Amol A. Verma

Importance: Quality improvement efforts on general medical wards are often driven by organisational or clinician priorities, and little is known about patient or caregiver priorities. Objective: To understand contributors to patients’ in-hospital experience and how patients and caregivers prioritise these for improvement. Methods: We recruited inpatients or their caregivers from the general medical wards of 10 hospitals across Canada from November 2023 to February 2025 to participate in a group concept mapping study. Group concept mapping is a participatory multistep mixed method used to understand how a group views a particular topic. We used a targeted recruitment strategy to ensure inclusion of individuals who identified as non-white, preferred a non-English language, were unhoused, had a history of a mental health condition or who had dementia. An open-ended prompt elicited narratives about the inpatient experience. These were iteratively summarised into statements, which subsequent participants thematically sorted and rated in response to two rating questions: (1) extent of negative impact on well-being and (2) prioritisation for quality improvement. Results: 302 participants (230 patients, 72 caregivers) responded to the open-ended prompt. These were synthesised into 935 ideas, summarised into 42 unique statements, which were then rated (N=203 participants) and sorted (N=50), generating a concept map with six clusters: Environment & Facilities, Personal Support, Delays, Communication, Nutrition, Inclusivity & Engagement. There was no correlation between rating results for impact to well-being versus prioritisation for quality improvement. The ‘Delays’, ‘Communication’ and ‘Inclusivity & Engagement’ conceptual clusters contained many of the highest-rated improvement priorities. Prioritisation was consistent across participant subgroups, despite variation in how these groups rated items on impact to well-being. Conclusion: This study provides a ranked list of priorities, setting an agenda that is grounded in the patient experience. Overall, patients’ and caregivers’ own negative inpatient experiences had little bearing on overall prioritisation for quality improvement.

Background: Intracranial metastatic disease is a severe complication of cancer that confers substantial morbidity and mortality. Patients with breast or lung cancer are at particularly elevated risk of IMD. Early identification of individuals at increased risk could enable targeted surveillance and timely intervention. Methods: We developed interpretable machine-learning competing-risk models to estimate the risk of intracranial metastatic disease among patients with breast or lung cancer. For each cancer type, cause-specific Cox models were combined via the Aalen–Johansen estimator to produce absolute risk estimates at one, three, and five years. Results: Here we show high test set discrimination for intracranial metastatic disease (Uno’s C-index: breast 0.95; lung 0.88) and favorable time-dependent precision–recall performance (AUPRC(t) at 1/3/5 years: breast 0.17/0.53/0.63; lung 0.37/0.61/0.64). Decision-curve analysis across relevant thresholds demonstrates greater net clinical benefit than baseline strategies. Model interpretability analysis identifies cancer stage as the dominant determinant in both cancers; in breast cancer, triple-negative and HER2-positive subtypes contribute additional risk, whereas in lung cancer, histology and tumor size are prominent contributors. Conclusions: Machine-learning based competing-risk survival models offer greater insight into prognostication of intracranial metastatic disease than baseline strategies. These findings support the potential of such models to strengthen personalized risk stratification and guide targeted surveillance for Intracranial metastatic disease among patients with breast or lung cancer.

The use of antibiotics in acute exacerbations of chronic obstructive pulmonary disease (AECOPD) with clear evidence of pneumonia is considered standard practice. However, without radiographic bacterial pneumonia, the net impact of antibiotics is equivocal. To study physician-level practice variation in antibiotic prescribing and associated outcomes for patients hospitalized with AECOPD without pneumonia. Retrospective cohort study. Patients admitted to general internal medicine wards across seven hospitals in Ontario, Canada, between April 2010 and December 2020 with AECOPD without pneumonia. Each hospitalization was attributed to the admitting physician. To avoid indication bias (sicker patients are more likely to receive antibiotics), the main exposure was the propensity of a patient's physician to prescribe antibiotics in AECOPD, measured by the proportion of their AECOPD patients treated with antibiotics. We studied four outcomes using multivariable regression to adjust for patient baseline characteristics: in-patient mortality, intensive care unit (ICU) transfer, 30-day hospital readmission, and hospital length of stay (LOS). The cohort included 2043 hospitalizations cared for by 106 physicians. Overall, 52.1% of patients were treated with antibiotics. Physician antibiotic prescribing ranged from 15.2 to 96.2% (median 69.2%, IQR 50.9 to 76.5). Physician propensity to prescribe antibiotics was not significantly associated with patient-level clinical outcomes, including in-patient mortality (adjusted odds ratio [aOR] 1.05, 95% confidence intervals [CI] = 1.00 to 1.10), ICU transfer (aOR 1.04, 95%CI = 1.00 to 1.09), 30-day readmission (aOR 1.01, 95%CI = 0.99 to 1.02), and hospital LOS (adjusted risk ratio 1.00, 95%CI = 0.99 to 1.00) (all not statistically significant). More than half of patients hospitalized to a medical ward with AECOPD without pneumonia were treated with antibiotics. Antibiotic prescribing varied widely across physicians, and greater prescribing was not associated with better outcomes.

Background and Aims: Spontaneous bacterial peritonitis (SBP) is a lethal complication of decompensated cirrhosis with a 90% mortality rate if untreated. Despite AASLD guidelines, a practice gap exists in fluid cell count (FCC) collection during paracentesis. We initiated a quality improvement (QI) project to increase FCC collection from 78% to 100% in an ambulatory setting between September 11, 2022, and April 17, 2023. Methods: We evaluated 233 ambulatory patient encounters for paracentesis procedures at a quaternary care centre in Toronto. A multidisciplinary QI initiative included focused groups, AASLD guideline education, a paracentesis bundle, and workflow streamlining. Descriptive statistics and statistical analyses using run charts and p-charts were conducted via QI Macros. Institutional QI Review Committee approval was obtained. Results: Baseline FCC adherence was 78%. Process mapping identified inconsistencies in documentation, label printing, and FCC collection. Bedside paracentesis bundles were introduced, increasing FCC collection marginally to 79.6% with no special cause variation. Subgroup analysis showed FCC collection rates of 100% among RNs/NPs, 97% among residents, and 92% among attending physicians. General hepatology had a 97% collection rate, while specialty hepatology had 40%. The SBP rate was 1.62% (N = 3). Discussion: This study highlights a persistent gap in FCC collection during ambulatory paracentesis. Understanding reasons for non-compliance and assessing the FCC collection’s impact on morbidity and mortality in ambulatory settings are next crucial steps to optimize care for this high-risk population.

Background: Bias evaluations of machine learning (ML) models often focus on performance in research settings, with limited assessment of downstream bias following clinical deployment. The objective of this study was to evaluate whether CHARTwatch, a real-time ML early warning system for inpatient deterioration, demonstrated algorithmic bias in model performance, or produced disparities in care processes, and outcomes across patient sociodemographic groups. Methods: We evaluated CHARTwatch implementation on the internal medicine service at a large academic hospital. Patient outcomes during the intervention period (November 1, 2020–June 1, 2022) were compared to the control period (November 1, 2016–December 31, 2019) using propensity score overlap weighting. We evaluated differences across key sociodemographic subgroups, including age, sex, homelessness, and neighborhood-level socioeconomic and racialized composition. Outcomes included model performance (sensitivity and specificity), processes of care, and patient outcomes (non-palliative in-hospital death). Results: Among 12 877 patients (9079 control, 3798 intervention), 13.3% were experiencing homelessness and 36.9% lived in the quintile with the highest neighborhood racialized and newcomer populations. Model sensitivity was 70.1% overall, with no significant variation across subgroups. Model specificity varied by age, 80 years: 84% (95%CI 79-88%), P < .001, but not other subgroups. CHARTwatch implementation was associated with an increase in code status documentation among patients experiencing homelessness, without significant differences in other care processes or outcomes. Conclusion: CHARTwatch model performance and impact were generally consistent across measured sociodemographic subgroups. ML-based clinical decision support tools, and associated standardization of care, may reduce existing inequities, as was observed for code status orders among patients experiencing homelessness. This evaluation provides a framework for future bias assessments of deployed ML-CDS tools.

Background: Acute severe ulcerative colitis (ASUC) is associated with significant morbidity. In patients with ulcerative colitis (UC), the estimated lifetime risk of developing severe colitis is 25%. Several gastrointestinal societies have provided recommendations on pathways of care for managing ASUC. The degree to which they are adhered to in different care settings remains unclear. Methods: We conducted a retrospective review using data from 7 acute-care hospitals collected through the general medicine inpatient initiative (GEMINI), a hospital research collaborative that collects administrative and clinical data from hospital information systems. We identified all patients with the most responsible inpatient discharge diagnosis of ulcerative colitis between April 2015 and December 2019. The primary outcome was the difference in hospital length of stay of patients admitted with ASUC based on hospital-type; community, academic, or inflammatory bowel disease (IBD)-focussed sites. Results: 765 eligible patients were identified between April 2015 and December 2019. The mean hospital length of stay was 9.21 days for the academic sites, 6.94 days for the community sites, and 8.03 for the IBD specialty centre (P = .094). Adverse events were uncommon overall. In our multiple logistic regression analysis, we identified that admission to an IBD-focussed centre compared to an academic centre, carried an odds ratio of 2.07 (95% CI, 1.16-3.78) for the outcome of inpatient-colectomy. Conclusions: The processes of care for patients with ASUC varied on the basis of the type of hospital they were admitted to, with the IBD specialty centre providing the most guideline adherent care. Low-cost interventions should be utilized to promote adherence to clinical practice recommendations.

Objectives: Electronic health records (EHRs) data are increasingly used for research and analysis, but there is little empirical evidence to inform how automated and manual assessments can be combined to efficiently assess data quality in large EHR repositories. Materials and Methods: The GEMINI database collected data from 462 226 patient admissions across 32 hospitals from 2021 to 2023. We report data quality issues identified through semi-automated and manual data quality assessments completed during the data collection phase. We conducted a simulation experiment to evaluate the relationship between the number of records reviewed manually, the detection of true data errors (true positives) and the number of manual chart abstraction errors (false positives) that required unnecessary investigation. Results: The semi-automated data quality assessments identified 79 data quality issues requiring correction, of which 14 had a large impact, affecting at least 50% of records in the data. After resolving issues identified through semi-automated assessments, manual validation of 2676 patient encounters at 19 hospitals identified 4 new meaningful data errors (3 in transfusion data and 1 in physician identifiers), distributed across 4 hospitals. There were 365 manual chart abstraction errors, which required investigation by data analysts to identify as “false positives.” These errors increased linearly with the number of charts reviewed manually. Simulation results demonstrate that all 3 transfusion data errors were identified with 95% sensitivity after manual review of 5 records, whereas 18 records were needed for the physician’s table. Discussion and Conclusion: The GEMINI approach represents a scalable framework for data quality assessment and improvement in multisite EHR research databases. Manual data review is important but can be minimized to optimize the trade-off between true and false identification of data quality errors.