Resident Excellence in Research and QI

Aneliya San, MD; Ilan Goldenberg, MD, Arwa Younis, MD, Kris Cutter, MS2, Scott McNitt, MS2, Nona Sotoodehnia, MD, Peter J. Kudenchuk, MD, Thomas D. Rea, MD, MPH, Dan E. Arking, PhD, Bronislava Polonski, MS2, Wojciech Zareba, MD, PhD, Mehmet K. Aktas, MD, MBA

Introduction:

QT prolongation is a known adverse effect of many medications, including class III antiarrhythmics such as dofetilide and sotalol, which block the IKr potassium current and increase the risk of torsades de pointes and sudden cardiac death. Women are disproportionately affected by drug-induced QT prolongation and associated arrhythmias, potentially due to sex hormone–mediated differences in cardiac repolarization. The influence of endogenous hormonal fluctuations during the menstrual cycle on arrhythmia risk in women taking QT-prolonging medications remains unclear. We hypothesized that variations in sex hormone levels across the menstrual cycle alter cardiac repolarization dynamics in this population.

Methods:

We prospectively enrolled 41 women—20 treated with either dofetilide or sotalol and 21 healthy controls. Each participant underwent three separate 7-day ambulatory ECG monitoring periods corresponding to different phases of the menstrual cycle. On the first day of each ECG period, concurrent saliva samples were collected for measurement of estradiol, progesterone, and testosterone. The primary ECG metrics analyzed were QT-Apex (an indicator of early repolarization) and QT interval (total repolarization time), both corrected for heart rate. Associations between hormone levels and repolarization indices were evaluated using linear mixed-effects regression models.

Results:

Among women treated with QT-prolonging drugs, QT-Apex duration showed a statistically significant inverse association with the progesterone-to-estradiol ratio (p = 0.018) and testosterone levels (p = 0.026), and a direct association with estradiol levels (p = 0.004). Similarly, QT interval duration was inversely associated with the progesterone-to-estradiol ratio (p = 0.012). No significant correlations between sex hormones and ECG parameters were observed in the control group.

Conclusion:

This study demonstrates that cardiac repolarization dynamics vary in relation to menstrual cycle–related hormone fluctuations in women treated with QT-prolonging drugs, but not in healthy controls. Specifically, increased estradiol levels and lower progesterone-to-estradiol ratios are associated with longer QT intervals, which may elevate the risk of ventricular tachyarrhythmia during specific menstrual phases. These findings suggest the need for personalized arrhythmia risk assessment and monitoring strategies in women undergoing treatment with QT-prolonging agents.

Natalia Paone, M.D., Erin Armenia M.D., J.P. Iskandar M.D., Piotr Karmilowicz M.D., Ryan O’Connor M.D., David Bass M.D., Jeffrey Alexis M.D., Sabu Thomas M.D., Frank Passero M.D., Farhan Bajwa M.D., Muhammed Inamullah Nouman M.B.P.S., Maria L. Mackin C.N.M.T., Ronald G. Schwartz M.D. M.S.

Introduction

Transthyretin amyloidosis can cause restrictive cardiomyopathy (ATTR CM), heart failure, and death. Metabolic imaging of ATTR CM with SPECT using bone-avid radiopharmaceuticals Tc-99m HMDP, PYP, and DPD in the absence of paraprotein has become a diagnostic standard of care, providing high sensitivity and specificity without routine endomyocardial biopsy (1) or cardiac MRI. Diagnostic challenges of SPECT include limited spatial resolution to ascertain radiopharmaceutical location (myocardium, blood pool, extra-cardiac), and variability of regional myocardial and rib uptake, affecting Perugini score (2). Both SPECT CT (3) and simultaneous dual isotope (SDI) SPECT with thallium-201 myocardial perfusion imaging (MPI) (4,5) have been reported to improve accuracy. An ultra-low radiation exposure alternative is serial dual radiopharmaceutical (SDR) CZT SPECT which substitutes Tc-99m sestamibi for thallium-201 to define precisely myocardium and assess its metabolic uptake of Tc-99m HMDP. However, SDR involves separate image acquisitions which may introduce motion artifact. This study aims to evaluate the incremental diagnostic value of SDR CZT SPECT compared to HMDP imaging alone and assess the impact of patient motion on diagnostic accuracy.

Methods

Secondary analysis of the quality assurance RedCap database of patients undergoing clinically indicated CZT SPECT imaging without need for research consent was approved by the RSRB. Two imaging experts retrospectively reviewed axial SPECT and SDR images of 101 consecutive patients referred for ATTR CM evaluation at URMC Nuclear Cardiology Laboratory (January 2023 to February 2024). Routine clinical protocol of SDR CZT SPECT includes metabolic imaging (10-12 minutes) at 60 minutes after injection of Tc-99m HMDP (8.0–9.9 mCi) followed by MPI (5 min) after Tc-99m sestamibi (8.0–9.9 mCi). Axial SPECT and planar images were compared to SDR imaging, categorized as positive, equivocal, or negative. Diagnostic reclassification, image quality, and motion impact on diagnosis were analyzed using Chi-square tests.

Results

Of 101 patients, 22 were diagnosed with ATTR CM. HMDP SPECT identified 20 positive, 23 equivocal, and 58 negative cases. SDR reclassified all 23 equivocal cases, confirming 2 additional positives (10% increase, p<0.005) and 21 negatives (36% increase, p<0.0001). Reclassified positives were supported by biopsy (n=1) and echocardiography with Tafamidis treatment (n=1). Image quality was excellent/good without motion in 85 cases and minor motion without diagnostic impact was noted in 16 cases.

Conclusion

SDR enhances accuracy of ATTR CM diagnosis using high-efficiency, high-resolution CZT SPECT by resolving equivocal cases on axial SPECT and planar imaging. Advantages of CZT SPECT SDR include rapid imaging (<17 min) with an 80-minute protocol, ultra-low radiation exposure (<5 mSv), and expedited referral for treatment without routine biopsy or cardiac MRI. These findings support SDR CZT SPECT as the preferred initial imaging modality, improving accuracy, efficiency and cost effectiveness. Further studies comparing SDR to SPECT and SPECT CT are warranted.

References

1. Hage FG et al. American Society of Nuclear Cardiology quality metrics for cardiac amyloid radionuclide imaging. J Nuclear Card 2024; 40: 102041. https://doi.org/10.1016/j.nuclcard.2024.102041.
2. Sperry BW, Vranian MN, Tower-Rader A, et al. Regional variation in technetium pyrophosphate uptake in transthyretin cardiac amyloidosis and impact on mortality. J Am Coll Cardiol Img 2018; 11:234–42.
3. Taha ZA, Alibazoglu D, Sabbour H, et al. Attacking the Achilles heel of cardiac amyloid nuclear scintigraphy: How to reduce equivocal and false positive studies. J. Nucl. Cardiol. 2023;30:1922–34. doi:10.1007/s12350-023-03214-6
4. Tamarappoo B, Otaki Y, Manabe O, et al.. Simultaneous Tc-99m PYP/Tl-201 dual-isotope SPECT myocardial imaging in patients with suspected cardiac amyloidosis. J Nucl Cardiol. 2020 Feb;27(1):28-37. doi: 10.1007/s12350-019-01753-5. Epub 2019 Jun 6. PMID: 31172386.
5. Armenia EM, Mackin ML, Karmilowicz P, et al. What is this image? 2023 image 1 results: Precision high-resolution CZT SPECT imaging to evaluate cardiac aTTR amyloidosis, Journal of Nuclear Cardiology 2023; 30: 1301-1307. https://doi.org/10.1007/s12350-023-03279-3

Filip Koritysskiy, M.D. and Bryan Redmond

Background:

Guideline-directed medical therapy (GDMT) is the cornerstone of heart failure (HF) management, but older adults may receive

suboptimal treatment due to comorbidities, polypharmacy concerns, or implicit provider biases. Hospital admission and discharge

represent key opportunities for medication optimization, permitting us to assess whether older adults receive GDMT at rates

comparable to younger adults. This study investigates differences in prescription rates of GDMT for HF — including beta-blockers,

renin-angiotensin system inhibitors (RASi), mineralocorticoid receptor antagonists (MRAs), and sodium-glucose co-transporter 2

inhibitors (SGLT2i) — across age groups at both admission and discharge.

Methods:

We conducted a retrospective cohort study of adult patients hospitalized for heart failure with an ejection fraction less than 40% at the

University of Rochester Medical Center over a 24-month period. Patients were stratified into two age groups: <65 years (younger

adults) and ≥65 years (older adults). GDMT prescription rates were extracted from electronic health records at both admission and

discharge. The primary outcome included the prescription rates of each GDMT class at admission and discharge. We used McNemar’s

tests to compare within group differences in medication use.

Results:

A total of 894 patients met inclusion criteria, with 45.2% aged ≥65. At admission, prescription rates were lower amongst older adult

patients for all medications except ACE inhibitors. However, all medications evaluated — including ACE inhibitors, ARBs, ARNIs,

mineralocorticoid receptor antagonists (MRAs), beta blockers, and SGLT2 inhibitors — showed significant increases in prescription

rate from admission to discharge for both young and older adult cohorts (McNemar's p < 0.01 for all comparisons). Importantly,

prescription rates for ARNIs and SGLT2 inhibitors increased substantially in both age groups, reflecting rapid uptake of recent

guideline updates.

Conclusions:

Hospitalization presents a critical opportunity to initiate or optimize guideline-directed medical therapy in patients with HFrEF. Our

findings demonstrate that both older and young adult patients alike experience significant improvements in heart failure

pharmacotherapy during admission, with high rates of initiation of novel agents such as ARNIs and SGLT2 inhibitors. Critically, our

older adult cohort showed equally strong or stronger medication optimization than their younger counterparts. These results

underscore hospitalization as a powerful opportunity to improve GDMT delivery, challenge outdated assumptions about age-related

treatment limitations, and reinforce the importance of equitable, evidence-based care across the lifespan.

Charles E Springer¹, AnnaLynn M. Williams¹, Andrea M Baran¹, Pinguang G Yang¹, Tina Faugh¹, Phil Rock¹, Jonathan W Friedberg¹, Paul M Barr¹, Patrick Reagan¹, Carla Casulo¹, Andrew G Evans¹, Walter R Burack¹, Clive S Zent¹

¹University of Rochester Medical Center, Rochester, NY

Introduction: Lymphoid malignancies are the fourth most prevalent cancer and rank as the sixth leading cause of cancer-related fatalities in the United States(1,2). B cell malignancies are classified according to the presumed stage of cellular maturation by analysis of morphology, histology, immunophenotype, and molecular characteristics(1,3). Survivors of lymphoid malignancies have been reported to have higher risks of developing second malignancies when compared to the general population(3,4). The cause of this increased propensity currently is unknown, but could be associated with genetic, immunologic, lifestyle, and environmental factors. We hypothesize that patients with a primary diagnosis of a B cell malignancy are at an increased risk of acquiring clonally unrelated, additional B cell lymphoid malignancies compared to a demographic-matched general population with no prior B cell malignancy.

Methods: An existing cohort of 4,325 patients at the University of Rochester Wilmot Cancer Institute (WCI) includes all adult patients with a confirmed diagnosis of primary lymphoid malignancies with a first visit to the WCI from April 1, 2014 to April 1, 2024. The reference population is the Surveillance, Epidemiology, and End Results (SEER) national cancer surveillance database, containing data from years 1975 to 2021 from large segments of the total population of the United States. This database serves as a representative sample of the United States population.

Results: We have determined the clonal relationship between initial and subsequent lymphoid malignancies for those of the WCI cohort with multiple diagnoses. We have estimated the prevalence and age- and sex-adjusted incidence rates of development of additional, clonally unrelated lymphoid malignancies in patients in the WCI mature B cell malignancy cohort. We now are comparing the WCI cohort’s incidence of an additional lymphoid malignancy to the expected rate of occurrence of initial lymphoid malignancies in the general population of the SEER database. This is achieved by using a Poisson Model to calculate age- and sex-adjusted standardized incidence ratios. Lastly, we will identify potential host factors associated with multiple B cell malignancies.

Conclusion: There are very limited investigations that quantify differences in risk of subsequent lymphoid malignancies in patients with an established diagnosis of one B cell malignancy(3). Understanding of the epidemiology and clinical presentation of additional B cell malignancies in patients with other lymphoid malignancies offers valuable insights to patient care. Determination of clonality is pertinent because those with a clonally related, transformed disease have a far poorer prognosis, whereas those with a subsequent but clonally distinct mature B cell malignancy have the same general prognosis as would anyone else with that same, specific, initial disease. As advancing medical treatment is leading to an increasing number of survivors of lymphoid malignancy, improving our knowledge of risk factors can provide pertinent information for physicians and survivors alike(2).

References:

1. Teras LR, D. C., Cerhan JR, Morton LM, Jemal A, Flowers CR (2016). "US lymphoid malignancy statistics by World Health Organization subtypes." CA Cancer J Clin 66(6): 443-459.

2. American Cancer Society, I. (2023). "American Cancer Center 2023 Facts and Figures."

3. Parikh SA, M. P., Zent CS, Evans AG (2020). "Multiple B cell malignancies in patients with chronic lymphocytic leukemia: epidemiology, pathology, and clinical implications." Leuk Lymphoma 61(5): 1037-1051.

4. Philip J. Meacham, A. M. W., Myla Strawderman, Andrea M., et al. (2020). "Additional B cell malignancies in patients with chronic lymphocytic leukemia/small lymphocytic lymphoma (CLL)." Leukemia Lymphoma 61(7): 1636-1644.

Rohith Palli, M.D., Ph.D., Jonathan Herington, and Nicole Wilson

Introduction: Advances in artificial intelligence (AI) have created a rapid proliferation of healthcare-related algorithms, but effects on health equity remain unclear. AI algorithms can instigate or perpetuate disparities in outcomes between protected groups [1–3]. Disparities can arise from underlying data, algorithms, application of results, or combinations of these.

Protected characteristics include race, gender, and social risk. Several indices have been used to assess geographic-associated social risk, including Area Deprivation Index (ADI) [4,5], developed specifically to assess healthcare outcomes with weight on socioeconomic status, and Social Vulnerability Index (SVI) [6], an index to assess resilience to disasters.

When using AI to predict necessary resources to treat traumatically injured patients upon arrival to the Emergency Department, biases may arise that result in unfair resource allocation. The Wilson ECLIPSe lab developed an AI algorithm that outperforms current manual triage of injured patients by predicting trauma activation level [7]. We tested this AI algorithm for bias and attempted to ameliorate any identified bias(es) using a tool designed to constrain an algorithm to various measures of fairness. We hypothesized that biases could be detected (and subsequently ameliorated) in resource allocation using four protected characteristics: gender, race, ADI, and SVI.

Methods: After IRB approval, training data and output from the published algorithm were used to construct descriptive bar plots aimed at demonstrating disparities with respect to accuracy for the four protected characteristics: gender, race, ADI, and SVI. Using cross-validation, we then generated logistic regression models to fit maximum difference between groups within protected characteristics to a range of maximums (0.01, 0.02, 0.05, 0.1, 0.5, 1.0) for true positive rate (TPR), false positive rate (FPR), and overall error (= FPR + false negative (FNR)  / TPR + true negative (TNR)). Calculations were performed using Jupyter [8] in a Conda [9] environment with Fairlearn version 0.10 [10].

Results: Large disparities were identified across all four characteristics (all p < 10^-9 by chi-square). Attempts to ameliorate disparities produced results dependent upon disparity constrained and extent of constraint. For the strictest 0.01 and 0.02 maximum difference, tradeoffs were frequent. For example, individuals with low SVI (= low social risk) had disproportionately low TPR (appropriate allocation of resources), which, when constrained to limit disparities in TP, led to increased FP (inappropriate allocation of resources, both p < 0.001 by ANOVA). Less stringent maximum difference limits (0.05, 0.1, 0.5, 1.0) consistently produced little effect. 

Conclusion: Algorithms created from healthcare data, even using rigorous methodology, will likely produce disparities. Understanding where and how disparities arise is key to creating trustworthy AI that can be confidently used in the clinical setting. However, disparity amelioration involves tradeoffs in accuracy and requires careful tailoring to the associated clinical and ethical scenarios.

Citations:

1. Vyas, D. A., Eisenstein, L. G. & Jones, D. S. Hidden in Plain Sight — Reconsidering the Use of Race Correction in Clinical Algorithms. https://doi.org/10.1056/NEJMms2004740 383, 874–882 (2020).
2. Rajkomar, A., Hardt, M., Howell, M. D., Corrado, G. & Chin, M. H. Ensuring fairness in machine learning to advance health equity. Ann. Intern. Med. 169, 866–872 (2018).
3. Obermeyer, Z., Powers, B., Vogeli, C. & Mullainathan, S. Dissecting racial bias in an algorithm used to manage the health of populations. Science 366, 447–453 (2019).
4. Stephens, C. Q. et al. Comparative Analysis of Indices for Social Determinants of Health in Pediatric Surgical Populations. JAMA Netw. open 7, e2449672 (2024).
5. Kind, A. J. H. & Buckingham, W. R. Making Neighborhood-Disadvantage Metrics Accessible — The Neighborhood Atlas. N. Engl. J. Med. 378, 2456–2458 (2018).
6. Registry, U. S. D. of H. and H. S. C. for D. C. and P. A. for T. S. and D. The Social Vulnerability Index (SVI). (2018). doi:10.3886/E101771V1
7. Liu, C. W. et al. Machine Learning Improves the Accuracy of Trauma Team Activation Level Assignments in Pediatric Patients. J. Pediatr. Surg. 59, 74–79 (2024).
8. Kluyver, T. et al. Jupyter Notebooks – a publishing format for reproducible computational workflows. Position. Power Acad. Publ. Play. Agents Agendas - Proc. 20th Int. Conf. Electron. Publ. ELPUB 2016 87–90 (2016). doi:10.3233/978-1-61499-649-1-87
9. Anaconda Software Distribution. Anaconda Doc. (2020). at <https://docs.anaconda.com/>
10. Bird, S. et al. Fairlearn: A toolkit for assessing and improving fairness in AI. (2020).

Anne Zhang, M.D., Dominick Roto, D.O.

Introduction: A physician’s ability to prognosticate the critically ill patient is a vital skill, though little is known about trainees’ ability to prognosticate, and how this may vary with clinical experience. The purpose of this study was to evaluate internal medicine residents and their ability to predict short- and long-term morbidity and mortality in critically ill patients, and how this varies by experience level. 

Methods: This was a single center observational study. Using RedCap, we developed a survey where residents were asked to identify patients under their direct care in the medical intensive care unit and respond to the following “Yes/No” questions for each patient: “Do you expect this patient will be alive in (X) month(s)?” and “Do you expect this patient will be both discharged from the hospital AND not readmitted in X month(s)?”. These two questions were used to assess mortality and morbidity outcome predictions, respectively. We asked the same questions at 1-month and 6-month time frames to assess short term and long-term outcome predictions. Respondents also listed variables that influenced their predictions. Using eRecord data, these predictions were blindly compared to actual patient outcomes. A two-tailed Z-test for proportions was conducted to compare the percentage of accurate predictions; a p value ≤ 0.05 was considered statistically significant. Participation was voluntary and uncompensated.

Results: From February 2024 to November 2024, 126 residents rotating through the medical intensive care unit received the survey; 30 surveys were completed (24% response rate) with predictions on 97 patients. Residents correctly predicted 1-month mortality in 78% (76/97) and 6-month mortality in 65% (56/86*) of cases (Z=1.99; p= 0.049*). Morbidity predictions were accurate in 78% (76/97) at 1 month and 78% (66/85*) at 6-months (Z=0.11, p=0.91). Amongst inaccurate predictions, residents often overestimated survival, predicting patients would survive at 1 month (17/21, 81%) and at 6 months (17/30, 57%). However, with morbidity predictions, residents tended to over-estimate on-going hospitalization status at 1 month (14/21, 67%) and 6 months (15/19, 79%). Common factors cited included admitting diagnosis, length of stay, and advanced age.

Conclusions: Internal medicine residents at Strong Memorial Hospital in Rochester, NY demonstrated reasonable accuracy in predicting short- and long-term outcomes in the critically ill patient. Residents were better at predicting short-term mortality compared to long-term mortality. Residents tended to be overly optimistic when asked to predict mortality. Interestingly, this did not appear to change based on training year. This small study serves to highlight that, while resident physicians can offer reasonably accurate prognostication, there is still substantial variability in outcomes, emphasizing the need to acknowledge one’s own limitations. 

*6-month outcome predictions: the total number of predictions does not total 97 because some time points have not been reached. The date of the last survey collected is 11/07/2025. Data collection ends 05/07/25. If selected, results will be updated for the final presentation.

Joseph Glick, M.D. (resident), Nina Rizk, M.D. (resident), Kavya Bana, M.D. (resident), Josiah Miller, M.D. (resident), Teresa Shannon, R.N.  (nurse manager), Catherine Glatz, M.D. (6-3400 Unit Medical Director), and Meghan K. Train, D.O. (6-1400 Unit Medical Director) 

Introduction: Excessive laboratory testing performed on hospitalized patients is harmful, costly, and often does not provide information that impacts clinical decision making. It can lead to iatrogenic anemia, additional costs, and reflex ordering of further unnecessary tests^1,2. Our project sought to decrease excessive laboratory testing across two medicine resident-staffed medical-surgical units at a quaternary care center with the aim to decrease median laboratory tests per patient stay by 10% over a 4-month period.

Methods: Automated electronic medical record (EMR) data mining was utilized to collect baseline data on the total number of laboratory tests performed per patient encounter. Encounters with a length of stay greater than 10 days were excluded to eliminate outliers, and lab tests classified as “Point of Care Testing” (POCT) were excluded to eliminate fingerstick glucose results. Manual verification was performed on an early data set. Data was broken into monthly subsets, and descriptive statistics were obtained for each month. Through a series of PDSA cycles, numerous interventions including visual aids, email reminders, check lists, and incorporating discussion of the frequency of laboratory testing into work and interdisciplinary rounds were utilized to improve "mindful" ordering by team members. At the end of the 4-month intervention period, and for an additional 12 months afterwards, the median number of laboratory tests per encounter was determined. A subsequent survey was sent to faculty and residents to collect quantitative and qualitative data on the project.

Results: A 12-month retrospective analysis revealed a baseline median of 11 laboratory tests per patient encounter, with an average of 130 encounters per month. Our interventions were successful in reducing the median total number of laboratory tests performed per encounter by 18%, from 11 to 9, during the intervention period. This was not sustained, however, with the median number of tests returning to the pre-intervention baseline for the subsequent 12 months. Survey results showed that most respondents (59%) felt comfortable with de-escalating labs on stable patients, and many (41%) reported discussing lab reduction least weekly. Analysis of qualitative themes demonstrated interest in utilizing the EMR to reduce the default number of days labs are ordered on admission, adding reminder functions about lab frequency to the EMR, and formalizing additional discussion about lab reduction on interdisciplinary and afternoon rounds.

Discussion: Through multiple interventions, a measurable reduction in total number of laboratory tests per patient encounter was observed, suggesting that routine laboratory testing may be excessively utilized for hospitalized patients. This result was transient, however, and returned to the baseline median after the intervention period. Future efforts will be aimed at implementing survey suggestions to promote a sustained response. Limitations included a small sample size and involvement of only resident staffed units.

References: 

Eaton, K. P., Levy, K., Soong, C., Pahwa, A. K., Petrilli, C., Ziemba, J. B., ... & Parsons, A. S. (2017). Evidence-based guidelines to eliminate repetitive laboratory testing. JAMA internal medicine, 177(12), 1833-1839.

Almeqdadi, M., Nair, H. K., Hill, J., Sanchez-Cruz, J., Nader, C., & Jaber, B. L. (2019). A quality improvement project to reduce overutilization of blood tests in a teaching hospital. Journal of Community Hospital Internal Medicine Perspectives, 9(3), 189-194.

Rebecca Lee, M.D., Farhan Bajwa, M.D.

Unreliable characterization of diastolic dysfunction on routine echocardiography can derail clinical decision‑making. An internal audit of reports highlighted the variation versus agreement among our three primary imaging hubs—Strong Memorial Hospital, Canal View, and Clinton Crossing. This quality‑improvement initiative therefore set out to quantify reporting concordance, address gaps in sonographer data acquisition, and identify knowledge gaps that contribute to physicians’ inaccurate grading of diastolic dysfunction.

We performed a cross‑sectional review of randomly selected studies from each location. For every case we scrutinized raw images and final interpretations, applied a standardized checklist to judge completeness of sonographer measurements, and recalculated diastolic indices before benchmarking staff conclusions against an expert panel. Metrics of interest were internal consistency, data completeness, and interpretive accuracy for each site.

Performance profiles diverged. Canal View excelled in documenting diastolic parameters (ninety percent of studies) yet fell short in measurement completeness (eighty‑three percent) and achieved correct grading in only two‑thirds of cases. Clinton Crossing captured every required measurement (one‑hundred percent) but referenced diastolic status in just one‑third of reports and reached sixty percent accuracy. Strong Memorial delivered balanced documentation and data capture (both eighty percent), but accuracy dipped to sixty‑five percent; complex referral cases with atrial fibrillation proved particularly challenging, revealing a specific educational need.

The project illuminated tailored opportunities for growth. Findings advocate a unified protocol that marries structured sonographer worksheets with templated physician reports, reinforced by targeted tutorials on advanced grading—especially for arrhythmias. Observed disparities likely stem from heterogeneity in training, hardware, case‑mix, and entrenched local habits, suggesting that similar multi‑site programs will encounter comparable patterns. By standardizing inputs and language, we anticipate sharper diagnostic precision, leaner downstream testing, and more confident therapeutic choices.

A six‑month post‑implementation audit is planned to measure gains in concordance, accuracy, and downstream clinical actions, providing a feedback loop for continuous improvement across the enterprise. Results will be shared at rounds to spur adoption.