Predicting Delirium in Elderly Patients: Development and Validation of a Risk-stratification Model

Age and Ageing 1996.25:31-3 Predicting Delirium in Elderly Patients: Development and Validation of a Risk-stratification Model S. T. O'KEEFFE, J. N. LAVAN Summary Delirium is a common and serious complication of acute illness in elderly patients. The aim of this study was to develop and validate a model for predicting development of in elderly medical inpatients who did not have on admission. Consecutive admissions to an acute geriatric unit underwent standardized cognitive assessment every 4 hours. Delirium was diagnosed according to DSM-3 criteria. Independent predictors of in a derivation group of 0 patients were determined using stepwise logistic regression analysis; the predictive model comprised dementia, severe illness and elevated serum urea. This model performed well in a validation group of 4 patients. We conclude that elderly medical patients can be stratified according to their risk for developing using a simple clinical model. Introduction Delirium, or acute confusional state, is a common and serious complication of hospitalization in elderly patients. Studies using formal diagnostic criteria for have shown that between and 22% of elderly medical patients are delirious on admission to hospital, and up to 30% develop after admission [1-6]. Delirious patients have increased morbidity, prolonged hospital stay and increased risk of entering long-term care on discharge from hospital compared with patients without [5,, ]. Early recognition of and treatment of the underlying cause can shorten the duration of [9]. Nevertheless, many patients do badly even with optimum management. Thus, prevention of should be a major focus of attention. This requires identification of risk factors for in order to predict which patients are at greatest risk. The purpose of this study was to derive and test a predictive model for based on the admission characteristics of elderly patients without admitted to an acute-care geriatric unit. Methods Subjects: The potential study population comprised 225 consecutive patients admitted over an 1-month period to a 24-bed acute-care geriatric unit within a university teaching hospital. The acute geriatric unit targets acutely ill elderly people who are frail and dependent in activities of daily living. Patients admitted electively for investigations, rehabilitation or respite care, patients with severe aphasia or deafness, patients who were expected to remain in hospital less than 4 hours and those who were not assessed within 4 hours of admission were excluded. The division of the total study population of 225 into a derivation group of 125 patients and a validation group of 0 patients occurred after completion of recruitment into the study but before any data analysis had been performed. Delirium was present on admission in 25 of 125 patients in the derivation group and 16 of 0 patients in the validation group; these patients are excluded from the present analysis. Clinical evaluation: All patients underwent detailed cognitive assessment within 4 hours of admission; over 5% of initial assessments were performed within 6 hours of admission. The initial assessment included a semi-structured interview using the Delirium Assessment Scale (DAS) [], a recently validated instrument for evaluating individual symptoms of, administration of the Folstein Mini- Mental State Examination (MMSE) [11, 12], and physical examination of the patient. Information about the previous cognitive and functional status, past medical history and about the pattern of onset of cognitive impairment was sought from family members, carers and the general practitioner and by inspection of old medical and nursing notes. Details of admission medications and their doses were recorded. All patients had serum electrolytes, full blood count and urinalysis performed on admission; other tests depended on the clinical circumstances. Patients underwent repeated cognitive assessment every 4 hours until discharge or death. Nursing and medical staff, including night staff, were interviewed daily regarding the patients. Diagnosis of : Diagnosis of was made using the Diagnostic and Statistical Manual (3rd edition) (DSM-3) criteria [13]. Operational definitions of the DSM-3 criteria proposed by workers at the Hospital of the University of Pennsylvania were used in this study, with slight modifications [6, 14]. We specified that all symptoms required for a

3* S. T. O'KEEFFE, J. N. LAVAN Table I. Admission characteristics of patients in derivation and validation groups Mean (SD) age (years) Female [n(%)] Long-term care patient [n(%)] Dementia [n(%)] Mean (SD) disability score Mean (SD) comorbidity index Severity of illness [n(%)]: mild moderate severe Incident [n(%)] >p< 0.001. Derivation group (n = 0) 2.(6.2) 63 (63) 20 (20) 24 (24) 2.6(1.) 1.9(0.) 32 (32) 41 (41) 2 (2) 2 (2) Validation group (n = 4) 1.2(4.4) 4 (56) 9 (11) 1 () 2.2(1.6)' 1.(0.9) 30 (36) 32 (3) 22 (26) 25 (30) diagnosis of had to be present during the same 4- hour period. Risk factor diagnosis: A diagnosis of dementia was made if there was evidence of cognitive impairment sufficient to interfere with social functioning or if the Blessed Dementia Rating Score was 4 or more with symptoms present for at least 6 months [15]. A subjective rating of overall illness severity as mild, moderate or severe was made on initial admission by the study physician based on the approach of Charlson and colleagues [16]. In a pilot study, inter-rater agreement between medical staff on our unit for rating severity was excellent (kappa scores >0.9). A weighted comorbidity index was used to quantify medical comorbidity [1]; this index takes into account the number and seriousness of cormorbid diseases. Dementia was excluded from calculation of the cormorbidity score in this study. Functional status was assessed using the Katz activities of daily living (ADLs) scale [1]; we calculated a disability score consisting of the number of ADLs with which the patient required assistance. Depression was diagnosed if the score on an admission 15- item Geriatric Depression score was 5 or more [19]. Visual or hearing impairment was recorded if these disabilities were severe enough, when those aids (hearing aids, spectacles) available on admission were used, to interfere with activities of daily living. Alcohol abuse was noted if a man reported taking more than four drinks a day or a woman more than two drinks a day, if alcohol abuse was reported by the patient or noted in previous medical records, or if a carer expressed any concern about the alcohol intake of a patient. The following criteria were used to define abnormal laboratory values: serum urea greater than mmol/1, sodium greater than 150 mmol/1 or less than 130 mmol/1, glucose greater than 16 mmol/1 or less than 3.5 mmol/1, albumin less than 30 mmol/1 and white blood cell count greater than 12 x 9 /l or less than 3.5 x 9 /l. Temperatures greater than 3 C or less than 35 C were also considered abnormal. Statistics: Unadjusted odds ratios (OR) and confidence intervals (CI) were calculated for each variable with incident as the dependent variable. Continuous variables were dichotomized at clinically meaningful cut-off points. Bivariate correlates of at a significance level of Table II. Risk factors for incident in derivation group % with (n = 2) % without (n = 2) Ddds ratios 95% CI) Abnormal serum sodium Dementia Elevated serum urea Severe illness Hypo-albuminaemia Abnormal temperature Age >0 years Abnormal white cell count Dependent in >2 ADLs Depression Alcohol abuse Visual impairment Long-term care patient Male sex 25 46 6 50 29 2 36 50 29 6 15 31 1 15 5. (1.5-22.0) *. (2.0-11.6) 1. (0.-.) 1.5 (1.9-.6) 3.0 (0.9-.5) 2.2 (0.-6.1) 1 1.9 (0.6-5.6) 24 1. (0.-4.4) 44 1.3 (0.6-2.) 2 3.9 (0.1-1.) ( 3. (0.5-2.0) 43 ( 3.5 (0.2-1.3)

PREDICTING DELIRIUM IN ELDERLY PATIENTS 319 Table III. Risk factors for incident in derivation group medications % with (n = 2) % without (n = 2) Odds ratios (95% CI) Neuroleptic agents Narcotic agents Anticholinergic agents NSAIDs Digoxin Histamine-2 blockers Benzodiazepines 1 14 14 23 2.0 (0.6-.2) 1.6 (0.4-6.1) 1.6 (0.4-6.1) 1.0 (0.4-3.0) 1.0 (0.3-2.) 0. (0.2-4.5) p < 0.1 were entered into stepwise logistic regression analysis. For variables with 50% or greater collinearity, only the variable most strongly related to in bivariate analysis was included in the multivariate analysis. Second-order interactions between main effects in the final model were examined. SAS statistical programs (SAS Institute, Cary, NC) were used for these analyses. The independent predictors of in the derivation group were combined to form a risk-stratification model which was tested in the validation group. Performance of the model in the two groups was quantified and tested using the areas under the receiving operating characteristic (ROC) curves [20]. A value for 1 for area corresponds to perfect prediction, while a value of 0.5 is equivalent to that expected by chance. Results Table I shows the characteristics of the derivation and validation groups. Functional impairment was greater in the derivation group; other differences in admission characteristics were not significant. Development of the predictive model: Delirium developed during hospitalization in 2% of patients in the derivation group. Of those who became delirious, 3% met diagnostic criteria within 5 days of admission. The relationship between and potential risk factors on admission in the derivation group is shown in Tables Table IV. Performance of the predictive model in the derivation and validation groups Points* Development of [nos. (%)] Derivation group (n = 0) 3/36 () /43 (23) 9/15(60) 6/6 (0) Validation group (n = 4) 5/3(13) /31 (32) 5/ (50) 5/5 (0) One point was added for each risk factor present. Chi-square for trend; derivation group, p < 0.02; validation group, p < 0.01. II and III. Four factors appeared to be strongly predictive of : history of chronic cognitive impairment, severe (as opposed to mild or moderate) illness, elevated serum urea and abnormal serum sodium. Abnormal body temperature or white blood cell count, low serum albumin and use of neuroleptics also appeared to confer an increased risk of, but these results did not reach statistical significance. Other variables examined in bivariate analysis which were not related to the risk of included number and type of admission diagnoses and number of admission medications. The predictive model derived by logistic regression analysis included dementia (adjusted OR 4.; 95% CI 1.4-15.2), severe illness (adjusted OR 5.6; 95% CI 1.- 1.2) and elevated serum urea (adjusted OR 5.1; 95% CI 1.-14.9). No significant interactions between these main effects were found. The beta coefficients from the regression equation for the three risk factors were very similar (chronic cognitive impairment = 1.54, severe illness = 1.2, abnormal blood urea =1.62), and the scoring system for the predictive model was derived by adding one point for each risk factor present. Performance of the predictive model: In the validation group, developed in 25 (30%) patients; developed within 5 days of admission in 2% of these patients. The relationship between risk score and the observed rates of in the derivation and validation groups is shown in Table IV. The area under the ROC curve was 0.9 (95% CI 0.69-0.90) for the derivation group and 0.5 (95% CI 0.63-0.6) for the validation group. Discussion Risk factors for identified in previous studies include advanced age, male sex, severity of illness, increased blood urea, dementia, fever or hypothermia, visual impairment, depression, alcohol abuse and use of psychoactive medications [1-6, -25]. However, several studies failed to distinguish between present on admission to hospital (prevalent ) and developing after admission to hospital (incident ). This distinction is necessary to

32O S. T. O'KEEFFE, J. N. LAVAN determine whether risk factors actually precede the development of. The strengths of the present study include the prospective design, the use of a standardized and validated instrument to identify and the performance of repeated assessments throughout hospitalization. Furthermore, the initial assessment of most patients within 6 hours of admission ensured adequate distinction of prevalent and incident. The predictive model derived in this study includes three well known risk factors for : severe illness, dementia and elevated blood urea nitrogen. This model performed well in an independent data set. The only previous study to derive and test a predictive model for in elderly people was conducted by Inouye et al. in Connecticut [3]. The results of the two studies are remarkably similar despite differences in how was diagnosed and in the definitions of risk factors. They identified visual impairment, severe illness, cognitive impairment and a high urea/creatinine ratio as independent predictors of ; this model also performed well in a validation cohort. Dementia has been identified as a risk factor for in medical [3-5], surgical [25] and psychiatric patients [26]. Pre-existing brain damage increases the vulnerability of the brain to physical insults. It is also possible that the disruption of admission to hospital contributes to the development of in demented patients [2]. In this study, severe illness, determined by the clinical judgement of the study physician, was an independent predictor of. This is consistent with the results of previous studies which suggested that the overall severity of illness is a more important predictor of outcome than the specific diagnoses [3-5]. Charlson and colleagues have shown that physicians' estimates of illness severity are an accurate predictor of in-hospital mortality [16]. Also, subjective judgement of illness severity by an experienced clinician is a better predictor of outcome in elderly general medical patients than 'physiological' scores such as APACHE [2]. Blood urea level greater than mmol/l was another independent predictor of. This level represented the upper limit of normal in Leask's study of blood results in healthy elderly [29]. In some patients, elevated urea is probably due to dehydration. There is good evidence that salt and water depletion or both can lead to [30]. Also, a disturbance such as an acute infection can cause both and dehydration [31]. In other patients, elevated urea probably represents a non-specific marker of underlying disease; this is also true of hypo-albuminaemia which was identified as a predictor of in several studies [5, 32]. The diagnosis of in this study was based on the DSM-3 criteria rather than the DSM-3R criteria. We, like others, believe that the DSM-3 criteria correspond more closely to the clinical features of [33]. Also, Liptzin and colleagues have shown that the DSM-3 criteria are more sensitive than the DSM-3R criteria [34]. The small sample size and the relatively small number of patients with in the derivation group are important limitations of our study. Several potential risk factors occurred in small numbers of patients in this study, and the confidence intervals for these variables are wide. Another limitation of our study is that we did not use formal screening instruments to detect visual impairment or alcohol abuse, although we did record the presence of these conditions according to predetermined criteria. Stepwise regression procedures are very powerful tools for deriving independent predictive variables. However, caution is necessary when interpreting the results of stepwise solutions. Very slight differences among the correlations between predictor and outcome variables can lead to major differences in which variables enter the final model using the stepwise approach. The relative importance of variables is not always reflected by the variables retained in a final stepwise model. Thus, it is not appropriate to say that a particular group of variables are the 'best' or 'most important' or to use a stepwise solution to try and explain a phenomenon such as. There are a number of possible applications for the predictive model derived in this study. Recognition and awareness of by medical and nursing staff on general medical and surgical wards is often poor [35]. Use of the simple model derived in this study to identify patients at high risk of developing might encourage improved surveillance by staff for symptoms of, thus allowing early treatment of the precipitating factors. Use of the model to stratify risk and identify high-risk patients may also facilitate research into strategies to prevent or minimize the impact of. References 1. Anthony JC, LeResche LA, Von-Korff MR, Niaz U, Folstein MF. Screening for on a general medical ward: the tachistoscope and a global accessibility rating. Gen Hosp Psychiatry 195;:36-42. 2. Jitapunkel S, Pillay I, Ebrahim S. Delirium in newly admitted elderly patients: a prospective study. Q J Med 1992;300:30-14. 3. Inouye SK, Viscoli CM, Horwitz RI, Hurst LD, Tinetti ME. A predictive model for in hospitalized elderly medical patients based on admission characteristics. Ann Intern Med 1993;119:44-1. 4. Francis J, Martin D, Kapoor WN. A prospective study of in hospitalized elderly. JAMA 1990;263: 9-1. 5. Levkoff SE, Evans DA, Liptzin B, et al. Delirium: the occurrence and persistence of symptoms among elderly hospitalized patients. Arch Intern Med 1992;152:334-40. 6. Johnson JC, Gottlieb GL, Sullivan E, et al. Using DSM-3 criteria to diagnose in elderly general medical patients. J Gerontol 1990;45:M113-19.. Hodkinson HM. Mental impairment in the elderly. J R Coll Physicians 193;:305-1.. Thomas RI, Cameron DJ, Fahs MJ. A prospective study

of and prolonged hospital stay. Arch Gen Psychiatry 19;45:93-4O. 9. Koizumi J, Shiraishi H, Ofuku K, Suzuki T. Duration of shortened by the correction of electrolyte imbalance. Jpn J Psychiatry Neurol 19;42:1-.. O'Keeffe ST. Rating the severity of : the Delirium Assessment Scale. Int J Geriatr Psychiatry 1994;9:551-6. 11. Folstein MF, Folstein SE, McHugh PR. Mini-mental state a practical method for grading the cognitive state of patients for the physician. J Psychiatr Res 195;12: 19-9. 12. O'Neill D, Condren L, O'Kelly F, et al. Cognitive impairment in the elderly. Ir Med J 19;1:11-13. 13. American Psychiatric Association. Diagnostic and statistical manual of mental disorders, 3rd edn. Washington DC: American Psychiatric Association, 190. 14. Gottlieb GL, Johnson J, Wanich C, Sullivan E. Delirium in the medically ill elderly: operationalizing the DSM-3 criteria. Int Psychogeriatr 1991;3:11-9. 15. Blessed G, Tomlinson BE, Roth M. The association between quantitative measures of dementia and of senile change in the cerebral grey matter of elderly subjects. Br J Psychiatry 196;114:9-11. 16. Charlson ME, Sax FL, MacKenzie CR, Fields SD, Braham RL, Douglas RG. Assessing illness severity: does clinical judgement work? J Chron Dis 196;39:439-52. 1. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.^ Chron Dis 19;4O:33-3. 1. Katz S, Akpom CA. A measure of primary sociobiological functions. Int J Health Serv 196;6:493-50. 19. Sheikh JI, Yesavage JA. Geriatric Depresson Scale (GDS): recent evidence and development of a shorter version. In: Brink TL, ed. Clinical gerontology: a guide to assessment and intervention. New York; Haworth Press, 196;165-2. 20. Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 192; 143:29-36.. Schor JD, Levkoff SE, Lipsitz LA, et al. Risk factors for in hospitalized elderly. JAMA 1992;26:2-31. 22. Williams MA, Cambell EB, Raynor WJ, Musholt MA, Mlynarczyk SM, Crane LF. Predictors of acute confusional states in hospitalized elderly patients. Res Nurse Health 195;35:0-. PREDICTING DELIRIUM IN ELDERLY PATIENTS 3 23. Rockwood K. Acute confusion in elderly medical patients. J Am Geriatr Soc 199;3:150-4. 24. Pompei P, Foreman M, Rudberg MA, Inouye SK, Braund V, Cassel CK. Delirium in hospitalized older persons: outcomes and predictors. J Am Geriatr Soc 1994;42:09-15. 25. Marcantonio ER, Goldman L, Mangione CM, et al. A clinical prediction rule for after elective noncardiac surgery. JAMA 1994;:134-9. 26. Koponen H, Stenback U, Mattila, Soininen H, Reinikainen PJ. Delirium among elderly people admitted to a psychiatric hospital: clinical course during the acute stage and one-year follow-up. Ada Psychiatr Scand 193;9: 59-5. 2. Grimley Evans J. The psychiatric aspects of physical disease. In: Levy R, Post F. eds. The psychiatry of late life. Oxford: Blackwell Scientific Publications, 192;114-42. 2. Young L, Barer D. Illness severity in elderly medical patients. Age Ageing 1995;24{suppl 2):. 29. Leask RGS, Andrews GR, Caird FI. Normal values for sixteen blood constituents in the elderly. Age Ageing 193;2:14-23. 30. Lipowski ZJ. Delirium: acute confusional state. Oxford: Oxford University Press, 1990. 31. Seymour DG, Henschke PJ, Cape RDT, Campbell AJ. Acute confusional states and dementia in the elderly: the role of dehydration/volume depletion, physical illness and age. Age Ageing 190;9:l3-46. 32. Dickson LR. Hypoalbuminemia in. Psychosomatics 1991 ;32:31-23. 33. Lipowski ZJ. Organic mental disorders and DSM-4. Am J Psychiatr 1990;14:94. 34. Liptzin B, Levkoff SE, Cleary PD, et al. An empirical study of diagnostic criteria for. Am J Psychiatry 1991;14:454-. 35. Lyness JM. Delirium: masquerades and misdiagnosis in elderly inpatients. J Am Geriatr Soc 1990;3:l235-. Authors' address Department of Geriatric Medicine, Royal Liverpool University Hospital, PO Box 14, Prescot Street, Liverpool L69 3BX Received in revised form 23 February 1996