2012 HSMR Methodology

Introduction | Results

On this page:

• Bibliography
• Diagnosis Groups

Technical Notes for the 2012 HSMR Public Release

Definition

The hospital standardized mortality ratio (HSMR) is a ratio of the actual number of in-hospital deaths in a region or hospital to the number that would have been expected based on the types of patients a region or hospital treats.

Method of calculation

Numerator: Observed deaths, or actual number of in-hospital deaths that occurred in a hospital or region.

Denominator: Expected deaths, or number of deaths that would have occurred in a hospital or region had the mortality of these patients been the same as the mortality of similar patients across the country, based on the reference year (2009–2010).

Regional or corporation-level HSMRs are calculated as the sum of observed deaths for all acute care sites divided by the sum of expected deaths for all acute care sites multiplied by 100. Regional and facility HSMR results are based on where patients were treated, not where they lived.

Inclusion criteria

1. Discharge between April 1 of a given year and March 31 of the following year.
2. Admission to an acute care institution.
3. Discharge with diagnosis group of interest (that is, one of the diagnosis groups that account for approximately 80% of in-hospital deaths).
4. Age at admission between 29 days and 120 years.
5. Sex recorded as male or female.
6. Length of stay up to 365 consecutive days.
7. Admission category is elective or emergent/urgent.
8. Canadian resident.

Exclusion criteria

1. Cadavers.
2. Stillborns.
3. Sign-outs (that is, discharged against medical advice).
4. Patients who did not return from a pass,
5. Neonates (age of admission less than or equal to 28 days).
6. Records with brain death as most responsible diagnosis code.
7. Records with palliative care as most responsible diagnosis code.

Logistic regression models

Some types of patients are more likely than others to die in hospital. For example, older patients and those with certain health problems on admission (comorbidities) are at higher risk. Since these risk factors can change over time and vary from place to place, the HSMR calculation takes these differences into account using a statistical technique known as logistic regression.

For each of the HSMR diagnosis groups, the HSMR logistic regression models are fitted with age, sex, length-of-stay (LOS) group, admission category, comorbidity group and transfers as independent variables. The models are based on data from all acute hospitals in Canada. Coefficients derived from the logistic regression models are used to calculate the probability of in-hospital death. The expected number of deaths for a hospital, corporation or region is based on the sum of the probabilities of in-hospital death for eligible discharges from that organization. The 95% confidence interval is calculated using Byar’s approximation.

Interpretation

A ratio equal to 100 suggests that there is no difference between a local mortality rate and the average national experience, given the types of patients cared for. An HSMR greater or less than 100 suggests that a local mortality rate is higher or lower, respectively, than the national experience.

The confidence intervals describe the precision of the HSMR estimate. The upper and lower confidence intervals are estimated to contain the true value of the HSMR 19 times out of 20 (95% confidence interval). A confidence interval that includes 100 suggests that the HSMR is not statistically different from the 2009–2010 baseline of 100.  HSMR results whose confidence interval does not include 100 and are therefore statistically different from the 2009–2010 baseline are denoted with a symbol in the report.

While HSMR adjusts for a number of factors affecting the risk of in-hospital mortality, it does not control for everything. Therefore, HSMR results are most useful in tracking trends over time.

Data source

Discharge Abstract Database, Canadian Institute for Health Information.
Hospital Morbidity Database, Canadian Institute for Health Information.

Reference period

The reference year for HSMR calculations is 2009–2010. To allow for comparisons over time, the coefficients derived from the model using the reference year are used to determine expected deaths for all reported years.

Comprehensiveness

Results are only reported for regions and acute care facilities that meet a statistical threshold for public reporting: at least 2,500 qualifying discharges in each of the last three years being reported i.e. 2009–2010, 2010–2011 and 2011–2012

Bibliography

Alexandrescu, R. et al. “Logistic vs Hierarchical Modeling: An Analysis of a Statewide Inpatient Sample.” Journal of American College of Surgeons 213, 3 (2011): pp. 392–401.

Bottle, A., B. Jarman and P. Aylin. “Hospital Standardized Mortality Ratios: Sensitivity Analyses on the Impact of Coding.” Health Services Research 46, 6 Part 1 (2011): pp. 1741–1761.

Bottle, A., B. Jarman and P. Aylin. “Strengths and Weaknesses of Hospital Standardised Mortality Ratios.” BMJ 342 (2011): p. 7116.

Breslow, N. E. and N. E. Day. Statistical Methods in Cancer Research: Volume II—The Design and Analysis of Cohort Studies. Lyon, France: International Agency for Research on Cancer, 1987.

Jarman, B. et al. “Explaining Differences in English Hospital Death Rates Using Routinely Collected Data.”BMJ 318 (1999): pp. 1515–1520, accessed from <http://bmj.bmjjournals.com/cgi/content/full/318/7197/1515>.

Jarman, B., A. Bottle and P. Aylin. “Monitoring Changes in Hospital Standardised Mortality Ratios.” BMJ 330 (2005): p. 329.

Quan, H. et al. “Updating and Validating the Charlson Comorbidity Index and Score for Risk Adjustment in Hospital Discharge Abstracts Using Data From 6 Countries.” American Journal of Epidemiology 173, 6 (2011): pp. 676–682.

Diagnosis Groups That Account for 80% of Acute Care In-Hospital Mortality

Contact us: hsmr@cihi.ca