Risk Factors for Delirium After On-pump Cardiac Surgery

A Systematic Review

Alex NC Gosselt; Arjen JC Slooter; Pascal RQ Boere; Irene J Zaal


Crit Care. 2015;19(346) 

In This Article



The review was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis statement recommendations for conducting a systematic review.[20] Both the study protocol[58] and the full search strategy[58] were registered before initiation of the search.

Eligible Studies

Five databases were searched (CINAHL, EMBASE, MEDLINE, the Cochrane Central Register for Controlled Trials, and the Cochrane Database of Systematic Reviews) for relevant articles or abstracts published from January 1990 through January 2015. With an online registered comprehensive search strategy using separately formulated strings for the domain (cardiac surgery patients) and the outcome (delirium) we searched for eligible studies. We deliberately left out the determinant (risk factors) to lower the possibility of missing relevant articles. We chose 1990 as the initial search year because this was the year that a well-defined screening instrument for delirium based on the Diagnostic and Statistical Manual of Mental Disorders III-revised (DSM III-R) definition was published.[21]

Study Selection

All abstracts and titles were screened in duplicate for potentially relevant studies, which were considered in full-text form by two authors (ANCG and PRQB). Whenever in doubt, studies were discussed with a third author (IJZ) before inclusion. We reviewed personal files, reference lists of review articles, and reference lists of eligible studies for additional investigations to identify additional relevant publications that were missed during the computerized search. We contacted the corresponding author if no full text version of the article was available to us or to inquire for missing data. Corresponding authors who failed to respond after the first contact were contacted one additional time over a 6-week period.

Cohort studies or randomized controlled trials were included that evaluated adults (age ≥18 years) undergoing cardiac surgery using CPB, where at least one potential risk factor for delirium was considered. The risk factor had to be present before delirium onset, and all patients had to have been evaluated for delirium at least once daily using a validated instrument. We excluded studies that also evaluated patients undergoing cardiac surgery without CPB or solely aortic surgery because of possible differences in the pathobiology of delirium. Further, cohort studies that failed to evaluate risk factors using a multivariable approach were excluded given the multitude of possible confounding variables. Articles published in a language other than English, Dutch or German were excluded.

Data Extraction

All data on the risk factor selection process and final association was independently extracted by two authors (ANCG and PRQB). Variables with an uncertain cause − effect relationship were excluded from the final tables. Discrepancies were resolved through discussion with a third author (IJZ) when necessary. In the case of separate studies using the same study cohort for the analysis, significant variables in both studies were counted only once.

Assessment of the Risk of Bias

Two authors (ANCG and PRQB) independently assessed the risk of bias in each included study by using adapted versions of the Scottish Intercollegiate Guidelines Network (SIGN) checklist for controlled trials and cohort studies.[22]

The SIGN RCT checklist considered randomization strategy, treatment allocation concealment, blinding, success of randomization, use of intention-to-treat principles and the completeness of the reported outcome data (see Additional file 1A http://www.ccforum.com/content/19/1/346/additional). The SIGN checklist for cohort studies was modified a priori by consensus. The checklist for cohort studies considered selection bias, performance bias, loss to follow up resulting in attrition bias, detection bias and statistical analysis. In order to be included in this study, a validated assessment for delirium had to be done either by use of the DSM III-R criteria[23] (or a newer version), or by a delirium screening tool which had to be validated for the specific population e.g., ventilated/ICU patients (Additional file 1B http://www.ccforum.com/content/19/1/346/additional).

If a checklist criterion was met, one point was assigned, studies that failed to meet a criterion or provided insufficient information, resulting in a 'cannot state', and received no points. The maximum attainable score was 8 points for cohort studies and 9 points for randomized controlled trials (RCTs). 'Not applicable' items were counted as one point as it was deemed not to be a consequence of insufficient study quality. Items deemed 'not applicable' did not illustrate methodological shortcomings and were counted as a point.

If a secondary multivariable analysis was performed within an RCT to identify independent risk factors other than the main factor under investigation, this analysis was separately assessed for bias using the cohort checklist. Any disagreement during the quality scoring process was resolved by discussion with a third author (IJZ). A priori, cohort studies and RCTs were deemed high quality (HQ) when the score was ≥7 and ≥8 points, respectively, acceptable quality (AQ) when the score was 5–6, and 6–7, respectively, and unacceptable quality when the score was ≤4, and ≤5 points, respectively.

Data Synthesis

Strength of evidence supporting an association between a risk factor and delirium was summarized using a semiquantitative approach. Variables that were analyzed using a multivariable approach, in at least two studies and that showed either a statistical significant association or an odds ratio ≤0.5 or ≥1.5 were included in the best evidence synthesis. Due to the fair amount of studies with small sample sizes we did not solely want to depend on statistical significance as this is highly dependent on sample size and other included variables. If an HQ study described a variable to be initially included in a stepwise selection process but was not presented in the final model it was categorized as having no association with delirium in a multivariable approach.

The available evidence for potential risk factors for postoperative delirium after cardiac surgery was quantitatively evaluated using three criteria: 1) the number of studies evaluating the variable, 2) the scored quality of each study evaluating the variable, and 3) the consistency of the association between the variable and risk of delirium. An association was deemed consistent when ≥75 % of the studies evaluating the variable reported the same direction of association. Variables found to have no association using a multivariable approach were also taken into account.

A variable evaluated at different time-points (e.g., perioperative or intraoperative blood product transfusion), or a variable for which slightly different definitions were used, was considered similar and therefore grouped. Details describing these differences are specified in Additional file 2 http://www.ccforum.com/content/19/1/346/additional.

We graded the strength of evidence for each association as follows: strong when the association was consistent in ≥2 HQ studies; moderate when consistent in 1 HQ study and ≥1 average quality (AQ) study or ≥3 AQ/LQ studies; inconclusive when the observed association was not consistent or was evaluated in 1 HQ, <3 AQ studies or solely in LQ studies. We concluded no evidence for an association if no significant association was found in HQ multivariable analysis or at least 3 HQ studies found no association in univariable analysis (Table 1).