Search

Artificial intelligence (AI) and machine learning (ML) are rapidly advancing fields with increasing utility in health care. We conducted a survey to determine the perceptions of Canadian vascular surgeons toward AI/ML.An online questionnaire was distributed to 162 members of the Canadian Society for Vascular Surgery. Self-reported knowledge, attitudes, and perceptions with respect to potential applications, limitations, and facilitators of AI/ML were assessed.Overall, 50 of the 162 Canadian vascular surgeons (31%) responded to the survey. Most respondents were aged 30 to 59 years (72%), male (80%), and White (67%) and practiced in academic settings (72%). One half of the participants reported that their knowledge of AI/ML was poor or very poor. Most were excited or very excited about AI/ML (66%) and were interested or very interested in learning more about the field (83.7%). The respondents believed that AI/ML would be useful or very useful for diagnosis (62%), prognosis (72%), patient selection (56%), image analysis (64%), intraoperative guidance (52%), research (88%), and education (80%). The limitations that the participants were most concerned about were errors leading to patient harm (42%), bias based on patient demographics (42%), and lack of clinician knowledge and skills in AI/ML (40%). Most were not concerned or were mildly concerned about job replacement (86%). The factors that were most important to encouraging clinicians to use AI/ML models were improvements in efficiency (88%), accurate predictions (84%), and ease of use (84%). The comments from respondents focused on the pressing need for the implementation of AI/ML in vascular surgery owing to the potential to improve care delivery.Canadian vascular surgeons have positive views on AI/ML and believe this technology can be applied to multiple aspects of the specialty to improve patient care, research, and education. Current self-reported knowledge is poor, although interest was expressed in learning more about the field. The facilitators and barriers to the effective use of AI/ML identified in the present study can guide future development of these tools in vascular surgery.

Objective The objectives of this study were to: 1) identify predictors of subsequent surgery following initial treatment of proximal humerus fractures (PHF); and 2) generate valid risk prediction tools to predict subsequent surgery. Methods We identified PHF patients ≥ 50 years from 2004 to 2015 using health datasets in Ontario, Canada. We used procedural codes to classify patients into treatment groups of: 1) surgical fixation; 2) shoulder replacement; and 3) conservative. We used procedural and diagnosis codes to capture subsequent surgery within two years post fracture. We developed regression models for two thirds of each group to identify predictors of subsequent surgery, and the regression equations to develop risk tools to predict subsequent surgery. We used the final third of each cohort to evaluate the discriminative ability of the risk tools using c-statistics. Results We identified 20,897 PHF patients, 2,414 treated with fixation, 1,065 with replacement, and 17,418 treated conservatively. Predictors of reoperation following fixation included bone grafting, and nail or wire fixation vs. plate fixation, while poor bone quality was associated with reoperation following initial replacement. In conservatively treated patients, more comorbidities were associated with subsequent surgery, while age 70+, and discharge home following presentation lowered the odds of subsequent surgery. The risk tools were able to discriminate with c-statistics of 0.75-0.88 (derivation) and 0.51-0.79 (validation). Conclusion Our risk tools showed good to strong discriminative ability for patients treated with fixation and conservatively. These data may be used as the foundation to develop a clinically informative tool. Level of evidence Prognostic Level II. See Instructions for Authors for a complete description of levels of evidence.

To characterize comparative risks and benefits of methadone versus buprenorphine/naloxone in a contemporary cohort where the unregulated drug supply is dominated by fentanyl.Population-based propensity-score matched cohort study conducted in Ontario, Canada among people aged 18+ initiating opioid agonist therapy (OAT) for an opioid use disorder between October 2016 and December 2018 (n = 18 880).Initiation of methadone versus buprenorphine/naloxone.The primary outcome was opioid overdose (fatal and non-fatal) while on treatment, with secondary outcomes including opioid overdose (first 30 days of treatment), treatment discontinuation, health-care interactions related to treatment of opioid use disorder, receiving a weekly supply of take-home doses and opioid overdose within 30 days of treatment discontinuation. Outcomes were assessed over 1 year.Overall, 7517 people initiating buprenorphine were matched to an equal number of methadone-treated individuals. Risk of opioid overdose while on treatment [hazard ratio (HR) = 0.50; 95% confidence interval (CI) = 0.37-0.68] or within the first 30 days of treatment (HR = 0.51, 95% CI = 0.31-0.85) was lower among buprenorphine recipients compared to methadone recipients. In secondary analyses, people initiating buprenorphine had a higher risk of treatment discontinuation within the first year (median time to discontinuation 104 versus 265 days, HR = 1.43, 95% CI = 1.37-1.49), had lower rates of health-care interactions for OUD (186.4 versus 254.3 per person-year; rate ratio = 0.73; 95% CI = 0.72-0.75), and a higher rate of receiving weekly take-home doses (HR = 2.33; 95% CI = 2.20-2.46). Overdose rates in the period following OAT discontinuation were higher than those observed while on treatment, but did not differ significantly by OAT type.Although treatment retention is higher among methadone recipients, overdose risk is also elevated compared to buprenorphine recipients. These findings demonstrate the benefits of any OAT on avoidance of overdose, particularly following treatment discontinuation and with the increasingly unpredictable drug supply in North America.

UNSTRUCTURED Artificial intelligence-based chatbots that fall in the category of Large Language Models, such as ChatGPT, have caught the world's attention. They boast a broad range of general capabilities and are believed to have the potential to revolutionize nearly every aspect of work and human interaction. The application of these tools to address administrative and resource challenges in healthcare delivery is promising. This paper discusses the use of these tools to introduce efficiency for frontline healthcare delivery and examines considerations and limitations that must be taken into account when using them in healthcare. Additionally, it explores potential future applications of LLMs in healthcare.

Background Population-based research examining geographic variability in psychotropic medication dispensing to children and youth and the sociodemographic correlates of such variation is lacking. Variation in psychotropic use could reflect disparities in access to non-pharmacologic interventions and identify potentially concerning use patterns. Methods We conducted a population-based study of all Ontario residents aged 0 to 24 years who were dispensed a benzodiazepine, stimulant, antipsychotic or antidepressant between January 1, 2018, and December 31, 2018. We conducted small-area variation analyses and identified determinants of dispensing using negative binomial generalized estimating equation models. Results The age- and sex-standardized rate of psychotropic dispensing to children and youth was 76.8 (range 41.7 to 144.4) prescriptions per 1000 population, with large variation in psychotropic dispensing across Ontario’s census divisions. Males had higher antipsychotic [rate ratio (RR) 1.40; 95% confidence interval (CI) 1.36 to 1.44) and stimulant (RR 1.75; 95% CI 1.70 to 1.80) dispensing rates relative to females, with less use of benzodiazepines (RR 0.85; 95% CI 0.83 to 0.88) and antidepressants (RR 0.81; 95% CI 0.80 to 0.82). Lower antipsychotic dispensing was observed in the highest income neighbourhoods (RR 0.72; 95% CI 0.70 to 0.75) relative to the lowest. Benzodiazepine (RR 1.12; 95% CI 1.01 to 1.24) and stimulant (RR 1.11; 95% CI 1.01 to 1.23) dispensing increased with the density of mental health services in census divisions, whereas antipsychotic use decreased (RR 0.82; 95% CI 0.73 to 0.91). The regional density of child and adolescent psychiatrists and developmental pediatricians (RR 1.00; 95% CI 0.99 to 1.01) was not associated with psychotropic dispensing. Conclusion We found significant variation in psychotropic dispensing among young Ontarians. Targeted investment in regions with long wait times for publicly-funded non-pharmacological interventions and novel collaborative service models may minimize variability and promote best practices in using psychotropics among children and youth.

Purpose: To estimate the positive predictive value (PPV) of Canadian ICD-10 diagnostic coding for the identification of hospitalization related to a diabetic foot ulcer (DFU). Methods: Hospitalizations related to a neuropathic and/or ischemic DFU were identified from the Discharge Abstract Database (DAD) records of a single Canadian tertiary care hospital between April 1, 2002 and March 31, 2019. The first coding approach required a most responsible diagnosis (MRDx) code for diabetes-specific foot ulceration or gangrene (DSFUG group). Three alternative coding approaches were also considered: MRDx code for lower-limb osteomyelitis (osteomyelitis group); lower-limb ulceration (LLU group); or lower-limb atherosclerotic gangrene (atherosclerosis group)—each in conjunction with a non-MRDx DSFUG code on the same DAD record. From all eligible DAD records, random samples were drawn for each coding group. DAD records were independently compared by a masked reviewer who manually abstracted data from the entire hospital record (reference standard). The PPV and 95% CI were generated. Results: Out of 1,460 hospitalizations, a total of 300, 50, 33 and seven records were included from the DSFUG, osteomyelitis, LLU and atherosclerosis samples, respectively. Compared to the reference standard, the PPV for all 390 records was 88.5% (95% CI 84.9 to 91.5). The DSFUG group had the highest PPV (90.0%, 95% CI 86.0 to 93.2), followed by the atherosclerosis (85.7%, 95% CI 42.1 to 99.6), LLU (84.9%, 95% CI 68.1 to 94.9) and osteomyelitis (82.0%, 95% CI 68.6 to 91.4) groups. Conclusion: Based on data from a Canadian tertiary care hospital, the specified coding algorithms can be used to identify and study the management and outcomes of people hospitalized with a DFU in Ontario.

OBJECTIVES: To quantify the real-world survival benefit of re-resection vs no re-resection in patients diagnosed with T1 bladder cancer (BC) at the population level. PATIENTS AND METHODS: Retrospective population-wide observational cohort study based on pathology reports linked to health administrative data. We identified patients who were diagnosed with T1 BC in the province of Ontario (01/2001-12/2015) and used billing claims to ascertain whether they received re-resection within 2-10 weeks. The time-dependent effect of re-resection on survival outcomes was modelled by Cox proportional hazards regression (unadjusted and adjusted for numerous assumed patient- and surgeon-level confounding variables). Effect measures were presented as hazard ratios (HRs) and 95% confidence intervals (CIs). RESULTS: We identified 7666 patients of which 2162 (28.7%) underwent re-resection after a median (interquartile range) time of 45 (35-56) days. Patients who received re-resection were less likely to die from any causes (HR 0.68, 95% CI 0.63-0.74, P

Points cles L’apprentissage machine — le developpement de systemes qui, a partir de donnees, apprennent a reconnaitre des tendances et a faire des predictions justes d’evenements a venir[1][1] — a un fort potentiel pour transformer le domaine de la sante. Les outils fondes sur

Points cles Dans un article connexe, Verma et ses collegues s’interessent a la maniere dont des solutions fondees sur l’apprentissage machine peuvent etre elaborees et mises en place pour appuyer la prise de decision medicale[1][1]. Les systemes d’aide a la prise de decision et

[See related articles at www.cmaj.ca/lookup/doi/10.1503/cmaj.202434][1] and [www.cmaj.ca/lookup/doi/10.1503/cmaj.210036][2] KEY POINTS In a companion article, Verma and colleagues discuss how machine-learned solutions can be developed and implemented to support medical decision-making.[1][3] Both

[See related articles at www.cmaj.ca/lookup/doi/10.1503/cmaj.202434][1] and [www.cmaj.ca/lookup/doi/10.1503/cmaj.202066][2] KEY POINTS Related articles have outlined problems with the development of machine-learned solutions for health care and suggested a framework for their optimal development.[1

[See related articles at [www.cmaj.ca/lookup/doi/10.1503/cmaj.202066][2]][2] and [[www.cmaj.ca/lookup/doi/10.1503/cmaj.210036][3]][3] KEY POINTS Machine learning — the process of developing systems that learn from data to recognize patterns and make accurate predictions of future events[1][3

OBJECTIVE School readiness is strongly associated with a child's future school success and well-being. The primary objective of this study was to determine whether meeting 24-hour movement guidelines (national physical activity, sedentary behaviors, and sleep recommendations) was associated with school readiness measured with mean scores in each of the 5 developmental domains of the Early Development Instrument (EDI) in Canadian children aged 4 to 6 years. Secondary objectives include examining the following: (1) the association between meeting 24-hour movement guidelines and overall vulnerability in school readiness and (2) the association between meeting individual physical activity, screen use and sleep recommendations, and overall school readiness. METHODS A prospective cohort study was performed using data from children (aged 4-6 years) who participated in a large-scale primary care practice-based research network. RESULTS Of the 739 participants (aged 5.9 + 0.12 years) in this prospective cohort study, 18.2% met the 24-Hour Movement Guidelines. Linear regression models (adjusted for child/family demographic characteristics, number of siblings, immigration status, and annual household income) revealed no evidence of an association between meeting all 24-hour movement guidelines and any of the 5 domains of the EDI (p > 0.05). Adjusted linear regression models revealed evidence of an association between meeting screen use guidelines and the "language and cognitive development" (β = 0.16, p = 0.004) domain, and for the sleep guideline, there was a statistically significant association with the "physical health and well-being" (β = 0.23, p = 0.001), the "language and cognitive development" (β = 0.10, p = 0.003), and the "communication skills and general knowledge" (β = 0.18, p < 0.001) domain. CONCLUSION Early lifestyle interventions targeting screen use and sleep may be beneficial for improving a child's readiness for school.

Opioid-related policies and guidelines increased the prevalence of rapid opioid dose tapering between 2016 and 2017 in Ontario, Canada, but instances were temporary and short-lived., Reports have emerged of abrupt tapering among recipients of long-term prescription opioids to conform new prescribing guidelines. We conducted a population-based, repeated cross-sectional time-series study among very high-dose (≥200 MME) opioid recipients in Ontario, Canada, to examine changes in the monthly prevalence of rapid tapering from 2014 to 2018, defined as recipients experiencing either a ≥50% reduction in daily doses or abrupt discontinuation sustained for 30 days. Interventional autoregressive integrated moving average models were used to test for significant changes following key guidelines and drug policies and programs. A sensitivity analysis examined rapid tapering sustained for 90 days. The monthly prevalence of rapid tapering events was stable from January 2014 to September 2016 (average monthly prevalence: 1.4%) but increased from 1.4% in October 2016 to 1.8% in April 2017 (P = 0.001), coincident with Ontario's Fentanyl Patch-for-Patch Return Program implementation. Transient spikes in the prevalence of rapid tapering also occurred 2 months after Ontario's delisting of publicly funded high-strength opioids and the release of updated Canadian Opioid Prescribing Guideline for Chronic Pain, reaching 2.3% in March 2017 and July 2017, respectively. However, this prevalence decreased to 1.2% in December 2018 (P < 0.0001). Although the prevalence of abrupt opioid discontinuation was lower, similar trends were observed. Our sensitivity analysis examining long-lasting rapid tapering found similar trends but lower prevalence, with no changes in complete discontinuation. These temporary increases in rapid tapering events highlight the need for improved communication and evidence-based resources for prescribers to minimize negative consequences of evolving policies and guidelines.

Background Machine learning carries considerable promise to improve healthcare delivery. Clinical outcomes that are objectively measured and have serious but preventable consequences are ideal targets for prediction and intervention. Hypoglycemia, defined as a blood glucose of 3.9 mmol/L (70 mg/dL) or lower, meets these criteria. Objective To predict the risk of hypoglycemia using machine learning techniques in hospitalized patients. Methods 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 8,492 GIM admissions and 8,044 CV admissions. Hypoglycemia 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 hypoglycemia 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%. Interpretation Machine learning approaches can accurately identify patients at high risk of hypoglycemia in hospital. Future work will involve evaluating whether implementing this model with targeted clinical interventions can improve clinical outcomes. This article is protected by copyright. All rights reserved.

The Standardized Reporting of Machine Learning Applications in Urology (STREAM-URO) framework was developed to provide a set of recommendations to help standardize how machine learning studies in urology are reported. This framework serves three purposes: (1) to promote high-quality studies and streamline the peer review process; (2) to enhance reproducibility, comparability, and interpretability of results; and (3) to improve engagement and literacy of machine learning within the urological community.

ObjectiveThere are many examples of machine learning based algorithms with impressive diagnostic characteristics. However, a few published studies have evaluated how well they perform when deployed into clinical care. The objective of this study was to evaluate the performance of a recently validated machine-learned model to predict inpatient hypoglycemia following its implementation into clinical care on cardiovascular and vascular surgery ward.MethodsWe conducted a prospective analysis of a machine learning algorithm to predict hypoglycemia. The algorithm was trained, validated, and tested using data from 2013 to 2019. We employed multiple supervised machine learning techniques (e.g., extreme gradient boosting) to predict inpatient hypoglycemia and severe hypoglycemia using a wide-range of patient-level data (i.e., features) including medications, labs, nursing notes, comorbid conditions, among others.ResultsOur study included 3989 hospitalizations during the pre-implementation period and 1916 post-implementation. Approximately one-third of patients were women, the median age was 66 years, 23% received metformin in hospital, 7% received a sulfonylurea, and the median length of stay was 6 days. During the pre-implementation period, more than 5% of patients experienced hypoglycemia during 9.4% (N=12/127 weeks) of study weeks as compared to 0% (N=0/79 weeks) of weeks during the post-implementation period (p=0.012). The weekly variability in the rates of hypoglycemia decreased by approximately 50% from the pre-implementation (standard deviation 1.8, variance 3.4) to implementation phase (standard deviation 1.3, variance 1.6; p=0.03). There was a week-to-week decrease in hypoglycemia rates by 0.03 events per week [95% CI: -0.04, -0.01] (p = 0.004) but no significant change in weekly rates of hyperglycemia (−0.04 [95% CI: -0.10, 0.01]; p=0.102). The severe hypoglycemia events per 100 patients per year was 1.3 pre-implementation and 1.1 following implementation.Discussion and ConclusionOur prospective analysis of a recently validated machine learned model to prevent hypoglycemia demonstrated a reduction in the rates of inpatient hypoglycemia. Our study suggests that machine learning methods can be leveraged to prevent inpatient hypoglycemia.

Opioids are commonly prescribed following childbirth, but data are lacking on trends in postpartum opioid prescribing over time. We examined whether a highly-publicized 2006 case report questioning the safety of codeine during lactation was associated with changes in postpartum opioid prescribing. We conducted a cross-sectional time series analysis of all publicly-funded prescriptions for opioids to postpartum women in Ontario, Canada from April 1, 2000 to March 31, 2017. The intervention was the publication of a case report in 2006 attributing the death of a breastfeeding neonate to maternal codeine use. The primary outcome was the rate of opioid prescribing to postpartum women. Among postpartum women eligible for prescription drug coverage, 17.5% filled an opioid prescription in the third quarter of 2006 (immediately prior to publication of the case report), with codeine representing 89.8% of all prescriptions. By the fourth quarter of 2010, only 12.2% of postpartum women filled an opioid prescription, representing a decline of 30% (p

Purpose: To estimate the positive predictive value of diagnosis and procedure codes for open and endovascular revascularization for peripheral artery disease (PAD) in Ontario administrative databases. Methods: We conducted a retrospective validation study using population-based Ontario administrative databases (2005-2019) to identify a random sample of 600 patients who underwent revascularization for PAD at two academic centres, based on ICD-10 diagnosis codes and Canada Classification of Health Intervention procedure codes. Administrative data coding was compared to the gold standard diagnosis (PAD vs. non-PAD) and revascularization approach (open vs. endovascular) extracted through blinded hospital chart re-abstraction. Positive predictive values and 95% confidence intervals were calculated. Combinations of procedure codes with or without supplemental physician claims codes were evaluated to optimize the positive predictive value. Results: The overall positive predictive value of PAD diagnosis codes was 87.5% (84.6%-90.0%). The overall positive predictive value of revascularization procedure codes was 94.3% (92.2%-96.0%), which improved through supplementation with physician fee claim codes to 98.1% (96.6%-99.0%). Algorithms to identify individuals revascularized for PAD had combined positive predictive values ranging from 82.8% (79.6%-85.8%) to 95.7% (93.5%-97.3%). Conclusion: Diagnosis and procedure codes with or without physician claims codes allow for accurate identifi-cation of individuals revascularized for PAD in Ontario administrative databases.

RESUME CONTEXTE: Les caracteristiques des patients, les soins cliniques, l’utilisation des ressources et les issues cliniques des personnes atteintes de la maladie a coronavirus 2019 (COVID-19) hospitalisees au Canada ne sont pas bien connus. METHODES: Nous avons recueilli des donnees sur tous les adultes hospitalises atteints de la COVID-19 ou de l’influenza ayant obtenu leur conge d’unites medicales ou d’unites de soins intensifs medicaux et chirurgicaux entre le 1er novembre 2019 et le 30 juin 2020 dans 7 centres hospitaliers de Toronto et de Mississauga (Ontario). Nous avons compare les issues cliniques des patients a l’aide de modeles de regression multivariee, en tenant compte des facteurs sociodemographiques et de l’intensite des comorbidites. Nous avons valide le degre d’exactitude de 7 scores de risque mis au point a l’externe pour determiner leur capacite a predire le risque de deces chez les patients atteints de la COVID-19. RESULTATS: Parmi les hospitalisations retenues, 1027 patients etaient atteints de la COVID-19 (âge median de 65 ans, 59,1 % d’hommes) et 783 etaient atteints de l’influenza (âge median de 68 ans, 50,8 % d’hommes). Les patients âges de moins de 50 ans comptaient pour 21,2 % de toutes les hospitalisations dues a la COVID-19 et 24,0 % des sejours aux soins intensifs. Comparativement aux patients atteints de l’influenza, les patients atteints de la COVID-19 presentaient un taux de mortalite perhospitaliere (mortalite non ajustee 19,9 % c. 6,1 %; risque relatif [RR] ajuste 3,46 %, intervalle de confiance [IC] a 95 % 2,56–4,68) et un taux d’utilisation des ressources des unites de soins intensifs (taux non ajuste 26,4 % c. 18,0 %; RR ajuste 1,50, IC a 95 % 1,25–1,80) significativement plus eleves, ainsi qu’une duree d’hospitalisation (duree mediane non ajustee 8,7 jours c. 4,8 jours; rapport des taux d’incidence ajuste 1,45; IC a 95 % 1,25–1,69) significativement plus longue. Le taux de rehospitalisation dans les 30 jours n’etait pas significativement different (taux non ajuste 9,3 % c. 9,6 %; RR ajuste 0,98 %, IC a 95 % 0,70–1,39). Trois scores de risque utilisant un pointage pour predire la mortalite perhospitaliere ont montre une bonne discrimination (aire sous la courbe [ASC] de la fonction d’efficacite du recepteur [ROC] 0,72–0,81) et une bonne calibration. INTERPRETATION: Durant la premiere vague de la pandemie, l’hospitalisation des patients atteints de la COVID-19 etait associee a des taux de mortalite et d’utilisation des ressources des unites de soins intensifs et a une duree d’hospitalisation significativement plus importants que les hospitalisations des patients atteints de l’influenza. De simples scores de risque peuvent predire avec une bonne exactitude le risque de mortalite perhospitaliere des patients atteints de la COVID-19.

Objective Stimulants are first-line pharmacotherapy for individuals with attention-deficit hyperactivity disorder. However, disparities in drug coverage may contribute to inequitable treatment access. In January 2018, the government of Ontario, Canada, implemented a publicly-funded program (OHIP+) providing universal access to medications at no cost to children and youth between the ages of 0 and 24. In April 2019, the program was amended to cover only children and youth without private insurance. We studied whether these policy changes were associated with changes in prescription stimulant dispensing to Ontario children and youth. Methods We conducted a population-based observational natural experiment study of stimulant dispensing to children and youth in Ontario between January 2013 and March 2020. We used interventional autoregressive integrated moving average models to estimate the association between OHIP+ and its subsequent modification with stimulant dispensing trends. Results The implementation of OHIP+ was associated with a significant immediate increase in the monthly rate of stimulant dispensing of 53.6 individuals per 100,000 population (95% confidence interval [CI], 36.8 to 70.5 per 100,000) and a 14.2% (95% CI, 12.8% to 15.6%) relative percent increase in stimulant dispensing rates between December 2017 and March 2019 (1198.6 vs. 1368.7 per 100,000 population). The April 2019 OHIP+ program amendment was associated with an increase in monthly stimulant dispensing trends of 10.2 individuals per 100,000 population (95% CI, 5.0 to 15.5), with rates increasing 7.5% (95% CI, 6.2% to 8.7%) between March 2019 and March 2020 (1368.7 vs. 1470.8 per 100,000 population). These associations were most pronounced among males, children and youth living in the highest income neighbourhoods and individuals aged 20 to 24. Conclusion A publicly-funded pharmacare program was associated with more children and youth being dispensed stimulants.

BACKGROUND AND AIMS: In June 2016, the Ontario, Canada government implemented the Ontario Naloxone Program for Pharmacies (ONPP), authorizing pharmacists to provide injectable naloxone kits at no charge to all Ontario residents. In March 2018, the program was amended to include intranasal naloxone and remove the requirement to present a government health card to the dispensing pharmacist. We examined whether these changes increased naloxone dispensing through the ONPP. DESIGN: Population-based time-series analysis using interventional autoregressive integrated moving average models. SETTING: Ontario, Canada. PARTICIPANTS: All Ontario residents between 1 July 2016 and 31 March 2020. MEASUREMENTS: Monthly rates of pharmacy naloxone dispensing. FINDINGS: Overall, 199 484 individuals were dispensed a naloxone kit during the study period. In the main analysis, the rate of pharmacy naloxone dispensing increased by 65.1% following program changes (55.6-91.8 kits per 100 000 population between February 2018 and May 2018; P = 0.01). In subgroup analyses, naloxone dispensing increased among individuals receiving opioid agonist therapy (OAT) (3374.9-7264.2 kits per 100 000 OAT recipients; P = 0.04) among individuals receiving other prescription opioids (192.8-381.8 kits per 100 000 population prescribed opioids; P

Data extraction from radiology free-text reports is time consuming when performed manually. Recently, more automated extraction methods using natural language processing (NLP) are proposed. A previously developed rule-based NLP algorithm showed promise in its ability to extract stroke-related data from radiology reports. We aimed to externally validate the accuracy of CHARTextract, a rule-based NLP algorithm, to extract stroke-related data from free-text radiology reports.Free-text reports of CT angiography (CTA) and perfusion (CTP) studies of consecutive patients with acute ischemic stroke admitted to a regional stroke center for endovascular thrombectomy were analyzed from January 2015 to 2021. Stroke-related variables were manually extracted as reference standard from clinical reports, including proximal and distal anterior circulation occlusion, posterior circulation occlusion, presence of ischemia or hemorrhage, Alberta stroke program early CT score (ASPECTS), and collateral status. These variables were simultaneously extracted using a rule-based NLP algorithm. The NLP algorithm's accuracy, specificity, sensitivity, positive predictive value (PPV), and negative predictive value (NPV) were assessed.The NLP algorithm's accuracy was 90% for identifying distal anterior occlusion, posterior circulation occlusion, hemorrhage, and ASPECTS. Accuracy was 85%, 74%, and 79% for proximal anterior circulation occlusion, presence of ischemia, and collateral status respectively. The algorithm confirmed the absence of variables from radiology reports with an 87-100% accuracy.Rule-based NLP has a moderate to good performance for stroke-related data extraction from free-text imaging reports. The algorithm's accuracy was affected by inconsistent report styles and lexicon among reporting radiologists.

The American Academy of Pediatrics recommends developmental surveillance and screening in early childhood in primary care. The 18-month visit may be an ideal time for identification of children with delays in language and communication, or symptoms of autism spectrum disorder (ASD). Little is known about the predictive validity of developmental screening tools administered at 18 months. Our objective was to examine the predictive validity of the Infant Toddler Checklist (ITC) at the 18-month health supervision visit, using school readiness at kindergarten age as the criterion measure.We designed a prospective cohort study, recruiting in primary care in Toronto, Canada. Parents completed the ITC at the 18-month visit. Teachers completed the Early Development Instrument (EDI) when the children were in Kindergarten, age 4-6 years. We calculated screening test properties with 95% confidence intervals (CIs). We used multivariable logistic and linear regression analyses adjusted for important covariates.Of 293 children (mean age 18 months), 30 (10.2%) had a positive ITC including: concern for speech delay (n = 11, 3.8%), concern for other communication delay (n = 13, 4.4%), and concern for both (n = 6, 2.0%). At follow-up (mean age 5 years), 54 (18.4%) had overall EDI vulnerability, 19 (6.5%) had vulnerability on the 2 EDI communication domains. The ITC sensitivity ranged from 11% to 32%, specificity from 91% to 96%, false positive rates from 4% to 9%, PPV from 16% to 35%, NPV from 83% to 95%. A positive ITC screen and ITC concern for speech delay were associated with lower scores in EDI communication skills and general knowledge (β = -1.08; 95% CI: -2.10, -0.17; β = -2.35; 95% CI: -3.63, -1.32) and EDI language and cognitive development (β = -0.62; 95% CI: -1.25, -0.18; β = -1.22; 95% CI: -2.11, -0.58).The ITC demonstrated high specificity suggesting that most children with a negative ITC screen will demonstrate school readiness at 4-6 years, and low false positive rates, minimizing over-diagnosis. The ITC had low sensitivity highlighting the importance of ongoing developmental surveillance and screening.

Background Use of neuraxial anaesthesia for open abdominal aortic aneurysm repair is postulated to reduce mortality and morbidity. This study aimed to determine the 90-day outcomes after elective open abdominal aortic aneurysm repair in patients receiving combined general and neuraxial anaesthesia vs general anaesthesia alone. Methods A retrospective population-based cohort study was conducted from 2003 to 2016. All patients ≥40 yr old undergoing open abdominal aortic aneurysm repair were included. The propensity score was used to construct inverse probability of treatment weighted regression models to assess differences in 90-day outcomes. Results A total of 10 447 elective open abdominal aortic aneurysm repairs were identified; 9003 (86%) patients received combined general and neuraxial anaesthesia and 1444 (14%) received general anaesthesia alone. Combined anaesthesia was associated with significantly lower hazards for all-cause mortality (hazard ratio [HR]=0.47; 95% confidence interval [CI], 0.37–0.61) and major adverse cardiovascular events (HR=0.72; 95% CI, 0.60–0.86). Combined patients were at lower odds for acute kidney injury (odds ratio [OR]=0.66; 95% CI, 0.49–0.89), respiratory failure (OR=0.41; 95% CI, 0.36–0.47), and limb complications (OR=0.30; 95% CI, 0.25–0.37), with higher odds of being discharged home (OR=1.32; 95% CI, 1.15–1.51). Combined anaesthesia was also associated with significant mechanical ventilation and ICU and hospital length of stay benefits. Conclusions Combined general and neuraxial anaesthesia in elective open abdominal aortic aneurysm repair is associated with reduced 90-day mortality and morbidity. Neuraxial anaesthesia should be considered as a routine adjunct to general anaesthesia for elective open abdominal aortic aneurysm repair.

BACKGROUND: Drugs are the fastest growing cost in the Canadian health care system, owing to the increasing number of high-cost drugs. The objective of this study was to examine the characteristics of high–drug-cost beneficiaries of public drug plans across Canada relative to other beneficiaries. METHODS: We conducted a cross-sectional study among public drug plan beneficiaries residing in all provinces except Quebec. We used the Canadian Institute for Health Information’s National Prescription Drug Utilization Information System to identify all drugs dispensed to beneficiaries of public drug programs in 2016/17. We stratified the cohort into 2 groups: high–drug-cost beneficiaries (top 5% of beneficiaries based on annual costs) and other beneficiaries (remaining 95%). For each group, we reported total drug costs, prevalence of high-cost claims (> $1000), median number of drugs, proportion of beneficiaries aged 65 or more, the 10 most costly reimbursed medications and the 10 medications most commonly reimbursed. We reported estimates overall and by province. RESULTS: High–drug-cost beneficiaries accounted for nearly half (46.5%) of annual spending, with an average annual spend of $14 610 per beneficiary, compared to $1570 among other beneficiaries. The median number of drugs dispensed was higher among high–drug-cost beneficiaries than among other beneficiaries (13 [interquartile range (IQR) 7–19] v. 8 [IQR 4–13]), and a much larger proportion of high–drug-cost beneficiaries than other beneficiaries received at least 1 high-cost claim (40.9% v. 0.6%). Long-term medications were the most commonly used medications for both groups, whereas biologics and antivirals were the most costly medications for high–drug-cost beneficiaries. INTERPRETATION: High–drug-cost beneficiaries were characterized by the use of expensive medications and polypharmacy relative to other beneficiaries. Interventions and policies to help reduce spending need to consider both of these factors.

BackgroundVariations in inpatient medical care are typically attributed to system, hospital or patient factors. Little is known about variations at the physician level within hospitals. We described the physician-level variation in clinical outcomes and resource use in general internal medicine (GIM).MethodsThis was an observational study of all emergency admissions to GIM at seven hospitals in Ontario, Canada, over a 5-year period between 2010 and 2015. Physician-level variations in inpatient mortality, hospital length of stay, 30-day readmission and use of ‘advanced imaging’ (CT, MRI or ultrasound scans) were measured. Physicians were categorised into quartiles within each hospital for each outcome and then quartiles were pooled across all hospitals (eg, physicians in the highest quartile at each hospital were grouped together). We report absolute differences between physicians in the highest and lowest quartiles after matching admissions based on propensity scores to account for patient-level variation.ResultsThe sample included 103 085 admissions to 135 attending physicians. After propensity score matching, the difference between physicians in the highest and lowest quartiles for in-hospital mortality was 2.4% (95% CI 0.6% to 4.3%, pConclusionsPatient outcomes and resource use in inpatient medical care varied substantially across physicians in this study. Physician-level variations in length of stay and imaging use were unlikely to be explained by patient factors whereas differences in mortality and readmission should be interpreted with caution and could be explained by unmeasured confounders. Physician-level variations may represent practice differences that highlight quality improvement opportunities.

Background: Machine learning (ML) is a rapidly advancing field with increasing utility in health care. We conducted a systematic review and critical appraisal of ML applications in vascular surgery. Methods: MEDLINE, Embase, and Cochrane CENTRAL were searched from inception to March 1, 2021. Study screening, data extraction, and quality assessment were performed by two independent reviewers, with a third author resolving discrepancies. All original studies reporting ML applications in vascular surgery were included. Publication trends, disease conditions, methodologies, and outcomes were summarized. Critical appraisal was conducted using the PROBAST risk-of-bias and TRIPOD reporting adherence tools. The PROSPERO registration number is CRD42021240310. Findings: We included 212 studies from a pool of 2,235 unique articles. ML techniques were used for diagnosis, prognosis, and image segmentation in carotid stenosis, aortic aneurysm/dissection, peripheral artery disease, diabetic foot ulcer, venous disease, and renal artery stenosis. The number of publications on ML in vascular surgery increased from 1 (1991-1996) to 118 (2016-2021). Most studies were retrospective and single center, with no randomized controlled trials. Area under the receiver operating characteristic curve (AUROC) ranged from 0.61-1.00, with 79.5% [62/78] studies reporting AUROC ≥ 0.80. Out of 22 studies comparing ML techniques to existing prediction tools, clinicians, or traditional regression models, 20 performed better and 2 performed similarly. Overall, 94.8% (201/212) studies had high risk-of-bias and adherence to reporting standards was poor with a rate of 41.4%. Interpretation: Machine learning has been increasingly applied to a broad range of vascular surgical conditions with good predictive value. However, overall risk-of-bias and reporting adherence are suboptimal. Future studies should consider standardized tools such as PROBAST and TRIPOD to improve study quality and clinical applicability. Funding: None. Declaration of Interest: All authors declare no competing interests

BACKGROUND Data extraction from radiology free-text reports is time-consuming when performed manually. Recently, more automated extraction methods using natural language processing (NLP) are proposed. A previously developed rule-based NLP algorithm showed promise in its ability to extract stroke-related data from radiology reports. OBJECTIVE We aimed to externally validate the accuracy of CHARTextract, a rule-based NLP algorithm, to extract stroke-related data from free-text radiology reports. METHODS Free-text reports of CT angiography (CTA) and perfusion (CTP) studies of consecutive patients with acute ischemic stroke admitted to a regional Stroke center for endovascular thrombectomy were analyzed from January 2015 - 2021. Stroke-related variables were manually extracted (reference standard) from the reports, including proximal and distal anterior circulation occlusion, posterior circulation occlusion, presence of ischemia, hemorrhage, Alberta stroke program early CT score (ASPECTS), and collateral status. These variables were simultaneously extracted using a rule-based NLP algorithm. The NLP algorithm's accuracy, specificity, sensitivity, positive predictive value (PPV), and negative predictive value (NPV) were assessed. RESULTS The NLP algorithm's accuracy was >90% for identifying distal anterior occlusion, posterior circulation occlusion, hemorrhage, and ASPECTS. Accuracy was 85%, 74%, and 79% for proximal anterior circulation occlusion, presence of ischemia, and collateral status respectively. The algorithm had an accuracy of 87-100% for the detection of variables not reported in radiology reports. CONCLUSIONS Rule-based NLP has a moderate to good performance for stroke-related data extraction from free-text imaging reports. The algorithm's accuracy was affected by inconsistent report styles and lexicon among reporting radiologists.

Background The capacity of general internal medicine (GIM) clinical teaching units has been strained by decreasing resident supply and increasing patient demand. The objective of our study was to compare the number of residents (supply) with the volume and duration of patient care activities (demand) to identify inefficiency. Methods Using the most recently available data from an academic teaching hospital in Toronto, Ontario, we identified each occurrence of a set of patient care activities that took place on the clinical teaching unit from 2015 to 2019. We completed a descriptive analysis of the frequencies of these activities and how the frequencies varied by hour, day, week, month and year. Patient care activities included admissions, rounds, responding to pages, meeting with patients and their families, patient transfers, discharges and responding to cardiac arrests. The estimated time to complete each task was based on the available data in our electronic medical record system and interviews with GIM physicians and trainees. To calculate resident utilization, the person-hours of patient care tasks was divided by the person-hours of resident supply. Resident utilization was computed for 3 scenarios corresponding to varying levels of resident absenteeism. Results During the study period, there were 14 581 consultations to GIM from the emergency department. Patient volumes tended to be highest during January and lowest during May and June, and highest on Monday morning and lowest on Friday night. Daily admissions to hospital from the emergency department were higher on weekdays than on weekends, and hourly admissions peaked at 8 am and between 3 pm and 1 am. Weekday resident utilization was generally highest between 8 am and 2 pm, and lowest between 1 am and 8 am. In a scenario in which all residents were present apart from those who were post-call, resident utilization generally never exceeded 100%; in scenarios in which at least 1 resident was absent owing to illness or vacation, it was common for resident utilization to approach or exceed 100%, particularly during daytime working hours. Interpretation Analyzing supply and demand on a GIM ward has allowed us to identify periods when supply and demand are not aligned and to demonstrate empirically the vulnerability of current staffing models. These data have the potential to inform and optimize scheduling on an internal medicine ward.

Cross-sector partnerships are vital for maintaining resilient health systems; however, few studies have sought to empirically assess the barriers and enablers of effective and responsible partnerships during public health emergencies. Through a qualitative, multiple case study, we analyzed 210 documents and conducted 26 interviews with stakeholders in three real-world partnerships between Canadian health organizations and private technology startups during the COVID-19 pandemic. The three partnerships involved: 1) deploying a virtual care platform to care for COVID-19 patients at one hospital, 2) deploying a secure messaging platform for physicians at another hospital, and 3) using data science to support a public health organization. Our results demonstrate that a public health emergency created time and resource pressures throughout a partnership. Given these constraints, early and sustained alignment on the core problem was critical for success. Moreover, governance processes designed for normal operations, such as procurement, were triaged and streamlined. Social learning, or the process of learning from observing others, offset some time and resource pressures. Social learning took many forms ranging from informal conversations between individuals at peer organisations (e.g., hospital chief information officers) to standing meetings at the local university’s city-wide COVID-19 response table. We also found that startups’ flexibility and understanding of the local context enabled them to play a highly valuable role in emergency response. However, pandemic fueled “hypergrowth” created risks for startups, such as introducing opportunities for deviation away from their core value proposition. Finally, we found each partnership navigated intense workloads, burnout, and personnel turnover through the pandemic. Strong partnerships required healthy, motivated teams. Visibility into and engagement in partnership governance, belief in partnership impact, and strong emotional intelligence in managers promoted team well-being. Taken together, these findings can help to bridge the theory-to-practice gap and guide effective cross-sector partnerships during public health emergencies.

Background Cerebral protection remains the cornerstone of successful aortic surgery; however, there is no consensus as to the optimal strategy. Objective To compare the safety and efficacy of innominate to axillary artery cannulation for delivering antegrade cerebral protection during proximal aortic arch surgery. Methods This randomized controlled trial (The Aortic Surgery Cerebral Protection Evaluation CardioLink-3 Trial, ClinicalTrials.gov Identifier: NCT02554032 ), conducted across 6 Canadian centers between January 2015 and June 2018, allocated 111 individuals to innominate or axillary artery cannulation. The primary safety outcome was neuroprotection per the appearance of new severe ischemic lesions on the postoperative diffusion-weighted-magnetic resonance imaging. The primary efficacy outcome was the difference in total operative time. Secondary outcomes included 30-day all-cause mortality and postoperative stroke. Results One hundred two individuals (mean age, 63 ± 11 years) were in the primary safety per-protocol analysis. Baseline characteristics between the groups were similar. New severe ischemic lesions occurred in 19 participants (38.8%) in the axillary versus 18 (34%) in the innominate group (P for noninferiority = .0009). Total operative times were comparable (median, 293 minutes; interquartile range, 222-411 minutes) for axillary versus (298 minutes; interquartile range, 231-368 minutes) for innominate (P for superiority = .47). Stroke/transient ischemic attack occurred in 4 (7.1%) participants in the axillary versus 2 (3.6%) in the innominate group (P = .43). Thirty-day mortality, seizures, delirium, and duration of mechanical ventilation were similar in both groups. Conclusions diffusion-weighted magnetic resonance imaging assessments indicate that antegrade cerebral protection with innominate cannulation is safe and affords similar neuroprotection to axillary cannulation during aortic surgery, although the burden of new neurological lesions is high in both groups.

Background The American Academy of Pediatrics recommends developmental screening at multiple visits using both a general developmental tool and an autism spectrum disorder (ASD)-specific tool. The Canadian Paediatric Society recommends screening at a single visit at 18 months. There is no consensus on which tool is best suited for one-time screening. The Infant Toddler Checklist (ITC) identifies children who are at risk for communication impairment, may detect ASD, and may be a useful screening tool at the 18-month visit. Objectives To examine the screening test accuracy of the ITC at the 18-month visit to predict school readiness at kindergarten age. Design/Methods This prospective cohort study included children who attended primary care health supervision visits in Toronto, Canada. Parents completed the ITC at the 18-month visit and teachers completed the Early Development Instrument (EDI - a population-level measure of school readiness in kindergarten) at 4-6 years. An ITC screen is positive if there is concern for expressive speech delay (speech composite below the 10th percentile) and/or other communication delay (social composite, symbolic composite or the total score below the 10th percentile). Children were considered overall vulnerable on the EDI if at least one of five domains was below the 10th percentile of the Ontario population: language and cognitive development; physical health and well-being; social competence; emotional maturity; communication skills and general knowledge. We calculated screening test properties with 95% confidence intervals (CIs), using EDI vulnerability as the criterion measure. We used multivariable regression models to examine the association between the ITC and EDI domains. Results Of 293 children, 30 (10%) had a positive ITC. At follow-up, 54 (18%) children had a teacher-reported EDI vulnerability. The specificity (range, 87%-96%) and negative predictive value (range, 83%-95%) for the ITC were high; false positive rate was low (range, 4%-13%); sensitivity was low (range, 11%-37%). A positive ITC was associated with a lower score in EDI language and cognitive development (b= -0.62, 95% CI: -1.25, -0.18; P=0.046) and EDI communication skills and general knowledge (b= -1.08, 95% CI: -2.10, -0.17; P=0.036). We found no evidence of an association between ITC and EDI vulnerability. Conclusion The ITC at 18 months had high specificity (87%-96%) suggesting that most children with a negative ITC will demonstrate school readiness at 4-6 years. False positive rates were low, minimizing over-diagnosis. The ITC, with its focus on speech and language, communication disorders and ASD, may be a candidate for screening at the 18-month visit.

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/ ).

Prior research has shown that concordance with the guideline-endorsed recommendation to re-resect patients diagnosed with primary T1 bladder cancer (BC) is suboptimal. Therefore, the aim of this population-based study was to identify factors associated with re-resection in T1 BC.We linked province-wide BC pathology reports (January 2001 to December 2015) with health administrative data sources to derive an incident cohort of patients diagnosed with T1 BC in the province of Ontario, Canada. Re-resection was ascertained by a billing claim for transurethral resection within 2 to 8 weeks after the initial resection, accounting for system-related wait times. Multivariable logistic regression analysis accounting for the clustered nature of the data was used to identify various patient-level and surgeon-level factors associated with re-resection. P values0.05 were considered statistically significant (2-sided).We identified 7,373 patients who fulfilled the inclusion criteria. Overall, 1,678 patients (23%) underwent re-resection. Patients with a more aggressive tumor profile and individuals without sufficiently sampled muscularis propria as well as younger, healthier and socioeconomically advantaged patients were more likely to receive re-resection (all p0.05). In addition, more senior, lower volume and male surgeons were less likely to perform re-resection for their patients (all p0.05).Only a minority of all patients received re-resection within 2 to 8 weeks after initial resection. To improve the access to care for potentially underserved patients, we suggest specific knowledge translation/exchange interventions that also include equity aspects besides further promotion of evidence-based instead of eminence-based medicine.

INTRODUCTION AND OBJECTIVE:A second transurethral resection within 2 to 6 weeks after the initial resection (i.e. re-resection) is recommended for patients diagnosed with primary T1 bladder cancer ...