Methodology paper for the General Medicine Inpatient Initiative Medical Education Database (GEMINI MedED): a retrospective cohort study of internal medicine resident case-mix, clinical care and patient outcomes

INTRODUCTION

Unwarranted variation in patient care among physicians is associated with negative patient outcomes and increased healthcare costs. Care variation likely also exists for resident physicians. Despite the global movement towards outcomes-based and competency-based medical education, current assessment strategies in residency do not routinely incorporate clinical outcomes. The widespread use of electronic health records (EHRs) may enable the implementation of in-training assessments that incorporate clinical care and patient outcomes.

METHODS AND ANALYSIS

The General Medicine Inpatient Initiative Medical Education Database (GEMINI MedED) is a retrospective cohort study of senior residents (postgraduate year 2/3) enrolled in the University of Toronto Internal Medicine (IM) programme between 1 April 2010 and 31 December 2020. This study focuses on senior IM residents and patients they admit overnight to four academic hospitals. Senior IM residents are responsible for overseeing all overnight admissions; thus, care processes and outcomes for these clinical encounters can be at least partially attributed to the care they provide. Call schedules from each hospital, which list the date, location and senior resident on-call, will be used to link senior residents to EHR data of patients admitted during their on-call shifts. Patient data will be derived from the GEMINI database, which contains administrative (eg, demographic and disposition) and clinical data (eg, laboratory and radiological investigation results) for patients admitted to IM at the four academic hospitals. Overall, this study will examine three domains of resident practice: (1) case-mix variation across residents, hospitals and academic year, (2) resident-sensitive quality measures (EHR-derived metrics that are partially attributable to resident care) and (3) variations in patient outcomes across residents and factors that contribute to such variation.

ETHICS AND DISSEMINATION

GEMINI MedED was approved by the University of Toronto Ethics Board (RIS#39339). Results from this study will be presented in academic conferences and peer-reviewed journals.

From compute to care: Lessons learned from deploying an early warning system into clinical practice

Background

Deploying safe and effective machine learning models is essential to realize the promise of artificial intelligence for improved healthcare. Yet, there remains a large gap between the number of high-performing ML models trained on healthcare data and the actual deployment of these models. Here, we describe the deployment of CHARTwatch, an artificial intelligence-based early warning system designed to predict patient risk of clinical deterioration.

Methods

We describe the end-to-end infrastructure that was developed to deploy CHARTwatch and outline the process from data extraction to communicating patient risk scores in real-time to physicians and nurses. We then describe the various challenges that were faced in deployment, including technical issues (e.g., unstable database connections), process-related challenges (e.g., changes in how a critical lab is measured), and challenges related to deploying a clinical system in the middle of a pandemic. We report various measures to quantify the success of the deployment: model performance, adherence to workflows, and infrastructure uptime/downtime. Ultimately, success is driven by end-user adoption and impact on relevant clinical outcomes. We assess our deployment process by evaluating how closely we followed existing guidance for good machine learning practice (GMLP) and identify gaps that are not addressed in this guidance.

Results

The model demonstrated strong and consistent performance in real-time in the first 19 months after deployment (AUC 0.76) as in the silent deployment heldout test data (AUC 0.79). The infrastructure remained online for >99% of time in the first year of deployment. Our deployment adhered to all 10 aspects of GMLP guiding principles. Several steps were crucial for deployment but are not mentioned or are missing details in the GMLP principles, including the need for a silent testing period, the creation of robust downtime protocols, and the importance of end-user engagement. Evaluation for impacts on clinical outcomes and adherence to clinical protocols is underway.

Conclusion

We deployed an artificial intelligence-based early warning system to predict clinical deterioration in hospital. Careful attention to data infrastructure, identifying problems in a silent testing period, close monitoring during deployment, and strong engagement with end-users were critical for successful deployment.

Outcomes for upper gastrointestinal bleeding during the first wave of the COVID-19 pandemic in the Toronto area

BACKGROUND AND AIM

Changes to endoscopy service availability during the COVID-19 pandemic may have affected management of upper gastrointestinal bleeding (UGIB). The aim of this study was to describe the impact of the pandemic on UGIB outcomes in the Toronto area in Canada.

METHODS

We described all adults admitted to general medicine wards or intensive care units at six hospitals in Toronto and Mississauga, Canada, with UGIB during the first wave of the COVID-19 pandemic (March 1 to June 30, 2020) and compared them with a historical cohort (March 1 to June 30, 2018 and 2019). We compared clinical outcomes (in-hospital mortality, length of stay, 30-day readmission, intensive care utilization, receipt of endoscopy, persistent bleeding, receipt of second endoscopy, and need for angiographic or surgical intervention) using multivariable regression models, controlling for demographics, comorbidities, and severity of clinical presentation.

RESULTS

There were 82.5 and 215.5 admissions per month for UGIB during the COVID-19 and control periods, respectively. There were no baseline differences between groups for demographic characteristics, comorbidities, or severity of bleeding. Patients in the COVID-19 group did not have significantly different unadjusted (3.9% vs 4.2%, P = 0.983) or adjusted mortality (adjusted odds ratio [OR] = 0.64, 95% confidence interval [CI] = 0.25-1.48, P = 0.322). Patients in COVID-19 group were less likely to receive endoscopy for UGIB in the unadjusted (61.8% vs 71.0%, P = 0.003) and adjusted (adjusted OR = 0.64, 95% CI = 0.49-0.84, P < 0.01) models. There were no differences between groups for other secondary outcomes.

CONCLUSIONS

While patients admitted for UGIB during the first wave of the pandemic were less likely to receive endoscopy, this had no impact on mortality or any secondary outcomes.

Prone positioning of patients with moderate hypoxaemia due to covid-19: multicentre pragmatic randomised trial (COVID-PRONE)

OBJECTIVES

To assess the effectiveness of prone positioning to reduce the risk of death or respiratory failure in non-critically ill patients admitted to hospital with covid-19.

DESIGN

Multicentre pragmatic randomised clinical trial.

SETTING

15 hospitals in Canada and the United States from May 2020 until May 2021.

PARTICIPANTS

Eligible patients had a laboratory confirmed or a clinically highly suspected diagnosis of covid-19, needed supplemental oxygen (up to 50% fraction of inspired oxygen), and were able to independently lie prone with verbal instruction. Of the 570 patients who were assessed for eligibility, 257 were randomised and 248 were included in the analysis.

INTERVENTION

Patients were randomised 1:1 to prone positioning (that is, instructing a patient to lie on their stomach while they are in bed) or standard of care (that is, no instruction to adopt prone position).

MAIN OUTCOME MEASURES

The primary outcome was a composite of in-hospital death, mechanical ventilation, or worsening respiratory failure defined as needing at least 60% fraction of inspired oxygen for at least 24 hours. Secondary outcomes included the change in the ratio of oxygen saturation to fraction of inspired oxygen.

RESULTS

The trial was stopped early on the basis of futility for the pre-specified primary outcome. The median time from hospital admission until randomisation was 1 day, the median age of patients was 56 (interquartile range 45-65) years, 89 (36%) patients were female, and 222 (90%) were receiving oxygen via nasal prongs at the time of randomisation. The median time spent prone in the first 72 hours was 6 (1.5-12.8) hours in total for the prone arm compared with 0 (0-2) hours in the control arm. The risk of the primary outcome was similar between the prone group (18 (14%) events) and the standard care group (17 (14%) events) (odds ratio 0.92, 95% confidence interval 0.44 to 1.92). The change in the ratio of oxygen saturation to fraction of inspired oxygen after 72 hours was similar for patients randomised to prone positioning and standard of care.

CONCLUSION

Among non-critically ill patients with hypoxaemia who were admitted to hospital with covid-19, a multifaceted intervention to increase prone positioning did not improve outcomes. However, wide confidence intervals preclude definitively ruling out benefit or harm. Adherence to prone positioning was poor, despite multiple efforts to increase it. Subsequent trials of prone positioning should aim to develop strategies to improve adherence to awake prone positioning.

STUDY REGISTRATION

ClinicalTrials.gov NCT04383613.

Outcomes in patients with and without disability admitted to hospital with COVID-19: a retrospective cohort study

Background:

Disability-related considerations have largely been absent from the COVID-19 response, despite evidence that people with disabilities are at elevated risk for acquiring COVID-19. We evaluated clinical outcomes in patients who were admitted to hospital with COVID-19 with a disability compared with patients without a disability.

Methods:

We conducted a retrospective cohort study that included adults with COVID-19 who were admitted to hospital and discharged between Jan. 1, 2020, and Nov. 30, 2020, at 7 hospitals in Ontario, Canada. We compared in-hospital death, admission to the intensive care unit (ICU), hospital length of stay and unplanned 30-day readmission among patients with and without a physical disability, hearing or vision impairment, traumatic brain injury, or intellectual or developmental disability, overall and stratified by age (≤ 64 and ≥ 65 yr) using multivariable regression, controlling for sex, residence in a long-term care facility and comorbidity.

Results:

Among 1279 admissions to hospital for COVID-19, 22.3% had a disability. We found that patients with a disability were more likely to die than those without a disability (28.1% v. 17.6%), had longer hospital stays (median 13.9 v. 7.8 d) and more readmissions (17.6% v. 7.9%), but had lower ICU admission rates (22.5% v. 28.3%). After adjustment, there were no statistically significant differences between those with and without disabilities for in-hospital death or admission to ICU. After adjustment, patients with a disability had longer hospital stays (rate ratio 1.36, 95% confidence interval [CI] 1.19–1.56) and greater risk of readmission (relative risk 1.77, 95% CI 1.14–2.75). In age-stratified analyses, we observed longer hospital stays among patients with a disability than in those without, in both younger and older subgroups; readmission risk was driven by younger patients with a disability.

Interpretation:

Patients with a disability who were admitted to hospital with COVID-19 had longer stays and elevated readmission risk than those without disabilities. Disability-related needs should be addressed to support these patients in hospital and after discharge.

Bedspacing and clinical outcomes in general internal medicine: A retrospective, multicenter cohort study

BACKGROUND

Admitting hospitalized patients to off-service wards ("bedspacing") is common and may affect quality of care and patient outcomes.

OBJECTIVE

To compare in-hospital mortality, 30-day readmission to general internal medicine (GIM), and hospital length-of-stay among GIM patients admitted to GIM wards or bedspaced to off-service wards.

DESIGN, PARTICIPANTS, AND MEASURES

Retrospective cohort study including all emergency department admissions to GIM between 2015 and 2017 at six hospitals in Ontario, Canada. We compared patients admitted to GIM wards with those who were bedspaced, using multivariable regression models and propensity score matching to control for patient and situational factors.

KEY RESULTS

Among 40,440 GIM admissions, 10,745 (26.6%) were bedspaced to non-GIM wards and 29,695 (73.4%) were assigned to GIM wards. After multivariable adjustment, bedspacing was associated with no significant difference in mortality (adjusted hazard ratio 0.95, 95% confidence interval [CI]: 0.86-1.05, p = .304), slightly shorter median hospital length-of-stay (-0.10 days, 95% CI:-0.20 to -0.001, p = .047) and lower 30-day readmission to GIM (adjusted OR 0.89, 95% CI: 0.83-0.95, p = .001). Results were consistent when examining each hospital individually and outcomes did not significantly differ between medical or surgical off-service wards. Sensitivity analyses focused on the highest risk patients did not exclude the possibility of harm associated with bedspacing, although adverse outcomes were not significantly greater.

CONCLUSIONS

Overall, bedspacing was associated with no significant difference in mortality, slightly shorter hospital length-of-stay, and fewer 30-day readmissions to GIM, although potential harms in high-risk patients remain uncertain. Given that hospital capacity issues are likely to persist, future research should aim to understand how bedspacing can be achieved safely at all hospitals, perhaps by strengthening the selection of low-risk patients.

Developing and validating natural language processing algorithms for radiology reports compared to ICD-10 codes for identifying venous thromboembolism in hospitalized medical patients

BACKGROUND

Identifying venous thromboembolism (VTE) from large clinical and administrative databases is important for research and quality improvement.

OBJECTIVE

To develop and validate natural language processing (NLP) algorithms to identify VTE from radiology reports among general internal medicine (GIM) inpatients.

METHODS

This cross-sectional study included GIM hospitalizations between April 1, 2010 and March 31, 2017 at 5 hospitals in Toronto, Ontario, Canada. We developed NLP algorithms to identify pulmonary embolism (PE) and deep venous thrombosis (DVT) from radiologist reports of thoracic computed tomography (CT), extremity compression ultrasound (US), and nuclear ventilation-perfusion (VQ) scans in a training dataset of 1551 hospitalizations. We compared the accuracy of our NLP algorithms, the previously-published "simpleNLP" tool, and administrative discharge diagnosis codes (ICD-10-CA) for PE and DVT to the "gold standard" manual review in a separate random sample of 4000 GIM hospitalizations.

RESULTS

Our NLP algorithms were highly accurate for identifying DVT from US, with sensitivity 0.94, positive predictive value (PPV) 0.90, and Area Under the Receiver-Operating-Characteristic Curve (AUC) 0.96; and in identifying PE from CT, with sensitivity 0.91, PPV 0.89, and AUC 0.96. Administrative diagnosis codes and the simple NLP tool were less accurate for DVT (ICD-10-CA sensitivity 0.63, PPV 0.43, AUC 0.81; simpleNLP sensitivity 0.41, PPV 0.36, AUC 0.66) and PE (ICD-10-CA sensitivity 0.83, PPV 0.70, AUC 0.91; simpleNLP sensitivity 0.89, PPV 0.62, AUC 0.92).

CONCLUSIONS

Administrative diagnosis codes are unreliable in identifying VTE in hospitalized patients. We developed highly accurate NLP algorithms to identify VTE from radiology reports in a multicentre sample and have made the algorithms freely available to the academic community with a user-friendly tool (https://lks-chart.github.io/CHARTextract-docs/08-downloads/rulesets.html#venous-thromboembolism-vte-rulesets).

Using machine learning to predict severe hypoglycaemia in hospital

AIM

To predict the risk of hypoglycaemia using machine-learning techniques in hospitalized patients.

METHODS

We conducted a retrospective cohort study of patients hospitalized under general internal medicine (GIM) and cardiovascular surgery (CV) at a tertiary care teaching hospital in Toronto, Ontario. Three models were generated using supervised machine learning: least absolute shrinkage and selection operator (LASSO) logistic regression; gradient-boosted trees; and a recurrent neural network. Each model included baseline patient data and time-varying data. Natural-language processing was used to incorporate text data from physician and nursing notes.

RESULTS

We included 8492 GIM admissions and 8044 CV admissions. Hypoglycaemia occurred in 16% of GIM admissions and 13% of CV admissions. The area under the curve for the models in the held-out validation set was approximately 0.80 on the GIM ward and 0.82 on the CV ward. When the threshold for hypoglycaemia was lowered to 2.9 mmol/L (52 mg/dL), similar results were observed. Among the patients at the highest decile of risk, the positive predictive value was approximately 50% and the sensitivity was 99%.

CONCLUSION

Machine-learning approaches can accurately identify patients at high risk of hypoglycaemia in hospital. Future work will involve evaluating whether implementing this model with targeted clinical interventions can improve clinical outcomes.

Internal medicine hospitalisations and liver disease: a comparative disease burden analysis of a multicentre cohort

BACKGROUND

Liver disease is an increasing burden on population health globally.

AIMS

To characterise burden of liver disease among general internal medicine inpatients at seven Toronto-area hospitals and compare it to other common medical conditions.

METHODS

Data from April 2010 to October 2017 were obtained from hospitals participating in the GEMINI collaborative. Using these cohort data from hospital information systems linked to administrative data, we defined liver disease admissions using most responsible discharge diagnoses categorised according to international classification of diseases, 10th Revision-enhanced Canadian version (ICD-10-CA). We identified admissions for heart failure, chronic obstructive pulmonary disease (COPD) and pneumonia as comparators. We calculated standardised mortality ratios (SMRs) as the ratio of observed to expected deaths.

RESULTS

Among 239 018 discharges, liver disease accounted for 1.7% of most responsible discharge diagnoses. Liver disease was associated with marked premature mortality, with SMR of 8.84 (95% CI 8.06-9.67) compared to 1.06 (95% CI 0.99-1.12) for heart failure, 1.05 (95% CI 0.96-1.15) for COPD and 1.28 (95% CI 1.20-1.37) for pneumonia. The majority of deaths were among patients younger than 65 years (57.7%) compared to 3.3% in heart failure, 5.6% in COPD and 10.7% in pneumonia. Liver disease patients presented with worse Laboratory-Based Acute Physiology Scores, were more frequently admitted to the intensive care unit (14.4%), incurred higher average total costs (median $6723 CAD), had higher in-hospital mortality (11.4%), and were more likely to be a readmission from 30 days prior (19.8%). Non-alcoholic fatty liver disease admissions increased from 120 in 2011-2012 to 215 in 2016-2017 (P < 0.01).

CONCLUSION

In Canada's largest urban centre, liver disease admissions resulted in premature morbidity and mortality with higher resource use compared to common cardio-respiratory conditions. Re-evaluation of approaches to caring for inpatients with liver disease is timely and justified.

Assessing the quality of clinical and administrative data extracted from hospitals: the General Medicine Inpatient Initiative (GEMINI) experience

OBJECTIVE

Large clinical databases are increasingly used for research and quality improvement. We describe an approach to data quality assessment from the General Medicine Inpatient Initiative (GEMINI), which collects and standardizes administrative and clinical data from hospitals.

METHODS

The GEMINI database contained 245 559 patient admissions at 7 hospitals in Ontario, Canada from 2010 to 2017. We performed 7 computational data quality checks and iteratively re-extracted data from hospitals to correct problems. Thereafter, GEMINI data were compared to data that were manually abstracted from the hospital's electronic medical record for 23 419 selected data points on a sample of 7488 patients.

RESULTS

Computational checks flagged 103 potential data quality issues, which were either corrected or documented to inform future analysis. For example, we identified the inclusion of canceled radiology tests, a time shift of transfusion data, and mistakenly processing the chemical symbol for sodium ("Na") as a missing value. Manual validation identified 1 important data quality issue that was not detected by computational checks: transfusion dates and times at 1 site were unreliable. Apart from that single issue, across all data tables, GEMINI data had high overall accuracy (ranging from 98%-100%), sensitivity (95%-100%), specificity (99%-100%), positive predictive value (93%-100%), and negative predictive value (99%-100%) compared to the gold standard.

DISCUSSION AND CONCLUSION

Computational data quality checks with iterative re-extraction facilitated reliable data collection from hospitals but missed 1 critical quality issue. Combining computational and manual approaches may be optimal for assessing the quality of large multisite clinical databases.

Caractéristiques et issues des hospitalisations pour les cas de COVID-19 et d’influenza dans la région de Toronto

CONTEXTE:

Les caractéristiques des patients, les soins cliniques, l’utilisation des ressources et les issues cliniques des personnes atteintes de la maladie à coronavirus 2019 (COVID-19) hospitalisées au Canada ne sont pas bien connus.

MÉTHODES:

Nous avons recueilli des données sur tous les adultes hospitalisés atteints de la COVID-19 ou de l’influenza ayant obtenu leur congé d’unités médicales ou d’unités de soins intensifs médicaux et chirurgicaux entre le 1^er novembre 2019 et le 30 juin 2020 dans 7 centres hospitaliers de Toronto et de Mississauga (Ontario). Nous avons comparé les issues cliniques des patients à l’aide de modèles de régression multivariée, en tenant compte des facteurs sociodémographiques et de l’intensité des comorbidités. Nous avons validé le degré d’exactitude de 7 scores de risque mis au point à l’externe pour déterminer leur capacité à prédire le risque de décès chez les patients atteints de la COVID-19.

RÉSULTATS:

Parmi les hospitalisations retenues, 1027 patients étaient atteints de la COVID-19 (âge médian de 65 ans, 59,1 % d’hommes) et 783 étaient atteints de l’influenza (âge médian de 68 ans, 50,8 % d’hommes). Les patients âgés de moins de 50 ans comptaient pour 21,2 % de toutes les hospitalisations dues à la COVID-19 et 24,0 % des séjours aux soins intensifs. Comparativement aux patients atteints de l’influenza, les patients atteints de la COVID-19 présentaient un taux de mortalité perhospitalière (mortalité non ajustée 19,9 % c. 6,1 %; risque relatif [RR] ajusté 3,46 %, intervalle de confiance [IC] à 95 % 2,56–4,68) et un taux d’utilisation des ressources des unités de soins intensifs (taux non ajusté 26,4 % c. 18,0 %; RR ajusté 1,50, IC à 95 % 1,25–1,80) significativement plus élevés, ainsi qu’une durée d’hospitalisation (durée médiane non ajustée 8,7 jours c. 4,8 jours; rapport des taux d’incidence ajusté 1,45; IC à 95 % 1,25–1,69) significativement plus longue. Le taux de réhospitalisation dans les 30 jours n’était pas significativement différent (taux non ajusté 9,3 % c. 9,6 %; RR ajusté 0,98 %, IC à 95 % 0,70–1,39). Trois scores de risque utilisant un pointage pour prédire la mortalité perhospitalière ont montré une bonne discrimination (aire sous la courbe [ASC] de la fonction d’efficacité du récepteur [ROC] 0,72–0,81) et une bonne calibration.

INTERPRÉTATION:

Durant la première vague de la pandémie, l’hospitalisation des patients atteints de la COVID-19 était associée à des taux de mortalité et d’utilisation des ressources des unités de soins intensifs et à une durée d’hospitalisation significativement plus importants que les hospitalisations des patients atteints de l’influenza. De simples scores de risque peuvent prédire avec une bonne exactitude le risque de mortalité perhospitalière des patients atteints de la COVID-19.

Characteristics and outcomes of hospital admissions for COVID-19 and influenza in the Toronto area

BACKGROUND:

Patient characteristics, clinical care, resource use and outcomes associated with admission to hospital for coronavirus disease 2019 (COVID-19) in Canada are not well described.

METHODS:

We described all adults with COVID-19 or influenza discharged from inpatient medical services and medical–surgical intensive care units (ICUs) between Nov. 1, 2019, and June 30, 2020, at 7 hospitals in Toronto and Mississauga, Ontario. We compared patient outcomes using multivariable regression models, controlling for patient sociodemographic factors and comorbidity level. We validated the accuracy of 7 externally developed risk scores to predict mortality among patients with COVID-19.

RESULTS:

There were 1027 hospital admissions with COVID-19 (median age 65 yr, 59.1% male) and 783 with influenza (median age 68 yr, 50.8% male). Patients younger than 50 years accounted for 21.2% of all admissions for COVID-19 and 24.0% of ICU admissions. Compared with influenza, patients with COVID-19 had significantly greater in-hospital mortality (unadjusted 19.9% v. 6.1%, adjusted relative risk [RR] 3.46, 95% confidence interval [CI] 2.56–4.68), ICU use (unadjusted 26.4% v. 18.0%, adjusted RR 1.50, 95% CI 1.25–1.80) and hospital length of stay (unadjusted median 8.7 d v. 4.8 d, adjusted rate ratio 1.45, 95% CI 1.25–1.69). Thirty-day readmission was not significantly different (unadjusted 9.3% v. 9.6%, adjusted RR 0.98, 95% CI 0.70–1.39). Three points-based risk scores for predicting in-hospital mortality showed good discrimination (area under the receiver operating characteristic curve [AUC] ranging from 0.72 to 0.81) and calibration.

INTERPRETATION:

During the first wave of the pandemic, admission to hospital for COVID-19 was associated with significantly greater mortality, ICU use and hospital length of stay than influenza. Simple risk scores can predict in-hospital mortality in patients with COVID-19 with good accuracy.

Comparing characteristics and outcomes of cancer to non-cancer patients admitted to general internal medicine (GIM)

21

Background: Cancer prevalence is rising and there is a corresponding increase in hospitalizations across the cancer continuum. However, little is known about the patterns of care and outcomes of cancer inpatients as administrative data may not capture in-hospital details including investigations and medications required for characterization. Understanding how cancer inpatients are managed and their outcomes can help to optimize care delivery. Methods: We conducted a multicenter study of all patients admitted to GIM at seven hospitals (Toronto, Canada) from 2010 to 2017 where we deterministically linked administrative data with each hospital’s electronic information (pharmacy, orders, notes, laboratory/imaging and results) at the patient level. Multivariable regression models compared characteristics and outcomes between cancer and non-cancer patients for the top 5 non-cancer patient discharge diagnoses. Results: Among 230,040 hospitalizations, 15% had cancer listed as an ICD-10 comorbidity. The most common cancer disease sites were gastrointestinal (20%), lung (13%) and leukemia (11%). The most common discharge diagnoses for cancer patients were disease progression (9%), palliative care (6%), pneumonia (4%), leukemia (4%) and lung cancer (4%), while for non-cancer patients were: heart failure (5%), pneumonia (5%), stroke (5%), COPD (5%) and urinary tract infections (5%). In general, compared to non-cancer patients, cancer patients were younger (70 vs 72), had greater length of stay (LOS; 6.4 vs 4.6 days), in-hospital mortality (16% vs 5%), ICU use (12% vs 11%), 30 day re-admission rate (17% vs 10%) and were more likely to receive CTs (64% vs 52%), MRIs (14% vs 12%) and interventional procedures (22% vs 8%) (p < 0.001, all comparisons). When evaluating the top 5 non-cancer patient discharge diagnoses, results (adjusted for age, gender, Charlson comorbidity score and hospital) were similar wherein cancer patients had a higher in-hospital mortality (aOR = 2.02 p < 0.001), 30 day re-admission rate (aOR = 1.09 p = 0.08) and were more likely to receive CTs (aOR = 1.88 p < 0.001), MRIs (aOR = 1.66 p < 0.001) or interventional procedures (aOR = 1.78 p < 0.001), despite similar mean LOS (5.7 vs 5.1 days p = 0.35). Results were similar across discharge diagnoses. Conclusions: Cancer patients represent a unique population on GIM and have higher resource use, mortality and LOS compared to non-cancer patients, with similar trends even for the same non-cancer diagnoses. Specialized models of care for hospitalized cancer patients may be warranted.

Accuracy of the Preferred Language Field in the Electronic Health Records of Two Canadian Hospitals

BACKGROUND

The collection of race, ethnicity, and language (REaL) data from patients is advocated as a first step to identify, monitor, and improve health inequities. As a result, many health care institutions collect patients' preferred languages in their electronic health records (EHRs). These data may be used in clinical care, research, and quality improvement. However, the accuracy of EHR language data are rarely assessed.

OBJECTIVES

This study aimed to audit the accuracy of EHR language data at two academic hospitals in Toronto, Ontario, Canada.

METHODS

The EHR language was compared with a patient's stated preferred language by interview. Language was dichotomized to English or non-English. Agreement between language documented in the EHR and patient-reported preferred language was calculated using sensitivity, specificity, and positive predictive value (PPV).

RESULTS

A total of 323 patients were interviewed, including 96 with a stated non-English preferred language. The sensitivity of the EHR for English-language preference was high at both hospitals: 100% at hospital A with a PPV of 88%, and 99% at hospital B with a PPV of 85%. However, the sensitivity of the EHR for non-English preference differed greatly between the two hospitals. The sensitivity was 81% with a PPV of 100% at hospital A and the sensitivity was 12% with a PPV of 60% at hospital B.

CONCLUSION

The accuracy of the EHR for identifying non-English language preference differed greatly between the hospitals studied. Language data must be accurate for it to be used, and regular quality assurance is required.

Empiric Antibiotic Treatment Thresholds for Serious Bacterial Infections: A Scenario-based Survey Study

BACKGROUND

Physicians face competing demands of maximizing pathogen coverage while minimizing unnecessary use of broad-spectrum antibiotics when managing sepsis. We sought to identify physicians' perceived likelihood of coverage achieved by their usual empiric antibiotic regimen, along with minimum thresholds of coverage they would be willing to accept when managing these patients.

METHODS

We conducted a scenario-based survey of internal medicine physicians from across Canada using a 2 × 2 factorial design, varied by infection source (undifferentiated vs genitourinary) and severity (mild vs severe) denoted by the Quick Sequential Organ Failure Assessment (qSOFA) score. For each scenario, participants selected their preferred empiric antibiotic regimen, estimated the likelihood of coverage achieved by that regimen, and considered their minimum threshold of coverage.

RESULTS

We had 238 respondents: 87 (36.6%) residents and 151 attending physicians (63.4%). The perceived likelihood of antibiotic coverage and minimum thresholds of coverage (with interquartile range) for each scenario were as follows: (1) severe undifferentiated, 90% (89.5%-95.0%) and 90% (80%-95%), respectively; (2) mild undifferentiated, 89% (80%-95%) and 80% (70%-89.5%); (3) severe genitourinary, 91% (87.3%-95.0%) and 90% (80.0%-90.0%); and (4) mild genitourinary, 90% (81.8%-91.3%) and 80% (71.8%-90%). Illness severity and infectious disease specialty predicted higher thresholds of coverage whereas less clinical experience and lower self-reported prescribing intensity predicted lower thresholds of coverage.

CONCLUSIONS

Pathogen coverage of 80% and 90% are physician-acceptable thresholds for managing patients with mild and severe sepsis from bacterial infections. These data may inform clinical guidelines and decision-support tools to improve empiric antibiotic prescribing.