Skip to main content
. 2023 Aug 26;27:329. doi: 10.1186/s13054-023-04621-4

Table 2.

GRADE evaluation of the certainty of evidence

Certainty assessment № of patients Effect Certainty Importance
№ of studies Study design Risk of bias Inconsistency Indirectness Imprecision Other considerations Haloperidol Placebo Relative (96.7%/98% CI)* Absolute (96.7%/98% CI)*
All-cause mortality (follow-up: range 28 days to 90 days)
5 Randomised trials Not seriousa Not serious Not serious Seriousb Nonec 272/789 (34.5%) 295/764 (38.6%) RR 0.89 (0.87 to 1.03) 42 fewer per 1.000 (from 50 fewer to 8 more)

⨁⨁⨁◯

Moderate

Critical
SAE/SAR (highest proportion) (follow-up: range 3 days to 90 days)
5 Randomised trials Not seriousa Seriousd Not serious Seriousj Nonec 311/789 (39.4%) 331/764 (43.3%) RR 0.94 (0.81 to 1.10) 26 fewer per 1.000 (from 82 fewer to 43 more)

⨁⨁◯◯

Low

Critical
SAEs/SARs (cumulated events) (follow-up: range 3 days to 90 days)
5 Randomised trials Not seriousa Seriousd,e Not serious Very seriousf Nonec 426/789 (54.0%) 440/764 (57.6%)

RR 0.97

(0.85 to 1.11)

17 fewer per 1.000 (from 86 fewer to 63 more)

⨁◯◯◯

Very low

Critical
Days alive without delirium or coma (follow-up: 14 days)
3 Randomised trials Not seriousg Not serious Not serious Not serioush Missing data have the potential to change evaluationc 7.2 6.8 MD 0.33 days more (0.31 fewer to 0.97 more)

⨁⨁⨁◯

Moderate

Important
Delirium severity
0 Randomised trials Not estimable Important
Cognitive function
0 Randomised trials Not estimable Critical
Health-related quality of life
0 Randomised trials Not estimable Critical
QTc prolongation (follow-up: range 3 days to 90 days)
3 Randomised trials Not serious Not serious Not serious Very seriousi Nonec 28/709 (3.9%) 18/683 (2.6%) RR 1.47§ (0.78 to 2.76) 12 more per 1.000 (from 5 fewer to 43 more)

⨁⨁◯◯

Low

Important

CI, confidence interval; MD, mean difference; RR, risk ratio

*TSA-adjusted 96.7% CI for primary outcomes and TSA-adjusted 98% CI for secondary outcomes

§95% CI

a3 of 5 trials were at overall low risk of bias. Trials with high risk of bias had small sample size and did not have much influence on the pooled effect estimate

bTSA showed that we did not have sufficient information to confirm or reject a 20% RRR. The cumulative z-curve did not cross monitoring boundaries. The accrued information size was between 50 and 100% of required information size, and therefore downgraded 1 level

c3 trials comparing haloperidol with placebo were identified in trial registers. These trials were terminated or had status unknown and trial results were not available. These unpublished trials were not considered as serious publication bias

dTrials did not adhere to ICH-GCP. Some trials reported SAEs others reported SARs

eI2 was 87% (P < 0.01) for cumulated SAEs. Inconsistencies in number of events, but overlapping CIs

fThe accrued information size was < 50% of the required information size estimated by TSA. No TSA monitoring boundaries were crossed. Therefore, we downgraded with 2 levels for imprecision

g3 of 4 trials were at low risk of bias and included the majority of patients

hTSA showed that the cumulative z-curve crossed the futility area with 80% of the required information size

iAccrued information was < 5% of the required information size. Downgraded 2 levels

jTSA reached futility for a RRR of 20%, but CI was wide and did include a potential 20% RRR but also a potential 10% RRI, indicating uncertainty. Downgraded 1 level