Development of a Prediction Model for Delirium in Hospitalized Patients with Advanced Cancer

Article information

Cancer Res Treat. 2024;56(4):1277-1287
Publication date (electronic) : 2024 February 26
doi : https://doi.org/10.4143/crt.2023.1243
1Division of Hematology and Medical Oncology, Department of Internal Medicine, Seoul National University Bundang Hospital, Seoul National University College of Medicine, Seongnam, Korea
2Center for Palliative Care and Clinical Ethics, Seoul National University Hospital, Seoul, Korea
3Palliative Care Center, Yonsei Cancer Center, Yonsei University Health System, Seoul, Korea
4Division of Medical Oncology, Yonsei Cancer Center, Yonsei University Health System, Seoul, Korea
5Yonsei Graduate School, Yonsei University College of Medicine, Seoul, Korea
6Division of Medical Oncology, Department of Internal Medicine, CHA Bundang Medical Center, CHA University School of Medicine, Seongnam, Korea
Correspondence: Beodeul Kang, Division of Medical Oncology, Department of Internal Medicine, CHA Bundang Medical Center, CHA University School of Medicine, 59 Yatap-ro, Bundang-gu, Seongnam 13496, Korea Tel: 82-31-780-3438 Fax: 82-31-780-3929 E-mail: wb0707@cha.ac.kr
Co-correspondence: Si Won Lee, Palliative Care Center, Yonsei Cancer Center, Yonsei University Health System, 50-1 Yonsei-ro, Seodaemun-gu, Seoul 03722, Korea Tel: 82-2-2228-4595 Fax: 82-2-2227-8084 E-mail: LEESIWON925@yuhs.ac
*Eun Hee Jung and Shin Hye Yoo contributed equally to this work.
Received 2023 November 21; Accepted 2024 February 23.

Abstract

Purpose

Delirium is a common neurocognitive disorder in patients with advanced cancer and is associated with poor clinical outcomes. As a potentially reversible phenomenon, early recognition of delirium by identifying the risk factors demands attention. We aimed to develop a model to predict the occurrence of delirium in hospitalized patients with advanced cancer.

Materials and Methods

This retrospective study included patients with advanced cancer admitted to the oncology ward of four tertiary cancer centers in Korea for supportive cares and excluded those discharged due to death. The primary endpoint was occurrence of delirium. Sociodemographic characteristics, clinical characteristics, laboratory findings, and concomitant medication were investigated for associating variables. The predictive model developed using multivariate logistic regression was internally validated by bootstrapping.

Results

From January 2019 to December 2020, 2,152 patients were enrolled. The median age of patients was 64 years, and 58.4% were male. A total of 127 patients (5.9%) developed delirium during hospitalization. In multivariate logistic regression, age, body mass index, hearing impairment, previous delirium history, length of hospitalization, chemotherapy during hospitalization, blood urea nitrogen and calcium levels, and concomitant antidepressant use were significantly associated with the occurrence of delirium. The predictive model combining all four categorized variables showed the best performance among the developed models (area under the curve 0.831, sensitivity 80.3%, and specificity 72.0%). The calibration plot showed optimal agreement between predicted and actual probabilities through internal validation of the final model.

Conclusion

We proposed a successful predictive model for the risk of delirium in hospitalized patients with advanced cancer.

Introduction

Delirium is a serious neurocognitive illness characterized by abrupt, fluctuating changes in mental status and consciousness, as well as cognitive functions, which is frequently observed in patients with advanced cancer [1]. Reported prevalence and incidence rates for delirium in this population from Sands et al. [2] range from 18 to 58% and 3.5 to 45%, respectively. Watt et al. [3] reported the median cumulative incidence and the median period prevalence of delirium in hospitalized patients was 29% and 60%, respectively. The presence of delirium in palliative setting can degrade the quality of life of patients with advanced cancer and their caregivers and make it difficult to evaluate symptoms, such as pain or dyspnea [4,5]. Moreover, delirium is associated with longer hospital stays, higher mortality and morbidity rates, and higher healthcare expenses [6]. Given the significant consequence of delirium, it is essential to predict the risk of developing delirium in advance and undertake steps to reduce it [7,8].

The etiology of delirium is a multi-factorial combination of predisposing factors, such as age or previous cognitive dysfunction and precipitating factors, such as metabolic derangements [9]. Among the potential etiologies, targeting the modifiable or reversible risk factors can be a good strategy for preventing delirium in patients with advanced cancer [7,9]. Several cohort studies of patients with advanced cancer in a palliative setting found that persistent symptoms, such as pain, drowsiness, use of medications, such as opioids and steroids, metabolic abnormalities, or dehydration were associated with an increased risk of delirium [4,10,11]. However, few studies have built a comprehensive prediction model incorporating multiple components, including predisposing and precipitating factors. Furthermore, apart from predictors of delirium in hospice or end-of-life cases, few prediction models targeted patients admitted to the general oncology ward for supportive cares. As it is often difficult to manage terminal delirium in a hospice setting, it is of value to predict delirium occurring in a relatively earlier setting [12].

Therefore, the current multicenter cohort study was performed to identify risk factors related to the occurrence of delirium and to propose a comprehensive prediction model in patients with advanced cancer who were hospitalized in oncology wards of tertiary cancer centers for supportive cares.

Materials and Methods

1. Study design and participants

This was a retrospective cohort study for hospitalized patients with advanced cancer at four tertiary cancer centers (Seoul National University Bundang Hospital, Yonsei Cancer Center, CHA University Bundang Hospital, and Seoul National University Hospital) in South Korea between January 2019 and December 2020. Among patients with advanced cancer hospitalized in oncology wards of the cancer centers in need of inpatient supportive care, patients older than 19 years were enrolled. We defined advanced cancer as a metastatic, or recurrent disease for solid tumors, relapsed/refractory disease for hematologic malignancies, and not treatable with curative intent. The exclusion criteria were as follows: (1) patients who were hard to communicate with because of active brain metastasis or leptomeningeal seeding, (2) patients who had previously been diagnosed with any neurocognitive disorder according to the code of the 10th revision of the International Classification of Diseases (ICD-10), such as dementia, Parkinson’s disease, Alzheimer’s disease other than delirium, (3) patients who were admitted due to active psychiatric symptoms resulting from underlying psychiatric disorder and (4) patients who discharged due to death. The last item of exclusion criteria was to exclude patients with terminal delirium. We defined terminal delirium as that accompanied by death within 15 days after the onset of delirium during inpatient care.

2. Study variables and data collection

Clinical and sociodemographic data were obtained from electrical medical records. Our dataset comprised general sociodemographic information (age, sex, health insurance, education level, use of glasses or hearing aids, living with family, smoking and alcohol consumption), clinical risk factors (vital signs and body mass index [BMI] at admission, laboratory findings, obesity, duration of admission, and data on chemotherapy during admission), past medical history, and concomitant drugs. Obesity was assessed in conformity with BMI, and a BMI of less than 23 kg/m2 was defined as non-obese. The laboratory findings within three days of admission were referred for analysis.

Regarding reliability, the occurrence of the delirium was measured by the diagnostic criteria used during the routine clinical practice. Patients were included if they were diagnosed with delirium according to ICD-10 criteria, referred for psychiatric consultation due to delirium, and administered medication to control delirium during their hospital stay. Due to the retrospective nature of the study, research team including a well-trained physician and academic nurses reviewed the medical records and confirmed the occurrence of the delirium.

3. Statistical analysis

The baseline characteristics between patients with and without delirium were compared using the student’s t test or Wilcoxon’s sign rank test for continuous variables and Pearson’s chi-square test or Fisher’s exact test for categorical variables. The variables comprised the following four categories: sociodemographic data, clinical characteristics, laboratory findings, and concomitant medication. We used mean imputation techniques to handle missing data for variables with missing values of less than 10%. Variables of four categories that were associated with the occurrence of delirium were evaluated by univariate logistic regression. Variables with more than 10% missing data (education, prothrombin time, C-reactive protein) were excluded from the univariate analysis. A stepwise backward procedure was used to identify the predictors of the occurrence of delirium. All predictors achieving a p-value below 0.05 were considered and sequentially removed if the p-value in the multivariate model was above 0.05. As no significant collinearity was observed, the variables that were statistically significant in the univariate analysis among the whole classified variables were entered into the multivariate logistic regression analysis. Using multivariate logistic regression, we confirmed each predictive model by consecutively adding the variables of the four categories.

A combination of sociodemographic data and clinical characteristics was set as the basic prediction model. Additional models were developed by sequentially adding laboratory findings and concomitant medication to the basic model, and the accuracy and discriminatory power of each prediction model were identified. We evaluated the performance of the predictive model in terms of discrimination and calibration. The discriminatory ability for each prediction model was measured by a receiver operating characteristic (ROC) curve to calculate the corresponding areas under the curve value. Values above 0.7 were suitable for discrimination [13]. Calibration for the final developed model, including significant variables in four categories, was assessed graphically by plotting to evaluate the agreement between the predicted probability and actual outcome. The final developed prediction model using multivariate logistic regression was internally validated by bootstrapping (1,000 repetitions).

A two-tailed p-value <0.05 was considered statistically significant, and confidence intervals (CIs) were calculated at a 95% confidence level. All data were analyzed using Statistical Package for the Social Sciences software ver. 22.0 (IBM SPSS Statistics, Armonk, NY) and R statistics ver. 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria).

Results

1. Patient characteristics between delirium and non-delirium groups

From January 2019 to December 2020, 2,152 patients were enrolled. A total of 127 patients (5.9%) developed delirium during hospitalization. The median age of patients with delirium was 69 years (range, 27 to 88 years). The delirium group had more male patients (70.1% vs. 57.7%, p=0.007), had more patients with hearing impairment (3.1% vs. 0.9%, p=0.037), had a longer length of hospital stay (13 days vs. 8 days, p < 0.001), and had fewer patients who underwent chemotherapy during hospitalization (15.7% vs. 28.8%, p=0.001) compared to the non-delirium group.

Regarding past medical history, the delirium group had a higher proportion of history of delirium (18.9% vs. 1.6%, p < 0.001), psychiatric disease (13.4% vs. 5.7%, p < 0.001), diabetes mellitus (30.7% vs. 21.6%, p=0.020), and cardiovascular disease (48.8% vs. 38.1%, p=0.019) compared to the non-delirium group. Concerning laboratory findings at admission, azotemia and electrolyte imbalances, including hyponatremia, hyperkalemia, and hypercalcemia, were more common in the delirium group than in the non-delirium group. In the delirium group, concomitant use of opioids (76.4% vs. 62.3%, p=0.002) and antibiotics (77.2% vs. 59.1%, p < 0.001) was statistically significantly higher than in the non-delirium group during hospitalization. Furthermore, the use of sedatives and antidepressants was more frequent in the delirium group (37.0% vs. 15.6%, p < 0.001) than in the non-delirium group. Baseline patient characteristics are summarized in Table 1.

Baseline characteristics of patients

2. Factors associated with the occurrence of delirium

By univariate analysis, among sociodemographic variables, age over 65, male sex, hearing impairment, smoking status, and non-obese range of BMI were associated with a higher risk of developing delirium. Among clinical variables, previous history of delirium, cardiovascular disease, diabetes mellitus, psychiatric diseases, length of hospital stay, or chemotherapy during hospitalization were significantly associated factors. Regarding laboratory findings, leukocytosis, elevated alkaline phosphatase, azotemia, and electrolyte imbalance were associated with delirium occurrence. The use of opioids, sedatives, antidepressants, and antibiotics during hospitalization were significant factors with delirium risk.

Multivariate analysis for variables among all four categories—sociodemographic features, clinical characteristics, laboratory findings, and medication—revealed that the occurrence of delirium significantly increased in patients whose age was over 65 years (odds ratio [OR], 1.86; 95% CI, 1.22 to 2.87; p=0.004), were non-obese (OR, 1.70; 95% CI, 1.07 to 2.80; p=0.029), had hearing impairment (OR, 3.38; 95% CI, 0.89 to 10.34; p=0.046), had a previous history of delirium (OR, 11.90; 95% CI, 6.28 to 22.42; p < 0.001), were hospitalized for longer than 8 days (OR, 3.24; 95% CI, 2.14 to 5.01; p < 0.001), who did not receive chemotherapy during hospitalization (OR, 1.77; 95% CI, 1.07 to 3.07; p=0.032), had elevated blood urea nitrogen (BUN) level (OR, 1.58; 95% CI, 1.06 to 2.36; p=0.024), had hypercalcemia (OR, 2.15; 95% CI, 0.98 to 4.33; p=0.042), and used sedatives (OR, 3.15; 95% CI 2.04 to 4.83; p < 0.001) (Table 2). The results of multivariate analysis on the risk of delirium according to the combination of each categorized variable were additionally presented as supplementary data (S1-S3 Tables).

Risk factor for occurrence of delirium on univariate analysis and multivariate analysis for all variables

3. Development of the prediction model for delirium and prediction ability

Table 3 lists the validity and accuracy statistics of the prediction model’s performance. The area under the ROC (AUROC) was approximately 0.8 in all four prediction models (Fig. 1A-D), and negative predictive values were greater than the positive predictive values in the overall prediction model. The highest AUROC of the predictive model was 0.831 (95% CI, 0.794 to 0.868), observed for the combination of all four classification variables, indicating good discriminant ability. Therefore, we designated the combination of all four categorized variables (sociodemographic data, clinical characteristics, laboratory findings, and concomitant medications) as the final prediction model. The sensitivity and specificity of the final prediction model were 80.3% and 72.0%, respectively. We performed internal validation of the final model using the bootstrap with 1,000 repetitions. The calibration plot of the internal validation set using bootstrapping resampling illustrated good accordance between the predicted and actual probabilities (Fig. 2).

Performance comparison of predictive models according to sociodemographic, clinical, laboratory and concomitant medication characteristics

Fig. 1.

Receiver operating characteristic curves of predictive models according to variables. (A) Sociodemographic+clinical characteristics. (B) Sociodemographic+clinical+laboratory characteristics. (C) Sociodemographic+clinical+concomitant medication characteristics. (D) Sociodemographic+clinical+laboratory+concomitant medication characteristics. AUROC, area under the receiver operating characteristic curve; CI, confidence interval.

Fig. 2.

Calibration plot for predictive model of delirium occurrence using bootstrapping method.

Discussion

This multicenter, retrospective cohort study developed and internally validated a delirium prediction model for patients with advanced cancer who were hospitalized in oncology wards in need of inpatient supportive care using predictors that can be readily measured. The prediction model, which consisted of four predisposing factors (old age, non-obesity, hearing difficulty, and previous history of delirium) and five precipitating factors (longer length of hospital stay, non-receipt of chemotherapy at this admission, uremia, hypercalcemia, and using sedatives), showed good performance and good discrimination capability. This model may contribute to timely and proper management of delirium by analyzing the risk factors and early detection of delirium, as compared to predicting irreversible terminal delirium.

Age [14], hearing impairment [10,14], and past delirium history [15] have been well identified as delirium risk factors in previous research of patients with advanced cancer, and were eventually incorporated in our study’s predictive model. However, our findings regarding the negative association between obesity and delirium risk are quite intriguing. Though obesity is a well-known risk factor for chronic cognitive impairment [16], mixed results have been reported about the association of BMI with delirium [17,18]. Existing research involving postoperative delirium [19] or intensive care unit delirium [20,21] as an outcome variable found that having a high BMI was a protective factor for delirium. “Obesity paradox,” which is a controversial issue, may support those results [22]. To our knowledge, as the association of BMI with delirium has not been well established in patients with advanced cancer, our findings, which reveal that non-obese patients were 1.7 times more likely than obese patients to develop delirium, are the first to show that such an obesity paradox may be applied to delirium in patients with advanced cancer [22,23]. Further research is required to elucidate and validate the correlation between delirium and BMI in patients with advanced cancer and the associated mechanisms.

Our findings suggesting a hospital stay of more than 8 days increases the risk of delirium are consistent with the findings of previous studies that indicated a positive relationship between hospital stay length and delirium risk [24,25]. Their relationship has not been fully clarified so far. When compared to those who do not have delirium, patients with delirium may have a longer hospital stay as a result of their poor outcome [24]. The length of hospital stay, however, may be a risk factor for delirium because the hospital environment itself may be a risk of delirium.

The results of our study, which indicated that patients who did not receive chemotherapy during hospitalization had a higher risk of delirium, reflect the circumstance of patients who were unable to receive chemotherapy rather than the effect of chemotherapy itself on the risk of delirium [14]. It has been reported that certain anticancer drugs that pass through the brain blood barrier can increase the risk of hyperactive delirium [26]; however, our study did not assess the influence of each patient’s specific anticancer drug on the risk of delirium. As this study focuses on patients with advanced cancer admitted to oncology wards of cancer centers, it is speculated that the risk of delirium may increase if chemotherapy is challenging because of a specific reason (e.g., deteriorating systemic conditions, low benefits when comparing gains and losses). However, the lack of investigation into the reason why the patients did not receive chemotherapy in our study makes it difficult to interpret the results. This aspect requires further research.

High BUN and hypercalcemia, which were incorporated in our final model, were previously referred to as acute renal failure, dehydration, and metabolic derangement, and were thought to present cognitive impairment as a common symptom. While previous research did not differentiate between specific laboratory parameters, instead of focusing on the overall impact of metabolic abnormalities [10,14,26], this study demonstrated that high BUN and hypercalcemia were significant in multivariate models after adjusting for several variables that were significant in the univariate analysis. As BUN and calcium can be easily measured during admission, using them to predict delirium can assist in identifying high-risk patients.

Our study’s strength is that it includes medication use, a risk factor for delirium that is frequently ignored in its predictive models. According to previous research [3,27], opioids are a well-known risk factor for delirium in patients with advanced cancer. However, in our study, sedatives were identified as predictors included in the final model, while opioids were not included as significant predictors after adjusting for multiple variables. The use of sedatives, such as benzodiazepines could be related to the mechanism of delirium development via overstimulation of the cortical gamma-aminobutyric acid system [28]. Our findings are consistent with those of a previous report [15] that hospitalized cancer patients exposed to benzodiazepines in excess of 2 mg per day are twice as likely to develop delirium as those who have not, as well as previous studies that used postoperative delirium as an outcome variable [29,30]. These findings suggest that more attention should be paid to the development of delirium in patients with advanced cancer who are administered sedatives, and that efforts should be made to limit the use of sedatives when they are not necessary.

In this study, some limitations must be addressed. First, this study contained only the training set without a validation cohort because of the limited number of participants though we internally validated the models through cross-validation with bootstrapped 95% CI. The validation step is needed using further cohort (internal) or independent data (external). Second, the incidence of delirium in this study was relatively lower (5.9%) than that reported in previous studies [2,3]. This low incidence rate may be due to the use of medical records to identify delirium, which would have missed cases. We also excluded patients who have neurocognitive disorders, which could be a risk factor for delirium. This may limit the generalizability of the data. Third, though we have attempted to contain several factors that can be potentially related to the occurrence of delirium, some relevant factors, such as the use of restraints or performance status of the patient might be missing [7]. Fourth, the retrospective design of this study may underestimate the occurrence of the hypoactive subtype of delirium. Not all patients wore the aid device for visual and hearing impairments, but impairments of visual and hearing were evaluated based on wearing devices due to the retrospective nature of the review of electrical medical records. Moreover, patient-reported outcomes, such as symptoms or quality of life could also not be fully obtained in this study. Development of a prediction model for delirium using further prospective study would help resolve these limitations.

Despite these limitations, exploring the predictive factors associated with the occurrence of delirium in patients with advanced cancer is warranted. The strength of our model is that we selected variables that are easily quantified among comprehensive characteristics in patient groups drawn from a large multicenter cohort. We also believe that given the predicted results of the model involve delirium that occurred before terminal delirium, this model will help us predict clinical outcomes that are more preventable.

In conclusion, we developed a model to predict the occurrence of delirium in patients with advanced cancer who were hospitalized in oncology wards of the cancer centers using sociodemographic and clinical features, including laboratory abnormalities and use of concomitant medication. This model could help clinicians estimate the risk of hospitalized patients developing delirium and pay special attention to those at high risk. However, it needs external validation and a prospective cohort study before considering the application in real practice setting.

Electronic Supplementary Material

Supplementary materials are available at Cancer Research and Treatment website (https://www.e-crt.org).

Notes

Ethical Statement

This study was approved by the Institutional Review Boards of each participating study center and conducted according to the Declaration of Helsinki (CHA University Bundang Hospital, 2021-03-054-002; Seoul National University Hospital, H-2103-028-1201; Seoul National University Bundang Hospital, B-2104/681-405; and Yonsei Cancer Center, 4-2021-0323). As the study is of a retrospective nature using de-identified data, patients’ consent to participate was waived by the institutional review board.

Author Contributions

Conceived and designed the analysis: Jung EH, Yoo SH, Lee SW, Kang B, Kim YJ.

Collected the data: Jung EH, Yoo SH, Lee SW, Kang B, Kim YJ.

Contributed data or analysis tools: Jung EH, Yoo SH, Lee SW, Kang B, Kim YJ.

Performed the analysis: Jung EH, Yoo SH, Lee SW, Kang B, Kim YJ.

Wrote the paper: Jung EH, Yoo SH, Lee SW, Kang B, Kim YJ.

Conflicts of Interest

Conflict of interest relevant to this article was not reported.

Acknowledgements

This research was supported by a grant from the Ministry of Health and Welfare, Republic of Korea (grant number: HC20C0040). The funders had no role in the study design, data collection, data analysis, data interpretation, or writing of the manuscript.

References

1. Caraceni A, Simonetti F. Palliating delirium in patients with cancer. Lancet Oncol 2009;10:164–72.
2. Sands MB, Wee I, Agar M, Vardy JL. The detection of delirium in admitted oncology patients: a scoping review. Eur Geriatr Med 2022;13:33–51.
3. Watt CL, Momoli F, Ansari MT, Sikora L, Bush SH, Hosie A, et al. The incidence and prevalence of delirium across palliative care settings: a systematic review. Palliat Med 2019;33:865–77.
4. Mercadante S, Adile C, Ferrera P, Cortegiani A, Casuccio A. Symptom expression in patients with advanced cancer admitted to an acute supportive/palliative care unit with and without delirium. Oncologist 2019;24:e358–64.
5. Mah K, Rodin RA, Chan VW, Stevens BJ, Zimmermann C, Gagliese L. Health-care workers’ hudgments about pain in older palliative care patients with and without delirium. Am J Hosp Palliat Care 2017;34:958–65.
6. Seiler A, Blum D, Deuel JW, Hertler C, Schettle M, Zipser CM, et al. Delirium is associated with an increased morbidity and in-hospital mortality in cancer patients: results from a prospective cohort study. Palliat Support Care 2021;19:294–303.
7. Featherstone I, Sheldon T, Johnson M, Woodhouse R, Boland JW, Hosie A, et al. Risk factors for delirium in adult patients receiving specialist palliative care: a systematic review and meta-analysis. Palliat Med 2022;36:254–67.
8. Lawlor PG, Davis DH, Ansari M, Hosie A, Kanji S, Momoli F, et al. An analytical framework for delirium research in palliative care settings: integrated epidemiologic, clinician-researcher, and knowledge user perspectives. J Pain Symptom Manage 2014;48:159–75.
9. Bush SH, Lawlor PG, Ryan K, Centeno C, Lucchesi M, Kanji S, et al. Delirium in adult cancer patients: ESMO Clinical Practice Guidelines. Ann Oncol 2018;29:iv143–65.
10. Seiler A, Schubert M, Hertler C, Schettle M, Blum D, Guckenberger M, et al. Predisposing and precipitating risk factors for delirium in palliative care patients. Palliat Support Care 2020;18:437–46.
11. Hamano J, Mori M, Ozawa T, Sasaki J, Kawahara M, Nakamura A, et al. Comparison of the prevalence and associated factors of hyperactive delirium in advanced cancer patients between inpatient palliative care and palliative home care. Cancer Med 2021;10:1166–79.
12. Mercadante S, Adile C, Ferrera P, Cortegiani A, Casuccio A. Delirium assessed by Memorial Delirium Assessment Scale in advanced cancer patients admitted to an acute palliative/supportive care unit. Curr Med Res Opin 2017;33:1303–8.
13. Nam BH, D’Agostino RB. Discrimination index, the area under the ROC curve. In : Huber-Carol C, Balakrishnan N, Nikulin MS, Mesbah M, eds. Goodness-of-fit tests and model validity Boston, MA: Birkhäuser Boston; 2002. p. 267–79.
14. Mercadante S, Masedu F, Balzani I, De Giovanni D, Montanari L, Pittureri C, et al. Prevalence of delirium in advanced cancer patients in home care and hospice and outcomes after 1 week of palliative care. Support Care Cancer 2018;26:913–9.
15. Gaudreau JD, Gagnon P, Harel F, Roy MA, Tremblay A. Psychoactive medications and risk of delirium in hospitalized cancer patients. J Clin Oncol 2005;23:6712–8.
16. Balasubramanian P, Kiss T, Tarantini S, Nyul-Toth A, Ahire C, Yabluchanskiy A, et al. Obesity-induced cognitive impairment in older adults: a microvascular perspective. Am J Physiol Heart Circ Physiol 2021;320:H740–61.
17. Feinkohl I, Winterer G, Pischon T. Obesity and post-operative cognitive dysfunction: a systematic review and meta-analysis. Diabetes Metab Res Rev 2016;32:643–51.
18. Hung KC, Chiu CC, Hsu CW, Ho CN, Ko CC, Chen IW, et al. Association of preoperative prognostic nutritional index with risk of postoperative delirium: a systematic review and meta-analysis. Front Med (Lausanne) 2022;9:1017000.
19. Nakatani S, Ida M, Wang X, Naito Y, Kawaguchi M. Incidence and factors associated with postoperative delirium in patients undergoing transurethral resection of bladder tumor. JA Clin Rep 2022;8:6.
20. Wang ML, Kuo YT, Kuo LC, Liang HP, Cheng YW, Yeh YC, et al. Early prediction of delirium upon intensive care unit admission: model development, validation, and deployment. J Clin Anesth 2023;88:111121.
21. Ko Y, Kim HE, Park JY, Kim JJ, Cho J, Oh J. Relationship between body mass index and risk of delirium in an intensive care unit. Arch Gerontol Geriatr 2023;108:104921.
22. Lee DH, Giovannucci EL. The obesity paradox in cancer: epidemiologic insights and perspectives. Curr Nutr Rep 2019;8:175–81.
23. Park Y, Peterson LL, Colditz GA. The plausibility of obesity paradox in cancer-point. Cancer Res 2018;78:1898–903.
24. Stevens LE, de Moore GM, Simpson JM. Delirium in hospital: does it increase length of stay? Aust N Z J Psychiatry 1998;32:805–8.
25. Inouye SK, Viscoli CM, Horwitz RI, Hurst LD, Tinetti ME. A predictive model for delirium in hospitalized elderly medical patients based on admission characteristics. Ann Intern Med 1993;119:474–81.
26. Matsuoka H, Yoshiuchi K, Koyama A, Otsuka M, Nakagawa K. Chemotherapeutic drugs that penetrate the blood-brain barrier affect the development of hyperactive delirium in cancer patients. Palliat Support Care 2015;13:859–64.
27. Matsuo N, Morita T, Matsuda Y, Okamoto K, Matsumoto Y, Kaneishi K, et al. Predictors of delirium in corticosteroid-treated patients with advanced cancer: an exploratory, multicenter, prospective, observational study. J Palliat Med 2017;20:352–9.
28. Gaudreau JD, Gagnon P. Psychotogenic drugs and delirium pathogenesis: the central role of the thalamus. Med Hypotheses 2005;64:471–5.
29. Litaker D, Locala J, Franco K, Bronson DL, Tannous Z. Preoperative risk factors for postoperative delirium. Gen Hosp Psychiatry 2001;23:84–9.
30. Marcantonio ER, Juarez G, Goldman L, Mangione CM, Ludwig LE, Lind L, et al. The relationship of postoperative delirium with psychoactive medications. JAMA 1994;272:1518–22.

Article information Continued

Fig. 1.

Receiver operating characteristic curves of predictive models according to variables. (A) Sociodemographic+clinical characteristics. (B) Sociodemographic+clinical+laboratory characteristics. (C) Sociodemographic+clinical+concomitant medication characteristics. (D) Sociodemographic+clinical+laboratory+concomitant medication characteristics. AUROC, area under the receiver operating characteristic curve; CI, confidence interval.

Fig. 2.

Calibration plot for predictive model of delirium occurrence using bootstrapping method.

Table 1.

Baseline characteristics of patients

Characteristic Total (n=2,152) Delirium group (n=127) Non-delirium group (n=2,025) p-value
Sociodemographic characteristic
 Age (yr), median (range) 64 (18-97) 69 (27-88) 63 (18-97) < 0.001
 Sex
  Male 1,257 (58.4) 89 (70.1) 1,168 (57.7) 0.007
  Female 895 (41.6) 38 (29.9) 857 (42.3)
 Marital status
  Married 1,776 (82.5) 107 (84.3) 1,669 (82.4) 0.532
  Not married 368 (17.1) 19 (15.0) 349 (17.2)
  Unknown 8 (0.4) 1 (0.8) 7 (0.3)
 Education (n=1,513)
  High school or less 984 (65.0) 51 (66.2) 933 (65.0)
  College/Graduate school 529 (35.0) 26 (33.8) 503 (35.0)
 Smoking status
  Non-smoker 1,475 (68.5) 74 (58.3) 1,401 (69.2)
  Ex-smoker 626 (29.1) 45 (35.4) 581 (28.7)
  Current smoker 51 (2.4) 8 (6.3) 43 (2.1)
 Alcohol consumption (n=2,151)
  Non-drinker 1,772 (82.4) 98 (77.2) 1,674 (82.7) 0.036
  1-3 times a week 265 (12.3) 16 (12.6) 249 (12.3)
  ≥ 4 times a week 114 (5.3) 13 (10.2) 101 (5.0)
 BMI (n=2,144)
  Obese 355 (16.6) 21 (16.5) 334 (16.6) 0.563
  Overweight 343 (16.0) 24 (18.9) 319 (15.8)
  Normal weight 980 (45.7) 51 (40.2) 929 (46.1)
  Underweight 466 (21.7) 31 (24.4) 435 (21.6)
 Living with family 1,430 (66.4) 82 (64.6) 1,348 (66.6) 0.698
 National health insurance 2,054 (95.4) 118 (92.9) 1,936 (95.6) 0.182
 Visual impairment (wearing glasses) 108 (5.0) 5 (3.9) 103 (5.1) 0.679
 Hearing impairment (using hearing aids) 22 (1.0) 4 (3.1) 18 (0.9) 0.037a)
Clinical characteristic
 Primary cancer
  Lung 442 (20.5) 49 (38.6) 393 (19.4) < 0.001
  Gastroesophageal and colorectal 610 (28.3) 31 (24.4) 579 (28.6)
  Hepatopancreatobiliary 391 (18.2) 21 (16.5) 370 (18.3)
  Genitourinary 148 (6.9) 7 (5.5) 141 (7.0)
  Hematology 91 (4.2) 3 (2.4) 88 (4.3)
  Breast 122 (5.7) 1 (0.8) 121 (6.0)
  Head and neck 139 (6.5) 3 (2.4) 136 (6.7)
  Others 209 (9.7) 12 (9.4) 197 (9.7)
 Past medical history
  History of delirium 56 (2.6) 24 (18.9) 32 (1.6) < 0.001a)
  History of cardiovascular disease 833 (38.7) 62 (48.8) 771 (38.1) 0.019
  History of diabetes mellitus 476 (22.1) 39 (30.7) 437 (21.6) 0.020
  History of respiratory disease 189 (8.8) 15 (11.8) 174 (8.6) 0.255
  History of liver disease 144 (6.7) 11 (8.7) 133 (6.6) 0.358
  History of psychiatric disease 132 (6.1) 17 (13.4) 115 (5.7) < 0.001
  History of CVA/head trauma 175 (8.1) 16 (12.6) 159 (7.9) 0.065
 Past history of admission 2,016 (93.7) 117 (92.1) 1,899 (93.8) 0.573
 Vital sign at admission
  Blood pressure
   SBP < 140 mmHg and DBP < 90 mmHg 1,624 (75.5) 89 (70.1) 1,535 (75.8) 0.166
   SBP ≥ 140 mmHg or DBP ≥ 90 mmHg 528 (24.5) 38 (29.9) 490 (24.2)
  Body temperature
   Normal temperature (< 38℃) 2,060 (95.7) 121 (95.3) 1,939 (95.8) 0.819
   Hyperthermia (≥ 38℃) 92 (4.3) 6 (4.7) 86 (4.2)
  Heart rate
   Normal rate (HR < 120/min) 2,024 (94.1) 114 (89.8) 1,910 (94.3) 0.050
   Tachycardia (HR ≥ 120/min) 128 (5.9) 13 (10.2) 115 (5.7)
  Respiratory rate
   Normal respiratory rate (RR < 20/min) 1,057 (49.1) 61 (48.0) 996 (49.2) 0.855
   Tachypnea (RR ≥ 20/min) 1,095 (40.9) 66 (52.0) 1,029 (50.8)
 Chemotherapy during hospitalization 604 (28.1) 20 (15.7) 584 (28.8) 0.001
 Duration of hospital admission (day), median (range) 8 (1-117) 13 (2-71) 8 (1-117) < 0.001
Laboratory finding
 Complete blood test
  Hb < 8 g/dL 183 (8.5) 16 (12.6) 167 (8.7) 0.099
  WBC > 9.8×103/μL 708 (32.9) 48 (37.8) 660 (32.6) 0.243
  PLT < 150×103/μL 559 (26.0) 41 (32.3) 518 (25.6) 0.096
 Liver function
  ALP > 129 IU/L 986 (45.8) 55 (43.3) 914 (45.1) 0.013
  AST > 40 IU/L 712 (33.1) 41 (32.3) 671 (33.1) 0.847
  ALT > 40 IU/L 466 (21.7) 20 (15.7) 446 (22.0) 0.097
  Total bilirubin > 1.2 mg/dL 488 (22.7) 27 (21.3) 461 (22.8) 0.744
  Albumin < 3.5 g/dL 1,240 (57.6) 81 (6.8) 1,159 (57.2) 0.165
  Total Protein < 6.5 g/dL 1,303 (60.5) 80 (63.0) 1,223 (60.4) 0.576
  Prothrombin time-INR > 1.2 (n=1,851) 559 (30.2) 48 (41.4) 511 (29.5) 0.009
 Renal function
  BUN > 20 mg/dL 802 (37.3) 72 (56.7) 730 (36.0) < 0.001
  Creatinine > 1.2 mg/dL 321 (14.9) 27 (21.3) 294 (14.5) 0.041
  eGFR < 60 mL/min/1.732 m2 362 (16.8) 28 (22.0) 334 (16.5) 0.112
  Uric acid > 8.6 mg/dL 135 (6.3) 6 (4.7) 129 (6.4) 0.573
  Na < 135 mEq/L 748 (34.8) 55 (43.3) 693 (34.2) 0.043
  K > 5.1 mEq/L 168 (7.8) 17 (13.4) 151 (7.5) 0.019
  Ca > 10.2 mg/dL 81 (3.8) 11 (8.7) 70 (3.5) 0.007a)
  P > 4.5 mg/dL 162 (7.5) 8 (6.3) 154 (7.6) 0.614
 Glucose < 74 mg/dL 54 (2.5) 4 (3.1) 50 (2.5) 0.557a)
 CRP > 0.3 mg/dL (n=1,928) 1,702 (88.3) 94 (81.0) 1,612 (88.9) 0.114
Medication
 Concomitant drugs
  Opioid 1,358 (63.1) 97 (76.4) 1,261 (62.3) 0.002
  Sedative 362 (16.8) 47 (37.0) 315 (15.6) < 0.001
  Antidepressant 249 (11.6) 22 (17.3) 227 (11.2) 0.044
  AED 271 (12.6) 19 (15.0) 252 (12.4) 0.408
  Cholinergic 172 (8.0) 8 (6.3) 164 (8.1) 0.508
  Anti-cholinergic 159 (7.4) 6 (4.7) 153 (7.6) 0.295
 Antibiotics use during hospital admission 1,295 (60.2) 98 (77.2) 1,197 (59.1) < 0.001

Values are presented as number (%) unless otherwise indicated. AED, antiepileptic drug; ALP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; BUN, blood urea nitrogen; CRP, C-reactive protein; CVA, cerebrovascular accident; DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate; Hb, hemoglobin; HR, heart rate; INR, international normalized ratio; PLT, platelet; RR, respiratory rate; SBP, systolic blood pressure; WBC, white blood cell.

a)

Fisher’s exact test was used for these variables to assess statistical significance of differences.

Table 2.

Risk factor for occurrence of delirium on univariate analysis and multivariate analysis for all variables

Univariate
Multivariate
OR (95% CI) p-value OR (95% CI) p-value
Sociodemographic characteristic
 Age at diagnosis (yr), ≥ 65 (vs. < 65) 2.07 (1.43-3.06) < 0.001 1.86 (1.22-2.87) 0.004
 Sex, female (vs. male) 0.58 (0.39-0.85) 0.006 0.66 (0.42-0.99) 0.051
 Marital status, not-married (vs. married) 1.22 (0.76-2.04) 0.439
 Smoking status
  Non-smoker 1.00 1.00
  Ex- or current smoker 1.61 (1.11-2.31) 0.011 1.09 (0.70-1.71) 0.696
 Alcohol consumption
  Non-drinker 1.00
  Drinker 1.00 (1.00-1.00) 0.677
 BMI
  Obese and overweight 1.00 1.00
  Non-obese 1.94 (1.25-3.01) 0.003 1.70 (1.07-2.80) 0.029
 Not living with family 1.09 (0.75-1.58) 0.643
 Without national health insurance 1.66 (0.76-3.21) 0.162
 Visual impairment (wearing glasses) 0.76 (0.27-1.73) 0.565
 Hearing impairment (using hearing aids) 3.62 (1.04-9.89) 0.022 3.38 (0.89-10.34) 0.046
Clinical characteristic
 No. of diseases with a history, > 1 (vs. ≤ 1) 0.92 (0.56-1.43) 0.710
  History of delirium 14.51 (8.18-25.47) < 0.001 11.90 (6.28-22.42) < 0.001
  History of cardiovascular disease 1.55 (1.08-2.22) 0.017 1.15 (0.74-1.77) 0.540
  History of diabetes mellitus 1.61 (1.08-2.37) 0.017 1.44 (0.92-2.22) 0.102
  History of respiratory disease 1.42 (0.78-2.42) 0.216
  History of liver disease 1.35 (0.67-2.46) 0.361
  History of psychiatric disease 2.57 (1.44-4.32) < 0.001 1.74 (0.92-3.14) 0.077
  History of CVA/Head trauma 1.69 (0.94-2.85) 0.060
 Past history of admission 1.32 (0.68-2.56) 0.415
 Vital sign at admission
  Blood pressure
   SBP ≥ 140 mmHg or DBP ≥ 90 mmHg 1.00
   SBP < 140 mmHg and DBP < 90 mmHg 0.75 (0.51-1.11) 0.147
  Body temperature
   Normal temperature (< 38℃) 1.00
   Hyperthermia (≥ 38℃) 1.12 (0.43-2.41) 0.796
  Heart rate
   Normal rate (HR < 120/min) 1.00 1.00
   Tachycardia (HR ≥ 120/min) 1.89 (0.99-3.35) 0.038 1.53 (0.73-2.97) 0.232
  Respiratory rate
   Normal respiratory rate (RR < 20/min) 1.00
   Tachypnea (RR ≥ 20/min) 1.05 (0.73-1.50) 0.801
 Duration of hospital admission (day), > 8 (vs. ≤ 8) 3.02 (2.06-4.51) < 0.001 3.24 (2.14-5.01) < 0.001
 Without chemotherapy during hospitalization 2.17 (1.36-3.63) 0.002 1.77 (1.07-3.07) 0.032
Laboratory findings
 Complete blood test
  Hb < 8 g/dL (vs. ≥ 8 g/dL) 1.10 (0.67-1.93) 0.714
  WBC ≤ 9.8×103/μL (vs. > 9.8×103/μL) 0.69 (0.48-0.99) 0.048 1.00 (0.66-1.52) 0.996
  PLT < 150×103/μL (vs. ≥ 150×103/μL) 1.39 (0.94-2.03) 0.096
 Liver function
  ALP > 129 IU/L (vs. ≤ 129 IU/L) 1.59 (1.11-2.29) 0.012 1.46 (0.98-2.17) 0.062
  AST > 40 IU/L (vs. ≤ 40 IU/L) 0.96 (0.65-1.40) 0.843
  ALT > 40 IU/L (vs. ≤ 40 IU/L) 0.66 (0.39-1.06) 0.098
  Total bilirubin > 1.2 mg/dL (vs. ≤ 1.2 mg/dL) 0.92 (0.58-1.40) 0.694
  Albumin < 3.5 g/dL (vs. ≥ 3.5g/dL) 0.76 (0.52-1.10) 0.149
  Total protein < 6.5 g/dL (vs. ≥ 6.5 g/dL) 1.09 (0.90-1.32) 0.388
 Renal function
  BUN > 20 mg/dL (vs. ≤ 20 mg/dL) 2.32 (1.62-3.35) < 0.001 1.58 (1.06-2.36) 0.024
  Cr > 1.2 mg/dL (vs. ≤ 1.2 mg/dL) 1.59 (1.00-2.44) 0.040 1.05 (0.62-1.74) 0.846
  eGFR < 60 mL/min/1.732 m2 (vs. ≥ 60 mL/min/1.732 m2) 1.50 (0.98-2.31) 0.065
  Na < 135 mEq/L (vs. ≥ 135 mEq/L) 1.47 (1.02-2.11) 0.038 1.01 (0.66-1.53) 0.979
  K > 5.1 mEq/L (vs. ≤ 5.1 mEq/L) 1.92 (1.09-3.20) 0.017 1.34 (0.69-2.51) 0.371
  Ca > 10.2 mg/dL (vs. ≤ 10.2 mg/dL) 2.65 (1.30-4.94) 0.004 2.15 (0.98-4.33) 0.042
  P > 4.5 mg/dL (vs. ≤ 4.5 mg/dL) 1.86 (1.07-3.22) 0.028 1.38 (0.70-2.73) 0.357
  Uric acid > 8.6 mg/dL (vs. ≤ 8.6 mg/dL) 1.01 (0.48-2.10) 0.990
  Glucose < 74 mg/dL (vs. ≥ 74 mg/dL) 1.37 (0.92-2.05) 0.120
Medication
 Concomitant drugs
  Opioid 1.96 (1.30-3.03) 0.002 1.24 (0.77-2.03) 0.387
  Sedative 3.19 (2.17-4.64) < 0.001 3.15 (2.04-4.83) < 0.001
  Antidepressant 1.66 (1.00-2.63) 0.039 1.15 (0.65-1.98) 0.617
  AED 1.24 (0.73-2.00) 0.408
  Cholinergics 0.76 (0.34-1.49) 0.469
  Anti-cholingergics 0.61 (0.23-1.29) 0.241
 Antibiotics use during hospital admission 2.34 (1.55-3.63) < 0.001 1.41 (0.88-2.31) 0.160

AED, antiepileptic drug; ALP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; BUN, blood urea nitrogen; CI, confidence interval; CVA, cerebrovascular accident; DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate; Hb, hemoglobin; HR, heart rate; OR, odds ratio; PLT, platelet; RR, respiratory rate; SBP, systolic blood pressure; WBC, white blood cell.

Table 3.

Performance comparison of predictive models according to sociodemographic, clinical, laboratory and concomitant medication characteristics

Model AUC Sensitivity (%) Specificity (%) PPV (%) NPV (%)
Sociodemographic+clinical characteristics 0.795 71.7 73.6 14.5 97.6
Sociodemographic+clinical+laboratory findings 0.809 75.6 73.2 15.0 97.9
Sociodemographic+clinical+concomitant medication 0.821 63.8 84.4 20.4 97.4
Sociodemographic+clinical+laboratory findings+concomitant medication 0.831 80.3 72.0 15.2 98.3

AUC, area under curve; NPV, negative predictive value; PPV, positive predictive value.