Nested case-control analysis of a retrospective cohort

Thiazolidinediones and cardiovascular outcomes in older patients with diabetes. Lipscombe LL, Gomes T, Lévesque LE, Hux JE, Juurlink DN, Alter DA. JAMA 2007; 298(22):2634–43.

Design Overview: Older Ontario diabetics who were using oral hypoglycemics were identified by anonymously linking the Ontario Diabetes Database (which was developed at ICES) to Ontario Drug Benefit Program prescription records. Cardiovascular outcomes and deaths were detected using the National Ambulatory Care System data (emergency room visits), Canadian Insitute for Health Information data (inpatient hospitalizations) and the ICES-enriched Registered Persons Database (death or last health care contact). Using the ICES data, the investigators were able to control for age, sex, diabetes duration, history of cardiovascular disease and co-morbid disease status. They were also able to determine drug use and its fluctuations over time and control for history of drug use and use of concomitant cardiovascular medications. Because this study was nested in a population-based cohort, the investigators were able to calculate the absolute risk associated with thiazolidinediones use in addition to the relative risk. 

Cohort with propensity-based matching

Antipsychotic drug use and mortality in older adults with dementia. Gill SS, Bronskill SE, Normand SL, Anderson GM, Sykora K, Lam K, Bell CM Lee PE, Fischer HD, Herrmann N, Gurwitz JH, Rochon PA. Ann Intern Med 2007; 146(11):775–86. 

Design Overview: The goal of the research was to examine the association between variations in use of antipsychotic drugs and variations in their clinical indications. Because Ontario has a centralized, standardized system for determining the need for nursing home care, these investigators were able to examine the impact of variations in antipsychotic drug use across nursing homes, assured that choice of nursing home was not correlated the need for these medications; i.e., this study took advantage of a natural experiment that was occurring across the province. A list of active nursing homes, their size and geographic location was available from the Ministry of Health. ICES data allowed for the identification of all nursing home residents at one point in time. Anonymously linked data provided information about the residents’ prescriptions (Ontario Drug Benefit Program Database) and physician claims (OHIP Data) and hospitalization data (CIHI Discharge Abstracts) provided information about indications for treatment and co-morbid illnesses. The prescribing rate of each facility was also compared to accepted benchmarks and standardized ratios computed that accounted for differences in the prevalence of clinical indications across nursing homes.

Cross-sectional time series

Rates of hyperkalemia after publication of the Randomized Aldactone Evaluation Study. Juurlink DN, Mamdani MM, Lee DS, Kopp A, Austin P, Laupacis A, Redelmeier D. N Engl J Med 2004; 351(6):543-51.

Design Overview: The combination of Aldactone (spironolactone) and angiotensin-converting enzyme (ACE) inhibitors for control of severe heart failure can result in life-threatening hyperkalemia. The investigators used the Ontario Drug Benefit Program computerized prescription records linked to hospitalization data (CIHI Discharge Abstracts) to assess the association between use of these drugs and subsequent hospital admissions involving a diagnosis of hyperkalemia. Associations with deaths during these hospitalizations were also assessed. The eight-year time span of the study was divided into 24 four-month intervals and time-series analysis was used to model the autocorrelations in these temporally-sequenced data. This analytic approach allowed the investigators to assess the immediate and delayed effects of the publication of the Aldactone trial results on usage, and to incorporate the impact of population-wide changes in the use of other drugs that could influence the risk of hyperkalemia.  

Self-matched case series

Guillain-Barré syndrome after influenza vaccination in adults: a population-based study. Juurlink D, Stukel T, Kwong J, Kopp A, McGeer A, Upshur R, Manuel D, Moineddin R, Wilson K. Arch Intern Med 2006; 166:2217–21.

Design Overview: Ontario has the largest mass vaccination program against influenza in the world. The investigators used the combination of physician claims (OHIP data) and hospitalization data (CIHI Discharge Abstracts) to identify a case series of Ontarians who had a vaccination and were hospitalized within 43 weeks for Guillain-Barré Syndrome. Subjects were excluded if they had any admission for GBS in the 18 months preceding the vaccination. The problem of confounding by indication is eliminated in this design because the case acts as her own control. The investigators were able to see a small, stable association using the self-matched case series that was not observable in a companion ecologic time series analysis of incident cases in Ontario. The longitudinal linkage of the OHIP and CIHI data made the use of this more powerful design possible.  

Multi-province cohort

Effectiveness of statins for secondary prevention in elderly patients after acute myocardial infarction: an evaluation of class effect. Zhou Z, Rahme E, Abrahamowicz M, Tu JV, Eisenberg MJ, Humphries K, Austin PC, Pilote L. CMAJ 2005; 172(9):1187–94.

Design Overview: Five different statins were examined for their relative effectiveness in preventing subsequent acute myocardial infarctions (AMI) in elderly persons in three provinces: Ontario, British Columbia and Quebec. Three comparable cohorts were identified using linked hospital discharge databases and physician and prescription claims databases. Death information was obtained from the provincial registry databases. Outcomes studied included subsequent AMI and/or death from any cause. All data were linked. Five statin groups were formed on the basis of the first statin prescribed. Covariates were determined from the hospital discharge and prescription databases and included: patient demographics, comorbidities, concurrent use of other major cardiac medications and previous use of statins in the year before the index AMI, in-hospital cardiac procedures, length of stay, specialty of treating physician, type of hospital, hospital volume and location, and availability of a cardiac catheterization facility. Multivariate Cox proportional hazards regression was used and data across provinces was pooled after checking for heterogeneity of effects.

Multi-province time series

Changes in rates of upper gastrointestinal hemorrhage after the introduction of cyclooxygenase-2 inhibitors in British Columbia and Ontario. Mamdani M, Warren L, Kopp A, Paterson JM, Laupacis A, Bassett K, Anderson GM. CMAJ 2006; 175(12):1535–8.

Design Overview: The investigators compared the use of cyclooxygenase-2 inhibitors and rates of hospitalizations for upper gastrointestinal bleeds over time in two provinces: British Columbia, which had very restrictive access to COX-2 inhibitors and Ontario, which had a less restrictive policy. The use of non-steroidal anti-inflammatory drugs was recorded for all persons over 66 years of age in the Ontario Drug Benefit database and the British Columbia Pharma Net Database, thus making this a population-based study in older adults. Hospitalization information was available from the Canadian Institute for Health Information Hospital Discharge Abstract database. Trends in potentially confounding medication use were also examined. This study demonstrated that although COX-2 inhibitors have a lower (but not zero) risk of GI bleed than other NSAIDs, drug benefit policy may actually increase the rate of bleeds at the population level by allowing use in those at risk of a bleed who would not otherwise have been on NSAIDs.  

Survival analysis with population-based case series data

Outcomes of surveillance mammography after treatment of primary breast cancer: a population-based case series. Paszat L, Sutradhar R, Grunfeld E, Gainford C, Benk V, Bondy S, Coyle D, Holloway C, Sawka C, Shumak R, Vallis K, van Walraven C. Breast Cancer Res Treat  2008; 114(1):169-78.

Design Overview: The effectiveness of the use of surveillance mammography after treatment for breast cancer is not well documented. The investigators used a linked database that included information about diagnosis from the Ontario Cancer Registry, episodes of breast surgery from the CIHI discharge abstract database, and information about mammography from the Ontario physician billings database. These data were enhanced with information abstracted from operative, pathology and mammography reports. They identified all women who had unilateral stage I or II primary breast cancer and sampled based on presence or absence of subsequent breast surgery. Surveillance mammograms were identified by reviewing mammogram reports. Outcomes included death form breast cancer, breast cancer recurrence and histologically confirmed subsequent breast cancers.  A Cox proportional hazards regression analysis computed the association between the use of surveillance mammography and breast cancer death adjusting for age, stage and treatment.