We identified 7950 articles from our initial search, and 7092 after initial deduplication (Figure 1). Following title and abstract screening, 97 records had full-text review and seven articles were included reporting six different studies (17, 30-34). We contacted 16 authors (including from these six studies) to request summary data, of which 10 authors replied: 4 authors shared data, 4 authors confirmed they did not report outcomes of interest stratified by delirium symptom domains, 1 study did not provide summary data due to a conflict of interest (i.e. the authors plan on publishing these data at a later date) and 1 author was unable to share data due to lack of patient consent. The main reason for exclusion of articles was that results stratified by delirium symptom domains were not reported and not provided by authors. Two articles reported results from the same cohort (17, 18): the article which provided additional prognostic information on a sub-domain of delirium was included in the synthesis of results (18).
All six studies were cohort studies, four prospective and two retrospective. Studies were published between 2013 and 2020 from the UK, USA, Italy, and Brazil (Table 1). Three studies were identified through conference abstracts; the authors of one study provided summary data (33) and authors of the other two studies shared the unpublished manuscript (30) or the accepted manuscript (34). All studies recruited hospitalised patients from acute medical wards, orthopaedic wards or emergency departments. The sample size varied from 108 to 2521, with a total of 6002 patients included in the review population of which 1112 (18.5%) had delirium as assessed by diagnostic criteria or a validated diagnostic instrument. Patients in one study all had a delirium diagnosis (30). The remaining five studies reported a prevalence of delirium ranging from 2.9% to 44.4% (this wide range reflecting different selection criteria). Mean or median age ranged from 77 to 84.4 years. Baseline co-morbidity data were collected in all studies and included dementia diagnosis (4 studies, prevalence 38.9% to 57%), cognitive impairment (1 study) and illness severity (3 studies).
Five studies used a delirium screening tool to assess delirium: the CAM (35) or its variants (short CAM, or s-CAM, and CAM for the Intensive Care Unit) and/or the DRS-R98 (11). Two studies reported diagnosing delirium according to the Diagnostic Statistical Manual of Mental Disorders, 4th Edition Text Revision (DSM-IV-TR) (30) or the International Classification of Mental and Behavioural Disorders 9th Revision Clinical Modification (ICD-9-CM) (31) (Table 1).
Arousal: Five studies assessed level of arousal with various tools: short-CAM (32, 34); Glasgow Coma Scale (34); Richmond Agitation Sedation Scale (RASS (17, 33)); and/or the Observational Scale for Level of Arousal (OSLA (30, 33). Studies which used a scale for assessing arousal dichotomised the scale to determine whether altered arousal was present or absent; the degree (or severity) of arousal disturbance was not considered. Diwell et al. (32) did not define ‘altered level of consciousness’, but it was clear from the paper that this term referred to altered level of arousal as assessed with the short CAM (36). One study distinguished between reduced and increased arousal (17), another study used GCS with scores ≤13 to indicate decreased arousal (34), and the remaining studies did not specify whether arousal was increased or decreased.
Attention: Attention was assessed in three studies, using either a computerised test of focused and sustained attention (Edinburgh Delirium Test Box (37)) (33), counting backwards from 20 to 1 and the Months of the Year backwards test (31), or patient observation (as part of the s-CAM) (32). As with altered arousal, inattention was coded in the studies as present or absent, and the degree of inattention was not considered in the analyses.
Other neuropsychological domains: in single studies, the presence of disorientation (name current year, month and time of day) and memory deficits (immediate and delayed recall), both part of the Short Blessed Test (38), and disorganised thinking (included in the CAM assessment (35)) was assessed. There were no studies that collected data on the association between psychotic features, visuospatial deficits or affective disturbances in delirium with mortality, or secondary non-mortality outcomes.
Three studies reported on the timing of the initial assessment: day 1 (33) or day 3 (17, 32) after admission or at time of enrolment (17). No other studies reported on timing and frequency of assessments.
One study reported the primary outcome of 30-day mortality (34). Three studies reported mortality at other follow-up timepoints, namely at 4, 6 and/or 12 months (17, 30, 33). One study reported in-hospital mortality (31) and another study reported mortality according to the UK Office of National Statistics where death was flagged by a certified death certificate, however details of the exact timepoint of this were not provided (32) (Table 1).
We found no studies which reported delirium symptom domains and any of the non-mortality secondary outcomes.
All studies reported unadjusted and adjusted data. Multivariable Cox proportional hazard models were performed to produce HRs of survival in four studies (17, 30, 32, 34), and a form of logistic regression with ORs was used in two studies (31, 33). All studies reported 95% CIs. One study reported model adjustments for prevalent delirium (34). No other studies adjusted their models for delirium or other domain impairments.
Risk of bias
Risk of bias varied across studies but was moderate to high overall (Figure 2). This was mainly due to: possible selection bias, either because studies recruited patients from a convenience sample and/or because of the retrospective study design used (17, 33, 34); lack of reported blinding of outcome assessments (17, 30, 31, 34); and the absence of a pre-published protocol to allow assessment of selective outcome reporting (all studies). The risk of bias was mostly low for the consideration of confounding variables, with sufficient information in individual studies regarding baseline variables such as dementia, co-morbidities, functional dependence, frailty, and illness severity, and inclusion of these variables in multivariate analyses. Most studies used a validated scale to assess delirium, although training and frequency of assessments were often poorly reported, resulting in unclear risk of bias (17, 32, 33). One study (31) diagnosed delirium by retrieving ICD-9 codes from web-based case reports without describing the method in detail (unclear risk of bias). Four studies (30, 31, 33, 34) addressed how missing data were treated and accounted for in the analysis (low risk of bias), whereas the remaining three studies (17, 32) did not report the number and reasons for loss to follow-up (unclear risk of bias).
Synthesis of results
Altered level of arousal
Altered (versus normal) level of arousal (with or without delirium) was associated with higher mortality at 30 days, and at 4, 6, and 12 months (17, 30, 33, 34) (Table 1 and Additional file 3: Table 2 expanded). Random effects meta-analysis indicated that presence of altered arousal (compared to normal arousal) was associated with higher mortality across follow-up time-points ranging from 30 days to 12 months (pooled OR 2.80, 95% CI 2.33-3.37, I2=0%, N=3481 (1040 delirium), risk of bias moderate-to-high: possible selection bias, unclear or lack of blinding of assessments and unclear risk of selective outcome reporting) (Fig 2). The quality of evidence for the association between altered arousal and higher mortality was moderate, downgraded because of high risk of bias (all but one study had unclear or high risk of bias ratings in at least three RoBANS domains), indirectness, and publication bias.
A sensitivity analysis excluding the Jackson et al. study (30) did not alter these findings (pooled OR 2.80; 95% CI 2.32-3.37).
Garcez et al. (34) was the only study reporting on the association between level of arousal and 30-day mortality, the primary outcome. In this retrospective study, altered (compared to normal) arousal was independently associated with higher 30-day mortality after adjusting for confounders including delirium, both when arousal was defined using s-CAM criteria (HR 2.33, 95% CI 1.66‒3.27) and GCS scores (HR 1.62, 95% CI 1.13‒2.33; GCS ≤ 13). The adjustment for delirium may have led to over-adjustment especially when using s-CAM criteria to define both arousal and delirium status.
Hall et al (33) reported that altered level of arousal was associated with increased risk of death at 12 months (OSLA: HR 2.21, 95% CI 1.01-4.86; RASS: HR 2.13, 95% CI 1.03-4.4).
Han et al. (17) compared the associations between each arousal subtype in delirium (normal/increased/decreased, using nondelirious patients as the reference group) and 6-month mortality in patients presenting in the emergency department. Of all three delirium arousal subtypes, delirium with normal arousal had the highest 6-month mortality (HR 3.1, 95% CI 1.3-7.4) although the sample size for this subgroup was very small (N=15). Furthermore, this was the only paper that found an association between increased arousal in delirium and mortality (HR 1.4, 95% CI 0.9-2.1; N=8), again using a very small sample size. The direction of the effect was similar to that found for decreased arousal in delirium (HR 1.4, 95% CI 0.9-2.1; N=132). A different paper describing the same cohort further reported that patients with altered arousal, irrespective of delirium status, were more likely to die within 6 months (HR 1.73, 95% CI 1.21-2.49) compared to those with normal arousal (18).
Jackson et al. (30) only included patients with delirium on admission who were followed up at 4 and 12 months. Delirium with reduced arousal and hypoactive motor symptoms in hospital was associated with higher mortality at 12 months when compared to delirium patients with hyperactive features (i.e. mixed and hyperactive motor subtypes; HR 1.09, 95% CI 1.02-1.18). In addition, analysis of arousal subtypes, informed by scores on the OSLA, DRS-R98 and clinical descriptions, found an association between the hypoactive (vs. hyperactive/mixed) subtype with higher mortality at 4 months (HR 3.18, 95% CI 1.13-8.93).
Diwell et al. (32) did not find a significant association between level of consciousness (i.e. altered arousal) and mortality in adjusted analysis (HR 1.33 95% CI 0.98, 1.79).
Results from the meta-analysis showed that presence of inattention (pooled across patients with and without delirium) was associated with higher mortality compared to normal attention and/or no delirium (pooled OR 2.57, 95% CI 1.74-3.80; I2=0%, risk of bias moderate: possible selection bias, unclear or lack of blinding of assessments, and unclear risk of selective outcome reporting; 3239 patients (387 delirium)). The quality of evidence for the association between inattention and higher mortality was low, downgraded because of risk of bias (two out of three studies had an unclear or high risk of bias in at least three RoBANS domains), imprecision, indirectness, and publication bias.
Only one of three original studies assessing attention reported a statistically significant association between inattention in patients with delirium and higher mortality. Bellelli and colleagues (31) reported an association between inattention in older patients (with and without delirium) and in-hospital mortality (OR 3.26, 95% CI 2.03-5.24) compared to a non-delirious group. It is noted that this study reported a lower prevalence rate of delirium (2.9%) than expected given the setting, suggesting that a proportion of cases with delirium were missed. Diwell and colleagues (32) reported a statistically significant association between inattention with higher mortality, but this association disappeared after adjustment for confounders (HR 1.24 95% CI 0.92, 1.67).
Other neuropsychological deficits
The evidence regarding other symptom domains of delirium and mortality is limited (two studies : 3131 patients, 343 with delirium (31, 32)). Bellelli and colleagues (31) found higher in-hospital mortality in delirium patients who had evidence of disorientation (OR 3.85, 95% CI 2.43-6.1) and memory deficits (OR 2.92, 95% CI 1.33-6.39) compared to non-delirious patients. Interestingly, the diagnosis of delirium itself was not a predictor of mortality in this study. The authors further noted that delirious patients mostly had multiple neurocognitive deficits and that patterns of inattention with memory and/or orientation deficits were shown to be more strongly associated with in-hospital mortality compared to having only one neurocognitive deficit.
Diwell and colleagues (32) found an association between the presence of disorganised thinking in patients with delirium (compared to no delirium) and higher mortality (timescale not specified but identified through the UK Office for National Statistics; HR 1.42, 95% CI 1.05-1.92).