Development and validation of a new drug-focused predictive risk score for postoperative delirium in orthopaedic and trauma surgery patients

Delirium is defined as an acute change in attention, awareness, and cognition [1]. It usually develops rapidly with a fluctuating course [1]. In elderly, hospitalized patients, it represents a severe complication with a prevalence ranging from 20% in general surgery up to 70% in intensive care units [2, 3]. After surgical intervention, postoperative delirium (POD) is one of the most common complications [4]. The clinical consequences of delirium are severe with an increase in mortality, length of hospital stay, and development of dementia or cognitive decline [5, 6].

Multiple risk factors determine the risk for delirium. Predisposing risk factors include advanced age, visual and hearing impairment, history of prior delirium, cognitive impairment, frailty, and comorbidities (i.e. cardiovascular or renal diseases). In addition, precipitating factors such as acute medical illness (i.e. infections, hypoglycaemia), trauma, surgical procedures, dehydration, pain, medication use, and drug withdrawal are relevant [2]. Preventive measures can reduce the occurrence of delirium; therefore, overall non-pharmacological measures are recommended for vulnerable patients [2, 7]. However, identifying patients at risk who are most likely to benefit from specific preventive measures remains challenging. One approach is the use of risk prediction scores based on pre- and perioperative risk factors for the identification of patients at high risk for delirium.

Drugs, especially substances targeting the central nervous system, are a proven risk factor for in-hospital delirium [8‐11]. However, a recent systematic review found that medication is not adequately considered in previously developed risk scores and should be addressed in future models [12]. Existing risk scores already considering medication include the risk factors polymedication (≥ five regular drugs) [13], psychoactive and anticholinergic drugs [14], or medication for insomnia treatment [15]. The Delirium Model (DEMO) includes drugs associated with delirium in a linear prediction model [16]. Recently, a medication-based prediction score for POD in surgical patients was developed by our group [17]; however, further revision concerning the weighing of risk factors and testing of the predictive performance is needed.

Importantly, patients at risk for POD should be identified at an early stage prior to surgery. One opportunity for a timely identification of patients at risk for drug-related POD is during pharmacist-led medication reconciliation (PhMR) at hospital admission. In addition, scores including delirium risk-increasing drugs should be easy to use for pharmacists involved in the hospital medication process, and variables should be readily accessible in clinical practice. Consequently, pharmacists can inform physicians about the patients’ individual risk and make preventive suggestions with a focus on minimizing the risk for drug-related POD.

Therefore, the aim of our study was to further develop and validate a risk score for POD including delirium risk-increasing drugs, which can be performed during PhMR at hospital admission by pharmacists based on the admission medication and other available risk factors. This risk score could identify patients at risk for drug-related POD benefiting from suggestions for drug therapy safety and additional preventive measures.

We developed and validated a new predictive risk score for postoperative delirium including delirium risk-increasing drugs (DRD score) based on preoperative risk factors available during pharmacist-led medication reconciliation at hospital admission. In a retrospective single-centre study including orthopaedic and trauma surgery patients, the four risk factors advanced age, reduced kidney function, high anticholinergic burden, and number of delirium risk-increasing drugs proved to be predictive in the final model after multivariable logistic regression analysis. The sensitivity of the score was good with 83.3% in the development and fair with 63.3% in the validation cohort, also a good specificity was achieved with 80.1% and 77.1%, respectively. Thus, the newly developed DRD score is a promising tool for the early and pragmatic identification of patients at risk for POD. After calculation during medication reconciliation at admission it allows a timely initiation of preventive measures. Considering the severe clinical consequences of POD and that 40% of delirium cases are possibly preventable [30], implementation of the score in the clinical routine has the potential to considerably improve patient safety.

Drugs are a well-described risk factor for delirium. Psychoactive drugs and drugs associated with brain-related adverse effects are commonly known and possibly modifiable risk factors for delirium [11, 14, 36]. Surprisingly, in several predictive risk scores developed in recent years, drugs are mostly neglected [25, 32, 33, 37], although medications may account for 12–39% of delirium cases [38]. Risk scores including drugs either require complex automated calculations [16] or show an oversimplified approach when only considering polymedication (≥ five regular drugs) [13]. Our list of delirium risk-increasing drugs includes drugs with effects on the central nervous system as well as drugs correlating with comorbidities associated with delirium (i.e. diabetes mellitus and COPD) [14, 34, 39]. In our analysis, we found that taking two or more delirium risk-increasing drugs was a significant risk factor for POD (OR 5.38, 95% CI 2.28–12.65). Identifying these drugs during PhMR appears feasible and easily applicable in clinical practice.

Neurotransmitter disturbance is a major mechanism in delirium pathophysiology, and a reduced cholinergic activity is associated with altered attention and delirium [2]. Anticholinergic drugs that decrease central cholinergic activity can therefore increase the risk for delirium. This anticholinergic activity can be estimated through anticholinergic burden scales. A preoperative high anticholinergic burden is significantly associated with incident delirium [40‐42], although contrary findings with no association have been reported in other studies [43]. Our study determined that a high anticholinergic burden was significantly associated with POD in both univariable and multivariable analyses.

Advanced age is a well-known risk factor for delirium [2, 7] and, accordingly, was proven to be a statistically significant factor in our study, as confirmed by multivariable analysis. However, thresholds for age as a risk factor vary in risk scores and evaluations published so far. In our study, two thresholds were evident: 66–75 years of age (OR 11.15) and > 75 years of age (OR 23.91); both of these thresholds were included in the score with distinct point assignments.

We found that a moderately decreased kidney function (eGFR < 60 ml/min/1.73m²) on the day of admission was significantly associated with POD (OR 3.89, 95% CI 1.21–12.11). To our knowledge, this predictor has not been considered in previously developed risk scores. An association between moderate renal impairment (eGFR 30–60 ml/min/1.73m², calculated using cystatin-based equations) and delirium was found in fracture patients aged 75–84 years [44]. End-stage renal failure was a consistent risk factor for POD, as reported in an umbrella review of systematic reviews [45]. We used a creatinine-based CKD-EPI equation to calculate the eGFR and found that moderately decreased values were associated with delirium.

Although male gender is included in previously developed risk scores [33, 37, 46], we did not determine this factor to be significant. The predictors smoking and heavy alcohol use are also represented in published risk models [14, 25, 32, 33], whereas for our cohort no associations were found. Underreporting might be a reason for this finding since documentation was based on self-reports. Although the exact correlation between smoking and delirium is unclear, acute nicotine withdrawal may increase the risk of POD [47, 48]. Since this might be especially relevant to patients with a high nicotine dependency, binary reporting of the smoking status may be inadequate.

The new DRD score was developed and validated in a cohort of orthopaedic and trauma surgery patients who are known to be at risk for POD due to multiple risk factors [49, 50]. The prevalence of POD in our study was 18.4% for patients > 65 years, which is comparable to previous findings. The overall incidence of delirium in hospitalized older adults was 23% according to a meta-analysis of 33 studies [2]. The incidence of POD varies depending on the type of surgery with ≥ 20% for major surgery, which includes interventions in orthopaedic and trauma surgery [2]. Due to the retrospective assessment of POD through documented 4AT scores, ICD-10 diagnosis, and chart review, underreporting is possible due to inappropriate documentation, especially for patients with hypoactive delirium. Nonetheless, a chart-based method for identifying POD is validated and frequently used [51].

In this study, patients with a DRD score of 4 or higher were classified as at risk for POD. This applies for all patients > 65 years with an intake of at least two delirium risk-increasing drugs. For patients with less than two delirium risk-increasing drugs, depending on age, four points can only be reached if one or both additional risk factors (high anticholinergic burden and reduced kidney function) apply. For the predictive performance of the DRD score in the development and validation cohorts, sufficient AUC-values with 0.887 (95% CI 0.845–0.930) and 0.805 (95% CI 0.736–0.874) were obtained. Also calibration plots showed good calibration for both test cohorts (Fig. 1). The specificity and NPV were good in both development and validation cohort, meaning patients without risk will be stratified correctly. The sensitivity was good in the development cohort (83.3%), but lower in the validation cohort (63.3%), meaning some patients at risk could be missed. However, since the DRD score will be a first screening during PhMR and additional screening will take place on the ward, the achieved sensitivity was judged as acceptable. For patients identified at high risk for drug-related POD, pharmacists can consequently perform a medication review and state suggestions for drug therapy safety as an additional preventive measure to reduce the risk for POD.

Our study has several limitations. Since we performed a single centre study in a specific patient cohort of orthopaedic and trauma surgery patients, the generalizability of our findings is unknown and should be addressed in further studies. A number of important predictors for POD are not included in the DRD score, such as dementia, cognitive impairment, previously developed delirium, hearing and visual impairment, physical status, type of surgical procedure, and severity of illness [7, 25, 32, 33, 37, 46]. This is primarily due to its focus on implementation during PhMR, and we were thus limited to factors available at this time point. Although some patients with dementia receive anti-dementia drugs, we are aware that patients with unrecognized or untreated dementia will not be assessed. Patients who develop POD because of other, not drug-related risk factors might not be predicted through the DRD score. However, the focus of our new score includes delirium risk-increasing drugs and it could be a trigger for pharmaceutical advice with the aim to erase or minimize the risk for drug-related POD. Thus, the primary aim of the DRD score is to identify patients at risk for drug-related POD who may benefit from suggestions for drug therapy safety.

There is some overlap between our list of delirium risk-increasing drugs and drugs included in the ACB score. For receiving corresponding score points, at least two delirium risk-increasing drugs or an ACB score ≥ 3 are necessary. An ACB score ≥ 3 can either be reached for multiple drugs with a low to moderate anticholinergic effect or for single drugs with a high anticholinergic effect (e.g. tricyclic antidepressants or antimuscarinic agents for the treatment of overactive bladder). For drugs with high anticholinergic properties, which are also classified as delirium risk-increasing drugs, a double rating will only occur if other delirium risk-increasing drugs are taken in addition. Besides anticholinergic properties, delirium risk-increasing drugs have various other effects on the central nervous system or correlate with comorbidities associated with POD. Thus, cases with a double rating can be justified, and an additional risk for POD can be proposed due to multiple drug-related central nervous effects and associated comorbidities. Furthermore, for both risk factors no statistical multicollinearity was determined and the overlap was therefore not considered to be decisive.

We performed an internal validation in orthopaedic and trauma surgery patients. External validation in different surgical patient cohorts is necessary to estimate the score performance in other settings and determine the generalizability of the DRD score. However, as a strength, this study was performed with real-life data and considered the feasibility in clinical practice.

The new DRD score is a predictive risk score assessable during pharmacist-led medication reconciliation at hospital admission and is suitable for identifying patients at risk for drug-related POD. In addition to general preventive measures, specific preventive measures concerning drug therapy safety should be implemented for identified patients to reduce the risk for POD.