Developing a Prediction Model for Post-Operative Delirium and Long-Term Outcomes Among Older Patients Receiving Elective Orthopedic Surgery: A Prospective Cohort Study in Taiwan

Abstract

This study aimed to develop a prediction model for post-operative delirium among older patients receiving elective orthopedic surgery and to evaluate its effectiveness in predicting long-term health outcomes. This prospective cohort study screened all subjects aged over 60 years who were admitted for elective orthopedic surgery in a tertiary medical center in Taiwan from April, 2011, to December, 2013. Demographic characteristics, surgery-related factors, and results of comprehensive geriatric assessment (CGA) were all used to develop the prediction model. Long-term health outcomes, including mortality, nursing home admission, and functional status in the first year after surgery, were used to further evaluate the effectiveness of the prediction model. Overall, 461 patients (median age, 73 years; interquartile range [IQR], 67–80 years; 42.3% males) were enrolled, and 37 patients (8.0%) developed post-operative delirium. Prediction models were developed on the basis of demographic characteristics and surgery-related factors (model 1) and of demographic characteristics, surgery-related factors, and geriatric assessment variables (model 2). Although both models effectively predicted the occurrence of post-operative delirium, duration of post-operative delirium, total hospital days, nursing home admission, and mortality, model 2 was more likely to differentiate cases with functional decline during the first year after surgery. In conclusion, a prediction model developed by using demographic characteristics, surgery-related factors, and results of CGA was highly predictive for post-operative delirium, as well as long-term health and functional outcomes.

Introduction

Identified as an aging country (percentage of people aged over 65 exceeding 7% of the total population) in 1993 and expected to be an aged country (percentage of people aged over 65 exceeding 14% of the total population) by 2017, Taiwan has become one of the fastest aging countries in the world.¹ This rapid demographic transition has posed various challenges to the healthcare system, and extensive attention has been paid to the increasing need for surgery among older people. Previous studies have shown that surgery for older patients is often associated with higher risk of mortality, functional decline, nursing home placement, and post-operative delirium.^2
–4 Delirium is defined as a rapid-onset acute disorder of consciousness disturbance and inattention and is closely related to various adverse health outcomes in older patients.⁵ The etiopathogenesis of delirium remains unclear⁶ and is often overlooked by clinicians.^7,8 Post-operative delirium is associated with higher in-hospital and long-term mortality, as well as longer hospital stay, longer intensive care unit stay, and higher chance of nursing home placement.^9

–12 Therefore, an effective prediction model for post-operative delirium is critical to the development of delirium prevention programs in high-risk patients.^13

–16

Because 40% of delirium among older hospital inpatients is potentially preventable,^14,15 it is expected that an effective prediction model will significantly improve the effectiveness of delirium prevention. A number of risk factors for post-operative delirium among older patients have been identified, including demographic and surgery-related factors.^17

–20 Past prediction models have varied from study to study due to differences in surgery types and risk factor profiles between studies.^21
–23 A previous study clearly showed that pre-operative comprehensive geriatric assessment (CGA) is an important indicator of post-operative outcome among older patients receiving elective surgery.²⁴ However, only parts of the CGA have been used for predicting post-operative delirium.^{21
–23,25,26} Moreover, most previously developed delirium prediction models focus on medical inpatients or hip surgery patients, but not specifically on elective orthopedic surgery patients. Therefore, the main aim of this study was to develop a new, more comprehensive model for predicting post-operative delirium in older patients receiving elective orthopedic surgery and to evaluate the effectiveness of this model in improving long-term health and functional outcome.

Methods

Study design

This study prospectively recruited two cohorts receiving elective orthopedic surgery in Kaohsiung Veterans General Hospital—the first between April, 2011, and March, 2012, and the second between January, 2013, and December, 2013. The inclusion criteria and study protocol were the same for both recruitments, but the purpose was different. The purpose of the first recruitment was to evaluate the impact of post-operative delirium on long-term outcome after orthopedic surgery,²⁷ and the second was to explore the association of genotypic biomarkers and post-operative delirium. All patients aged 60 years and older admitted to orthopedic wards for elective orthopedic surgery were screened (204 cases in first cohort and 257 cases in the second cohort). A total of 11 orthopedic surgeons performed surgery for these patients, such as spine surgery, hip arthroplasty, and knee arthroplasty. Two geriatricians regularly visited the study subjects in their orthopedic wards to provide interdisciplinary orthogeriatric services. Patients with the following conditions were excluded: (1) Unable to complete CGA, (2) unable to provide informed consent, (3) having limited life expectancy due to terminal illnesses, and (4) delirium identified before enrollment. The study protocol was approved by the Institutional Review Board of Kaohsiung Veterans General Hospital and written informed consent was obtained from all participants.

Pre-operative and peri-operative assessments

All patients completed an interview questionnaire to collect their demographic profile, clinical characteristics (body mass index, Charlson's Comorbidity Index [CCI], and types of surgery)²⁸ on admission, a pre-operative survey and CGA within the first 24 hr after hospital admission, and all surgical procedures within the first 48 hr after admission. The contents of the CGA included self-reported visual and hearing impairment, poly-pharmacy (defined as currently using more than four prescription drugs for over 2 weeks), depressive symptoms (assessed on the 15-item Chinese Geriatric Depression Scale [GDS-15])^29,30 nutritional status (defined by Mini-Nutritional Assessment [MNA]),³¹ cognitive function (determined on admission by the Chinese version of the Mini-Mental State Examination [MMSE]),³² severity of pain (assessed on a Visual Analogue Scale [VAS]),³³ the baseline activities of daily living (ADL) (evaluated by Barthel Index [BI]),³⁴ and Instrumental Activities of Daily Living (IADL) (evaluated by Lawton–Brody IADL scale).³⁵ Information about surgery-related factors, including type of anesthesia (spinal or general anesthesia), American Society of Anesthesiologists (ASA) class, pre-operative laboratory data (total white cell counts, hemoglobin, serum levels of sodium, potassium, blood urea nitrogen, and creatinine), and blood transfusion during surgery were obtained from surgical records.

Outcomes measures

Primary outcome

Post-operative delirium was assessed by trained research nurses 24 hr after surgery to exclude the residual effect of anesthesia, and delirium was assessed by nursing staff everyday thereafter. The primary assessment, the Confusion Assessment Method (CAM), used information from the primary care nurses and primary caregivers.³⁶ All primary care nurses and research nurses received training for delirium assessment and use of the CAM tool before the study began. If delirium was detected by the CAM tool, the senior psychiatrist was informed to confirm the diagnosis of delirium using Diagnostic and Statistical Manual of Mental Disorders, 4th Edition (DSM-IV) diagnostic criteria. All patients with diagnoses confirmed by the psychiatrist were assigned to the post-operative delirium group. Multiple episodes of delirium in the same patient were counted only once.

Secondary outcomes

The secondary outcomes of this study were the adverse impacts of delirium and duration of post-operative delirium³⁷ on hospital length of stay, ADL and IADL impairment, nursing home admissions, and mortality in the first year after surgery. Research nurses evaluated the ADL and IADL scores and number of nursing home admissions and deaths from the participants or their primary caregivers by telephone interview at 1, 3, 6, and 12 months after surgery. Moreover, functional decline was defined as having a lower ADL and IADL score at follow-up than before surgery.

Statistical analysis

In this study, continuous variables are expressed as median with interquartile range (IQR), and categorical data are expressed as percentages. The chi-squared test or Fisher exact test was used to compare dichotomous and ordinal variables in those with and without delirium, and an independent Student t-test or Mann–Whitney U-test was used to compare continuous variables when appropriate. Logistic regression analysis was used to determine independent risk factors for post-operative delirium and the clinical outcomes. All risk factors were dichotomized for logistic regression and to develop a scoring system for prediction models. The study subjects were divided into an older group (age 75 years and older) and younger group (age less than 75 years) according to the previous literature.^21,38 The highest quintile of the CCI score (CCI≥2) was defined as the highest burden of post-operative mortality and death and revisiting the Emergency Department within 72 hr.^39

–42 ADL dependence was defined as a BI score less than or equal to 75. BI ≤75 has been reported to influence length of stay.^43,44 Cognitive impairment was defined as a MMSE score lower than 24,³² and presence of depressive symptoms was defined as a GDS-15 score of more than 4.^29,30 In laboratory data, a ratio of blood urea nitrogen/serum creatinine of 18 or more⁴⁵ and hemoglobin of less than 12 grams/dL were both regarded as indications of dehydration and anemia, respectively.⁴⁶

In univariate analysis, variables with a p value above 0.1 or the prevalence rate below 5% were excluded for the logistic regression model to identify independent risk factors. Finally, the independent risk factors were generated using multivariate logistic regression analysis with backward elimination in both models.

Model 1 was developed on the basis of clinical characteristics and pre- and post-operative-related factors as in most previous studies. Model 2 was developed on the basis of model 1 factors plus risk factors identified from results of CGA. To avoid over-fitting the data, we generated 1000 random bootstrap samples from the original cohort using the bootstrap resampling method, and the risk factors with the significance above the 0.05 level were excluded.⁴⁷ Total risk score of each model was the sum of the rounded-off adjusted odds ratio values of all remaining variables in each model. The performance of the prediction model was evaluated using receiver operating characteristic (ROC) analysis, and the best model was selected by comparing the area under the ROC curve (AUC) derived from the risk scores using the β coefficients from the logistic regression model.⁴⁸ The Youden Index (the maximum sum of sensitivity and 1 − specificity) was used to determine the optimal cutoff point of the prediction model.⁴⁹ Finally, trends in secondary outcome rates and risk ratio (odds ratio [OR]]) were assessed for internal validation. One thousand bootstrap samples were also resampled to obtain 95% confidence interval (CI) of odds ratios from the 2.5^th and 97.5^th percentiles of the bootstrap distribution. All statistical analyses were performed using IBM SPSS version 21 (SPSS Inc., Chicago, IL). For all tests, a p value (two-tailed) less than 0.05 was considered statistically significant.

Results

Demographic characteristics

Overall, 461 patients (median age, 73 years; IQR 67–80; male 42.3%) admitted for spinal (n=150; 32.5%), hip (n=111; 24.1%), and knee (n=200; 43.4%) surgery were enrolled, and 37 of them (8.0%, 37/461) developed post-operative delirium. Table 1 describes the demographic characteristics of all study subjects. Comparisons of the demographic characteristics between those who developed post-operative delirium, and those who did not found that patients with post-operative delirium had significantly older age (48.6% vs. 23.8% in older group, p<0.001), poorer baseline physical function (24.3% vs. 5.0% of ADL scores <75, p<0.001), poorer cognitive function (59.5% vs. 37.0% of MMSE scores<24, p<0.007), and higher disease burden (40.5% vs. 16.5% of CCI≥2, p<0.001). On average, post-operative delirium persisted for 2.7±2.5 days and the hospital length of stay of patients with post-operative delirium tended to be longer (10 [7–12] vs. 7 [6–9] days, p<0.001).

Table 1.

Demographic Data, Surgery-Related Factors, and Comprehensive Geriatric Assessment Variables for 461 Study Subjects

	n=461 (% or median+IQR)
Demographic variables
Age, years	73 (67–80)
Educational level	6 (0–9)
Male gender	195 (42.3%)
BMI	27 (24–29)
CCI	0 (0–1.0)
Surgery categories
Spine	150 (32.5%)
Hip	111 (24.1%)
Knee	200 (43.4%)
Geriatric assessment variables
Polypharmacy	203 (44.0%)
Visual impairment	306 (66.4%)
Hearing impairment	86 (18.7%)
ADL (Barthel Index)	90 (90–100)
MMSE	25 (22–28)
GDS-15	53 (11.5%)
Risk of malnutrition (MNA-SF≥12)	57 (12.4%)
Pain VAS score	5 (4–6)
Routine pre- & perioperative variables
Anesthesia method
Spinal anesthesia	210 (45.6%)
General anesthesia	251 (54.4%)
ASA class
ASA: 1 and 2	372 (80.7%)
ASA: 3	89 (19.3%)
Blood transfusion	102 (22.1%)
WBC	6590 (5630–7690)
Hgb <12	116 (25.2%)
Sodium (Na)	141 (139–142)
Potassium (K)	4.1(3.9–4.4)
BUN/Cr≥18	236 (51.2%)

IQR, interquartile range; BMI, body mass index; CCI, Charlson's Comorbidity Index; ADL, Activities of Daily Living; MMSE, Mini-Mental State Examination; MNA-SF, Mini-Nutritional Assessment–Short Form; GDS-15, Geriatric Depression Score; VAS, Visual Analogue Scale; ASA, American Society of Anesthesiologists; WBC white blood cells; BUN/Cr, blood urea nitrogen to creatinine ratio.

Prediction models for post-operative delirium

Table 2 summarizes the percentage of patients with each risk factor and the crude odds ratio with 95% confidence interval (95% CI) for post-operative delirium. Variables entered in the final prediction models included clinical characteristics (age, gender, CCI≥2, and type of surgery), surgery-related factors (ASA class, blood transfusion, hemoglobin <12 grams/dL, and serum blood urea nitrogen/serum creatinine ≥18), and results of CGA (poly-pharmacy, hearing impairment, impairment of cognition, risk of malnutrition, and impairment of baseline ADL ≤75).

Table 2.

Comparisons of Demographic Data, Surgery-Related Factors, and Comprehensive Geriatric Assessment Variables Between Subjects with and without Post-Operative Delirium

	Post-operative delirium (+) n=37 (% or median±IQR)	Post-operative delirium (–) n=424 (% or median±IQR)	OR	CI	p value	In multivariate model
Demographic variables
Age≥75	18 (48.6%)	101 (23.8%)	3.030	1.531–5.994	0.001	Yes
Male gender	24 (64.9%)	171 (40.3%)	2.731	1.353–5.513	0.005	Yes
Educational level			1.033	0.967–1.104	0.329
BMI	26.0 (23.0–28.6)	27.0 (24.0–29.0)	0.973	0.895–1.058	0.522
CCI≥2	15 (40.5%)	70 (16.5%)	3.448	1.704–6.975	0.001	Yes
Surgery categories						Yes
Spine	20 (54.1%)	130 (30.7%)	3.265	1.441–7.396	0.005
Hip	8 (21.6%)	103 (24.3%)	1.648	0.617–4.401	0.319
Knee	9 (24.3%)	191 (45.0%)	Reference	Reference	Reference
Comprehensive geriatric assessment variables
Poly-pharmacy	22 (59.5%)	181 (42.7%)	1.969	0.994–3.902	0.052	Yes
Visual impairment	29 (78.4%)	277 (65.3%)	1.924	0.858–4.315	0.112
Hearing impairment	14 (37.8%)	72 (17.0%)	2.976	1.461–6.060	0.003	Yes
ADL (Barthel Index)≤75	9 (24.3%)	21 (5.0%)	6.168	2.585–14.720	<0.001	Yes
Impairment of cognition (MMSE<24)	22 (59.5%)	157 (37.0%)	2.494	1.257–4.949	0.009	Yes
GDS-15≥5	30 (81.1%)	378 (89.2%)	1.917	0.797–4.612	0.146
Risk of malnutrition (MNA-SF≥12)	13 (35.1%)	44 (10.4%)	4.678	2.224–9.840	<0.001	Yes
Pain VAS score	5 (4–6.5)	5 (4–6)	1.197	0.962–1.489	0.107
Routine pre- and peri-operative variables
Anesthesia method			1.602	0.794–3.231	0.188
Spinal anesthesia	13 (35.1%)	197 (46.5%)
General anesthesia	24 (64.9%)	227 (53.5%)
ASA class			2.832	1.393–5.759	0.003	Yes
ASA: 1 and 2	23 (62.2%)	349 (82.3%)
ASA: 3	14 (37.8%)	75 (17.7%)
Blood transfusion	18 (48.6%)	84 (19.8%)	3.835	1.928–7.626	<0.001	Yes
WBCs	5910 (4745–7555)	6630 (5670–7698)	0.869	0.714–1.059	0.164
Hgb <12	14 (37.8%)	102 (24.1%)	1.922	0.954–3.872	0.068	Yes
Sodium (Na)	140 (138–143)	141 (139–142)	0.982	0.894–1.079	0.712
Potassium (K)	4.2 (4.0–4.4)	4.1 (3.9–4.4)	1.456	0.664–3.189	0.348
BUN/Cr ≥18	15 (40.5%)	221 (52.1%)	0.626	0.316–1.240	0.180	Yes

IQR, inter-quartile range; OR, odds ratio; CI, confidence interval; BMI, body mass index; CCI, Charlson's Comorbidity Index; ADL, Activities of Daily Living; MMSE, Mini-Mental State Examination; GDS-15, Geriatric Depression Score; MNA-SF, Mini-Nutritional Assessment–Short Form; ASA, American Society of Anesthesiologists; WBCs, white blood cells; Hgb, hemoglobin; BUN/Cr, blood urea nitrogen to creatinine ratio.

Table 3 shows the adjusted OR and risk scores for both multivariate models. Four variables (age >75 years, CCI≥2, blood transfusion, and ASA class) were associated with post-operative delirium in model 1, and seven variables (age ≥75 years, male gender, CCI≥2, blood transfusion, risk of malnutrition, cognitive impairment, and type of surgery) were independently associated with post-operative delirium in model 2. Only six independent risk factors were selected for the development of prediction model 2, after the bootstrap process excluded age ≥75 years. The individual risk score of each risk factor derived from the rounded-off adjusted relative risk values and the summation of total risk score for each model are also shown in Table 3.

Table 3.

Multivariate Logistic Regression Analysis of Independent Risk Factors for Postoperative Delirium and the Development of Risk Scoring Systems

Prediction model	Adjusted OR	CI	Bootstrapping selection	Risk scores
Prediction model 1
Age ≥75	3.633	1.582–8.340	Yes	3
CCI ≥2	2.879	1.338–6.196	Yes	2
Blood transfusion	4.211	2.009–8.826	Yes	4
ASA 3	2.466	1.143–5.319	Yes	2
Total risk score				11
Prediction model 2
Age ≥75	2.943	1.230–7.039	No
Male gender	2.84	1.202–6.706	Yes	2
CCI ≥2	3.09	1.405–6.795	Yes	3
MMSE <24	3.181	1.410–7.173	Yes	3
Risk of malnutrition (MNA)	3.261	1.420–7.488	Yes	3
Blood transfusion	2.841	1.239–6.515	Yes	2
Type of surgery			Yes
Knee and hip	Reference	Reference		0
Spine	2.313	1.010–5.295		2
Total risk score				15

Model 1: Demographic data+routine pre- and peri-operative survey (ASA group, blood transfusion, Hgb<12 grams/L, BUN/Cr≥18).

Model 2: Model 1+ non-official CGA assessment (poly-pharmacy, hearing impairment, cognitive impairment based on MMSE, risk of malnutrition based on MNA, ADL≥75).

OR, odds ratio; CI, confidence interval; CCI, Charlson's Comorbidity Index; MMSE, Mini-Mental State Examination; MNA, Mini-Nutritional Assay; ASA, American Society of Anesthesiologists; Hgb, hemoglobin; BUN/Cr, blood urea nitrogen to creatinine ratio; ADL, Activities of Daily Living.

Overall, both models effectively predicted post-operative delirium (Table 4). The AUC derived from the summation of risk scores was 0.763 in model 1 (95% CI 0.690–0.835), and 0.814 in model 2 (95% CI 0.746–0.882). In model 1, the cutoff point ≥5 had a sensitivity of 62.2%, specificity of 77.4%, and Youden Index of 0.396, whereas the cutoff point ≥4 and ≥6 had a sensitivity of 64.9% and 51.4%, specificity of 65.8% and 87.3%, and Youden Index of 0.307 and 0.387. In model 2, the cutoff point ≥5 had a sensitivity of 83.8%, specificity of 65.3%, and Youden Index of 0.491, whereas the cutoff point ≥6 and ≥7 had a sensitivity of 64.9% and 56.8%, specificity of 76.9% and 84.2%, and Youden Index of 0.418 and 0.410. Therefore, the cut-points ≥5 in model 1 and model 2 were selected as the criteria for post-operative delirium prediction because of their higher Youden Index. The positive and negative predictive values and likelihood ratios for both models are also listed in Table 4.

Table 4.

Sensitivity, Specificity, and Youden Index at Different Cutoff Points of Both Models

Cutoff points	AUC	CI	Sensitivity N	%	Specificity N	%	PPV N	%	NPV N	%	LR+	LR−	Youden index
Model 1 (0–11)	0.763	0.690–0.835
≥4			24/37	64.9	279/424	65.8	24/169	14.2	279/292	95.5	1.898	0.494	0.307
≥5			23/37	62.2	328/424	77.4	23/119	19.3	328/342	95.9	2.752	0.374	0.396
≥6			19/37	51.4	370/424	87.3	19/73	26.0	370/388	95.4	4.047	0.207	0.387
Model 2 (0–15)	0.814	0.746– 0.882
≥5			31/37	83.8	277/424	65.3	31/178	17.4	277/283	97.9	2.415	0.682	0.491
≥6			24/37	64.9	326/424	76.9	24/122	19.7	326/339	96.2	2.810	0.397	0.418
≥7			21/37	56.8	357/424	84.2	21/88	23.9	357/373	95.7	3.595	0.268	0.410

AUC, area under the curve; CI, confidence interval; PPV, positive predictive value; NPV, negative predictive value; LR+ and LR−, positive and negative likelihood ratios.

Prediction of long-term health outcomes

For long-term health outcomes, only data from patients enrolled in the first year (n=204) were analyzed to ensure at least 1 year of follow-up. For analysis of adverse outcomes, nursing home admission was combined with mortality as a composite outcome indicator to avoid inferential errors resulting from the fact that patients who died during the follow-up period would never be placed in nursing homes.⁵⁰ Table 5 shows the effectiveness of both models in predicting post-operative delirium and long-term health outcomes. Both models significantly predicted adverse outcomes (nursing home admission and mortality) during the 1^st and 3^rd month of follow-up. Moreover, in addition to the composite outcome, model 2 also significantly predicted post-discharge functional decline (ADL decline in 6 months, OR=2.462, 95% CI 1.081–5.605; ADL decline in 12 months, OR=2.533, 95% CI 1.115–5.755; IADL decline in 3 months, OR=2.554, 95% CI 1.320–4.943; IADL decline in 6 months, OR=3.176, 95% CI 1.577–6.397; IADL decline in 12 months, OR=2.904, 95% CI 1.456–5.792). The final model was validated internally through bootstrapping and performed well (Table 5). Moreover, the mean total number of hospital days and the mean duration of post-operative delirium were longer among subjects in the higher-risk group (total hospital days, 9.93±4.88 vs. 7.89±3.57 days in model 1, p value<0.001; 9.46±4.24 vs. 7.76±3.79 days in model 2, p value<0.001; duration of post-operative delirium, 0.61±1.79 days vs. 0.08±0.47 days in model 1, p value <0.001; 0.45±1.20 days vs. 0.07±0.85 days in model 2, p value<0.001).

Table 5.

Clinical Outcomes Undergoing Elective Orthopedic Surgery Based on Two Prediction Models

	Predicting model 1 with cutoff point ≥5					Predicting model 2 with cutoff point ≥5
Outcome measures	Yes	No	OR	95% CI	Bootstrapping OR 95% CI ^#a	Yes	No	OR	95% CI	Bootstrapping OR 95% CI ^#a
Postoperative delirium (n=461)
Yes	23/119 (19.3%)	14/342 (4.1%)	5.613	2.781–11.329	2.781–11.329	32/173 (18.5%)	5/288 (1.7%)	9.736	3.971–23.869	4.433–29.524
Post-discharge outcomes^&b
(Total numbers=204)
ADL decline
At 3^rd month (n=188)	9/45 (20.0%)	20/143 (14.0%)	1.538	0.644–3.670	0.554–3.747	13/64 (20.3%)	16/124 (12.9%)	1.867	0.839–4.152	0.846–4.148
At 6^th month (n=169)	9/40 (22.5%)	19/129 (14.7%)	1.681	0.692–4.083	0.593–4.252	13/57 (22.8%)	15/112 (13.4%)	2.462	1.081–5.605	1.096–5.561
At 12^th month (n=175)	10/40 (25.0%)	18/135 (13.3%)	2.167	0.907–5.176	0.794–4.849	14/57 (24.6%)	14/118 (11.9%)	2.533	1.115–5.755	1.020–6.443
IADL decline
At 3^rd month (n=181)	15/43 (34.9%)	38/138 (27.5%)	1.410	0.679–2.925	0.649–3.374	24/61 (39.3%)	29/120 (24.2%)	2.554	1.320–4.943	1.298–5.166
At 6^th month (n=176)	14/40 (35.0%)	31/136 (22.8%)	1.824	0.850–3.913	0.753–4.310	22/58 (37.9%)	23/118 (19.5%)	3.176	1.577–6.397	1.596–6.465
At 12^th month (n=175)	14/40 (35.0%)	33/135 (24.4%)	1.664	0.779–3.556	0.653–3.706	22/57 (38.6%)	25/118 (21.2%)	2.904	1.456–5.792	1.479–5.638
Nursing home admission and death after discharge
At 1^st month (n=191)	6/45 (13.3%)	1/146 (0.7%)	22.308	2.608–190.805	3.596–42.085	7/64 (10.9%)	0/127 (0.0%)	^c
At 3^rd month (n=187)	5/44 (11.4%)	1/143 (0.7%)	18.205	2.066–160.417	2.662–36.799	6/63 (9.5%)	0/124 (0.0%)	^c

1000 random bootstrap samples from the original cohort for cross-validation.

For studying post-operative delirium, all patients in two cohorts were enrolled. For studying post-discharge outcomes, patients in the 1^st year cohort were enrolled.

All patients admitted to nursing homes and mortalities at the 1^st and 3^rd months after discharge are grouped in the group with scores ≥5 in prediction model 2.

Discussion

This prospective study identified risk factors for post-operative delirium among older patients receiving elective orthopedic surgery and also developed prediction models for post-operative delirium on the basis of these risk factors. Notably, the prevalence of post-operative delirium in this study was lower than that in a previous study of hip fractures (8% vs. 9–15%),⁵¹ but was similar to that in another study of patients receiving elective orthopedic surgery.⁵² Results of multivariate logistic regression analysis identified four independent predictors (age ≥75 years, CCI≥2, receiving blood transfusion, and ASA class ≥3) of post-operative delirium in model 1, and six predictors (male, CCI≥2, MMSE<24, risk of malnutrition, receiving blood transfusion, and type of surgery) in model 2. In this study, we have developed and successfully cross-validated two prediction models for post-operative delirium. Although both prediction models were highly effective in predicting post-operative delirium, model 2 was significantly more effective than model 1 and good success in predicting both medical and functional outcomes. Therefore, using CGA in the development of the prediction model for post-operative delirium was superior to the conventional approach.

In this study, we developed prediction models (based on demographic characteristics, pre- and peri-operative factors, and results of CGA) that were more comprehensive than previously developed. Both prediction models showed good predictive power for post-operative delirium (AUC 0.763 in model 1 and 0.814 in model 2). Inouye et al. developed a prediction model for delirium on the basis of medical risk factors, including demographic factors, physical function, cognitive function, biomedical factors, and psychosocial factors.²¹ The final model consisted of low MMSE score, visual impairment, dehydration, and severity of acute illness.²¹ However, the prediction model assigned one point to each of the risk factors, thereby disregarding the differences in the weights of individual risk factors. Kalisvaart et al. validated Inouye's prediction model for elderly patients receiving hip surgery, but were unable to use visual impairment and dehydration to predict post-operative delirium.²² Moreover, some important risk factors, such as age, type of admission, and surgery-related factors, were not included in Inouye's model. Results of this study were highly compatible with those in the above-mentioned studies. Although the risk factors of post-operative delirium in this study were not significantly different from other studies,^13,17

–20 the only surgery-related factor predictive for post-operative delirium in patients undergoing elective orthopedic surgery was blood transfusion. Vochteloo et al. reported that allogeneic blood transfusion was associated with post-discharge mortality in the first 3 months, prolonged hospital length-of-stay, and the occurrence of post-operative delirium.⁵³ The present study also identified blood transfusion as a risk factor for adverse health outcomes.

The usefulness of pre-operative CGA in predicting post-operative outcomes in older surgical patients with relatively stable medical conditions has been demonstrated in a previous study.²⁴ Nevertheless, most CGA-related risk factors identified in this study (i.e., poly-pharmacy, malnutrition, symptoms of pain, and burden of illness) usually were not included in previous studies because CGA is not routinely performed in orthopedic wards.^21
–23 Using patients similar to our patients, Freter et al. developed the Delirium Elderly At-Risk (DEAR) instrument to predict post-operative delirium among elective orthopedic patients and improve measures to prevent post-operative delirium.²⁶ Although the DEAR instrument included pre-operative risk factors (cognitive impairment, sensory impairment, ADL dependence, substance abuse, and older age), only cognitive impairment and substance abuse were identified as independent risk factors by their study.

The current study evaluated, as risk factors, the severity of disease burden and important pre- and peri-operative factors, which were not included in the DEAR instrument.⁵⁴ Malnutrition has been reported to be associated with surgical and functional outcomes, as well as increased mortality rate, prolonged length of hospital stay, higher risk of nursing home admission, and higher risk of functional dependence.^55

–61 However, the DEAR instrument also did not include malnutrition assessment. Unlike the DEAR instrument, our prediction model 2 did not include age but rather the presence of geriatric syndromes, which may be used as a proxy for age. This study clearly showed the great diversity of risk factors for post-operative delirium, suggesting that an integrated and comprehensive approach is needed in the development of prediction models. Therefore, routine CGA in older patients receiving elective orthopedic surgery is very useful in risk factor identification and the design of post-operative delirium prevention programs.

Despite efforts to avoid them, there were several limitations. First, the results of this study have not yet been validated in an independent external cohort. However, the effectiveness of our prediction model has been validated by bootstrapping procedures, which may partly overcome the lack of external validation. Second, the low prevalence of post-operative delirium in this study may bias the statistical analysis. However, in this study, delirium was assessed by trained primary nurses and research nurses, and its diagnosis was confirmed by a senior psychiatrist as soon as the symptoms were identified. The strict diagnostic algorithm for delirium may explain the low prevalence of post-operative delirium; yet the prevalence of delirium was higher in this study than in a previous report from Taiwan.³⁸ Third, this study enrolled patients receiving certain types of elective orthopedic surgery, which may not be representative of all orthopedic surgical interventions. Despite the above-mentioned limitations, the prediction model developed in this study was extremely effective in predicting post-operative delirium as well as long-term health and functional outcomes.

Conclusions

The prevalence of older patients receiving elective orthopedic surgery was approximately 8% in this study, and a prediction model developed on the basis of risk factors (demographic characteristics, surgery-related factors) with/without CGA results was highly effective in predicting post-operative delirium and adverse medical outcomes, such as post-operative mortality, nursing home admission, duration of post-operative delirium, and total hospital days. Adding the CGA results to the prediction model increased its ability to predict functional outcome 6 and 12 months after discharge. Results of this study suggested the usefulness of the orthogeriatrics program in elective orthopedic surgery for delirium prevention program development in the clinical services. Preventing delirium in older patients receiving orthopedic surgery is highly dependent on identification of high-risk subjects, and the highly effective prediction model developed in this study may facilitate the provision of care to these patients.

Footnotes

Acknowledgment

The study was supported by the Veteran Affairs Commission of Taiwan. The study group thanks all staff in the Orthopedics Department for their valuable assistance in data collection.

Author Disclosure Statement

No competing financial interests exist.

References

Chen

. Meeting the challenges of elder care in Taiwan's aging society. J Clin Gerontol Geriatr, 2010; 1:2–4.

Liu

, Leung

. Predicting adverse postoperative outcomes in patients aged 80 years or older. J Am Geriatr Soc, 2000; 48:405–412.

Leung

, Dzankic

. Relative importance of preoperative health status versus intraoperative factors in predicting postoperative adverse outcomes in geriatric surgical patients. J Am Geriatr Soc, 2001; 49:1080–1085.

Oresanya

, Lyons

, Finlayson

. Preoperative assessment of the older patient: A narrative review. JAMA, 2014; 311:2110–2120.

Inouye

, Westendorp

, Saczynski

. Delirium in elderly people. Lancet, 2014; 383:911–922.

Chu

, Liang

, Lin

, Chow

, Pan

, Chou

, Lu

. Biomarkers of delirium: Well evidenced or not?. J Clin Gerontol Geriatr, 2011; 2:100–104.

Élie

, Rousseau

, Cole

, Primeau

, McCusker

, Bellavance

. Prevalence and detection of delirium in elderly emergency department patients. CMAJ, 2000; 163:977–981.

Inouye

, Foreman

, Mion

, Katz

, Cooney

Jr.

Nurses' recognition of delirium and its symptoms: Comparison of nurse and researcher ratings. Arch Intern Med, 2001; 161:2467–2473.

Marcantonio

, Flacker

, Michaels

, Resnick

. Delirium is independently associated with poor functional recovery after hip fracture. J Am Geriatr Soc, 2000; 48:618–624.

10.

Robinson

, Raeburn

, Tran

, Angles

, Brenner

, Moss

. Postoperative delirium in the elderly: Risk factors and outcomes. Ann Surg, 2000; 249:173–178.

11.

Gottesman

, Grega

, Bailey

, Pham

, Zeger

, Baumgartner

, Selnes

, McKhann

. Delirium after coronary artery bypass graft surgery and late mortality. Ann Neurol, 2010; 67:338–344.

12.

Witlox

, Eurelings

, de Jonghe

, Kalisvaart

, Eikelenboom

, van Gool

. Delirium in elderly patients and the risk of postdischarge mortality, institutionalization, and dementia: A meta-analysis. JAMA, 2010; 304:443–451.

13.

Sanders

, Pandharipande

, Davidson

, Ma

, Maze

. Anticipating and managing postoperative delirium and cognitive decline in adults. BMJ, 2011; 343:d4331.

14.

Inouye

, Bogardus

Jr , Charpentier

, Leo-Summers

, Acampora

, Holford

, Cooney

Jr . A multicomponent intervention to prevent delirium in hospitalized older patients. N Engl J Med, 1999; 340:669–676.

15.

Marcantonio

, Flacker

, Wright

, Resnick

. Reducing delirium after hip fracture: A randomized trial. J Am Geriatr Soc, 2001; 49:516–522.

16.

Neuman

, Silber

, Magaziner

, Passarella

, Mehta

, Werner

. Survival and functional outcomes after hip fracture among nursing home residents. JAMA Intern Med, 2014; 174:1273–1280.

17.

Schor

, Levkoff

, Lipsitz

, Reilly

, Cleary

, Rowe

, Evans

. Risk factors for delirium in hospitalized elderly. JAMA, 1992; 267:827–831.

18.

Levkoff

, Evans

, Liptzin

, Cleary

, Lipsitz

, Wetle

, Reilly

, Pilgrim

, Schor

, Rowe

. Delirium. The occurrence and persistence of symptoms among elderly hospitalized patients. Arch Intern Med, 1992; 152:334–340.

19.

, Chang

, Chen

. Overview of studies related to geriatric syndrome in Taiwan. J Clin Gerontol Geriatr, 2012; 3:14–20.

20.

Pompei

, Foreman

, Rudberg

, Inouye

, Braund

, Cassel

. Delirium in hospitalized older persons: Outcomes and predictors. J Am Geriatr Soc, 1994; 42:809–815.

21.

Inouye

, Viscoli

, Horwitz

, Hurst

, Tinetti

. A predictive model for delirium in hospitalized elderly medical patients based on admission characteristics. Ann Intern Med, 1993; 119:474–481.

22.

Kalisvaart

, Vreeswijk

, de Jonghe

, van der Ploeg

, van Gool

, Eikelenboom

. Risk factors and prediction of postoperative delirium in elderly hip-surgery patients: Implementation and validation of a medical risk factor model. J Am Geriatr Soc, 2006; 54:817–822.

23.

Van den Boogaard

, Schoonhoven

, Maseda

, Plowright

, Jones

, Luetz

, Sackey

, Jorens

, Aitken

, van Haren

, Donders

, van der Hoeven

, Pickkers

. Recalibration of the delirium prediction model for ICU patients (PRE-DELIRIC): A multinational observational study. Intensive Care Med, 2014; 40:361–369.

24.

Partridge

, Harari

, Martin

, Dhesi

. The impact of pre-operative comprehensive geriatric assessment on postoperative outcomes in older patients undergoing scheduled surgery: A systematic review. Anaesthesia., 2014; 69:8–16.

25.

Harari

, Hopper

, Dhesi

, Babic-Illman

, Lockwood

, Martin

. Proactive care of older people undergoing surgery (‘POPS’): Designing, embedding, evaluating and funding a comprehensive geriatric assessment service for older elective surgical patients. Age Ageing, 2007; 36:190–196.

26.

Freter

, Dunbar

, MacLeod

, Morrison

, MacKnight

, Rockwood

. Predicting post-operative delirium in elective orthopaedic patients: The Delirium Elderly At-Risk (DEAR) instrument. Age Ageing, 2005; 34:169–171.

27.

Liang

, Chu

, Chou

, Lin

, Lu

, Hsu

, Chen

. Interrelationship of postoperative delirium and cognitive impairment and their impact on the functional status in older patients undergoing orthopaedic surgery: a prospective cohort study. PLoS One, 2014; 9:e110339.

28.

Charlson

, Pompei

, Ales

, MacKenzie

. A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation. J Chronic Dis, 1987; 40:373–383.

29.

Liao

, Yeh

, Ko

, Luoh

, Lu

. Geriatric depression scale (GDS)—a preliminary report of reliability and validity for Chinese version. Changhua Med J, 1995; 1:11–17. (in Chinese)

30.

Nyunt

, Fones

, Niti

, Ng

. Criterion-based validity and reliability of the Geriatric Depression Screening Scale (GDS-15) in a large validation sample of community-living Asian older adults. Aging Ment Health, 2009; 13:376–382.

31.

Kaiser

, Bauer

, Ramsch

, Uter

, Guigoz

, Cederholm

, Thomas

, Anthony

, Charlton

, Maggio

, Tsai

, Grathwohl

, Vellas

, Sieber

; MNA-International Group. Validation of the Mini Nutritional Assessment Short-Form (MNA-SF): A practical tool for identification of nutritional status. J Nutr Health Aging, 2009; 13:782–788.

32.

Folstein

, Folstein

, McHugh

. Mini-mental status: A practical method for grading the cognitive state of patients for the clinical use. J Psychiatr Res, 1975; 12:189–198.

33.

Huskisson

. Measurement of pain. Lancet, 1974; 2:1127–1131.

34.

Collin

, Wade

, Davies

, Horne

. The Barthel ADL Index: A reliability study. Int Disabil Stud, 1988; 10:61–63.

35.

Lawton

, Brody

. Assessment of older people: Self-maintaining and instrumental activities of daily living. Gerontologist, 1969; 9:179–186.

36.

Inouye

, Van Dyck

, Alessi

, Balkin

, Siegal

, Horwitz

. Clarifying confusion: The confusion assessment method. A new method for detection of delirium. Ann Intern Med, 1990; 113:941–948.

37.

Bellelli

, Mazzola

, Morandi

, Bruni

, Carnevali

, Corsi

, Zatti

, Zambon

, Corrao

, Olofsson

, Gustafson

, Annoni

. Duration of postoperative delirium is an independent predictor of 6-month mortality in older adults after hip fracture. J Am Geriatr Soc, 2014; 62:1335–1340.

38.

Maldonado

. Delirium in the acute care setting: Characteristics, diagnosis and treatment. Crit Care Clin, 2008; 24:657–722, vii.

39.

Núñez

, Núñez

, Fácila

, Bertomeu

, Llàcer

, Bodí

, Sanchis

, Sanjuán

, Blasco

, Consuegra

, Martínez

, Chorro

. Prognostic value of the Charlson comorbidity index at 30 days and 1 year after acute myocardial infarction. Rev Esp Cardiol (Engl Ed), 2004; 57:842–849.

40.

Wang

, Chew

, Kung

, Chung

, Lee

. The use of Charlson comorbidity index for patients revisiting the emergency department within 72 hours. Chang Gung Med J, 2007; 30:437–444.

41.

Wang

, Lin

, Tzao

, Lee

, Huang

, Hsu

. Comparison of Charlson comorbidity index and Kaplan-Feinstein index in patients with stage I lung cancer after surgical resection. Eur J Cardiothorac Surg, 2007; 32:877–881.

42.

Guzzo

, Dluzniewski

, Orosco

, Platz

, Partin

, Han

. Prediction of mortality after radical prostatectomy by Charlson comorbidity index. Urology, 2010; 76:553–557.

43.

Supervía

, Aranda

, Márquez

, Aguirre

, Skaf

, Gutiérrez

. Predicting length of hospitalisation of elderly patients, using the Barthel Index. Age Ageing, 2008; 37:339–342.

44.

Uyttenboogaart

, Stewart

, Vroomen

, De Keyser

, Luijckx

. Optimizing cutoff scores for the Barthel index and the modified Rankin scale for defining outcome in acute stroke trials. Stroke, 2005; 36:1984–1987.

45.

Seymour

, Henschke

, Cape

, Campbell

. Acute confusional states and dementia in the elderly: The role of dehydration/volume depletion, physical illness and age. Age Ageing, 1980; 9:137–146.

46.

Andrès

, Serraj

, Federici

, Vogel

, Kaltenbach

. Anemia in elderly patients: New insight into an old disorder. Geriatr Gerontol Int, 2013; 13:519–527.

47.

Efron

, Tibshirani

. An Introduction to the Bootstrap. New York: Chapman & Hall, 1993.

48.

Hanley

, McNeil

. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology, 1982; 143:29–36.

49.

Biggerstaff

. Comparing diagnostic tests: A simple graphic using likelihood ratios. Stat Med, 2000; 19:649–663.

50.

Gent

, Sackett

. The qualification and disqualification of patients and events in long-term cardiovascular clinical trials. Thromb Haemost, 1979; 41:123–134.

51.

Rudolph

, Marcantonio

. Review articles: Postoperative delirium: Acute change with long-term implications. Anesth Analg, 2011; 112:1202–1211.

52.

Dai

, Lou

, Yip

, Huang

. Risk factors and incidence of postoperative delirium in elderly Chinese patients. Gerontology, 2000; 46:28–35.

53.

Vochteloo

, Borger van der Burg

, Mertens

, Niggebrugge

, de Vries

, Tuinebreijer

, Bloem

, Nelissen

, Pilot

. Outcome in hip fracture patients related to anemia at admission and allogeneic blood transfusion: An analysis of 1262 surgically treated patients. BMC Musculoskelet Disord, 2011; 12:262.

54.

Noimark

. Predicting the onset of delirium in the post-operative patient. Age Ageing, 2009; 38:368–373.

55.

Gibbs

, Cull

, Henderson

, Daley

, Hur

, Khuri

. Preoperative serum albumin level as a predictor of operative mortality and morbidity: Results from the National VA Surgical Risk Study. Arch Surg, 1999; 134:36–42.

56.

Schalk

, Visser

, Deeg

, Bouter

. Lower levels of serum albumin and total cholesterol and future decline in functional performance in older persons: The Longitudinal Aging Study Amsterdam. Age Ageing, 2004; 33:266–272.

57.

Sullivan

, Sun

, Walls

. Protein-energy undernutrition among elderly hospitalized patients: A prospective study. JAMA, 1999; 281:2013–2019.

58.

Guigoz

. The Mini Nutritional Assessment (MNA) review of the literature—What does it tell us?. J Nutr Health Aging, 2006; 10:466–485; discussion 485–487.

59.

Drevet

, Bioteau

, Mazière

, Couturier

, Merloz

, Tonetti

, Gavazzi

. Prevalence of protein-energy malnutrition in hospital patients over 75 years of age admitted for hip fracture. Orthop Traumatol Surg Res, 2014; 100:669–674.

60.

Koval

, Maurer

, Su

, Aharonoff

, Zuckerman

. The effects of nutritional status on outcome after hip fracture. J Orthop Trauma, 1999; 13:164–169.

61.

Kim

, Moon

, Lim

, Yoon

, Min

, Lee

, Park

. Prediction of survival, second fracture, and functional recovery following the first hip fracture surgery in elderly patients. Bone, 2012; 50:1343–1350.