Dataset Information

Machine Learning to Develop and Internally Validate a Predictive Model for Post-operative Delirium in a Prospective, Observational Clinical Cohort Study of Older Surgical Patients.

ABSTRACT:

Background

Our objective was to assess the performance of machine learning methods to predict post-operative delirium using a prospective clinical cohort.

Methods

We analyzed data from an observational cohort study of 560 older adults (≥ 70 years) without dementia undergoing major elective non-cardiac surgery. Post-operative delirium was determined by the Confusion Assessment Method supplemented by a medical chart review (N = 134, 24%). Five machine learning algorithms and a standard stepwise logistic regression model were developed in a training sample (80% of participants) and evaluated in the remaining hold-out testing sample. We evaluated three overlapping feature sets, restricted to variables that are readily available or minimally burdensome to collect in clinical settings, including interview and medical record data. A large feature set included 71 potential predictors. A smaller set of 18 features was selected by an expert panel using a consensus process, and this smaller feature set was considered with and without a measure of pre-operative mental status.

Results

The area under the receiver operating characteristic curve (AUC) was higher in the large feature set conditions (range of AUC, 0.62-0.71 across algorithms) versus the selected feature set conditions (AUC range, 0.53-0.57). The restricted feature set with mental status had intermediate AUC values (range, 0.53-0.68). In the full feature set condition, algorithms such as gradient boosting, cross-validated logistic regression, and neural network (AUC = 0.71, 95% CI 0.58-0.83) were comparable with a model developed using traditional stepwise logistic regression (AUC = 0.69, 95% CI 0.57-0.82). Calibration for all models and feature sets was poor.

Conclusions

We developed machine learning prediction models for post-operative delirium that performed better than chance and are comparable with traditional stepwise logistic regression. Delirium proved to be a phenotype that was difficult to predict with appreciable accuracy.

SUBMITTER: Racine AM

PROVIDER: S-EPMC7878663 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Introduction: Risk assessment for post-operative delirium (POD) is poorly developed. Improved metrics could greatly facilitate peri-operative care as costs associated with POD are staggering. In this preliminary study, we develop a novel stress-diathesis model based on comprehensive pre-operative psychiatric and neuropsychological testing, a blood oxygenation level-dependent (BOLD) magnetic resonance imaging (MRI) carbon dioxide (CO2) stress test, and high fidelity measures of intra-operative parameters that may interact facilitating POD. Methods: The study was approved by the ethics board at the University of Manitoba and registered at clinicaltrials.gov as NCT02126215. Twelve patients were studied. Pre-operative psychiatric symptom measures and neuropsychological testing preceded MRI featuring a BOLD MRI CO2 stress test whereby BOLD scans were conducted while exposing participants to a rigorously controlled CO2 stimulus. During surgery the patient had hemodynamics and end-tidal gases downloaded at 0.5 hz. Post-operatively, the presence of POD and POD severity was comprehensively assessed using the Confusion Assessment Measure -Severity (CAM-S) scoring instrument on days 0 (surgery) through post-operative day 5, and patients were followed up at least 1 month post-operatively. Results: Six of 12 patients had no evidence of POD (non-POD). Three patients had POD and 3 had clinically significant confusional states (referred as subthreshold POD; ST-POD) (score ≥ 5/19 on the CAM-S). Average severity for delirium was 1.3 in the non-POD group, 3.2 in ST-POD, and 6.1 in POD (F-statistic = 15.4, p < 0.001). Depressive symptoms, and cognitive measures of semantic fluency and executive functioning/processing speed were significantly associated with POD. Second level analysis revealed an increased inverse BOLD responsiveness to CO2 pre-operatively in ST-POD and marked increase in the POD groups when compared to the non-POD group. An association was also noted for the patient population to manifest leucoaraiosis as assessed with advanced neuroimaging techniques. Results provide preliminary support for the interacting of diatheses (vulnerabilities) and intra-operative stressors on the POD phenotype. Conclusions: The stress-diathesis model has the potential to aid in risk assessment for POD. Based on these initial findings, we make some recommendations for intra-operative management for patients at risk of POD.

Project description:BackgroundPost-operative delirium (POD) presents as a serious neuropsychiatric syndrome in the elderly undergoing thoracic and abdominal surgery, which is mostly associated with poor prognosis. The Age-adjusted Charlson Comorbidity Index (ACCI) has been widely recognized as an independently predictive factor for overall survival rate and mortality in various surgeries. However, no studies demonstrated the potential relationship between ACCI and POD. The current study was to explore the correlation between ACCI and POD, and determine the predictive effect of ACCI on POD in the elderly after thoracic and abdominal surgery.Materials and methodsTotal 184 patients (≥60 years) who underwent thoracic and abdominal surgery from 2021.10 to 2022.5 were enrolled in this prospective observational cohort study. ACCI was calculated by weighting comorbidities and age. POD was diagnosed using Confusion Assessment Method (CAM) twice a day in the first 3 days after surgery. The Visual Analog Scale (VAS) was applied to measure pre-operative and post-operative pain at rest and in motion. All demographic and perioperative data were compared in patients with POD and without POD. ACCI and other variables were analyzed by univariate and multivariate logistic regression analysis. The characteristic curve of receiver operating characteristic (ROC) was used to further evaluate the accuracy of ACCI to predict POD.ResultsPost-operative delirium was diagnosed in 36 of 184 patients included in our study. The prevalence of POD in the elderly after thoracic and abdominal surgery was 19.6%. The outcomes by multivariate regression analysis showed the independent risk factors for POD were ACCI (OR: 1.834; 95%CI: 1.434-2.344; P < 0.001), pre-operative Mini-Mental State Examination (MMSE) scores (OR: 0.873; 95%CI: 0.767-0.994; P = 0.040), serum albumin (OR: 0.909; 95%CI: 0.826-1.000; P = 0.049) and pain scores in the post-operative third day (OR: 2.013; 95%CI: 1.459-2.778; P < 0.001). ACCI can predict POD more accurately with the largest area under curve (AUC) of 0.794 and sensitivity of 0.861, respectively.ConclusionAge-adjusted Charlson Comorbidity Index, pre-operative MMSE scores, serum albumin and post-operative pain were independently associated with POD in geriatric patients following thoracic and abdominal surgery. Moreover, ACCI may become an accurate indicator to predict POD early.

Project description:BackgroundPost-operative delirium (POD) is a common complication in elderly patients undergoing surgery, but the underpinning causes are not clear. We hypothesized that inflammaging, the subclinical low and chronic grade inflammation characteristic of old people, can contribute to POD onset. Accordingly, we investigated the association of pre-operative and circulating cytokines in elderly patients (>65 years), admitted for elective and emergency surgery.MethodsThis is a secondary analysis of a sub-cohort of patients belonging to a previous large case-control study, where 351 patients were clinically and cognitively thoroughly characterized, together with the assessment of POD (47 patients) by confusion assessment method and delirium rating scale. Seventy-four pre-operative plasma samples were selected from a larger bio-bank and they included 37 subjects with POD and 37 without POD. Inflammaging related cytokines, i.e., IL-1β, IL-2, IL-6, IL-8, IL-10, and TNF-α, were assayed by ELISA in pre-operative blood samples; univariate and multivariable analyses have been applied to identify cytokines independently associated to POD. Associations of cytokine levels with functional status, cognitive decline, intra-hospital mortality, and comorbidity were also analyzed independently of POD onset.ResultsHigh IL-6 and low-IL-2 levels were significantly associated with POD. After adjustment for potential confounders in multivariate analysis, high level of pre-operative IL-6 was confirmed to be significantly associated with risk of POD onset. High level of IL-6 was also associated with several baseline features (including poor functional status, cognitive impairment, emergency admission, and higher comorbidity burden) and intra-hospital mortality.ConclusionPre-operative, high-plasma level of IL-6 (≥9 pg/mL) was significantly associated with POD onset. We propose IL-6 as an additional risk factor of POD onset together with the previously identified factors. Discovery of all risk factors contributing to POD onset will permit to improve hospitalized patient management and the decrease of healthcare cost.