Evaluating gender bias in ML-based clinical risk prediction models: A study on multiple use cases at different hospitals.

BACKGROUND: An inherent difference exists between male and female bodies, the historical under-representation of females in clinical trials widened this gap in existing healthcare data. The fairness of clinical decision-support tools is at risk when developed based on biased data. This paper aims to quantitatively assess the gender bias in risk prediction models. We aim to generalize our findings by performing this investigation on multiple use cases at different hospitals. METHODS: First, we conduct a thorough analysis of the source data to find gender-based disparities. Secondly, we assess the model performance on different gender groups at different hospitals and on different use cases. Performance evaluation is quantified using the area under the receiver-operating characteristic curve (AUROC). Lastly, we investigate the clinical implications of these biases by analyzing the underdiagnosis and overdiagnosis rate, and the decision curve analysis (DCA). We also investigate the influence of model calibration on mitigating gender-related disparities in decision-making processes. RESULTS: Our data analysis reveals notable variations in incidence rates, AUROC, and over-diagnosis rates across different genders, hospitals and clinical use cases. However, it is also observed the underdiagnosis rate is consistently higher in the female population. In general, the female population exhibits lower incidence rates and the models perform worse when applied to this group. Furthermore, the decision curve analysis demonstrates there is no statistically significant difference between the model's clinical utility across gender groups within the interested range of thresholds. CONCLUSION: The presence of gender bias within risk prediction models varies across different clinical use cases and healthcare institutions. Although inherent difference is observed between male and female populations at the data source level, this variance does not affect the parity of clinical utility. In conclusion, the evaluations conducted in this study highlight the significance of continuous monitoring of gender-based disparities in various perspectives for clinical risk prediction models.

Intraoperative Blood Pressure Management and Its Effects on Postoperative Delirium After Cardiac Surgery: A Single-Center Retrospective Cohort Study.

OBJECTIVES: There is accumulating evidence that blood pressure management might be associated with end-organ dysfunction after cardiac surgery. This study aimed to investigate the impact of intraoperative hypotension (IOH) on adverse neurologic outcomes and mortality. DESIGN: A single-center retrospective cohort study. SETTING: The Heart and Diabetes Centre Bad Oeynhausen NRW, Ruhr-University Bochum. PARTICIPANTS: This retrospective cohort study included 31,315 adult patients who underwent elective cardiac surgery at the authors' institution between January 2009 and December 2018. INTERVENTIONS: All cardiac surgery procedures except assist device implantation, organ transplantation, and emergency surgery. MEASUREMENTS AND MAIN RESULTS: Adverse neurologic outcomes were defined as postoperative delirium and stroke. IOH was defined as mean arterial pressure below 60 mmHg for >2 minutes. The frequency of IOH episodes and the cumulative IOH duration were recorded. The association between IOH and adverse neurologic outcomes was examined with unadjusted statistical analysis and multiple logistic regression analysis. Eight hundred forty-nine (2.9%) patients developed postoperative stroke, and 2,401 (7.7%) patients developed postoperative delirium. The frequency of IOH episodes was independently associated with postoperative delirium in the multiple logistic regression analysis (odds ratio 1.02, 95% CI 1.003-1.03, p < 0.001), whereas there was no association between it and stroke. CONCLUSION: This large retrospective monocentric cohort study revealed that increased episodes of IOH were associated with the risk of developing postoperative delirium after cardiac surgery. This might have important clinical implications with respect to careful and precise hemodynamic monitoring and proactive treatment, especially in patients with increased risk for postoperative delirium.

The Efficacy and Safety of a Low-Dose Tranexamic Acid Bolus-Only Protocol Compared with the Moderate-Dose Protocol in Valvular Heart Surgery.

In 873 propensity score-matched pairs of patients undergoing valvular heart surgery, we compared a "moderate dose" of tranexamic acid (TXA) protocol (group 1; median TXA dose: 24 mg/kg body weight) with a 1.5-g "bolus-only" protocol (group 2; median TXA dose: 19 mg/kg body weight). The number of transfused patients was higher in group 2 than in group 1 (74.5 vs 66.0%, p < 0.001), as was the number of transfused red blood cell concentrates (p = 0.001). The risks of re-exploration and convulsive seizures were similar between groups (p > 0.50). Data indicate an impaired efficacy following the "bolus-only" protocol, without a significant safety improvement.

How Would We Treat Our Own Cystic Fibrosis With Lung Transplantation?

Lung transplantation is the only therapy for patients with end-stage lung disease. In advanced lung diseases such as cystic fibrosis (CF), life expectancy increases, and it is important to recognize extrapulmonary comorbidities. Cardiovascular involvement, including pulmonary hypertension, right-heart failure, and myocardial dysfunction, are manifest in the late stages of CF disease. Besides right-heart failure, left-heart dysfunction seems to be underestimated. Therefore, an optimal anesthesia and surgical management risk evaluation in this high-risk patient population is mandatory, especially concerning the perioperative use of mechanical circulatory support. The use of an index case of an older patient with the diagnosis of cystic fibrosis demonstrates the importance of early risk stratification and strategy planning in a multidisciplinary team approach to guarantee successful lung transplantation.

Moderate Dose of Tranexamic Acid and Complications after Valvular Heart Surgery.

BACKGROUND: In valvular open-heart surgery, data regarding the effect of a moderate dose of tranexamic acid (TXA) on clinical outcomes are limited. METHODS: Out of a cohort of 13,293 patients, we performed a propensity-score-matched analysis in 6,106 patients and assessed the risk of convulsive seizures (CS, primary endpoint), stroke, renal replacement therapy, and mortality (secondary endpoints). In the entire study cohort of 13,293 patients, we also assessed the multivariable-adjusted association of CS with postoperative outcomes. RESULTS: The risk of CS was significantly higher in the TXA group (2.4%; n = 72) than in the non-TXA group (1.0%; n = 32), with a relative risk ratio (RR) of 2.28 (95% confidence interval [CI]: 1.50-3.47; p < 0.001). The risk of CS was also higher in patients receiving TXA doses ≥25 mg/kg body weight (3.7%; n = 40) than in patients receiving <25 mg/kg body weight (1.6%; n = 32; p < 0.001). Perioperative secondary clinical endpoints and 1-year mortality did not differ significantly between study groups (p-value > 0.05). Compared with non-CS patients (n = 13,000), patients with nonhemorrhagic, nonembolic CS (n = 253) revealed higher multivariable-adjusted in-hospital risks of stroke (RR: 3.82 [95% CI: 2.44-5.60; p < 0.001]) and mortality (RR: 2.07 [95% CI: 1.23-3.48; p = 0.006]), and a higher 1-year mortality risk (RR: 1.85 [95% CI: 1.42-2.41; p < 0.001]). CONCLUSION: A moderate TXA dose was associated with a significantly higher risk of seizure, but not with other clinical complications such as stroke, renal replacement therapy, and mortality. However, in the small group of patients experiencing a seizure, the risks of stroke and short- and mid-term mortality were substantially higher than in patients not experiencing a seizure, indicating that the use of a low-dose TXA protocol (<25 mg/kg body weight) should be considered.

Tranexamic Acid and the Risk of Delirium after Off-Pump Surgery.

We investigated whether in patients undergoing off-pump coronary artery bypass grafting surgery a single bolus of 1 g tranexamic acid (TXA) impacts the risk of postoperative delirium using the propensity score matching approach. In 2,757 pairs, the risk of delirium was 4.2% (TXA group) and 5.0% (non-TXA group), with a relative risk in the TXA versus the non-TXA group of 0.83 (95% confidence interval: 0.65-1.07; p = 0.16). There was no significant interaction between TXA administration and renal function on the risk of delirium (p = 0.12). Data indicate that a single bolus of 1 g TXA does not increase the risk of delirium in patients undergoing off-pump surgery.

Machine Learning-Based Prediction Models for Different Clinical Risks in Different Hospitals: Evaluation of Live Performance.

BACKGROUND: Machine learning algorithms are currently used in a wide array of clinical domains to produce models that can predict clinical risk events. Most models are developed and evaluated with retrospective data, very few are evaluated in a clinical workflow, and even fewer report performances in different hospitals. In this study, we provide detailed evaluations of clinical risk prediction models in live clinical workflows for three different use cases in three different hospitals. OBJECTIVE: The main objective of this study was to evaluate clinical risk prediction models in live clinical workflows and compare their performance in these setting with their performance when using retrospective data. We also aimed at generalizing the results by applying our investigation to three different use cases in three different hospitals. METHODS: We trained clinical risk prediction models for three use cases (ie, delirium, sepsis, and acute kidney injury) in three different hospitals with retrospective data. We used machine learning and, specifically, deep learning to train models that were based on the Transformer model. The models were trained using a calibration tool that is common for all hospitals and use cases. The models had a common design but were calibrated using each hospital's specific data. The models were deployed in these three hospitals and used in daily clinical practice. The predictions made by these models were logged and correlated with the diagnosis at discharge. We compared their performance with evaluations on retrospective data and conducted cross-hospital evaluations. RESULTS: The performance of the prediction models with data from live clinical workflows was similar to the performance with retrospective data. The average value of the area under the receiver operating characteristic curve (AUROC) decreased slightly by 0.6 percentage points (from 94.8% to 94.2% at discharge). The cross-hospital evaluations exhibited severely reduced performance: the average AUROC decreased by 8 percentage points (from 94.2% to 86.3% at discharge), which indicates the importance of model calibration with data from the deployment hospital. CONCLUSIONS: Calibrating the prediction model with data from different deployment hospitals led to good performance in live settings. The performance degradation in the cross-hospital evaluation identified limitations in developing a generic model for different hospitals. Designing a generic process for model development to generate specialized prediction models for each hospital guarantees model performance in different hospitals.

Preoperative Risk Factors and Early Outcomes of Delirium in Valvular Open-Heart Surgery.

BACKGROUND: Compared with coronary artery bypass grafting surgery, data regarding postoperative delirium are scant in valvular open-heart surgery. Therefore, the goal of this retrospective study was to investigate the incidence, preoperative risk factors, and early outcomes of delirium in a large group of patients undergoing valvular open-heart surgery. METHODS: In 13,229 patients with isolated valvular or combined valvular and bypass surgery, the incidence of postoperative delirium was assessed until discharge. Independent risk factors of delirium were evaluated by multivariable logistic regression analysis. Moreover, we assessed the multivariable-adjusted risk of prolonged intensive care unit (ICU) stay (>48 hours) and in-hospital mortality in patients with delirium. RESULTS: Overall, the incidence of postoperative delirium was 8.4%. The incidence in patients experiencing a postoperative stroke or seizure was 23.1 and 29.7%, respectively. Twelve preoperative risk factors, mostly nonmodifiable, were independently associated with the risk of delirium, including advanced age, renal impairment, stroke, the need for emergency surgery, and severe preoperative anemia (hemoglobin < 9 g/dL). Postoperative delirium was associated with an adjusted odds ratio (OR) of prolonged ICU stay of 9.48 (95% confidence interval [CI]: 7.96-11.30). Adjusted in-hospital mortality was, however, significantly lower in patients with delirium versus patients without delirium (OR, 0.56; 95% CI: 0.38-0.83). CONCLUSION: In valvular open-heart surgery, postoperative delirium is a frequent neurological complication that is associated with other postoperative neurological complications and several, mostly nonmodifiable, preoperative risk factors. Although postoperative delirium was associated with a significantly increased risk of prolonged ICU stay, this did not translate into an increased short-term mortality.

Impairment of Cognitive Function in Different Domains Early After Lung Transplantation.

In this prospective observational pilot study patients with the diagnosis of end-stage lung disease and listed for lung transplantation underwent a cognitive function test battery before and after lung transplantation to investigate postoperative cognitive function in three domains (visual and verbal memory, executive functioning, concentration/speed of processing). Additionally we investigated intraoperative risk factors for postoperative cognitive dysfunction. In total, 24 patients were included in this pilot study. The incidence of postoperative cognitive dysfunction was 58.3%. In the cognitive dysfunction group, the domains executive functioning and concentration/attention were significantly impaired whereas memory was not affected. Patients with cognitive impairment had a significantly longer ICU stay. The strongest independent risk factor for the development of cognitive dysfunction was operation time. No influence of cerebral oxygen desaturations on cognitive dysfunction was found. This might have important implications for early psychological rehabilitation strategies in this high-risk patient collective.

Artificial intelligence predicts delirium following cardiac surgery: A case study.

Delirium is a highly relevant complication of surgical interventions. Current research indicates that despite increased awareness for delirium, it is often overlooked. We implemented an AI-based tool to monitor delirium in cardiac surgery patients in our specialist clinic. This appears to be a promising approach to improve detection of delirium, especially for underrecognized forms and in peripheral wards without intensive screening. We present a case in which the AI identified delirium, confirmed by our routine screening and specialist evaluation.

Risk-adjusted analysis of long-term outcomes after on- versus off-pump coronary artery bypass grafting.

OBJECTIVES: Recent data suggested that off-pump coronary artery bypass (OPCAB) may carry a higher risk for mortality in the long term when compared to on-pump coronary artery bypass (ONCAB). We, therefore, compared long-term survival and morbidity in patients undergoing ONCAB versus OPCAB in a large single-centre cohort. METHODS: A total of 8981 patients undergoing isolated elective/urgent coronary artery bypass grafting between January 2009 and December 2019 were analysed. Patients were stratified into 2 groups (OPCAB n = 6649/ONCAB n = 2332). The primary end point was all-cause mortality. Secondary endpoints included repeat revascularization, stroke and myocardial infarction. To adjust for potential selection bias, 1:1 nearest neighbour propensity score (PS) matching was performed resulting in 1857 matched pairs. Moreover, sensitivity analysis was applied in the entire study cohort using multivariable- and PS-adjusted Cox regression analysis. RESULTS: In the PS-matched cohort, 10-year mortality was similar between study groups [OPCAB 36.4% vs ONCAB 35.8%: hazard ratio (HR) 0.99, 95% confidence interval (CI) 0.87-1.12; P = 0.84]. While 10-year outcomes of secondary endpoints did not differ significantly, risk of stroke (OPCAB 1.50% vs ONCAB 2.8%: HR 0.51, 95% CI 0.32-0.83; P = 0.006) and mortality (OPCAB 3.1% vs ONCAB 4.8%: HR 0.65, 95% CI 0.47-0.91; P = 0.011) at 1 year was lower in the OPCAB group. In the multivariable- and the PS-adjusted model, mortality at 10 years was not significantly different (OPCAB 34.1% vs ONCAB 35.7%: HR 0.97, 95% CI 0.87-1.08; P = 0.59 and HR 1.01, 95% CI 0.90-1.13; P = 0.91, respectively). CONCLUSIONS: Data do not provide evidence that elective/urgent OPCAB is associated with significantly higher risks of mortality, repeat revascularization, or myocardial infarction during late follow-up when compared to ONCAB. Patients undergoing OPCAB may benefit from reduced risks of stroke and mortality within the first year postoperatively.

Effects of controlled ovarian stimulation on vascular barrier and endothelial glycocalyx: a pilot study.

PURPOSE: Controlled ovarian stimulation significantly amplifies the number of maturing and ovulated follicles as well as ovarian steroid production. The ovarian hyperstimulation syndrome (OHSS) increases capillary permeability and fluid extravasation. Vascular integrity intensely is regulated by an endothelial glycocalyx (EGX) and we have shown that ovulatory cycles are associated with shedding of EGX components. This study investigates if controlled ovarian stimulation impacts on the integrity of the endothelial glycocalyx as this might explain key pathomechanisms of the OHSS. METHODS: Serum levels of endothelial glycocalyx components of infertility patients (n=18) undergoing controlled ovarian stimulation were compared to a control group of healthy women with regular ovulatory cycles (n=17). RESULTS: Patients during luteal phases of controlled ovarian stimulation cycles as compared to normal ovulatory cycles showed significantly increased Syndecan-1 serum concentrations (12.6 ng/ml 6.1125th-19.1375th to 13.9 ng/ml 9.625th-28.975th; p=0.026), indicating shedding and degradation of the EGX. CONCLUSION: A shedding of EGX components during ovarian stimulation has not yet been described. Our study suggests that ovarian stimulation may affect the integrity of the endothelial surface layer and increasing vascular permeability. This could explain key features of the OHSS and provide new ways of prevention of this serious condition of assisted reproduction.

Associations of preoperative stroke and tranexamic acid administration with convulsive seizures in valvular open-heart surgery.

In cardiac surgery, use of the antifibrinolytic agent tranexamic acid (TXA) and acute perioperative stroke are both associated with convulsive seizures. We hypothesized that an older (preoperative) stroke increases the risk of TXA-associated seizures as well. To test this hypothesis, we retrospectively analyzed data from 16,110 patients who had undergone open-heart valvular surgery at our institution between 2009 and 2020. The dosing of TXA was moderate. Use of TXA and a history of stroke were both independently associated with convulsive seizure with an adjusted odds ratio (OR) of 2.40 (95%CI: 1.71-3.37) and 1.79 (95%CI: 1.27-2.54), respectively. Compared to patients without TXA administration, the adjusted OR of experiencing a seizure in TXA patients without a history of stroke was 2.44 (95%CI: 1.71-3.46) and in patients receiving TXA with a history of stroke 4.30 (95%CI: 2.65-6.99). However, there was no significant interaction between TXA use and preoperative stroke on convulsive seizures (P = 0.77). Compared to patients without seizure, for patients with seizure, the inverse probability-weighted ORs of in-hospital mortality and 30-day mortality were 3.58 (95%CI: 2.20-5.83) and 4.04 (95%CI: 2.34-6.98), respectively. We conclude that, in patients undergoing open-heart surgery, a history of stroke is independently associated with convulsive seizures but is not a contraindication for TXA use.

A scalable approach for developing clinical risk prediction applications in different hospitals.

OBJECTIVE: Machine learning (ML) algorithms are now widely used in predicting acute events for clinical applications. While most of such prediction applications are developed to predict the risk of a particular acute event at one hospital, few efforts have been made in extending the developed solutions to other events or to different hospitals. We provide a scalable solution to extend the process of clinical risk prediction model development of multiple diseases and their deployment in different Electronic Health Records (EHR) systems. MATERIALS AND METHODS: We defined a generic process for clinical risk prediction model development. A calibration tool has been created to automate the model generation process. We applied the model calibration process at four hospitals, and generated risk prediction models for delirium, sepsis and acute kidney injury (AKI) respectively at each of these hospitals. RESULTS: The delirium risk prediction models have on average an area under the receiver-operating characteristic curve (AUROC) of 0.82 at admission and 0.95 at discharge on the test datasets of the four hospitals. The sepsis models have on average an AUROC of 0.88 and 0.95, and the AKI models have on average an AUROC of 0.85 and 0.92, at the day of admission and discharge respectively. DISCUSSION: The scalability discussed in this paper is based on building common data representations (syntactic interoperability) between EHRs stored in different hospitals. Semantic interoperability, a more challenging requirement that different EHRs share the same meaning of data, e.g. a same lab coding system, is not mandated with our approach. CONCLUSIONS: Our study describes a method to develop and deploy clinical risk prediction models in a scalable way. We demonstrate its feasibility by developing risk prediction models for three diseases across four hospitals.

Non-invasive evaluation of macro- and microhemodynamic changes during induction of general anesthesia - A prospective observational single-blinded trial.

BACKGROUND: Hypotension and bradycardia are known side effects of general anesthesia, while little is known about further macro- and microhemodynamic changes during induction. Intriguing is furthermore, why some patients require no vasopressor medication to uphold mean arterial pressure, while others need vasopressor support. OBJECTIVE: Determination of macro- and microhemodynamic changes during induction of general anesthesia. METHODS: We enrolled 150 female adults scheduled for gynaecological surgery into this prospective observational, single-blinded trial. Besides routinely measuring heart rate (HR) and mean arterial blood pressure (MAP), the non-invasive technique of thoracic electrical bioimpedance was applied to measure cardiac output (CO), cardiac index (CI), stroke volume (SV), stroke volume variability (SVV) and index of myocardial contractility (ICON) before induction of anesthesia, 7 times during induction, and, finally, after surgery in the recovery room. Changes in microcirculation were assessed using sidestream dark field imaging to establish the perfused boundary region (PBR), a validated gauge of glycocalyx health. Comparisons were made with Friedman's or Wilcoxon test for paired data, and with Mann-Whitney-U test for unpaired data, with post-hoc corrections for multiple measurements by the Holm-Bonferroni method. RESULTS: 83 patients did not need vasopressor support, whereas 67 patients required therapy (norepinephrine, atropine or cafedrine/theodrenaline) to elevate MAP values to ≥70mmHg during induction, 54 of these receiving norepinephrine (NE) alone. Pre-interventional (basal) values of CO, CI, ICON, SV and SVV were all significantly lower in the group of patients later requiring NE (p < 0.04), whereas HR and MAP were identical for both groups. HR, MAP and CO decreased from baseline to 12 min after induction of general anesthesia in both the patients without and those with NE support. Heart rate decreased significantly by about 25% in both groups (-19 to -21 bpm). The median individual decrease of MAP amounted to -26.7% (19.7/33.3, p < 0.001) and -26.1% (11.6/33.2, p < 0.001), respectively, whereas for CO it was -40.7% (34.1/50.1, p < 0.001) and -43.5% (34.8/48.7). While these relative changes did not differ between the two groups, in absolute values there were significantly greater decreases in CO, CI, SV and ICON in the group requiring NE. Noteably, NE did not restore ICON or the other cardiac parameters to levels approaching those of the group without NE. PBR was measured in a total of 84 patients compiled from both groups, there being no intergroup differences. It increased 6.4% (p < 0.001) from pre-induction to the end of the operation, indicative of damage to microvascular glycocalyx. CONCLUSION: Non-invasive determination of CO provides additional hemodynamic information during anesthesia, showing that induction results in a significant decrease not only of MAP but also of CO and other cardiac factors at all timepoints compared to baseline values. The decrease of CO was greater than that of MAP and, in contrast to MAP, did not respond to NE. There was also no sign of a positive inotropic effect of NE in this situation. Support of MAP by NE must consequently result from an increase in peripheral arterial resistance, posing a risk for oxygen supply to tissue. In addition, general anesthesia and the operative stimulus lead to an impairment of the microcirculation.