A Novel Nomogram Developed Based on Preoperative Immune Inflammation-Related Indicators for the Prediction of Postoperative Delirium Risk in Elderly Hip Fracture Cases: A Single-Center Retrospective Cohort Study

Introduction

Because of steadily increasing population aging globally, hip fracture cases are expected to increase to about 6.1 million by 2050, indicating a yearly escalation of 1–3%,^1,2 which represents an important public health challenge in elderly individuals.³ Hip fractures impose a significant economic burden on the society and are responsible for disability and multiple human ailments.⁴ They are mainly treated by surgical operations, which unfortunately can induce many complications affecting distinct organs.^5,6 Delirium is a serious complication occurring in elderly individuals administered hip fracture surgery, with an incidence of 10–62%.^7,8

Postoperative delirium (POD) refers to acute changes in the fluctuating mental state of patients within 7 days after anesthesia or before discharge, primarily manifested as unclear consciousness, inattention, psychomotility disorders and sleep-wake cycle disorders, with rapid onset and short disease course, as well as potential missed diagnosis and misdiagnosis.^9,10 POD is considered a major complication in geriatric individuals administered surgery.¹¹ Despite the transient and reversible nature of POD in the vast majority of patients, it may result in neuropsychiatric conditions, prolonged hospitalization, elevated treatment cost and reduced ability to perform daily living activities, particularly in elderly individuals. Additionally, POD has potential associations with long-term cognitive impairment, dementia and even elevated mortality.^12–16 Although delirium has a high incidence, most cases are not detected in clinic.^17,18 Consequently, identifying molecular markers applicable for early detection of delirium is highly important both for patients and healthcare professionals.

The pathogenetic mechanism of delirium is not fully defined, but inflammation is thought to play a role in delirium onset.^19–21 Increasing evidence suggests elevated neutrophils and decreased lymphocytes in elderly delirium cases.²² New non-specific inflammatory indicators have been developed, eg, monocyte-to-lymphocyte (MLR), neutrophil-to-lymphocyte (NLR) and platelet-to-lymphocyte (PLR) ratios and systemic immune-inflammatory (SII) and systemic inflammatory response (SIRI) indexes.^23,24 The latter parameters better mirror systemic inflammatory response, with high availability and low cost. They play crucial roles in multiple cancers, autoimmune disorders, and cardiovascular diseases.^25–28 For example, SIRI, a ratio considering neutrophil, monocyte, and lymphocyte counts, is a critical biomarker of cancer development and progression. Recently, Lu et al¹¹ also demonstrated the SIRI reflects the degree of chronic inflammation in elderly patients after hip arthroplasty. Additionally, hypoalbuminemia independently predicts POD in surgically treated patients, suggesting malnutrition is associated with POD.^29,30 Albumin (Alb), an important player in acute inflammation, can be utilized to examine the nutritional status of individuals administered surgery. Hu and collaborators³¹ reported that Alb-derived indexes based on inflammation and nutrition may be employed for POD prediction in geriatric individuals administered THA. Such albumin-related inflammatory and nutritional parameters include neutrophil-to-albumin (NAR) and CRP-to-albumin (CAR) ratios and prognostic nutritional index (PNI).

However, few studies have examined whether SII, SIRI, NLR, MLR, PLR, NAR, CAR and PNI are involved in POD in geriatric individuals with hip fractures. The nomogram model is a commonly employed prognostic tool in medicine to generate a single numerical probability of a given clinical event through integration of different prognostic variates, meeting the need for integrated biological and clinical models and promoting personalized medicine to assist clinical decision making. The present work aimed to determine potential risk factors for POD in geriatric hip fracture cases, with the goal of developing a nomogram to predict the risk of POD with high accuracy in geriatric individuals with hip fractures.

Materials and Methods

Patient Data

This observational, analytical, retrospective cohort case-control study collected the clinical data of geriatric hip fracture cases administered surgical procedures, including internal fixation surgery, total hip arthroplasty or artificial femoral head replacement, in the First People’s Hospital of Neijiang, between January 2018 and December 2023. Data collection was performed in an independent manner by the first and second authors, with any discrepancy resolved by consensual discussion. The present study followed the 1964 helsinki Declaration and was approved by the Ethics Committee of the First People’s Hospital of Neijiang, who required no informed consent because of the retrospective nature of the current analysis.

Diagnosis was confirmed by X-ray and computed tomography (CT) with 3D reconstruction. Figure 1 presents the study flowchart. Inclusion criteria were: (1) diagnosis of hip fractures such as femoral neck, intertrochanteric and subtrochanteric fractures and first treatment by internal fixation surgery, total hip arthroplasty or artificial femoral head replacement; (2) Han nationality; (3) age ≥60 years; (4) traumatic factors including fall, body twisting, weight lifting fail, car accident, etc.; (5) availability of clinicodemographic data and laboratory findings. Exclusion criteria were: (1) multiple or pathological hip fractures; (2) neurological and psychiatric diseases; (3) postoperative infections; (4) treatment with antipsychotic medications in the last 3 months; (5) incomplete data.

Figure 1 The enrollment flowchart.

POD Diagnosis

POD was diagnosed per the Confusion Assessment Method (CAM) criteria.^32–34 the CAM scale was assessed twice a day at the same period (10:00 am and 5:00 pm) every-day. POD incidence was recorded only within 7 postoperative days. The CAM questionnaire was administered by an experienced physician (the first author) based on the following criteria: (1) inattention; (2) acute onset and fluctuating course; (3) altered consciousness; and (4) disorganized thinking. Delirium was reflected by criteria (1) and (2), combined with either criterion (3) or (4). Firstly, individuals with typical words of delirium recorded postoperatively were retained. Then, cases administered drugs for POD were added. Thirdly, individuals uttering words of delirium or administered drugs for delirium preoperatively were excluded. Finally, preliminary diagnoses were confirmed by neurologists per the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) criteria.³⁵

Data Collection

Clinicodemographic variables included gender, age, body mass index (BMI), smoking, drinking, hypertension, heart disease, cerebrovascular diseases, diabetes, abnormal renal function, fracture type, surgical method, surgery time, intraoperative blood loss, anesthesia method, ASA class (I/II/III/IV/V), red blood cell (RBC), hemoglobin (HGB), erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), total protein (TP), albumin (ALB), NLR, PLR, MLR, SII, SIRI, aggregate index of systemic inflammation (AISI), NAR, lymphocyte-to-monocyte ratio (LMR), CAR, and PNI.

Systemic Inflammatory, Albumin-Related Inflammatory and Nutritional Biomarkers

Systemic inflammatory, albumin-related inflammatory and nutritional biomarkers were calculated using the equations in Table 1.

Table 1 The Definitions of Systemic Inflammatory Markers, Albumin-Associated Inflammatory and Nutritional Markers

Statistical Analysis

The included cases were randomized into the training and validation sets at 7:3. Data with non-normal distribution were reported as median and interquartile range. In univariable analysis, categorical variates were assessed by the chi-square or Fisher’s exact test, while continuous variates were compared by the Student’s t-test or rank-sum test. In the training set, the least absolute shrinkage and selection operator (LASSO) was applied in multivariable analysis to determine independent risk factors for developing a nomogram for POD prediction. The nomogram’s performance was assessed by generating receiver operating characteristic (ROC) and calibration curves, with areas under the ROC curve (AUCs) ranging between 0.5 (no discriminant) and 1 (perfect discriminant). The net clinical benefit of the nomogram was examined by decision curve analysis (DCA). P<0.05 suggested statistical significance. R 4.2.2 was employed for data analysis.

Results

Baseline Features of the Patients

Totally 437 geriatric patients underwent surgery for hip fractures and met the eligibility criteria. The incidence of delirium was 13.96% (61/437). Then, the cases were assigned to the training and validation cohorts at 7:3 by computer-based randomization. The baseline clinicodemographic features of patients in both cohorts are summarized in Table 2.

Table 2 Patient Demographics and Baseline Characteristics

Similar gender distributions were found in the training (N=306) and internal validation (N=131) cohorts (p=0.125). Patient ages were also similar in both sets (p=0.419), ie, 80±8 and 80±9 years in the training and internal validation cohorts, respectively. Likewise, BMI, smoking status, drinking habits, fracture classifications, and the prevalence rates of hypertension, heart disease, cerebrovascular diseases, diabetes and abnormal renal function did not differ significantly between cohorts. Notably, various surgical methods and associated parameters such as surgery time, intraoperative blood loss, anesthesia method, and ASA classification also showed no significant differences. Laboratory indexes such as RBC, HGB, ESR, CRP, TP, ALB, NLR, PLR, MLR, SII, SIRI, AISI, NAR, LMR, CAR, and PNI were comparable between the two cohorts, with no statistical significance (p>0.05). These findings suggest a high degree of similarity in baseline characteristics in the training and internal cohorts in our predictive study.

Various indexes were next compared by the Wilcoxon or chi-square test between the POD and No-POD groups. In the training set, gender (P=0.047), heart disease (P=0.035), fracture classification (P=0.010), surgical method (P=0.003), NLR (P<0.001), PLR (P<0.001), MLR (P<0.001), SII (P<0.001), SIRI (P<0.001), AISI (P<0.001), NAR (P<0.001) and LMR (P<0.001) showed significant differences (Table 3).

Table 3 Comparison of Variables Between POD Group and No-POD Group

Nomogram Developed Based on Logistic Regression Analysis

Candidate predictors, ie, gender, heart disease, classification of fracture, surgical method, NLR, PLR, MLR, SII, SIRI, AISI, NAR and LMR, were entered in the initial model, and 2 potential predictive factors were finally retained after LASSO regression analysis in the training set (Table 4 and Figure 2A). Figure 2B shows cross-validation errors, with the best model in cross-validation including 2 variables (NLR and SIRI).

Table 4 The Coefficients of Lasso Regression Analysis

Figure 2 Lasso regression cross-validation plot (A) and lasso regression coefficient path plot (B).

ROC curve analysis of NLR and SIRI yielded AUCs of 0.919 and 0.991, respectively (>0.5) (Figure 3). Further multivariable analyses were carried out in the training cohort (Table 5). A simple-to-use nomogram incorporating one independent predictive factor, SIRI, was developed (Figure 4). The nomogram had outstanding predictive performance (Figure 5), with AUCs of 0.991 (95% CI 0.983~0.998) and 0.986 (95% CI 0.966~1.000) in the training and internal validation sets, respectively. Calibration analysis of the nomogram (Figure 6A and B) showed a high concordance between the actual and predicted probabilities of POD occurrence. These data indicate the novel nomogram can be accurately used to predict POD in these patients. DCA curve analysis of the nomogram is depicted in Figure 7A and B, revealing the nomogram provides overt net benefits for POD prediction in clinic.

Table 5 Results of Multivariate Logistic Regression for Training Cohort

Figure 3 ROC curve analysis 2 candidate diagnostic indicators.

Figure 4 Nomogram of probability to develop postoperative delirium risk in elderly Hip fracture patients using preoperative immune inflammation-related indicators. To use the nomogram, draw an upward vertical line from each covariate to the points bar to calculate the number of points. Based on the sum of the covariate points, draw a downward vertical line from the total point’s line to calculate the probability of developing postoperative delirium.

Figure 5 ROC curve for the nomogram based on the training cohort (The AUC is 0.991) and internal validation cohort (The AUC is 0.986).

Figure 6 (A) Calibration curves of the nomogram for predicting postoperative delirium from the training cohort; (B) Calibration curves of the nomogram for predicting postoperative delirium from the internal validation cohort.

Figure 7 Decision curve analysis (DCA) of the nomogram: (A) The DCA curve of the training cohort; (B) The DCA curve of the internal validation cohort.

Discussion

Hip fractures threaten the health of geriatric individuals and decrease their quality of life, with multiple complications, including bedsores, lung infections.³⁶ Currently, geriatric hip fracture cases are mostly treated by surgical procedures such as arthroplasty and fracture internal fixation.³⁷ However, these surgical treatments may also result in multiple complications, of which POD is a major postoperative complication in geriatric hip fracture cases. POD pathogenesis is very complex and remains unclear so far. In recent years, many hypotheses have emerged,^38–40 including the central neurotransmitter theory, the theory of changes in brain metabolic level, the surgical stress theory, and the inflammation hypothesis. This study aimed to establish a novel inflammatory composite scoring system and to construct a nomogram model that could predict POD in hip fracture cases. The novel nomogram model may be utilized as an important strategic guide for perioperative management and targeted to screen patients for POD risk before surgery for early prevention and treatment.

Previous studies have demonstrated that NLR has significant associations with POD and cognitive decline.^41,42 Findings by Wen and colleagues revealed that preoperative SIRI and NLR levels are correlated with hip arthroplasty in elderly individuals, with SIRI independently predicting hip arthroplasty, corroborating our findings.^11,43 Additionally, stress response, associated with inflammation, trauma, and surgical and anesthetic procedures, activates the peripheral immune system, increases neutrophil and monocyte contents, and reduces lymphocyte amounts.⁴⁴ Upon activation, neutrophils and monocytes secrete anaerobic free radicals, chemokines, and inflammatory cytokines, as a potential mechanism for POD. Increasing evidence suggests preoperative inflammatory mediators, inflammation and immune responses induced by surgery or anesthesia contribute to the pathogenetic mechanism of POD.^45,46 Since neutrophils, lymphocytes, and monocytes represent crucial components of the peripheral immune system, cerebral immune-inflammatory responses are induced by proinflammatory cytokines produced by circulating immune cells, which might activate microglia and thus cause POD.^47,48 Therefore, NLR and SIRI as comprehensive inflammatory indicators might help predict immune and inflammatory disorders.

In this study, POD incidence in 437 elderly hip fractures was 13.96% (61/437). This finding suggests almost one-sixth of the patients experienced POD, and individuals with higher NLR and SIRI levels were prone to develop POD, corroborating previously reported data.^11,31 This study firstly explored the associations of 12 indicators, including gender, heart disease, fracture classification, surgical method, NLR, PLR, MLR, SII, SIRI, AISI, NAR and LMR, and POD. The results showed that these 12 indexes were closely related to POD occurrence in hip fracture cases. In this study, LASSO regression analysis optimized the 2 included indicators, ie, NLR and SIRI, which showed AUCs of 0.919 and 0.991 in ROC curve analysis, respectively. These AUC values were greater than 0.5, indicating high predictive value and clinical significance. As demonstrated above, NLR and SIRI were both higher in POD patients before surgery, and SIRI had a greater AUC compared with NLR, indicating that preoperative SIRI is a better indicator to predict POD than NLR. Logistic regression analysis revealed SIRI as an independent risk factor for POD in hip fracture cases. The current findings suggest the preoperative inflammatory condition should be examined in elderly individuals scheduled for hip fracture surgery.

However, this study had limitations. By excluding patients with dementia, this study excluded one of the groups of patients at highest risk for delirium. Despite the relatively large sample size, the nomogram model was not verified by external data sets. Our analysis of gender, heart disease, fracture classification, surgical method, NLR, PLR, MLR, SII, SIRI, AISI, NAR and LMR at a single time point rather than studying their dynamic changes may hamper the understanding POD development. Besides, there are many potential influencing factors of POD, and the nomogram model may miss other important risk factors. Also, the study only screened POD within 7 days after surgery, which this might be a limitation of the study. Finally, this was a single-center study, and generalization of the proposed nomogram model may require further validation.

Conclusion

In summary, the novel nomogram constructed in the present study has a satisfactory accuracy in predicting POD. Therefore, assessing preoperative immune inflammation-related indicators in elderly hip fracture cases, combined with using the nomogram model constructed in the current study, may provide early detection of patients at high risk of POD and improve perioperative treatment strategies.