The detection of delirium in admitted oncology patients: a scoping review

Key summary points Aim To understand the validation of delirium detection tools in medical oncology, as well as identify data on incidence, prevalence and reversibility in this setting. Findings Of twelve studies, only four used case ascertainment methods following published recommendations, six studies had a low risk of bias. Message In delirium tool validation studies in the oncology setting, choice of appropriate gold standard for case ascertainment is a critical factor. New tools and new validations are not recommended, rather the critical application of existing tools depending on appropriate validation and clinical practicality for the setting. Supplementary Information The online version contains supplementary material available at 10.1007/s41999-021-00586-1.


Background and aim
Delirium is a neurocognitive syndrome characterised by an altered level of arousal, altered awareness and cognition, and a reduced ability to direct, focus, sustain, and shift attention [1,2]. Delirium is associated with increased morbidity and mortality, longer length of stay [3,4], and marked distress for cancer patients, their families and staff [5,6]. Delirium is common in hospitalised patients [1,2], and outcomes can be improved via prevention [7,8] and effective management [9]. The use of validated assessments improves detection and provides earlier identification of patients with delirium [10].
Under-diagnosis of delirium is an important issue in clinical settings [11]; outcomes are worse if the diagnosis of delirium is delayed or missed entirely [12]. Studies of general hospital patients indicate that pain, younger age, correct orientation in person, place and time, and previous psychiatric diagnosis, especially bipolar disorder or psychosis, are important risk factors for the diagnosis of delirium being missed [13]. One study has shown increasing age, poor performance on cognitive testing and lower serum albumin to be associated with a higher risk of delirium in the hospitalised cancer patients, however, less is known about factors which increases the misdiagnosis of delirium in cancer populations or whether there are specific clinical factors which can be used to mitigate risk [3].
The majority of epidemiological studies in delirium have targeted people over 65 years of age [14]. Although guidelines for the management of delirium in cancer settings exist [15], fewer studies have primarily focussed on adults (defined as 18yrs or older) in an acute hospital, oncology, inpatient setting [16][17][18]. More commonly studies including cancer patients have been in "stand-alone" palliative care units [19,20], or subsets of cancer inpatient cohorts on the basis of palliative care [21][22][23] or psychiatry consultation/liaison services in acute hospitals [11]. A recent review of delirium in the palliative care setting yielded a point prevalence estimate of 35% [95% confidence interval (CI) = 0.29-0.40] at inpatient admission. [20] Studies indicate that in the palliative care cancer setting at least, whilst the prevalence of delirium is high, it remains reversible in approximately half of cases [24]. These data also lend support to the case for improved detection. Of interest, reversibility in the palliative care setting although not a universal possibility, has been associated with factors such as delirium aetiology specifically opioid, or other psychoactive medication, or dehydration, and where there is a less severe cognitive disturbance or absence of organ failure [9,25].
We chose a scoping review methodology because initial searches yielded few returns in the target setting. We also chose to take a broad approach to clarify key concepts in delirium detection in cancer settings and identify key concepts and gaps in the evidence base [26]. Our review explores the literature in relation to delirium detection and missed delirium in the inpatient oncology setting, and clinical factors associated with misdiagnosis. The aim of this scoping review is to synthesise knowledge and identify gaps relating to detection tool selection, incidence, prevalence and reversibility of delirium in hospitalised, adult patients with cancer.

Patients and methods
The target population was admitted, adult, oncology patients in an acute-hospital or comprehensive cancer centre. The research questions were: 1. Which instruments are most commonly used to detect delirium? 2. Which reference standards have been used to measure rates of delirium and compare the performance of delirium screening instruments? 3. What is the incidence and prevalence of delirium in the target setting? and 4. What is the rate of reversibility of delirium in the target setting?
Our search strategy centred on four key domains; delirium, cancer, inpatient oncology, and delirium detection. Full inclusion criteria were: original study, English language, for inclusion the focus of the study must be syndromic delirium e.g. not: confusion, cognitive impairment, acute brain syndrome. The target population is patients with cancer and the setting is adult inpatient oncology, studies not relevant to this population were excluded. Specifically, the target setting was oncology wards in acute hospitals including tertiary referral and cancer centres. Studies of non-oncology ward patients were included if the oncology population could be abstracted from a broader study e.g., hospital-wide point prevalence, subset of cancer patients within an index population of older patients with cancer. Studies set in palliative care populations in a "stand alone" inpatient unit or hospice were only included if the setting was combined oncology and palliative care, for example a comprehensive cancer centre. To meet inclusion a delirium assessment with a validated objective tool, or clinical diagnostic criteria was also required.
Studies were excluded if they were solely conducted in the following settings or populations; haematology or nonsolid haematological malignancy, non-cancer palliative care, perioperative including surgical oncology, or alcohol withdrawal delirium.
The reason for excluding non-solid haematological malignancy was the consideration that illness trajectories and treatment protocols in this population may differ a great deal from those of solid cancers, similarly for non-cancer palliative care patients. The exclusion of surgical oncology and peri-operative settings was pragmatic as those patients may be admitted to surgical wards with a different background for staff and potentially different delirium aetiologies. Understanding delirium in these patient cohorts is important and we hope that future work will address areas not included in our review as has been the case in recent multicentre delirium prevalence studies [27,28].
All authors and an academic liaison-librarian were involved in an iterative process to determine search terms. MEDLINE, CINAHL, PsycINFO, EMBASE and SCOPUS databases were searched. Publication date was limited from 1st of January 1996 to 12th of August 2017. A full list of keywords and Medical Sub-heading (MeSH) is available in Appendix 1.
Independent title, abstract, full-text review and cross check was carried out by MBS and IW, using COVIDENCE [29] software, with conflicts resolved by consensus. Where the same study was reported in more than one manuscript, additional information from related or subsequent publications was included where possible [9,[30][31][32][33][34]. Study heterogeneity was not objectively tested, but the overall lower quality of several included studies and issues with reference standards seemed to suggest meta-analysis would not be meaningful, but sources of bias and generalisability were assessed using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) system. [35] Two authors (MBS and IW) independently piloted the QUADAS-2 and subsequently quality considerations and information synthesis was reviewed by all authors consensus was achieved through discussion.
For the purposes of this study, we defined a delirium reference standard as one which determined diagnostic assignment based on an instrument which used an independent reference-rater evaluation [36]. This last point, although identified in the literature was also arrived at via an iterative process that revealed unclear distinctions between screening tools and case ascertainment or diagnostic criteria upon which case identification was verified among included studies. Examples of reference standards in the basis of these criteria are the World Health Organization (WHO) International Classification of Diseases, 10th Revision (ICD-10) [37] or the American Psychiatric Association Diagnostic and Statistical Manual of Mental Disorders (DSM) [1,38] criteria, applied by a psychiatrist or consultant physician. On the basis of these criteria, the Confusion Assessment Method (CAM) meets reference standard criteria for case ascertainment, only in studies where reference-rater training in the use of the CAM is explicitly-stated [39]. This follows published recommendations for valid use of the CAM [39] along with diagnostic assignment in delirium research [36]. With regard to detection tool we use the term detection instrument (or tool) to include screening tools or other instruments put forward as standardised methods to identify delirium.
With regard to protocol registration, on inception authors were advised that PROSPERO did not currently accept registrations for scoping reviews and was unable to accept our application for protocol registration. The following is an accurate description of our methodology and further information is available on request. The data that support the findings of this study are available from the corresponding author.

Search results
The search date was August 12, 2017. Returns were as follows: Medline (211), EMBASE (684), SCOPUS (97), PsycINFO (52) and CINAHL (47). A total of 805 studies were identified with an additional 91 titles added from hand search. Although most duplicates were removed prior, for pragmatic reasons, final removal of duplicates and screening of abstract and date of publication for hand search returns was held over until full-text review. Hand search consisted of hand search of reference lists from included studies as well as search of authors PDF library using delirium as a title search.
In total, 91 studies were retained for full-text review, and 12 studies remained for data abstraction and synthesis ( Fig. 1)

Characteristics of included studies
Study recruitment periods ranged from 1997 to 2015. Study design of all 12 studies was observational: six were prospective, six were retrospective. Two studies were secondary analyses of data from prior prospective studies. Tables 1 and 2 provide detailed data for the included studies related to the research questions. Table 3 summarises quality considerations according to QUADAS-2 criteria [35]. Table 4 provides a summary Studies were grouped into three categories on the basis of the clinical setting: (1) inpatient, acute-hospital or comprehensive cancer centre oncology ward; (2) older oncology patients (patients > 65 years, admitted to acute hospitals under any admitting team, with cancer as the primary diagnosis or co-morbidity); and (3) palliative care ward in acute-hospital or comprehensive cancer centre (APCU). The rationale for this grouping was based in the observation that clinical care for oncology patients occurs largely in one of these three settings, but that from the point of view of research, these settings tend to be studied independently; we also wanted to decrease heterogeneity within subgroups, but facilitate understanding the use of delirium detection tools across the spectrum of admitted adult patients with cancer to improve care in this cohort.
Five studies [3,31,[40][41][42] were in the adult-oncology setting. Three of these approached all patients on nominated days [40][41][42], while two studies approached all admitted patients [3,43]. Three studies were of older cancer patients [27,32,44]. One of these was a point prevalence study in which patients over 65 years were recruited from more than 100 hospitals across several regions of Italy during one 24-h period [27]. Unpublished subset data on patients in this last cohort, were provided by the author. (personal communication G Bellelli, October 2017) [45] A further study recruited all patients aged 65 years or older admitted to the general medicine or oncology ward in two Dutch teaching hospitals [44]. The remaining study in older oncology patients, was a secondary analysis of a subset of cancer patients from a previous study, composed of patients from three North American centres [32]. Four studies [9,[46][47][48] were in an APCU. Three of these [46][47][48] were retrospective and based in the same health care facility.
Three studies focused on patients with cancer referred to consultation psychiatry services and reported misdiagnosis of neuropsychiatric conditions, with two studies reporting a missed diagnosis of delirium in 46%, and a further study reporting 63% missed cases [11,13,49] .

Patient recruitment and demographics
Patient characteristics were described in varying detail: four studies [3,9,30,40] specified histological diagnosis of cancer, and three specified consecutive recruitment [27,34,46]. Four studies gave a detailed description of recruitment [3,44,50], and seven provided the number of eligible patients when providing number of participants. Eight studies [3,9,31,41,42,[46][47][48] presented flow diagrams or data accounting for eligible patients not included in recruitment or analysis. Ten of 12 studies reported primary cancer types. All reported age; the average of the mean age (years) in each setting were as follows: oncology 59, older cancer 78, and APCU 60. Six studies reported length of hospital stay [3,32,41,44,47,48]; these were reported as mean or median, and ranged from 3 to 9.8 days. Clinical information describing cancer stage, co-morbidity burden, overall illness severity, functional status or vital status at discharge, were not uniformly described. Only one of five studies in the adult oncology setting provided detailed information that described markers of burden of disease [41]. Six of 12 studies across all setting subgroups reported the stage of cancer in terms of metastatic versus loco-regional disease [9,[41][42][43][44]47]. One study reported the number of patients receiving anti-cancer treatment [44].
Scoping questions; data relating to our four research questions 1. Which instruments are most commonly used to detect delirium?
Of the studies meeting our inclusion criteria, five used previously validated instruments for clinical detection of delirium: Nursing Delirium Screening scale (NuDESC, n = 1); Delirium Observational Screening Scale (DOSS, n = 1); four A's test (4AT, n = 1); Neelon and Champagne (NEECHAM, n = 1); and, Memorial Delirium Assessment Scale (MDAS, n = 4). One study tested a novel delirium screening tool (Single Question in Delirium; SQiD) and one tested cognitive measures (Clock Drawing Test, Mini Cognitive, Digit Span Test) against a reference standard. Six studies included a second delirium detection tool, as presented in Table 2. 2. Which reference standards have been used to measure rates of delirium?
Four of 12 studies met criteria for a delirium reference standard for case ascertainment [36]. Two studies in the adult oncology setting used diagnostic criteria, namely the ICD 10 [40] (assessor characteristics were not stated), and DSM IV/IVR (assessed by final year psychiatry fellow or psychiatrist) [42]. Two studies used the CAM in a way that met criteria for use as a reference standard, including an account of assessor training [3,31] .
Seven studies used a screening tool alone as the basis of case ascertainment of delirium: MDAS (n = 4); CAM (n = 2); DOSS (n = 1); 4AT (n = 1); NEECHAM (n = 1); Nu-DESC (n = 1), and one used a battery of tests of cognition (n = 1). Neither of these two studies using the CAM as the basis of delirium case ascertainment, specified assessor training [40,44]. Of the prospective studies, Bellelli    [41], and three studies used a cut-off score on the MDAS to identify delirium cases on chart review [46][47][48]. Studies comparing MDAS, MMSE, 4AT, NEECHAM tools for detection in clinical practice compared to a reference standard were not identified in our target settings, so it is not possible to ascertain the rate of missed delirium from the available literature.
3. What is the incidence and prevalence of delirium in this setting?
Rates of delirium incidence and prevalence reported by studies in this review are presented in Table 2. Table 4 presents a summary of tools used and delirium rates established on that basis. Consecutive or non-consecutive recruitment is also reported to aid interpretation of delirium rates.
Delirium reversibility was reported in three studies, two of these were in APCU settings; 46/94 (49%) [9], 68/229 (26%) [46], and one in older patients with cancer 13/43 (30%) [33]. Of four studies reporting reversibility [9,32,46,47], two [9,32] did not explicitly state how reversibility was defined; one used the MDAS or clinical documentation to determine delirium reversibility and another used a MDAS cut off score at day five. Although one of these [47] referenced a predating publication, this could not be found. The other three used MDAS scores collected as were more likely to have a resolution but found only prior cognitive impairment to be negatively associated with delirium reversal [33].
In the Lawlor study delirium associated with opioids and non-opioid psychoactive medication and dehydration were more likely to be reversed while non-reversed delirium was more common when associated with a respiratory infection, pulmonary cancer and metabolic causes [9].

Discussion
We identified significant knowledge gaps regarding epidemiological characteristics of delirium in oncology inpatients. A variety of delirium screening tools were identified, but few studies used accepted diagnostic or reference standards for case ascertainment. Sources of bias included study design and generalisability. A small number of eligible studies reported reversibility of delirium.
Delirium is a multifactorial syndrome. The relationship of delirium risk with demographic factors such as age and clinical factors (e.g. cancer diagnosis), is complex. One of the studies in the older cancer sub-group provided comparative figures of delirium point prevalence in cancer and non-cancer patients; 19.2% (n = 323) for patients with cancer compared to 23.9% (n = 1544) of the patients with no cancer diagnosis (p = 0.06) [27]. Within the included studies, but also in non-cancer settings reported elsewhere, multiple delirium risk factors such as co-morbidities, presence of physiological disturbance and medication burden are factors that seem to be constants in understanding overall delirium risk. [7,9,13,41,43,[51][52][53][54][55][56][57][58][59] Delirium screening tools have been developed, and validated, predominantly in older, hospitalised patients [60][61][62][63][64].

3
The tools for which psychometric properties have been assessed in cancer in-patients in the acute setting, are the Nu-DESC and MDAS [31,34,65]. There is a clear rationale for use of the Nu-DESC as a delirium screening tool [31]. The 4AT has been tested in "stand-alone" palliative care, inpatient settings [66] and for older adults admitted to hospital [50,67,68]. The MMSE can be used to screen for cognitive impairment it has been found to have poor performance as a bedside tool for identifying delirium [69,70]. The Confusion Assessment Method [36] (CAM) has several versions [67], and has well-established psychometric properties [61]. In the main, studies have supported the use of the CAM for delirium screening in research settings providing there is strict adherence to operator training, however, one study suggests even in the context of strict adherence to CAM training, sensitivity of the CAM may not be sustained [68]. Our review found that most studies using the CAM for case ascertainment did not describe the training staff underwent [3,37,39,41] and one, described difficulty in attaining adequate training in a clinically embedded research context [42,71].
Four of the twelve included studies used the MDAS as a basis for case confirmation of delirium [9,[46][47][48]. The MDAS was designed specifically to rate delirium severity [65], it has face validity and uptake, further formal validation studies for its use as a delirium screening tool would build on the existing psychometric data and help to reinforce the attributes of the tools [64]. One perspective looks at the balance between the positive features of usability of the MDAS compared with some other tools, and the effect of the breakdown and operationalisation of delirium features within the MDAS which does not support the syndromic nature of delirium diagnosis in terms of coexisting core features. Although it identifies delirium symptoms, regardless of the cut-off score specified to identify delirium, the MDAS risks false positive results, as patients with delirium symptoms who do not fit the core diagnostic criteria for syndromic delirium as characterised by coexistent core features may be labelled case positive. Several studies in this review used the MDAS alone for case ascertainment, which may bias reported detection rates [48,64].
Clinical operationalisation appeared to be the major driver of choice of delirium screening tools. Delirium diagnosis is complex, multidimensional and not intuitive for bedside staff. DSM 5 criteria require 5 characteristics and so while screening tools may gain in usability through operationalisation they lose precise application of the necessarily coexistent core features that define delirium. In the research setting, we found clear demarcation between the index tool and the chosen reference standard was not always evident. A blurring of the distinction between screening tools and diagnostic reference standards used for case confirmation for validation purposes was found. More specifically where references standards were other than DSM or ICD based reporting of reference-rater training was at times lacking. The importance of tool selection to fit the intended purpose is an important finding of our review.
In APCUs delirium rates were higher than in oncology inpatients but given methodological constraints in studies within this setting, results may not be representative. The use of the MDAS may have contributed to inflated delirium rates reflecting the way the tool is operationalised. In the older cancer patient cohorts, differences in delirium incidence and prevalence might be accounted for by study heterogeneity and patient recruitment. This is an important issue for future work, as establishing delirium incidence and prevalence in inpatient oncology settings is an important step in management. Better understanding of how to use available tools will improve management and inform education initiatives in this setting.
Criteria for delirium reversal were inadequately defined in studies, making it difficult to compare delirium reversibility across studies. These data may be further constrained by retrospective methodology, the absence of a diagnostic reference standard, or study flow reliant on clinical documentation. Ascertainment of delirium reversibility requires prospective, longitudinal study design, use of a robust diagnostic standard and explicit definition of delirium reversal. Assessment of delirium reversibility is an important issue for consideration in the design of future studies.
Patient selection, choice of the delirium screening tool and the choice of the diagnostic reference standard, were all identified as a source of bias on QUADAS-2 criteria [35]. Recruitment flow was also an important consideration. For example, patient selection methods at times risked exclusion of potentially delirious patients due to retrospective design, convenience sampling, and ascertainment bias.
Adherence to consensus recommendations for reporting patient characteristics and wherever possible the use of assessment tools and delirium reference standards will improve epidemiological studies of delirium in this setting [36,72,73] .
Limitations to our review include those related to the methodology of the original studies as well as a limitation to the English language. The search was limited to publications between 1996 and 2017. The discussion has aimed to identify recent updates in the area, again these are largely limited to aged care or stand-alone settings, with one systematic review of delirium in palliative care fining an incidence of 9-57% across hospital palliative care consultative services, with a majority of patients having cancer diagnoses [20]. a further systematic review, again in the palliative care setting, identified 14 delirium detection tools and heterogeneity of methods [23], Important questions for future work include which tools translate 1 3 well to inpatient oncology from aged care and stand-alone inpatient palliative care settings, which tools are most suitable for patients, carers and staff, and which reference standards are most appropriate. Requirements for clinical and research uses of detection tools will differ according to purpose, however establishing methodical approaches to the detection of delirium in either setting is a prerequisite to determining the incidence, prevalence and reversibility of delirium for oncology inpatients. Maintaining a clear accountability for the validation and purpose of the tool, and its psychometric characteristics when applying it to clinical screening/detection is critical as is the requirement in research uses to select a reference standard with established reference-rater methodology, is extremely important.
Choosing a tool for delirium detection in the clinical oncology setting will vary according to operational issues such as staff training and preference, however, it is important that tools are fit for purpose, and where possible, have been validated in the same clinical setting. While patient profiles may be similar across palliative care, aged care and some oncology inpatient settings, staff competencies will be more specifically related to setting. Delirium detection and diagnosis must be a core competency for clinical teams in acute settings, however, operational characteristics may render a tool selection may vary according to operational setting, the exact tool chosen is not as important as the review of characteristics that makes it fit for purpose/setting. Our review, found gaps in the validation of tools in for use in oncology inpatients. At present extrapolation from findings in other acute hospital settings, such as aged care, may help support a more robust selection for this population for the time being. As further validation occurs in acute oncology settings the evidence base for selection tools to detect the presence or resolution of delirium in this clinical setting should improve.
The knowledge gaps identified to generate new hypotheses for future investigation. We recommend the optimal description of patient characteristics, selection of delirium detection tools appropriate to the setting, use of reproducible methods of patient selection and diagnostic assignment using a reference standard with appropriate reference rater methodology. Our results indicate that a determination of the incidence, prevalence, and reversibility of delirium in the inpatient cancer population is both lacking and overdue. Addressing these knowledge gaps will help to provide a more robust evidence base to inform ongoing efforts for effective prevention, detection and management of delirium in the inpatient oncology setting.