• The Our Patients patient demographics table now includes 3 additional patient-level variables. These variables are No Health Card Number, No Postal Code, and Discharged to Long Term Care. These are presented as percentages in the Our Patients table.
• The Our Patients table now includes 2 additional neighborhood-level variables. These variables are Percentage Visible Minority and After Tax Income. Neighborhood-level variables are from Statistics Canada's 2016 census, linked to hospitalizations at the level of dissemination area. Each patient is assigned a value based on their dissemination area, and median (25th - 75th) are presented in the Our Patients table.
• The Advanced Imaging indicator now considers two imaging measurements from the same imaging order as a single imaging test. This is a small update to correct for duplicate measurements, which only affects a minority of hospitals.
• The 7-Day Readmission and 30-Day Readmission indicator definitions have been updated to further align with CIHI's definition. First, episodes of care with the last record as self sign-out are excluded from the denominator. Second, episodes of care that involve obstetric delivery are excluded from the numerator. Third, episodes of care where any encounter (including transfers) involves mental heath, palliative care, or chemotherapy are excluded from the numerator. Elective admissions are excluded from the numerator only if the initial encounter in the readmission is coded as elective. These updates will generally result in slightly lower readmission rates.
• Added text to the Appropriate RBC Transfusion indicator page legend to clarify that below average is undesirable and above average is desirable.
• For risk adjustment of In-Hospital Mortality, COVID-19 as a risk factor has been added to models for all diagnosis groups where COVID-19 is not the most responsible discharge diagnosis. Hospitalizations where COVID-19 is the most responsible diagnosis continue to be modeled as a separate diagnosis group in risk adjustment. This is based on a 2022 update to CIHI's HSMR.
• The In-Hospital Mortality indicator page, Exclude filter on the top left, was modified for risk-adjusted results. When the filter is set to "Nothing", hospitalizations with palliative care as the most responsible discharge diagnosis are now included in determining the diagnosis groups accounting for 80% of deaths that are included in the risk-adjusted results. The default setting of the in-hospital mortality indicator page is still to exclude hospitalizations with palliative care as the most responsible discharge diagnosis when determining these diagnosis groups

This report was optimized for 1080p screens on modern browsers such as Google Chrome, Microsoft Edge, and Mozilla Firefox in full screen mode. The report will read well on higher resolutions and will default to a mobile "scrollable" layout for resolutions less than 768p. For resolutions between 768p and 1080p, the report may be uncomfortable to read. In this situation, experimenting with the "zoom" feature on your browser may improve readability.

There are several reasons why data may not be shown in the report, including the absence of data, identified data quality issues, or the suppression of data. Data on fewer than 6 admissions are suppressed to reduce the privacy risk that could lead to re-identification of individuals and residual disclosure of information.

This report includes General Medicine (GIM) patients at GeMQIN participating hospitals between . Based on guidance from your hospital, the following criteria was used to identify General Medicine patients at your hospital:

The data for this report were collected by GEMINI. GEMINI is one of Canada's largest hospital patient data repositories for research. Established in 2015 as a hospital research collaborative based out of Unity Health Toronto with seven hospitals, GEMINI currently holds data on >1.8 million admissions from over 30 Ontario hospitals. GEMINI collects data for all medical (including general medicine, cardiology, oncology, etc.) and intensive care hospitalizations, covering >60% of all inpatients across the province.

GEMINI contains administrative and clinical data extracted from hospital information systems and administrative databases. Administrative data include variables standardized for reporting to the Canadian Institute for Health Information (e.g., patient demographics, admission and discharge dates, diagnosis codes). Clinical data include rich and granular variables from the electronic health record (e.g., patient vital signs, laboratory test results, medication orders). GEMINI receives hospital data every 3 months and has established analytical processes to handle the large volume and broad range of data.

GEMINI data are collected through research ethics board-approved protocols and are governed by the GEMINI data governance policies.

The GEMINI team has undertaken extensive efforts to validate the data that were collected. These include checking data quality by testing for missing data, consistency over time, and implausible values. We have developed workflows for preparing research-ready data that include deidentification, quality control, standardization and validation. Our methodology for assessing data quality has been rigorously validated (98%-100% accuracy), as described in this article. In addition, a data validator at each hospital manually compared the extracted data from GEMINI with the data in the hospital's electronic health records.

Below is a list of hospitals included in this report.

• Brampton Civic Hospital - William Osler Health System
• Cortelluci Vaughan Hospital - Mackenzie Health
• Etobicoke General Hospital - William Osler Health System
• Grand River Hospital
• Greater Niagara General Site - Niagara Health
• Hamilton General Hospital - Hamilton Health Sciences
• Humber River Hospital
• Juravinski Hospital - Hamilton Health Sciences
• Kingston General Hospital - Kingston Health Sciences Centre
• Mackenzie Richmond Hill Hospital - Mackenzie Health
• Markham Stouffville Hospital - Oak Valley Health
• Mount Sinai Hospital - Sinai Health
• Michael Garron Hospital - Toronto East Health Network
• North York General Hospital
• Sault Area Hospital
• St. Catharines Site - Niagara Health
• St. Joseph's Health Centre - Unity Health Toronto
• St. Mary's General Hospital
• St. Michael's Hospital - Unity Health Toronto
• Sunnybrook Health Sciences Centre
• Thunder Bay Regional Health Sciences Centre
• University Hospital - London Health Science Centre
• Victoria Hospital - London Health Science Centre
• Welland Hospital Site - Niagara Health

The indicators in this report were selected by the GeMQIN Report Development Committee, which includes the program's provincial clinical leads, physicians and interdisciplinary health professionals, hospital administrators, quality improvement experts, and researchers. Part of GeMQIN's ongoing work over the coming years will be to further develop quality indicators that are relevant to hospital medicine, and we welcome feedback and participation in that process. If you have any feedback about these reports, please contact us at GeMQIN@ontariohealth.ca.

Details and definitions can be found on the "Definitions, Methods, and QI Tips" button on the top-left of each indicator page. The table below provides a summary. Refer to the OurPractice Background and Indicator Details document for additional information.

Table 1: Definitions and Calculations for OurPractice Report Indicators

Indicator	Encounter-level Definition	Hospital-level Calculation	Risk Adjustment	Exclusions
Total Length of Stay (LOS)	Number of days from admission to discharge	Median total length of stay	Yes	Transfers in from, or out to, another acute care institution Total LOS greater than 365 days
Acute Length of Stay	Number of days from admission to discharge, excluding ALC days	Median acute length of stay	Yes	Transfers in from, or out to, another acute care institution Total LOS greater than 365 days
Alternate Level of Care	Number of days spent on an alternate level of care service	Percentage of all inpatient days that were ALC Median number of ALC days among hospitalizations with ALC days	No	Total LOS greater than 365 days
7-Day Readmission	Readmission to any medicine or ICU service at a GeMQIN hospital within 7 days of discharge	Percentage of discharges that were followed by a 7-day readmission	Yes	Episodes of care with invalid health card number, ending in death, with at least one record for palliative care or mental health as most responsible diagnosis, transferred to and discharged from a hospital outside of GeMQIN, with the last record as self sign-out.
30-Day Readmission	Readmission to any medicine or ICU service at a GeMQIN hospital within 30 days of discharge	Percentage of discharges of care that were followed by a 30-day readmission	Yes	Episodes of care with invalid health card number, ending in death, with at least one record for palliative care or mental health as most responsible diagnosis, transferred to and discharged from a hospital outside of GeMQIN, with the last record as self sign-out.
In-Hospital Mortality	Death occurring in hospital	Percentage of hospitalizations that ended in death in hospital	Yes	Age greater than 120 years Total LOS greater than 365 days Palliative care as most responsible discharge diagnosis Medical assistance in dying
Advanced Imaging	Total number of CT, MRI, and ultrasound tests from admission to discharge	Mean number of tests	Yes	Total LOS greater than 365 days
Routine Bloodwork	Total number of CBC and electrolyte tests from admission to discharge	Mean number of tests	Yes	Total LOS greater than 365 days
Appropriate RBC Transfusion	RBC transfusion with a most recent hemoglobin value < 80g/L within 48 hours prior to transfusion	Percentage of RBC Transfusions that were appropriate	No	RBC transfusions with no pre-transfusion hemoglobin values within 48 hours prior (these are rare, ~2%)

Abbreviations: ALC, alternate level of care; CBC, complete blood count; CT, computerized tomography; GeMQIN, General Medicine Quality Improvement Network; ICU, intensive care unit; LOS, length of stay; MRI, magnetic resonance imaging; RBC, red blood cell

The 30-day timeframe to measure readmission is what is most commonly used to report hospital readmissions. However, recent research (Graham et al, 2018) suggests that earlier readmissions (e.g., within 7 days) are more likely to be preventable and amenable to hospital-based interventions, while later readmissions (e.g., within 30 days) are more likely to be preventable and improved upon by home and community-based interventions.

Readmission indicators exclude episodes of care with at least one record for mental health as the most responsible discharge diagnoses, and do not consider readmissions for mental health as readmissions. This is to align with CIHI's reporting, and because we do not comprehensively capture all mental health hospitalizations. Mental health hospitalizations may go to non-medical services, and if included, may bias the reported data.

Different hospitals treat different types of patients. This makes it difficult to assess hospital performance in a fair manner. For example, certain hospitals may treat sicker patients and thus may appear to perform worse despite delivering high quality care. Risk adjustment aims to address these differences by holding constant case mix and patient severity so that fair hospital assessments can be made.

Risk adjustment is based on comparing what we observed during the reporting period to what we should expect based on a hospital's case mix and patient severity. Our expectation is based on flexible regression models trained on historical data at all hospitals during the four years immediately before the reporting period. These regression models consider a patient's age, sex, Charlson comorbidity index score at admission (Quan et al., 2011), a modified laboratory-based acute physiology score (mLAPS) based on 12 biochemical parameters at admission (Escobar et al., 2008, van Walraven et al., 2010, Roberts et al., 2023), and whether the admission was elective or urgent. Separate regression models are fit for each diagnosis group, allowing each diagnosis group to have their own associations between risk adjustment variables and quality indicators. These regression models also consider additional variables specific to each quality indicator (see "What risk adjustment was applied to each quality indicator?" below for more details). Refer to the OurPractice Background and Indicator Details for further information.

Our risk adjustment is based on methodology from the Canadian Institute for Health Information (CIHI) and Kaiser Permanente. Age, sex, Charlson comorbidity index score at admission, a modified laboratory-based acute physiology score at admission (mLAPS), and whether the admission was elective or urgent are included in the base set of variables for all risk adjustment models. This mLAPS score is based on Kaiser Permanente methodology (Escobar et al., 2008, van Walraven et al., 2010) and our modification involves removing troponin from the score because there is no way to reconcile high-sensitivity troponin tests when calculating the score (Roberts et al., 2023). Please see Table 2 below for a summary of risk adjustment. Refer to the OurPractice Background and Indicator Details document for additional information. Variables for the readmission indicators are collected from the index admission.

Table 2: Summary of Risk Adjustments for OurPractice Report Indicators

Indicator	Additional Adjustment Variables	Risk Adjustment Model	Hospital Assessment Method	Note
Total Length of Stay	Number of acute care hospitalizations in the past 6 months	Semiparametric ordinal regression	Negative binomial GLMM	None
Acute Length of Stay	Number of acute care hospitalizations in the past 6 months	Semiparametric ordinal regression	Negative binomial GLMM	None
7-Day Readmission	Number of acute care hospitalizations in the past 6 months	Logistic regression	Observed to expected ratio	Consistent with CIHI, this calculation groups contiguous inpatient hospitalizations within GeMQIN into a single episode of care. The following are not counted as readmissions: readmissions where the initial encounter was elective, readmissions involving at least 1 record for chemotherapy for neoplasm, palliative care, obstetric delivery, mental health, medical assistance in dying.
30-Day Readmission	Number of acute care hospitalizations in the past 6 months	Logistic regression	Observed to expected ratio	Consistent with CIHI, this calculation groups contiguous inpatient hospitalizations within GeMQIN into a single episode of care. The following are not counted as readmissions: readmissions where the initial encounter was elective, readmissions involving at least 1 record for chemotherapy for neoplasm, palliative care, obstetric delivery, mental health, medical assistance in dying.
In-Hospital Mortality	Total length of stay, Transfer in from acute care institution, COVID-19 as a risk factor	Logistic regression	Observed to expected tatio	Consistent with CIHI's HSMR, risk-adjusted mortality is restricted to the top diagnosis groups accounting for ~80% of all in-hospital mortality.
Advanced Imaging	Number of acute care hospitalizations in the past 6 months	Semiparametric ordinal regression	Negative binomial GLMM	None
Routine Bloodwork	Number of acute care hospitalizations in the past 6 months	Semiparametric ordinal regression	Negative binomial GLMM	None

Abbreviations: CIHI, Canadian Institute for Health Information; GeMQIN, General Medicine Quality Improvement Network; GLMM, generalized linear mixed model; HSMR, hospital standardized mortality ratio

Hospital assessments of binary indicators (i.e., in-hospital mortality, 7-day readmission, 30-day readmission) use an observed to expected ratio (O:E ratio) framework consistent with CIHI's calculation of the Hospital Standardized Mortality Ratio. In this framework, the observed number of events during the reporting period is compared to the expected number of events. This expected number of events is the sum of expected values (i.e., predicted probabilities) from risk adjustment models for all encounters at that hospital during the reporting period. A 95% confidence interval is calculated around this ratio, and the ratio is multiplied by the network-wide rate to produce a standardized rate.

Hospital assessments of continuous and count indicators (i.e., total length of stay, acute length of stay, routine bloodwork, advanced imaging) use a random intercept framework. A regression is fit with expected values from risk adjustment models as a fixed effect and random intercepts for each hospital. The hospital-level random intercepts represent the difference between observed values at a given hospital and network-wide observed values, after accounting for differences in case mix and patient severity (holding expected values constant). A 95% confidence interval is calculated for each hospital's random intercept, and the intercept is multiplied by the network-wide average to produce a standardized value.

Risk-adjusted values should be interpreted in the context of their 95% confidence intervals. These intervals reflect uncertainty in the hospital's risk-adjusted values. A hospital whose entire interval is below the expected value (denoted by the solid vertical black line) will be classified as "below average" and colored blue. A hospital whose entire interval is above the expected value will be classified as "above average" and colored magenta. A hospital whose interval contains the expected value will be classified as "average" and colored gray because there is no evidence that the hospital was different than average. Note that "above average" and "below average" comment on the direction of the effect and should not be interpreted as "good" or "bad" (for example, a risk-adjusted mortality rate that is below average may be desirable).

Each hospital's risk-adjusted value should be interpreted based on its position relative to the expected value (solid vertical black line). Risk-adjusted values are not designed for direct comparison between individual hospitals, and risk-adjusted values are not designed for ranking hospitals relative to one another.

Unadjusted numbers are simple summary statistics that describe what was observed during the reporting period. Unadjusted numbers do not take into account differences in case mix or patient severity. Risk-adjusted numbers compare what was observed during the reporting period to what would be expected based on the case mix and patient severity at a given hospital. For binary outcomes, each hospital is assigned a value representing the difference between what was observed and what was expected. A value above 1 means that observed values were higher than expected and a value below 1 means that observed values were lower than expected. Each hospital's value is multiplied by the combined average value at all hospitals to produce a "standardized" value. For example, if a hospital had 1.2 times more deaths than expected and the in-hospital mortality rate was 6% at all hospitals, the risk-adjusted mortality rate at that hospital would be 1.2 x 6% = 7.2%. A similar concept is applied to numeric and count outcomes, except hospital comparisons are performed on the log scale so the expected value is defined by 0 instead of 1.

Unadjusted values may be meaningfully different than risk-adjusted values for total length of stay, acute length of stay, and in-hospital mortality. Length of stay values differ because unadjusted values are medians while risk-adjusted values are estimated means. Mortality values differ because risk adjustment only considers diagnosis groups accounting for 80% of mortality, consistent with CIHI's HSMR (i.e., risk-adjusted mortality estimates are based on a subgroup of patients from high-mortality diagnosis groups).

All data in this report have been recalculated using the most recent data available, and in some cases updated methodology. There may be small variations between values in this report and the previous report. Variations due to updated methodology are documented above in "What has changed since the previous report?".

Three hospitals included in the previous report are not included in this report. Two hospitals not included in the previous report are included in this report. As a result, numbers in this report that rely on network-wide data may have changed compared to the previous report (e.g., cards on the left of indicator pages, time trend plots on the bottom-left, expected values in risk-adjusted plots in the middle and top-right).

For each hospitalization, the most responsible inpatient discharge diagnosis is coded using ICD-10 codes in hospital administrative data. We use the Clinical Classifications Software Refined (CCSR) to group ICD-10 codes into clinically meaningful diagnosis groups. This approach allows us to aggregate >70,000 unique ICD-10 diagnosis codes into ~540 mutually exclusive categories across 22 body systems. When a proxy most responsible discharge diagnosis is present, that code is used in place of the most responsible discharge diagnosis.

We define region based on the Ontario Health Region within which a hospital is located.

These data are intended to help hospitals understand the quality of general medicine care and inform quality improvement efforts. There are sometimes large differences in quality indicator performance between hospitals. These differences may be due to variation in processes and quality of care, case mix, and/or patient characteristics. Risk adjustment aims to standardize for case mix and patient severity so that hospital estimates can be attributed to quality of care. It should be acknowledged that risk adjustment is imperfect and is unable to consider factors that are not available in hospital or administrative health data.

We encourage hospitals to interpret these data with their local context in mind.

The COVID-19 pandemic has evolved substantially over time. Ontario’s population has become increasingly vaccinated, the demographics of people infected with COVID-19 have shifted, new treatments have become available for both mild and severe COVID-19, and the virulence of dominant COVID-19 variants has shifted. The time-varying nature of the COVID-19 pandemic makes it difficult to estimate baseline risk of patients hospitalized with COVID-19 using historical data, leading to an overestimation of baseline risk during the reporting period. As such, observed in-hospital mortality is lower than expected mortality based on risk-adjustment models at most hospitals. Because of this, we do not present any diagnosis-specific risk-adjusted estimates for COVID-19.