Institute for Clinical Evaluative Sciences

The propensity score is the probability of treatment assignment conditional on observed baseline characteristics. The propensity score allows one to design and analyze an observational (nonrandomized) study so that it mimics some of the particular characteristics of a randomized controlled trial. In particular, the propensity score is a balancing score: conditional on the propensity score, the distribution of observed baseline covariates will be similar between treated and untreated subjects. I describe 4 different propensity score methods: matching on the propensity score, stratification on the propensity score, inverse probability of treatment weighting using the propensity score, and covariate adjustment using the propensity score. I describe balance diagnostics for examining whether the propensity score model has been adequately specified. Furthermore, I discuss differences between regression-based methods and propensity score-based methods for the analysis of observational data. I describe different causal average treatment effects and their relationship with propensity score analyses.

The number of published systematic reviews of studies of healthcare interventions has increased rapidly and these are used extensively for clinical and policy decisions. Systematic reviews are subject to a range of biases and increasingly include non-randomised studies of interventions. It is important that users can distinguish high quality reviews. Many instruments have been designed to evaluate different aspects of reviews, but there are few comprehensive critical appraisal instruments. AMSTAR was developed to evaluate systematic reviews of randomised trials. In this paper, we report on the updating of AMSTAR and its adaptation to enable more detailed assessment of systematic reviews that include randomised or non-randomised studies of healthcare interventions, or both. With moves to base more decisions on real world observational evidence we believe that AMSTAR 2 will assist decision makers in the identification of high quality systematic reviews, including those based on non-randomised studies of healthcare interventions<i>.</i>

The propensity score is a subject's probability of treatment, conditional on observed baseline covariates. Conditional on the true propensity score, treated and untreated subjects have similar distributions of observed baseline covariates. Propensity-score matching is a popular method of using the propensity score in the medical literature. Using this approach, matched sets of treated and untreated subjects with similar values of the propensity score are formed. Inferences about treatment effect made using propensity-score matching are valid only if, in the matched sample, treated and untreated subjects have similar distributions of measured baseline covariates. In this paper we discuss the following methods for assessing whether the propensity score model has been correctly specified: comparing means and prevalences of baseline characteristics using standardized differences; ratios comparing the variance of continuous covariates between treated and untreated subjects; comparison of higher order moments and interactions; five-number summaries; and graphical methods such as quantile-quantile plots, side-by-side boxplots, and non-parametric density plots for comparing the distribution of baseline covariates between treatment groups. We describe methods to determine the sampling distribution of the standardized difference when the true standardized difference is equal to zero, thereby allowing one to determine the range of standardized differences that are plausible with the propensity score model having been correctly specified. We highlight the limitations of some previously used methods for assessing the adequacy of the specification of the propensity-score model. In particular, methods based on comparing the distribution of the estimated propensity score between treated and untreated subjects are uninformative.

Routinely collected health data, obtained for administrative and clinical purposes without specific a priori research goals, are increasingly used for research. The rapid evolution and availability of these data have revealed issues not addressed by existing reporting guidelines, such as Strengthening the Reporting of Observational Studies in Epidemiology (STROBE). The REporting of studies Conducted using Observational Routinely collected health Data (RECORD) statement was created to fill these gaps. RECORD was created as an extension to the STROBE statement to address reporting items specific to observational studies using routinely collected health data. RECORD consists of a checklist of 13 items related to the title, abstract, introduction, methods, results, and discussion section of articles, and other information required for inclusion in such research reports. This document contains the checklist and explanatory and elaboration information to enhance the use of the checklist. Examples of good reporting for each RECORD checklist item are also included herein. This document, as well as the accompanying website and message board (http://www.record-statement.org), will enhance the implementation and understanding of RECORD. Through implementation of RECORD, authors, journals editors, and peer reviewers can encourage transparency of research reporting.

The propensity score is defined as a subject's probability of treatment selection, conditional on observed baseline covariates. Weighting subjects by the inverse probability of treatment received creates a synthetic sample in which treatment assignment is independent of measured baseline covariates. Inverse probability of treatment weighting (IPTW) using the propensity score allows one to obtain unbiased estimates of average treatment effects. However, these estimates are only valid if there are no residual systematic differences in observed baseline characteristics between treated and control subjects in the sample weighted by the estimated inverse probability of treatment. We report on a systematic literature review, in which we found that the use of IPTW has increased rapidly in recent years, but that in the most recent year, a majority of studies did not formally examine whether weighting balanced measured covariates between treatment groups. We then proceed to describe a suite of quantitative and qualitative methods that allow one to assess whether measured baseline covariates are balanced between treatment groups in the weighted sample. The quantitative methods use the weighted standardized difference to compare means, prevalences, higher-order moments, and interactions. The qualitative methods employ graphical methods to compare the distribution of continuous baseline covariates between treated and control subjects in the weighted sample. Finally, we illustrate the application of these methods in an empirical case study. We propose a formal set of balance diagnostics that contribute towards an evolving concept of 'best practice' when using IPTW to estimate causal treatment effects using observational data.

In a study comparing the effects of two treatments, the propensity score is the probability of assignment to one treatment conditional on a subject's measured baseline covariates. Propensity-score matching is increasingly being used to estimate the effects of exposures using observational data. In the most common implementation of propensity-score matching, pairs of treated and untreated subjects are formed whose propensity scores differ by at most a pre-specified amount (the caliper width). There has been a little research into the optimal caliper width. We conducted an extensive series of Monte Carlo simulations to determine the optimal caliper width for estimating differences in means (for continuous outcomes) and risk differences (for binary outcomes). When estimating differences in means or risk differences, we recommend that researchers match on the logit of the propensity score using calipers of width equal to 0.2 of the standard deviation of the logit of the propensity score. When at least some of the covariates were continuous, then either this value, or one close to it, minimized the mean square error of the resultant estimated treatment effect. It also eliminated at least 98% of the bias in the crude estimator, and it resulted in confidence intervals with approximately the correct coverage rates. Furthermore, the empirical type I error rate was approximately correct. When all of the covariates were binary, then the choice of caliper width had a much smaller impact on the performance of estimation of risk differences and differences in means.

BACKGROUND: Although the relation between hospital volume and surgical mortality is well established, for most procedures, the relative importance of the experience of the operating surgeon is uncertain. METHODS: Using information from the national Medicare claims data base for 1998 through 1999, we examined mortality among all 474,108 patients who underwent one of eight cardiovascular procedures or cancer resections. Using nested regression models, we examined the relations between operative mortality and surgeon volume and hospital volume (each in terms of total procedures performed per year), with adjustment for characteristics of the patients and other characteristics of the providers. RESULTS: Surgeon volume was inversely related to operative mortality for all eight procedures (P=0.003 for lung resection, P<0.001 for all other procedures). The adjusted odds ratio for operative death (for patients with a low-volume surgeon vs. those with a high-volume surgeon) varied widely according to the procedure--from 1.24 for lung resection to 3.61 for pancreatic resection. Surgeon volume accounted for a large proportion of the apparent effect of the hospital volume, to an extent that varied according to the procedure: it accounted for 100 percent of the effect for aortic-valve replacement, 57 percent for elective repair of an abdominal aortic aneurysm, 55 percent for pancreatic resection, 49 percent for coronary-artery bypass grafting, 46 percent for esophagectomy, 39 percent for cystectomy, and 24 percent for lung resection. For most procedures, the mortality rate was higher among patients of low-volume surgeons than among those of high-volume surgeons, regardless of the surgical volume of the hospital in which they practiced. CONCLUSIONS: For many procedures, the observed associations between hospital volume and operative mortality are largely mediated by surgeon volume. Patients can often improve their chances of survival substantially, even at high-volume hospitals, by selecting surgeons who perform the operations frequently.

BACKGROUND: The magnitude, consistency, and manner of association between sedentary time and outcomes independent of physical activity remain unclear. PURPOSE: To quantify the association between sedentary time and hospitalizations, all-cause mortality, cardiovascular disease, diabetes, and cancer in adults independent of physical activity. DATA SOURCES: English-language studies in MEDLINE, PubMed, EMBASE, CINAHL, Cochrane Library, Web of Knowledge, and Google Scholar databases were searched through August 2014 with hand-searching of in-text citations and no publication date limitations. STUDY SELECTION: Studies assessing sedentary behavior in adults, adjusted for physical activity and correlated to at least 1 outcome. DATA EXTRACTION: Two independent reviewers performed data abstraction and quality assessment, and a third reviewer resolved inconsistencies. DATA SYNTHESIS: Forty-seven articles met our eligibility criteria. Meta-analyses were performed on outcomes for cardiovascular disease and diabetes (14 studies), cancer (14 studies), and all-cause mortality (13 studies). Prospective cohort designs were used in all but 3 studies; sedentary times were quantified using self-report in all but 1 study. Significant hazard ratio (HR) associations were found with all-cause mortality (HR, 1.240 [95% CI, 1.090 to 1.410]), cardiovascular disease mortality (HR, 1.179 [CI, 1.106 to 1.257]), cardiovascular disease incidence (HR, 1.143 [CI, 1.002 to 1.729]), cancer mortality (HR, 1.173 [CI, 1.108 to 1.242]), cancer incidence (HR, 1.130 [CI, 1.053 to 1.213]), and type 2 diabetes incidence (HR, 1.910 [CI, 1.642 to 2.222]). Hazard ratios associated with sedentary time and outcomes were generally more pronounced at lower levels of physical activity than at higher levels. LIMITATION: There was marked heterogeneity in research designs and the assessment of sedentary time and physical activity. CONCLUSION: Prolonged sedentary time was independently associated with deleterious health outcomes regardless of physical activity. PRIMARY FUNDING SOURCE: None.

Competing risks occur frequently in the analysis of survival data. A competing risk is an event whose occurrence precludes the occurrence of the primary event of interest. In a study examining time to death attributable to cardiovascular causes, death attributable to noncardiovascular causes is a competing risk. When estimating the crude incidence of outcomes, analysts should use the cumulative incidence function, rather than the complement of the Kaplan-Meier survival function. The use of the Kaplan-Meier survival function results in estimates of incidence that are biased upward, regardless of whether the competing events are independent of one another. When fitting regression models in the presence of competing risks, researchers can choose from 2 different families of models: modeling the effect of covariates on the cause-specific hazard of the outcome or modeling the effect of covariates on the cumulative incidence function. The former allows one to estimate the effect of the covariates on the rate of occurrence of the outcome in those subjects who are currently event free. The latter allows one to estimate the effect of covariates on the absolute risk of the outcome over time. The former family of models may be better suited for addressing etiologic questions, whereas the latter model may be better suited for estimating a patient's clinical prognosis. We illustrate the application of these methods by examining cause-specific mortality in patients hospitalized with heart failure. Statistical software code in both R and SAS is provided.

Researchers are increasingly using the standardized difference to compare the distribution of baseline covariates between treatment groups in observational studies. Standardized differences were initially developed in the context of comparing the mean of continuous variables between two groups. However, in medical research, many baseline covariates are dichotomous. In this article, we explore the utility and interpretation of the standardized difference for comparing the prevalence of dichotomous variables between two groups. We examined the relationship between the standardized difference, and the maximal difference in the prevalence of the binary variable between two groups, the relative risk relating the prevalence of the binary variable in one group compared to the prevalence in the other group, and the phi coefficient for measuring correlation between the treatment group and the binary variable. We found that a standardized difference of 10% (or 0.1) is equivalent to having a phi coefficient of 0.05 (indicating negligible correlation) for the correlation between treatment group and the binary variable.

BACKGROUND: Comorbidity measures are necessary to describe patient populations and adjust for confounding. In direct comparisons, studies have found the Elixhauser comorbidity system to be statistically slightly superior to the Charlson comorbidity system at adjusting for comorbidity. However, the Elixhauser classification system requires 30 binary variables, making its use for reporting and analysis of comorbidity cumbersome. OBJECTIVE: Modify the Elixhauser classification system into a single numeric score for administrative data. METHODS: For all hospitalizations at the Ottawa Hospital, Canada, between 1996 and 2008, we determined if International Classification of Disease codes for chronic diagnoses were in any of the 30 Elixhauser comorbidity groups. We then used backward stepwise multivariate logistic regression to determine the independent association of each comorbidity group with death in hospital. Regression coefficients were modified into a scoring system that reflected the strength of each comorbidity group's independent association with hospital death. RESULTS: Hospitalizations that were included were 345,795 (derivation: 228,565; validation 117,230). Twenty-one of the 30 groups were independently associated with hospital mortality. The resulting comorbidity score had an equivalent discrimination in the derivation and validation groups (overall c-statistic 0.763, 95% CI: 0.759-0.766). This was similar to models having all Elixhauser groups (0.760, 95% CI: 0.756-0.764) or significant groups only (0.759, 95% CI: 0.754-0.762), but significantly exceeded discrimination when comorbidity was expressed using the Charlson score (0.745, 95% CI: 0.742-0.749). CONCLUSION: When analyzing administrative data, the Elixhauser comorbidity system can be condensed to a single numeric score that summarizes disease burden and is adequately discriminative for death in hospital.

BACKGROUND: The importance of heart failure with preserved ejection fraction is increasingly recognized. We conducted a study to evaluate the epidemiologic features and outcomes of patients with heart failure with preserved ejection fraction and to compare the findings with those from patients who had heart failure with reduced ejection fraction. METHODS: From April 1, 1999, through March 31, 2001, we studied 2802 patients admitted to 103 hospitals in the province of Ontario, Canada, with a discharge diagnosis of heart failure whose ejection fraction had also been assessed. The patients were categorized in three groups: those with an ejection fraction of less than 40 percent (heart failure with reduced ejection fraction), those with an ejection fraction of 40 to 50 percent (heart failure with borderline ejection fraction), and those with an ejection fraction of more than 50 percent (heart failure with preserved ejection fraction). Two groups were studied in detail: those with an ejection fraction of less than 40 percent and those with an ejection fraction of more than 50 percent. The main outcome measures were death within one year and readmission to the hospital for heart failure. RESULTS: Thirty-one percent of the patients had an ejection fraction of more than 50 percent. Patients with heart failure with preserved ejection fraction were more likely to be older and female and to have a history of hypertension and atrial fibrillation. The presenting history and clinical examination findings were similar for the two groups. The unadjusted mortality rates for patients with an ejection fraction of more than 50 percent were not significantly different from those for patients with an ejection fraction of less than 40 percent at 30 days (5 percent vs. 7 percent, P=0.08) and at 1 year (22 percent vs. 26 percent, P=0.07); the adjusted one-year mortality rates were also not significantly different in the two groups (hazard ratio, 1.13; 95 percent confidence interval, 0.94 to 1.36; P=0.18). The rates of readmission for heart failure and of in-hospital complications did not differ between the two groups. CONCLUSIONS: Among patients presenting with new-onset heart failure, a substantial proportion had an ejection fraction of more than 50 percent. The survival of patients with heart failure with preserved ejection fraction was similar to that of patients with reduced ejection fraction.

BACKGROUND: The CONSORT statement is intended to improve reporting of randomised controlled trials and focuses on minimising the risk of bias (internal validity). The applicability of a trial's results (generalisability or external validity) is also important, particularly for pragmatic trials. A pragmatic trial (a term first used in 1967 by Schwartz and Lellouch) can be broadly defined as a randomised controlled trial whose purpose is to inform decisions about practice. This extension of the CONSORT statement is intended to improve the reporting of such trials and focuses on applicability. Methods At two, two-day meetings held in Toronto in 2005 and 2008, we reviewed the CONSORT statement and its extensions, the literature on pragmatic trials and applicability, and our experiences in conducting pragmatic trials. Recommendations We recommend extending eight CONSORT checklist items for reporting of pragmatic trials: the background, participants, interventions, outcomes, sample size, blinding, participant flow, and generalisability of the findings. These extensions are presented, along with illustrative examples of reporting, and an explanation of each extension. Adherence to these reporting criteria will make it easier for decision makers to judge how applicable the results of randomised controlled trials are to their own conditions. Empirical studies are needed to ascertain the usefulness and comprehensiveness of these CONSORT checklist item extensions. In the meantime we recommend that those who support, conduct, and report pragmatic trials should use this extension of the CONSORT statement to facilitate the use of trial results in decisions about health care.

BACKGROUND: The Randomized Aldactone Evaluation Study (RALES) demonstrated that spironolactone significantly improves outcomes in patients with severe heart failure. Use of angiotensin-converting-enzyme (ACE) inhibitors is also indicated in these patients. However, life-threatening hyperkalemia can occur when these drugs are used together. METHODS: We conducted a population-based time-series analysis to examine trends in the rate of spironolactone prescriptions and the rate of hospitalization for hyperkalemia in ambulatory patients before and after the publication of RALES. We linked prescription-claims data and hospital-admission records for more than 1.3 million adults 66 years of age or older in Ontario, Canada, for the period from 1994 through 2001. RESULTS: Among patients treated with ACE inhibitors who had recently been hospitalized for heart failure, the spironolactone-prescription rate was 34 per 1000 patients in 1994, and it increased immediately after the publication of RALES, to 149 per 1000 patients by late 2001 (P<0.001). The rate of hospitalization for hyperkalemia rose from 2.4 per 1000 patients in 1994 to 11.0 per 1000 patients in 2001 (P<0.001), and the associated mortality rose from 0.3 per 1000 to 2.0 per 1000 patients (P<0.001). As compared with expected numbers of events, there were 560 (95 percent confidence interval, 285 to 754) additional hyperkalemia-related hospitalizations and 73 (95 percent confidence interval, 27 to 120) additional hospital deaths during 2001 among older patients with heart failure who were treated with ACE inhibitors in Ontario. Publication of RALES was not associated with significant decreases in the rates of readmission for heart failure or death from all causes. CONCLUSIONS: The publication of RALES was associated with abrupt increases in the rate of prescriptions for spironolactone and in hyperkalemia-associated morbidity and mortality. Closer laboratory monitoring and more judicious use of spironolactone may reduce the occurrence of this complication.

BACKGROUND: The health implications of regional differences in Medicare spending are unknown. OBJECTIVE: To determine whether regions with higher Medicare spending provide better care. DESIGN: Cohort study. SETTING: National study of Medicare beneficiaries. PATIENTS: Patients hospitalized between 1993 and 1995 for hip fracture (n = 614,503), colorectal cancer (n = 195,429), or acute myocardial infarction (n = 159,393) and a representative sample (n = 18,190) drawn from the Medicare Current Beneficiary Survey (1992-1995). EXPOSURE MEASUREMENT: End-of-life spending reflects the component of regional variation in Medicare spending that is unrelated to regional differences in illness. Each cohort member's exposure to different levels of spending was therefore defined by the level of end-of-life spending in his or her hospital referral region of residence (n = 306). OUTCOME MEASUREMENTS: Content of care (for example, frequency and type of services received), quality of care (for example, use of aspirin after acute myocardial infarction, influenza immunization), and access to care (for example, having a usual source of care). RESULTS: Average baseline health status of cohort members was similar across regions of differing spending levels, but patients in higher-spending regions received approximately 60% more care. The increased utilization was explained by more frequent physician visits, especially in the inpatient setting (rate ratios in the highest vs. the lowest quintile of hospital referral regions were 2.13 [95% CI, 2.12 to 2.14] for inpatient visits and 2.36 [CI, 2.33 to 2.39] for new inpatient consultations), more frequent tests and minor (but not major) procedures, and increased use of specialists and hospitals (rate ratio in the highest vs. the lowest quintile was 1.52 [CI, 1.50 to 1.54] for inpatient days and 1.55 [CI, 1.50 to 1.60] for intensive care unit days). Quality of care in higher-spending regions was no better on most measures and was worse for several preventive care measures. Access to care in higher-spending regions was also no better or worse. CONCLUSIONS: Regional differences in Medicare spending are largely explained by the more inpatient-based and specialist-oriented pattern of practice observed in high-spending regions. Neither quality of care nor access to care appear to be better for Medicare enrollees in higher-spending regions.

BACKGROUND: Asthma is a major cause of disability, health resource utilization and poor quality of life world-wide. We set out to generate estimates of the global burden of asthma in adults, which may inform the development of strategies to address this common disease. METHODS: The World Health Survey (WHS) was developed and implemented by the World Health Organization in 2002-2003. A total of 178,215 individuals from 70 countries aged 18 to 45 years responded to questions related to asthma and related symptoms. The prevalence of asthma was based on responses to questions relating to self-reported doctor diagnosed asthma, clinical/treated asthma, and wheezing in the last 12 months. RESULTS: The global prevalence rates of doctor diagnosed asthma, clinical/treated asthma and wheezing in adults were 4.3%, 4.5%, and 8.6% respectively, and varied by as much as 21-fold amongst the 70 countries. Australia reported the highest rate of doctor diagnosed, clinical/treated asthma, and wheezing (21.0%, 21.5%, and 27.4%). Amongst those with clinical/treated asthma, almost 24% were current smokers, half reported wheezing, and 20% had never been treated for asthma. CONCLUSIONS: This study provides a global estimate of the burden of asthma in adults, and suggests that asthma continues to be a major public health concern worldwide. The high prevalence of smoking remains a major barrier to combating the global burden of asthma. While the highest prevalence rates were observed in resource-rich countries, resource-poor nations were also significantly affected, posing a barrier to development as it stretches further the demands of non-communicable diseases.

Propensity score methods are increasingly being used to estimate causal treatment effects in observational studies. In medical and epidemiological studies, outcomes are frequently time-to-event in nature. Propensity-score methods are often applied incorrectly when estimating the effect of treatment on time-to-event outcomes. This article describes how two different propensity score methods (matching and inverse probability of treatment weighting) can be used to estimate the measures of effect that are frequently reported in randomized controlled trials: (i) marginal survival curves, which describe survival in the population if all subjects were treated or if all subjects were untreated; and (ii) marginal hazard ratios. The use of these propensity score methods allows one to replicate the measures of effect that are commonly reported in randomized controlled trials with time-to-event outcomes: both absolute and relative reductions in the probability of an event occurring can be determined. We also provide guidance on variable selection for the propensity score model, highlight methods for assessing the balance of baseline covariates between treated and untreated subjects, and describe the implementation of a sensitivity analysis to assess the effect of unmeasured confounding variables on the estimated treatment effect when outcomes are time-to-event in nature. The methods in the paper are illustrated by estimating the effect of discharge statin prescribing on the risk of death in a sample of patients hospitalized with acute myocardial infarction. In this tutorial article, we describe and illustrate all the steps necessary to conduct a comprehensive analysis of the effect of treatment on time-to-event outcomes.

Colorectal cancer (CRC) ranks third among the most commonly diagnosed cancers worldwide, with wide geographical variation in incidence and mortality across the world. Despite proof that screening can decrease CRC incidence and mortality, CRC screening is only offered to a small proportion of the target population worldwide. Throughout the world there are widespread differences in CRC screening implementation status and strategy. Differences can be attributed to geographical variation in CRC incidence, economic resources, healthcare structure and infrastructure to support screening such as the ability to identify the target population at risk and cancer registry availability. This review highlights issues to consider when implementing a CRC screening programme and gives a worldwide overview of CRC burden and the current status of screening programmes, with focus on international differences.

OBJECTIVE: Accurate information about the magnitude and distribution of diabetes can inform policy and support health care evaluation. We linked physician service claims (PSCs) and hospital discharge abstracts (HDAs) to determine diabetes prevalence and incidence. RESEARCH DESIGN AND METHODS: A retrospective cohort was constructed using administrative data from the national HDA database, PSCs for Ontario (population 11 million), and registries carrying demographics and vital statistics. All HDAs and PSCs bearing a diagnosis of diabetes (ICD9-CM 250) were selected for 1991-1999. Two previously reported algorithms for identification of diabetes were applied as follows: "1-claim" (any HDA or PSC showing diabetes) and "2-claim" (one HDA or two PSCs within 2 years showing diabetes). Incident cases were defined as individuals who met the criteria for diabetes for the first time after at least 2 years of observation. For validation, diagnostic data abstracted from primary care charts (n=3,317) of 57 randomly selected physicians were linked to the administrative data cohort, and sensitivity and specificity were calculated. RESULTS: -In 1998, 696,938 individuals met the 1-claim criteria and 528,280 met the 2-claim criteria. Sensitivity for diabetes was 90 and 86%; for the 1- and 2-claim algorithms, specificity was 92 and 97%, respectively, and positive predictive values were 61 and 80%, respectively. Using the 2-claim algorithm, the all-age prevalence increased from 3.2% in 1993 to 4.5% in 1998 (6.1% in adults). Incidence remained stable. CONCLUSIONS: Administrative data can be used to establish population-based incidence and prevalence of diabetes. Diabetes prevalence is increasing in Ontario and is considerably higher than self-reported rates.

BACKGROUND: Colonoscopy is advocated for screening and prevention of colorectal cancer (CRC), but randomized trials supporting the benefit of this practice are not available. OBJECTIVE: To evaluate the association between colonoscopy and CRC deaths. DESIGN: Population-based, case-control study. SETTING: Ontario, Canada. PATIENTS: Persons age 52 to 90 years who received a CRC diagnosis from January 1996 to December 2001 and died of CRC by December 2003. Five controls matched by age, sex, geographic location, and socioeconomic status were randomly selected for each case patient. MEASUREMENTS: Administrative claims data were used to detect exposure to any colonoscopy and complete colonoscopy (to the cecum) from January 1992 to an index date 6 months before diagnosis in each case patient and the same assigned date in matched controls. Exposures in case patients and controls were compared by using conditional logistic regression to control for comorbid conditions. Secondary analyses were done to see whether associations differed by site of primary CRC, age, or sex. RESULTS: 10 292 case patients and 51 460 controls were identified; 719 case patients (7.0%) and 5031 controls (9.8%) had undergone colonoscopy. Compared with controls, case patients were less likely to have undergone any attempted colonoscopy (adjusted conditional odds ratio [OR], 0.69 [95% CI, 0.63 to 0.74; P < 0.001]) or complete colonoscopy (adjusted conditional OR, 0.63 [CI, 0.57 to 0.69; P < 0.001]). Complete colonoscopy was strongly associated with fewer deaths from left-sided CRC (adjusted conditional OR, 0.33 [CI, 0.28 to 0.39]) but not from right-sided CRC (adjusted conditional OR, 0.99 [CI, 0.86 to 1.14]). LIMITATION: Screening could not be differentiated from diagnostic procedures. CONCLUSION: In usual practice, colonoscopy is associated with fewer deaths from CRC. This association is primarily limited to deaths from cancer developing in the left side of the colon.