Leveraging explainable artificial intelligence to optimize clinical decision support

OBJECTIVE

To develop and evaluate a data-driven process to generate suggestions for improving alert criteria using explainable artificial intelligence (XAI) approaches.

METHODS

We extracted data on alerts generated from January 1, 2019 to December 31, 2020, at Vanderbilt University Medical Center. We developed machine learning models to predict user responses to alerts. We applied XAI techniques to generate global explanations and local explanations. We evaluated the generated suggestions by comparing with alert's historical change logs and stakeholder interviews. Suggestions that either matched (or partially matched) changes already made to the alert or were considered clinically correct were classified as helpful.

RESULTS

The final dataset included 2 991 823 firings with 2689 features. Among the 5 machine learning models, the LightGBM model achieved the highest Area under the ROC Curve: 0.919 [0.918, 0.920]. We identified 96 helpful suggestions. A total of 278 807 firings (9.3%) could have been eliminated. Some of the suggestions also revealed workflow and education issues.

CONCLUSION

We developed a data-driven process to generate suggestions for improving alert criteria using XAI techniques. Our approach could identify improvements regarding clinical decision support (CDS) that might be overlooked or delayed in manual reviews. It also unveils a secondary purpose for the XAI: to improve quality by discovering scenarios where CDS alerts are not accepted due to workflow, education, or staffing issues.

One Confirmed and 2 Suspected Cases of Heartland Virus Disease

Heartland virus is a tickborne phlebovirus first identified in Missouri in 2009; 11 human cases have been reported in the literature. Reported hallmarks of infection have included fever, malaise, anorexia, gastrointestinal complaints, thrombocytopenia, neutropenia, and aminotransferase elevations. We report 1 confirmed and 2 suspected cases and discuss implications for case-finding.

Systems-based hematology: highlighting successes and next steps

Systems-based hematology is dedicated to improving care delivery for patients with blood disorders. First defined by the American Society of Hematology in 2015, the idea of a systems-based hematologist arose from evolving pressures in the health care system and increasing recognition of opportunities to optimize the quality and cost effectiveness of hematologic care. In this review, we begin with a proposed framework to formalize the discussion of the range of initiatives within systems-based hematology. Classification by 2 criteria, project scope and method of intervention, facilitates comparison between initiatives and supports dialogue for future efforts. Next, we present published examples of successful systems-based initiatives in the field of hematology, including efforts to improve stewardship in the diagnosis and management of complex hematologic disorders (eg, heparin-induced thrombocytopenia and thrombophilias), the development of programs to promote appropriate use of hematologic therapies (eg, blood products, inferior vena cava filters, and anticoagulation), changes in care delivery infrastructure to improve access to hematologic expertise (eg, electronic consultation and disorder-specific care pathways), and others. The range of projects illustrates the broad potential for interventions and highlights different metrics used to quantify improvements in care delivery. We conclude with a discussion about future directions for the field of systems-based hematology, including extension to malignant disorders and the need to define, expand, and support career pathways.

Management of the bleeding patient receiving new oral anticoagulants: a role for prothrombin complex concentrates

Ease of dosing and simplicity of monitoring make new oral anticoagulants an attractive therapy in a growing range of clinical conditions. However, newer oral anticoagulants interact with the coagulation cascade in different ways than traditional warfarin therapy. Replacement of clotting factors will not reverse the effects of dabigatran, rivaroxaban, or apixaban. Currently, antidotes for these drugs are not widely available. Fortunately, withholding the anticoagulant and dialysis are freqnently effective treatments, particularly with rivaroxaban and dabigatran. Emergent bleeding, however, requires utilization of Prothrombin Complex Concentrates (PCCs). PCCs, in addition to recombinant factor VIIa, are used to activate the clotting system to reverse the effects of the new oral anticoagulants. In cases of refractory or emergent bleeding, the recommended factor concentrate in our protocols differs between the new oral anticoagulants. In patients taking dabigatran, we administer an activated PCC (aPCC) [FELBA] due to reported benefit in human in vitro studies. Based on human clinical trial evidence, the 4-factor PCC (Kcentra) is suggested for patients with refractory rivaroxaban- or apixaban-associated hemorrhage. If bleeding continues, recombinant factor VIIa may be employed. With all of these new procoagulant agents, the risk of thrombosis associated with administration of factor concentrates must be weighed against the relative risk of hemorrhage.

Not only in trauma patients: hospital-wide implementation of a massive transfusion protocol

OBJECTIVES

To review outcomes of massive transfusion protocol (MTP) activation and determine the impact of MTP implementation on blood bank use.

BACKGROUND

MTP has been established to rapidly provide plasma and packed red blood cells in ratios approaching 1 : 1. Due to availability, MTP has been utilised in non-traumatic haemorrhage despite evidence of benefit in this population. Our hospital-wide implementation of MTP was reviewed for propriety, outcomes and effect on blood bank resources.

METHODS

Retrospective cohort study of patients receiving transfusion after MTP activation from October 2009 to 2011. Underlying medical conditions and baseline medication use were determined. In-hospital and 24-h mortality were compared with evaluation for confounding by Acute Physiology And Chronic Health Evaluation (APACHE) score and duration of MTP activation. Blood product use before and after MTP implementation was reviewed.

RESULTS

MTP activation occurred in 62 trauma and 63 non-trauma patients. Non-trauma patients were older, had more underlying medical conditions and higher APACHE scores compared with trauma patients; 24-h mortality was higher in trauma compared with non-trauma patients (27·4 vs 11·1%, P = 0·02). There was no significant difference of in-hospital mortality. Transfusion ratio did not differ between trauma and non-trauma patients and was not associated with mortality even when MTP activation duration and APACHE score were considered. Hospital-wide blood product use did not change with MTP implementation.

CONCLUSIONS

MTP may be successfully used in trauma and non-trauma settings without significantly impacting overall blood product utilisation. Inclusion of non-trauma patients into prospective studies of resuscitation with blood products is warranted to ensure benefit in these patients.

Pregnancy and thrombophilia

The physiological changes that occur during pregnancy create a hypercoagulable milieu. This hypercoagulable state is thought to be protective, especially at the time of labor, preventing excessive hemorrhage. The presence of hereditary or acquired causes of thrombophilia during pregnancy tilts the balance in favor of unwanted venous thromboembolism and adverse pregnancy outcomes due to vascular uteroplacental insufficiency. These adverse pregnancy outcomes include recurrent pregnancy losses, intrauterine fetal death, intrauterine growth retardation, preeclampsia and placental abruption. Much of the current data with regards to the association of the different thrombophilias and pregnancy-related complications are based on retrospectively designed studies. This lack of randomization, in-homogeneity of patient populations, varying case definitions, selection biases and inadequately matched control populations, have given rise to conflicting data with regard to screening for, and treatment of, pregnant women with suspected thrombophilias. The limited data that we have support the use of anticoagulant drugs for the prevention of pregnancy-related complications in the setting of thrombophilia. Heparin and low-molecular-weight heparins are the anticoagulant drugs of choice as they do not cross the placental barrier and, hence, do not cause fetal anticoagulation or teratogenicity. Warfarin can be used from the 12th week of gestation onwards but is preferably reserved for the postpartum period.

Is prophylaxis required for delivery in women with factor VII deficiency?

Factor VII (fVII) deficiency is a rare congenital bleeding disorder in which fVII activity level and bleeding tendency do not completely correlate. Pregnancy and delivery present a significant haemostatic challenge to women with fVII deficiency. Treatment with recombinant factor VIIa (rfVIIa) carries a thrombotic risk and the literature is not clear whether prophylaxis is necessary prior to delivery. The aim of this study was to define management, haemorrhagic and thrombotic complications of pregnant women with fVII deficiency through a systematic review. Medical databases (PubMed, MEDLINE, CINAHL, Academic Search Premier, Cochrane Library, Web of Science and Scopus) were searched using "factor VII deficiency" and "pregnancy" or "surgery." Overall 34 articles, four abstracts, and three institutional cases were reviewed. Literature from 1953 to 2011 reported 94 live births from 62 women with fVII deficiency. The median fVII activity was 5.5%. Haemostatic prophylaxis was used in 32% of deliveries. Without prophylaxis, 40 vaginal deliveries and 16 caesarean sections were completed. The odds of receiving prophylaxis were 2.9 times higher in women undergoing caesarean section compared to vaginal delivery. Post-partum haemorrhage occurred in 10% of deliveries with prophylaxis and 13% of deliveries without prophylaxis. The fVII level did not significantly differ between women who did and did not receive prophylaxis. We present the only systematic review of the management of pregnancy in fVII deficient women. No difference in post-partum haemorrhage was seen in deliveries with and without prophylaxis. Therefore, we recommend that rfVIIa be available in the case of haemorrhage or surgical intervention, but not as mandatory prophylaxis.

Whole blood tissue factor procoagulant activity remains detectable during severe aplasia following bone marrow and peripheral blood stem cell transplantation

Using a novel whole blood assay, we recently demonstrated that tissue factor procoagulant activity (TF PCA) is present in normal individuals. Preliminary experiments suggested that this activity is localized in the mononuclear cell fraction. Postulating that whole blood TF PCA would therefore be undetectable when monocytes and neutrophils are absent from peripheral blood, we assayed TF PCA during the peri-transplant period in 15 consecutive patients undergoing allogeneic (n = 12) or autologous (n = 3) bone marrow transplantation (BMT) or peripheral blood stem cell transplantation (PBSCT). Baseline (pre-transplant) mean TF PCA was higher in patients compared to normal controls (P <0.005). Unexpectedly, although TF PCA during the period of profound aplasia was significantly reduced compared to baseline (p <0.05), fully 55% of the initial activity remained detectable. During the engraftment phase, TF PCA returned to pre-transplant levels, with a linear correlation between monocyte counts and TF PCA (r = 0.63). In contrast to normal whole blood, incubation of aplastic samples with E. Coli lipopolysaccharide ex vivo failed to induce TF PCA. Throughout the period of study--but especially during the aplastic phase--the absolute number of circulating endothelial cells (CECs) that were TF antigen-positive was increased compared to normals (P <0.001). However, removal of these cells from whole blood samples failed to significantly diminish total TF PCA indicating that CECs alone could not account for the detectable TF PCA during aplasia. We conclude that neither circulating mature myelo-monocytic cells nor endothelial cells can account for all the functionally intact TF in peripheral blood. Further studies are needed to identify the other source(s) of TF PCA.