | 111 (87) |
| Test comparison strategy | ||||
| Direct comparison only | 8 (15) | 8 (28) | 0 | 16 (13) |
| Indirect comparison only—comparative studies available | 26 (49) | 10 (34) | 4 (9) | 40 (32) |
| Indirect comparison only—no comparative studies available | 2 (4) | 6 (21) | 1 (2) | 9 (7) |
| Both direct and indirect comparison | 17 (32) | 5 (17) | 0 | 22 (17) |
| None | 0 | 0 | 40 (89) | 40 (32) |
| Method used for test comparisonb | ||||
| Meta-regression—hierarchical model | 18 (34) | 0 | 0 | 18 (14) |
| Meta-regression—SROC regression | 2 (4) | 0 | 0 | 2 (2) |
| Meta-regression—ANCOVA | 2 (4) | 0 | 0 | 2 (2) |
| Meta-regression—logistic regression | 1 (2) | 0 | 0 | 1 (1) |
| Univariate pooling of difference in sensitivity and specificity or DORs | 6 (11) | 0 | 0 | 6 (5) |
| Naïve (comparison of pooled estimates from separate meta-analyses) | 0 | 0 | ||
| Z-test | 15 (28) | 0 | 0 | 15 (12) |
| Paired t-test | 1 (2) | 0 | 0 | 1 (1) |
| Unpaired t-test | 1 (2) | 0 | 0 | 1 (1) |
| Chi-squared test | 1 (2) | 0 | 0 | 1 (1) |
| Comparison of Q* statistic and their SEsc | 1 (2) | 0 | 0 | 1 (1) |
| Overlapping confidence intervals | 0 | 3 (10) | 0 | 3 (2) |
| Narrative | 0 | 9 (31) | 4 (9) | 13 (10) |
| None | 0 | 14 (48) | 40 (89) | 54 (43) |
| Unclear | 5 (9) | 3 (10) | 1 (2) | 9 (7) |
| Relative measures used to summarize differences in test accuracy | 18 (34) | 0 | 0 | 18 (14) |
| Multiple thresholds included | 13 (25) | 12 (41) | 17 (38) | 42 (33) |
| If multiple thresholds included, were they accounted for in the comparative meta-analysis (meta-analysis at each threshold or fitted appropriate model) | ||||
| Yes | 6 (46) | 0 | 0 | 6 (46) |
| No | 4 (31) | 0 | 0 | 4 (31) |
| Unclear | 3 (23) | 0 | 0 | 3 (23) |
Abbreviations: ANCOVA, analysis of covariance; DOR, diagnostic odds ratio; SE, standard error; SROC, summary receiver operating characteristic.
Numbers in parentheses are column percentages unless otherwise stated. Percentages may not add up to 100% because of rounding.
a
Numbers in parentheses are row percentages.
b
These methods either involve a comparative meta-analysis or follow-on from a meta-analysis of each test individually.
c
Moses et al. [11] proposed the Q* statistic as an alternative to the area under the curve. Q* is the point on the SROC curve where sensitivity is equal to specificity, that is, the intersection of the summary curve and the line of symmetry.