Citations

Introduction: Predictive algorithms using multiple clinical variables can identify patients at higher risk for abdominal aortic aneurysms (AAA), however they are cumbersome and cannot used directly by patients. A simple, cheap, noninvasive, and readily available (virtual) screening tool for identification of AAA could aid in screening algorithms. Aims: We sought to determine if a machine learning algorithm could predict the presence of an AAA by using a image of the face Methods: Diagnostic imaging studies were extracted from the electronic health record (EHR) from 5/7/2018 through 1/1/2023 and analyzed by a natural language processing algorithm previously validated to identify abdominal aortic aneurysms. Patient EHR profile pictures were extracted and matched to imaging studies. Various deep neural network (DNN) architectures were explored, all trained as classifiers on the face images along with clinical variables to predict clinical outcomes. All models used were CNNs with an EfficientNet architecture. Model performance was evaluated by standard metrics such as the area under the curve (AUC), specificities (Sp), and sensitivities (Sn). All performance metrics reported results from the test subset of the primary dataset and did not contain any observations considered during training. Results: A total of 5522 patients with facial images and AAA diagnostic studies were analyzed, among whom 1314 (23.8%) had a AAA. The optimal model had an AUC of 0.72 (Sn 0.44, Sp 0.83) for predicting AAA and remained similar in women (AUC 0.7), men (AUC 0.65), age >65 (AUC 0.63), and age <=65 (AUC 0.72) (Table). Conclusions: Using only an image of a patients face, the presence of an AAA could be identified with moderate performance in men, women, and in those less than 65. The model’s performance in demographics not typically screened for AAA may allow for targeted screening programs in these groups.

Introduction: Warfarin is a commonly prescribed anticoagulant for treating atrial fibrillation, mechanical valves, and venous thromboembolism. Warfarin dose management remains challenging due to dosing variability between patients and warfarin’s narrow therapeutic window. Time in therapeutic range (TTR) is critical to warfarin’s safety and efficacy, but TTR typically remains low (40-50%) in community practices. Specialized anticoagulation clinics and protocolized approaches can increase TTR but have great administrative burdens and health care costs. Aim: To develop standardized optimal warfarin dose decision support using a machine learning model based on time series anticoagulation data and patient demographic characteristics Methods: The dataset included 12,497 warfarin patients monitored in the anticoagulation tracker of electronic medical records across the Mayo Clinic Enterprise. The dataset contained time series anticoagulation data (warfarin dose, INR, INR target range) and patient demographic characteristics. We implemented an offline deep reinforcement learning model (DRL) to predict the cumulative warfarin dose for the days until the next INR test based on a patient’s historical anticoagulation data. Prior approaches utilized traditional supervised learning methods such as regression or Long Short Term Memory (LSTM) neural networks to approximate a function mimicking the behavior of the training data. On the other hand, DRL learned an optimal dosing policy through continuous interaction and feedback from the training data. The key advantage of DRL is the model can learn to behave differently (and potentially better) for suboptimal clinical states in the data such as overdosing or underdosing. To evaluate the DRL model we compared the predicted warfarin doses with the physician-prescribed doses Results: DRL model’s prediction accuracy was 96.96%, outperforming our implementation of a baseline LSTM model with a prediction accuracy of 70.58%. Further evaluation of the DRL model indicated that the model correctly adjusted the warfarin dose at time steps when patients had out-of-range INRs. Conclusions: Offline deep reinforcement learning demonstrates potential in supporting warfarin dose management to maximize TTR.

Introduction: Pulmonary vein thrombosis (PVT) is often diagnosed incidentally but data on its management and outcomes are sparse. Methods: We queried our institutional health record to identify patients ≥18 years of age diagnosed with PVT. Demographics, medications, risk factors, diagnostics, treatment, and sequelae were collected by manual chart review. Suspected tumor thrombus cases were excluded. Results: Seventy-two patients with PVT were identified (median age 62 years, 50% male). The cohort was mostly white (85%), overweight (BMI 26 kg/m), with frequent comorbidities including hypertension (44%), hyperlipidemia (38%), atrial fibrillation/flutter (29%), and history of venous thromboembolism (21%). The most common PVT risk factors were active malignancy (54%), lobectomy (21%), surgery within 30 days (19%), and extrinsic vein compression (15%). Most PVT were diagnosed by CT (86%) and affected a single vein (86%), mostly commonly the left upper PV (32%). Laboratory studies revealed median platelet count 243 х 10 9 cells/L (n=58) and D-dimer elevated in 6 patients when measured (67%, n=9). About one-third of patients were on therapeutic anticoagulation (37%) at PVT diagnosis. Most patients were started on new anticoagulation (78%, n=56), with unfractionated heparin as the most common initial therapy (68%). Of those anticoagulated, few were treated with a limited duration (20%, n=11, median 54 d, IQR 95) while most received indefinite anticoagulation (80%, n=45), primarily warfarin (45%), apixaban (24%), or enoxaparin (21%). Serial imaging was obtained in 69% (n=50, median 65 d, IQR 116) and demonstrated resolution of PVT in 32 (64%) patients. Common PVT complications noted were as follows: 4 year risk of LA thrombus (26%, 12/16 anticoagulated), mechanical ventilation (13%), pneumonia (9%), and ischemic stroke (9%, 5/6 anticoagulated). Conclusions: PVT is often associated with active malignancy, lobectomy, recent surgery, and extrinsic vein compression. Most PVT resolve on serial imaging. Clinicians should be aware of potential high-risk sequelae including LA thrombus and stroke as well as high anticoagulation failure rates when managing PVT.

Purpose: The aim of this study was to evaluate the frequency of ocular imaging in patients with hereditary hemorrhagic telangiectasia (HHT) at a large academic medical institution, characterize the choroidal vascular changes on indocyanine green angiography and optical coherence tomography, and explore the relationship of these ocular features with HHT genotypes and systemic manifestations. Methods: This is a retrospective review of patients with HHT who were seen in the Mayo Clinic ophthalmology department between 2011 and 2021. Results: Of the 71 patients with HHT identified, 7 patients (10%) had qualifying imaging defined as radial or raster spectral-domain optical coherence tomography scans and/or indocyanine green angiography. Four patients (57%) had enlarged choroidal vessels detected on optical coherence tomography and/or indocyanine green angiography. Three patients (43%) had retinal telangiectasias. Patients with choroidal changes (mean age 67 years) had the endoglin gene mutation, gastrointestinal arteriovenous malformations, and facial and mucocutaneous telangiectasias. By contrast, patients without choroidal changes (mean age 45 years) included the activin A receptor-like kinase 1 gene mutation. Conclusion: Only a small proportion of patients with HHT undergo ophthalmic imaging at the authors’ institution. The presence of choroidal abnormalities is associated with the endoglin gene mutation, older age, corroborating retinal vascular changes, and increased systemic manifestations.

Introduction: Use of validated risk stratification tools for work up of suspected pulmonary embolism (PE) is an endorsed recommendation of the American Society of Hematology and the American College of Emergency Physicians. Due to variations in clinical practice, electronic health record (EHR) systems, and medical provider trainings, the use, documentation, and implementation of these tools vary widely. Current processes do not allow for electronic capture and quality reporting at most institutions. Methods: We sought to design and implement clinical pre-test probability (PTP) tools at three large healthcare institutions for use in emergency departments (ED). At each site, a hematologist partnered with ED physician leadership and local information technology experts to design and implement PTP tools in clinical workflows. Overall, 38 EDs (high volume tertiary EDs and smaller regional EDs) located in the United States along the East Coast and Midwest regions were involved. A major goal of implementation was integration into clinical workflow and automatic documentation. After design and implementation, awareness and education of the new EHR tool was distributed electronically to ED providers. After implementation, use of PTP tools was analyzed between September 12, 2022 and January 11, 2023. PTP use was examined as percent of visits for which patients underwent CT pulmonary angiography (CTPA). Results: Each institution chose the 3 tier Wells' Score for implementation while one site designed an integrated Wells'/PERC or YEARS score calculator. Two sites implemented the tool as an optional flowsheet and one site implemented the tool as mandatory parameters within the orders for PE imaging studies. The mandatory tool could be bypassed with prespecified selections or if a D-dimer existed within 48 hours . Complete details of the site-specific implementation are shown in Table 1. Over the 4-month evaluation timeframe, there was a total of 270,214 ED encounters. Uptake and utilization of the PTP tools are shown in Table 2. Use of PTP was highest at the site with forced PTP documentation which ranged from 49-53% of ordered CTPAs, compared to Site 2 where use was 2-3%, and Site 1 where use ranged from 1-3%. At Site 1, use of PTP increased slightly over the study period with signs that PE yield on imaging was increasing as well (3.4% to 5.9%). At Site 2, PE yield on imaging was overall high (9-10%), and remained similar with similar use of PTP tools over the study period. Use of PTP and PE yield (6-8%) on imaging also remained similar throughout the study at Site 3. Conclusion: Forced use of PTP within the orders for CTPA led to the highest use of PTP (Site 3) but this did not result in meaningfully higher PE yield on CTPA compared to the other sites. PE yield at each site was at or above the 5% yield for PE on CTPA which has been previously reported. Given that PTP uptake was relatively stable over the measurement period at Site 2 and 3, this indicates the process had mostly stabilized and that other strategies are needed to improve uptake of PTP, reduce utilization of CTPA, and further increase yield on CTPA. Increasing PTP use and rising PE yield at Site 1 demonstrated ongoing potential for improvements past this study timeframe.

Intro: Venous thromboembolism (VTE) is a common cause of death among cancer patients, second only to cancer progression. Identifying factors predicting worse outcomes among cancer patients with VTE would be useful clinically. For cancer patients with pulmonary emboli (PE), it remains unclear whether concomitant leg deep vein thrombosis (DVT) as a source for further embolization adversely impacts this risk. The aim of this project was to determine the impact of residual DVT on adverse outcomes among cancer patients suffering acute PE. Methods: Consecutive patients with confirmed acute PE (March 1, 2013 - June 30, 2021) underwent ultrasound (US) imaging and were divided into two groups depending on the presence or absence of leg DVT. PE severity was classified based on AHA criteria. Patients were followed prospectively for VTE recurrence, major and clinically relevant non-major bleeding (CRNMB) and all-cause mortality. Results: Over the study interval, 1,035 cancer patients with PE were enrolled (Table 1). Of these, 375 (36.2%) had leg DVT confirmed by US. Patients with leg DVT were older (64.2 vs 61.4 years, p = 0.01), heavier (90.2 vs 85.1 kg, p &lt;0.001), with lower metastatic disease prevalence (49.1% vs 57.1%, p = 0.01) compared to those without leg DVT. Cancer type distribution also differed between groups. PE patients with DVT more often had prostate, genitourinary, and hematologic cancers while those without DVT more often had pancreatic and breast cancers. Patients with DVT more often had symptomatic PE (48.2% vs. 35.9%, p &lt; 0.001), massive (2.9% vs 0.5%, p &lt; 0.001), and submassive (22.0% vs 7.6%, p &lt; 0.001). Patients with DVT more often received unfractionated heparin (39.5% vs 18.7%, p &lt; 0.001) whereas those without DVT were more frequently treated with low molecular weight heparin (57.5% vs. 44.3%, p &lt;0.001), and apixaban (16.4% vs. 11.7%, p &lt; 0.05). Anticoagulant duration, VTE recurrence, major bleeding and CRNMB were similar between groups (Table 2). Conclusion: Among a large cohort of cancer patients with pulmonary emboli, significant clinical and cancer specific differences exist comparing those with and without leg DVT.While theoretically increasing the risk of recurrent embolization, the presence of residual DVT does not appear to adversely impact important outcomes among cancer patients adequately treated with anticoagulation.

Introduction: ASH Guidelines for diagnosis of pulmonary embolism (PE) start with risk assessment using a pretest probability tool (PTP) followed by D-dimer testing or CT pulmonary angiography (CTPA) depending on risk. We aimed to develop an electronic clinical quality measure (eCQM) to encourage broader use of a validated PTP scoring tool in emergency departments (EDs). The Centers for Medicare & Medicaid Services require reporting of eCQMs which consist of value sets or lists of standardized codes linked to patient data within the electronic health record. We sought to identify a standardized way to identify CTPA and abnormal D-dimer tests using administrative codes (i.e. CPT, LOINC, SNOMEDCT codes) across three institutions. Methods: Value sets identifying CTPA and D-dimers were reviewed from existing eCQMs. Three institutions representing 38 EDs from across the country submitted data from all ED encounters between September 12, 2022 and January 11, 2023. Imaging types were reviewed from each of the CPT codes and LOINC (Logical Observation Identifiers Names and Codes). We determined the number of ED encounters, pre-test probability tool use, and diagnosis of PE using different codes. We calculated sensitivity, specificity, positive and negative predictive value (PPV and NPV, respectively). Results: Over the 4-month study period, 270,214 encounters were identified across 38 EDs. Each institution identified CTPA with site-specific codes (Table 1). The Full CTPA Value Set identified 55% more encounters than the site-specific codes (Full set 15,054/9679 site-specific codes). LOINC code 88322-3 was only used at one site. CPT code 71275 had the best sensitivity and specificity, but still had false positives (PPV 82%) (Table 2). D-dimer values were identified as LOINC code 48065-7 and 91556-1 at Site 1 and 48067-3 at Site 2. SNOMEDCT codes were not used at any site despite being the way other eCQMs identify elevated D-dimer results. Different D-dimer tests with different normal ranges were used at each site. All sites did not have an electronically extractable positive indicator for D-dimer. Conclusion: Due to persistent false positives using CPT Code 71275 for identification of CTPA, using this code would lead to inclusion of patients for whom PTP was not required or recommended. Additionally, coding for an abnormal D-dimer test result are not standardized across institutions. Therefore, administrative codes cannot be used to develop eCQMs whose aim is to evaluate whether CTPA is ordered appropriately based on the PTP risk level and laboratory testing.

Introduction Pulmonary Embolism (PE) is a major cause of morbidity and mortality worldwide. Diagnosis is essential as untreated PE has a 30% mortality rate. Currently, computer tomography pulmonary angiogram (CTPA) is the standard of care to diagnose PE. However, this modality is expensive, exposes patients to radiation and increases risk of contrast induced nephropathy. Furthermore, only 5% of CTPAs performed are positive for PE signifying that many unnecessary tests are being performed. To address this gap, Pre-Test Probability (PTP) tools have been developed to safely exclude patients with low risk of PE and limit iatrogenic harm from over testing. Despite the tools being developed over 20 years ago, they are not widely available in electronic health record (EHR) systems and even when they are built in, uptake remains low. Methods Three large integrated health systems using Epic EHR systems across five states (MA, WI. MN, AZ, FL) built electronic PTP assessments into their EHR systems to safely exclude patients with low risk of PE and limit over testing. Two sites made the PTP optional, and one site made it a forced function (providers were forced to complete PTP prior to ordering CTPA unless a D-dimer result was present in the prior 48 hours). To assess the usability of these EHR tools, the team conducted a survey with ED clinicians across the three test sites. The survey was created to identify which PE risk stratification methods were employed by ED clinicians (EHR embedded PTP tool, apps on their phone, web applications like MDcalc, calculations from memory, or if they relied on clinical gestalt), and to identify barriers to using the EHR embedded PTP. Results Site 1 had 47 respondents, Site 2 had 94 respondents, and Site 3 had 16 respondents. Two key findings emerged from the survey (Figure 1). First, clinicians used a variety of risk stratification approaches and few of them documented their risk assessments in structured fields in the medical record. Second, clinical gestalt was the most used PE risk stratification approach, and this was true across all three test sites. In addition, the health system with a PTP prompt before clinicians could proceed with CTPA orders (Site 3) did not have a higher percentage of clinicians who reported using the EHR embedded tool than sites that did not prompt clinicians to fill in the PTP. Clinicians cited a variety of different barriers to using the EHR embedded PTP tool (Table 1). Conclusion While many clinicians indicated intermittent use of PTP criteria to aid in the diagnosis of PE, risk factors and scores from phone apps, web tools, and calculations from memory are unlikely to be documented in the patient's medical record, which means that they are not accessible to all members of the ED care team and cannot be used for electronic clinical quality measure. The biggest hurdles to increasing the use of structured PTP assessments in the ED setting to safely exclude PE and limit iatrogenic harm from over testing are (1) obtaining buy-in from ED clinicians that the PTP tools provide value over clinical gestalt, at least for a subset of patients who could most benefit, and (2) identifying best practices for incorporating PTP tools into EHR systems that make the tools easier to access and use 3) developing and implementing effective education and awareness programs around the PTP tools. This survey adds to the body of literature on PTP assessments in the ED setting and can help inform further research that is needed to continue to improve the diagnosis of PE. Funding: This project is funded in part by the Gordon and Betty Moore Foundation through Grant GBMF10778 to the American Society of Hematology.

Purpose/Introduction: Patients with cancer and venous thromboembolism (VTE) have higher complication rates including thrombosis recurrence and bleeding. Real world prospective clinical outcome estimates of VTE management comparing cancer and non-cancer patients are limited. To assess VTE recurrence, major bleeding, and clinically relevant non-major bleeding (CRNMB) in patients with cancer and without cancer, the prospective Mayo Clinic Thrombophilia Clinic Registry was analyzed. Methods: Upon recruitment, consecutive patients with confirmed acute VTE (03/01/2013 - 04/30/2023) were treated in a standardized, guideline-sanctioned, protocol driven strategy incorporating shared patient-decision making. Patients were divided into groups based on cancer status. After enrollment, patients were actively followed at 3 month intervals, in person whenever feasible, by mailed questionnaire or scripted phone interview to assess vital status, medication compliance, VTE recurrence, major bleeding and CRNMB. Results: Over the study time-frame, 2,064 patients (53.8% male, 46.2% female) with a cancer and 2,647 patients (54.9% male, 45.1% female) without cancer were enrolled. The most common cancers were gastrointestinal (n=423, 20.5%), pancreatic (n=287, 13.9%), genitourinary (n=198, 9.6%), hematologic (n=171, 8.3%), and lung cancer (n=170, 8.2%). Patients with cancer were older, had lower weight and lower platelet counts compared to non-cancer patients (Table1). Pulmonary embolism (PE, 52.6% vs 43.7%, p&lt;0.001), upper extremity DVT (9.1% vs 6.0%, p&lt;0.001), and splanchnic DVT (11.1% vs 6.9%, p&lt;0.001) were more frequent among cancer patients. In contrast, leg DVT was more frequent among non-cancer patients (65.4% vs.47.2%, p&lt;0.001). While mean duration of anticoagulation was similar between groups, notable differences in 3 month and &gt; 9 month durations were evident as were initial anticoagulant choices (Table 1). Despite a well-organized protocol driven and guideline sanctioned management strategy, patients with active cancer experienced a 2.2-fold higher rate of VTE recurrence (p&lt;0.001) and a 1.8 fold higher rate of major bleeding (p&lt;001) compared to non-cancer patients (Table 2). CRNMB rates did not differ by cancer status. Conclusions: In this large prospective, guideline-driven and patient-centric registry of VTE management, patients with cancer had significantly higher rate of VTE recurrence and major bleeding, compared to non-cancer patients. These data provide important estimates for power calculations for future randomized trials of VTE treatment.

Background: Reduced or low-dose direct oral anticoagulants (DOACs) have been studied in randomized control trials for the extended prevention of venous thromboembolism (VTE) after 6 months of treatment at full, therapeutic doses. The real-world effectiveness as well as utilization of this approach in clinical practice compared to continuing at therapeutic doses is unclear. Aims: Examine the effectiveness, safety and utilization of reduced-dose DOACs in cancer patients and overweight patients. Methods: Consecutive patients with VTE were identified using the Mayo Clinic VTE Registry from March 1st, 2013 to December 31st, 2021. Patients were followed prospectively for outcomes of VTE recurrence, death, major bleeding and clinically relevant non major bleeding (CRNMB) either in person or by survey. Patients with recurrent VTE or bleeding during the first 3 months were excluded from further analysis. After completion of anticoagulation treatment for 3 months with either rivaroxaban or apixaban, patients continuing anticoagulation were evaluated in a nested case-control study. Results: A total of 404 patients (283 (70%) on apixaban and 121 (30%) on rivaroxaban) were identified in the low dose DOAC and 3060 in the full dose anticoagulation group. Overall, the mean age was 60.7 years (SD 14.4), mean weight was 89.9 kg (SD 23.8) and 55.1% were males. Within the full cohort, patients with active cancer were less likely to be prescribed low dose DOACs (HR 0.56; 95%CI 0.42-0.73; p&lt;0.001). However, low dose DOAC prescription was more likely in patients with pulmonary embolism alone (HR 1.48; 95%CI 1.21-1.80; p&lt;0.001) or in combination with DVT (HR 1.54; 95%CI 1.25,1.90; p&lt;0.001). Patients transitioned to low dose DOACs had similar age, sex and weight compared to the full dose anticoagulation group, however, fewer patients with weight ≥120 kg had reduced dosing (HR 0.68; 95%CI 0.46,0.99; p=0.046). The mean time to start of a low dose DOAC was 5.1 months overall, 5.4 months in cancer patients, and 3.6-4.9 months in non-cancer patients (Table 1). Non-cancer patients had significantly lower transition numbers in those ≥120kg (HR 0.62; 95%CI 0.40,0.97; p=0.038). In non-cancer patients In comparing weight, ≥120kg to &lt;120kg, in the low dose group, only death was statistically different with higher risk in those ≥120kg (HR 6.44; 95%CI1.27,32.81; p=0.025). Cancer patients transitioned to low dose DOACs around 5 months on average with a median of 3.3 months. Apixaban continued to be the dominant DOAC used (78%) in the low dose group. The cancers with the most patients on low dose DOACs include lymphoma, prostate, melanoma, ovarian, and breast. There was no significant difference in VTE recurrence or death in the cancer group on different DOAC doses. Conclusion: Our results show that anticoagulant prescribers are not always waiting until 6 months to transition to low doses and low doses are being utilized in cancer patients despite lack of cancer-specific studies during the analyzed time period. Although the sample size limits the precision of this analysis, we did not find statistically significant increases in VTE recurrence with low dose DOACs in either cancer patients or with elevated body weight. We did find CRNMB but not MB was observed more frequently in weights above 120kg, possibly from unadjusted confounding in this unadjusted analysis.