Delirium is characterized by an acute onset and fluctuations in cognition, attention and consciousness (ICD-10, DSM V). The term “delirium” is defined in ICD-10 (F05.0) and ICD-11 (6D70 Delirium Foundation¹) and characterized by a state of disturbed attention and awareness (i.e., reduced orientation to the environment) that develops over a short period of time and tends to fluctuate (WHO, 2021). In literature, the term “acute encephalopathy” is often used instead of delirium, focusing on pathophysiological aspects of brain dysfunction (Schieveld et al., 2019; Slooter and Stevens, 2020). Delirium is often an incident in medical emergencies (Inouye, 2006), such as sepsis or stroke (Wilson et al., 2020) but also becomes an incident of postoperative delirium (POD), especially in older adults after hip fracture (Kim et al., 2020) or cardiac surgery (Aldecoa et al., 2017). The incidence of POD depends on several predisposing risk factors that are not modifiable, such as age, multimorbidity, stroke, dementing disorders caused by neurodegeneration or other causes, and partially modifiable factors, such as mild cognitive impairment, impaired sensory function, high anxiety levels, and polypharmacy (Guenther et al., 2016) and/or frailty (Leung et al., 2011). Further precipitating factors include the type of surgery, type and duration of anesthesia, and usage of cardiopulmonary bypass (CPB) in heart surgery.

The number of elective surgeries in older adults has dramatically increased within the last years. More than 150,000 knee and hip replacement and 415,000 heart surgery procedures [Operation and Procedure Classification System (OPS) Codes 5–35 to 5–37] have been documented per year in Germany, most of which were performed in patients who are over 70 years of age. On average, hip replacement rates increased by 30% between 2007 and 2017². Especially, older patients with cognitive impairment are overrepresented in German hospitals and are at a high risk for further cognitive decline (Bickel et al., 2018; Nationale Demenz Strategie, 2020, RKI³).

A better understanding of the individual risk for the incidence of POD is important, as its occurrence, severity, and duration have been shown to be significantly decreased by modifying risk factors and proper disease management (RR 0.69, 95% CI 0.59–0.81) (Siddiqi et al., 2016; Hughes et al., 2020). Most studies, however, are done in groups with heterogenic surgery procedures without elaborated preoperative cognitive assessments (Aldecoa et al., 2017; Wilson et al., 2020). Preoperative cognitive state is a key risk factor for delirium (Oh et al., 2015) but cannot be estimated objectively in medical or surgical emergencies. Fever, pain, anxiety, blood loss, sedation, and/or lack of time prevent a valid assessment of cognitive domains, such as attention, verbal and non-verbal memory, visuospatial ability, reasoning, and language performance. In contrast to emergency surgery, elective surgery allows preoperative cognitive and further assessments. Objective memory impairments in older adults (Wolfsgruber et al., 2014; Hagen et al., 2015) are not only relevant for the risk of developing dementia but also for developing POD (Aldecoa et al., 2017; Culley et al., 2017; Wilson et al., 2020). Preliminary support is provided for an interaction of diatheses (vulnerabilities) and intra-operative stressors on POD (El-Gabalawy et al., 2017). POD is not only an indicator of brain vulnerability but might also moderate and/or accelerate cognitive decline, especially in patients suffering from dementia (Fong et al., 2015a).

Physical vulnerability (e.g., clinical frailty) is an additional risk factor for POD in cardiac surgery, as recently shown (Itagaki et al., 2020): only older patients who were frail and cognitively impaired showed a significantly increased risk of POD. In elective surgery, most surgeons and anesthesiologists do not assess cognition by screening, although European and American guidelines recommend it at Grade A, which is, however, based on studies with mixed design quality (Aldecoa et al., 2017).

PAWEL-R (“Patientensicherheit, Wirtschaftlichkeit und Lebensqualität” [transl.: “Patient security, economy and life quality”], R for “Risikoschätzung” [transl.: “Risk estimation”]) addresses risk profiles of POD as an add-on study of the PAWEL trial. The PAWEL study focused on the prevention and management of POD (Sanchez et al., 2019) in orthopedic, general, and cardiovascular surgery. The first aim of PAWEL-R was to estimate a more general delirium risk model for elective procedures in cardiac but also for general and orthopedic surgery, such as stepwise cognitive and geriatric assessments. The second aim of PAWEL-R is the development of a pragmatic clinical delirium risk score applicable to patients who are over 70 years of age undergoing elective surgery [PAWEL and the sub-study PAWEL-R are funded by the German Innovation Funds (Innovationsfonds; AZ: VF1_2016-201)].

PAWEL-R investigated clinical, psychosocial, and biological risk factors in 880 patients who are over 70 years of age at five surgical centers in southwest Germany: University Hospital Tübingen, Medical Center Stuttgart (Klinikum Stuttgart), University Hospital Ulm, University Hospital Freiburg im Breisgau, and the Medical Centers Karlsruhe (HELIOS Klinik, ViDiaKliniken). Patients were included if they were scheduled for elective surgery to heart, vessels, abdomen, big joints (hip, knee, shoulder), or spine with an estimated duration of more than 60 min (cut-to-suture time). The estimated clinical survival time was set to be at least 15 months, such that, e.g., surgery of advanced cancer was excluded. Patients who did not speak sufficient German were excluded. The date of first enrollment in the control arm was July 11th, 2017, and the last enrollment was on January 15th, 2019.

PAWEL-R was able to estimate stepped delirium risk models for elderly patients undergoing elective surgery in five medical centers. By logistic regression analyses in the whole groups of the participants (lumping various surgical procedures), we determined the following independent main risk factors: high ASA scores, longer expected cut-to-suture time, and preoperative cognitive impairment. Age per se, excessive alcohol consumption, active smoking, and visual or hearing impairments were not predictive in the sample.

In the standard clinical risk prediction (Table 4, model CLIN), about 24% of variance (AUC 0.74) were explained in multivariate logistic regression analysis by clinical and perioperative factors. Adding cognitive screening assessments improved the explained variance to about 30% (models CLIN-COG and CLIN-COG-F), while the inclusion of a geriatric assessment (dichotomized clinical frailty index) only increased it to 25.6% (model CLIN-F).

In the Successful Aging after Elective Surgery (SAGES) cohort (Racine et al., 2020), a stepwise logistic regression and machine learning algorithms were used to predict POD in elective surgery of 560 older adults. With a limited data set of 18 items in different machine learning methods, they showed AUCs from 0.53 to 0.57, while in another analysis, adding cognitive performance using the modified Mini-Mental-State 3MS (Teng and Chui, 1987) showed intermediate AUCs (0.53–0.68). With the full data of 71 items, the AUCs in this study were 0.62 up to 0.71. Thus, all prediction models in SAGES were numerically less accurate (Racine et al., 2020) than the shorter models.

The improved predictive accuracy in PAWEL-R might be due to a higher POD incidence in the included cardiovascular surgery group and the usage of the MoCA and Digit Span backward—instead of the MMS or 3MS. The finding of a doubling of delirium risk in the presence of cognitive impairment as evinced by the MoCA scores is comparable with other studies (Racine et al., 2018) and illustrates the importance of comprehensive preoperative cognitive screening. The suggestion of using a lower cut-off score of <24 points could be validated in the CHAID analysis, which showed a risk split at this cut-off.

The phenotype of presurgical cognitive impairment might be important. There is an ongoing debate whether the MoCA (Nasreddine et al., 2005) can be divided into its seven subscores to discriminate different cognitive domains like memory, language, attention, and executive functions. A Memory Index Score (MoCA-MIS) (Julayanont et al., 2014) has been calculated, but the memory index depends on additional questions and scoring from the delayed recall of the five items. As these additional data from delayed recall are not available in the PAWEL sample, cognitive subdomains were not included in the models. All MoCA subscores but naming, however, showed a significant difference between non-delirious and delirious patients. The differences were most pronounced for the memory domain, followed by language and orientation. As the phenotyping of MCI into subtypes like amnestic and non-amnestic MCI is predictive for cognitive decline in dementia (Petersen et al., 2001), we will further analyze MOCA subdomains such as TMT and Digit Span and will differentiate the impact of memory, attention, language, and executive domains on POD risk and potential postoperative cognitive dysfunction (POCD) (Fong et al., 2015b).

Most studies on POD prediction in older adults include multimorbidity or the CCI based on 19 disorders. The acquisition and analysis of the personal medical history is, however, time-consuming. The combination of multimorbidity indices with a frailty score, the easy and short assessment by the seven intuitive levels of the CFS Score (Jones et al., 2005) increased the delirium prediction substantially. In this study, the CFS with a cut-off at 3 vs. 4 (mild frailty in need of a walking aid) added significantly to overall predictive performance for POD.

Regarding different types of surgery, differences between cardiovascular surgery (mostly with CPB) and orthopedic and general surgery were only subtle: The models in orthopedic-general surgery explained less variance and exhibited lower predictive accuracy, possibly because of the lower overall POD incidence in orthopedic and general surgery with overall shorter delirium duration (see Supplementary Material). However, in both groups, the prediction models significantly benefitted from adding cognitive assessments such as MoCA or Digit Span backward. As POD is often an acute reversible brain disorder, brain vulnerability is a central issue and may reflect the finding of a preoperative cognitive impairment as highly predictive of POD. In addition, the results underline the impact of the use of short clinical frailty indices and cognitive impairment as independent risk factors (for cardiovascular surgery, compare Jung et al., 2015).

Overall, cardiovascular surgery is associated with a significantly higher delirium risk than other elective surgeries. Cardiovascular surgery has several procedural implications. In the sample, 87% of the patients received a CPB, which is associated with hypoperfusion of the brain, a higher incidence of delirium (Danielson et al., 2018; Bernardi et al., 2019), and microembolic showers to the cerebral circulation. In the multivariate and tree analysis, however, it was suggested that not the usage of CPB per se but the longer duration of cardio-vascular surgery as compared with orthopedic or general surgery, together with the high incidence of severe heart failure, may be the main factor for the higher POD incidence in cardiovascular surgery.

The POD incidence in PAWEL-R was comparable with recent meta-analyses: in vascular surgery, the POD incidence was 23.4% ranging from 4 to 39% (Oldroyd et al., 2017). In elective gastrointestinal surgery, the incidence of POD was between 8 and 55% (Scholz et al., 2016). In the latter analysis, an increased ASA score of III or more was a significant risk factor for POD, while pre-surgical cognitive assessments were not available. The recently reported prevalence of 34% for POD (Itagaki et al., 2020) after cardiac surgery could be confirmed by PAWEL-R. At the same time, older adults who were frail and cognitively impaired showed a significantly increased risk of POD. Another study with 220 cardiac surgery patients and a POD incidence of 52% (Rudolph et al., 2009) identified four independent risk factors for delirium: MMSE impairment, prior stroke or transient ischemic attack, depression, and abnormal albumin. Delirium risk increased more than four-fold from the lowest to highest risk levels. In this study, however, despite the fact that cranial computed tomography (CCT) or magnetic resonance imaging (MRI) were not available, predictive accuracy (with an overall AUC of 0.78) was comparable with studies such as neuroimaging. We did not include further comprehensive clinical assessments and biomarkers in the scope of this study.

This risk study initially addressed 18 risk factors. It is not feasible in routine clinical practice to perform such comprehensive risk assessments. Several assessments are quite time consuming and not easily obtained. For example, the TMT-A and TMT-B proved to be time consuming and were obtained from <87% of the sample (Table 1). Likewise, assessments of hyposmia, which was predictive for POD in cardiac surgery (Brown et al., 2015), could just be obtained in <90% of the patients due to temporal constraints and self-reported difficulties to smell. Thus, while the assessments of hyposmia (OR of 1.47) and TMT (TMT-A univariate OR 2.7; TMT-B univariate OR 1.87) might also have independent predictive power, they seem less practicable in routine care.

Precipitating and perioperative risk factors are important to predict POD. The well-known ASA classification and the expected cut-to-suture time have such a predictive power in older adults. Impaired cognitive performance is a major independent risk factor. The MoCA test is such a widely known cognitive assessment and proved to be predictive of delirium, but in itself is a relatively time-consuming tool with a duration of at least 10 min (Nasreddine, 2020). In addition, physical and medical history perioperative factors operationalized in a clinical frailty score together with the type of surgery and its duration exhibited high predictive accuracy. Using <10 risk factors (especially ASA, multimorbidity, expected cut-to-suture time, MoCA, Digit Span backward) we were able to show significant predictive accuracy for delirium in a CHAID analysis and in Model CLIN-COG with an AUC of 0.8. We recommend efficient cognitive items for further evaluation and use in clinical practice.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Ethics Commission of the Faculty of Medicine of the Eberhard-Karls University and University Hospital Tübingen with number 233/2017BO1 on October 12, 2017 and by the Ethics Commission of the University of Potsdam with number 38/2017 on December 11, 2017. The study was registered on the German Clinical Trials Register (DRKS-ID: DRKS00012797) in July, 2017. The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

GE, CT, and MR designed the study, planned the data collection, performed the statistical analysis, and prepared the manuscript. MC and YK contributed to the data collection and performed the statistical analysis. CA, BM, and MD were involved in designing the study, data collection, and data interpretation. MH, OF, OK, LC, CM, FK, FD, AS, SW, and EM were involved in the data collection and critically revised the manuscript for the final approval of the version to be published. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.