Skip to main content

Table 3 Statistical methods applied to delirium incidence

From: Heterogeneity in design and analysis of ICU delirium randomized trials: a systematic review

Statistical method

Overalla

Primary outcomea

Secondary outcomea

n = 56

n = 48

n = 8

Two-sample test for proportionsb

51 (91)

45 (94)

6 (75)

Two-sample test for meansc

1 (2)

1 (2)

0 (0)

Non-parametric testd

2 (4)

1 (2)

1 (13)

Binomial regression modele

1 (2)

0 (0)

1 (13)

Longitudinal regression modelf

2 (4)

1 (2)

1 (13)

Survival analysisg

9 (16)

6 (13)

3 (38)

Competing risk survival analysish

1 (2)

1 (2)

0 (0)

Joint modeli

1 (2)

1 (2)

0 (0)

  1. Values in the table are count (%). Several statistical methods may be reported for each outcome; therefore, column counts (%s) will not sum to the number of primary or secondary outcomes or 100%
  2. aThe sample size, n, reported as “overall” is the total number of delirium incidence outcomes, both primary and secondary, whereas the sample size reported for primary and secondary delirium incidence outcomes is the number of trials. A trial may report multiple delirium incidence outcomes, e.g., delirium incidence by 14 or 28 days as the primary and secondary outcomes, respectively. There were a total of 56 delirium incidence outcomes reported by 50 of the 65 trials; 42, 6, and 2 trials reported only a primary, both a primary and secondary, or only a secondary delirium incidence outcome, respectively
  3. bTwo-sample test for proportions includes two-sample test for proportions assuming normally distributed sample proportions, Fisher’s exact test, chi-square test, and logistic regression model
  4. cTwo-sample test for means includes two-sample t test, analysis of variance, or linear regression model
  5. dNon-parametric test for continuous or ordinal outcomes includes Mann-Whitney test, Wilcoxon rank-sum test, Kruskal-Wallis test, and the proportional odds logistic regression model
  6. eBinomial regression model defines the number of days with delirium as the binomial outcome and the number of days in the ICU as the offset/denominator
  7. fLongitudinal regression model includes marginal longitudinal logistic regression models for daily delirium and random effects logistic regression models for daily delirium
  8. gSurvival analysis defined the outcome as time from randomization to delirium onset with patients censored at ICU discharge or death; statistical comparisons were made using the log-rank test or the Cox proportional hazards regression model
  9. hCompeting risk survival analysis defined the outcome as time from randomization to delirium onset with (i) patients censored at ICU discharge and death defined as a competing risk or (ii) ICU discharge and death defined as competing risks; statistical comparisons were made using the Fine and Gray competing risk model
  10. iJoint model refers to the joint model for recurrent event outcomes (e.g., recurrent delirium events) with terminating event (e.g., ICU discharge or death) proposed by Rondeau [23]